Cost Savings From an mHealth Tool for Improving Medication Adherence

ABSTRACT

Objective: To determine the health care cost savings from the Wellth app, a mobile health intervention that uses financial incentives to increase medication adherence.

Study Design: An observational study of members in one of Arizona’s Medicaid managed care plans, part of Arizona Health Care Cost Containment System (AHCCCS), using the Wellth app from March 28, 2020, to January 12, 2021. One-to-one matching was used to identify comparable nonparticipants, and a difference-in-differences approach was used to estimate the impact of the Wellth intervention on outcomes defined over the 9 months before and after using Wellth.

Methods: An AHCCCS managed care health plan provided claims data that contained drug prescription, health care utilization, and health care cost information for all participants, and Wellth provided app usage data and contextual information about the Wellth intervention.

Results: On average, the Wellth intervention increased medication adherence by 5.0 percentage points (95% CI, 2.9-7.1; P = .008) and reduced emergency department (–0.02; 95% CI, –0.03 to –0.01; P = .002), inpatient (–0.04; 95% CI, –0.06 to –0.02; P = .001), and mental health clinic (–0.06; 95% CI, –0.10 to –0.01; P = .013) visits relative to nonparticipants over 9 months. Short-term reductions in utilization had an estimated mean cost savings over 9 months of $88.15 (95% CI, $31.07-$136.40), with greater reductions for those with chronic obstructive pulmonary disease, schizophrenia, or major depression.

Conclusions: Given the relatively low cost of the Wellth intervention, our findings provide preliminary evidence of cost savings from implementing Wellth among adults with several common chronic conditions.

Am J Manag Care. 2024;30(10):e289-e296. https://doi.org/10.37765/ajmc.2024.89621

Takeaway Points

Medication nonadherence for chronic conditions results in $100 billion to $300 billion in avoidable health care costs annually in the US and can result in out-of-pocket costs as high as 110% of annual household income for Medicaid beneficiaries. Our study findings show that the Wellth smartphone app successfully increased medication adherence and lowered unnecessary health care utilization and costs among Medicaid beneficiaries self-managing chronic conditions over a 9-month period. Given the low cost of the Wellth program, these findings demonstrate the health care cost savings from using this mobile health (mHealth) medication adherence solution among Medicaid beneficiaries.

• Managed care decision makers should consider these findings when deciding what types of behavioral interventions and supports to offer insurance plan members, particularly those managing chronic conditions, to reduce avoidable health care expenses.
• This study builds on the success of other financial incentive–based interventions and demonstrates that an mHealth platform can effectively deliver incentives that produce sustained benefits post intervention.
• This research emphasizes the importance of behavior change and healthy habits for reducing avoidable health care utilization and costs associated with chronic conditions.

Approximately 60% of US adults have a chronic health condition, and 40% of US adults have multiple chronic conditions.¹ Self-management of chronic conditions between clinic visits is key for improving individual health outcomes, slowing disease progression, and reducing the economic burden of chronic conditions in the US.^2,3 One crucial form of self-management is adherence to prescribed medications; however, an estimated 50% of patients with a chronic condition are nonadherent to their prescribed medication regimen,^4,5 resulting in $100 billion to $300 billion in avoidable health care costs every year.^6,7 Nonadherence to medications for managing chronic conditions can result in out-of-pocket costs as high as 110% of annual household income for Medicaid beneficiaries,⁸ and nonadherence will become an even larger public health and financial crisis as the prevalence of chronic conditions continues to rise in the US.⁹

Incentive-based interventions have effectively improved several self-management behaviors, such as physical activity,^10-15 smoking cessation,^16,17 and medication adherence,^18,19 but they tend to be resource intensive and difficult to scale.²⁰ One way to increase the scalability of incentive-based interventions is through the use of mobile health (mHealth) tools, such as a smartphone app that can measure medication adherence and immediately deliver incentives for demonstrating adherence.^21-26 Wellth is one example of a smartphone app that is commercially available for all iOS and Android smartphones that delivers financial incentives to improve medication adherence without requiring additional time or resources from health care providers. Specifically, image recognition algorithms are used to immediately verify a daily photograph that Wellth users submit through the app of their prescribed medication(s) in their hand to avoid losing a small amount from their larger monthly financial incentive. Recent studies have shown that Wellth is feasible and acceptable in a wide range of populations, but to date, the efficacy of Wellth has not been rigorously tested.^22-25

Financial incentives can improve medication adherence by combatting individuals’ forgetfulness^27-30 and present bias (ie, the tendency to discount future health benefits relative to more immediate costs).^31,32 Therefore, the effectiveness of incentives may vary between chronic conditions that impose different immediate costs, such as adverse effects from the medication, health symptoms, and social stigma. Given these factors, examining the effect of Wellth on medication adherence across chronic conditions is a novel test of financial incentives delivered through an mHealth app.

