JMIR Publications

ABSTRACT

Background:

Text message reminders have been used to promote many health behaviors such as improving diet and physical activity, managing chronic health conditions, reminding patients about medical appointments, and supporting medication adherence across a range of health conditions. Despite their promise, developing effective reminders tailored to specific patient populations is resource intensive. Artificial intelligence (AI) may facilitate item development and online research panels may provide an efficient way to test message content with target users prior to implementing large-scale trials.

Objective:

This study aimed to (1) develop a library of antidepressant adherence-promoting text messages that are perceived as helpful, (2) test whether an online panel approach can be utilized to evaluate them and (3) identify message characteristics perceived as most helpful by depressed patients taking antidepressant medication.

Methods:

Eighty-three text message reminders were developed based on barriers to adherence and behavior change technique pairings, with about half authored by the study team and half generated by AI. Using an online panel, we recruited 181 U.S. adults currently prescribed an antidepressant medication. Each participant rated a subset of messages on how much they thought each would help them remember to take their medication. Associations between message characteristics and ratings were estimated using generalized linear models. Survey weights were used in analyses to align the sample with national antidepressant user demographics.

Results:

The online panel was able to rapidly recruit a sample of participants that provided 7,520 item ratings in total. AI generated messages were rated significantly more helpful than those authored by humans (adjusted mean difference = 0.24 on a 5-point scale, 95% CI: 0.12–0.36, p < .001). Messages addressing delayed symptom benefit were preferred over other adherence barriers, and behavioral change techniques emphasizing self-monitoring, habit formation, and natural consequences (i.e., highlighting outcomes of one’s behavior) received significantly higher ratings than those using external influence.

Conclusions:

Online panels offer a rapid, scalable approach to evaluating text message reminders targeted to patients currently taking antidepressants. AI can efficiently generate message content perceived to be helpful in promoting adherence. In addition, behavioral change techniques and barrier-focused content can guide message development and effectively support adherence.