Diagnostic Accuracy of Pediatric Teledermatology Using Parent-Submitted Photographs: A Randomized Clinical Trial

Key Points

Question  Can parent-provided smartphone photographs be used for the diagnosis of pediatric skin conditions?

Findings  In this randomized clinical trial of 40 patient-parent dyads, overall concordance between photograph-based vs in-person diagnosis was 83%. Concordance was 89% in a subgroup of 37 cases with photographs considered of high enough quality to make a diagnosis.

Meaning  Parent-provided smartphone photographs are typically of sufficient quality to permit accurate diagnosis of pediatric skin conditions.

Importance  Advances in smartphone photography (both quality and image transmission) may improve access to care via direct parent-to-clinician telemedicine. However, the accuracy of diagnoses that are reliant on parent-provided photographs has not been formally compared with diagnoses made in person.

Objective  To assess whether smartphone photographs of pediatric skin conditions taken by parents are of sufficient quality to permit accurate diagnosis.

Design, Setting, and Participants  A prospective study was conducted among 40 patient-parent dyads at a pediatric dermatology clinic at the Children’s Hospital of Philadelphia from March 1 to September 30, 2016, to assess concordance between diagnoses made by an independent pediatric dermatologist based on in-person examination and those based on parental photographs. Half of the patient-parent dyads were randomized for a secondary analysis to receive instructions on how best to take photographs with smartphones. Clinicians were blinded to whether parents had received photography instructions.

Exposures  Half of the patient-parent dyads received a simple, 3-step instruction sheet on how best to take photographs using a smartphone (intervention group); the other half did not (control group).

Main Outcomes and Measures  Concordance between photograph-based vs in-person diagnosis in the intervention vs control groups, as quantified using Cohen κ, a measure of interrater agreement that takes into account the possibility of agreement occurring by chance.

Results  Among the 40 patient-parent dyads (22 female children and 18 male children; mean [SD] age, 6.96 [5.23] years), overall concordance between photograph-based vs in-person diagnosis was 83% (95% CI, 71%-94%; κ = 0.81). Diagnostic concordance was 89% (95% CI, 75%-97%; κ = 0.88) in a subgroup of 37 participants with photographs considered of high enough quality to make a diagnosis. No statistically significant effect of photography instructions on concordance was detected (group that received instructions, 85%; group that did not receive instructions, 80%; P = .68). In cases of diagnostic disagreement, appropriate follow-up was suggested.

Conclusions and Relevance  Parent-operated smartphone photography can accurately be used as a method to provide pediatric dermatologic care.

Trial Registration  clinicaltrials.gov Identifier: NCT03246945

Pediatric dermatologists are in short supply, with fewer than 300 board-certified physicians serving the nearly 75 million children in the United States. Furthermore, skin conditions comprise 10% to 30% of nearly 200 million annual pediatric outpatient visits.^1-3 Although many skin conditions can be handled without input from a subspecialist, one survey revealed that more than 80% of primary care pediatricians reported that there were too few pediatric dermatologists to meet the needs of patients in their practices.⁴ This deficit of subspecialists results in barriers to accessing care.

In adults, store-and-forward teledermatology has been shown to improve access to specialty care, consistently provide accurate diagnosis, and reduce time to treatment, while achieving high patient satisfaction and cost-effectiveness.^5-7 To evaluate pediatric clinician-to-clinician teledermatology, Heffner et al⁸ compared pediatric dermatologists’ diagnoses made in person vs using photographs and history taken by a pediatrician and found an 82% concordance rate. Similarly, in a retrospective cohort study, Philp et al⁹ found that pediatric dermatologists were able to render a diagnosis via photography in 75% of cases.

Enhancements in smartphones (eg, optics, storage, encryption, image processing, and data transmission) have made photographs submitted by parents increasingly viable for teledermatology diagnosis.^10,11 For example, a 2012 study assessing infantile hemangiomas showed that more than 85% of parent photographs were of sufficiently high quality.¹² The availability of this technology has also greatly increased, with nearly three-quarters of all adults, and more than 90% of adults younger than 35 years, now owning a smartphone.¹³ Pediatric dermatologists also appear to be highly interested in teledermatology.¹⁰

Despite this interest, the accuracy of teledermatology has not been evaluated specifically for the pediatric population using photographs taken by parents, to our knowledge. Thus, our study’s objective was to evaluate the quality of parent photography and the concordance between diagnoses made from in-person and photograph-based examinations. Furthermore, we explored whether a simple instruction sheet provided to parents would improve image quality and diagnostic accuracy.

