Imaging modalities to inform the detection and diagnosis of early caries

Summary of findings 1. Summary of findings table ‐ main results

Question	What is the diagnostic accuracy of radiographic methods for the detection and diagnosis of early dental caries?
Population	Asymptomatic children or adults presenting for clinical examination (clinical studies); extracted teeth of children or adults (in vitro studies). Clinical or in vitro studies which intentionally included dentine and frank cavitations for assessment were excluded
Index test	Intra and extraoral radiographic caries detection methods completed on intact teeth including: analogue (conventional) radiographs; digital radiographs; and CBCT
Comparator test	Estimates were compared across different radiographic methods. A separate review in this series explores the comparative accuracy of visual classification, fluorescence‐based, radiograph, and transillumination methods of detection and diagnosis
Target condition	Early dental caries (positivity threshold of early caries or beyond)
Reference standard	Histology, excavation, enhanced visual examination
Action	If dental caries can be detected at an early stage then remedial action can be taken to arrest or even reverse decay, and potentially prevent restorations
Diagnostic stage	Aimed at the general dental practitioner assessing regularly attending patients for early stage caries
Quantity of evidence	77 studies providing 104 datasets; 9331 lesions in 15,518 tooth sites or surfaces (median 63% prevalence)
Findings
Sensitivity (95% CI)^a		0.47 (0.40 to 0.53)
Specificity (95% CI)^a		0.88 (0.84 to 0.92)
Outcome	Numbers applied to a hypothetical cohort of 1000 tooth surfaces Effect per 1000 tooth surfaces assessed (95% CI)		Test accuracy Certainty of the evidence
Outcome	Pre‐test probability 28%^b	Pre‐test probability 63%^b	Test accuracy Certainty of the evidence
True positives (tooth surfaces with early enamel caries)	130 (113 to 148)	293 (254 to 332)	Sensitivity ‐ ⊕⊕⊝⊝ LOW Specificity ‐ ⊕⊕⊝⊝ LOW
False negatives (tooth surfaces incorrectly classified as not having early enamel caries)	150 (132 to 167)	337 (298 to 376)
True negatives (tooth surfaces without early enamel caries)	636 (606 to 660)	327 (312 to 339)
False positives (tooth surfaces incorrectly classified as having early enamel caries)	84 (60 to 114)	43 (31 to 58)
Limitations ‐ factors that may decrease the certainty of the evidence
Risk of bias	No studies were considered to be at low risk of bias overall. Across the 4 domains, the patient selection domain had the highest proportion of studies judged at high risk of bias (56%). We judged the index test, reference standard, and flow and timing domains to have a lower proportion of high risk of bias studies (39%, 16%, and 9% respectively). All but 1 study avoided a case‐control design, 57% avoided inappropriate exclusions, whilst only 9% reported the use of random or consecutive sampling. Most included studies were in vitro studies using histology as the reference standard, and likely to correctly classify the target condition, however some studies used an imperfect reference standard such as excavation or a visual examination with or without separation of the approximal surfaces
Applicability of evidence to question	16 studies (21%) were considered to have low concern for applicability across all domains. Applicability of patients and setting was of high concern in 75% of studies, mostly because they investigated extracted teeth and, as the objective of this review was to inform general clinical practice, there may be concerns that such results may have limited relevance to a clinical setting. The conduct or interpretation of the index test was of high concern in 29% of studies. There was low concern that the target condition as defined by the reference standard does not match the review question in all studies
Certainty of the evidence	We rated the certainty of the evidence as low, and downgraded 2 levels in total for risk of bias due to limitations in the design and conduct of the included studies, indirectness arising from the in vitro studies, and inconsistency of the results
^aThere was variability in the results of the individual studies, with sensitivities which ranged from 0 to 0.96 and specificities from 0 to 1.00. The variability in the results was partly explained by different methods evaluated. ^bHypothetical cohorts of 1000 tooth sites or surfaces are presented for numbers estimated at prevalence of 28% and 63% of enamel caries prevalence. Based on consultation with clinical colleagues, the lower prevalence figure addresses concerns that the higher prevalence value is not representative of the general population and is taken from the level of cavitated teeth in the UK Adult Dental Health Survey (Steele 2011). The higher prevalence figure is the median prevalence of early caries reported in the included studies. CBCT: cone beam computed tomography; CI: confidence interval.

Summary of findings 2. Summary of findings table ‐ comparison of tests

Question	What is the diagnostic accuracy of radiographic methods for the detection and diagnosis of early dental caries?
Population	Asymptomatic children or adults presenting for clinical examination (clinical studies); extracted teeth of children or adults (in vitro studies). Clinical or in vitro studies which intentionally included dentine and frank cavitations for assessment were excluded
Index test	Intra and extraoral radiographic caries detection methods completed on intact teeth including: analogue (conventional) radiographs; digital radiographs; and CBCT For the purposes of this review the positivity threshold was caries in enamel
Comparator test	Estimates were compared across different radiographic methods. A separate review in this series explores the comparative accuracy of visual classification, fluorescence‐based, radiograph, and transillumination methods of detection and diagnosis
Target condition	Early dental caries (positivity threshold of early caries or beyond)
Reference standard	Histology, excavation, enhanced visual assessment
Action	If dental caries can be detected at an early stage then remedial action can be taken to arrest or even reverse decay, and potentially prevent restorations
Diagnostic stage	Aimed at the general dental practitioner assessing regularly attending patients for early stage caries
Quantity of evidence	77 studies providing 104 datasets; 9331 lesions in 15,518 tooth surfaces (median 63% prevalence)
Findings: analysis comparing analogue radiographs, digital radiographs, CBCT. Consequences in a cohort of 1000 tooth sites or surfaces
Test	Datasets	Tooth surfaces	Sensitivity (95% CI)	Specificity (95% CI)	Pre‐test probability 28%		Pre‐test probability 63%
Test	Datasets	Tooth surfaces	Sensitivity (95% CI)	Specificity (95% CI)	Missed	Overdiagnosis	Missed	Overdiagnosis
Analogue radiograph	55	8589	0.44 (0.38 to 0.50)	0.90 (0.86 to 0.93)	157 (139 to 174)	75 (53 to 104)	353 (312 to 392)	38 (27 to 53)
Digital radiograph	42	5936	0.49 (0.42 to 0.56)	0.87 (0.82 to 0.91)	143 (125 to 162)	92 (66 to 128)	323 (280 to 365)	47 (34 to 66)
CBCT	9	1081	0.60 (0.51 to 0.68)	0.81 (0.73 to 0.88)	112 (89 to 137)	134 (89 to 196)	253 (200 to 307)	69 (46 to 101)
The addition of test type to the model resulted in a meaningful difference to the sensitivity and specificity estimates: Chi² = 32.44, df = 4, P < 0.001
Interpretation: these results should be interpreted taking into account the factors that limit the certainty of the evidence as indicated in summary of findings Table 1
Using analogue as the reference standard: difference in sensitivity for CBCT = ‐0.16 (‐0.23 to ‐0.09), P < 0.001; digital = ‐0.04 (‐0.09 to ‐0.01), P = 0.018 difference in specificity for CBCT = 0.08 (0.02 to 0.14), P = 0.007; digital = 0.02 (‐0.01 to 0.06), P = 0.107
CBCT: cone beam computed tomography; CI: confidence interval; df: degrees of freedom.

Background

Cochrane Oral Health (COH) has undertaken several systematic reviews of diagnostic test accuracy (DTA) on methods to inform the detection and diagnosis of early dental caries. The suite of systematic reviews forms part of a UK National Institute for Health Research (NIHR) Cochrane Programme Grant Scheme and involved collaboration with the Complex Reviews Support Unit. The reviews follow standard Cochrane DTA methodology and are differentiated according to the index test under evaluation. A generic protocol serves as the basis for the suite of systematic reviews (Macey 2018).

Caries is an entire disease process, which can be stabilised and sometimes reversed if diagnosed and treated early on in the disease process (Fejerskov 2015; Pitts 2009).

Most high‐income countries around the world have evidenced a reduction in caries incidence in children and adolescents, and in some Scandinavian countries preventive programmes have almost eradicated caries, but such activities have not been widely replicated in other locations (Pitts 2017). The 2015 Global Burden of Disease study identified dental caries as the most prevalent, preventable condition worldwide (Feigin 2016; Kassebaum 2015), affecting 60% to 90% of children and the majority of adults of the world's population (Petersen 2005). Furthermore, despite a reduction in caries in some industrialised countries, the global incidence of untreated caries was reported to be 2.4 billion in 2010 (Feigin 2016; Kassebaum 2015; World Health Organization 2017) and continues to increase year on year. In the UK, statistics indicate that the primary reason for childhood (aged 5 to 9 years) hospital admissions is for the extraction of teeth (Public Health England 2014). Longitudinal studies have shown that those who experience caries early in childhood will have an increased risk of severe caries in later life, and that the disease trajectory will be steeper than those without early caries experience (Broadbent 2008; Hall‐Scullin 2017).

Untreated caries can lead to episodes of severe pain and infection, often requiring treatment with antibiotics. Dental anxiety, resulting from the failure to treat caries and the subsequent need for more invasive management, can adversely affect a person's future willingness to visit their dentist, leading to a downward spiral of oral disease (Milsom 2003; Thomson 2000). If left to progress, treatment options are limited to restoration or extraction, requiring repeated visits to a dental surgery or even to a hospital (Featherstone 2004; Fejerskov 2015; Kidd 2004).

The cost of treating caries is high. In the UK alone, the National Health Service (NHS) spends around GBP half a billion every year in treating the disease. Hidden costs also exist, and the related productivity losses are high, estimated at USD 27 billion globally in 2010 (Listl 2015).

Caries detection and diagnosis will most often be undertaken at a routine dental examination, by a general dental practitioner. However, caries detection can additionally be employed in secondary care settings, school, or community screening projects and epidemiology or research studies (Braga 2009; Jones 2017). The traditional method of detecting dental caries in clinical practice is a visual‐tactile examination often with supporting radiographic investigations. This combination of methods is believed to be successful at detecting caries that has progressed into dentine and reached a threshold where a restoration may be necessary (Kidd 2004). The detection of caries earlier in the disease continuum could lead to stabilisation of disease or even possible remineralisation of the tooth surface, thus preventing the patient from entering a lifelong cycle of restoration (Pitts 2017; Splieth 2020). However, early caries is difficult to detect visually, and the use of radiographs provides limited ability to detect small changes in dental enamel (Ismail 2007).

Detection and diagnosis at the initial (non‐cavitated) and moderate levels of caries is fundamental in achieving the promotion of oral health and prevention of oral disease (Fejerskov 2015; Ismail 2013). The prevalence of this early caries state is not often reported in dental epidemiology, with most reports preferring to focus on cavitated/dentinal lesions which may be easier to detect. For example, the most recent UK Adult Dental Health survey reported that 31% of the sample had untreated caries into dentine (Steele 2011; White 2012), and a US study reported levels of cavities at 15.30% in 12‐ to 19‐year olds (Dye 2015). However, one UK survey of children identified "clinical decay experience" which incorporates any enamel breakdown and all other form of caries and reported a prevalence of 63% in 15‐year olds (Vernazza 2016).

A wide variety of management options are available under NHS care at different thresholds of disease, ranging from non‐operative preventive strategies such as improved oral hygiene, a reduced sugar diet, and the application of topical fluoride, to minimally invasive treatments (e.g. sealing the affected surface of the tooth, or 'infiltrating' the demineralised tissue with resins), through to step‐wise caries removal and restoration for extensive lesions.

With advances in technology over the last two decades, alternative methods of detection have become available, such as digital radiography, fluorescence, transillumination, and electrical conductance‐based devices. The adjunctive use of such methods could support the detection and diagnosis of caries at an early stage of decay. In turn, early detection could lead to less invasive treatment with preservation of more tooth tissue, and potentially result in reduced cost of care to the patient and to healthcare services.

Target condition being diagnosed

The term dental caries is used to describe the mechanism which can ultimately lead to the breakdown of the tooth surface which results from an imbalance in the activity within the biofilm (or dental plaque) on the surface of the tooth within the oral cavity (Kidd 2016). This imbalance is due to bacterial breakdown of sugars in the diet which leads to the production of acid and demineralisation of the tooth. Disease progression can be moderated by improved oral hygiene practices together with the influx of fluoride through toothpaste and other available fluoride sources. However, the levels of sugar consumption observed in many populations will often outweigh the benefits of fluoride (Hse 2015). Ultimately, carious lesions may develop and destroy the structure of the tooth.

The most common surfaces for caries to manifest are on the biting (occlusal) surface or the tooth surface which faces an adjacent tooth (approximal surfaces); although smooth surfaces on the flat exterior of teeth adjacent to the tongue, cheeks, and lips can be affected. The severity of disease is defined by the depth of demineralisation of the tooth's structure and whether the lesion is active or arrested. Caries lesions confined to tooth enamel have the potential to be stabilised or even reversed. The progression of carious lesions into the deeper aspects of dentine and pulp of the tooth may be arrested, for example using silver diamine fluoride (Urquhart 2019), but will often require restorative treatment (Bakhshandeh 2018; Kidd 2004).

