The effect of interactive reminders on medication adherence: A randomized trial

The effect of interactive reminders on medication adherence: A randomized trial

Accepted Manuscript The effect of interactive reminders on medication adherence: A randomized trial Hengchen Dai, David Mao, Kevin G. Volpp, Heather ...

493KB Sizes 0 Downloads 30 Views

Accepted Manuscript The effect of interactive reminders on medication adherence: A randomized trial

Hengchen Dai, David Mao, Kevin G. Volpp, Heather E. Pearce, Michael J. Relish, Victor F. Lawnicki, Katherine L. Milkman PII: DOI: Reference:

S0091-7435(17)30266-9 doi: 10.1016/j.ypmed.2017.07.019 YPMED 5098

To appear in:

Preventive Medicine

Received date: Revised date: Accepted date:

19 January 2017 23 May 2017 14 July 2017

Please cite this article as: Hengchen Dai, David Mao, Kevin G. Volpp, Heather E. Pearce, Michael J. Relish, Victor F. Lawnicki, Katherine L. Milkman , The effect of interactive reminders on medication adherence: A randomized trial, Preventive Medicine (2017), doi: 10.1016/j.ypmed.2017.07.019

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT 1 The Effect of Interactive Reminders on Medication Adherence: A Randomized Trial Hengchen Dai, PhDa; David Mao, BAb; Kevin G. Volpp, MD, PhDb,c; Heather E. Pearce, RPhd; Michael J. Relish, MSd; Victor F. Lawnicki, PhDd; Katherine L. Milkman, PhDb,c a

Olin Business School, Washington University in St. Louis. 1 Brookings Dr, St. Louis, MO 63130, USA

b

The Wharton School, University of Pennsylvania. 3620 Locust Walk, Philadelphia, PA 19104, USA

c

The Perelman School of Medicine, University of Pennsylvania. 3400 Civic Center Blvd, Philadelphia,

T

PA 19104, USA d

IP

Humana Inc. 500 West Main Street, Louisville, KY 40202, USA

CR

Corresponding author: Hengchen Dai [email protected]

Keywords: Reminder Systems; Communication; Behavioral; Medication Adherence; Health Promotion

CE

PT

ED

M

AN

US

Abstract: Expanding on evidence that interventions to improve health are more effective when informed by behavioral science, we explore whether reminders designed to harness behavioral science principles can improve medication adherence. We conducted a randomized controlled trial with 46,581 U.S. participants with commercial or Medicare Advantage insurance from Humana. Participants were randomly assigned to one of four experimental conditions. Participants in the usual care condition only received standard mailings that the insurer usually sends. In addition to the standard mailings, participants in the other three conditions also received (1) mailings that reminded them to take a target medication (basic reminder condition), (2) reminders that prompted them to predict their medication adherence in the next 30 days (prediction condition), or (3) reminders that prompted them to commit to a self-determined level of adherence for the next 30 days (commitment condition). We sent these mailings once a month for three months from November, 2014 through January, 2015, and tracked prescription refills. We find that, during the mailing period, reminders increased adherence by 0.95 percentage points (p < 0.05), and this effect was driven by the prediction and commitment conditions; during the three-month post-mailing period, reminders increased adherence by 0.98 percentage points (p < 0.05), and this effect was driven by the basic reminder and commitment conditions. The reminders increased medication adherence by 0.7 pills per dollar spent over our 181 day study period. Word count of the main text: 1996 (maximum = 2000 words); Word count of the abstract: 230 (maximum

AC

= 250 words)

Trial registry name: Prediction of Behavior Registration identification number: NCT02411006 URL for the registry: https://clinicaltrials.gov/ct2/show/NCT02411006

ACCEPTED MANUSCRIPT 2 Introduction The low rate of medication adherence in the United States is estimated to cost Americans between $100 billion and $289 billion annually.1 Forgetfulness is considered to be a key contributor to patients’ failure to take their medications as prescribed2. To overcome forgetfulness, a number of recent studies have examined whether devices that remind people to take their prescribed medication improve medication adherence. We conducted a large randomized controlled trial (RCT) to examine the efficacy

T

of not only basic reminders that simply tackle forgetfulness but also reminders that are designed to tackle

IP

another barrier of adherence—procrastination and lack of motivation3,4. We evaluated their impact on medication adherence over a three-month intervention period and three-month post-intervention period

CR

using data on prescription refills.

