The development of a protocol for diagnosing hand dermatitis from photographic images

A hand photography protocol was needed to ascertain the presence and severity of dermatitis in a trial testing the effectiveness of a behaviour change intervention to prevent hand dermatitis in nurses.

assessment of whether hand dermatitis was present or not, as this was the primary outcome of the trial. In addition, we required a method that would reliably distinguish dermatitis towards the milder end of the spectrum.
The purpose of this article is to describe the 3 distinct stages that we employed in developing a new hand photography protocol for the Skin Care Intervention in Nurses (SCIN) trial in the United Kingdom.
This new protocol offers a method for diagnosing hand dermatitis and its severity that relies on dermatologist and research nurse inspection of hand photographs from research participants (in lieu of physical examinations), with comparisons then being made from standardized images contained in the Coeraands et al photographic guide. 11 The stages include: (1) developing a standardized procedure for hand photography; (2) a stepwise validation process of rules for the study dermatologists to diagnose and determine the severity of the hand dermatitis; and (3) training by a dermatologists of a research nurse to screen out hand photographs of study participants without dermatitis ("clear cases"). In developing the new method, we had several requirements: 1. The method had to measure the presence or absence of hand dermatitis as well as severity.
2. The method could not involve physical examination of the participants, as that would be logistically very difficult, expensive, and likely to result in poor response rates. 3. The method had to be objective and not based on self-reporting, as self-reporting tends to result in over-reporting of hand dermatitis.
4. The severity scale needed to be able to distinguish dermatitis towards the milder end of the disease spectrum.

| Study background
The SCIN trial is a national multicentre cluster randomized controlled trial examining the effectiveness of a complex intervention to reduce the prevalence and incidence of hand dermatitis in at-risk nurses working in the NHS in the United Kingdom. 1  The primary outcome measure was the difference in the point prevalence of hand dermatitis between participants in the intervention arm and the control arm of the trial from baseline (T1) to 12 months (T2) on photographs assessed by the 2 study dermatologists.
2.2 | Development of the hand photography procedure and fieldworker training (Stage 1) In collaboration with a medical photographer, we developed a detailed hand photography procedure to standardize the collection, screening and assessment of hand photographs. This provided fieldworkers with step-by-step instructions on setting up and using high-resolution digital SLR cameras for taking the hand photographs from each participant ( Figure S1). A flexible grey/white photographic exposure card was used as a background screen when the photographs were taken.
The hand photography procedure required fieldworkers to check the correct settings of camera set-up functions, that the camera flash was switched on, and that a minimum distance (75 cm) of the camera from the participants' hands was maintained. 11 Before the trial started, we trained fieldworkers in the use of the photography protocol, including practical photography demonstrations. During the follow-up period, we also provided participants with an opportunity to take hand photographs on their smart phones and send them to the research team via email. Specific instructions on how to take and send in hand selfie photographs were sent to participants, and these were based on key aspects of the main photography protocol. Agreement/disagreement on the severity of hand dermatitis was not assessed during the validation process, as we realized early on that the likelihood of our 2 dermatologists agreeing on the severity grading (5 grades) at 4 different sites was likely to be poor, and that perfect agreement according to each site was not necessary for our study, which sought to establish a global estimate of hand dermatitis severity. We took the pragmatic view that each participant's overall severity of hand dermatitis would be defined as the most severe combined score from both dermatologists on the Coenraads et al scale from their 4 hand photographs. Agreement between the 2 dermatologists on the binary rating (Yes/No) was assessed by the use of Cohen's kappa statistic.
In a prior feasibility study before the setting of agreement rules of diagnosing dermatitis between the same dermatologists, we found moderate (κ = 0.5) interobserver agreement in the assessment of photographs. This was mainly attributable to disagreement on the threshold of very mild vs no dermatitis. The study dermatologists therefore established rules for undertaking the assessments in the main study.
To complete this task, we undertook the following stepwise validation process. The study dermatologists were provided with hand photographs from an initial sample of 70 cases (study 1) from the main study population to independently assess for dermatitis, followed by a further enriched sample of 71 cases (study 2) with a high percentage of dermatitis cases (as identified by the chief investigator). To minimize bias, we ensured that the study dermatologists remained blinded to any other participant information, such as self-reported information in the questionnaires or each other's independent assessment outcomes. The study dermatologists independently scored the hand photographs by using the photographic assessment guide developed by Coenraads et al. 11 Discordant cases were then identified by the central trial team and sent back to the study dermatologists, who remained blinded to other information about the participants for their follow-up joint assessment. Both dermatologists looked at the discordant cases together, and explained why one or the other had decided that the participant had some degree of hand dermatitis. Very often, these discordant cases were very difficult to judge, so a set of rules were developed; these are shown in Table S1. The study dermatologists met and jointly refined these "mini-rules" for deciding whether a case met the criteria for dermatitis. This validation process was repeated again (study 3). A final arbitrator (an independent dermatologist) was available for consultation in circumstances where the study dermatologists were unable to agree. The intraobserver error was calculated to determine the degree of error in the dermatologist assessments. Figure 1 outlines the flowchart for assessing hand photographs.