Variants of the DMFT (decayed, missing, and filled teeth) scale have typically been used in the assessment of disease severity in epidemiological and research studies. Within the D (decayed) component there are four clinically detectable thresholds applied as indicators for diagnosis and treatment planning, often labelled as D₁, D₂, D₃, and D₄ (Anaise 1984) (Additional Table 1). Typically the D₃ threshold, with only lesions extending into dentine classed as carious, has been used to determine the presence of caries (Pitts 1988; Shoaib 2009). These four categories have formed the basis for expanded indices based on visual characteristics such as the International Caries Detection and Assessment System (ICDAS) (Ekstrand 2007; Ismail 2007).

Table 1. Classification of levels of caries levels

DMFT classification	Definition (Pitts 2001 )
0	Sound (non‐diseased)
D₁	Non‐cavitated yet clinically detectable enamel lesions with intact surfaces
D₂	Cavitated lesion penetrating the enamel or shadowing
D₃	Cavity progressing past the enamel‐dentine junction into dentine
D₄	Cavity progressing into pulp

DMFT = decayed, missing, and filled teeth.

Treatment of caries

There are many varied treatment options available to the dental clinician, dependent on the thresholds of observed disease. Initial caries can be addressed without surgical intervention using preventive and remineralising approaches such as plaque control, dietary advice, and application of fluoride (Kidd 2016). Minimally invasive treatments for initial caries are available, such as sealing the affected surface of the tooth, or 'infiltrating' the pores of the demineralised tissue with resins. Patients with severe caries may require step‐wise or selective caries removal and restoration of advanced lesions.

A caries management pathway, informed by diagnostic information, can be beneficial in guiding the clinician towards prevention or a treatment plan. One recently developed care pathway is the International Caries Classification and Management System (ICCMS) (Ismail 2015). The system presents three forms of management in the care pathway:

when dentition is sound the clinician proceeds with preventive strategies to prevent sound surfaces from developing caries;
non‐invasive treatment of the lesion to arrest the decay process and encourage remineralisation, preventing initial lesions from progressing to cavitated decay; and
management of more severe caries through excavation and restoration or potentially extraction.

At the core of this care pathway is the ability to accurately detect early caries and optimise preventative strategies through excavation methods that preserve tooth tissue, and restoration or potential extraction in more severe cases. The detection and diagnosis of early caries remains challenging, and the likelihood of undiagnosed early disease is high (Ekstrand 1997). In such instances, the opportunity for preventing initial lesions from progressing to cavitated decay, or even reversing the disease process, is missed, and disease progresses to cavitated decay where restoration is required (Ekstrand 1998).

Index test(s)

The cornerstone of caries detection is a visual and tactile dental examination, and the ability of clinicians to accurately detect disease in this way has been researched for over half a century (Backer Dirks 1951). Tests may be suitable at different stages of the care pathway (Bloemendal 2004; Fyffe 2000), and the use of additional detection tools can support the detection, diagnosis, and monitoring process. The generic protocol (Macey 2018) provides information regarding alternative index methods of caries detection and diagnosis in this suite of Cochrane Reviews.

The use of dental imaging is a standard component of the clinical examination. The bitewing radiograph is a radiograph of the crowns of posterior teeth and their immediate supporting tissues. It is one of the most common dental diagnostic tools and is used to aid the diagnosis of caries or periodontal diseases and also to assess the integrity of existing restorations. Internationally, there is variation in the recall duration, but guidelines from the Faculty of General Dental Practice (UK), for example, suggest that bitewing radiographs are taken six monthly for people at high caries risk, annually for people at moderate caries risk, and at 12‐ to 18‐month intervals in children at low risk and 24‐month intervals for adults (Goodwin 2017).

The two most significant advances in this area within the last 20 years are the replacement of analogue imaging with digital imaging, and the introduction of cone beam computed tomography (CBCT), the latter providing the potential for three‐dimensional (3D) imaging. This review focuses on radiographic caries detection, and incorporates various adjuncts to the typical visual‐tactile examination including analogue and digital radiographs. Due to the high radiation dose, the utility of CBCT may rest in the use of 3D technology as a potential reference standard for in vivo studies.

Digital imaging can be direct or indirect. Direct digital imaging uses charge‐coupled devices to produce an instant chairside radiographic image. Indirect digital imaging using storage phosphor plates is less efficient and requires the additional step of a laser system to transform the X‐ray exposed phosphor into a visible image. However, despite the longer time taken to produce an image, indirect digital imaging has other advantages: it is easier to use, it has lower reject rates, and imaging plates are cheaper to replace than the direct imaging sensors. In addition, indirect digital imaging is more comfortable for patients and consequently, it is the first choice in the dental hospital for general use. Both direct and indirect imaging systems bring benefits over conventional analogue imaging and have replaced the need for the acquisition, storage, and disposal of hazardous chemicals, a dark room, the cost and storage of unexposed film, and the physical filing and security of patient hard‐copy films. Exposure factors can be set to a lower dose, as the post‐processing digital functionality allows for change in contrast, lightening and darkening, size of image, and easy magnification. These features all compensate for a reduction in spatial resolution, in comparison to film. Images can be stored in an electronic patient record, annotated and shared between clinicians. Progression of lesions, including early caries, can be evaluated on dual monitors, or split screens.

CBCT is an inherently digital system, that was designed as a lower dose alternative to the cross‐sectional imaging provided by conventional computed tomography (CT). It does not demonstrate soft tissue detail, but the tissue contrast of air, bone, and dental tissues make it the preferred modality for pre‐implant planning. In unrestored teeth, caries is believed to be well demonstrated, and due to the relief of superimposition provided by cross‐sectional imaging, more accurately shown than in film‐based (analogue) or digital imaging. However, the most significant artefact in CT or CBCT is that of beam hardening created by high attenuation objects such as dental restorations. Opaque streaks mask the adjacent tissues. Thus, CBCT is not recommended to be used to detect recurrent caries. However, the use of CT/CBCT is expensive and carries a higher radiation dose than that of conventional imaging, and is typically reserved for cases where significant clinical uncertainty remains following standard investigations. Despite the clinical benefits, the societal efficacy of CBCT is low and it is therefore not typically recommended as a modality for the routine detection of caries in specialist guideline documents.

Clinical pathway

The process proceeding from a dental patient attending for a routine examination and a caries assessment being undertaken potentially has four intertwined stages: screening, detection (case‐finding), diagnosis, and treatment planning. If the presenting patient is at some risk of disease but seemingly asymptomatic then this can be considered as a screening exercise (Wilson 1968) to detect initial caries in individuals who do not yet have symptoms. Since caries is a dynamic process the pure detection of the disease at a single time point is insufficient to inform the future care of the patient, and additionally the depth and severity of demineralisation, allied to a decision on the caries activity levels, are combined to reach a diagnosis (Ismail 2004; Nyvad 1997). This diagnosis then feeds into a caries management pathway once the patient's history, personal oral care, and risk factors have been considered. A comprehensive methodology has been developed, the International Caries Classification and Management System (ICCMS™), that "helps practitioners to intuitively and systematically collect and analyze personal and clinical data to develop comprehensive patient care plans" that go beyond restorative care (Ismail 2015). A version for use in clinical practice, CariesCare, has been developed by an international team (Martignon 2019).

Figure 1 presents the key elements of the ICCMS process and these reviews could inform the process at 'Keystone 3' where diagnosis is an indefinable component.

Figure 1

Keystones of the International Caries Classification and Management System (ICCMS™).
Copyright© 2018 Ismail AI, Pitts NB, Tellez M. The International Caries Classification and Management System (ICCMS™) an example of a caries management pathway. BMC Oral Health 2015;15(Suppl 1):S9. Reproduced with permission.

Role of index test(s)

In clinical practice, a typical visual or visual‐tactile oral examination would always be undertaken as part of the clinical examination. Supplementing the visual‐tactile examination with radiographic methods (an 'add‐on') at appropriate time points can support the practitioner to resolve uncertainties regarding the presence or absence of early decay, and also to communicate the results of the clinical assessment to the patient. The information from caries detection (including assessment of severity of disease) will be an integral part of diagnosis, which additionally incorporates patient history, risk factors, and treatment planning protocols.

Alternative test(s)

Alternative tests include.

Fluorescence: the breakdown of enamel alters the characteristics of its structure, when exposed to light‐inducing fluorescence diseased teeth respond differently to sound teeth. There is potential for mineral loss to be quantified and used to aid the diagnostic decision and treatment pathway (Angmar‐Månsson 2001; Matos 2011). Fluorescence is typically divided into laser fluorescence and light fluorescence (i.e. DIAGNOdent and quantitative light‐induced fluorescence (QLF) type devices) (Macey 2020).
Visual or visual‐tactile examination: identifying caries according to visual appearance, aided by a dental mirror and probe, on clean and dry teeth.
Fibre‐optic transillumination (FOTI): uses a light emitted from a handheld device which when placed directly onto the tooth illuminates the tooth (Pretty 2006). Any demineralisation should appear as shadows in the tooth due to the disruption of the tooth's structure (Macey 2021).
Electrical conductance: the demineralisation of the tooth is reported to have an effect on the tooth's electrical conductance. This is measured by placing a probe on the tooth which measures any potentially higher conductivity which occurs due to carious lesions being filled with saliva (Tam 2001).

Rationale

Despite technological advances, the usual method of caries detection is currently based upon information from visual‐tactile clinical examination, with or without radiographs. Bader 2002 completed an extensive review of in vitro studies investigating the accuracy of visual examination, radiographs, fibre‐optic, electrical conductance, and fluorescence in the primary and permanent dentition. This review was limited to studies with a histological reference standard and grouped studies according to index test, disease threshold (enamel or dentinal lesions), and tooth surfaces (occlusal or proximal); a meta‐analysis was not undertaken and the authors graded the quality of the available evidence as low (Bader 2002). This review pre‐dates the development of meta‐analysis methods for DTA reviews recommended in the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy (Deeks 2013). More recently systematic reviews of caries detection have been completed which explore the use of fluorescence (Gimenez 2013), visual inspection (Gimenez 2015), and radiographs (Schwendicke 2015).

In this Cochrane Review we have included contemporary studies irrespective of publication language and status, and built upon existing research to incorporate methodological developments by: expanding the search strategy to capture all relevant evidence, applying appropriate hierarchical analysis (Dinnes 2016), and assessing the body of evidence using GRADE (Schünemann 2020; Schünemann 2020a) to facilitate the production of 'Summary of findings' tables.

Objectives

To establish the accuracy of imaging modalities to detect and inform the diagnosis of non‐cavitated enamel only coronal dental caries in children and adults. The specific research questions addressed in this Cochrane Review were:

what is the diagnostic test accuracy of different methods for:
- different purposes (case finding or detection and diagnosis in a clinical setting);
- in different populations (children: primary/mixed dentition, adolescents: immature permanent dentition, or adults: mature permanent dentition);
- when a comparison is made between different methods.

Secondary objectives

Where sufficient studies were available we investigated the following areas of potential heterogeneity:

the use of different reference standards;
tooth surface (occlusal, proximal, or smooth surface);
prevalence of dentinal caries;
participants or teeth with previously applied restorations (secondary caries) and pit and fissure sealants.

Methods

Criteria for considering studies for this review

Types of studies

We included diagnostic test accuracy (DTA) designs that were:

studies with a single set of inclusion criteria that compared a diagnostic test with a reference standard. We included prospective studies that evaluated the diagnostic accuracy of single index tests, and studies that directly compared two or more index tests;
randomised controlled trials (RCTs) of the diagnostic test accuracy of one or more index tests in comparison, or versus a no test option;
'case‐control' type accuracy studies where different sets of criteria were used to recruit those with or without the target condition, although prone to bias some novel systems may be identifiable through this design alone;
reporting at either the patient, tooth, or tooth surface level, however only those reporting at the tooth surface level were included in the primary analysis;
in vivo or in vitro studies.

In vitro studies are those in which teeth have been extracted prior to the initiation of the study, and assessed by the index test and reference standard (usually histology). This scenario is not representative of the typical clinical setting. In vivo studies recruit participants and carry out the index tests and reference standard on teeth in the oral cavity, usually without extraction of the teeth. A histology reference standard would not usually be employed in an vivo study with the exception of teeth indicated for orthodontic, periodontal, or third molar extraction, or for primary teeth close to exfoliation.

Studies were ineligible for inclusion where:

artificially created carious lesions were used in the assessment;
studies used and reported the results of two or more tests in combination; or
an index test was used during the excavation of dental caries to ascertain the optimum depth of excavation.

Participants

Participants asymptomatic for dental caries who may have early, undetected caries at the point of recruitment. Studies that explicitly recruited participants with more advanced lesions that were obviously into dentine or frankly cavitated were excluded. Studies that recruited participants referred to secondary care for restorative treatment were also excluded as there is a likelihood that more advanced caries (into dentine or pulp) would be present and readily detectable without the need for the index tests investigated in this review.