Our basic reminders tackle forgetfulness by simply encouraging people to take pills regularly.

US

Inspired by the potential of behavioral science for promoting health behavior,5,6 our interactive reminders are designed to leverage several behavioral science principles that may increase patients’ motivation to

AN

refill prescriptions and take medicines as prescribed. First, past research has shown that merely asking people about their intentions and plans to engage in a behavior (e.g., to make blood donations, to get vaccinations) can increase actual engagement.7,8 One explanation for this effect is that people prefer to

M

behave consistently with intentions that they have stated, even when their intentions are private.8,9 Second, research has shown that setting specific goals motivates people to work harder and achieve better

ED

performance than they would in the absence of such goals.10 Building on these insights, we designed two types of interactive reminders that prompted people to form intentions about taking medications and

PT

encouraged them to set specific adherence goals. One type of reminder prompted participants to predict their future medication adherence levels, and another type of reminder prompted participants to commit to

CE

a medication adherence level. Our RCT compares receiving no reminders with receiving reminders and explores whether interactive reminders are particularly impactful.

AC

Methods

In November 2014, 46,581 participants with employer-sponsored commercial or Medicare Advantage insurance from Humana were deemed eligible for inclusion in our randomized controlled trial. These participants were taking oral medication for cholesterol, diabetes, or blood pressure control, were expected to refill their medication every 30 days, and had 40-80% medication adherence in the 12 months prior to randomization (Figure 1). Participants were enrolled by Humana. This study involved four primary experimental conditions. After consulting with Humana, we set our intervention period to three months—a reasonable starting point for examining the impact of our intervention on adherence among participants with 30-day scripts. Participants in the usual care condition received standard engagement messages that the insurer regularly sends. Participants in the other three

ACCEPTED MANUSCRIPT 3 conditions received three additional reminder mailings, with each mailing sent around the 15th of each month from November, 2014 through January, 2015. In the basic reminder condition, these mailings simply reminded participants to take a target medication (for cholesterol, diabetes, or blood pressure).11 In the prediction condition, our mailings prompted participants to predict the number of days in the next 30 when they would take their medication.7-10 In the commitment condition, the mailings prompted participants to commit to a self-determined number of days over the next 30 to take their medication.7-10

T

All reminder mailings included a sticky note that participants could place somewhere (e.g. on a

IP

refrigerator) as a reminder, and those in the prediction and commitment conditions were encouraged to record their prediction or commitment on the sticky note.

CR

Furthermore, inspired by past research showing that people are more likely to follow through on plans that are made public,9 we randomly assigned half of the participants in the prediction and

US

commitment conditions to receive encouragement to share their prediction or commitment with Humana online, via text, or via postcard. Thus, both the prediction and commitment conditions consisted of two

AN

sub-groups. After institutional review board waiver of informed consent, researchers at the University of Pennsylvania used R, version 3.1.1 (R Core Team) to randomly assign participants in a 1:1:1:1:1:1 ratio to one of six groups, without blocking or other restrictions (Figure 1).

M

Statistical analyses

We used pharmacy claims data to calculate each participant’s proportion of days covered (PDC),

ED

defined as the number of days when he/she had any pills in the medication category listed on his/her mailing divided by the number of days in the observation period. In our primary analysis, we use PDC as

PT

our proxy for adherence, a continuous measure that is widely used by researchers to measure medication adherence.12-15 We used ordinary least squares regressions in STATA, version 14 (StataCorp) to predict

CE

PDCs on an intent-to-treat basis, controlling for PDC during the six months prior to the start of our experiment (May 15, 2014 to November 14, 2014), gender, ethnicity, the linear and squared terms of age,

AC

log-transformed income, and the type of medication that was listed on the reminders. The regressions had an 80% power to detect a minimum of a 1.2-percentage-point difference between the usual care and each treatment condition with α=.05. We also evaluated adherence using a binary measure of optimal adherence which was set equal to one if a patient’s PDC in a given period was equal to or greater than 80%, and zero otherwise. This less sensitive measure of adherence has also been widely used in past research, and we used logistic regressions to predict this secondary outcome variable.14,15 Results All conditions were balanced on available demographic variables (age, gender, ethnicity, income) and adherence prior to our randomized control trial, suggesting that our randomization was successful. Encouraging participants to publicly share their prediction or commitment (i.e., by conveying it to

