Background Primary nonadherence occurs when new prescriptions are not dispensed. Little is known about how to reduce primary nonadherence. We performed a randomized controlled trial to evaluate an automated system to decrease primary nonadherence to statins for lowering cholesterol.
Methods Adult members of Kaiser Permanente Southern California with no history of statin use within the past year who did not fill a statin prescription after 1 to 2 weeks were passively enrolled. The intervention group received automated telephone calls followed 1 week later by letters for continued nonadherence; the control group received no outreach. The primary outcome was a statin dispensed up to 2 weeks after delivery of the letter. Secondary outcomes included refills at intervals up to 1 year. Intervention effectiveness was determined by intent-to-treat analysis and Fisher exact test. Subgroups were examined using logistic regression.
Results There were 2606 participants in the intervention group and 2610 in the control group. Statins were dispensed to 42.3% of intervention participants and 26.0% of control participants (absolute difference, 16.3%; P < .001). The relative risk for the intervention vs control group was 1.63 (95% CI, 1.50-1.76). Intervention effectiveness varied slightly by age (P = .045) but was effective across all age strata. Differences in the frequency of statin dispensations persisted up to 1 year (P < .001).
Conclusions The intervention was effective in reducing primary nonadherence to statin medications. Because of low marginal costs for outreach, this strategy appears feasible for reducing primary nonadherence. This approach may generalize well to other medications and chronic conditions.
Clinical trials have confirmed the efficacy of many medications, yet taking medications as prescribed can be challenging to patients for a variety of reasons. Quiz Ref IDRetrospective studies1 of adherence to medications for the prevention of cardiovascular events among patients with coronary artery disease have found an association with improved cardiovascular outcomes. Decreased hospitalization costs appear to offset increased medication costs because of better adherence.2-4 The linkage between medication adherence and improved outcomes thus appears sufficient to support interventions to improve adherence, even though additional resources are required.5 Automated outreach to targeted populations may be a relatively efficient strategy to address nonadherence in large populations.
The problem of medication adherence is often divided into 2 main types: primary nonadherence, which occurs when a newly prescribed medication is never taken, and secondary nonadherence (persistence), which occurs when a medication is not taken as prescribed. Quiz Ref IDReasons for both primary and secondary nonadherence to chronic disease medications include 3 main patient factors: (1) side effects and safety, (2) need and effectiveness, and (3) affordability.6 Although reasons for primary and secondary nonadherence may overlap, the frequency of specific reasons may differ among types of nonadherence.6 For example, because patients have limited experience with a new medication, fear of side effects may be more common among patients with primary nonadherence. Primary nonadherence can be difficult to capture because measurement depends on prescription and dispensation data. There appear to be few trials on interventions to reduce primary nonadherence, although systematic reviews demonstrate a variety of interventions for secondary nonadherence.7,8
We tested an automated outreach intervention to increase dispensations of the prescribed medication near the prescription date. We chose to study dyslipidemia management as a clinical scenario that may typify common problems associated with primary nonadherence. We developed an intervention based on the belief that primary nonadherence might decrease if patients received encouraging prompts and information about the benefits of therapy. The intervention consisted of a telephone call followed later by a letter for patients with continued nonadherence.
This randomized controlled trial evaluated a strategy of automated messaging to prompt adherence to a newly prescribed 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitor (ie, statin) for dyslipidemia. The study was conducted from April through the middle of June 2010. Adherence rates were compared between an intervention group and a usual care (no population outreach) control group. The prescribing physician was not involved in the outreach. All data were passively collected from existing electronic records. The Southern California Kaiser Permanente Institutional Review Board approved the study. Informed consent was waived. All analyses were performed using SAS statistical software, version 9.1 (SAS Institute, Inc).
The study was conducted at Kaiser Permanente Southern California (referred to as the health plan), an integrated health care system that provides comprehensive care to 3.4 million members at 14 medical centers and 197 medical offices. Members have similar health service benefits. More than 90% of members have a pharmacy benefit that covers all or a portion of medication costs. Clinical practice guidelines developed by the health plan are routinely accessible for clinician use. Among these are guidelines for dyslipidemia control using statins, other medications, and behavioral change.
We identified a group of patients who were prescribed a statin as a new medication. A new medication was operationalized as a prescription for a statin or combination drug containing a statin and no record of such a drug dispensed within 365 days before the index prescription date. Participants were limited to those with 1 or more years of membership from the prescription date and no gap in enrollment more than 30 days during the past year. An age limit of 24 years and older at the time of the prescription was required because of infrequent statin prescriptions in younger individuals and controversial use of statins early in life.9 Members who had no record of the statin prescription being filled at a health plan pharmacy after 1 to 2 weeks were considered nonadherent and eligible for the study. The participant identification process is shown in Figure 1.
