Skin-in-Sight: a Longitudinal Cohort Infrastructure for Standardized Remote Monitoring and the Evaluation of Digital Care Innovations in Chronic Skin Diseases

Overview

Psoriasis, atopic dermatitis (eczema) and hidradenitis suppurativa are long-term skin conditions that flare and settle over time. Most of what happens to the skin happens at home, between hospital appointments. A dermatologist usually sees the skin only on the day of the visit, which may not reflect how the condition has been over the preceding months. This study sets up a long-term research collection, called a cohort, in which people with one of these three conditions take part from home. Once a month, participants use a certified digital health app on their own phone to answer a short set of questions about their symptoms and about how their skin affects daily life, and to take a standard set of photographs of their skin. Twice a year they answer a longer set of questions. Taking part takes about five minutes in most months and about twenty-five minutes twice a year. The app gives step-by-step guidance on lighting, distance and framing at the start of every photo session, so that the photographs can be compared from month to month. The photographs are reviewed by trained assessors, who score how severe the skin disease looks using the scoring systems established in dermatology research. Three assessors score each set of photographs independently, and their scores are combined according to rules that are set in advance. With the participant's permission, this information is combined with information already recorded during routine care, such as changes in treatment and hospital visits; with pharmacy records showing which medicines were dispensed; and with publicly available daily measurements of air quality, sunshine, ultraviolet radiation and pollen for the area where the participant lives. Bringing these together makes it possible to study how these conditions change over time, what happens in the period before a flare, and how treatment and the living environment relate to the course of the disease. The collected data are also intended to support the development of computer-based tools that could in future help assess skin disease from photographs. Taking part does not change the care a participant receives. The study does not provide medical advice, and the data collected are not used for diagnosis, treatment decisions or routine follow-up. Participants who experience worsening symptoms contact their treating physician as usual. The cohort is also designed so that studies of digital care tools can later be carried out within it, among participants who have agreed in advance to be approached. Each of those studies is registered separately.

