The American Journal of Managed Care January 2010
Mail-Order Pharmacy Use and Adherence to Diabetes-Related Medications
Patients with diabetes were more likely to have good medication adherence if they refilled their medications by mail versus at localpharmacies.
Objective: To investigate whether patients who use mail-order pharmacies were more likely to have good medication adherence than patients who use local pharmacies.
Study Design: Cross-sectional investigation.
Methods: We conducted cross-sectional analyses of patients from the Kaiser Permanente Northern California (KPNC) diabetes registry who received a new antiglycemic, antihypertensive, or lipid-lowering index medication between January 1, 2006, and May 31, 2006. We defined good adherence as medication availability at least 80% of the time (ie, a continuous measure of medication gaps value of ≤20%) and compared adherence between mail-order users (≥66% of refills by mail) and KPNC local pharmacy users (all refills in person). Adherence was calculated from the initial dispensing through 15 months of follow-up, medication discontinuation, or May 31, 2007, whichever came first. We analyzed the data using multivariate logistic regression models, after determining that unmeasured patient-level factors and self-selection did not significantly bias our analyses.
Results: A total of 13,922 eligible patients refilled an index medication. Compared with those who used only local KPNC pharmacies, patients who received medications by mail were more likely to have good adherence (84.7% vs 76.9%, P <.001). After adjusting for potential confounders, including days’ supply and out-of-pocket costs, mailorder users had better adherence to antiglycemic, antihypertensive, and lipid-lowering medications (P <.001 for all).
Conclusions: Compared with patients who obtained medication refills at local pharmacies, patients who received them by mail were more likely to have good adherence. The association between mail-order use and medication adherence should be evaluated in a randomized clinical trial.
(Am J Manag Care. 2010;16(1):33-40)
In this study of an integrated delivery system, the use of mail-order pharmacies was associated with a higher likelihood of good medication adherence.
- Minority patients and patients of low socioeconomic status were less likely to refill their medications by mail.
- Although these findings should be confirmed in other analyses and in a randomized controlled trial, they suggest the intriguing potential of organization-level interventions to improve adherence, particularly among vulnerable populations, by promoting increased mail-order use.
The use of mail-order pharmacies may streamline the refill process because mailed medications eliminate the need for travel to the pharmacy. Home delivery may be particularly beneficial for patients with disabilities, inadequate access to transportation, or time constraints.10 Patients who use mail-order pharmacies purchase more medications than patients using local walk-in pharmacies.11-13 However, this difference could represent factors other than better adherence such as self-selection into mail order by adherent patients. In addition, most mail-order pharmacies mandate that patients purchase a 90-day medication supply, unlike local pharmacies. Medication purchases by mail-order users may not reflect actual adherence if providers alter the regimen after 30 to 60 days and patients are obligated to waste pills.13 Although one-third of chronic disease prescriptions in the United States are filled by mail,14 there has been little evaluation of patients’ propensity to use mail-order pharmacies or whether the use of mail-order pharmacies is associated with medication adherence.
Among patients with diabetes in an integrated delivery system, we examined (1) the demographic and clinical factors associated with the use of mail-order pharmacies and (2) the likelihood of good adherence to newly prescribed antiglycemic, antihypertensive, and lipid-lowering medications, comparing mail-order pharmacy users with local pharmacy users. We used a data set in which the days’ supply is similar for mail-order and local pharmacies, and we used statistical techniques to assess for selection bias. We hypothesized that patients who used a mail-order pharmacy would be more likely to have good medication adherence over a 12-month period compared with patients who used only local pharmacies.
We conducted this study within Kaiser Permanente Northern California (KPNC), a fully integrated health system that provides comprehensive medical care to more than 3 million members. The KPNC membership includes employed persons and retirees and approximates the general population of northern California racially/ethnically and socioeconomically. The study protocol was approved by the KPNC Institutional Review Board. We selected subjects from the KPNC diabetes registry, established in 1993.15 The registry is updated annually by adding patients with diabetes identified from automated databases of pharmacy data, laboratory data, hospitalization records, and outpatient diagnoses.
Kaiser Permanente Northern California has more than 120 local walk-in pharmacies, located on-site within outpatient clinics and hospital facilities. We identified each patient’s clinic and local pharmacy based on his or her prior utilization or the utilization patterns of other patients in the same zip code.16 Since 1999, KPNC has maintained a mail-order pharmacy distribution system, in coordination with local KPNC pharmacies. After completing a simple enrollment process, patients can obtain medications by mail. Although most new prescriptions are filled in the local pharmacy, KPNC patients also have the option of filling new prescriptions by mail, with telephone access to a pharmacist to answer any medication-related questions. There is no minimum days’ supply required for mail-order delivery at KPNC, which typically dispenses 100-day medication supplies through mail-order and local KPNC pharmacies. Some patients have a financial incentive to use mail order, typically a lower copayment for the same days’ supply obtained by mail versus at a local KPNC pharmacy.