Studies recruiting children, adolescents, and adults were all eligible for inclusion, as this allowed for the analysis of the diagnostic test accuracy of index tests for primary, mixed, and permanent dentition.

Index tests

We included intra and extraoral radiographic caries detection methods completed on intact teeth including:

analogue (conventional) radiographs;
digital radiographs; and
cone beam computed tomography (CBCT).

These index tests could be used as a new adjunct to the typical clinical examination or as a replacement for aspects of the current standard examination (e.g. digital radiography to replace analogue radiograph).

Where studies used multiple examiners the most appropriate examiner to the research question was selected. For example, if the study used dental students, general dental practitioners, and restorative or radiography consultants, then the results of the general dental practitioner were extracted. In the scenario where multiple examiners were stated to have similar skills and experience, then the mean sensitivity and specificity values were extracted if available, otherwise the first set of reported results was extracted.

Target conditions

Coronal caries: initial stage decay, defined as initial or incipient caries or non‐cavitated enamel lesions. Specifically where there is a detectable change in enamel evident which is not thought to have progressed into dentine, on occlusal, approximal surfaces, or smooth surfaces.

Reference standards

A number of different reference standards have been used in primary DTA studies. The only way of achieving a true classification of caries presence and depth is to extract and section the tooth and then perform a histological assessment (Downer 1975; Kidd 2004). This approach is commonly undertaken on previously extracted teeth for in vitro studies but unethical for a healthy population in clinical (in vivo) studies. The only scenario where histology could be appropriate for studies undertaken in a primary or secondary care dental setting would be where a tooth has been identified as requiring extraction (ideally for a non‐caries related reason, such as orthodontic extraction, or third molar extraction), the index test could be applied prior to extraction, and followed by a histological reference standard.

The optimum reference standard was histology. Alternative, acceptable reference standards for this review included operative exploration, for patients with suspected early caries lesions, where a clinician removes caries with a dental burr (drill) in preparation for a restoration and reports the depth of decay for in vivo studies and enhanced visual examination (using a validated classification system) for patients where no initial decay is observed with the index test. Operative exploration was considered to be an acceptable reference standard on the understanding that in vivo studies require a separate reference standard for tooth sites not requiring treatment and verification bias would therefore be introduced. Enhanced visual examination with tooth separation was also considered an acceptable reference standard for approximal surfaces, but is the least preferred of all the options.

Studies that employed a composite reference standard based on the results of information from multiple sources were eligible for inclusion.

Search methods for identification of studies

Electronic searches

Cochrane Oral Health's Information Specialist conducted systematic searches in the following databases without language or publication status restrictions:

MEDLINE Ovid (1946 to 31 December 2018) (Appendix 1);
Embase Ovid (1980 to 31 December 2018) (Appendix 2).

Searching other resources

The following trial registries were searched for ongoing studies:

US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (clinicaltrials.gov; searched 31 December 2018) (Appendix 3);
World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch; searched 31 December 2018) (Appendix 4).

We searched the reference lists of included papers and previously published systematic reviews for additional publications not identified in the electronic searches.

Data collection and analysis

Selection of studies

Two review authors independently screened and assessed the results of all searches for inclusion. Any disagreements were resolved through discussion and, where necessary, consultation with another clinical or methodological member of the author team. Studies were excluded if they failed to present the data in the format of a 2 x 2 contingency table, or if they failed to report prevalence at the enamel threshold to enable a 2 x 2 table to be constructed. An adapted PRISMA flowchart was used to report the study selection process (McInnes 2018). Once agreement for inclusion was reached, the studies were categorised according to their index test, the tooth surface, and the dentition of the participants.

Data extraction and management

Two review authors extracted data independently and in duplicate using a piloted data extraction form based on the review inclusion criteria. Disagreements were resolved through discussion by the review team. Where data had been reported for multiple tooth surfaces or dentitions, data were extracted separately for each. We contacted study authors to obtain missing data or characteristics which were not evident in the published paper.

We recorded the following data for each study:

sample characteristics (age, sex, socioeconomic status, risk factors where stated, number of patients/carious lesions, lesion location);
setting (country, disease prevalence, type of facility);
the type of index test(s) used (category, name, conditions (i.e. clean/dried teeth), positivity threshold);
study information (design, reference standard, case definition, training and calibration of personnel);
study results (true positive, true negative, false positive, false negative, any equivocal results, withdrawal).

Assessment of methodological quality

We used the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS‐2) to assess the risk of bias and applicability of the primary diagnostic studies over the four domains of participant selection, index test, reference standard, and flow and timing (Whiting 2011). 'Review specific' descriptions of how the QUADAS‐2 items were contextualised and implemented are detailed in the accompanying checklist (Additional Table 2).

Table 2. QUADAS‐2 tool

Item	Response (delete as required)
Participant selection – Risk of bias
1) Was a consecutive or random sample of participants or teeth used?	Yes – where teeth or participants were selected consecutively or allocated to the study via a randomisation process No – if study described another method of sampling Unclear – if participant sampling is not described
2) Was a case‐control design avoided?	Yes – if case‐control clearly not used No – if study described as case‐control or describes sampling specific numbers of participants with particular diagnoses Unclear – if not clearly described
3) Did the study avoid inappropriate exclusions (e.g. inclusion of caries into dentine)?	Yes – if the study clearly reports that included participants or teeth were apparently healthy or caries into dentine were excluded No – if lesions were included that showed caries into dentine or exclusions that might affect test accuracy (e.g. teeth with no caries) Unclear – if not clearly reported
Could the selection of participants have introduced bias?
If answers to all of questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) and 3) was 'no'	Risk is High
If answers to any of questions 1) and 2) and 3) was 'unclear'	Risk is Unclear
Participant selection – Concerns regarding applicability
1) Does the study report results for participants or teeth selected by apparent health or suspected early caries (i.e. studies do not recruit patients who are known to have advanced caries into dentine)?	Yes – if a group of participants or teeth has been included which is apparently healthy or indicative of early caries No – if a group of participants or teeth has been included which is suspected of advanced caries Unclear – if insufficient details are provided to determine the spectrum of participants or teeth
2) Did the study report data on a per‐patient rather than on a tooth or surface basis?	Yes – if the analysis was reported on a surface or tooth basis No – if the analysis was reported on a per‐patient basis Unclear ‐ if it is not possible to assess whether data are presented on a per‐patient or per‐tooth basis
3) Did the study avoid an in vitro setting which required the usage of extracted teeth?	Yes – if the participants were recruited prior to tooth extraction No – if previously extracted teeth were used in the analysis Unclear – if it was not possible to assess the source and method of recruiting of included participants/teeth
Is there concern that the included participants or teeth do not match the review question?
If answers to all of questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) and 3) was 'no'	Risk is High
If answers to any of questions 1) and 2) and 3) was 'unclear'	Risk is Unclear
Index test ‐ Risk of bias (to be completed per test evaluated)
1) Was the index test result interpreted without knowledge of the results of the reference standard?	Yes – if the index test described is always conducted and interpreted prior to the reference standard result, or for retrospective studies interpreted without prior knowledge of the reference standard No – if index test described as interpreted in knowledge of reference standard result Unclear – if index test blinding is not described
2) Was the diagnostic threshold at which the test was considered positive pre‐specified?	Yes – if threshold was pre‐specified (i.e. prior to analysing the study results) No – if threshold was not pre‐specified Unclear – if not possible to tell whether or not diagnostic threshold was pre‐specified
For visual and radiograph tests only: 3) For studies reporting the accuracy of multiple diagnostic thresholds for the same index test or multiple index tests, was each threshold or index test interpreted without knowledge of the results of the others?	Yes – if thresholds or index tests were selected prospectively and each was interpreted by a different clinician or interpreter, or if study implements a retrospective (or no) cut‐off (i.e. look for deepest/most severe lesion first) No – if study states reported by same reader Unclear ‐ if no mention of number of readers for each threshold or if pre‐specification of threshold not reported N/A ‐ multiple diagnostic thresholds not reported for the same index test
Could the conduct or interpretation of the index test have introduced bias?
For visual and radiographic studies item 3) to be added
If answers to all of questions 1) and 2) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) was 'no'	Risk is High
If answers to any of questions 1) and 2) was 'unclear'	Risk is Unclear
Index test ‐ Concerns regarding applicability
1) Were thresholds or criteria for diagnosis reported in sufficient detail to allow replication?	Yes – if the criteria for detection or diagnosis of the target disorder were reported in sufficient detail to allow replication No – if the criteria for detection or diagnosis of the target disorder were not reported in sufficient detail to allow replication Unclear ‐ if some but not sufficient information on criteria for diagnosis to allow replication were provided
2) Was the test interpretation carried out by an experienced examiner?	Yes – if the test clearly reported that the test was interpreted by an experienced examiner No – if the test was not interpreted by an experienced examiner Unclear – if the experience of the examiner(s) was not reported in sufficient detail to judge or if examiners described as 'Expert' with no further detail given
Is there concern that the included participants do not match the review question?
If the answer to question 1) and 2) was 'yes'	Concern is Low
If the answer to question 1) and 2) was 'no'	Concern is High
If the answer to question 1) and 2) was 'unclear'	Concern is Unclear
Reference standard ‐ Risk of bias
1) Is the reference standard likely to correctly classify the target condition?	Yes – if all teeth or surfaces underwent a histological or excavation reference standard No – if a final diagnosis for any participant or tooth was reached without the histological or excavation reference standards Unclear – if the method of final diagnosis was not reported
2) Were the reference standard results interpreted without knowledge of the results of the index test?	Yes – if the reference standard examiner was described as blinded to the index test result No – if the reference standard examiner was described as having knowledge of the index test result Unclear – if blinded reference standard interpretation was not clearly reported
Could the reference standard, its conduct, or its interpretation have introduced bias?
If answers to questions 1) and 2) was 'yes'	Risk is Low
If the answer to question 1) and 2) was 'no'	Concern is High
If the answer to question 1) and 2) was 'unclear'	Concern is Unclear
Reference standard ‐ Concerns regarding applicability
1) Does the study use the same definition of disease positive as the prescribed in the review question?	Yes ‐ same definition of disease positive used, or teeth can be disaggregated and regrouped according to review definition No ‐ some teeth cannot be disaggregated Unclear ‐ definition of disease positive not clearly reported
Flow and timing ‐ Risk of bias
1) Was there an appropriate interval between index test and reference standard (in vivo studies less than 3 months, in vitro no limit but must be stored appropriately)?	Yes ‐ if study reports index and reference standard had a suitable interval or storage method No ‐ if study reports greater than 3‐month interval between index and reference standard or inappropriate storage of extracted teeth prior to reference standard Unclear ‐ if study does not report interval or storage methods between index and histological reference standard
2) Did all participants receive the same reference standard?	Yes ‐ if all participants underwent the same reference standard No ‐ if more than 1 reference standard was used Unclear ‐ if not clearly reported
3) Were all participants included in the analysis?	Yes ‐ if all participants were included in the analysis No ‐ if some participants were excluded from the analysis Unclear ‐ if not clearly reported
If answers to questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any one of questions 1) or 2) or 3) was 'no'	Risk is High
If answers to any one of questions 1) or 2) or 3) was 'unclear'	Risk is Unclear

N/A = not applicable; QUADAS‐2 = Quality Assessment of Diagnostic Accuracy Studies 2.

A 'Risk of bias' judgement ('high', 'low', or 'unclear') was made for each domain. Generally, where the answers to all signalling questions within a domain were judged as 'yes' (indicating low risk of bias for each question) then the domain was judged to be at low risk of bias. If any signalling question was judged as 'no', indicating a high risk of bias, the domain was scored as high risk of bias. This was followed by a judgement about concerns regarding applicability for the participant selection, index test, and reference standard domains. Results of the assessment of methodological quality are presented graphically.

Participant selection domain (1)

This domain refers to the selection of observations within the primary studies, either patient or tooth sample selection. The selection of patients has a fundamental effect on the estimated accuracy of an index test. The disease stages of sound and enamel should be represented in the sample and children, adolescents, and adults should be represented in the included studies to allow a complete appraisal of a test's potential to correctly classify disease in different populations.

It was acceptable for studies to focus on early enamel lesions for a specific surface (occlusal, approximal, or smooth) or dentition (primary or mixed, immature permanent, permanent). For a low risk of bias judgement, the inclusion of study participants or teeth meeting the eligibility criteria should be consecutive or random as inappropriate exclusions may lead to an over‐ or under‐estimation of the test's ability to detect disease. Additionally, the prevalence and severity of disease reported was used to inform the applicability of this test to a wider population.

Study results should be reported at a tooth or surface level, as opposed to patient level, due to the potential for the index test and reference standard to be reporting on different sites within the same mouth.