ACCEPTED MANUSCRIPT 4 Humana) did not significantly influence adherence rates (see supplemental materials for details). Thus, we focus on the differences between our usual care, basic reminder, prediction, and commitment conditions below. During the mailing period (November 15, 2014 to February 14, 2015), mean PDC in the usual care condition was 60.69%. The three reminder conditions, combined, increased PDCs by 0.95 percentage points (95% confidence interval [CI], 0.26% to 1.65%)—a significant but small amount. This effect was

T

driven by prediction and commitment reminders, which increased PDCs by 1.07 percentage points (95%

IP

CI, 0.30% to 1.85%) and 1.00 percentage points (95% CI, 0.22% to 1.77%), respectively. Basic reminders insignificantly increased PDCs by 0.64 percentage points (95% CI, -0.26% to 1.53%).

CR

During the post-mailing period (February 15 to May 14, 2015), mean PDC in the usual care condition was 58.20%. The three reminder conditions, combined, increased PDCs significantly by 0.98

US

percentage points (95% CI, 0.15% to 1.81%). This was driven by basic and commitment reminders, which increased PDCs by 1.21 percentage points (95% CI, 0.14% to 2.28%) and 1.09 percentage points

AN

(95% CI, 0.16% to 2.02%), respectively. Prediction reminders insignificantly increased PDCs by 0.76 percentage points (95% CI, -0.17% to 1.69%). Table 1 summarizes these results. Our secondary analyses considering ―optimal‖ adherence (PDC≥80%) as a binary outcome measure yield directionally similar

M

results, though statistical significance is weaker for this less sensitive outcome (Supplementary Table 1). Subgroup analysis

ED

We separated our sample by median PDC during the six-month pre-mailing period (57.07%) and estimated the effect of our intervention on each subgroup.16 Among the participants whose PDC was

PT

above median prior to the mailing period, the three reminder conditions, combined, did not significantly improve medication adherence during the mailing period (mean difference from the usual care condition

CE

= 0.46 percentage points, 95% CI, -0.40% to 1.31%) or during the post-mailing period (mean difference from the usual care condition = 0.45 percentage points, 95% CI, -0.68% to 1.58%). Among participants

AC

whose PDC was below median prior to the mailing period, the three reminder conditions, combined, significantly improved medication adherence by 1.50 percentage points (95% CI, 0.41% to 2.58%) during the mailing period and 1.54 percentage points (95% CI, 0.33% to 2.76%) during the post-mailing period, compared to the usual care condition. We observed similar patterns when we compared different types of reminders with the usual care condition (rather than pooling all reminders before making this comparison). Participants with above-median baseline PDC did not statistically significantly benefit from any type of reminder during the mailing or post-mailing period; participants with below-median baseline PDC benefited significantly from prediction and commitment reminders during the mailing period, and from basic reminders and commitment reminders during the post-mailing period. Table 1 summarizes these results. However, note that in our OLS regressions to predict PDCs, the interaction terms between

ACCEPTED MANUSCRIPT 5 indicators for our reminder conditions and an indicator for having a below-median (vs. above-median) baseline PDC are not statistically significant, indicating that the qualitatively better adherence observed among those with relatively lower PDCs described above is not significant. Discussion We show that reminders, especially interactive ones, improve medication adherence and that the effects persist even after mailings stop. Our subgroup analysis suggests that reminder mailings may be

T

more effective for those who have low prior adherence.

IP

Overcoming low medication adherence is a challenge. Even RCTs that test expensive, aggressive reminders (e.g., pill bottle caps with a digital timer) show null effects.16 The effects of our intervention are

CR

small in terms of absolute magnitude, and if implemented alone, reminder mailings would have limited clinical impact. However, mailed reminders can easily supplement more intensive adherence

US

interventions at very low cost. They scale easily and are inexpensive; in contrast, methods that have been shown to successfully improve medication adherence to a substantial degree are often complex to scale,

AN

labor-intensive, and expensive.12,14

As an example, we compare our intervention with a highly-cited program that reduced medication copayments and significantly increased adherence by 5.4 percentage points. We calculate the increase in

M

medication adherence per dollar spent using the following formula: (the average increase in PDC per person due to an intervention) × (the number of days in the observation period) / (the cost of the

ED

intervention per person). In our study, reminders increased the average PDC by 0.95 percentage points during our 92-day mailing period and by 0.98 percentage points during our 89-day post-mailing period.