Although financial incentives have been shown to increase medication adherence, evidence of their benefits on subsequent reductions in health care utilization and costs is limited. This retrospective study estimated the effects of the Wellth app on both medication adherence and health care utilization and costs using Medicaid insurance claims among a large sample (N = 6716) of adults managing 1 or more chronic conditions through daily medication. We used a machine learning–based matching design to estimate the benefits of financial incentives delivered through the Wellth app among Medicaid beneficiaries in Arizona who were enrolled in the same Medicaid managed care organization. The primary outcomes were changes in (1) medication adherence measured by proportion of days covered and (2) health care utilization and costs over 9 months, and we hypothesized that all these outcomes would be significantly improved among Wellth participants relative to those who did not participate in Wellth. These findings are crucial for demonstrating the cost-effectiveness of financial incentive–based interventions delivered via an mHealth tool and will inform providers’ and policy makers’ future adoption of such interventions.

METHODS

Design and Participants

From March 28, 2020, through January 12, 2021, a private Medicaid managed care organization in Arizona (hereafter referred to as the Medicaid provider) offered the Wellth intervention to all its members who had been diagnosed with at least 1 of 7 chronic conditions and demonstrated low medication adherence over the past 12 months. The 7 chronic conditions were diabetes, chronic obstructive pulmonary disease (COPD), asthma, schizophrenia, bipolar disorder, major depression, and opioid use disorder. Past medication adherence was determined using Medicaid prescription drug claims data and the Medicaid provider’s own calculation of members’ proportion of days covered (PDC), which is an estimate of the proportion of days with medication available based on the prescription or refill dates and days supplied. The Medicaid provider considered low medication adherence to be less than 80% PDC, and it paid for the full cost of the Wellth intervention ($225 per person) for any eligible member who agreed to participate until a total of 3358 participants were enrolled.

All eligible members with low adherence (N = 21,904) received an introductory letter and postcard from the Medicaid provider and Wellth, which contained information on the Wellth intervention and a link to the Wellth intervention interest form. At the time of recruitment, Arizona’s Medicaid agency, called the Arizona Health Care Cost Containment System (AHCCCS), limited annual incentives for Medicaid beneficiaries to $150 through policy, which, as operationalized by the managed care entity, limited the length of the Wellth intervention. As a result, the Wellth intervention varied based on a member’s gender and diabetes status, as existing financial incentives were already available for female members and those with diabetes (eAppendix Table 1 [eAppendix available at ajmc.com]). For example, female members with diabetes could receive up to $75 for complete adherence during the Wellth intervention, whereas male members without diabetes could receive up to $125. These differences resulted from the existing managed care health plan financial incentives framework of $25 for receiving a cervical cancer screening (women only), hemoglobin A_1C testing (patients with diabetes only), and/or a flu shot that were offered by the Medicaid provider. The intervention also differed in length (90 or 180 days), with the 90-day length used when incentives were limited to $75 because Wellth determined that spreading $75 across a 180-day intervention would result in monthly incentive amounts that were too small to be motivating. All intervention participants provided informed consent allowing Wellth to use their Medicaid insurance claims data and app data for research purposes. The protocol for this study was deemed exempt from review by the institutional review board of Arizona State University because all intervention and outcome data were deidentified.