| Dermatology research nurse training (Stage 3)
Because of the large number of hand photographs collected during the trial, we appointed a dermatitis research nurse to screen out all of the photographs in which no dermatitis was evident. This reduced the time spent by the dermatologists, as they assessed only those images for which the dermatology research nurse was unsure or sure that dermatitis was present. One of the study dermatologists provided the nurse with 2-hour training sessions, including the following assessment principles: (1) a quick look for abnormal erythema (or surface changes), by using pattern recognition skills; (2) if suspicious areas were identified, images were enlarged to lifesize (but not beyond) to determine whether the abnormality was dermatitis (poorly defined erythema with surface change such as scaling, lichenification, or vesicles); and (3) if the research nurse ruled out evidence of dermatitis on first inspection, a final inspection was carried out by the research nurse on high-risk areas such as fingers, interdigital webspaces, or around rings if worn, and easily missed areas such as the wrist. We ensured that the dermatology research nurse was also aware of the agreed rules that the study dermatologists would adhere to during their own assessment process.
To ensure that the screening by the dermatology research nurse had high specificity, we conducted a subgroup reliability analysis. A subsample of 250 cases (images of the dorsum of the right hand only) from the main study population that were initially assessed by the dermatology research nurse as "clear" (no dermatitis) were sent to one of the study dermatologists for assessment (study 4), as this is the area where occupational hand dermatitis is most likely to be seen.

| Procedure for taking hand photographs (Stage 1)
We trained 97 local fieldworkers from 35 participating sites in the use of the hand photography protocol. To differentiate the specific time points at which the hand photographs were taken ("recruitment", T = 0 months; or "follow-up", T = 12 months), we used specific photographic label cards containing unique sequence codes, to which the dermatologists and research nurse were blinded. We sent regular reminders to fieldworkers to ensure that the correct label cards were being used during the follow-up period. Moreover, it became evident, following the recruitment period, that fieldworkers occasionally forgot to use the camera flash when taking hand photographs. This meant that there were a number of sets of hand photographs (n = 10) that could not be included in the final dataset, owing to the difficulties in conducting a reliable assessment, because of their poor image quality.