The goal of the intervention was to provide educational information and an encouraging prompt to adhere to a recently prescribed statin. The type and sequence of automated messaging were based on health plan outreach team experience and previous research showing the effectiveness of an automated telephone call followed by a letter for diabetes test monitoring.10Quiz Ref IDThe time frame for intervention was guided by prior work (led by T.C.C.) on dispensation rates in members who were prescribed a statin in the months just before the study. In these analyses, 18.4% of new statin prescriptions remained unfilled after 12 weeks. Among those who filled their prescription, 82.2% to 90.1% did so by 1 to 2 weeks after the prescription date. With a plan to conduct the intervention in weekly batches, it was determined that contact 1 to 2 weeks after the prescription date was a practical time frame for initiating outreach. This time frame allowed sufficient time for most of the patients to fill a prescription, thus reducing the need for outreach, while intervening soon enough to avoid an unhelpful delay.
Participants were contacted 1 to 2 weeks after the prescription date by an automated telephone call to retrieve a personalized message from the health plan (eAppendix). If no one answered, messages were left on answering machines directing participants to call a toll-free number to retrieve their message. Busy signals resulted in up to 2 more attempts to make telephone contact on subsequent days. Calls were made between 10 AM and 8 PM. One week after the initiation of calls, participants who still did not fill their prescription were sent a letter. The letters were expected to arrive 9 to 11 days after the first outreach contact by telephone. More than 95% of all health plan members have a telephone number on record, and more than 99% have an address.
Telephone calls began with a standard greeting saying that the message was from Kaiser Permanente. The message could be retrieved through interactive messaging during the call or by dialing a toll-free number. The personalized message conveyed that a statin drug was prescribed by their clinician and there was no record of the drug being dispensed by health plan pharmacies. The potential importance of the medication was described, and participants were encouraged to either have the prescription filled or contact the prescribing physician. The contact number of the local health plan pharmacy was provided. The telephone message was accessed in either English or Spanish and was approximately 40 seconds in duration. The letter was printed on one side in English and the other side in Spanish, and the text occupied approximately half a page. Letters were signed using the prescribing physician's name, a standard outreach practice in the health plan.
Enrollment, randomization, and study flow
Study participants were enrolled in weekly batches for 10 weeks (Figure 1). Each batch underwent the same process of study steps as shown in Figure 2. Enrollment and follow-up were passive. Members of the health plan who were prescribed a statin during a 1-week interval from Monday to Sunday were identified on Sunday night using HL7 electronic health records (Health Level Seven, Inc). The dates of prescriptions and dispensations were recorded. Algorithms were developed to identify and remove prescriptions that were canceled. Age and membership exclusions were applied. Pharmacy records were checked to exclude members who had a history of a statin dispensation during the year before the prescription date. On Sunday night, 1 to 2 weeks after the prescription date, pharmacy dispensation records were checked to exclude members who had filled their prescription. The remaining members were eligible for the study; all eligible members were enrolled.
On Monday morning, a quality check was manually performed by reviewing a sample of medical records for assurance that no obvious errors in electronic identification were made—a standard quality check for health plan outreach. A study programmer used computer-generated random numbers to sort participants into the intervention and control groups in equal proportion (day 0). Assignment was concealed from study investigators and analysts. Telephone contact information for the intervention group was sent to the telephone outreach organization for calls to be made (on days 0-2). One week after telephone calls were completed, dispensations from health plan pharmacies were checked. Address contact information for intervention group participants who still had no prescription dispensed was sent to the letter outreach organization (day 7) for printing (day 8) and mailing (day 9).
The primary outcome was dispensation of a statin between the first telephone call (day 0, the randomization day) and up to 2 weeks after the expected delivery of the letter, totaling 25 days from randomization and 32 to 39 days from the date of the prescription. Sensitivity analyses included outcomes 1 week earlier and 1 week later. This time frame was chosen to focus on primary nonadherence rather than secondary nonadherence. In a secondary analysis to evaluate long-term differences in medication dispensations, the number of dispensations was determined at 139, 253, and 367 days after randomization (114, 228, and 342 days after the primary outcome). These intervals were chosen because statins are often prescribed for a 100-day supply—the maximum allowed by the health plan—plus a 2-week delay for refilling. Therefore, by 139 days, for example, everyone who filled a prescription once during the primary study period had reasonable opportunity to obtain an “on schedule” refill, indicating secondary adherence.
