This multicenter, parallel RCT (ChiCTR2000031474) was conducted within the pediatric dermatology departments of 12 tertiary public hospitals across China, including locations in Chongqing, Beijing, Liaoning, Shanghai, Shenzhen, Zhejiang, Hunan, Xi’an, Suzhou, Tianjin, Kunming, and Guangdong (Multimedia Appendix 1). The trial adhered to the CONSORT (Consolidated Standards of Reporting Trials) reporting guidelines (Checklist 1).

Children aged 0 to 6 years diagnosed with AD according to the American Academy of Dermatology Consensus criteria were eligible for inclusion if they presented with moderate-to-severe AD (SCORAD ≥25; Investigator’s Global Assessment [IGA] ≥3) and had caregivers proficient in reading Chinese characters and using a smartphone. All participants were recruited through in-person clinical visits. Exclusion criteria included children with severe infections, psychiatric disorders, primary or secondary immune deficiencies, malignancies, or other medical conditions that significantly impair quality of life. Additionally, caregivers with mental disorders or cognitive impairment, prior participation in any patient-caregiver education program, and severe AD requiring systemic treatment were excluded.

Most caregivers were the patients’ parents, and a minority were grandparents. All caregivers lived in the same household as the child, as families of “left-behind children,” defined as children who remain in their hometowns in China for more than 6 months under the care of grandparents or other relatives while both parents migrate to urban areas for work, were excluded to avoid potential confounding due to social and environmental separation. Caregiver demographic details (eg, age and education level) were not collected, as the original study design focused on patient-level outcomes.

Eligible patients were randomly assigned in a 1:1 ratio to either the intervention group (smartphone-based digital education program plus standard care) or the control group (conventional outpatient consultation only). Randomization was conducted using a computer-generated number table with a block size of 4. The digital educational intervention protocol was developed and administered through a secure cloud-based platform operated by an independent clinical research organization (CRO) not involved in trial execution. This third-party CRO was responsible for maintaining the randomization database, delivering standardized digital educational content (including interactive modules, video tutorials, and self-assessment tools), and monitoring intervention adherence by automated adherence reminders (SMS text messaging or telephone) and digital footprint tracking (platform logins and content interaction time). Crucially, the CRO had no involvement in participant recruitment, clinical management, or outcome evaluation processes to ensure blinding integrity.

Assessments were conducted at 4, 8, 12, 24, 36, and 52 weeks following randomization. Disease severity was evaluated using SCORAD, IGA, Peak Pruritus Numerical Rating Scale (PP-NRS) for pruritus, and the Patient-Oriented Eczema Measure (POEM). Quality of life was assessed using the Dermatitis Family Impact (DFI), the CDLQI, and the Infant’s Dermatitis Quality of Life Index (IDQOL). The assessments were completed as follows: If the child was able to attend an in-person visit, the evaluations were performed by physicians not involved in the educational intervention; if not, caregivers uploaded photos and completed the PP-NRS, POEM, Quality of Life (QoL) questionnaires, and SCORAD self-assessment (Multimedia Appendix 9) on the platform. Subsequently, physicians not involved in the educational intervention performed the IGA and SCORAD assessments based on the uploaded photos and the SCORAD self-assessment form.

In addition, the timing of each relapse was meticulously recorded. Relapse was defined as an increase of ≥10 points in the SCORAD score relative to the value recorded at the conclusion of the 2-week acute-phase treatment. The primary endpoint was the relapse rate at 12 weeks. The secondary endpoints included the changes in disease severity scores (SCORAD, PP-NRS, and POEM) and quality of life scores (CDLQI/IDQOL and DFI) from week 2 (end of the acute-phase treatment) to weeks 4, 8, 12, 24, 36, and 52 in both groups.

Statistical analyses were performed using R (version 4.0.5; R Foundation for Statistical Computing). Sample size estimation was based on an expected 12-week relapse rate of 30% in the digital education group and 45% in the control group, which were derived from a multicenter randomized controlled clinical study on long-term intermittent maintenance therapy for children with AD conducted in China [31]. Assuming a 1:1 allocation ratio and a 30% expected dropout rate, an empirically supported adjustment commonly recommended in sample size estimation to maintain statistical power for both completer and intention-to-treat (ITT) analyses [32], a minimum of 229 patients per group was required to achieve 80% power with an α level of .05.

