This prospective, open-label, unblinded pilot randomized trial was conducted at the respiratory outpatient clinic of China-Japan Friendship Hospital, Beijing, China. Patients were recruited through physician recommendation during outpatient visits or self-referral in response to advertisements posted in the outpatient area. All potential participants underwent eligibility screening by the research team, which included review of medical records and spirometry results obtained within the previous 6 months, and were invited to an introductory meeting to review the study details and ask questions. Those who met the eligibility criteria and agreed to participate provided written informed consent, and baseline information was then collected.

At baseline, demographic and clinical characteristics, medical and smoking history, and exacerbation history of the participants were collected. All participants completed the modified Medical Research Council (mMRC), COPD Assessment Test (CAT), St George’s Respiratory Questionnaire for Chronic Obstructive Pulmonary Disease (SGRQ-C), Patient Health Questionnaire-9 (PHQ-9), Generalized Anxiety Disorder-7 (GAD-7), and Test of Adherence to Inhalers (TAI) questionnaires [18]. Trained assessors evaluated participants’ inhaler technique using the metered-dose inhaler score [19] and performed physical examination (height, weight, waist and hip circumference, blood pressure, resting pulse rate, and oxygen saturation). Follow-up assessments were conducted at weeks 4, 12, and 24. This study was reported in accordance with the CONSORT (Consolidated Standards of Reporting Trials) guidelines (Checklist 1).

Patients were eligible if they (1) fully understood and were willing to sign the written informed consent, (2) were local residents diagnosed with COPD by a physician, (3) were aged 40 years or older, (4) were prescribed a metered-dose inhaler as long-term treatment, and (5) were able to use a smartphone app after training. COPD diagnosis was confirmed during screening by reviewing medical records and spirometry results obtained within the previous 6 months, including a documented postbronchodilator FEV₁/FVC (forced expiratory volume in 1 second/forced vital capacity) <0.70.

Participants were randomized in a 2:1 ratio (intervention:control) stratified into 2 strata according to baseline CAT score (<10 vs ≥10) using a computer-generated randomization sequence. Neither researchers nor participants were blinded to group assignment.

The initial sample size was estimated based on a previous questionnaire conducted by our research group, in which 258 of 737 (35.01%) patients with COPD were classified as having good adherence. Assuming a 15% absolute increase in the proportion of patients with good adherence in the intervention group at week 24, with an α of 0.05 and 90% power, the estimated sample size was 400 participants in the intervention group and 200 in the control group, allowing for approximately 20% loss to follow-up. However, as this was a pilot study, recruitment was conducted within a predefined study period and was constrained by available funding and research personnel.

The multicomponent digital therapeutic program comprised an EIM and a remote management platform for both patients and clinicians, supporting both patient self-management and clinician oversight (Figure 1A).

The same EIM was compatible with any pressurized metered-dose inhaler (pMDI) device regardless of whether the prescribed regimen was dual bronchodilation therapy, consisting of a long-acting beta-agonist and a long-acting muscarinic antagonist, or triple therapy, consisting of a long-acting beta-agonist, a long-acting muscarinic antagonist, and inhaled corticosteroids, as it was externally attached to the pMDI. It passively recorded each inhalation event with a time stamp and transmitted the data via a cellular network to the patient’s mobile phone. The adherence data were then synchronized with the patient app and, with patient consent, transmitted to a cloud-based data center. A built-in buzzer delivered audible reminders at scheduled dosing times for the intervention group. The device emitted a daily “heartbeat” signal to confirm proper functioning (Figure 1B). The safety of the device has been tested by the China Mobile Electronic Component Testing Center.

The management app enabled patients to visualize adherence data through graphs and tables, generate monthly adherence reports, complete a daily COPD Exacerbation Recognition Tool (CERT) diary [20], and record vital signs. Additional features included accessing step counts and weather information, setting medication reminders, self-reporting of CAT and mMRC scores, and viewing educational content. Clinicians could remotely monitor patient data via mobile or desktop devices, identify high-risk individuals, and initiate timely, individualized communication (Figure 1C).

Participants in the intervention group received an EIM with a reminder paired with a mobile management app, which delivered personalized reminders and adherence feedback based on EIM data. They were instructed to complete a daily CERT diary and record vital signs in the app, which displayed trends and generated monthly reports on adherence, physiological parameters, and symptoms. Participants in the control group received an EIM without reminders, solely for recording inhaler use. They also had the app installed solely to relay EIM data, without access to adherence feedback or educational content, or clinician access for review. Both groups continued to receive usual care from their physicians at each scheduled follow-up visit.

The primary outcomes were the proportion of participants with good adherence (TAI score ≥46) and the average daily adherence (calculated as the ratio of actual inhalations over prescribed inhalations) recorded by EIM during the study period. The TAI questionnaire assessed self-reported adherence during the preceding week and was administered at baseline and each scheduled follow-up visit. Adherence was continuously recorded by the EIM and summarized as the average daily adherence over the corresponding follow-up intervals (weeks 0‐4, 4‐12, and 12‐24). The secondary outcomes included CAT score, mMRC score, PHQ-9 score, GAD-7 score, lung function, and the number of exacerbations between groups.