| Establishing agreed assessment rules for diagnosing hand dermatitis and for ascertaining dermatitis severity (Stage 2)
From the initial sample of 70 sets of hand photographs from the main study sent to the study dermatologists for independent assessment as part of our validation process (study 1), we found they agreed on 66 of 70 (94%) cases and disagreed on 4 of 70 (6%) (κ = 0.30). The proportion of agreements vs disagreements from the follow-up enriched sample of 71 sets of hand photographs sent to the study dermatologists for independent assessment (study 2) is shown in Table 1 (κ = −0.14). After joint discussion, the study dermatologists agreed on all 29 cases that they had previously disagreed on.
Of the additional 100 photographs from the main trial that were sent to the study dermatologists for their independent assessment as part of our final validation process (study 3), a further 12 (12%) discordant cases required joint deliberation. Following discussion, the study dermatologists agreed on all of the 12 cases. The final arbitrator was not used during the development of the photography protocol or during the main trial. This stepwise validation procedure allowed the study dermatologists to further refine their rules for diagnosing hand dermatitis until the interobserver agreement exceeded a κ score of 0.60. A full list of the mini-rules is shown in Table S1.
The joint review of discordant cases showed that one of the dermatologists had a lower threshold for diagnosing dermatitis than the other dermatologist. In particular, one of them was more likely to grade dryness as meeting the criteria for dermatitis. Therefore, the study dermatologists agreed to exclude very borderline cases of noninflamed dermatitis as not meeting the criteria for dermatitis. Agreement/disagreement on the severity of hand dermatitis was not assessed during the validation process. Table 2 shows the results from the intraobserver assessment of the 71 cases that were randomly selected from the baseline database and were reassessed by the dermatology research nurse, and of the 53 cases that were randomly selected from the baseline database and were reassessed by the study dermatologists. Figure 2 is an example that shows early signs of hand dermatitis on which both study dermatologists agreed during their independent assessment; Figure 3 shows a moderate case of dermatitis; and Figure 4 shows dry and crinkly skin that was assessed as "clear".

| Dermatology research nurse training (Stage 3)
From the subgroup analysis of the 250 cases (images of the dorsum of the right hand only) that were screened by the nurse and categorized as    "clear," the study dermatologists found that 2 "positive" (0.8%) cases of hand dermatitis had potentially been missed (study 4). The study dermatologists suggested that both cases could be considered to be possible cases of dermatitis, because one image showed dermatitis on the right lateral surface of the right thumb (ie, not the back of the right hand, which was the primary site for the subgroup analysis), and the other showed dermatitis on the right index finger, although the photograph was underexposed and was difficult to interpret.

| DISCUSSION
We developed a novel and practical photography protocol that was suitable for use in a large-scale multicentre research trial examining hand dermatitis prevention in nurses. The hand photography procedure provided a useful instructional guide to promote the standardization of hand photography for later diagnostic assessment. During the stepwise validation procedure, we gained a number of important insights into the complexities of the independent assessment process, which required careful deliberation and refinement. This played an important role in formulating an agreed list of assessment rules to use as a reference guide during the study. We found that hand photographs taken by trained field workers using high-resolution digital SLR cameras provided a practical method for collecting the data on the presence or absence of dermatitis in participants who were geographically dispersed across the United Kingdom. We successfully trained a dermatology research nurse to competently prescreen hand photographs as "clear" (no dermatitis), "positive" (present dermatitis), or "not sure," thereby reducing the assessment burden on the study dermatologists.
The use of a broad range of hand photographs, showing varying degrees of asymptomatic and symptomatic dermatitis, played an important role during the dermatology research nurse training sessions.
An important observation from our study is that high-quality photographic images of hands will always show small areas of scaling, erythema and surface changes that could be deemed to be very early signs of hand dermatitis. This observation reinforces the view that hand dermatitis is a continuum from surface damage to frank dermatitis with cardinal signs such as lichenification and vesicles. Furthermore, we found that agreement between the dermatologists on moderate or severe cases was very good, whereas obtaining agreement on the gradation between very mild and simply dry "overwashed" hands is more difficult, and therefore to be expected. To address this issue, we incorporated a joint assessment procedure and mini-rules that the study dermatologists followed when assessing borderline cases, to minimize the risk of misdiagnosis. Such an approach will always be needed in population-based (as opposed to clinicbased) studies, in which the threshold for diagnosing disease is blurred and difficult to assess.