Inclusion criteria consisted of patients being younger than 18 years; new patients or urgent visits in a pediatric dermatology practice; and parent with English fluency, smartphone ownership, and ability to download the MyChart application (available on Android and Apple devices; Epic Systems Corporation). The study protocol in Supplement 1 (NCT03246945) was approved by the Children’s Hospital of Philadelphia institutional review board. Parents (or legal guardians) provided written consent for both themselves and their children. Written patient assent was obtained when appropriate.

For the primary aim, we used a prospective cohort design to assess concordance between diagnoses based on in-person examinations and those based on parental photographs. After enrollment, the parent took photographs of the child’s skin condition in the clinic examination room using a personal smartphone. If the child presented for more than 1 skin condition, the parent was instructed to choose the primary condition. Images were uploaded into the patient’s electronic health record securely using the MyChart application, which the parent downloaded to the smartphone. The parent then completed a single-page survey to gather basic information about the child (age, sex, and medical history), the child’s skin condition (location, duration, and associated symptoms), and willingness to use teledermatology. Race and ethnicity were self-reported. Subsequently, all children were seen in person by 1 of 2 possible physicians (M.J.P./L.A.C.-S.) for their scheduled visit. Given variation in photograph quality (lighting, number of pictures taken, and patient compliance), a simple, 3-step instruction sheet on smartphone photography was developed (eTable in Supplement 2).

For the secondary aim, we embedded a 2-arm, parallel-group randomized trial within the cohort to assess the effect of the photography instructions on the in-person photograph-based vs examination-based diagnoses. On enrollment, the patient-parent dyad was randomly assigned to the study arm (instruction sheet provided) or the control arm (no instruction sheet provided). Patients were randomized in a 1:1 manner using simple, sequential allocation without stratification. Generation of the allocation sequence occurred before enrollment. A clinical investigator assigned participants using this sequence during participant enrollment. We estimated that a sample size of 40 patients would yield 80% power to detect a difference of 1 point in the photograph quality rating scale (PQRS) score between the 2 groups, assuming α = .05, and normally distributed data with a variance of 2.5 for both groups.

All photographs were evaluated using the PQRS based on the following 5 criteria: clarity, perspective, darkness, brightness, and color (eFigure in Supplement 2). Each criterion was rated on an integer scale from 0 to 2, yielding a total score ranging from 0 (lowest quality) to 10 (highest quality). The teledermatology physician was given only the photographs and basic survey information and was asked to make a diagnosis. Diagnoses were considered concordant if the same diagnosis was provided by the in-person physician and the teledermatologist and considered discordant if a different diagnosis was provided by the 2 physicians.

Baseline and demographic characteristics were summarized by standard descriptive summaries (eg, means and SDs for continuous variables, such as age, and percentages for categorical variables, such as sex). The Mann-Whitney test was used to compare the image quality and the number of images taken between groups. Rates of concordance were expressed using percentages and the Cohen κ coefficient and were compared between groups using the χ² test. Logistic regression was used to determine whether sample characteristics influenced concordance, and multilevel mixed-effects ordered logistic regression was used to determine whether sample characteristics influenced image quality. Individual images (level 1) were nested in cases (level 2) for the multilevel image quality model. Sample characteristics (fixed effects) in these models included patient age, patient sex, smartphone age, smartphone operating system, whether photographic instructions were provided, diagnostic category, and number of photographs taken. Mean PQRS score and number of photos taken were also included as fixed effects in the concordance models. Analyses were performed using Stata, version 13.1 (StataCorp LP). P < .05 (2-sided) was considered statistically significant.

Forty-eight patient-parent dyads were invited to participate in this study, with 40 completing the study protocol between March 1 and September 30, 2016 (Figure). New enrollment was ended when target numbers were reached. Failure to complete the protocol occurred with technical issues (eg, inability to download the application or upload photographs) and time constraints. Twenty dyads were randomly assigned to the study arm (instruction sheet provided) and 20 dyads were assigned to the control arm (no instruction sheet provided). Our sample represented a wide range of ages (mean [SD] age, 6.96 [5.23] years) and both sexes approximately equally (Table 1). Most parents (25 [63%]) used a version of the Apple iPhone 5 or 6. The remainder of parents used smartphones running the Android operating system, with 10 (25%) using a version of the Samsung Galaxy and 5 (13%) using LG, Google, or HTC smartphones. All smartphone cameras used by parents had at least an 8-megapixel resolution, none was more than 4 years old, and more than two-thirds were less than 2 years old. The diagnoses rendered in this study by the in-person dermatologist, along with the diagnostic category by which they were grouped, are listed in Table 2.