Recruitment was completed between March 28, 2020, and June 29, 2020. In addition to mailers and emails, eligible members received recruitment phone calls from Wellth staff, but not all members were called (66%; 14,451/21,904 received recruitment calls) before Wellth reached the enrollment target set by the Medicaid provider. These recruitment calls were first made to randomly selected eligible members in the lowest quartile of medication adherence. The recruitment calls then were made to members in progressively higher quartiles of adherence until the enrollment target was reached. Only a few members in the top quartile of adherence (n = 707) were called and recruited before the enrollment target was reached. Those who declined to participate, did not answer, or did not receive the recruitment calls before the enrollment target was reached were included in the analyses as potential controls (Figure 1). Recruitment calls were made in either English or Spanish based on a primary language spoken variable that was collected by the Medicaid provider.

Intervention

The Wellth intervention uses insights from behavioral economics to improve medication adherence. Specifically, participants are eligible for monthly financial incentives that are conditional on submitting daily photographic evidence of their pill taking through the Wellth app. The intervention leverages loss aversion framing (ie, the tendency of people to prefer avoiding losses to acquiring equivalent gains)^33,34 to increase the effectiveness of incentives. At the start of each 30-day period, participants are shown their potential maximum incentive amount and then are penalized $2 per day if they fail to submit a photo(s) of their prescribed pill(s) in their hand. If a participant’s prescription requires multiple doses per day, then photos of each dose in the day are required to avoid the $2 penalty. No participant has to pay Wellth if the amount they are penalized in the pay period exceeds the maximum incentive amount (eg, if 10 or more days are missed, resulting in a loss of $20 or greater, and the maximum incentive amount is $20, then the participant receives $0 in incentives). To prevent participants from uploading old photos, the Wellth app requires that a new photo be taken through the in-app camera feature for each dose. Wellth uses image recognition algorithms to instantly verify if the correct type and quantity of pills are shown in a participant’s submitted photo. Wellth then immediately provides a notification to the participant that their photo submission was accepted and that they avoided losing $2 that day (see eAppendix Table 2 for details on the incentive structures).

In addition to financial incentives, Wellth provides daily reminders at and after the time of day that participants selected for taking their pills. The first daily reminder is sent at the participant’s selected pill-taking time, and subsequent reminders are then sent 1.5 hours and 3 hours after the selected time if the participant has yet to complete their photo submission for that dose. The first 2 reminders are delivered as push notifications, and the final reminder is sent via SMS message. If participants miss more than 4 daily submissions during a week, a Wellth patient coordinator calls the participant to ask whether they are having technical difficulties using the app or to provide help navigating other medication adherence barriers. Following the conclusion of the Wellth intervention, participants in this study could continue to use the app to track their pill taking and receive reminders, but they were no longer eligible for incentives, were no longer contacted by Wellth patient coordinators, and no longer received messages from the app if they missed any submissions.

Outcomes

The primary outcome was change in medication adherence from the 9 months before to the 9 months after the Wellth intervention started. Medication adherence was measured by PDC for the prescription associated with each member’s chronic condition. For members with multiple chronic conditions, the prescription with the highest PDC was used to generate conservative estimates for the impact of the Wellth intervention on adherence. We also examined the impact of the Wellth intervention on changes in health care utilization and costs from the 9 months before to the 9 months after the Wellth intervention started. Health care utilization was measured as the number of unique visits (ie, health insurance encounters) within 5 categories based on the type of service provided: emergency department, inpatient, outpatient, mental health clinic, and substance use treatment center. Due to time and other resource limitations among the Medicaid provider’s data team, health care costs were not available for all insurance encounters, so we estimated costs based on Medicaid reimbursement cost data that were available only for a randomly generated subset of the insurance claims (32%; 1,293,524/4,082,188). The change in health care costs from the Wellth intervention was the product of the change in utilization and the mean cost associated with each type of service, and 95% CIs for these estimates were also estimated.

Statistical Analysis

Because participation in the Wellth intervention was not randomly assigned, we used a machine learning technique to construct a propensity score–matched sample of Wellth participants and nonparticipants. Specifically, we calculated each member’s likelihood of Wellth participation using a random forest model based on members’ age, gender, primary language spoken, race, ethnicity, employment status, diagnosed chronic condition(s), quartile of adherence that determined recruitment call priority, medication adherence, health care utilization, and health care costs in the 9 months before Wellth intervention recruitment began. A random forest model was used because empirical and simulation-based research has shown improved control for selection bias over the traditional logistic regression approach when using high-dimensional data.^35-37 We then used one-to-one propensity score matching (with replacement) to identify an equal-sized group of nonparticipants to act as the control group in the subsequent analyses. Given the large reservoir of potential matches (n = 18,501), a caliper for selecting matches was not needed (the mean [SD] propensity score difference was 0.06 [0.21]), and a match was found for all Wellth participants.