Study Design and Participants
From the diabetes registry, we selected a subset (n = 23,488) of patients who were aged at least 18 years by January 1, 2006; had a pharmacy benefit; and had been prescribed a new diabetes-related medication (no recorded use in the previous 24 months) between January 1, 2006, and May 31, 2006. To most accurately estimate pharmacy utilization, we excluded members who lacked drug benefits during the study period and had little incentive to use KPNC pharmacies (approximately 4% of patients). Diabetes-related medications included antiglycemic, antihypertensive, or lipid-lowering medications. If patients were prescribed multiple new diabetes-related medications during this period, the earliest prescription was defined as the index medication. We estimated adherence for this medication using a cross-sectional study design (described herein). The study window started at the initial dispensing and continued for 15 months, until discontinuation of the medication, or until May 31, 2007, whichever came first. Patients were considered to have discontinued their medication if they did not obtain a refill within 90 days after their existing supply had run out.
To assess adherence to the index medication, we used a pharmacy utilization–based measure of secondary adherence, the continuous measure of medication gaps (CMG).17-19 The CMG uses refill data to determine the cumulative period for which no medication was available to the patient (gaps), dividing the number of days for which the patient did not have the medication by the number of days in the study window for that participant. The CMG values range from 0% (completely adherent) to 100% (completely nonadherent). We dichotomized CMG values, classifying values of 20% or less as good and greater than 20% as inadequate. This cutoff has been used in previous studies3-5 that examined the relationship between medication adherence and hospitalizations and mortality rates. Because the CMG cannot be calculated without at least 1 complete refill interval, we excluded 5613 patients who did not fill their prescription at least twice. The CMG cannot be reliably calculated for insulin given the flexible dosing instructions, so we also excluded 658 patients whose index medication was insulin. Patients with oral index medications who were taking insulin concurrently were not excluded.
We defined 2 patient groups for our analytic comparisons: mail-order users and local pharmacy users. Because the proportion of patients using mail-order exclusively was small (<2% of the sample), we considered patients to be mail-order users (n = 2595) if they used this system predominantly (≥66% of refills by mail). Local pharmacy users (n = 11,327) obtained all their index medication refills in person. Patients who used mail-order intermittently (1%-65% of refills [n = 3295]) were excluded from the analysis.
We included several demographic characteristics in our analyses, including race/ethnicity (non-Latino white, African American, Latino, Asian/Pacific Islander, Native American, mixed race/ethnicity, and missing race/ethnicity). As a contextual measure of socioeconomic status, we used a neighborhood deprivation index20 based on principal components analysis of the following 8 census tract variables from the 2000 US Census: percentage of households in poverty, percentage of households receiving public assistance, percentage of female-headed households with children, percentage of households earning less than $30,000 annually, percentage with less than a high school education, percentage in crowded housing (>1 person per room), percentage unemployed, and percentage of men in management or professional occupations. The socioeconomic deprivation score was standardized to a mean (SD) of 0 (1), with negative scores indicating less deprived neighborhoods.
The number of comorbid conditions for each patient was ascertained from the KPNC outpatient clinical records database for the 18 months before the first refill. Patients were classified by their type of insurance (non-Medicare commercial, Medicare Part D with a group insurer, or Medicare Part D from the individual insurance market) and by whether they had a financial incentive to order by mail. We calculated the linear distance from the patient’s home address to his or her local KPNC pharmacy using MapMarker geocoding software (Pitney Bowes MapInfo, Troy, NY). We categorized the days’ supply of the index medication dispensed at the first refill into 1 to 30, 31 to 60, 61 to 90, or more than 90 days. Finally, we measured the duration of therapy for the index medication, defined as the number of days from the first fill through the last fill before discontinuation or the end of the study window.
We estimated separate logistic regression models using Stata version 9.2 (StataCorp LP, College Station, TX) to examine (1) variables associated with mail-order pharmacy use and (2) differences in the probability of good medication adherence between mail-order and local pharmacy users. In each analysis, we included the following as covariates: age; sex; race/ethnicity; neighborhood socioeconomic deprivation score; number of comorbidities; smoking status; use of nonformulary medications, antidepressants, or insulin; insurance type; financial incentive to order by mail; distance from the patient’s home to the local pharmacy; an indicator for whether the index medication was generic or brand name; length of therapy with the index medication; and medical facility dummy variables. To facilitate interpretability of the results, we estimated predicted percentages (95% confidence intervals) for groups of interest, holding all other covariates at their mean value.
Observational studies are susceptible to more bias than experimental ones when studying causal relationships. Omitted (unmeasured) variable bias, or self-selection (eg, differences in intrinsic patient motivation to use mail order), is of particular concern in this study. We used health econometric techniques, specifically a bivariate probit (BVP) model incorporating an instrumental variable, to assess whether our estimates from the logistic regression predicting medication adherence were subject to omitted variable bias. A nonlinear BVP model is more appropriate than a standard linear instrumental variable model when the outcome (eg, medication adherence) and the treatment variable (eg, mail-order pharmacy use) are binary.21,22