PT

Thus, our reminder intervention was estimated to increase the number of pills taken by 0.7 pills per dollar spent over 181 days, or (0.95%*92 + 0.98%*89)/$2.535. In comparison, reducing copayments was

CE

estimated to produce an average increase of 0.02 pills per dollar spent over 394 days.15 We use the median observation period (394 days), average cost ($926), and average increase in PDC (5.4 percentage points)

Limitations

AC

in our estimate (i.e., (5.4%*394)/$926).

Several limitations of this study should be acknowledged. First, once a month, on average, all of our study participants received usual care mailings from their insurer, which sometimes contained medication adherence reminders, and participants may also have received reminders from other sources (e.g., the pharmacies where they refilled prescriptions). Thus, our observed reminder effect reflects the benefits of adding an (interactive) reminder to standard reminders from Humana and other sources, and our study may underestimate the impact that reminders could have on people who do not already receive any form of reminder.

ACCEPTED MANUSCRIPT 6 Second, limited by the data that the pharmacy benefit manager (PBM) had, we calculate medication adherence based on prescription refills. While prescription refill patterns are often used as a proxy for medication adherence,14,15 they do not measure actual pill intake. Changing medication intake is psychologically different than changing prescription refills. Unfortunately, it is extremely challenging, if not impossible, to measure actual medication intake, especially in large populations. Third, while our choice of tracking PDCs for three months during the post-mailing period is

T

consistent with the length of the follow-up periods in many other RCTs examining medication

IP

adherence,17,18 the long-term effects of our interventions beyond our follow-up period are unknown. Finally, we only examined customers with 30-day refill frequencies, and reminders may work differently

CR

for customers on different refill schedules. Conclusion

US

Our randomized controlled trial reveals that reminders – particularly interactive ones – have a positive and persistent, although modest, effect on medication adherence. Given the large challenge of

AN

overcoming poor medication adherence, a multifaceted approach seems necessary.14 Future investigations

AC

CE

PT

ED

M

of how to use reminders as a supplement to other medication adherence programs would be valuable.

ACCEPTED MANUSCRIPT 7 Conflict of Interest Disclosures All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Dai reports funding in part by grants from Humana during the conduct of the study. Mr Mao reports funding in part by grants from Humana during the conduct of the study. Dr Volpp reports grants from Humana during the conduct of the study as well as grants from Hawaii Medical Services Agency, Discovery (South Africa), Merck, Weight Watchers, and CVS outside of the submitted work; he

T

has received consulting income from CVS and VALHealth and is a principal in VALHealth, a behavioral

IP

economics consulting form. Dr Pearce reports employment by Humana, the funding organization, during the course of the study. Mr Relish reports employment by Humana, the funding organization, during the

CR

course of the study. Dr Lawnicki reports employment by Humana, the funding organization, during the course of the study. Dr Milkman reports grants from Humana during the conduct of the study and

US

personal fees from serving on the Health Prize Academic Advisory Board, outside the submitted work. No other disclosures were reported.

AN

Funding/Support: This research was funded by research grants from Humana. Role of the Funder/Sponsor: The funding organization, in collaboration with the academic investigators, provided input on the design and conduct of the study; oversaw the collection and management of the

M

data; and contributed to the preparation, review, and approval of the manuscript, as well as the decision to submit the manuscript for publication. The funding organization had no role in the analysis or

ED

interpretation of the data.

Additional Contributions: The authors thank Bradford Tuckfield, PhD, Berkadia, for his contributions

PT

to study conception and design, who received funding in part by grants from Humana during the conduct of the study; Hae Nim (Sunny) Lee, BA, University of Pennsylvania, for her contributions to study

CE

conception and design and data analysis, who received funding in part by grants from Humana during the conduct of the study; and Joelle Friedman, MPA, University of Pennsylvania, for her help with

AC

acquisition of data, who was not compensated for her contributions. The authors also thank Humana Inc., especially George J. Spurlock, MBA, Peinie P. Young, PharmD, BCACP, Shannon Clark, MBA, Vicki L. Vogel, BA, and Tova Levin, MBA, for their administrative and technical support.