We used an intent-to-treat analytic approach and compared the change in outcomes from the 9 months before to the 9 months after Wellth began for each Wellth participant (and their matched nonparticipant) to estimate the effect of the Wellth intervention (ie, a difference-in-differences estimate). Measuring outcomes over the 9 months after the Wellth intervention started provided results that readily generalize to other managed care plans that face similar limitations on intervention duration (eAppendix Tables 1 and 2). A censored tobit regression model was used to estimate effects on PDC, which ranges from 0 to 1, and ordinary least squares (OLS) regression was used to estimate effects on health care utilization, which is a continuous measure of average monthly visits. We also examined the effect of the Wellth intervention separately according to the 7 chronic conditions included in the sample. All hypothesis tests were 2-sided with an unadjusted type I error rate of 0.05, and all analyses and visualizations were conducted using Stata/MP 16.1 (StataCorp LLC).

RESULTS

A total of 21,904 members were eligible for participation in the Wellth intervention. Table 1 presents the sample characteristics of the 3358 members who volunteered to participate in the Wellth intervention, the 18,546 members who did not participate in the Wellth intervention, and the 3358 nonparticipants who were included in the matched sample. The differences in characteristics between Wellth participants and nonparticipants, along with 2-sided t tests, are also displayed in Table 1.

Table 2 shows the distribution of the outcome measures among the 3358 Wellth participants, the 18,546 total nonparticipants, and the 3358 nonparticipants who were included in the matched sample. The outcome measures in Table 2 include PDC and emergency department, inpatient, outpatient, mental health clinic, and substance use treatment center visits per month over the 9 months prior to the start of the Wellth intervention. The differences in outcomes between the Wellth participants and nonparticipants, along with 2-sided t tests, are also displayed in Table 2.

Tobit regression models were used to estimate the effect of the Wellth intervention on changes in PDC and health care utilization over 9 months. The estimated change in PDC is displayed in Table 3, which shows the estimated effect of the Wellth intervention both with and without including all the observable characteristics (listed in Table 1). Figure 2 displays the OLS regression coefficient and the 95% CI for the estimated reduction in visits among the 5 types of health care services: emergency department, inpatient, outpatient, mental health clinic, and substance use treatment center. The full regression results are displayed in the eAppendix (eAppendix Table 3).

To better understand what types of participants had the largest reductions in health care utilization, we estimated separate OLS regression models for subsamples based on chronic condition(s). eAppendix Figures 1 and 2 show the Wellth intervention effect and 95% CI for the reductions in emergency department and inpatient visits, respectively, as these are the 2 most expensive types of health care utilization (see eAppendix Table 4 for the distribution of Medicaid reimbursement rates for each of the 5 types of health care services). The corresponding regression results are displayed in the supplemental materials (eAppendix Tables 5 and 6).

Given the significant reductions in inpatient visits across multiple conditions, we estimated the health care cost reductions associated with the reductions in inpatient visits observed among each of the 7 chronic conditions. eAppendix Table 7 displays the mean inpatient cost reductions for each chronic condition, which are calculated as the product of the mean inpatient cost and the estimated mean reduction in the number of visits for each chronic condition. The 95% CI uses the lower (and upper) bounds for both the inpatient cost and estimated reduction in inpatient visits.

DISCUSSION

Our findings show that the Wellth intervention, an mHealth tool that delivers behavioral economics–based financial incentives for improving medication adherence, significantly increased medication adherence and reduced health care utilization and costs among a population of Medicaid beneficiaries who were managing 1 or more chronic conditions and were considered low adherers (< 80%) by their Medicaid program. Specifically, emergency department, inpatient, and mental health clinic visits were all reduced over 9 months relative to nonparticipants. We additionally found that the reductions in emergency department and inpatient visits were largest among members with COPD and mental health conditions. These reductions in health care utilization were associated with large decreases in Medicaid reimbursement costs (approximately $288.98 in reduced emergency department and inpatient visit costs for those with COPD, for example). Given the small per-person cost of the Wellth intervention (roughly $125 in incentives and $100 in administrative costs), these findings suggest that the Wellth intervention is cost saving for several common chronic conditions (eg, COPD, schizophrenia, and major depression) over 9 months and that Wellth is likely cost saving for many other chronic conditions over a period longer than 9 months.

