Predictors of Pharmacy-Based Measurement and Self-Report of Antidepressant Adherence: Are Individuals Overestimating Adherence?

Participants were Veterans Health Administration (VHA) patients enrolled in a randomized controlled trial examining a peer-support program conducted from February 2010 through September 2013 that was aimed at improving depressive symptoms and quality of life. This study was a secondary analysis of participants’ self-report assessments and VA pharmacy records at baseline, six months, and 12 months. Participants were recruited from 15 VHA mental health clinics affiliated with four Midwestern U.S. Department of Veterans Affairs (VA) health care systems, including community-based outpatient centers. Patients were eligible for the study if they were receiving VA mental health treatment for depression (but not using mental health providers outside the VA), had significant levels of depressive symptoms (score ≥10 on the Patient Health Questionnaire–9) or current decrements in functioning (score ≥10 on the Work and Social Adjustment Scale), and had at least one prior trial of psychotherapy or an antidepressant. Participants with a diagnosis of active substance dependence in the past 12 months or substance abuse in the past six months were ineligible. A total of 443 eligible patients enrolled in the study. The study was approved by each VA’s institutional review board, and written informed consent was obtained from all participants.

We assessed predictors of adherence at six and 12 months as measured by MPRs and self-report on the Brief Medication Questionnaire (BMQ). BMQ results and MPRs were available for 281 patients at six months and for 265 patients at 12 months, MPRs only were available for 96 patients at six months and for 99 patients at 12 months, and BMQ results only were available for ten patients at six months and for 12 patients at 12 months.

MPRs were constructed by using VA pharmacy data by dividing the days’ supply of antidepressant medication received by the days’ supply needed for a patient to take prescribed doses from baseline to six months and from six to 12 months. An MPR of ≥80% was used as the threshold for good adherence (11). If a participant switched antidepressants, we assumed the old drug was taken until the new one was started. If multiple antidepressants were taken concurrently, a weighted average of both drugs’ MPR was used.

The BMQ was used to assess self-reported adherence from baseline to six months and from six to 12 months. Poor adherence was defined as reporting any missed pills or taking medication for fewer than seven days in the past week.

Demographic predictors included gender, race (white and other), Hispanic ethnicity, age group (young adults, ages 18–39; middle-aged adults, ages 40–64; and older adults, age ≥65), occupation (employed, retired, and unemployed), income (<$15,000; $15,000 to <$25,000; $25,000 to <$50,000; and ≥$50,000), married or cohabiting, and education (high school or less, some college, and college or more).

Social support was assessed with the Interpersonal Support Evaluation List (12). The scale ranges from 0 to 120, with higher scores indicating greater levels of support.

The Depression Coping Self-Efficacy Scale assesses confidence in one’s ability to perform coping activities for depression (13), with higher scores indicating greater self-efficacy.

The severity of depression was assessed at baseline with the Beck Depression Inventory (scores of 0–13 indicating minimal depression; 14–19, mild; 20–28, moderate; and ≥29, severe) (14). Side effects were assessed with an open-ended BMQ question asking whether any of the medications taken by the patients bothered them in any way. If participants reported that their antidepressants were bothersome, they were categorized as experiencing a side effect (any versus none). Participants were considered to have a complex medication regimen if they were taking drugs from two or more classes of psychotropic medications (antidepressants, anticonvulsants, antianxiety drugs, and antipsychotics).

Descriptive statistics were calculated for all study variables. Logistic regressions were run by using the previously listed demographic, social, psychological, and clinical variables as predictors of pharmacy-based adherence (MPRs) and self-reported adherence (BMQ results) at six and 12 months.

[A complete list of sample characteristics is available as an online supplement to this report.] At six months, MPRs were available for 377 participants, and BMQ results were available for 291 participants. Of these, 36% (N=135) of participants were considered poorly adherent on the basis of MPRs and 24% (N=71) were considered poorly adherent on the basis of BMQ results. Nearly 22% (N=40) of the 183 participants who were considered adherent on the basis of their MPRs were considered nonadherent on the basis of their BMQ responses. Over 32% (N=68) of the 211 participants who were considered adherent on the basis of their BMQ were considered nonadherent on the basis of their MPRs.

