Quantitative Analysis

Descriptive analyses were performed on data collected using the app and through the various questionnaires (Figure 2). Data collected from the app includes usage logs (logging all interactions with the app including features that were accessed, frequency, the time spent on a certain feature, etc), activities registered by patients using the app, their medication compliance, and evolution of various physiological parameters. Data collected from the questionnaires included overall physical activity levels and quality of life indicators using International Physical Activity Questionnaire (IPAQ) [27], HeartQoL [28], and the 5-level EuroQol 5-Dimensions (EQ-5D) questionnaires [29]. Clinical parameters such as weight, blood pressure, heart rhythm, exercise capacity (maximal oxygen consumption [VO₂ max]), glucose (glycated hemoglobin [HbA_1c]), and lipid profile (low-density lipoprotein and high-density lipoprotein [HDL] cholesterol) were collected using blood tests and ergo-spirometry tests at the hospital and rehabilitation center. All measurements were procured at baseline, the crossover point (month 2), and the end (month 4). The gathered data were used to observe mean differences between various parameters during different stages of the study, compare assessments, and make preliminary estimates of the influence of the app on the different evaluation metrics (parameters listed in Multimedia Appendix 3). To gain a precise insight on the actual influence of the app, we used the overall app usage percentage of each patient as a weighting variable against which all parameters were weighted. The overall app use percentage was computed using the number of days of actual app use against the total days that a patient was in the app usage phase. We then performed a detailed weighted statistical analysis on the data according to the intention-to-treat principle by the assigned treatment group. Nonparametric alternatives (such as Wilcoxon 2-sample test) were used for parametric statistics (t tests) in case assumptions for the latter were violated.

For all the evaluations, the level of significance was 2-sided alpha=.05. First, we tested if the data were normally distributed using 4 tests for normality—Shapiro-Wilk test, Kolmogorov-Smirnov test, Cramer-von Mises test, and the Anderson-Darling test. When the data were normally distributed, we used Student t test, and when the data were not normally distributed, we used signed rank test to determine significant effects (Figure 3).

Flowchart showing the statistical analysis process to evaluate the effect of HeartHab on various health metrics.

Being a crossover study design, we also evaluated if there were any significant carryover effects. The primary objective to determine this was to see if any significance observed during the first phase with patients in group 1 (using the app in the first phase) was carried over when they switched to usual care in phase 2 as a means to get an indication on maintenance effects. The process followed to evaluate the carryover effect is detailed in the flowchart (Figure 4).

Flowchart showing the process of statistical analysis to identify significant carryover effects. UC: under control; SUM: summation.