Background

Cellulitis is the most common presenting dermatologic diagnosis in the ED that results in admission [1], with an estimated 2.9 million cellulitis related ED visits in 2013 alone [2]. Diagnosis relies on the interpretation of subjective (erythema/edema) and non-specific (fever, leukocytosis, tenderness) physical exam findings and patient reported symptoms, all of which can be present in a number of clinical mimickers, commonly referred to as pseudocellulitis [3, 4] These ambiguities result in high misdiagnosis rates (28–74.3%) by primary teams [4,5,6,7,8,9] compared to in-person dermatology consults, which are costly to payers and harmful to patients [4, 5, 10, 11]. With no test or data point serving as gold-standard, in-person dermatology diagnosis remains the current gold-standard for cases of presumed cellulitis and is often employed by primary teams [5, 6, 11]. Access to in-person dermatology consults, however, remains limited in scope, and teledermatology has been studied as a means to fill these gaps in access [12,13,14,15,16,17,18].

Teledermatology augmented with the use of standardized questionnaires and additional imaging has been previously investigated in the setting of pigmented lesions, but never for cellulitis [19, 20]. Additional diagnostic tools, such as the ALT-70 Score for Cellulitis and thermal imaging devices—which measure temperature gradients between affected and unaffected skin—have been developed to help physicians differentiate between cellulitis and pseudocellulitis cases [21]. Investigations of thermal imaging have been particularly promising in patients with unilateral lower extremity cellulitis, where minute temperature differentials less than < 0.5 °C have been shown to differentiate cellulitis and pseudocellulitis with strong sensitivity [21]. However, while these predictive systems have been investigated for in-person cellulitis diagnosis, they have not been studied in the teledermatology setting [22,23,24], where cellulitis has been found to be the most common presumed diagnosis by consulting teams [18].

In this study, we sought to determine the impact of a standardized questionnaire and thermal imaging on the teledermatology diagnostic accuracy and confidence for cases of presumed cellulitis among a cohort of dermatologists with variable experience with inpatient consults.

Materials and methods

This multi-centered, survey-based study consisted of ten randomly ordered real-life ED cases which were initially diagnosed as cellulitis by the primary team. Each of these cases was evaluated in-person by a board-certified dermatologist, who categorized them as either cellulitis or pseudocellulitis. This result was confirmed with 30-days follow-up to assure patient improvement and to determine if the diagnosis had been changed as an outpatient. The survey was distributed to board-certified dermatologists, and their responses were compared to the referent standard of an in-person dermatology consultation.

Survey design

Patient images and data (including thermal images following a previously described protocol) [22] were compiled and organized into a survey using Qualtrics XM [25]. Each survey contained 10 randomly ordered patient cases representing either cellulitis or pseudocellulitis. Phototypes were selected to represent a range of Fitzpatrick skin types: FST 1 (n = 3), FST 2 (n = 2), FST 3 (n = 3), FST 5 (n = 1) and FST 6 (n = 1). For each case, the physician was provided, in sequential order, (1) a detailed patient history and physical mirroring the initial consultation request from the ED, (2) a cellulitis questionnaire filled out by an attending dermatologist (Table S1), and (3) skin thermal images of the affected and unaffected sites (Fig. 1). The participant was asked to answer a series of 6 diagnostic and management-related questions after each set of information, as well as rate their diagnostic confidence on a 4-point Likert scale, and their responses were locked once they continued on to the next incremental set of patient information.

Fig. 1
figure 1

Diagram of information flow for each case. For each patient case presented in the survey, respondents were first shown the history and physical, followed by responses to the cellulitis questionnaire, and finally thermal imaging. At each level, respondents answered a series of questions related to diagnosis and management, and their responses were locked when progressing to the subsequent level

Full size image

Survey distribution

A link to the survey was emailed to attending dermatologists at the following academic institutions: (1) Department of Dermatology, Brigham and Women’s Hospital (Boston, MA, USA), (2) Division of Dermatology, The Ohio State University Wexner Medical Center (Columbus, OH, USA), (3) Department of Dermatology, Perelman School of Medicine (Philadelphia, PA, USA), and (4) Department of Dermatology, UCSF School of Medicine (San Francisco, CA, USA). We surveyed a convenience sample of twenty physicians selected from these institutions from December 2018 to February 2019, with an emphasis on having a mix of physicians with varying amounts of inpatient and consultative dermatology exposure. Prior experience interpreting thermal images was not required, nor was any training provided prior to initiating the survey. This project was approved by the Partners Healthcare Institutional Review Board.

Outcome measures and data analysis

The primary outcome was diagnostic accuracy among attending dermatologists with each additional layer of information. The secondary outcomes were diagnostic accuracy for cellulitis and pseudocellulitis cases individually, diagnostic confidence, and management decisions among cases of cellulitis versus cases of pseudocellulitis at each level of information. We summarized cases for each physician for descriptive statistics on diagnostic accuracy, diagnostic confidence, and management decisions. Post hoc pairwise comparisons were performed when omnibus tests were statistically significant. Statistical significance of differences by information presented and frequency of consults was assessed using mixed effects logistic regression with random intercepts to account for within-physician correlation of responses. We used a misdiagnosis rate from previous work to estimate the absolute risk reduction via telemedicine with each increasing level of information. The empirical bootstrap confidence interval was estimated from 1000 resamples from the current and control study. Analyses were performed using SAS 9.4 (SAS Institute, Cary, NC).

