Types of studies

We will include studies that allow for assessment of binomial diagnostic accuracy metrics, i.e. sensitivity and specificity, positive and predictive values, and diagnostic odds ratio of point‐of‐care tests (POCT) in the diagnosis of sickle cell disease (SCD). All the tests should have an identified reference standard.

We will include cross‐sectional studies, also known as single‐gated studies, describing participants receiving the index test and reference standard as (Mathes 2019):

1. studies in which all participants, or their samples, receive a concurrent single index test and a reference standard during the period of the study;

2. studies in which all participants, or their samples, receive either the index test or reference standard retrospectively; and

3. studies in which all participants, or their samples, receive one index test and more than one reference standard concurrently.

We will also include case‐control studies, also known as two‐gated studies, which include participants with or without a prespecified outcome ascertainment of SCD, using the reference standard before the index test. In this case, all participants' samples should be subject to both the index test and reference standard concurrently.

We will exclude studies in which participants did not receive both the index test and reference standard, either concurrently or retrospectively, such as when some participants received the index test, and others received the reference standard.

Participants

This review seeks to include studies with participants receiving a POCT for SCD, specifically, the common SCD genotypes types, such as two HbS together (HbSS), HbS with another abnormal haemoglobin gene (HbSC, HbSD), and HbSβº thalassaemia. The POCTs are relevant for participants of all age groups receiving a screening test for SCD. Participants may be asymptomatic, such as newborns and pregnant women, or symptomatic and undergoing an assessment for SCD.

We will exclude studies that only assess the diagnostic accuracy of POCT among participants with diagnosed SCD, e.g. those receiving hydroxyurea or any HbF‐stimulating chemical (Alapan 2016). We will exclude studies without a control group.

Index tests

The index tests in this review will include POCTs that have been developed to test for SCD. The POCT will be used either for the screening of asymptomatic participants or the diagnosis of symptomatic participants. We will define the index tests as portable tests, conducted onsite with the individual, with a short turnaround time, facilitating a quick referral to a comprehensive care centre for those receiving a positive diagnosis of SCD. Currently, a number of technologies have been developed, including lateral flow immunoassays, paper‐based hb solubility tests, micro‐engineering hb electrophoresis, and density‐based multiphase systems (McGann 2017).

We will exclude solubility and sodium metabisulfite tests that are unable to distinguish between sickle cell trait (HbAS) and SCD.

Target conditions

The target condition is SCD, a heterogenous genetic condition, characterised by the presence of abnormal haemoglobin S (HbS). Although there are many genotypes of SCD, some are more common, such as HbSS, HbSC, and HbSβº thalassaemia; and are responsible for the majority of the morbidity due to SCD (NIH 2014; Rees 2010). The diagnosis of SCD involves the separation and identification of the different types of hb, using different techniques. The diagnosis of some types of SCD often gives similar results, and is at times indistinguishable, e.g. HbSS and HbSβº thalassaemia, are indistinguishable on qualitative assessment, and HbSC might be wrongly classified as HbAS in some tests (NIH 2014).

Reference standards

The traditional methods for diagnosing SCD are routinely used in identifying, separating, and quantifying the normal and variant haemoglobin molecules, taking advantage of charges on the haemoglobin molecules (McGann 2017). There are four methods of diagnosing SCD, used either alone or in combination in a testing algorithm; these include classical electrophoresis (cellulose acetate and citrate agar), isoelectric focusing (IEF), high‐performance liquid chromatography (HPLC), and capillary zone electrophoresis (CZE). All methods are considered to have sensitivity and specificity of at least 99%.

• Classic electrophoresis contains a support medium containing a two‐tiered electrophoretic protocol at alkaline pH ~ 8.6 and acidic pH ~ 6.0, which differentially separates mobile, charged, haemoglobin molecules (McGann 2017). The results are interpreted using visual estimation of band intensities and scanning densitometry.

• IEF is another electrophoretic technique that is an equilibrium process, in which haemoglobins migrate within a gradient to a position of zero net charge, or isoelectric point (pI).

• HPLC provides precise quantification of haemoglobins at a specific pH. The haemoglobin molecules that adhere to a negatively‐charged resin column are removed from the column by a positively‐charged solution. This solution is added in increasing concentration, and competes for binding to the negatively‐charged resin.

• CZE has an alkaline buffer in an electrical field to separate haemoglobin (Alapan 2016; McGann 2017). The haemoglobins migrate toward the cathode in the capillary tube by electro‐osmotic flow and optical density, detected by an ultraviolet system.

Any of two methods are used sequentially to diagnose SCD, and these have a sensitivity and specificity of over 99%. DNA‐based assays have also been developed to detect point mutations in the β globin gene, using restriction enzyme digestion, and amplifying the fragments using the polymerase chain reaction (PCR (Alapan 2016)).