Index test domain (2)

The nature of the index tests and the visual presentation of the disease means that it should be feasible to ensure that the index test is conducted prior to the reference standard. The index test should be taken before the extraction of a tooth for any histological analysis or before in situ excavation of a tooth is undertaken. To minimise potential for bias, it is preferable for separate examiners to carry out the index test and reference standard. Where the index test is performed on previously extracted teeth, there should have been efforts made to replicate the setting of the oral cavity. This would often be achieved by placing teeth in a model which replicates the jaw and using either a water bath or plastic/resin block to replicate the effect of the cheek. The threshold of disease positive and negative should have been determined prior to analysis and be reflective of the participants recruited to the study.

Given the subjective nature of the interpretation of test results where thresholds of both initial and more severe assessments of disease are considered within a primary study, there may be potential for information bias. For example, if the assessor judgement is uncertain between caries into enamel or caries into dentine, the interpretation of the first threshold would influence the decision made on the second threshold. This can be mitigated by the use of different examiners to undertake separate assessments for each of the different thresholds.

Reference standard domain (3)

The reference standard assessment should be carried out by a different examiner to the index test, as the subjective interpretation of the reference standard could be compromised by knowledge of the index test results. However, where a tooth had been extracted, sectioned and prepared for histological evaluation it is extremely unlikely that an examiner would be able to recall the specific tooth or participant and the results from the index test results.

Time delays between index test and reference standard should be under three months for in vivo studies.

Ideally, each participating tooth or patient within a study should receive the same reference test. This is possible in the in vitro setting as each selected tooth can have a histological assessment applied. In vitro studies may have applied the same reference standard to all participants. If a study allocated participants or specific teeth to different reference standards then reasons for this allocation should have been clearly reported. All reference standards should have been completed without knowledge of the index test results.

Flow and timing domain (4)

The index test should have been conducted prior to the reference standard. If the reference standard used was tooth separation, radiograph, or excavation then there should be less than three months between index test and reference standard. Caries is a slow growing disease so minimal changes should be experienced within this time frame. All included teeth in the sample should receive both an index test and reference standard. Where studies report some teeth having an index test but not a reference standard, a reason should be clearly reported, such as teeth being broken during sectioning, this would not influence the risk of bias decision.

Statistical analysis and data synthesis

The threshold of interest was between sound teeth and early enamel caries. Estimates of diagnostic accuracy were expressed as sensitivity and specificity with 95% confidence intervals (CI) for each study and for each available data point if there were multiple index tests, dentition, or surfaces reported within a single study. When there were two or more test results reported in the same study, we included them as separate datasets. For both the overall analysis of all included datasets and the analysis of different imaging modalities we also indicated the 95% prediction regions as an indication of the region in which the sensitivity and specificity of a future study could be expected to lie given the results of the studies that have already been observed and included in the analysis.

Hierarchical models were used for data synthesis. The data were extracted for the target condition of early caries (caries into enamel) at the tooth site or surface level. This target condition has been consistently used across the suite of caries detection reviews. Study estimates of sensitivity and specificity were plotted on coupled forest plots and in receiver operating characteristic (ROC) space. Meta‐analysis was conducted which combined the results of studies for each index test using a bivariate model to estimate the summary values of sensitivity and specificity at a common threshold (Chu 2006; Reitsma 2005). Data were input to Review Manager 5 (Review Manager 2020) and displayed in coupled forest plots. Analysis was conducted using xtmelogit and the METANDI package in Stata (Harbord 2009; Stata 14; Takwoingi 2016), and the MetaDTA interactive web‐based tool (Freeman 2019). We used meta‐regression with xtmelogit to compare the accuracy of different imaging methods, study types, and dentitions separately. We added the different imaging methods, study designs (in vivo, in vitro) and dentition as covariates to the bivariate model, assuming equal variances, and used a likelihood ratio test to formally assess the significance of any model comparisons (Macaskill 2010; Takwoingi 2016). Initially we allowed the covariate effects to be assessed on both sensitivity and specificity. If a difference in sensitivity and or specificity was observed then further investigations were undertaken to determine whether the differences could be attributed to sensitivity or specificity, initially by assuming the same sensitivity but allowing specificity to vary and then assuming the same specificity but allowing sensitivity to vary (Takwoingi 2016). Where sufficient studies allowed we intended to explore the assumption of equal variances by performing a likelihood ratio test comparing the model that included the sensitivity and specificity covariates and assumed equal variances for each covariate with the model that included the sensitivity and specificity covariates but allowed for separate variances for the sensitivity and specificity of each covariate. Due to the large number of parameters to be estimated we had problems fitting these models, and consequently reverted to the assumption of equality of variances for the analyses.

Investigations of heterogeneity

The investigation of each potential source of heterogeneity was considered individually. Initially, a visual inspection of the clinical and methodological characteristics of the included studies, coupled forest plots, and summary ROC plots were used to form the basis of the assessment of heterogeneity. Where sufficient numbers of studies allowed, meta‐regression analyses were carried out to explore possible sources of heterogeneity. Formal model comparisons were undertaken as indicated in Statistical analysis and data synthesis.

The sources of heterogeneity (specified a priori) were the different reference standards used, tooth surfaces, prevalence of caries into dentine (classed as low (0% to 14%), medium (15% to 34%), and high prevalence (≥ 35%), as per the other reviews in this series), and studies including previously applied restorations (secondary caries) or pit and fissure sealants. Each potential source of heterogeneity was investigated separately.

Sensitivity analyses

Where a sufficient number of studies investigated the same index test, the following sensitivity analyses were performed to assess the impact on summary estimates of restricting the analyses according to the following criteria:

low prevalence of dentine caries (i.e. less than 15%);
low risk of bias for an index test;
low risk of bias for a reference standard.

Assessment of reporting bias

Methods currently available to assess reporting or publication bias for diagnostic studies may lead to uncertainty and misleading results from funnel plots (Deeks 2005; Leeflang 2008), therefore we did not perform reporting bias tests in the reviews.

Summary of findings and assessment of the certainty of the evidence

We reported our results for our primary objective following GRADE methods (Schünemann 2020; Schünemann 2020a), and using the GRADEPro online tool (www.guidelinedevelopment.org). To enhance readability and understanding, we presented test accuracy results in natural frequencies to indicate numbers of false positives and false negatives. The certainty of the body of evidence was assessed with reference to the overall risk of bias of the included studies, the indirectness of the evidence, the inconsistency of the results, the imprecision of the estimates, and the risk of publication bias. We categorised the certainty of the body of evidence, as high, moderate, low, or very low.

Results

Results of the search

The search identified 5933 results. After an initial screen of titles and abstracts 123 studies were potentially eligible for inclusion. Upon closer inspection of the full papers this number reduced to the 77 studies which are included in this review (Figure 2). An inability to create a 2 x 2 table of the results and studies that stated their intention to include more advanced lesions that were obviously into dentine or frankly cavitated were the main reason for exclusion of studies. Studies and their reasons for exclusion are detailed in the Characteristics of excluded studies table.

Figure 2

Study flow diagram.

We included 77 unique studies providing 104 datasets in the review. Two studies reported results for both the primary and permanent dentition (Ekstrand 2011; Souza 2014), three studies reported approximal and occlusal surfaces (Ariji 1998; Hintze 2003; Simon 2016). The results of multiple index tests were reported for 18 studies: 15 studies (Abesi 2012; Ariji 1998; Ashley 1998; Astvaldsdottir 2012; Costa 2002; Da Silva 2010; Erten 2005; Firestone 1998; Kalathingal 2007; Mitropoulos 2010; Pontual 2010; Ramezani 2016; Rathore 2012; Rockenbach 2008; Svanaes 2000) reported on two index tests, and three studies (Safi 2015; Senel 2010; Tarim 2014) reported on three index tests. Taking into account the multiplicity of dentitions, surfaces, and index tests within each study provided 104 datasets for the analysis, which reported a total of 15,518 tooth sites or surfaces. All included studies were published between 1986 and 2019 inclusive, with 44 (57%) studies conducted from 2010 onwards. Most studies were conducted in Brazil (35%), Germany (8%), Iran (8%), Denmark (6%), Switzerland (6%), Turkey (8%), and the US (6%). 17 (22%) studies were carried out in a clinical setting with an in vivo study design, 60 (78%) were classed as in vitro studies using extracted teeth.

A reference standard of histology was used in 64 studies (83%), excavation in 5 studies (6%), and an enhanced visual reference standard using separation with bands was used in 8 (10%) studies.

Most included studies assessed analogue radiography (51 studies and 55 datasets) followed by digital radiography (40 studies and 42 datasets), and cone beam computed tomography (CBCT) (7 studies and 7 datasets). 15 studies directly compared two tests of which 12 studies directly compared analogue and digital (Abesi 2012; Ariji 1998; Ashley 1998; Astvaldsdottir 2012; Costa 2002; Da Silva 2010; Erten 2005; Firestone 1998; Mitropoulos 2010; Pontual 2010; Rockenbach 2008; Svanaes 2000), and three studies compared digital with CBCT (Kalathingal 2007; Ramezani 2016; Rathore 2012). Three studies directly compared three tests, analogue, digital, and CBCT (Safi 2015; Senel 2010; Tarim 2014). 2 studies (4 datasets) reported on both the primary and permanent dentition, 55 studies (78 datasets) on the permanent dentition only, and 20 studies (22 datasets) on the primary dentition only. Most studies (42 studies and 57 datasets) reported on approximal surfaces only, with 31 studies (39 datasets) reporting results for the occlusal surfaces only. 3 studies (8 datasets) reported results for both the approximal and occlusal surfaces. We categorised the prevalence of caries into dentine, D₃ level, as high (> 35%) in 30 datasets (29%), medium (15% to 34%) in 40 datasets, and low (< 15%) in 22 studies. Insufficient information was available to make a judgement about the prevalence of D₃ caries in 12 datasets.

Our search identified some studies that had evaluated the use of dental imaging for the evaluation of secondary caries, but these were considered ineligible due to the inability to construct 2 x 2 tables for data at the enamel level, or tooth sites with artificially created lesions. No study intentionally included restored teeth in its sample or reported the inclusion of sealants that would allow us to evaluate early caries lesions.

Methodological quality of included studies

The methodological quality of the 77 included studies is summarised across the QUADAS‐2 domains in Figure 3 and the individual study results are shown in Figure 4. No studies were considered to be at low risk of bias across all four domains but 16 studies (21%) were judged to have low concern for applicability across all domains (Behere 2011; Cinar 2013; Freitas 2016; Hintze 2003; Jablonski‐Momeni 2017; Kockanat 2017; Kucukyilmaz 2015; Matos 2011; Mialhe 2003; Mortensen 2018; NCT02657538; Novaes 2009; Novaes 2010; Rocha 2003; Shimada 2014; Souza 2014). The patient selection domain had the highest proportion of high risk of bias studies (43 studies, 56%), we judged the index test, reference standard, and flow and timing domains to have a lower proportion of high risk of bias studies (30, 12, and 7 studies respectively). Applicability of patients and setting was of high concern in 58 (75%) studies, the conduct or interpretation of the index test was of high concern in 22 (29%) studies. There was low concern that the target condition as defined by the reference standard did not match the review question in all 77 studies.

Figure 3

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study.

Figure 4

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies.

Seven studies (9%) reported the use of random or consecutive sampling (Jablonski‐Momeni 2017; Matos 2011; Novaes 2009; Novaes 2010; Novaes 2012; Simon 2016; Souza 2018), 44 studies did not, instead choosing to select teeth, and in 26 studies it was not possible to determine how the sample of participants was obtained and an unclear judgement was made. All but one study (Senel 2010) avoided a case‐control study design. 44 studies (57%) avoided inappropriate exclusions, 32 (42%) failed to provide sufficient detail in the description of the sample population to satisfy us that exclusions were appropriate. Applicability of the sample to the population was of high concern for studies which investigated extracted teeth. As the objective of this review was to inform general clinical practice there may be concerns that results from laboratory‐based studies may have limited relevance to a clinical setting, and that extrapolation of results is inappropriate.

Index tests were interpreted without the knowledge of the reference standard in 72 (94%) studies, and with the knowledge of the reference standard in four studies due to: the same assessor applying/conducting the index test and reference standard (Rathore 2012), the index test being interpreted with knowledge of the reference standard (Lazarchik 1995), the index test preceding the reference standard (Matos 2011; Mortensen 2018). A pre‐specified threshold for interpretation of the index test was used in 74 (96%) studies. 22 (29%) studies were found to be of high concern regarding the applicability of the index test, this is due to examiners.