Preliminary data identified a nonadherence rate of 25% to 30% after 1 to 2 weeks from the prescription date and 20% after 1 to 2 months under usual care conditions. We aimed for sufficient power to detect a 5% difference in adherence between the study arms based on a response rate of 20% in the control arm. Use of a significance level of .05, 90% power, a 2-sided test of proportions, and equal-sized groups required 1504 participants per group. Preliminary work suggested we would accumulate approximately 500 participants per week; because of the low costs of data collection and to ensure additional power for subgroup analyses, we planned a 10-week intervention for approximately 5000 participants.
Measurement of covariates
Covariates included age, sex, race/ethnicity, language preference (routinely collected by the health plan), and a pharmacy drug benefit that reduces the cost of prescribed medications. Socioeconomic indicators of income and educational level were gathered to further describe the cohort. These variables were imputed by address of residence and were not used in main results. Participants who died or disenrolled were detected using health plan vital statistics and administrative files. Health plan vital statistics historically miss 5% to 9% of deaths after later reconciliation with California Vital Records and the Social Security Death Index.
The effects of randomization were assessed with the χ2 and t tests. All participants' data were analyzed according to initial randomization (intent-to-treat) whether or not the participant was successfully contacted. Dispensations were compared between the intervention and control groups using 2-by-2 contingency tables and the Fisher exact test (2-sided) for dependency between the study groups. In secondary analyses, logistic regression was used to examine dispensations by participant characteristics. P < .05 was used to guide determination of statistical significance. Intervention effectiveness was examined by sex, age, race, and language preference using regression interaction terms.
There were 2606 participants in the intervention group and 2610 participants in the control group. Participant characteristics and the results of randomization are given in Table 1. The mean patient age was 56.1 years, and 50.6% of the participants were female. English was the preferred spoken language in 82.1% of participants and Spanish in 15.6%. The racial and ethnic distribution was similar to the entire health plan membership (data not shown). A total of 90.7% of participants had a pharmacy drug benefit. The mean (SD) low-density lipoprotein cholesterol level based on the last value during the year before the randomization date was 146 (36) mg/dL (to convert to millimoles per liter, multiply by 0.0259). Among those randomized, 17 had no telephone contact because of a nonworking number, 2 died, and 45 disenrolled from the health plan during the study period.
The main results (Table 2) indicate that Quiz Ref IDthe proportion of participants who were dispensed a statin 25 days after randomization and up to 2 weeks after expected delivery of the letter was greater in the intervention group compared with the control group (42.3% vs 26.0%; absolute difference, 16.3%; P < .001). The relative risk (RR) for adherence in the intervention vs control group was 1.63 (95% CI, 1.50-1.76); the odds ratio (OR) for comparison with logistic regression (Table 3) was 2.08 (95% CI, 1.85-2.35). The corresponding RR of nonadherence was 0.78 (95% CI, 0.75-0.81). When the outcome was measured 1 week earlier and 1 week later, the RR for adherence was similar (1.63 and 1.58, respectively; both P < .001). Secondary adherence (number of dispensations at 3 intervals in the following year) was always greater (P < .001) in the intervention arm (Table 4), although the difference decreased over time. Among those who were dispensed a medication during the study period, the refill rate was similar (eg, by 114 days, the proportion with a refill was 35.1% and 35.5% in the intervention and control groups, respectively).
Quiz Ref IDTwo participant characteristics were found to predict adherence regardless of study group (Table 3). Spanish speakers were more likely to have a statin dispensed (OR, 1.32; 95% CI, 1.06-1.65; P = .01), and a pharmacy drug benefit was strongly associated with increased adherence (OR, 10.05; 95% CI, 6.85-14.75; P < .001). Intervention effectiveness across subgroups is shown in Figure 3. A significant interaction with the intervention was found for age (P = .045). The intervention was effective in all age categories (P = .003 to < .001), although point estimates suggested a slightly greater effect in participants 50 years and older.
This method of population outreach using automated messaging to decrease primary medication nonadherence for a commonly prescribed cholesterol-lowering drug was highly successful. After detecting nonadherence to a prescription written 1 to 2 weeks earlier, a telephone call, followed a week later by a letter that reinforced the reasons to use the medication and encouraged action, resulted in a 16% absolute and 22% RR decrease in primary nonadherence at 25 days. In so doing, the size of the population with primary nonadherence was reduced with an automated program that is easily expanded. Given how common primary nonadherence appears to be for chronic diseases,6 if similar results are realized for different medications and conditions, the population effect is potentially large.