As expected, average medication adherence was significantly improved for Wellth participants compared with nonparticipants over the 9 months after the Wellth intervention (roughly 5 percentage points higher), which likely drove the corresponding reductions in health care utilization. To further examine the relationship between medication adherence and health care utilization, we first identified the Wellth participants who maintained medication adherence above 90% during the entire Wellth intervention, which was just over half of Wellth participants (53.9%). We then used OLS regression models to estimate separate Wellth effects for those with high (> 90%) vs low medication adherence during the intervention and found that those with high adherence also had the largest reductions in emergency department, inpatient, and substance use treatment center visits (eAppendix Table 8). These findings support the conclusion that increased medication adherence resulting from the Wellth intervention contributed to lower health care utilization and costs among Wellth participants.

The success of the Wellth intervention relative to other financial incentives–based or mHealth interventions that have shown smaller or insignificant effects^38-40 can be attributed to several factors. First, Wellth carefully frames the financial incentives to leverage individuals’ loss aversion, which has been shown to significantly influence behavior relative to traditional incentives-based interventions.^34,41-43 Second, Wellth asked participants to set a time of day for taking their pills and then provided daily reminders at and after participants’ self-selected pill-taking time. These reminders are likely effective at reducing forgetfulness^44-46 and may also be helpful for promoting habit formation.⁴⁷ Finally, relative to other mHealth medication adherence interventions, the Wellth app interface and daily procedures are easy to use and perform. Previous Wellth interventions found that participants preferred submitting a daily pill-taking photo through the Wellth app over other methods of monitoring adherence (eg, electronic monitoring using pill bottle caps or ecological momentary assessments).^24,25

Limitations

Despite the large sample size and objective outcome measures, there were limitations to this study. First, this was not a randomized controlled trial, and thus we did not have a true control group to use in our analyses. Instead, our matched sample included individuals who either declined to participate, did not answer recruitment calls, or were never contacted by Wellth before the enrollment target was reached. As a result, everyone in the sample received some form of Wellth recruitment materials (eg, emails, mailers) and some may have received recruitment calls, which might have influenced their medication adherence and health care utilization and costs. Additionally, there were several observable differences between Wellth participants and nonparticipants at baseline that were not completely resolved through the matching procedures. These sample differences, and the potential for unobservable differences between those who volunteered for the Wellth intervention and those who did not, highlight the need for additional research using a more robust study design.

Second, intervention recruitment occurred around the start of the COVID-19 pandemic, which likely had its own effect on medication adherence and health care utilization and costs. Specifically, health care utilization and costs were likely reduced for everyone given concerns about COVID-19 and the shortages of hospital beds and health care workers, and medication adherence may have been improved for those who started working from home, where they were closer to their medications throughout the day. However, we expect that COVID-19 impacted both participants and nonparticipants equally and that this confounding effect likely reduces our estimates (decreased utilization left less room for improvements due to the Wellth intervention), so it may be that Wellth would have even greater benefits under regular conditions.

Third, our data for nonparticipants did not indicate the number of prescribed pills per day. Although more than 73% of participants were on a once-a-day regimen and we did not observe significant differences in the likelihood of receiving Wellth rewards between those on a once-a-day regimen and those with a multiple-daily-dose regimen during the Wellth intervention, future research should investigate potential heterogenous Wellth intervention effects by daily dosage.

Finally, the effect of the Wellth intervention on health care utilization may be underestimated because of the short intervention duration and 9-month outcome measures. Typically, Wellth interventions are 12 months or longer, but the available incentive limits at the time of the study ($150) resulted in a shorter 90- or 180-day intervention, as described in eAppendix Table 1. Additionally, we expect that the benefits from improved medication adherence are experienced over time and that greater reductions in health care utilization may have been observed if the outcomes had been measured over a longer time horizon.

