Patient engagement with prescription refill text reminders across time and major societal events

Key Findings.

• •Patients still engage and reply to text messages sent related to medication adherence despite major social events, notably the coronavirus disease 2019 (COVID-19) pandemic and the 2020 presidential election.
• •Text messaging is a way to engage patients in times of uncertainty.
• •Messaging patients via text for prescription adherence can be studied for adherence in the midst of modern society.

Introduction

Many patients with chronic diseases face challenges in adhering to medications as prescribed.¹ Alerts to remind patients to refill medications may facilitate improvements in adherence.² However, patient engagement with such reminders varies, likely due to a variety of factors. We hypothesized that patient interaction with alerts may be partially dependent on temporal changes in the external environment. In this study, we specifically evaluated response to text message alerts during the coronavirus disease 2019 (COVID-19) pandemic lockdown and the 2020 election season.

Methods

We conducted a secondary analysis of The Nudge Study (Personalized Patient Data and Behavioral Nudges to Improve Adherence to Chronic Cardiovascular Medications; ClinicalTrials.gov Identifier: NCT03973931) to assess changes in patient engagement with medication adherence text message reminders over time. The study included 4 arms: standard care (no messaging); generic messages; optimized messages; and chatbot messaging. Patients who filled their prescriptions within the health care system (HCS) were assigned to an intervention arm and when the apparent refill gap was ≥7 days were messaged 5 times over 10 days reminding them to refill their medication. Messaging stopped if participants replied “DONE” to indicate that they had refilled their prescription or “STOP” to opt out of the study. Patients also could respond to the text messages with questions or other comments. Messages resumed for subsequent coverage gaps. Any response from patients was considered receipt and acknowledgment of the message.

The timing of the study crossed both the COVID-19 pandemic and the 2020 United States election. We created 3 time periods during study conduct: (1) pre-COVID baseline—October 2, 2019, to March 13, 2020—from study start until pandemic onset; (2) COVID-19 lockdown—March 14, 2020, to July 31, 2022—from declaration of pandemic stay-at-home orders through the transition to “living with and adapting to the pandemic conditions” in Colorado; and (3) election season—August 1, 2020, to November 30, 2020—marked by a contentious election and higher-volume messaging related to election campaigns, noted by the platform we used to send text messages to patients. We analyzed 5201 patients: 843 (16.2%) from the Veterans Affairs Eastern Colorado Health Care System and 4358 (83.8%) from Denver Health Medical Center. Of the patients analyzed, 3502 (67.3%) were enrolled in the pre-COVID period, 1180 (22.7%) in the initial COVID-19 period, and 519 (10.0%) in the 2020 election period. Table 1 provides a breakdown of patient demographics and conditions that qualified patients for the study stratified by time period of enrollment. Patient demographics and qualifying conditions differed over time, driven in part by disparate populations in each HCS and varying proportion of patients coming from each HCS over time. Given that sampled data from this observational analysis are subject to confounding across the time periods of interest, we attempted to account for imbalances through adjustment for HCS, individual patient demographics, and qualifying conditions in the model.

Table 1.

Breakdown of patient demographics stratified by period of enrollment

	Prepandemic (10/02/19–03/13/20)	Lockdown (03/14/20–07/31/21)	2020 Election (08/01/20–11/07/20)	P value
N	3502	1180	519
Age (y)				<.001
0–40	0.078 (274)	0.061 (72)	0.048 (25)
41–50	0.159 (558)	0.150 (177)	0.098 (51)
51–60	0.289 (1013)	0.275 (325)	0.225 (117)
61–70	0.3 (1049)	0.301 (355)	0.276 (143)
71–80	0.136 (476)	0.168 (198)	0.291 (151)
81–100	0.038 (132)	0.045 (53)	0.062 (32)
Gender				<.001
Female	0.497 (1741)	0.429 (506)	0.293 (152)
Race				<.001
White	0.692 (2425)	0.777 (917)	0.813 (422)
Black	0.195 (684)	0.110 (130)	0.073 (38)
Multiple	0.007 (26)	0.007 (8)	0.002 (1)
Other	0.097 (338)	0.086 (101)	0.056 (29)
Ethnicity				<.001
Hispanic	0.579 (2028)	0.532 (628)	0.364 (189)
Health care system				<.001
VA	0.087 (305)	0.229 (270)	0.516 (268)
Denver Health	0.913 (3197)	0.771 (910)	0.484 (251)
Qualifying condition
Atrial fibrillation	0.051 (177)	0.047 (56)	0.067 (35)	.21
Coronary artery disease	0.124 (434)	0.118 (139)	0.160 (83)	.04
Diabetes	0.555 (1942)	0.470 (555)	0.366 (190)	<.001
Hyperlipidemia	0.427 (1495)	0.443 (523)	0.435 (226)	.61
Hypertension	0.777 (2722)	0.772 (911)	0.796 (413)	.55

Values are given as proportion (total N).

P values are from multiple degree of freedom χ² tests.

VA = Veterans Affairs.

This discrete time hazards model was fit adjusting for patient characteristics, medication gaps, and HCS using a time-varying predictor for period to account for patient temporal transitions.³ Risk standardized survival curves and 95% bootstrapped confidence intervals (CIs) were estimated over the 21 and 30 days of postenrollment follow-up for message and opt-out responses, respectively.

Results

Time to message response (ie, “DONE”) and opt-out message response (ie, “STOP”) were not significantly different during the 3 periods after adjustment (Figures 1 and 2). After 3 weeks, the estimated proportion with a message response was 36.6% (95% CI 34.8%–38.3%), 36.8% (95% CI 34.1%–39.4%), and 37.4% (95% CI 33.7%–41.6%), and estimated proportion with an opt-out response at 30 days was 3.6% (95% CI 2.9%–4.3%), 2.8% (95% CI 1.8%–3.8%), and 4.2% (95% CI 2.6%–6.1%) for the pre-COVID, COVID-19, and election period groups, respectively.

Figure 1.

Time to patient text message response by time period (risk standardized). CI = confidence interval; COVID = coronavirus disease 2019.

Figure 2.

Time to patient opt-out by time period (risk standardized). Abbreviations as in Figure 1.

Discussion

We found that patients on chronic cardiovascular medications who received text message refill reminders when a ≥7-day coverage gap was detected consistently responded affirmatively to these alerts about one-third of the time and consistently opted out <5% of the time. Consistency in response was maintained despite the most disruptive pandemic in 100 years and one of the most contentious presidential elections in history.

Conclusion

These findings suggest that patient interactions with text messaging reminders are perhaps less likely related to temporal external changes and more fundamentally rooted in ingrained behaviors. As such, efforts to use and improve text messaging reminders directed to patients should recognize the apparent consistency of smartphone interactions over time.