Results

Twenty attending dermatologists across the four academic institutions participated in the study. Seven (35%) reported conducting consults only when on-call, one (5%) reported conducting consults less than one month per year, and twelve (60%) reported conducting consults more than one month per year.

Diagnostic accuracy and confidence

The overall (i.e., cellulitis and pseudocellulitis) average diagnostic accuracy (mean ± standard error) was 84 ± 4% when provided initial history and physical (H&P), 86 ± 3% with the subsequent addition of cellulitis-specific questionnaire, and 89 ± 3% with the final addition of thermal imaging (p = 0.23, Tables 1). Correct diagnosis of patients with cellulitis increased with each additional level of information, 76 ± 6% for H&P, 82 ± 5% for adding the cellulitis questionnaire, and 88 ± 4% with thermal imaging (p = 0.049). Following significant omnibus testing, post hoc pairwise comparisons found that for cellulitis cases, compared to H&P, the odds of a correct diagnosis were higher after the cellulitis questionnaire (OR 1.44, 95% CI 0.76–2.71) and after thermal imaging (OR 2.35, 95% CI 1.19–4.65). Correct diagnosis of patients with pseudocellulitis did not change significantly with added information, starting at 97 ± 2% with only H&P, and 94 ± 3% after both addition of the cellulitis questionnaire and thermal imaging (p = 0.57).

Table 1 Provider-level accuracy and confidence by patient information presented
Full size table

When asked to rate their diagnostic confidence at each level of each patient case, the percentage of dermatologists who rated their confidence as certain (either “Completely” or “Somewhat”) increased from 72 ± 5% after H&P, to 77 ± 4% after cellulitis-specific questionnaire, and to 79 ± 4% after thermal imaging (p = 0.28, Table 1).

Comparison of dermatologists by frequency of inpatient consults

The diagnostic accuracy and confidence of dermatologists who regularly conduct inpatient consults (n = 13) was similar to those who perform consults when on call/weekends (n = 7), with no significant differences noted for neither accuracy nor confidence, regardless of the level of information provided (Table 2).

Table 2 Provider-level accuracy and confidence by frequency of inpatient dermatology consults
Full size table

Management decisions

Over 96% of dermatologists recommended continuing antibiotics for cases of true cellulitis at all three levels of information (H&P, cellulitis-questionnaire, and thermal images, Table 1). Likewise, ≥ 86% of dermatologists recommended discontinuing antibiotics for cases of pseudocellulitis at all 3 levels of information. Hospital admission was recommended for cases of cellulitis between 61 and 63% of the time, and discharge was recommended for cases of pseudocellulitis between 95 and 96% of the time, depending on the level of information provided.

Among all ten teledermatology cases, additional diagnostic labs ordered by attendings remained largely unchanged depending on the amount of clinical information provided (median difference in percentage ordered 2.0%, SD = 1.5, Table S2). The same remained true when analyzing specifically cases of cellulitis (median difference in percentage ordered 1.7%, SD = 2.0) and pseudocellulitis (median difference in percentage ordered = 2.5%, SD = 1.3). When comparing diagnostic lab ordering trends between cellulitis and pseudocellulitis cases, the largest difference was seen for CBC (mean percentage ordered for cellulitis cases = 86–87%, mean percentage ordered for pseudocellulitis = 21–25%). “No additional workup” was chosen on average by 57–60% of attendings for pseudocellulitis cases, while only 6–9% chose this option for cellulitis cases. Decision to biopsy was largely similar between the two diagnoses, with 4–6% and 4–5% of attendings ordering for cases of cellulitis and pseudocellulitis, respectively.

Comparison with in-person diagnosis of non-dermatologists

When comparing the cellulitis misdiagnosis rates of the dermatologists in our study with the in-person misdiagnosis rate of non-dermatology providers from previously published data [25], the absolute risk reduction for a teledermatology consultation was 5.6% (95% CI − 6.6 to 17.0%) when only given initial H&P, 11.6% (95% CI − 0.6 to 22.2%) with addition of cellulitis questionnaire, and 17.6% (95% CI 6.7 to 29.8%) with the final thermal imaging (Table 3). After addition of both the questionnaire and thermal imaging, the number needed to treat (NNT) was 5.7 (95% CI 3.4 to 14.9).

Table 3 Number needed to treat calculation for teledermatology consult of cellulitis
Full size table

Discussion

These findings suggest that teledermatology can be used by attending dermatologists to confidently and accurately diagnose both cases of cellulitis and pseudocellulitis, corroborating previous findings [17, 18]. Our results expand existing knowledge by demonstrating that augmented teledermatology with additional data from a standardized cellulitis questionnaire and thermal imaging can further improve diagnostic accuracy for cellulitis specifically, while pseudocellulitis diagnostic accuracy remained ≥ 94% regardless of the amount of clinical data. Decisions considered essential to patient safety (i.e., continuing antibiotics for cases of true cellulitis) were consistently appropriate, and these, as well as all other management decisions, remained largely stable despite access to increasing amounts of patient data.