Rationale. Chronic inflammatory skin diseases follow a fluctuating course, and the determinants of that course largely operate outside the clinical encounter: treatment adherence, self-management capacity, and environmental exposure to air pollution, ultraviolet radiation and pollen. These determinants are held in separate information systems and are rarely assembled for the same patient over time. Routine care records disease activity only at the moments a patient attends, and validated severity instruments such as PASI and EASI, although recommended by guidelines, are applied inconsistently in practice. Remote photographic follow-up is a plausible way to close this gap, but patient-acquired photographs vary considerably in quality, and only a modest proportion are judged clearly sufficient for clinical decision-making. This cohort therefore combines a standardized home photography procedure with a pre-specified, multiple-rater scoring procedure, and links the resulting measurements to routine care data, national pharmacy dispensing data and daily environmental exposure data. Objectives. The primary objectives are: (1) to establish a longitudinal cohort infrastructure for patients with chronic inflammatory skin diseases that permits the embedding and evaluation of multiple randomized digital care interventions using a cohort multiple randomized controlled trial design; (2) to prospectively collect standardized longitudinal data comprising serial home photographs, patient-reported outcome measures, and clinically relevant disease outcomes derived from routine care; and (3) to create a dataset suitable for the development, training and validation of models for automated or semi-automated assessment of disease severity and disease impact from photographs and patient-reported measures. The secondary objectives are: to describe disease trajectories over time, including patterns of stability and flare; to evaluate current dermatological care pathways by comparing outcomes and healthcare use between digitally supported and standard outpatient follow-up; to explore factors associated with disease worsening; to assess the feasibility, completeness and adherence of long-term home-based photograph and questionnaire collection in routine care; and to quantify the reliability of photograph-based severity scoring across raters of differing professional background, including the effect of access to the clinical record on scoring consistency. Design and setting. Prospective observational cohort at two hospital dermatology departments in the Netherlands, designed as reusable research infrastructure. No intervention is assigned at cohort level; participants receive usual care. Participants are informed by their treating physician during a routine outpatient visit, contacted subsequently by the research team, given a reflection period of at least one week, and included after electronic informed consent. Measurement schedule. Participants complete a baseline questionnaire set and a baseline photograph set at inclusion. Thereafter a fixed monthly schedule applies, at approximately day 28, identical for every participant and deliberately not responsive to symptom worsening. Each monthly session comprises a short questionnaire set and a standardized home photograph set; an extended questionnaire set is added twice yearly. Automated reminders are sent after two days and after one week, with research team contact where needed. Photograph acquisition. Standardization is achieved through disease-specific written photo protocols and in-app capture guidance on lighting, background, distance, framing and flash, presented at the start of every session. No automated image quality algorithm is applied at the point of capture. For psoriasis and atopic dermatitis, each session begins with a question about currently visible facial disease, followed by three standardized facial photographs where applicable, and by trunk, arm and leg photographs from every participant at every session. For hidradenitis suppurativa the session proceeds region by region (axillae, groin, gluteal, inframammary, perineal or genital, and other), with a presence question and, where lesions are present, counts of nodules, abscesses and draining tunnels; photography of each region is offered but optional. For every region the dataset records one of three states: photographed and scored, reported as affected but not photographed, or reported as clear, so that an absent photograph is not interpreted as absent disease. Photograph scoring. Before cohort initiation, all personnel involved in scoring, including dermatologists, clinician-researchers and trained medical students, complete a calibration session against a reference set of annotated photographs, covering the instruments applicable to their assigned disease group. Each submitted photograph set is scored independently by three calibrated raters, at least two of whom score from the photographs alone without access to the participant's clinical record at the time of scoring. Rater assignments and scoring timestamps are recorded. Where image quality prevents assessment, the affected components are recorded as missing and the reason documented in one of six predefined categories: blur or motion artefact; insufficient or uneven lighting; incorrect framing or distance; incomplete coverage of the affected area; identifiable features visible outside the consented scope; and other. Consensus is the mean of the three scores for continuous instruments and the median for ordinal staging. Sets exceeding the pre-specified discordance threshold are referred to an adjudicating rater, either the principal investigator or a designated senior dermatologist not involved in the original scoring of that set, whose score replaces the consensus label. All submitted photographs are retained regardless of assessed quality; the accumulated quality annotations are intended to serve as labelled training data for the future development of automated image quality assessment tools, which do not presently exist within the study. Linked data. Disease exacerbations, healthcare use, care setting and associated costs are derived from routine electronic health records and hospital procedural codes. Dispensing data are obtained from the national pharmacy information system. Environmental exposures are treated as exposure variables rather than outcomes and are linked to each participant by residential postal code and measurement date: air quality parameters from the national Luchtmeetnet network operated by RIVM, using the monitoring station nearest the residential postal code or, where no station lies within a defined distance, modelled national concentration grids; daily sunshine duration, global radiation, ambient temperature and ultraviolet index from the national meteorological institute, using the automatic weather station closest to the participating sites; and daily tree and grass pollen concentrations from the two national reference stations. No pollen monitoring station is located in the study region, so pollen exposure is approximated from national reference stations and modelled estimates; this is acknowledged as a limitation in pollen-related analyses. Participant characteristics. Participant characteristics are recorded at inclusion and updated at least annually. These comprise demographics (age, sex, height, weight, body mass index, family situation, educational level, employment status), clinical background and lifestyle (smoking status, allergies, comorbidities and their treatment, current medication use), environmental and exposure-related factors (childhood and current residency, pets, Fitzpatrick skin type based on self-reported burning and tanning response), and disease-related variables (diagnosis, disease duration, baseline severity), together with disease-specific characteristics such as atopic comorbidity. These variables are used descriptively and as covariates; they are not outcome measures. Analysis. Repeated measurements within individuals are analysed using linear mixed-effects models for continuous measures and appropriate generalized models for count and binary measures. Associations between environmental exposures and disease activity are explored using time-series and mixed-effects approaches, including lagged exposures where relevant; these analyses are exploratory. Because all participants share a single regional exposure series for several parameters, between-person contrasts in environmental exposure are limited and inference rests mainly on within-person temporal variation. Agreement between raters is quantified using intraclass correlation coefficients for continuous instruments and weighted kappa for ordinal staging, reported per disease population and, where numbers permit, per rater group. Participants who leave the study are compared with those who remain, using their last available measurements, to characterize any systematic differences; all participants contribute data up to the point of leaving. Sample size. No formal sample size calculation is performed at cohort level. The cohort is an infrastructure intended to support multiple analyses and embedded evaluations rather than a single predefined comparison, and the anticipated enrollment reflects feasibility and representativeness of the population treated at the participating departments. Sample size calculations for embedded interventions are performed separately and described in the corresponding intervention protocols. Cohort evaluation. A formal evaluation is conducted every two years from the date of first inclusion, by the principal investigator, the coordinating investigator, a paediatrician, an epidemiologist and a patient panel representative. It covers participant experience, data quality and completeness, attrition, progress of embedded evaluations, and continued scientific and clinical justification. Targets are pre-specified for this evaluation, assessed overall and per disease group, with photograph-related targets assessed separately for hidradenitis suppurativa given that photography is optional in that group: at least 70% average monthly questionnaire completion, at least 70% average monthly photograph set submission, at least 65% of submitted photograph sets of sufficient quality to be scored, at least 75% of quality-passed sets scored by all three raters within 90 days, no more than 20% annual attrition among active participants, and a minimum of 30 active participants per disease group at the first evaluation and 50 thereafter. These are pre-specified operational targets for the evaluation, not hypotheses and not predictions; failure to meet a target triggers a documented action plan and, for attrition, may lead to stopping or substantially modifying that part of the study. Embedded evaluations. Randomized evaluations of digital care components may be embedded using a cohort multiple randomized controlled trial design, in which eligible participants are randomly selected and invited, with outcomes compared against eligible participants not selected. The present record concerns the cohort infrastructure only. Each embedded intervention is submitted as a separate ethics application and registered as a separate interventional study referencing this record. Data retention. Research data are retained for a minimum of 15 years.