Predictors of Antidepressant Nonadherence Among Older Veterans With Depression
Abstract
Objective:
Most depression among older adults is treated in primary care, and many patients do not adhere to medication treatment. The U.S. Department of Veterans Affairs (VA) has recently introduced initiatives to address such treatment gaps. This study examined patient-reported antidepressant nonadherence during the acute treatment period (first four months after a prescription) and identified predictors of nonadherence in a sample of older veterans.
Methods:
This was a prospective, observational study of 311 participants ages 60 and older who received care at three VA medical centers and who had a diagnosis of clinically significant depression and were given a new outpatient antidepressant prescription. Participants completed an initial interview and a follow-up interview at four months after treatment recommendation. Antidepressant adherence was measured with a well-validated self-report measure.
Results:
At four months, 29% of participants reported nonadherence to their antidepressant medication. In unadjusted analyses, nonadherence was significantly associated with being African American, having no spouse or significant other, having greater general medical comorbidity, and receiving the prescription in a primary care setting (versus a specialty mental health setting). In logistic regression models controlling for several variables (demographic, illness, and functional status variables; type of antidepressant; contact with a therapist; medical setting; and site of recruitment), two predictors remained significantly associated with nonadherence: African-American race (odds ratio [OR]=4.23, p<.001) and general medical comorbidity (OR=1.33, p=.002).
Conclusions:
The two main predictors of nonadherence among older adults with depression were African-American race and general medical comorbidity. Results suggest that significant needs remain for important patient subgroups to improve antidepressant adherence.
Antidepressant nonadherence among elderly persons is common (1). A recent study found that approximately 40% of elders newly prescribed an antidepressant in university primary care clinics were nonadherent at a four-month follow-up (2). Nonadherence was even higher among African-American elders in the study. In a previous study, we found that over a third of older veterans with depression did not receive any treatment, and among those who received medications, another third failed to fill their prescriptions consistently (3). Recent initiatives in the U.S. Department of Veterans Affairs (VA), including the Primary Care Mental Health Integration (PC-MHI) program, have been introduced to address such treatment gaps.
Although the PC-MHI program has resulted in improvements in “first-generation” problems of depression treatment, such as screening, case finding, and treatment delivery (4–6), the extent of “second-generation” problems, such as adherence-based barriers to successful depression treatment, merits examination. Much research has focused on the primary care provider’s role and on organizational factors in the inadequate treatment of patients with depression, but less research has assessed patient factors, such as adherence behaviors. Further understanding of nonadherence and its predictors among older veterans is critical to improving interventions for enhanced depression care.
This study evaluated antidepressant nonadherence and correlates among older veterans with depression whose physicians had recommended antidepressant treatment. Study goals were twofold: first, to examine antidepressant nonadherence during the acute treatment period (first four months) and, second, to identify predictors of antidepressant nonadherence. Self-reported adherence rates were also compared with pharmacy data–generated medication possession ratios (MPRs), which provided an objective measure of antidepressant coverage over the study period.
Methods
This study received approval from the institutional review boards of the VA Ann Arbor Healthcare System and Wayne State University.
Sample
Study participants were recruited from three VA medical centers in Michigan (Ann Arbor, Detroit, and Battle Creek) from 2008 to 2011. Study inclusion criteria required participants to be ≥60 years old, have a depression diagnosis, and have received a new antidepressant prescription from a VA outpatient clinic within the previous week. Antidepressant medication qualified as new if the participant had not received it during the past six months; antidepressants added to augment existing antidepressants qualified as new unless the clinician’s note and dosage range indicated intent to treat sleep or to assist with smoking cessation. As in our prior study (2), patients were screened and included if they had clinically significant depression (score of ≥5 on the Patient Health Questionnaire [PHQ-9]) (7) and did not have cognitive impairment (fewer than three errors on the Six-Item Screener [8]). Participants were given an initial semistructured telephone interview as soon as possible after the antidepressant recommendation; the mean±SD time to initial interview from treatment recommendation date was 48.0±12.0 days. The time lag reflected the period it took to identify potentially eligible patients from new antidepressant fills, screen charts for eligibility, mail a letter to the patient regarding a possible study phone call, and then recontact the patient by phone to request participation. The consent form was then mailed back by the participant prior to the initial interview.