Our study included 87 images from the 40 patients. The median PQRS score was 9 of 10. No sample characteristics (ie, patient age, patient sex, smartphone age, or smartphone operating system) had a significant effect on image quality according to our multilevel mixed-effects ordered logistic regression model. The component of the PQRS with the highest score was color (84 images [97%] received a score of 2 on the 0-2 scale, representing not altered), while the lowest component was clarity, the criterion evaluating photograph blurriness (only 37 [43%] received a score of 2, representing in focus).

Overall concordance between photograph-based vs in-person diagnosis was 83% (95% CI, 71%-94%; κ = 0.81). Of the 40 dyads’ photographs, 3 diagnoses were unable to be provided by the remote dermatologist owing to poor photograph quality. Diagnostic concordance was 89% (95% CI, 75%-97%; κ = 0.88) in the subgroup of 37 dyads with photographs considered of high enough quality to make a diagnosis. Image quality differed significantly based on whether or not a diagnosis could be provided (mean PQRS score with diagnosis rendered, 8.9; mean score with no diagnosis, 7.0; P = .001). Diagnostic accuracy varied by diagnostic category (Table 2). For example, the concordance for birthmarks was 100%, 92% for rashes, and 64% for alopecia-related diagnoses. Of the 4 cases that were misdiagnosed, there were 3 cases of alopecia and 1 nodule (Table 3). Concordance did not vary by demographic or other sample characteristics (ie, patient age, patient sex, smartphone age, smartphone operating system, or number of photographs taken).

Half of the parents participating in the study received instructions for taking high-quality images (eTable in Supplement 2). The group with instructions had increased average image quality and mean number of images provided by each parent, although these trends did not reach statistical significance. Providing instructions significantly increased the quality of the best image submitted by each parent, a metric that takes into account both individual image quality and number of images taken (mean PQRS score with instructions, 9.5; mean score without instructions, 8.9; P = .04). No statistical difference in diagnostic concordance was seen for dyads who were provided photographic instructions (85%) compared with those in the group who did not receive instructions, including cases in which no diagnosis could be made (80%; P = .68).

Parents were asked to rate their willingness to use a teledermatology application to communicate with a pediatric dermatologist at our institution rather than wait for an in-person appointment. On a scale of 1 (not willing) to 10 (very willing), the median response was 8 and the mode was 10. Parents were also asked to select the price they would be willing to pay for a virtual visit from a list ranging from $0 to $200 in $20 increments. Most respondents (32 [80%]) were willing to pay to use the application, with a median price of $20 (range, $0-$160).

Our study shows that, in most cases, parents can take photographs of sufficient quality to allow for accurate teledermatology diagnoses of pediatric skin conditions. Our results are consistent with previous literature studying teledermatology in adults showing concordance between primary care teledermatology consultations and in-person dermatologist diagnoses.^5,14-18 Our study also demonstrated potential for use of teledermatology as triage in a pediatric setting. For example, urgent dermatology clinic follow-up was recommended in cases of infantile hemangiomas that presented on the lower lip and upper eyelid, whereas reassurance was suggested for a hemangioma that presented on the calf.

Diagnostic accuracy varied by diagnostic category. When dealing with categories with low concordance, such as alopecia and nodules and tumors, teledermatology practitioners may need to be cautious about attempting definitive diagnoses in some cases and may need to refer patients for in-person consultation. For these cases, teledermatology may still serve as a triage tool. For example, patients with suspicious nodules could be referred for expedited appointments in specialty clinics, whereas patients with isolated alopecia could be scheduled for routine visits. Conversely, in diagnostic categories with high concordance, such as birthmarks and rashes, certain cases could be definitively diagnosed and treated exclusively using teledermatology (eg, mild acne).

The greatest number of incorrect diagnoses occurred in cases of alopecia, likely because such diagnoses can be difficult to differentiate without dermoscopy and a thorough history. In 2 misdiagnosed cases of alopecia, the correct diagnosis could have been rendered with a small amount of additional history screening for trichotillosis. Unlike the standard paper intake forms used in this study, mobile applications could adapt questions based on specific diagnoses and patient input, much like the way that skilled clinicians tailor their history taking based on the particular patient, responses to previous questions, and differential diagnosis. In addition, mobile applications could integrate features, such as secure text messaging, to allow asynchronous, or even real-time, communication with patients.