Of the 77 included studies we judged that the choice, conduct, or interpretation of the reference standard in 56 (73%) studies to be at low risk of bias, and 12 (16%) to be at high risk (Braun 2017; Espelid 1986; Jablonski‐Momeni 2012; Matos 2011; Mialhe 2003; Mortensen 2018; NCT02657538; Novaes 2009; Novaes 2010; Novaes 2012; Shimada 2014; Souza 2018). We judged 11 studies (14%) as unlikely to correctly classify our target condition of enamel caries (Bussaneli 2015; Espelid 1986; Jablonski‐Momeni 2017; Matos 2011; Mialhe 2003; Mortensen 2018; Novaes 2009; Novaes 2010; Novaes 2012; Shimada 2014; Souza 2018). This was typically due to the different reference standards used across the studies (e.g. excavation of teeth judged to require restoration, or the use of visual examination with or without separation of the proximal surfaces). It was unclear whether independent examiners assessed the reference standard in 30 (39%) studies however, we were satisfied that the reference standard examiner was independent in 44 (57%) studies. No studies were considered to be at high concern for applicability for this domain.

Sixty‐nine (90%) studies were judged to be at low risk of bias for patient flow. The time interval between administration of the index test and reference standard was unclear in 4 (5%) studies (Freitas 2016; Jablonski‐Momeni 2017; Senel 2010; Shimada 2014). There was variation in the reference standards in three studies due to different reference standards for suspected enamel and dentine caries (enhanced visual examination or excavation). 17 studies (22%) excluded some observations from the analysis (Ariji 1998; Astvaldsdottir 2012; Bahrololoomi 2015; Bussaneli 2015; Ekstrand 2011; Erten 2005; Goel 2009; Hintze 1996; Jablonski‐Momeni 2017; Ko 2015; Kockanat 2017; Matos 2011; Mialhe 2003; Mortensen 2018; Pontual 2010; Shimada 2014; Wenzel 2002). Where studies clearly reported that teeth did not receive the reference standard due to broken teeth during the sectioning procedure then this did not affect the overall bias result for this domain.

Findings

summary of findings Table 1 reports the results of the 77 included studies which generated 104 datasets. We rated the certainty of the evidence as low and downgraded two levels in total for risk of bias due to limitations in the design and conduct of the included studies, indirectness arising from the in vitro studies, and inconsistency of the results.

Radiological assessment was used to detect early/enamel caries in 15,518 tooth sites or surfaces, with a median prevalence in the included studies of 61%. The principal findings of this review are reported for all included datasets assessing a range of imaging modalities for detecting enamel caries. There was substantial variability in the accuracy estimates reported in the included studies. Sensitivities ranged from 0 to 0.96 and specificities from 0 to 1.00. The coupled forest plots of sensitivity and specificity for each of the included datasets are illustrated in Figure 5. For all radiographic methods the estimated summary sensitivity and specificity point was 0.47 (95% confidence interval (CI) 0.40 to 0.53) and 0.88 (95% CI 0.84 to 0.92), respectively. Figure 6 presents the summary receiver operating characteristic (SROC) plot with the summary point plotted along with the 95% confidence region and prediction region. The plot clearly illustrates the substantial variability in the estimates from the included studies, and this is further reflected in the wide 95% confidence region that covers very low values of sensitivity and specificity.

Figure 5

Forest plot of 104 datasets from 77 studies.

Figure 6

Summary receiver operating characteristic (SROC) plot of 104 datasets from 77 studies.

It should be noted that as 18 studies included in the meta‐analysis reported the use of more than one method on the same tooth surfaces, or a single method on different dentition or on different tooth surfaces, then there is some non‐independence of data in this analysis.

Different imaging modalities

We explored the comparative effects of the different methods using information from all 104 datasets (summary of findings Table 2). The most frequently reported methods were analogue radiographs (55 datasets from 51 studies, 8589 tooth sites or surfaces) and digital radiographs (42 datasets from 40 studies, 5936 tooth sites or surfaces). The performance of CBCT was less frequently studied in only seven studies (7 datasets, 993 tooth sites or surfaces). The estimates of specificity were fairly consistent across the test types, but estimates for sensitivity were much lower for analogue and digital than for CBCT (Figure 7). However, the individual accuracy estimates and the 95% prediction regions clearly illustrate the substantial variability in estimates from the included studies and this is reflected in the wide 95% confidence regions for each of the imaging modalities. We explored the effect of different imaging modalities by including a covariate in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

Figure 7

Forest plot of different imaging: analogue, digital, cone beam computed tomography (CBCT).

0.44 (95% CI 0.38 to 0.50) and 0.90 (95% CI 0.86 to 0.93) for analogue;
0.49 (95% CI 0.42 to 0.56) and 0.87 (95% CI 0.82 to 0.91) for digital;
0.60 (95% CI 0.51 to 0.68) and 0.81 (95% CI 0.73 to 0.88) for CBCT.

The summary points for all three methods are displayed in Figure 8. The addition of test type to the model resulted in a meaningful difference to the sensitivity or specificity estimates (Chi²(4) = 32.44, P < 0.001). Further investigation revealed a difference in the summary sensitivity values across the different methods (Chi²(2) = 22.04, P < 0.001), with highest sensitivity observed for CBCT, and a difference in the summary specificity values across the different methods (Chi²(2) = 11.33, P < 0.005), with highest specificity observed for analogue radiographs.

Figure 8

Summary receiver operating characteristic (SROC) plot of different imaging: analogue, digital, cone beam computed tomography (CBCT).

As summary estimates differed according to imaging method, further exploration of the data was undertaken separately for the 97 analogue and digital datasets and for the seven CBCT datasets. This distinction has clinical applicability as CBCT are not routinely indicated for the diagnosis of caries and hence rarely undertaken by the general dental practitioner.

Different purposes (case finding or detection and diagnosis in a clinical setting)

Of the total 104 datasets, 87 datasets were obtained from in vitro studies (11,827 tooth sites or surfaces) and 17 from in vivo studies (3691 tooth sites or surfaces).

Of the 97 non‐CBCT datasets, 80 datasets were obtained from in vitro studies (10,834 tooth sites or surfaces) and 17 from in vivo studies (3691 tooth sites or surfaces) (Figure 9). We explored the effect of studies in a non‐clinical or clinical setting by including a covariate in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

Figure 9

Forest plot of in vivo/in vitro (97 non‐cone beam computed tomography (non‐CBCT) datasets).

in vitro 0.44 (95% CI 0.37 to 0.50) and 0.89 (95% CI 0.84 to 0.92);
in vivo 0.56 (95% CI 0.43 to 0.69) and 0.89 (95% CI 0.78 to 0.95).

The summary points are illustrated in Figure 10. The addition of a covariate for in vivo or in vitro evaluation to the model did not result in any meaningful difference to the sensitivity or specificity estimates (Chi²(2) = 3.48, P = 0.18).

Figure 10

Summary receiver operating characteristic (SROC) plot of in vivo/in vitro (97 non‐cone beam computed tomography (non‐CBCT) datasets).

We were unable to establish the effect of the reference standard for the seven CBCT datasets as all were in vitro studies.

Different dentition

Of the total 104 datasets, 80 datasets were obtained from studies evaluating the permanent dentition (11,424 tooth sites or surfaces) and 24 from the primary dentition (4094 tooth sites or surfaces).

Of the 97 non‐CBCT datasets with 14,525 tooth sites or surfaces, 73 datasets (10,431 tooth sites or surfaces) evaluated the permanent dentition and 24 (4094 tooth sites or surfaces) evaluated the primary dentition. A coupled forest plot (Figure 11) illustrates the reported sensitivity and specificity estimates. We explored the effect of the permanent or primary dentition by including a covariate for dentition in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

Figure 11

Forest plot of dentition (97 non‐cone beam computed tomography (non‐CBCT) datasets).

permanent dentition 0.44 (95% CI 0.37 to 0.51) and 0.90 (95% CI 0.85 to 0.93);
primary dentition 0.52 (95% CI 0.40 to 0.63) and 0.87 (95% CI 0.77 to 0.93).

The summary points for the different dentitions are displayed in Figure 12. The addition of dentition to the model did not result in any meaningful difference to the sensitivity or specificity estimates (Chi²(2) = 1.34, P = 0.51).

Figure 12

Summary receiver operating characteristic (SROC) plot of dentition (97 non‐cone beam computed tomography (non‐CBCT) datasets).

We were unable to establish the effect of dentition for the seven CBCT datasets as all datasets evaluated the permanent dentition.

Investigations of heterogeneity

Tests for heterogeneity were carried out separately for the analogue and digital datasets and the CBCT dataset. Where possible meta‐regression was used to explore the potential sources of heterogeneity.

Reference standard

Of the total 104 datasets, 91 datasets used a reference standard of histology (12,255 tooth sites or surfaces), eight used a visual reference standard (2450 tooth sites or surfaces), and five used excavation (813 tooth sites or surfaces).

Of the 97 non‐CBCT datasets, 84 used a reference standard of histology (11,262 tooth sites or surfaces), eight used a visual reference standard (2450 tooth sites or surfaces), and five used excavation (813 tooth sites or surfaces) (Figure 13). We explored the effects of excavation, histology, or enhanced visual assessment by including a covariate for reference standard in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

Figure 13

Forest plot of reference standard (97 non‐cone beam computed tomography (non‐CBCT) datasets).

excavation 0.51 (95% CI 0.27 to 0.74) and 0.96 (95% CI 0.80 to 0.99);
histology 0.46 (95% CI 0.39 to 0.53) and 0.88 (95% CI 0.83 to 0.92);
visual 0.47 (95% CI 0.29 to 0.66) and 0.91 (95% CI 0.77 to 0.97).

The summary points for the different reference standards are illustrated in Figure 14. The addition of the reference standard covariate to the model did not result in any meaningful difference to the sensitivity or specificity estimates (Chi²(4) = 2.98, P = 0.56).

Figure 14

Summary receiver operating characteristic (SROC) plot of reference standard (97 non‐cone beam computed tomography (non‐CBCT) datasets).

We were unable to establish the effect of reference standard for the seven CBCT datasets as all used a reference standard of histology.

Tooth surface

Of the total 104 datasets, 61 evaluated proximal surfaces (11,089 tooth sites or surfaces) and 43 evaluated occlusal surfaces (4429 tooth sites or surfaces).

Of the 97 non‐CBCT datasets 58 (10,437) evaluated proximal surfaces and 39 (4088) evaluated occlusal surfaces (Figure 15). We explored the effect of the proximal or occlusal tooth surfaces by including a covariate for dentition in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

Figure 15

Forest plot of tooth surface (97 non‐cone beam computed tomography (non‐CBCT) datasets).

proximal tooth surfaces 0.47 (95% CI 0.39 to 0.55) and 0.89 (95% CI 0.84 to 0.93);
occlusal tooth surfaces 0.45 (95% CI 0.35 to 0.54) and 0.88 (95% CI 0.80 to 0.93).

The summary points for the different reference standards are illustrated in Figure 16. The addition of tooth surface to the model did not result in any meaningful difference to the sensitivity or specificity estimates (Chi²(2) = 0.72, P = 0.70).

Figure 16

Summary receiver operating characteristic (SROC) plot of tooth surface (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Of the seven CBCT datasets, three (652 tooth sites or surfaces) evaluated proximal surfaces and four (341 tooth sites or surfaces) evaluated the occlusal surfaces. Assuming a common variance the summary points for sensitivity and specificity were:

proximal surfaces 0.68 (95% CI 0.42 to 0.86) and 0.90 (95% CI 0.77 to 0.96);
occlusal surfaces 0.76 (95% CI 0.55 to 0.89) and 0.54 (95% CI 0.31 to 0.75).

The addition of tooth surface to the model did not result in any meaningful difference to the sensitivity or specificity estimates (Chi²(2) = 5.68, P = 0.06).

Prevalence of caries into dentine (D₃)

The datasets were categorised according to the prevalence of caries into dentine with low (0% to 14%), medium (15% to 34%), and high prevalence (≥ 35%), as per the other reviews in this series. Where studies did not report the prevalence of dentine caries an estimation of the level was made wherever possible, based on the description in the paper and the reporting of the enamel prevalence. The prevalence of dentine caries was high in 30 datasets (3400 tooth sites or surfaces), medium in 40 datasets (5565 tooth sites or surfaces), and low in 22 datasets (4775 tooth sites or surfaces). The prevalence of dentine caries was not reported and could not be estimated in the remaining 12 datasets (1778 tooth sites or surfaces).

Of the 97 non‐CBCT datasets, the prevalence of dentine caries was high in 30 datasets (3400 tooth sites or surfaces), medium in 34 datasets (4632 tooth sites or surfaces), and low in 22 datasets (4775 tooth sites or surfaces). The prevalence of dentine caries was not reported and could not be estimated in the remaining 11 datasets (1718 tooth sites or surfaces) (Figure 17). For the purposes of analysis the high and unreported prevalence categories were combined.

Figure 17

Forest plot of caries into dentine prevalence category (97 non‐cone beam computed tomography (non‐CBCT) datasets).

We explored the effect of the prevalence of caries into dentine by including a covariate in the bivariate model to assess the effects on sensitivity or specificity or both. Assuming a common variance the summary points for sensitivity and specificity were:

low prevalence 0.36 (95% CI 0.25 to 0.47) and 0.91 (95% CI 0.85 to 0.95);
medium prevalence 0.51 (95% CI 0.40 to 0.61) and 0.87 (95% CI 0.78 to 0.92);
high or unreported prevalence 0.50 (95% CI 0.40 to 0.59) and 0.89 (95% CI 0.82 to 0.93).