All 158 randomized participants were included in the intention-to-treat analysis for primary and secondary outcomes. Continuous data were presented as mean with SD or median with IQR and analyzed using 2-tailed Student t tests or Mann-Whitney U tests as appropriate, and categorical data were presented as numbers with percentages and analyzed using the chi-square test or Fisher exact test. Two-sided tests were used throughout, and a value of P<.05 was considered statistically significant. Statistical analyses were conducted using IBM SPSS (version 29.0.1.0) for data management and descriptive statistics, R (version 4.4.2; R Foundation for Statistical Computing) for linear mixed-effects modeling and advanced analyses, and GraphPad Prism (version 10.4.1) for data visualization.

The primary outcome was analyzed using a linear mixed-effects model, with follow-up adherence measurements as the dependent variable, and group, time, and their interaction as fixed effects. A random intercept was included to account for within-participant correlation, and an unstructured covariance matrix was used for repeated measures over unequally spaced visits. Secondary outcomes were analyzed as follows: acute exacerbations using negative binomial regression; lung function and SGRQ scores using linear mixed models (after the Levene test indicated heterogeneity of variance, P=.046); and PHQ-9, GAD-7, mMRC, and CAT scores using linear mixed models. All models were adjusted for corresponding baseline values.

Subgroup analyses for mMRC and CAT scores examined potential effect modification by including group, time, subgroup, and their 3-way interaction terms in linear mixed models, with a significance threshold of P<.10 for interaction terms and not adjusted for multiple comparisons because it was considered exploratory. For other secondary outcomes, potential effect modification was examined using linear regression models that included group, subgroup, and their interaction term.

The study was approved by the China-Japan Friendship Hospital Ethics Committee (approval number 2022-KY-129) and was conducted in accordance with the Declaration of Helsinki (1975, as revised in 2008) and applicable Chinese regulations. Written informed consent was obtained from all participants before enrollment. Participant data were deidentified prior to analysis and stored on secure password-protected servers accessible only to authorized study personnel. Participants did not receive any financial compensation for participation. As appreciation for their time and contribution, participants were offered 1 complimentary lung function test at the 24-week follow-up and permanent access to the EIM device and integrated digital management platform after study completion, regardless of study group.

Participant recruitment was conducted between March 6, 2023, and January 31, 2024, and follow-up was completed on July 17, 2024. In the context of the COVID-19 pandemic during the recruitment period, a total of 158 eligible participants provided written informed consent by the end of the predefined study period, with 106 assigned to the intervention group and 52 to the control group. Participants continued their physician-prescribed inhaled therapy throughout the study. A total of 19 participants in the intervention group and 10 in the control group withdrew during the study, of whom 11 discontinued due to pMDI cessation (Figure 2). Among the remaining participants, 2 in the intervention group (1 switched from dual bronchodilation to triple therapy and 1 switched from triple therapy to dual bronchodilation) and 1 in the control group (switched from dual bronchodilation to triple therapy) had their inhaled regimens switched by their physicians during the study, without changing the pMDI device. No other changes in inhaled therapy were documented.

Baseline characteristics were similar between groups. A total of 129 (81.6%) participants were male, and the mean age was 64 (SD 8.30) years. The mean number of acute exacerbations in the previous year (including mild, moderate, and severe) was 0.95 (SD 1.34), and 72.2% (n=114) of participants were classified as having good adherence based on the self-reported TAI questionnaire (Table 1).

Except for 2 participants in the intervention group with sensor detachment, 156 participants had at least one valid daily adherence record. The median number of valid monitoring days was 161.5 (IQR 86.3‐168.0). TAI scores were collected at 3 follow-up visits, with response rates of 82.9% (n=131; 88 intervention vs 43 control), 75.3% (n=119; 77 vs 42), and 79.1% (n=125; 85 vs 40). No statistically significant differences were observed between groups in the proportion of participants classified as having good TAI adherence at any visits (Table 2). Linear mixed-effects modeling indicated that neither the main effect of group (P=.64), time within group (P=.051), nor the group × time interaction (P=.66) reached statistical significance.

Over the 24-week study period, the electronically measured medication adherence was significantly higher in the intervention group than the control group (59.52% vs 33.11%; P=.009; Figure 3A). The between-group difference increased over time: 85.27% vs 76.79% during weeks 0 to 4 (P=.04), 74.11% vs 31.47% during weeks 4 to 12 (P=.004), and 43.01% vs 11.90% during weeks 12 to 24 (P=.03; Table 2 and Figure 3B). The proportion of days with 100% of prescribed inhalations taken was also higher in the intervention group than in the control group (31.5% vs 16.7%; P=.007). Linear mixed-effects modeling revealed a significant main effect of group (P=.004) and time within groups (P<.001), whereas the group × time interaction did not reach statistical significance (P=.24).