Electronic searches

The Cochrane Cystic Fibrosis and Genetic Disorders Group's Information Specialist will conduct a systematic search of the Group's Haemoglobinopathies Trials Register for relevant studies, using the terms: sickle cell OR (haemoglobinopathies AND general) AND screening OR diagnosis.

The Haemoglobinopathies Trials Register is compiled from electronic searches of the Cochrane Central Register of Controlled Trials (CENTRAL), updated following each new issue of the Cochrane Library, and weekly searches of MEDLINE. Unpublished work is identified by searching the abstract books of five major conferences: the European Haematology Association conference; the American Society of Hematology conference; the British Society for Haematology Annual Scientific Meeting; the Caribbean Public Health Agency Annual Scientific Meeting (formerly the Caribbean Health Research Council Meeting); and the National Sickle Cell Disease Program Annual Meeting. For full details of all searching activities for the register, please see the relevant section of the Cochrane Cystic Fibrosis and Genetic Disorders Group's website.

The Cochrane Child Health Information Specialist will support the team by conducting the search. Please see Appendix 1 for the search strategies:

• MEDLINE Ovid (1946 onwards);

• Embase Ovid (1974 onwards);

• CINAHL Plus EBSCO (Cumulative Index to Nursing and Allied Health Literature; 1937 onwards);

• Global Health – CAB Direct (1910 onwards);

• ISRCTN registry (www.isrctn.com);

• US National Institutes of Health Ongoing Trials Register ‐ ClinicalTrials.gov (www.clinicaltrials.gov);

• World Health Organization International Clinical Trials Registry Platform (WHO ICTRP; apps.who.int/trialsearch).

Searching other resources

We will check the bibliographies of included studies and any relevant systematic reviews identified, for further references to relevant studies.

We will also consult experts in the field of SCD for suggestions of any studies not found through our electronic searching. We will identify these experts through the co‐ordinating centre for the Sickle Cell Disease Implementation Consortium (SCDIC), supported by the National Institutes of Health.

Selection of studies

Two review authors (IK and EK) will independently select studies in two phases. In the first phase, two review authors will independently screen the titles and abstracts of identified reports. During this phase, we will exclude reports by relevance, according to the title and abstract; all included studies will report on SCD. In phase two, two review authors (IK and EK) will screen the full text of the remaining studies, applying the inclusion and exclusion criteria. We will resolve any disagreements in the second phase through consensus, or when required, by consulting with a third review author (DM). When there are insufficient data to determine the eligibility of a potential eligible study, we will contact the study authors for more information.

We will show the process of study selection in a PRISMA flow chart (Figure 2).

Figure 2

---

---

Study flow diagram

We will summarise information from the included and excluded studies, and reasons for exclusion, in separate tables.

Data extraction and management

We will use a data extraction form to record key study characteristics and data, which we extract from all eligible studies. Two review authors (IK and EK) will independently extract the data from included publications into the data tables. A third review author (RM) will verify the data on a random 10% of the included studies. We will resolve any disagreements on data extraction by discussion and consensus; however, if we are not able to reach a consensus, a third review author (DM) will help resolve any issues. We will extract key summaries from the studies, including: unique study identifier; study title; name(s) of first author(s); diagnostic index test and its manufacturer; type of rapid diagnostic test; study setting; publication date; publication number; study design; mean age of participants (±2 standard deviations (SDs)); cadre of health professionals performing the test (laboratory or non‐laboratory staff); study setting (laboratory or field); reference tests; sample size; follow‐up duration; true positives (TP); true negatives (TN); false positives (FP); false negatives (FN) for HbS, HbC, and HbA; and information to answer the questions in the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS‐2 (Whiting 2011)). We will construct a 2 x 2 diagnostic contingency table for each index test (see below). We will contact authors of studies for missing information.

Assessment of methodological quality

We will use the QUADAS‐2 tool to assess the quality of the included studies with respect to the risk of bias and their applicability. Briefly, QUADAS‐2 defines four domains: participant selection; index test; reference standard; and flow and timing. We will assess the risk of bias for each of these domains using the signalling questions, and a question to judge applicability of participant selection, index test, and reference standard. For each domain, the signalling questions guide the risk of bias assessment to low risk, high risk, or unclear, depending on whether the responses are yes, no, or unclear. If the answers to all signalling questions for a domain are yes, then the risk of bias is low. If the answer to any question is no, then a potential risk of bias exists (Whiting 2011).

The adaption of the QUADAS‐2 in the review is informed by the inclusion and exclusion criteria in terms of the participants, index tests, and reference standards, as set out in the table below.

Statistical analysis and data synthesis

We will extract and analyse data using Review Manager 5 (RevMan 5) software (Review Manager 2020). We will also compare the results of each diagnostic POCT with the reference standard. We will construct a 2 x 2 table for each test and hb in the usual way.