ACCEPTED MANUSCRIPT 8 References 1. Viswanathan M, Golin CE, Jones CD, et al. Interventions to improve adherence to self-administered medications for chronic diseases in the United States: a systematic review. Ann Intern Med. 2012; 157(11): 785-795. 2. Gadkari AS, McHorney CA. Unintentional non-adherence to chronic prescription medications: how unintentional is it really? BMC Health Serv Res. 2012; 12(1): 98.

IP

Behaviour Questionnaire (BBQ). Patient Educ Couns. 2006; 64(1): 50-60.

T

3. George J, Mackinnon A, Kong DC, Stewart K. Development and validation of the Beliefs and

4. Israni A, Dean C, Kasel B, Berndt L, Wildebush W, Wang CJ. Why do patients forget to take

CR

immunosuppression medications and miss appointments: can a mobile phone app help?. JMIR Public Health Surveill. 2016; 2(1): e15.

US

5. 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.

AN

6. Li M, Chapman GB. Nudge to health: Harnessing decision research to promote health behavior. Soc Personal Psychol Compass. 2013; 7(3): 187-198.

M

7. Godin G, Sheeran P, Conner M, Germain M. Asking questions changes behavior: mere measurement effects on frequency of blood donation. Health Psychol. 2008; 27(2): 179-184.

ED

8. Milkman KL, Beshears J, Choi JJ, Laibson D, Madrian BC. Using implementation intentions prompts to enhance influenza vaccination rates. Proc Natl Acad Sci U S A. 2011; 108(26): 10415-10420.

PT

9. Cialdini R. Influence: The Psychology of Persuasion, Revised Edition. New York, New York: Harper Business; 2006.

CE

10. Locke EA, Latham GP. A theory of goal setting & task performance. Englewood Cliffs, New Jersey: Prentice-Hall, Inc; 1990. 11. Szilagyi PG, Bordley C, Vann JC, et al. Effect of patient reminder/recall interventions on immunization rates: a review. JAMA. 2000; 284(14): 1820-1827.

AC

12. Lee JK, Grace KA, Taylor AJ. Effect of a pharmacy care program on medication adherence and persistence, blood pressure, and low-density lipoprotein cholesterol: a randomized controlled trial. JAMA. 2006; 296(21): 2563-71. 13. Lin EH, Katon W, Rutter C, et al. Effects of enhanced depression treatment on diabetes self-care. Ann Fam Med. 2006; 4(1): 46-53. 14. Osterberg L, Blaschke T. Adherence to medication. N Engl J Med. 2005; 353(5): 487-97. 15. Choudhry NK, Avorn J, Glynn RJ, et al. Full coverage for preventive medications after myocardial infarction. N Engl J Med. 2011; 365(22): 2088-2097. 16. Choudhry NK, Krumme AA, Ercole PM, et al. Effect of reminder devices on medication adherence: The REMIND randomized clinical trial. JAMA Intern Med. 2017; 177(5): 624-631.

ACCEPTED MANUSCRIPT 9 17. Simoni JM, Pantalone DW, Plummer MD, Huang B. A randomized controlled trial of a peer support intervention targeting antiretroviral medication adherence and depressive symptomatology in HIVpositive men and women. Health Psychol. 2007; 26(4): 488-495.

AC

CE

PT

ED

M

AN

US

CR

IP

T

18. Murray MD, Young J, Hoke S, et al. Pharmacist intervention to improve medication adherence in heart failure: a randomized trial. Ann Intern Med. 2007; 146(10): 714-25.

ACCEPTED MANUSCRIPT

AN

US

CR

IP

T

10

AC

CE

PT

ED

M

Figure 1

ACCEPTED MANUSCRIPT 11

Table 1 Reminder Conditions Interactive Reminder: Prediction

Interactive Reminder: Commitment

15,500

15,503

47.64% 54.00% 18.90% 69.38 (11.59) $54,720

46.23% 53.74% 19.32% 69.32 (11.52) $54,644

46.04% 54.45% 18.97% 69.36 (11.37) $54,512

(46,974)

(46,974)

(48,226)

(48,054)

55.36%

55.33%

55.18%

55.37%

55.45%

(23.73%)

(23.48%)

(23.83%)

(23.85%)

(23.67%)