The summary points are illustrated in Figure 18. The addition of D₃ prevalence category to the analysis did not result in a meaningful difference in sensitivity or specificity (Chi²(4) = 4.59, P = 0.33).

Figure 18

Summary receiver operating characteristic (SROC) plot of caries into dentine prevalence category (97 non‐cone beam computed tomography (non‐CBCT) datasets).

We were unable to establish the effect of prevalence of caries into dentine for the seven CBCT datasets as all but one reported a category of medium prevalence.

Sensitivity analysis ‐ direct comparisons of test type

In a change to the protocol we were able to directly compare different tests based on data from studies which had applied more than one index test with a reference standard (within‐person analysis or paired studies). The methods for exploration of covariate effects on the accuracy estimates for the direct comparisons are as stated above.

Direct comparison of analogue and digital

Thirty‐two datasets from 15 studies (2152 tooth sites or surfaces) provided data to enable a direct comparison of analogue and digital methods of caries detection (Abesi 2012; Ariji 1998; Ashley 1998; Astvaldsdottir 2012; Costa 2002; Da Silva 2010; Erten 2005; Firestone 1998; Pontual 2010; Rockenbach 2008; Safi 2015; Senel 2010; Svanaes 2000; Tarim 2014; Mitropoulos 2010). The results from the bivariate analyses (same variance) were as follows (Figure 19):

Figure 19

Summary receiver operating characteristic (SROC) plot of within‐person analysis analogue and digital radiographs.

all 32 datasets sensitivity 0.44 (95% CI 0.33 to 0.56) and specificity 0.90 (95% CI 0.84 to 0.93), similar to values for the 97 datasets (sensitivity 0.46 (95% CI 0.40 to 0.51) and specificity 0.89 (95% 0.86 to 0.92));
analogue sensitivity 0.41 (95% CI 0.30 to 0.53) and specificity 0.91 (95% CI 0.86 to 0.94);
digital sensitivity 0.47 (95% CI 0.36 to 0.59) and specificity 0.88 (95% CI 0.83 to 0.92).

There was evidence of a difference in accuracy estimates according to test type (Chi²(2) = 10.34, P = 0.006). Further investigation revealed that there was a difference in sensitivity (Chi²(2) = 8.06, P = 0.005) but no difference in specificity (Chi²(2) = 2.27, P = 0.13).

Direct comparison of digital and CBCT

Twelve datasets from six studies (925 tooth sites or surfaces) provided data to enable a direct comparison of digital and CBCT methods of caries detection (Kalathingal 2007; Ramezani 2016; Rathore 2012; Safi 2015; Senel 2010; Tarim 2014). The results from the bivariate analyses (same variance) were as follows (Figure 20):

Figure 20

Summary receiver operating characteristic (SROC) plot of within‐person analysis digital radiographs and cone beam computed tomography (CBCT).

all 12 datasets sensitivity 0.73 (95% CI 0.61 to 0.82) and specificity 0.73 (95% CI 0.39 to 0.92);
CBCT sensitivity 0.76 (95% CI 0.65 to 0.85) and specificity 0.69 (95% CI 0.35 to 0.90);
digital sensitivity 0.69 (95% CI 0.57 to 0.80) and specificity 0.76 (95% CI 0.43 to 0.93).

There was evidence of a difference in accuracy estimates according to test type (Chi²(2) = 7.81, P = 0.02). Further investigation revealed that there was a difference in sensitivity (Chi²(2) = 5.09, P = 0.02) but no difference in specificity (Chi²(2) = 2.72, P = 0.10).

Direct comparison of analogue and CBCT

Six datasets from three studies (737 tooth sites or surfaces) provided data to enable a direct comparison of analogue and CBCT methods of caries detection (Safi 2015; Senel 2010; Tarim 2014). We were unable to pool the results and so the results of the individual studies are included in Figure 21.

Figure 21

Summary receiver operating characteristic (SROC) plot of within‐person analysis analogue radiographs and cone beam computed tomography (CBCT).

Sensitivity analysis ‐ quality items

In the protocol we proposed sensitivity analysis was restricting studies to:

low risk of bias for the index test domain;
low risk of bias for the reference standard domain; and
low prevalence of dentine caries (i.e. less than 35%).

The poor overall quality of many of the studies precluded any meaningful sensitivity analysis, as to do so would have meant discarding a substantial number of included studies from the analysis. For example, many studies failed to clearly report the level of cavitation that was present in the eligible population from which participants were recruited for the obtained sample. Descriptions of the inclusion criteria used in many of the studies were vague at best.

For all radiographic methods across the 104 datasets the estimated summary sensitivity and specificity point was 0.47 (95% CI 0.40 to 0.53) and 0.88 (95% CI 0.84 to 0.92) respectively.

Thirty‐five datasets were judged as low risk of bias for the index test (Abesi 2012; Astvaldsdottir 2012; Attrill 2001; Diniz 2010; Diniz 2011; Ekstrand 2011; Erten 2005; Espelid 1986; Firestone 1998; Haak 2003; Haiter‐Neto 2007; Haiter‐Neto 2008; Haiter‐Neto 2009; Huysmans 1997; Isidor 2009; Kalathingal 2007; Kockanat 2017; Kulczyk 2014; Kutcher 2006; Lussi 2006; Mendes 2006; Mialhe 2003; Mitropoulos 2010; Neuhaus 2011; Pakkala 2012; Rockenbach 2008; Schulze 2004; Shimada 2014; Simon 2016; Souza 2014; Souza 2018; Svanaes 2000; Tarim 2014; Xavier 2011; Zangooei 2010). Restricting the analysis to these datasets resulted in estimates of sensitivity and specificity of 0.51 (95% CI 0.43 to 0.61) and 0.88 (95% CI 0.82 to 0.92), a marginal change to the overall estimates.

Twenty‐one datasets were judged as high or unclear risk of bias for the reference standard (Apostolopoulou 2009; Bahrololoomi 2015; Braun 2017; Bussaneli 2015; Bussaneli 2015a; Costa 2002; Espelid 1986; Jablonski‐Momeni 2012; Jablonski‐Momeni 2017; Ko 2015; Matos 2011; Mialhe 2003; Mortensen 2018; NCT02657538; Neuhaus 2011; Novaes 2009; Novaes 2010; Novaes 2012; Shimada 2014; Souza 2018; Tonkaboni 2019). Removing these datasets from the analysis resulted in estimates of sensitivity and specificity of 0.51 (95% CI 0.45 to 0.57) and 0.87 (95% CI 0.83 to 0.90), a marginal change to the overall estimates.

The reported or imputed prevalence of dentine caries was low (< 15%) in 22 datasets (Behere 2011; Haiter‐Neto 2007; Haiter‐Neto 2008; Haiter‐Neto 2009; Hintze 1996; Hintze 2003; Huysmans 1997; Isidor 2009; Ko 2015; Matos 2011; Mialhe 2003; NCT02657538; Novaes 2009; Novaes 2010; Pakkala 2012; Pontual 2010; Rocha 2003; Souza 2014 (two datasets); Souza 2018; Xavier 2011). Restricting the analysis to these 22 datasets, the sensitivity and specificity estimates of these datasets was 0.34 (95% CI 0.25 to 0.45) and 0.91 (95% CI 0.87 to 0.94), a meaningful decrease in the overall sensitivity estimate.

Discussion

Summary of main results

Radiographs have long been used as a vital tool in the detection of dental caries by the general dental practitioner and other dental professionals. This systematic review specifically investigated the ability of analogue and digital radiographs, and cone beam computed tomography (CBCT) to detect early or enamel caries, with the rationale that once caries at this level is observed, then caries management strategies can be deployed to restrict disease progression and limit the need for future restorations. Many potentially eligible studies did not meet the inclusion criteria for this review even though they investigated an index test of interest with an appropriate reference standard, the reason being that they did not report the data in a 2 x 2 table or in a format to enable a 2 x 2 table to be constructed. Despite these limitations, 104 datasets from 77 studies were available for analysis.

We judged the overall certainty of the evidence to be low, and downgraded two levels in total for study limitations, indirectness due to the many in vitro studies included, and inconsistency of the results.

There are inherent difficulties in conducting diagnostic accuracy studies to detect and inform the diagnosis of caries, not least the choice of a suitable reference standard. The preferred reference standard is histology, but this requires a tooth to be extracted and sectioned prior to microscopic investigation. Clearly this is not achievable in a clinical study unless participants have teeth that have previously been indicated for orthodontic or third molar extraction or extraction due to periodontal disease. In younger children or adolescents clinical studies may be undertaken on teeth that are due to exfoliate. Such studies may evoke applicability concerns due to the specific subpopulations studied who may not necessarily be representative of the wider population. There are therefore difficult trade‐offs to be made between patient selection, applicability of study participants, and choice of reference standard which means that designing a study which minimises risk of bias and maximises applicability is particularly difficult. No studies in this review were judged as low risk of bias with low concern for applicability for all domains. Participant selection was the domain where we observed the highest percentage of studies judged to be at high risk of bias. The sample patients, teeth, or surfaces should be recruited consecutively or randomly with clearly reported methods to avoid any suggestion that observations are included that are more complex or straightforward to diagnose which would clearly introduce potential bias to the sample. There were also substantial applicability concerns for this domain, due to many studies being undertaken in an in vitro setting. We judged only 45% of studies as low risk of bias for the index test domain, for reasons including a lack of independence of examiners for multiple assessments within the same study, or in vitro studies that did not create a model for the teeth that attempted to recreate the soft tissues and facial structure. Specifically, the placement/position of teeth should be comparable to teeth within the jaw, and an attempt made to recreate the effect that the cheek would have on the efficacy of a radiograph. This was most often achieved with a water bath or a plastic or resin barrier being placed between the tooth and the radiograph device. Applicability was of high concern for in vitro studies that failed to replicate a clinical context.

We judged most studies to be at low risk of bias for the reference standard. This was due to the volume of in vitro studies which used a reference standard of histology. The studies that used enhanced visual methods as a reference standard were deemed to be at high risk of bias for this domain.

The main findings of this review are reported overall for all technologies and irrespective of study design and patient characteristics.