CONCLUSIONS

Our findings demonstrate the effectiveness of Wellth, a behavioral economics–based mHealth tool, for improving medication adherence and reducing health care utilization and costs among Medicaid beneficiaries managing a range of chronic conditions. Wellth participants showed significantly higher medication adherence over a 9-month period relative to a sample of matched nonparticipants. Furthermore, the improvements in medication adherence reduced health care utilization, particularly for those with COPD and mental health conditions, and the associated cost reductions in many cases exceeded the cost of administering Wellth. These findings provide preliminary evidence for the cost-effectiveness of the Wellth intervention among adults with chronic conditions, which can help address the growing economic and health care burdens of chronic conditions in the US.

Acknowledgments

The authors would like to thank Matthew Loper, Samantha Cordero, and Coby Kramer-Golinkoff for facilitating their research collaboration with the Medicaid provider and for providing important program details during the writing process.

Author Affiliations: College of Health Solutions, Arizona State University (CS), Phoenix, AZ; RAND Corporation (SL), Santa Monica, CA; College of Medicine, University of Arizona (PR), Phoenix, AZ; Department of Psychology, University of Georgia (SC), Athens, GA; School of Public Health, University of Minnesota (PH), Minneapolis, MN.

Source of Funding: None.

Author Disclosures: Dr Stecher and Dr Linnemayr have R01 grants from the National Institutes of Health. Dr Huckfeldt has received grants from the National Institutes of Health and the Agency for Healthcare Research and Quality and contract funding from CMS. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (CS, SL, PH); acquisition of data (CS); analysis and interpretation of data (CS, SL, PR, PH); drafting of the manuscript (CS, SL, SC); critical revision of the manuscript for important intellectual content (CS, SL, PR, SC, PH); statistical analysis (CS, PH); administrative, technical, or logistic support (CS, SC); and supervision (CS, SL, PR, PH).

Address Correspondence to: Chad Stecher, PhD, Arizona State University, 500 N 3rd St, Phoenix, AZ 85004. Email: chad.stecher@asu.edu.

REFERENCES

1. Carney TJ, Wiltz JL, Davis K, Briss PA, Hacker K. Advancing chronic disease practice through the CDC Data Modernization Initiative. Prev Chronic Dis. 2023;20:E110. doi:10.5888/pcd20.230120

2. Asch DA, Muller RW, Volpp KG. Automated hovering in health care—watching over the 5000 hours. N Engl J Med. 2012;367(1):1-3. doi:10.1056/NEJMp1203869

3. Allegrante JP, Wells MT, Peterson JC. Interventions to support behavioral self-management of chronic diseases. Annu Rev Public Health. 2019;40(1):127-146. doi:10.1146/annurev-publhealth-040218-044008

4. Colantonio LD, Huang L, Monda KL, et al. Adherence to high-intensity statins following a myocardial infarction hospitalization among Medicare beneficiaries. JAMA Cardiol. 2017;2(8):890-895. doi:10.1001/jamacardio.2017.0911

5. Nieuwlaat R, Wilczynski N, Navarro T, et al. Interventions for enhancing medication adherence. Cochrane Database Syst Rev. 2014;2014(11):CD000011. doi:10.1002/14651858.CD000011.pub4

6. Fast facts: health and economic costs of chronic diseases. CDC. July 12, 2024. Accessed September 15, 2024. https://www.cdc.gov/chronic-disease/data-research/facts-stats/index.html

7. Iuga AO, McGuire MJ. Adherence and health care costs. Risk Manag Healthc Policy. 2014;7:35-44. doi:10.2147/RMHP.S19801

8. Chapel JM, Ritchey MD, Zhang D, Wang G. Prevalence and medical costs of chronic diseases among adult Medicaid beneficiaries. Am J Prev Med. 2017;53(6)(suppl 2):S143-S154. doi:10.1016/j.amepre.2017.07.019

9. Chronic disease. CDC. Accessed September 15, 2024. https://www.cdc.gov/chronicdisease/data/statistics.htm

10. Barte JCM, Wendel-Vos GCW. A systematic review of financial incentives for physical activity: the effects on physical activity and related outcomes. Behav Med. 2017;43(2):79-90. doi:10.1080/08964289.2015.1074880

11. Mitchell MS, Orstad SL, Biswas A, et al. Financial incentives for physical activity in adults: systematic review and meta-analysis. Br J Sports Med. 2020;54(21):1259-1268. doi:10.1136/bjsports-2019-100633