55.37% (23.78 %)

61.49% N/A N/A

60.67% N/A N/A

61.16% 0.50% 0.64%

61.76% 1.12%* 1.07%**

61.79% 1.09%* 1.00%*

61.65% 0.98%* 0.95%*

All Conditions Combineda

Usual Care

Basic Reminder

46,581

7,812

7,766

46.31% 54.20% 19.04% 69.32 (11.49) $54,621

45.66% 54.81% 18.79% 69.21 (11.56) $54,692

Number of participants Baseline sample characteristics % Male % Caucasian % Black Mean Age (SD) Mean Income (SD) Mean adherence (PDC) during the sixmonth pre-mailing period: May 15, 2014 to November 14, 2014 (SD) b

(47,983)

C A

D E

M

I R

C S

U N

A

T P

E C

Results - overall analysis During the three-month mailing period: November 15, 2014 to February 14, 2015 Mean medication adherence (PDC) Unadjusted increase OLS regression-adjusted

T P

Remind er Conditi ons, Combin ed 38,769 46.44% 54.07% 19.10% 69.35 (11.47) $54,606 (47,908 )

ACCEPTED MANUSCRIPT 12 increasec During the three-month post-mailing period: February 15, 2014 to May 14, 2015 Mean medication adherence (PDC) Unadjusted increase OLS regression-adjusted increase Results - subgroup analysisb,d Participants with above-median baseline PDC During the three-month mailing period: November 15, 2014 to February 14, 2015 Mean medication adherence (PDC) Unadjusted increase OLS regression-adjusted increase During the three-month post-mailing period: February 15, 2014 to May 14, 2015 Medication adherence (PDC) Unadjusted increase OLS regression-adjusted increase

*

59.39% 1.22%*

59.18% 1.02%*

0.76%

1.09%*

0.98%*

C A

58.17% N/A

59.26% 1.09%

N/A

N/A

1.21%*

58.94% 0.78%

76.36% 0.21%

76.89% 0.74%

76.87% 0.71%

76.78% 0.62%

N/A

0.11%

0.49%

0.60%

0.46%

71.43% N/A

T P

70.95% N/A

71.52% 0.57%

71.38% 0.43%

71.68% 0.73%

71.53% 0.57%

N/A

N/A

0.51%

0.24%

0.62%

0.45%

46.32%

45.29%

46.01%

46.76%

46.56%

46.53%

T P

I R

C S

U N

A

76.68% N/A

76.16% N/A

D E

N/A

E C

Participants with below-median baseline PDC During the three-month mailing period: November 15, 2014 to February 14, 2015 Mean medication adherence (PDC)

59.01% N/A

M

ACCEPTED MANUSCRIPT 13 Unadjusted increase OLS regression-adjusted

N/A

N/A

0.72%

1.47%*

1.27%+

increase N/A N/A 1.20%+ 1.73%** 1.41%* During the three-month post-mailing period: February 15, 2014 to May 14, 2015 Medication adherence (PDC) 46.62% 45.47% 47.04% 46.62% 46.98% Unadjusted increase N/A N/A 1.57%+ 1.15% 1.50%* OLS regression-adjusted increase N/A N/A 1.95%* 1.33%+ 1.56%* + p<0.1 * p<0.05 ** p<0.01 *** p<0.001 Standard deviations in parentheses a. Our full sample involved participants in the United States with commercial or Medicare Advantage insurance with Humana from November 15, 2014 to May 14, 2015. b. Increases are relative to the usual care condition.

T P

I R

C S

46.85% 1.37%* 1.55%*

U N

A

M

1.24%* 1.50%* *

c. Our regression controlled for PDC during the six months prior to the start of our experiment (May 15, 2014 to November 14, 2014), gender, ethnicity, the linear and squared terms of age, log-transformed income, and the type of medication that was listed on the reminders. d. We separated our sample by median PDC during the six-month pre-mailing period (57.07%) and estimated the effect of our intervention for each subgroup.

D E

T P

C A

E C

ACCEPTED MANUSCRIPT 14

AC

CE

PT

ED

M

AN

US

CR

IP

T

Highlights 1. Mailed reminders, especially interactive reminders, improve medication adherence 2. Increases in medication adherence persist after reminder mailings stop 3. Reminders can be an inexpensive supplement to other medication adherence programs