TL;DR
This study uses latent-class growth modeling (LCGM) to analyze the usage patterns of a caloric-monitoring mobile app (iDAT) among patients with type 2 diabetes. It reveals three distinct usage trajectories: minimal, intermittent-waning, and consistent users, with females and those with higher exercise motivation being more likely to be consistent users. This work highlights the potential of LCGM for understanding app engagement in healthcare settings.
What Problem Does This Paper Address?
Managing type 2 diabetes effectively relies heavily on patient self-management, and mobile health (mHealth) interventions show promise in supporting this. However, simply developing an app isn't enough; understanding how patients actually use these tools over time is crucial for assessing their real-world impact and for designing more effective interventions. Previous research often provided only basic usage statistics, making it difficult to discern meaningful patterns of engagement. This paper addresses this gap by investigating the short-term (8-week) usage trajectories of a specific caloric-monitoring app, the "interactive Diet and Activity Tracker" (iDAT), among type 2 diabetes patients in a primary care setting, and identifying patient characteristics associated with these trajectories.
Methods
The core methodological innovation here is the application of Latent-Class Growth Modeling (LCGM) to analyze app usage data. LCGM is a statistical technique that identifies distinct groups (latent classes) of individuals who exhibit similar patterns of change (trajectories) over time.
Here's a breakdown of the approach:
- Data Collection: 84 patients with type 2 diabetes, who were new to the iDAT app, were enrolled. Their weekly app usage was monitored for 8 weeks. Weekly usage was defined as any food or exercise entry within that week. Baseline data included demographics, motivation for diet and exercise, diabetes self-efficacy (using the Diabetes Empowerment Scale-Short Form - DES-SF), and clinical variables (BMI, blood pressure, HbA1c).
- LCGM Application: PROC TRAJ software was used to apply LCGM to the weekly iDAT app usage data. The model aimed to identify the optimal number of latent trajectory groups and their shapes. Model fit statistics (like BIC) were used to determine the best-fitting model.
- Trajectory Identification: Based on the LCGM results, three distinct usage trajectories were identified and labeled:
- Minimal Users: Little to no app use, or use only in the first two weeks.
- Intermittent-Waning Users: Occasional weekly use, primarily in the first four weeks.
- Consistent Users: Weekly use throughout most or all of the 8-week period.
- Predictor Analysis: Univariate and multivariate logistic regression analyses were performed to identify patient characteristics (demographics, motivation scores, self-efficacy) that predicted membership in each trajectory group, with "Minimal Users" serving as the reference group.
Datasets
The study utilized data collected from 84 patients with type 2 diabetes. This data included:
- App Usage Data: Weekly records of food and exercise entries from the iDAT app over an 8-week period. This data was provided by the Singapore Health Promotion Board (HPB).
- Baseline Questionnaire Data: Information gathered at the start of the study, including:
- Demographics (age, gender, ethnicity, marital status, education, employment).
- Clinical variables (BMI, blood pressure, HbA1c).
- Social lifestyle factors.
- Smartphone characteristics and usage patterns.
- Motivation scores for diet and exercise.
- Diabetes self-efficacy scores (using the DES-SF).
No new datasets were introduced or benchmarks created; the study focused on analyzing existing app usage and patient characteristics.
Key Insights and Ideas
- Usage isn't Uniform: The most striking insight is that app usage is not a monolithic behavior. Even within a relatively short timeframe, distinct patterns of engagement emerge, ranging from near-zero use to consistent, sustained use. This challenges the assumption that all users will engage with an app in a similar way.
- LCGM as a Powerful Tool: The paper successfully demonstrates that LCGM is a valuable statistical framework for dissecting complex longitudinal usage patterns in mHealth interventions. It moves beyond simple averages to reveal underlying group behaviors, which is critical for understanding user engagement.
- Motivation Matters (Especially for Exercise): Higher motivation to exercise was a significant predictor for both intermittent-waning and consistent app use. This suggests that individuals who are already motivated to improve their physical activity are more likely to leverage tools like iDAT to support their efforts.
- Gender Differences in Engagement: Females were significantly more likely to be consistent users of the iDAT app compared to males. This finding, while requiring further exploration, points to potential gender-specific preferences or needs in mHealth tool adoption.
Key Results
- Three Trajectories Identified:
- Minimal Users: 78.6% (66/84) of participants.
- Intermittent-Waning Users: 11.9% (10/84) of participants.
- Consistent Users: 9.5% (8/84) of participants.
- Predictors of Consistent Use:
- Females had significantly higher odds of being Consistent Users compared to males (OR 19.55, 95% CI 1.78-215.42).
- Subjects with higher exercise motivation scores at baseline also had significantly higher odds of being Consistent Users (OR 4.89, 95% CI 1.80-13.28).
- Predictors of Intermittent-Waning Use:
- Subjects with higher exercise motivation scores at baseline had significantly higher odds of being Intermittent-Waning Users (OR 1.82, 95% CI 1.00-3.32).
Limitations and Future Work
- Short Observation Period: The study focused on an 8-week period. Longer-term usage patterns and their impact on clinical outcomes are not captured.
- App Database Limitations: The iDAT app's backend only provided weekly usage data. Daily input data would have offered a more granular view of usage patterns.
- Self-Motivation Reliance: The study was conducted in a primary care setting without external motivators (reminders, financial incentives), which reflects real-world usage but might limit the generalizability to more supported intervention settings.
- Sample Size: While 84 participants is reasonable for LCGM, a larger sample size could provide more statistical power and potentially reveal more nuanced trajectories or predictors.
- Future Work: The authors suggest that future research should investigate the psychological constructs influencing app adoption and explore how app design features can facilitate or hinder engagement. They also propose using LCGM for future studies to determine the effectiveness of caloric-monitoring apps on clinical outcomes. The study's findings can inform the design and sample size estimation for such future investigations.
Summary updated