All analyses followed the ITT principle, which includes all randomized participants in the groups to which they were originally assigned. This approach maintains the integrity of randomization and provides a conservative estimate of treatment effectiveness in real-world clinical settings. To address missing data, multiple imputation by chained equations (MICEs) was performed under the assumption that the data were missing at random. Missing at random implies that the probability of missingness depends only on observed data and not on unobserved data. MICEs generate multiple plausible values for missing data based on observed variables, and we used 5 imputations (m=10) to account for uncertainty. The results from the imputed datasets were combined according to Rubin’s rules. Normality of continuous variables was evaluated using the Kolmogorov-Smirnov test, a nonparametric test that compares the empirical distribution of the data with a theoretical normal distribution to assess deviations from normality. Data with normal distributions were presented as mean (SD) and compared using independent t tests. Nonnormally distributed data were presented as median (IQR) and analyzed using the Mann-Whitney test.

For the primary outcome, the 12-week relapse rate was compared between groups using the chi-square test. Time to relapse over 100 days was visualized using Kaplan-Meier curves and compared using the log-rank test. Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% CIs, both unadjusted and adjusted for age and sex.

For continuous outcomes repeatedly measured over time (SCORAD, POEM, PP-NRS, IDQOL/CDLQI, and DFI), population-averaged generalized estimating equations were used with a Gaussian family, identity link, exchangeable correlation structure, and robust SEs to account for within-subject correlations. Each model was adjusted for baseline score, age, and sex. Missing data in these repeated measures were handled using MICEs (10 imputations), with final estimates combined using Rubin’s rules.

A 2-sided P value of .04 was considered statistically significant.

This study was reviewed and approved by the Medical Research Ethics Committee of the Children’s Hospital of Chongqing Medical University (approval No. 2019-44-1). Written informed consent was obtained from all participants and their guardians before their involvement in the study. All collected data were deidentified to protect participant privacy and confidentiality. No compensation was provided to the participants.

The multicenter trial was conducted across 12 tertiary dermatology centers in China from July 2020 to March 2021, with a detailed CONSORT flow presented in Figure 2. Of 980 screened children with moderate-to-severe AD, 615 (62.8%) met inclusion criteria and were subsequently randomized into the digital education group (n=307, 49.9%) and the control group (n=308, 50.1%). Significant differences in attrition rates were observed at 12 weeks, with 20.2% (62/307) in the digital education group and 27.6% (85/308) in the control group (95% CI 0.66‐0.98; χ²₁=4.1; P=.04). Ultimately, 48.62% (299/615) completed the 52-week follow-up, comprising the per-protocol population: digital education (n=148, 24.1%) and control (n=151, 24.6%).

Baseline characteristics demonstrated adequate randomization balance (Table 1), with no significant differences observed in gender distribution (men: 52.1% vs 50.8%; P=.74), mean age (3.2, SD 1.8, vs 3.4, SD 1.7 y; P=.21), or disease severity measures, including SCORAD (49 [40, 59] vs 47.5 [37, 55]; P=.29), PP-NRS (6 [6, 8] vs 6 [5, 8]; P=.31) and POEM (15 [11, 20] vs 15 [11, 19]; P=.50). Similarly, no significant differences were noted in quality-of-life measures, including DFI (9 [4, 16] vs 10 [5,16]; P=.77) and QoL scores (13 [11, 15] vs 14 [11, 16]; P=.11). Quality of life was assessed using the CDLQI for children and IDQOL for infants.

Intervention adherence, as measured by digital footprint tracking, demonstrated that 58.0% (178/307) of participants in the digital education group maintained regular engagement, averaging 1.64 (SD 0.38) sessions per week, and 26.7% (82/307) accessed the expedited physician consultation portals during disease flares. In the control arm, clinic attendance at the 52-week follow-up was 49.0 % (151/308). The 9.0% absolute difference in adherence between groups was statistically significant (Z=2.3, 95% CI 1.15%-16.85%; P=.03).