The intervention was designed to address primary nonadherence and leave to other programs the issue of secondary nonadherence. It is therefore encouraging to see a difference in effect up to 1 year, albeit decreased in size. The difference between the study groups in long-term adherence appears to be due to increased numbers of participants starting use of the medication because the refill rate among all those dispensed a medication during the study period was essentially the same (35.1% and 35.5% in the intervention and control groups, respectively) in both study groups. This rather disappointing measure of first refill and the overall low rate of “on schedule” dispensations up to 1 year point to the difficult problem of secondary medication adherence. The practice of pill splitting might lead to an underestimation of secondary adherence but should not affect the difference between study groups.
The intervention was similarly effective across sex and age groups. We found no evidence of differences in response by race, although our results may have been affected by 22.9% missing data. Spanish speakers were more likely to have their prescription dispensed after 1 to 2 weeks regardless of study group, indicating less nonadherence or delayed dispensation requests. A drug benefit (as opposed to no drug cost coverage) was strongly associated with medication adherence, possibly indicating that costs affected the decision to fill the new prescription. It might be expected that the intervention would be less effective in patients without a drug benefit, but that was not the case. We did not collect data on differing copayment amounts for those with a drug benefit. Studies of changes in medication copayments have found an effect on medication adherence in the United States.4,11-16
The approach appeared to be well accepted by health plan members: in a survey study of a sample of intervention and control participants from this study (led by T.H.), there were no complaints or negative responses to the study intervention. The study intervention was similar to outreach commonly used in the health plan for other purposes. Although a detailed cost analysis was not attempted, the marginal costs of the telephone calls and mailings were approximately $1.70 per person. These costs were based on arrangements for large-scale outreach using existing vendors who processed our study participants like other patients scheduled for outreach.
Our study results may have varied with an alternative definition of primary nonadherence. For other chronic conditions and medications, the effects of our intervention may differ. Our intervention required access to data on both prescriptions and tracking of dispensed medications. Although the numbers of persons in health care systems with access to these data may increase substantially, many health care plans and private practices do not have access to both these types of data. Some medications would be filled outside the health plan in both study groups. The primary nonadherence rate we observed (18.4% after 12 weeks) was similar to studies6,17-19 of adherence in different settings (mean of approximately 15%), suggesting a problem of similar magnitude. In other Kaiser Permanente regions, primary nonadherence to cholesterol-lowering drugs ranged from 8.5% to 12.6% using end points of 60 and 30 days, respectively.18,19 These differences may be due to differing populations and definitions of primary nonadherence.
In summary, a telephone call followed by a letter 1 week later was effective in reducing primary nonadherence to a statin prescription in patients who had no record of a statin dispensed the year before and up to 2 weeks after the prescription date. The suitability for outreach to large populations makes this an attractive strategy to help reduce the numbers of patients with primary nonadherence and better target those who remain nonadherent with complementary, more resource-intensive programs.
Correspondence: Stephen F. Derose, MD, MS, Department of Research and Evaluation, Kaiser Permanente Southern California, 100 S Los Robles Ave, Second Floor, Pasadena, CA 91188 (Stephen.F.Derose@kp.org).
Accepted for Publication: July 21, 2012.
Published Online: November 26, 2012. doi:10.1001/2013.jamainternmed.717
Author Contributions:Study concept and design: Derose, Green, Tunceli, Cheetham, and Reynolds. Acquisition of data: Cheetham and Vansomphone. Analysis and interpretation of data: Derose, Green, Marrett, Tunceli, Cheetham, Chiu, Harrison, Reynolds, and Scott. Drafting of the manuscript: Derose. Critical revision of the manuscript for important intellectual content: Derose, Green, Marrett, Tunceli, Cheetham, Chiu, Harrison, Reynolds, Vansomphone, and Scott. Statistical analysis: Derose, Tunceli, and Chiu. Obtained funding: Derose. Administrative, technical, and material support: Green, Marrett, Harrison, Vansomphone, and Scott. Study supervision: Derose.
Conflict of Interest Disclosures: Ms Marrett is an employee of Merck. Dr Tunceli is an employee of Merck and owns stock in the company.
Funding/Support: This research was supported by grants from Merck Sharp & Dohme Corp, a subsidiary of Merck & Co Inc, Whitehouse Station, New Jersey.
Additional Contributions: The Kaiser Permanente Southern California outreach team personnel supported the implementation of this study. Portia Summers, BS, reviewed and prepared the manuscript. Sponsors were consulted in the design of the study, including the analytic approach, and provided interpretation of results.
1.Ho PM, Magid DJ, Shetterly SM,
et al. Medication nonadherence is associated with a broad range of adverse outcomes in patients with coronary artery disease.