12. Milkman KL, Gromet D, Ho H, et al. Megastudies improve the impact of applied behavioural science. Nature. 2021;600(7889):478-483. doi:10.1038/s41586-021-04128-4

13. Acland D, Levy MR. Naiveté, projection bias, and habit formation in gym attendance. Manage Sci. 2015;61(1):146-160. doi:10.1287/mnsc.2014.2091

14. Carrera M, Royer H, Stehr M, Sydnor J. Can financial incentives help people trying to establish new habits? experimental evidence with new gym members. J Health Econ. 2018;58:202-214. doi:10.1016/j.jhealeco.2018.02.010

15. Charness G, Gneezy U. Incentives to exercise. Econometrica. 2009;77(3):909-931. doi:10.3982/ECTA7416

16. Cahill K, Hartmann-Boyce J, Perera R. Incentives for smoking cessation. Cochrane Database Syst Rev. 2015;(5):CD004307. doi:10.1002/14651858.CD004307.pub5

17. Volpp KG, Troxel AB, Pauly MV, et al. A randomized, controlled trial of financial incentives for smoking cessation. N Engl J Med. 2009;360(7):699-709. doi:10.1056/NEJMsa0806819

18. Petry NM, Rash CJ, Byrne S, Ashraf S, White WB. Financial reinforcers for improving medication adherence: findings from a meta-analysis. Am J Med. 2012;125(9):888-896. doi:10.1016/j.amjmed.2012.01.003

19. Stecher C, Mukasa B, Linnemayr S. Uncovering a behavioral strategy for establishing new habits: evidence from incentives for medication adherence in Uganda. J Health Econ. 2021;77:102443. doi:10.1016/j.jhealeco.2021.102443

20. Michie S, van Stralen MM, West R. The behaviour change wheel: a new method for characterising and designing behaviour change interventions. Implement Sci. 2011;6:42. doi:10.1186/1748-5908-6-42

21. Iribarren SJ, Akande TO, Kamp KJ, Barry D, Kader YG, Suelzer E. Effectiveness of mobile apps to promote health and manage disease: systematic review and meta-analysis of randomized controlled trials. JMIR Mhealth Uhealth. 2021;9(1):e21563. doi:10.2196/21563

22. Guinart D, Sobolev M, Patil B, Walsh M, Kane JM. A digital intervention using daily financial incentives to increase medication adherence in severe mental illness: single-arm longitudinal pilot study. JMIR Ment Health. 2022;9(10):e37184. doi:10.2196/37184

23. Riegel B, Stephens-Shields A, Jaskowiak-Barr A, Daus M, Kimmel SE. A behavioral economics-based telehealth intervention to improve aspirin adherence following hospitalization for acute coronary syndrome. Pharmacoepidemiol Drug Saf. 2020;29(5):513-517. doi:10.1002/pds.4988

24. Granek B, Evans A, Petit J, et al. Feasibility of implementing a behavioral economics mobile health platform for individuals with behavioral health conditions. Health Technol (Berl). 2021;11:505-510. doi:10.1007/s12553-020-00492-9

25. Giordano NA, Riman KA, French R, et al. Comparing medication adherence using a smartphone application and electronic monitoring among patients with acute coronary syndrome. Appl Nurs Res. 2021;60:151448. doi:10.1016/j.apnr.2021.151448

26. Nadadur S. Medication adherence app for food pantry clients with diabetes: a feasibility study. J Nurse Pract. 2022;18(8):897-903. doi:10.1016/j.nurpra.2022.05.011

27. Stawarz K, Rodríguez MD, Cox AL, Blandford A. Understanding the use of contextual cues: design implications for medication adherence technologies that support remembering. Digit Health. 2016;2:2055207616678707.
doi:10.1177/2055207616678707

28. Mira JJ, Lorenzo S, Guilabert M, Navarro I, Pérez-Jover V. A systematic review of patient medication error on self-administering medication at home. Expert Opin Drug Saf. 2015;14(6):815-838. doi:10.1517/14740338.2015.1026326

29. Gadkari AS, McHorney CA. Unintentional non-adherence to chronic prescription medications: how unintentional is it really? BMC Health Serv Res. 2012;12:98. doi:10.1186/1472-6963-12-98