For all included studies, the summary estimates were sensitivity of 0.47 (95% confidence interval (CI) 0.40 to 0.53) and specificity 0.88 (95% CI 0.84 to 0.92). There was substantial variability observed in the accuracy estimates of the individual studies, especially for sensitivity. This was reflected in the 95% prediction region which indicated that the estimates of sensitivity and specificity of a future study could fall within a broad range given the results of the studies that have already been observed and included in this analysis. It should be noted that 18 studies included in this meta‐analysis reported the use of more than one method on the same tooth surfaces, or a single method on different dentition or on different tooth surfaces, therefore there is some non‐independence of data in this analysis. In a hypothetical cohort of 1000 tooth sites or surfaces with a prevalence of 63% (the median prevalence observed in studies included in the meta‐analysis), the sensitivity of 0.47 and specificity of 0.88 would result in 337 not being identified as having early caries when actually there was caries present (false negatives) and 43 being identified as having caries when it did not exist (false positives) (summary of findings Table 1). The consequences of these misclassifications are concerning. The false‐negative results imply that patients who would be considered for treatment would not receive it, and an opportunity to prevent the further development of a carious lesion would be missed. Given the recall period for routine dental examinations and the slow growing nature of the disease the clinician may be reassured that the lesion would be identified at the patient's next appointment. However, guidelines of the Faculty of General Dental Practice (UK) and other professional bodies recommend appropriate time intervals between radiographs, and so subsequent clinical examinations are unlikely to be undertaken with the radiograph adjunct. The false‐positive results for enamel caries would mean that topical fluoride or other minimally invasive treatments would be deployed when not required, the impact on the patient would be low. The prevalence applied to this scenario is potentially inflated due to many of the included studies being based on extracted teeth. Using the median prevalence of caries of any severity from a national epidemiological study (Steele 2011) the median prevalence is lower, 28%. When this lower prevalence is applied to a hypothetical cohort of 1000 tooth sites or surfaces, the revised numbers are 150 being incorrectly classified as healthy and 84 as having caries when the true disease state of the tooth site or surface is sound.
Differences on summary estimates according to analogue, digital, and CBCT were observed (Chi²(2) = 32.44, P < 0.001). Further investigation revealed a difference in the summary sensitivity values across the different technologies (Chi²(2) = 22.04, P < 0.001) and in specificity values (Chi²(2) = 11.33, P < 0.005), with the highest sensitivity observed for CBCT and the highest specificity observed for analogue radiographs (summary of findings Table 2). As with the overall analysis, there were concerns regarding the broad range of the 95% prediction interval and expectations for future studies with all three modalities. All subsequent analyses were undertaken separately for the analogue and digital devices and for the CBCT modalities. CBCT was included in the analysis to provide a complete overview of all relevant technologies, however this method would not typically be undertaken for the detection of enamel caries due to the high doses of radiation that patients would be exposed to.
There was no meaningful difference in observed accuracy estimates obtained from the in vivo or in vitro settings (Chi²(2) = 3.48, P = 0.18). One reason for the variability in any observed sensitivity estimates could be whether the study was undertaken in a laboratory setting and evaluated extracted teeth (in vitro) or whether it was a clinical based study on teeth in situ (in vivo). In vivo studies are closest to evaluating the use of different technologies for diagnosis on a presenting patient. In vitro studies more closely mimic case finding or case detection. There was no meaningful difference observed in the investigations of study setting. Specificity results were virtually identical but the sensitivity was slightly higher in the in vivo studies (0.56 (95% CI 0.43 to 0.69)) than for the in vitro studies (0.44 (95% CI 0.37 to 0.50)). This may have been a result of the imperfect reference standards used in the in vivo setting which may have failed to identify the same cases of caries as the radiograph and therefore create an inflated sensitivity. It would be reasonable to assume that better quality radiographic standards can be expected in an in vivo research project than in everyday practice, and therefore the in vivo results reported in this review can be considered an optimum level of performance, and, in reality, diagnostic accuracy may be expected to be poorer and more variable.
Summary values of sensitivity and specificity values were according to dentition, tooth surface, reference standard, and prevalence of dentinal caries in the study sample.
- As the depth of enamel is thinner and as caries progresses more rapidly in the primary dentition, leading to more severe decay with greater expedience, we considered it important to explore potential differences in results according to the primary and permanent dentition. The need for prevention in permanent teeth is even more important as individuals aim to retain permanent teeth across their lifespan. Most studies evaluated permanent teeth, but there were no meaningful differences observed in the summary estimates of the primary or permanent dentition.
- The tooth surface result is interesting as the radiograph is particularly valuable as a tool for detecting caries on proximal surfaces which are difficult to access. These surfaces cannot be easily viewed when performing a visual assessment and other types of technologies or devices (investigated in this series of Cochrane Reviews) are difficult to apply to proximal surfaces, so radiographs are considered to be of particular benefit to the clinician for this aspect of the clinical examination. On the occlusal surface the invaginated anatomy of the fissure means that early caries occurs on the walls of the fissure in its depth. In this area an X‐ray beam has to potentially travel through much more sound enamel and dentine buccal and lingual to the lesion before it hits the radiographic film or sensor, which would attenuate the X‐ray beam, making early detection radiographically very difficult. On the proximal surface however, as the tooth tissues curve toward the contact with the adjacent tooth, this sound tooth tissue buccal and lingual to an early lesion is much less, with much less attenuation of the X‐ray beam occurring making it potentially easier to detect proximal lesions radiographically. In this meta‐analysis the proportion of studies with a prevalence of dentine lesions > 15% was far higher in the studies evaluating the occlusal surfaces than those evaluating the proximal surfaces, making the classification task easier on the occlusal surfaces than may ordinarily be observed in clinical practice. The performance on occlusal surfaces may therefore, in part, be an artefact of a higher proportion of more severe decay in the study samples.
- The results suggest that there is no difference between the ability of radiographs to detect caries on the occlusal or proximal surfaces with analogue or digital radiographs, or with CBCT. For CBCT however, the specificity for the occlusal surfaces 0.54 (95% CI 0.31 to 0.75) was markedly lower than for the proximal surfaces 0.90 (95% CI 0.77 to 0.96).
- No differences were observed according to reference standard or prevalence of dentinal caries in the study sample.

Strengths and weaknesses of the review

The strengths of this review are the comprehensive literature search and rigorous application of appropriate methodology which ensured that all screening, inclusion decisions, and data extraction were performed in duplicate and with clinical and methodological expertise. A clear and reproducible method was used for the application of methodological quality decisions and data analysis. The relatively large volume of data enabled meta‐analysis using hierarchical bivariate methods which have been shown to be mathematically superior to simpler methods commonly used in systematic reviews of diagnostic test accuracy in oral health (Dinnes 2016). We elected to initially consider all studies irrespective of study design and patient characteristics, and to formally investigate covariates of interest through meta‐analysis, in contrast to the qualitative approach taken in other systematic reviews.

This systematic review builds upon existing literature. Bader 2002 completed a systematic review of radiograph devices (alongside visual, fluorescence, and other methods), but limited the inclusion criteria to studies where the index test was verified by a histological reference standard. Only two included studies reported on the assessment of enamel lesions. This Cochrane Review is a significant update which broadens the available reference standards to include visual and excavation in addition to histology, and substantially increases the number of included studies that have reported data at the enamel threshold. A later review by Gomez 2013 also investigated a variety of methods for caries detection, limited to non‐cavitated carious lesions. Study quality was assessed using a caries‐specific rating scale but no meta‐analysis was carried out. Potential sources of heterogeneity were explored but no formal analysis was undertaken.

Of most relevance to this review, Schwendicke 2015 presented a comprehensive review of radiograph techniques at three levels of dentine lesions, cavitated lesions, and any lesions. This review included 117 studies with a search date current up to September 2014. Our review updates this review but focuses on the enamel threshold of disease so although reporting a lower number of studies, the number of included studies reporting data at the enamel threshold has increased. In contrast to the Schwendicke 2015 review, the use of the bivariate methodology for the estimation of summary points and formal investigation of potential sources of heterogeneity through meta‐regressions is of substantial benefit, along with assessment of the certainty of the evidence. However, our final conclusion is similar, that digital and analogue radiographs show limited sensitivity for detecting early caries, in contrast to CBCT.

One weaknesses of this review is the large volume of ineligible studies due to insufficient information to extract or construct a 2 x 2 table for estimates of sensitivity and specificity. In the fluorescence review of this series (Macey 2020) we observed that 55 out of the 133 studies that were otherwise eligible for inclusion in the review did not permit the construction of a 2 x 2 table. There is no reason to expect that the proportion of potentially eligible studies for this review would be different. Many potentially eligible studies did not present the numbers of true positives, true negative, false positives, and false negatives at the enamel threshold. Instead they reported sensitivity, specificity, and area under the curve as their primary results, and often did not include prevalence of caries at the enamel threshold.

A significant source of bias particularly evident in the in vitro studies was that the participants or teeth were selected, with the potential that teeth were selected where the spectrum of disease was not representative and where classification of disease was artificially facilitated leading to inflated accuracy estimates.

The inclusion criteria of the review were selected to ensure that the focus was placed on the early detection of caries or caries limited to enamel. With the best of intentions studies could easily attempt to recruit sound or non‐cavitated teeth but when investigated with a thorough/complete reference standard, it become apparent that some tooth sites or surfaces, when viewed during participant selection, actually harboured dentinal caries. The concern of the review team was that if studies intentionally recruited dentinal lesions, then there would be a simplification of the detection and diagnostic decision as more advanced lesions obviously into dentine or frankly cavitated are generally easier to detect than earlier lesions which are limited to enamel. A further complication arose where some studies were poorly reported or lacked clarity on the selection criteria that they imposed on their sample. We took the position that unless the authors clearly stated that frank or dentinal cavities were intentionally included, then we were unable to exclude the study from the review. Due to incomplete reporting in the primary studies this decision has been difficult to apply consistently and a consequence is that some well‐reported studies have been excluded due to their clarity of reporting, whereas studies which may have intentionally included dentinal lesions but failed to explicitly report this fact have been included. To our best efforts this has been accounted for in the QUADAS‐2 assessment. We accept that this may leave the review open to criticism, however we would reiterate that this review intended to synthesise the evidence on early lesions and the inclusion of frank or dentinal cavities was not applicable to this question. The analysis by prevalence of caries at the dentinal level allowed us to investigate the effect of the inclusion of tooth sites or surfaces with dentinal decay. This analysis indicated that there was no meaningful difference in the accuracy estimates of studies with high or low prevalence of dentinal caries, and we can therefore conclude that the results of this review are robust, and the results confirm this to be the case.

An area of concern arose where the reference standard was histology or enhanced visual assessment and it was not clearly reported whether the same examiners conducted the index test and reference standard assessments. This issue was logged in the characteristics tables but our interpretation was that this would not affect the conduct or interpretation of the histological reference standard as it was hard to see how an examiner would remember the specific detail of a radiograph and recall it during the examination of a sectioned tooth. It is also important to note that for an enhanced visual assessment to be considered acceptable as a reference standard, then separators needed to be applied to allow the approximal surface to be clearly viewed.

Applicability of findings to the review question

Clearly there are concerns regarding the clinical applicability of the findings of this review. This is highlighted by the fact that the meta‐analysis is dominated by in vitro studies which are not conducted in a clinical setting which is representative of general dental practice. There is also concern that some of the in vitro studies also failed to make a reasonable attempt to replicate the conditions found in the oral cavity. However, the formal analysis concluded that there was no observed differences in the summary estimates from in vivo or in vitro studies.

We did not include any studies that evaluated caries at the enamel threshold adjacent to existing restorations (none were found in the search looking at enamel caries), and it is very unlikely that caries adjacent to existing restorations within the same tooth would be restricted to enamel.

Figure 1

Figure 2

Study flow diagram.

Figure 3

Risk of bias and applicability concerns summary: review authors' judgements about each domain for each included study.

Figure 4

Risk of bias and applicability concerns graph: review authors' judgements about each domain presented as percentages across included studies.

Figure 5

Forest plot of 104 datasets from 77 studies.

Figure 6

Summary receiver operating characteristic (SROC) plot of 104 datasets from 77 studies.

Figure 7

Forest plot of different imaging: analogue, digital, cone beam computed tomography (CBCT).

Figure 8

Summary receiver operating characteristic (SROC) plot of different imaging: analogue, digital, cone beam computed tomography (CBCT).

Figure 9

Forest plot of in vivo/in vitro (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 10

Summary receiver operating characteristic (SROC) plot of in vivo/in vitro (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 11

Forest plot of dentition (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 12

Summary receiver operating characteristic (SROC) plot of dentition (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 13

Forest plot of reference standard (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 14

Summary receiver operating characteristic (SROC) plot of reference standard (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 15

Forest plot of tooth surface (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 16

Summary receiver operating characteristic (SROC) plot of tooth surface (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 17

Forest plot of caries into dentine prevalence category (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 18

Summary receiver operating characteristic (SROC) plot of caries into dentine prevalence category (97 non‐cone beam computed tomography (non‐CBCT) datasets).

Figure 19

Summary receiver operating characteristic (SROC) plot of within‐person analysis analogue and digital radiographs.

Figure 20

Summary receiver operating characteristic (SROC) plot of within‐person analysis digital radiographs and cone beam computed tomography (CBCT).

Figure 21

Summary receiver operating characteristic (SROC) plot of within‐person analysis analogue radiographs and cone beam computed tomography (CBCT).