The ITT analysis revealed a significant reduction in relapse rates in the digital education group at 12 weeks (16.6% [51/307] vs 24.0% [74/308]; relative risk [RR] 0.69, 95% CI 0.50‐0.96; χ²₁=5.1; P=.02). However, there was no statistically significant difference between the 2 groups at other time points: 4 weeks (8.79% vs 12.99%; RR 0.68, 95% CI 0.42‐1.09; χ²₁=2.8; P=.10), 8 weeks (15.64% vs 21.43%; RR 0.73, 95% CI 0.52‐1.02; χ²₁=3.4; P=.07), 24 weeks (21.82% vs 27.60%; RR 0.79, 95% CI 0.60‐1.04; χ²₁=2.8; P=.10), 36 weeks (23.13% vs 29.22%; RR 0.79, 95% CI 0.60‐1.03; χ²₁=3.0; P=.09), and 52 weeks (24.76% vs 31.17%; RR 0.79, 95% CI 0.62‐1.03, χ²₁=3.1; P=.08).

Kaplan-Meier analysis (100-d follow-up) further supported the 12-week finding, showing a significant difference in relapse risk between groups (HR] 0.688, 95% CI 0.490‐0.966; χ²₁=4.7; P=.03; Figure 3). In a Cox proportional hazards model adjusting for age and sex, the HR remained significant (adjusted HR 0.665, 95% CI 0.469‐0.943; P=.02).

Changes in disease severity scores (including SCORAD, PP-NRS, and POEM) and QoL scores (IDQOL/CDLQI and DFI) from week 2 (the end of the acute treatment phase) were compared between the 2 groups at weeks 4, 8, 12, 24, 36, and 52. Using population-averaged GEE models adjusted for baseline score, age, and sex, the results demonstrated no statistically significant differences between the groups in the magnitude of score changes at any of the predefined follow-up time points (Multimedia Appendix 10).

This multicenter RCT demonstrated that a structured, smartphone-based multimodal digital education program for caregivers, when integrated with standard care, significantly reduced the substantial proportion of early relapses in children aged 0 to 6 years with moderate-to-severe AD, compared to conventional outpatient consultation alone. This early benefit was further supported by Kaplan-Meier survival analysis showing a clearer separation of relapse-free curves during the early follow-up period. Consistent with the study’s primary objective, these results confirm that the digital educational intervention effectively achieved short-term relapse prevention. However, differences in relapse rates appeared to diminish over time, and no notable differences were observed in secondary end points assessing disease severity or quality of life across the 52-week follow-up.

The observed reduction in early relapse aligns with the core principle of AD management, which emphasizes the critical role of effective patient and caregiver education in improving clinical outcomes [8-12]. The success of our digital program likely stems from its multimodal design, which combined frequent, brief multimedia modules with a “Real-Time Clinician Access” pathway (“Green Channel”). This combination probably enhanced caregivers’ ability to detect early signs of worsening and initiate prompt interventions, thereby preventing progression to full-blown relapse, even in the absence of measurable changes in objective severity indices. The platform’s adherence rate suggested better engagement compared with what is often observed in traditional formats [33], indicating improved feasibility and user engagement.

Our findings are consistent with prior evidence demonstrating that structured education improves short-term management and adherence in pediatric AD [16-23,34-37]. However, they also highlight a common challenge: achieving long-term disease modulation with brief interventions is difficult. The diminishing effect on relapse prevention after the initial 12 weeks may be attributed to several factors. First, the structured program builds knowledge and habits, but its impact may wane without ongoing reinforcement [38]. Second, the natural relapsing-remitting course of AD may eventually overshadow the effects of a time-limited intervention. Furthermore, despite good initial adherence, participant engagement with digital health interventions often declines over time [39].