These results contribute to growing data related to not only the utility of teledermatology, but also the structure of teledermatology. Augmented teledermatology with standardized questionnaires [19] and additional technology [20] has been studied in the setting of cutaneous oncology, but not for presumed cellulitis [17, 18]. We found that augmentation of standard store-and-forward teledermatology, with the addition of a standardized cellulitis questionnaire and thermal imaging, may improve accuracy for cases of cellulitis (OR 2.35, 95% CI 1.19–4.65), and that this increased accuracy will benefit patients who can’t receive in-person dermatology consultation (NNT 5.7 (95% CI 3.4–14.9).

While the cellulitis questionnaire is free and easy to interpret by any admitting clinician, thermal imaging requires monetary investment, effort, and training on the part of the primary treatment team. Furthermore, we provided thermal image data to teledermatologists without specific training, and in our study, it was unknown what level of experience, if any, participating dermatologists had in interpreting thermal images. While the potential benefits of thermal imaging include increased diagnostic accuracy of cellulitis as demonstrated in this and previous studies, thermal imaging has also been shown to underperform relative to the ALT-70 predictive models [22]. Thus, the overall cost–benefit ratio of thermal imaging must, therefore, be evaluated further, as well as comparing the impact it has on diagnosis between dermatologists with and without previous experience in interpreting thermal imaging data.

Attending dermatologists were able to accurately diagnose regardless of their experience with inpatient dermatology consultations, reducing concerns that accuracy via teledermatology would be limited to attendings with extensive experience with inpatient cellulitis diagnosis [17]. This finding increases the pool of dermatologists who could potentially aid in the management of presumed cellulitis at smaller, non-academic hospitals that often lack inpatient consult services and disproportionately admit patients presenting with skin disease [15, 26].

Decisions around laboratory testing remained largely stable for cases of cellulitis and pseudocellulitis despite increasing amounts of clinical information. Importantly, the decision to continue antibiotics for cases of true cellulitis consistently remained above 95%, higher than the percentage of dermatologists who accurately diagnosed those cases as cellulitis. Paradoxically, the recommendation to discontinue antibiotics for cases of pseudocellulitis decreased from 93% with initial H&P to 86% with the addition of standardized questions and thermal images. While still consistent with previously reported data on antibiotic discontinuation for cases of pseudocellulitis diagnosed via teledermatology (87%) [18], this finding suggests worsening antibiotic stewardship in scenarios with greater information transfer.

Decisions regarding admission or discharge are difficult to interpret, as there are no consensus guidelines. Factors external to what was presented in our cases may guide clinical choices and the data presented in these cases was not sufficiently comprehensive to include all factors that a physician may consider when making admission decisions. That said, the decision to recommend discharge for cases of pseudocellulitis was consistently ≥ 95% among our cohort of teledermatologists, similar to previous findings among patients that received in-person dermatology consultation [4]. For cellulitis, fewer of our study’s teledermatologists recommended admission compared to dermatologists who consulted in-person [5]. This may be related to this study’s inability to recreate the same risk threshold dermatologists would encounter when making real-time decisions for patients directly under their care. Future studies should analyze trends in admission recommendations for cases of true cellulitis that received teledermatology as the primary dermatology consultation.

The disconnection between improving diagnostic accuracy and confidence with additional patient information and a lack of change in testing and evaluation suggests that the latter may be a function of intrinsic clinician preference and practice habits. Standardizing approaches and reducing unnecessary testing may be better addressed by point-of-care decision support that helps physicians link the utility of their requested tests in the specific clinical scenario they are evaluating.

Our findings regarding diagnostic accuracy and antibiotic stewardship with augmented teledermatology are similar to previous studies investigating in-person dermatology consultation [4, 8]. These improved management decisions, when considered along with the incidence of cellulitis and the cost of US hospital admissions, has the potential for significant economic and patient safety outcomes [10]. These findings are reinforced in the context of the current COVID-19 pandemic, during which teledermatology has been expanded to limit the number of person-to-person contacts and decrease personal protective equipment use [27].

These results must be considered within the context of our study design. The official diagnosis for each case was determined by in-person dermatology consultation, and while each case had 30 days of follow-up, the diagnosis may not represent the true diagnosis. Although all survey cases were real clinical cases presenting to the ED, the attendings who took our survey were not actively treating these patients, and their risk threshold and decision making may change in real life settings. All survey cases initially presented to the emergency department of a single academic institution, a setting for which improved diagnostic accuracy would be useful given the high incidence of cellulitis presenting to the ED [1]. All respondents were members of academic departments, and future studies should include dermatologists practicing in additional settings.

Conclusions

In summary, teledermatology is an effective means for dermatologists to accurately diagnose cellulitis and form patient management and disposition recommendations, and accuracy specifically may benefit from additional data via a cellulitis questionnaire and thermal imaging. Data from real-life encounters are necessary to determine the longitudinal impact on patient outcomes and the effect of these interventions on physician workflow and costs.