At 12 months, MPRs were available for 364 participants, and BMQ results were available for 277 participants. Of these, 35% (N=126) of participants were considered poorly adherent on the basis of MPRs, and 18% (N=49) were considered poorly adherent on the basis of the BMQ. Thirteen percent (N=24) of the 185 participants who were considered adherent on the basis of their MPRs were considered nonadherent on the basis of the BMQ. More than 26% (N=57) of the 218 participants who were considered adherent on the basis of their BMQ were considered nonadherent on the basis of their MPRs. Results for MPRs were available at both six and 12 months for 352 participants, and BMQ results were available at both six and 12 months for 237 participants. In a display of the dynamic nature of adherence, MPR results for participants with MPRs at both follow-ups indicated that adherence status (good versus poor) changed between six and 12 months for 27% (N=96) of participants; BMQ results for participants with a BMQ at both follow-ups indicated that adherence status changed between six and 12 months for 24% (N=56) of participants.

Table 1 presents results of the full logistic regression analyzing predictors of adherence. At six months, MPRs indicated that individuals of a race other than white (odds ratio [OR]=.40, p=.01) or of Hispanic or Latino ethnicity (OR=.23, p=.03) were less likely than whites and non-Hispanics, respectively, to be adherent. Individuals with a complex medication regimen were more likely than participants without a complex medication regimen to be adherent (OR=2.64, p=.003). At 12 months, men (OR=.32, p=.02) and individuals of Hispanic or Latino ethnicity (OR=.17, p=.04) were less likely than women or non-Hispanics, respectively, to be adherent. Middle-aged (OR=4.79, p=.01) and older (OR=5.17, p=.03) adults and individuals with a complex medication regimen (OR=3.79, p=.001) were more likely than younger adults and participants without a complex medication regimen, respectively, to be adherent.

No significant predictors were found for self-reported adherence at six months. At 12 months, individuals who were unemployed were significantly less likely than participants who were employed (OR=.19, p=.01) to report being adherent. Participants with moderate (OR=.06, p=.01) or severe depression (OR=.05, p=.01) were less likely than participants with minimal depression to report being adherent. Individuals with a complex medication regimen were more likely than individuals without a complex medication regimen to report adherence (OR=4.10, p=.001).

In line with reports for both VA and non-VA samples (1–4,10), 35% and 36% of patients in ongoing depression treatment had MPRs indicating poor adherence at six and 12 months, respectively, whereas 24% and 18% reported poor adherence on the BMQ at six and 12 months, respectively. Approximately, 13%−32% of individuals who were considered adherent on one measure were classified as nonadherent on the other measure. This discordance was likely the result of limitations in both measures and the different time frames assessed by these measures, but both measures may convey important information. When adherence is measured with self-report, individuals may exaggerate their adherence because of social desirability or because they forget having missed doses. However, a BMQ indicating poor adherence may be specific for short-term issues, given that patients are unlikely to report missing medication if they are taking it regularly. Thus patient report of poor adherence on the BMQ could lead to productive discussions between clinicians and patients and be a short-term marker of adherence for researchers.

The MPR measure may be biased upward (if patients fill prescriptions but do not ingest the medication) or downward (if patients’ medication is discontinued in conjunction with a clinician). However, in general, the MPR is the upper limit of the amount that the patient might have taken, given that it represents medications “on hand” over a longer period of time. This measure could lead to clinician-patient discussions regarding longer-term patterns in medication use (for example, taking short breaks in medication use) and may be a more sensitive marker for researchers of longer-term adherence. This study suggests that self-reported adherence may be more affected by patients’ current clinical and social circumstances, whereas MPR-based adherence may be more affected by the complexity of medication regimens and demographic factors.

In support of prior research, the most consistent predictor of adherence was having a complex medication regimen (10). However, taken together, study findings indicated that certain demographic groups (such as younger patients and members of racial and ethnic minority groups), patients with higher levels of depressive symptoms, and the unemployed may be targeted for efforts to improve adherence. Of interest, we found no significant associations between adherence and social support or patient-reported self-efficacy for managing depression, suggesting that these constructs may not be important factors in interventions to improve adherence.

Limitations of the study included generalizability, given that the sample predominantly consisted of male veterans who were likely to be older and not employed. Our moderate sample size may have limited our ability to detect modest effects on adherence. Social desirability may influence self-reports of adherence. Although pharmacy benefits through the VA are generally more generous than Medicare and many veterans prefer these benefits, we may not have captured all medications paid for by other insurers (15).

Self-report assessments of adherence were higher than estimates of adherence obtained from pharmacy-based assessments, and different patterns of predictors were associated with the two measures. This study and prior research indicate that men, the unemployed, persons from racial groups other than white, and persons who are not Hispanic or Latino may be more likely to benefit from targeted interventions for improving adherence. Patients with higher levels of depressive symptoms or adverse environmental circumstances may also benefit from more frequent self-report assessments and prompt discussion of issues related to medication if poor adherence is reported.