30. Pérez-Jover V, Sala-González M, Guilabert M, Mira JJ. Mobile apps for increasing treatment adherence: systematic review. J Med Internet Res. 2019;21(6):e12505. doi:10.2196/12505

31. Mogler BK, Shu SB, Fox CR, et al. Using insights from behavioral economics and social psychology to help patients manage chronic diseases. J Gen Intern Med. 2013;28(5):711-718. doi:10.1007/s11606-012-2261-8

32. Prakash AM, He QC, Zhong X. Incentive-driven post-discharge compliance management for chronic disease patients in healthcare service operations. IISE Trans Healthc Syst Eng. 2019;9(1):71-82. doi:10.1080/24725579.2019.1567630

33. Tversky A, Kahneman D. Advances in prospect theory: cumulative representation of uncertainty. J Risk Uncertain. 1992;5:297-323. doi:10.1007/BF00122574

34. Kahneman D, Tversky A. Prospect theory: an analysis of decision under risk. Econometrica. 1979;47(2):363-391. doi:10.2307/1914185

35. Zhao P, Su X, Ge T, Fan J. Propensity score and proximity matching using random forest. Contemp Clin Trials. 2016;47:85-92. doi:10.1016/j.cct.2015.12.012

36. Watkins S, Jonsson-Funk M, Brookhart MA, Rosenberg SA, O’Shea TM, Daniels J. An empirical comparison of tree-based methods for propensity score estimation. Health Serv Res. 2013;48(5):1798-1817. doi:10.1111/1475-6773.12068

37. Ferri-García R, Rueda MDM. Propensity score adjustment using machine learning classification algorithms to control selection bias in online surveys. PLoS One. 2020;15(4):e0231500. doi:10.1371/journal.pone.0231500

38. Wiecek E, Torres-Robles A, Cutler RL, Benrimoj SI, Garcia-Cardenas V. Impact of a multicomponent digital therapeutic mobile app on medication adherence in patients with chronic conditions: retrospective analysis. J Med Internet Res. 2020;22(8):e17834. doi:10.2196/17834

39. Thirumurthy H, Asch DA, Volpp KG. The uncertain effect of financial incentives to improve health behaviors. JAMA. 2019;321(15):1451-1452. doi:10.1001/jama.2019.2560

40. Garza KB, Owensby JK, Braxton Lloyd K, Wood EA, Hansen RA. Pilot study to test the effectiveness of different financial incentives to improve medication adherence. Ann Pharmacother. 2016;50(1):32-38. doi:10.1177/1060028015609354

41. Ghesla C, Grieder M, Schmitz J, Stadelmann M. Pro-environmental incentives and loss aversion: a field experiment on electricity saving behavior. Energy Policy. 2020;137:111131. doi:10.1016/j.enpol.2019.111131

42. Fryer RG Jr, Levitt SD, List J, Sadoff S. Enhancing the efficacy of teacher incentives through framing: a field experiment. Am Econ J Econ Policy. 2022;14(4):269-299. doi:10.1257/pol.20190287

43. Tversky A, Kahneman D. Loss aversion in riskless choice: a reference-dependent model. Q J Econ. 1991;106(4):1039-1061. doi:10.2307/2937956

44. Vervloet M, Linn AJ, van Weert JC, de Bakker DH, Bouvy ML, van Dijk L. The effectiveness of interventions using electronic reminders to improve adherence to chronic medication: a systematic review of the literature. J Am Med Inform Assoc. 2012;19(5):696-704. doi:10.1136/amiajnl-2011-000748

45. Basit SA, Mathews N, Kunik ME. Telemedicine interventions for medication adherence in mental illness: a systematic review. Gen Hosp Psychiatry. 2020;62:28-36. doi:10.1016/j.genhosppsych.2019.11.004

46. Khonsari S, Subramanian P, Chinna K, Latif LA, Ling LW, Gholami O. Effect of a reminder system using an automated short message service on medication adherence following acute coronary syndrome. Eur J Cardiovasc Nurs. 2015;14(2):170-179. doi:10.1177/1474515114521910

47. Lally P, Gardner B. Promoting habit formation. Health Psychol Rev. 2013;7(suppl 1):S137-S158. doi:10.1080/17437199.2011.603640