1. TP ‒ true positive, i.e. the number of participants in a study with a positive result from the index test of any of the conditions of SCD, confirmed by a positive result from the reference standard

2. FN ‒ false negative, i.e. the number of participants in a study with a negative result from the index test of any of the conditions of SCD, but positive result from the reference standard

3. FP ‒ false positive, i.e. the number of participants in a study with a positive result from the index test of any of the conditions of SCD, but a negative result from the reference standard

4. TN ‒ true negative, the number of participants in a study with a negative result from the index test of any of the conditions of SCD, confirmed by a negative result from the reference standard

From these we can compute:

1. sensitivity (sn) ‒ TP/(TP + FN);

2. specificity (sp) ‒ TN/(FP + TN);

3. positive predictive value (PPV) ‒ TP/(TP/FP);

4. negative predictive value (NPV) ‒ TN/(TN + FN);

5. positive likelihood ratio ‒ sensitivity/(1 ‐ specificity); and

6. negative likelihood ratio ‒ (1 ‐ sensitivity)/specificity.

We will extract the TP, FP, FN, and TN from each study for our data set. If a study does not have sufficient data to construct the 2 x 2 table, but provides the sensitivity and specificity, we will back‐calculate the data for each cell.

For each test, we will display study‐specific sensitivities and specificities and their corresponding 95% confidence intervals (CIs) in paired forest plots. We will also display study‐specific sensitivities and sensitivities in the summary receiver operator characteristic (ROC), using different symbols or colours where applicable. The size of points will be scaled proportionally to the precision of sensitivities and specificities. We anticipate performing a diagnostic meta‐analysis to determine the average sensitivity and specificity. We will assess the heterogeneity of the studies, assessing the thresholds of the index tests and differences in the reference standards. We will generate a summary ROC curve using bivariate models to estimate the joint sensitivity and specificity with 95% CI, because there are several POCTs with various techniques and thresholds. We will perform the analysis of the bivariate model in Stata (Stata 2017), using generalised linear mixed models through the metandi approach (Chu 2006). We will use the bivariate model to adjust for covariates, such as settings and symptomatic versus asymptomatic individuals. We will import the results into RevMan 5 to generate the summary curve (Review Manager 2020) .

The different methods for the index tests will include paper‐based haemoglobin solubility, lateral flow assay, density‐based, micro‐engineering, or smartphone technology. We will determine the index tests' ROC curves and area under the curve (AUC) for the different graphs. We will also conduct bivariate models with random‐effects models for the index test methods extended to differences in their variances.

We will determine the sensitivities and specificities in different settings, i.e. controlled and field settings. We define controlled settings as those in which the index test is performed in a laboratory environment, or if the test is conducted by trained laboratory personnel; field settings will include clinic, community, or bedside environments. We will also assess whether the participants were symptomatic or asymptomatic before the test.

Investigations of heterogeneity

We will explore heterogeneity by visually inspecting the paired forest plots of sensitivities and specificities and the colour‐coded summary ROC (SROC). We will explore sources of heterogeneity using a bivariate model and adding covariates, such as the type of SCD, because we anticipate differences in accuracy of identifying these conditions. Other covariates will include the different levels of HbF, which affect the accuracy of tests for SCD (Edoh 2006; Hebbel 2018). We will assess performance of the tests and note if they were performed by a non‐laboratory health professional, such as nurse, medical doctor, or community health worker. The use of POCT might differ according to the expertise of the tester, as tests in non‐laboratory settings are more likely to be used by non‐laboratory health professionals.

Sensitivity analyses

We will conduct sensitivity analyses by alternatively including and removing studies judged to have either a high or unclear risk of bias in any of the QUADAS‐2 domains assessed: participant selection, index test, reference standard, and flow and timing. We will compare the results at each step to a model in which all the studies are included, to assess the robustness of the test accuracy estimates. We will keep studies with a low risk of bias in any of the domains in the model at all iterations.

Assessment of reporting bias

There are currently no recommended methods for assessing reporting bias in Cochrane Reviews of diagnostic test accuracy.

Summary of findings of the review

We will summarise the results for each index test in separate tables. We will summarise the participants or population, prior testing, settings, index test and method, reference standard, and target condition. We will also report the numbers of true‐positives, true‐negatives, false‐positives, and false‐negatives per 1000 tested (Bossuyt 2013). Since GRADE for Diagnostic Test Accuracy reviews is still under development (Gopalakrishna 2014), we will consider the following to assess the strength of the evidence.

1. Precision of study estimates

2. Heterogeneity in study findings

3. Risk of bias

4. Concerns about applicability

5. Indirect test comparisons

These issues cover the main GRADE domains (except publication bias).