No statistically significant between-group differences were observed for SGRQ-C scores, PHQ-9 score, GAD-7 score, mMRC grade, CAT score, or lung function at week 24 (Table 3).

No significant between-group differences were detected for continuous CAT and mMRC scores either; however, longitudinal analyses showed significant changes over time in both groups (main effect of time, both P<.001). Compared with baseline, the mean change in CAT score was −7.09 (95% CI −10.55 to −3.62) at week 4,−8.60 (95% CI −12.03 to −5.17) at week 12, and −8.72 (95% CI −12.18 to −5.25) at week 24 (all P<.001). The mean change in mMRC grade was −0.59 (95% CI −0.97 to −0.22), −0.87 (95% CI −1.25 to −0.50), and −0.93 (95% CI −1.32 to −0.54) at weeks 4, 12, and 24 (Figure 4). Exploratory generalized estimating equation analysis showed that the intervention group had higher odds of achieving the minimal clinically important difference in CAT scores over follow-up (odds ratio 2.21, 95% CI 1.11‐4.41; P=.02), with a significant group × time interaction (P<.001). However, no individual follow-up time point remained statistically significant after Bonferroni correction for multiple comparisons (Figure 5).

During the 24-week follow-up, the mean number of moderate-to-severe exacerbations was 0.11 (SD 0.89) in the intervention group and 0.44 (SD 0.90) in the control group, with 8 (7.5%) and 10 (19.2%) participants, respectively, experiencing at least one event. After adjustment for baseline exacerbation history, negative binomial regression showed no statistically significant between-group differences in moderate-to-severe exacerbations (incidence rate ratio 0.68, 95% CI 0.24‐1.94; P=.47) or severe exacerbations (incidence rate ratio 0.54, 95% CI 0.15‐2.00; P=.36; Figure 5). Baseline exacerbation history was a strong predictor of subsequent events, with each additional exacerbation at baseline associated with a 3.69-fold higher risk of exacerbation at 24 weeks (risk ratio 3.69, 95% CI 2.37‐5.75; P<.001).

To further explore potential factors associated with adherence and clinical outcomes, we conducted post hoc exploratory correlation analyses, subgroup analyses, and multivariable regression.

Exploratory Pearson correlation analysis of the main outcome and baseline information showed that the average adherence of the intervention group was positively correlated with age (Pearson correlation coefficient, r=0.31; P=.03) and the presence of an acute exacerbation in the past year (r=0.225, P=.02), whereas no significant correlation was observed in the control group (r=0.049; P=.28).

Exploratory subgroup analyses for secondary outcomes were conducted according to age (<60 vs≥60 y), sex, GOLD (Global Initiative for Chronic Obstructive Lung Disease) stage, baseline TAI score (<46 vs ≥ 46), baseline mMRC grade (<2 vs≥2), history of exacerbations (0 vs ≥1), and electronic adherence (<80% vs ≥80%). For acute exacerbations at week 24, potential effect modification was observed by mMRC grade (P=.02), GOLD stage (P=.03), and adherence (P=.045). Patients with mMRC ≥2, GOLD stage 3, or adherence ≥80% showed greater reductions with the intervention (mean differences −0.68, 95% CI −1.10 to −0.26; P=.002; −1.75, 95% CI −2.72 to −0.78; P<.001; and −0.79, 95% CI −1.32 to −0.26; P=.004) compared with the respective comparator subgroups. No significant differential response pattern was observed for the other subgroups.

To identify characteristics associated with high adherence (adherence ≥80%), we performed univariate logistic regression analyses using baseline variables. Eleven variables were associated with high adherence (P<.20 or area under the curve [AUC]>0.6). Among these, GOLD group (AUC=0.631; P=.03) and BMI (AUC=0.614; P=.03) exhibited the highest predictive ability. A multivariable logistic model incorporating these 2 variables yielded a mean cross-validated AUC of 0.611 (SD 0.046) after 5-fold cross-validation, with a sensitivity of 31.7% and a specificity of 83.3%.

The EIM demonstrated satisfactory performance in tracking medication adherence, but the completion rates of actively reported measures via the patient mobile app were notably low. During the 168-day follow-up period, the average completion rate for the daily CERT diary was only 7.0% (SD 23.7%), accompanied by a 2.3% (SD 16.5%) completion rate for body temperature logs and 5.3% (SD 23.2%) for oxygen saturation recordings. This finding may be attributed to the predominantly older adult study population, who generally exhibit lower proficiency and frequency in smartphone use. These results underscore the need for future digital therapeutic tools to incorporate more automated and user-friendly symptom monitoring modalities that minimize reliance on active patient input. Despite the low engagement with active app features, the end-of-study user satisfaction survey revealed favorable attitudes toward the digital therapeutic software. Among the 158 participants, 101 (63.9%) reported being “very satisfied” with the device, and 98 (62.0%) rated it as very convenient to use.

No adverse events were reported during the study.