Because 1 of the 2 in-person clinicians specializes in pediatric hair conditions, our study included an above-average proportion of such cases. Given the disproportionately large number of cases of alopecia, as well as the inability of the telemedicine clinician to ask additional questions or request more photos, our study likely underestimates the potential pediatric teledermatology diagnostic concordance rate. Furthermore, since patients were already being seen by a pediatric dermatologist and parents knew that the photographs taken for this study would not affect their child’s diagnoses and treatments, they may not have been as diligent with photograph quality compared with parents using a teledermatology application from home when their child’s care depends on image quality.

Although diagnostic concordance was higher in the group that received instructions, this improvement did not reach statistical significance. This trend should be investigated further in a larger study, as this one was likely underpowered to detect an effect given the low number of incorrectly diagnosed cases. Nevertheless, including similar instructions may be useful in future teledermatology services. In recent years, the utility of patient photography instructions may have become somewhat reduced owing to modern user-friendly smartphone design, which maximizes image quality regardless of user ability; this design includes standard features such as high-resolution image sensors and high-quality lenses, as well as sophisticated lighting, image stabilization, and focusing algorithms. Furthermore, while the instructions discussed optimal lighting, most parental photographs were taken in the well-lit clinic examination room. Future studies should be performed using real-life conditions to assess diagnostic concordance and other aspects of care.

Although this study simulated rather than performed an actual teledermatology encounter, our findings contribute to the large body of literature that supports the potential of teledermatology to provide quality care. Direct-to-patient pediatric teledermatology could also improve access for patients whose families face geographic, scheduling, or financial limitations¹⁹ and could reduce wait times. Our survey indicated parent enthusiasm about using a pediatric teledermatology application if one were available rather than waiting for an in-person appointment. Most parents were willing to pay an amount similar to, or in excess of, the cost of their visit copayment for the service.

We agree with recent recommendations that telemedicine is best performed by physicians who are part of a patient’s medical home or health system.^19-21 When a teledermatology diagnosis is uncertain or requires further management, there must be a system in place to facilitate appropriate follow-up with the patient’s primary clinician, dermatologist, or other specialists. Without the opportunity for in-person follow-up, patients may not receive accurate diagnoses when teledermatology is insufficient, may be sent to emergency departments, or may not receive further care at all.

Providing care integrated into local health networks also enables clinicians practicing telemedicine to view a patient’s existing medical record and readily communicate with the patient’s established medical professionals. In our study, additional history that could have been gleaned from the electronic health record may have allowed for more accurate diagnoses. Being part of a local network would also allow telemedicine clinicians to order diagnostic studies or laboratory tests and prescribe medications while preventing siloed or fragmented care by integrating directly into patient records.

Limitations of our study included the small sample size; a setting that was urban, academic, and clinic based; and use of only 1 pediatric dermatologist to provide remote diagnoses. The small sample size may have limited our ability to show an effect from providing instructions. The clinic-based setting may have altered factors such as lighting, among others. The urban academic setting and single physician providing remote diagnoses may make our findings less generalizable.

Parents can reliably take high-quality photographs of their child’s skin condition using smartphone cameras. This finding suggests that direct-to-patient pediatric teledermatology should not be limited by image quality, especially when appropriate photography instructions are provided.

Corresponding Author: Patrick J. McMahon, MD, Section of Dermatology, Division of General Pediatrics, Children’s Hospital of Philadelphia, 3550 Market St, Second Floor, Philadelphia, PA 19104 (mcmahonp@email.chop.edu).

Accepted for Publication: August 24, 2017.

Published Online: November 15, 2017. doi:10.1001/jamadermatol.2017.4280

Author Contributions: Drs O’Connor and McMahon had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: O’Connor, Winston, McMahon.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: O’Connor, Jew, McMahon.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: O’Connor, Jew, Winston.

Obtained funding: Winston.

Administrative, technical, or material support: O’Connor, Jew, Winston.

Study supervision: O’Connor, Perman, Castelo-Soccio, Winston, McMahon.

Conflict of Interest Disclosures: None reported.

Funding/Support: Funding was provided for research assistance through the Office of Entrepreneurship and Innovation at Children’s Hospital of Philadelphia.

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Jules B. Lipoff, MD, Department of Dermatology, Perelman School of Medicine, University of Pennsylvania, reviewed the manuscript. He was not compensated for his contribution.