A critical component contributing to the early success was the integrated “Real-Time Clinician Access,” used by a notable portion of participants during flares or uncertainty. This feature provided timely guidance, effectively replicating the benefit of rapid in-person intervention during exacerbations without the associated logistical burdens. This mechanism, improved early recognition and intervention facilitated by accessible education and on-demand clinical support, provides a plausible explanation for the significant reduction in early relapse rates, underscoring that the primary value of the intervention lies in optimizing the behavioral and self-management aspects of AD care.

The absence of significant differences in disease severity and QoL scores suggests that while the intervention improved caregivers’ short-term flare management, it did not fundamentally alter the underlying core inflammatory process within the assessed follow-up period. This distinction has practical implications: digital education may be most valuable as a tool to reduce short-term exacerbations and health care utilization, complementing rather than replacing ongoing medical and pharmacological strategies needed to modify underlying disease activity.

Traditional educational formats (face-to-face counseling [34-37], workshops [16-19], and eczema schools [20-23] are effective but face significant scalability challenges due to their resource-intensive nature, geographical barriers, and time constraints for families and clinicians [35]. Passive educational materials, such as leaflets and videos, are limited by a lack of interactivity and personalization [26,36,37]. Our study addresses these limitations by leveraging widespread smartphone access to deliver a scalable, multimodal intervention. The platform’s use of brief, engaging modules (animated stories, videos, and illustrated texts) delivered 3 times weekly made essential educational topics accessible to caregivers with varying literacy levels and busy schedules.

Several limitations must be acknowledged. First, smartphone proficiency was an enrollment requirement, potentially excluding digitally underserved groups (eg, from rural or socioeconomically disadvantaged backgrounds) and limiting the generalizability of our findings to urban or digitally connected families. Future implementations should consider hybrid delivery models combining digital modules with printed materials or in-person sessions to enhance inclusivity. Second, understanding of the educational materials was not formally assessed, which may influence outcomes. Third, although early retention differed between groups, a considerable number of participants did not complete long-term follow-up, potentially affecting the accuracy of long-term outcome estimates. Fourth, although outcome assessors were blinded, caregivers and clinicians were not, creating potential for performance or reporting bias. Fifth, caregiver demographic variables (eg, age, education, and occupation) were not collected, limiting our ability to explore effect modification by caregiver characteristics. While the exclusion of ‘left-behind children’ and the predominance of parent caregivers likely reduced heterogeneity in living arrangements and direct caregiving availability, unmeasured socioeconomic or educational differences could still influence the uptake and use of digital education. Finally, while the control group also had access to the platform for noneducational purposes (eg, data entry, random allocation, and scheduling visits), they did not receive the structured educational modules or ‘Green Channel’ clinician access. Therefore, the observed effect likely reflects the combined impact of the educational content and the interactive digital delivery modality, rather than platform access alone. Future trials should include comparative arms using alternative digital formats (eg, app-based platforms and interactive chatbots) to disentangle which specific features most strongly influence adherence and relapse outcomes.

In conclusion, this large multicenter RCT demonstrates that a structured, smartphone-based digital education program for caregivers, featuring interactive modules and real-time clinician support, can significantly reduce early relapse rates in young children with moderate-to-severe AD. This finding indicates that scalable digital education can strengthen short-term relapse prevention by optimizing caregiver empowerment and flare management, a strategy particularly relevant in regions with high smartphone penetration.

The broader implications of our study are 3-fold. First, given the high burden of AD and the scalability of digital tools, such interventions hold promise for improving access to quality education and potentially reducing health care disparities. Second, the lower dropout rate in the intervention group suggests that well-designed digital platforms may improve retention in long-term pediatric dermatology research and care. Finally, the transient nature of the benefit underscores that digital education should be viewed as a complementary adjunct to, not a replacement for, ongoing medical management.

Future research should prioritize (1) developing hybrid models that integrate digital education with targeted in-person support to enhance emotional connection and engagement; (2) designing strategies for sustained engagement, such as “booster” modules and adaptive content; (3) ensuring digital accessibility for underserved populations to address health equity; (4) conducting rigorous economic evaluations; and (5) conducting head-to-head comparisons of different digital features and mechanistic studies to identify the active ingredients of digital education.