Test 1

All

Test 2

Analogue

Test 3

Digital

Test 4

CBCT

Test 5

All non‐CBCT

Summary of findings 1. Summary of findings table ‐ main results

Question	What is the diagnostic accuracy of radiographic methods for the detection and diagnosis of early dental caries?
Population	Asymptomatic children or adults presenting for clinical examination (clinical studies); extracted teeth of children or adults (in vitro studies). Clinical or in vitro studies which intentionally included dentine and frank cavitations for assessment were excluded
Index test	Intra and extraoral radiographic caries detection methods completed on intact teeth including: analogue (conventional) radiographs; digital radiographs; and CBCT
Comparator test	Estimates were compared across different radiographic methods. A separate review in this series explores the comparative accuracy of visual classification, fluorescence‐based, radiograph, and transillumination methods of detection and diagnosis
Target condition	Early dental caries (positivity threshold of early caries or beyond)
Reference standard	Histology, excavation, enhanced visual examination
Action	If dental caries can be detected at an early stage then remedial action can be taken to arrest or even reverse decay, and potentially prevent restorations
Diagnostic stage	Aimed at the general dental practitioner assessing regularly attending patients for early stage caries
Quantity of evidence	77 studies providing 104 datasets; 9331 lesions in 15,518 tooth sites or surfaces (median 63% prevalence)
Findings
Sensitivity (95% CI)^a		0.47 (0.40 to 0.53)
Specificity (95% CI)^a		0.88 (0.84 to 0.92)
Outcome	Numbers applied to a hypothetical cohort of 1000 tooth surfaces Effect per 1000 tooth surfaces assessed (95% CI)		Test accuracy Certainty of the evidence
Outcome	Pre‐test probability 28%^b	Pre‐test probability 63%^b	Test accuracy Certainty of the evidence
True positives (tooth surfaces with early enamel caries)	130 (113 to 148)	293 (254 to 332)	Sensitivity ‐ ⊕⊕⊝⊝ LOW Specificity ‐ ⊕⊕⊝⊝ LOW
False negatives (tooth surfaces incorrectly classified as not having early enamel caries)	150 (132 to 167)	337 (298 to 376)
True negatives (tooth surfaces without early enamel caries)	636 (606 to 660)	327 (312 to 339)
False positives (tooth surfaces incorrectly classified as having early enamel caries)	84 (60 to 114)	43 (31 to 58)
Limitations ‐ factors that may decrease the certainty of the evidence
Risk of bias	No studies were considered to be at low risk of bias overall. Across the 4 domains, the patient selection domain had the highest proportion of studies judged at high risk of bias (56%). We judged the index test, reference standard, and flow and timing domains to have a lower proportion of high risk of bias studies (39%, 16%, and 9% respectively). All but 1 study avoided a case‐control design, 57% avoided inappropriate exclusions, whilst only 9% reported the use of random or consecutive sampling. Most included studies were in vitro studies using histology as the reference standard, and likely to correctly classify the target condition, however some studies used an imperfect reference standard such as excavation or a visual examination with or without separation of the approximal surfaces
Applicability of evidence to question	16 studies (21%) were considered to have low concern for applicability across all domains. Applicability of patients and setting was of high concern in 75% of studies, mostly because they investigated extracted teeth and, as the objective of this review was to inform general clinical practice, there may be concerns that such results may have limited relevance to a clinical setting. The conduct or interpretation of the index test was of high concern in 29% of studies. There was low concern that the target condition as defined by the reference standard does not match the review question in all studies
Certainty of the evidence	We rated the certainty of the evidence as low, and downgraded 2 levels in total for risk of bias due to limitations in the design and conduct of the included studies, indirectness arising from the in vitro studies, and inconsistency of the results
^aThere was variability in the results of the individual studies, with sensitivities which ranged from 0 to 0.96 and specificities from 0 to 1.00. The variability in the results was partly explained by different methods evaluated. ^bHypothetical cohorts of 1000 tooth sites or surfaces are presented for numbers estimated at prevalence of 28% and 63% of enamel caries prevalence. Based on consultation with clinical colleagues, the lower prevalence figure addresses concerns that the higher prevalence value is not representative of the general population and is taken from the level of cavitated teeth in the UK Adult Dental Health Survey (Steele 2011). The higher prevalence figure is the median prevalence of early caries reported in the included studies. CBCT: cone beam computed tomography; CI: confidence interval.

Summary of findings 1. Summary of findings table ‐ main results

Summary of findings 2. Summary of findings table ‐ comparison of tests

Question	What is the diagnostic accuracy of radiographic methods for the detection and diagnosis of early dental caries?
Population	Asymptomatic children or adults presenting for clinical examination (clinical studies); extracted teeth of children or adults (in vitro studies). Clinical or in vitro studies which intentionally included dentine and frank cavitations for assessment were excluded
Index test	Intra and extraoral radiographic caries detection methods completed on intact teeth including: analogue (conventional) radiographs; digital radiographs; and CBCT For the purposes of this review the positivity threshold was caries in enamel
Comparator test	Estimates were compared across different radiographic methods. A separate review in this series explores the comparative accuracy of visual classification, fluorescence‐based, radiograph, and transillumination methods of detection and diagnosis
Target condition	Early dental caries (positivity threshold of early caries or beyond)
Reference standard	Histology, excavation, enhanced visual assessment
Action	If dental caries can be detected at an early stage then remedial action can be taken to arrest or even reverse decay, and potentially prevent restorations
Diagnostic stage	Aimed at the general dental practitioner assessing regularly attending patients for early stage caries
Quantity of evidence	77 studies providing 104 datasets; 9331 lesions in 15,518 tooth surfaces (median 63% prevalence)
Findings: analysis comparing analogue radiographs, digital radiographs, CBCT. Consequences in a cohort of 1000 tooth sites or surfaces
Test	Datasets	Tooth surfaces	Sensitivity (95% CI)	Specificity (95% CI)	Pre‐test probability 28%		Pre‐test probability 63%
Test	Datasets	Tooth surfaces	Sensitivity (95% CI)	Specificity (95% CI)	Missed	Overdiagnosis	Missed	Overdiagnosis
Analogue radiograph	55	8589	0.44 (0.38 to 0.50)	0.90 (0.86 to 0.93)	157 (139 to 174)	75 (53 to 104)	353 (312 to 392)	38 (27 to 53)
Digital radiograph	42	5936	0.49 (0.42 to 0.56)	0.87 (0.82 to 0.91)	143 (125 to 162)	92 (66 to 128)	323 (280 to 365)	47 (34 to 66)
CBCT	9	1081	0.60 (0.51 to 0.68)	0.81 (0.73 to 0.88)	112 (89 to 137)	134 (89 to 196)	253 (200 to 307)	69 (46 to 101)
The addition of test type to the model resulted in a meaningful difference to the sensitivity and specificity estimates: Chi² = 32.44, df = 4, P < 0.001
Interpretation: these results should be interpreted taking into account the factors that limit the certainty of the evidence as indicated in summary of findings Table 1
Using analogue as the reference standard: difference in sensitivity for CBCT = ‐0.16 (‐0.23 to ‐0.09), P < 0.001; digital = ‐0.04 (‐0.09 to ‐0.01), P = 0.018 difference in specificity for CBCT = 0.08 (0.02 to 0.14), P = 0.007; digital = 0.02 (‐0.01 to 0.06), P = 0.107
CBCT: cone beam computed tomography; CI: confidence interval; df: degrees of freedom.

Summary of findings 2. Summary of findings table ‐ comparison of tests

Table 1. Classification of levels of caries levels

DMFT classification	Definition (Pitts 2001 )
0	Sound (non‐diseased)
D₁	Non‐cavitated yet clinically detectable enamel lesions with intact surfaces
D₂	Cavitated lesion penetrating the enamel or shadowing
D₃	Cavity progressing past the enamel‐dentine junction into dentine
D₄	Cavity progressing into pulp
DMFT = decayed, missing, and filled teeth.

Table 1. Classification of levels of caries levels

Table 2. QUADAS‐2 tool

Item	Response (delete as required)
Participant selection – Risk of bias
1) Was a consecutive or random sample of participants or teeth used?	Yes – where teeth or participants were selected consecutively or allocated to the study via a randomisation process No – if study described another method of sampling Unclear – if participant sampling is not described
2) Was a case‐control design avoided?	Yes – if case‐control clearly not used No – if study described as case‐control or describes sampling specific numbers of participants with particular diagnoses Unclear – if not clearly described
3) Did the study avoid inappropriate exclusions (e.g. inclusion of caries into dentine)?	Yes – if the study clearly reports that included participants or teeth were apparently healthy or caries into dentine were excluded No – if lesions were included that showed caries into dentine or exclusions that might affect test accuracy (e.g. teeth with no caries) Unclear – if not clearly reported
Could the selection of participants have introduced bias?
If answers to all of questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) and 3) was 'no'	Risk is High
If answers to any of questions 1) and 2) and 3) was 'unclear'	Risk is Unclear
Participant selection – Concerns regarding applicability
1) Does the study report results for participants or teeth selected by apparent health or suspected early caries (i.e. studies do not recruit patients who are known to have advanced caries into dentine)?	Yes – if a group of participants or teeth has been included which is apparently healthy or indicative of early caries No – if a group of participants or teeth has been included which is suspected of advanced caries Unclear – if insufficient details are provided to determine the spectrum of participants or teeth
2) Did the study report data on a per‐patient rather than on a tooth or surface basis?	Yes – if the analysis was reported on a surface or tooth basis No – if the analysis was reported on a per‐patient basis Unclear ‐ if it is not possible to assess whether data are presented on a per‐patient or per‐tooth basis
3) Did the study avoid an in vitro setting which required the usage of extracted teeth?	Yes – if the participants were recruited prior to tooth extraction No – if previously extracted teeth were used in the analysis Unclear – if it was not possible to assess the source and method of recruiting of included participants/teeth
Is there concern that the included participants or teeth do not match the review question?
If answers to all of questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) and 3) was 'no'	Risk is High
If answers to any of questions 1) and 2) and 3) was 'unclear'	Risk is Unclear
Index test ‐ Risk of bias (to be completed per test evaluated)
1) Was the index test result interpreted without knowledge of the results of the reference standard?	Yes – if the index test described is always conducted and interpreted prior to the reference standard result, or for retrospective studies interpreted without prior knowledge of the reference standard No – if index test described as interpreted in knowledge of reference standard result Unclear – if index test blinding is not described
2) Was the diagnostic threshold at which the test was considered positive pre‐specified?	Yes – if threshold was pre‐specified (i.e. prior to analysing the study results) No – if threshold was not pre‐specified Unclear – if not possible to tell whether or not diagnostic threshold was pre‐specified
For visual and radiograph tests only: 3) For studies reporting the accuracy of multiple diagnostic thresholds for the same index test or multiple index tests, was each threshold or index test interpreted without knowledge of the results of the others?	Yes – if thresholds or index tests were selected prospectively and each was interpreted by a different clinician or interpreter, or if study implements a retrospective (or no) cut‐off (i.e. look for deepest/most severe lesion first) No – if study states reported by same reader Unclear ‐ if no mention of number of readers for each threshold or if pre‐specification of threshold not reported N/A ‐ multiple diagnostic thresholds not reported for the same index test
Could the conduct or interpretation of the index test have introduced bias?
For visual and radiographic studies item 3) to be added
If answers to all of questions 1) and 2) was 'yes'	Risk is Low
If answers to any of questions 1) and 2) was 'no'	Risk is High
If answers to any of questions 1) and 2) was 'unclear'	Risk is Unclear
Index test ‐ Concerns regarding applicability
1) Were thresholds or criteria for diagnosis reported in sufficient detail to allow replication?	Yes – if the criteria for detection or diagnosis of the target disorder were reported in sufficient detail to allow replication No – if the criteria for detection or diagnosis of the target disorder were not reported in sufficient detail to allow replication Unclear ‐ if some but not sufficient information on criteria for diagnosis to allow replication were provided
2) Was the test interpretation carried out by an experienced examiner?	Yes – if the test clearly reported that the test was interpreted by an experienced examiner No – if the test was not interpreted by an experienced examiner Unclear – if the experience of the examiner(s) was not reported in sufficient detail to judge or if examiners described as 'Expert' with no further detail given
Is there concern that the included participants do not match the review question?
If the answer to question 1) and 2) was 'yes'	Concern is Low
If the answer to question 1) and 2) was 'no'	Concern is High
If the answer to question 1) and 2) was 'unclear'	Concern is Unclear
Reference standard ‐ Risk of bias
1) Is the reference standard likely to correctly classify the target condition?	Yes – if all teeth or surfaces underwent a histological or excavation reference standard No – if a final diagnosis for any participant or tooth was reached without the histological or excavation reference standards Unclear – if the method of final diagnosis was not reported
2) Were the reference standard results interpreted without knowledge of the results of the index test?	Yes – if the reference standard examiner was described as blinded to the index test result No – if the reference standard examiner was described as having knowledge of the index test result Unclear – if blinded reference standard interpretation was not clearly reported
Could the reference standard, its conduct, or its interpretation have introduced bias?
If answers to questions 1) and 2) was 'yes'	Risk is Low
If the answer to question 1) and 2) was 'no'	Concern is High
If the answer to question 1) and 2) was 'unclear'	Concern is Unclear
Reference standard ‐ Concerns regarding applicability
1) Does the study use the same definition of disease positive as the prescribed in the review question?	Yes ‐ same definition of disease positive used, or teeth can be disaggregated and regrouped according to review definition No ‐ some teeth cannot be disaggregated Unclear ‐ definition of disease positive not clearly reported
Flow and timing ‐ Risk of bias
1) Was there an appropriate interval between index test and reference standard (in vivo studies less than 3 months, in vitro no limit but must be stored appropriately)?	Yes ‐ if study reports index and reference standard had a suitable interval or storage method No ‐ if study reports greater than 3‐month interval between index and reference standard or inappropriate storage of extracted teeth prior to reference standard Unclear ‐ if study does not report interval or storage methods between index and histological reference standard
2) Did all participants receive the same reference standard?	Yes ‐ if all participants underwent the same reference standard No ‐ if more than 1 reference standard was used Unclear ‐ if not clearly reported
3) Were all participants included in the analysis?	Yes ‐ if all participants were included in the analysis No ‐ if some participants were excluded from the analysis Unclear ‐ if not clearly reported
If answers to questions 1) and 2) and 3) was 'yes'	Risk is Low
If answers to any one of questions 1) or 2) or 3) was 'no'	Risk is High
If answers to any one of questions 1) or 2) or 3) was 'unclear'	Risk is Unclear
N/A = not applicable; QUADAS‐2 = Quality Assessment of Diagnostic Accuracy Studies 2.

Table 2. QUADAS‐2 tool

Table Tests. Data tables by test

Test	No. of studies	No. of participants
1 All Show forest plot	78	15518

2 Analogue Show forest plot	52	8589

3 Digital Show forest plot	41	5936

4 CBCT Show forest plot	7	993

5 All non‐CBCT Show forest plot	77	14525

Table Tests. Data tables by test