Am Heart J. 2008;155(4):772-77918371492
PubMedGoogle ScholarCrossref 2. McCombs JS, Nichol MB, Newman CM, Sclar DA. The costs of interrupting antihypertensive drug therapy in a Medicaid population.
Med Care. 1994;32(3):214-2268145599
PubMedGoogle ScholarCrossref 3.Muszbek N, Brixner D, Benedict A, Keskinaslan A, Khan ZM. The economic consequences of noncompliance in cardiovascular disease and related conditions: a literature review.
Int J Clin Pract. 2008;62(2):338-35118199282
PubMedGoogle ScholarCrossref 4.Choudhry NK, Fischer MA, Avorn J,
et al. The implications of therapeutic complexity on adherence to cardiovascular medications.
Arch Intern Med. 2011;171(9):814-82221555659
PubMedGoogle ScholarCrossref 5.Ho PM, Bryson CL, Rumsfeld JS. Medication adherence: its importance in cardiovascular outcomes.
Circulation. 2009;119(23):3028-303519528344
PubMedGoogle ScholarCrossref 6.Gadkari AS, McHorney CA. Medication nonfulfillment rates and reasons: narrative systematic review.
Curr Med Res Opin. 2010;26(3):683-70520078320
PubMedGoogle ScholarCrossref 7.Schedlbauer A, Davies P, Fahey T. Interventions to improve adherence to lipid lowering medication.
Cochrane Database Syst Rev. 2010;(3):CD00437120238331
PubMedGoogle Scholar 8.Haynes RB, Ackloo E, Sahota N, McDonald HP, Yao X. Interventions for enhancing medication adherence.
Cochrane Database Syst Rev. 2008;(2):CD00001118425859
PubMedGoogle Scholar 9.Pletcher MJ, Hulley SB. Statin therapy in young adults: ready for prime time?
J Am Coll Cardiol. 2010;56(8):637-64020705221
PubMedGoogle ScholarCrossref 10.Derose SF, Nakahiro RK, Ziel FH. Automated messaging to improve compliance with diabetes test monitoring.
Am J Manag Care. 2009;15(7):425-43119589010
PubMedGoogle Scholar 11.Fung V, Mangione CM, Huang J,
et al. Falling into the coverage gap: Part D drug costs and adherence for Medicare Advantage prescription drug plan beneficiaries with diabetes.
Health Serv Res. 2010;45(2):355-37520050931
PubMedGoogle ScholarCrossref 12.Gleason PP, Gunderson BW, Gericke KR. Are incentive-based formularies inversely associated with drug utilization in managed care?
Ann Pharmacother. 2005;39(2):339-34515644478
PubMedGoogle ScholarCrossref 13.Hsu J, Price M, Huang J,
et al. Unintended consequences of caps on Medicare drug benefits.
N Engl J Med. 2006;354(22):2349-235916738271
PubMedGoogle ScholarCrossref 14.Maciejewski ML, Farley JF, Parker J, Wansink D. Copayment reductions generate greater medication adherence in targeted patients.
Health Aff (Millwood). 2010;29(11):2002-200821041739
PubMedGoogle ScholarCrossref 15.Yang W, Kahler KH, Fellers T,
et al. Copayment level, treatment persistence, and healthcare utilization in hypertension patients treated with single-pill combination therapy.
J Med Econ. 2011;14(3):267-27821446895
PubMedGoogle ScholarCrossref 16.Choudhry NK, Avorn J, Glynn RJ,
et al; Post-Myocardial Infarction Free Rx Event and Economic Evaluation (MI FREEE) Trial. Full coverage for preventive medications after myocardial infarction.
N Engl J Med. 2011;365(22):2088-209722080794
PubMedGoogle ScholarCrossref 17.Fischer MA, Stedman MR, Lii J,
et al. Primary medication non-adherence: analysis of 195,930 electronic prescriptions.
J Gen Intern Med. 2010;25(4):284-29020131023
PubMedGoogle ScholarCrossref 18.Karter AJ, Parker MM, Moffet HH, Ahmed AT, Schmittdiel JA, Selby JV. New prescription medication gaps: a comprehensive measure of adherence to new prescriptions.
Health Serv Res. 2009;44(5, pt 1):1640-166119500161
PubMedGoogle ScholarCrossref 19.Raebel MA, Ellis JL, Carroll NM,
et al. Characteristics of patients with primary non-adherence to medications for hypertension, diabetes, and lipid disorders.
J Gen Intern Med. 2012;27(1):57-6421879374
PubMedGoogle ScholarCrossref