Participants were reinterviewed four months after they received the recommendation for antidepressant treatment. Four-month follow-up was chosen as the primary outcome point because early-phase treatment is a critical period associated with increased risk of poor outcomes, including vulnerability to suicide (9–11). Interviews were conducted by trained research assistants. All participants provided written informed consent before study participation.
Research Assessments
The initial interview included collection of patient demographic data, including age, gender, race, education level, and marital status. Illness burden of depression was assessed with the PHQ-9 (7), and anxiety was assessed with the Anxiety Sensitivity Index–Revised (12) subscales 1 and 4, which measure beliefs and fears about somatic sensations, and the anxiety subscale of the Hospital Anxiety and Depression Scale (13). Medical illness burden was assessed with the modified Charlson Comorbidity Index (14) on the basis of ICD-9 codes from problem summary lists in participants’ medical records during the year before the initial interview; these data were acquired from the Veterans Health Administration administrative databases. An estimate of functional status was based on the Instrumental Activities of Daily Living Scale (15), perceived global functioning was measured by the 12-item Short-Form Health Survey (16), and cognitive executive functioning was measured on the basis of the letter-number sequencing subscale of the Wechsler Memory Scale (17).
Data on geographic site of care (Ann Arbor, Battle Creek, or Detroit) and medical setting (primary care versus specialty mental health care, based on the provider who made the new antidepressant treatment recommendation) were also collected from the electronic medical record. Information regarding whether participants were engaged in therapy was obtained from a Brief Medication Questionnaire (BMQ) (18) item that asks “Are you seeing a therapist?” Participant self-report of psychotherapy, counseling, or group therapy was counted in the affirmative.
At four-month follow-up, the antidepressant prescribed dosage was verified by using the electronic medical record, and depression medication adherence was assessed by using the validated self-report adherence question from the BMQ. This item asks participants about how consistently they took their medication on a daily basis in the week prior to the interview and has been found to be significantly correlated with pharmacy refill records. As in prior studies (19–22) that considered adherence of less than 80% to be inadequate, patients who missed two or more daily doses in a given week were rated as nonadherent. Participants who never initiated the antidepressant medication (either by not filling or never taking the prescription) were also recorded as nonadherent.
Because pharmacy data are collected for veterans obtaining prescriptions through VA pharmacies, these data were used to calculate MPRs. The MPR is the number of days of medication supplied divided by the number of days in the time interval. The MPR provided an objective measure of antidepressant coverage over the study period, augmenting the BMQ self-report. An MPR threshold of less than 80% was considered inadequate coverage (23).
Statistical Analysis
Descriptive statistics were calculated for the total sample. Logistic regression analysis was used to evaluate associations between nonadherence during the first four months and the following initial variables: demographic, illness burden, functional status, site of care, and medical setting. Potential explanatory variables were entered sequentially in a predefined order of variable sets (blocks) in order to examine the influence on medication nonadherence of each set of variables in a sequential way. The block order of entry was determined so that the potentially confounding contextual variables were entered first to control for their effects: demographic characteristics were entered first, followed by illness burden variables, functional status variables, and finally, site of recruitment and medical setting. At each step, relevant variables were kept in the model when the next block of variables was considered. As a measure of the strength of the association, odds ratios (ORs) were obtained based on each model. Multicollinearity across the potential explanatory variables within each block of variables was checked. In the study sample, 4% (N=11) had missing data for at least one risk factor. Missing values were handled with a multiple-imputation strategy in SAS; imputation was done by using a multivariate normal model that included all covariates as well as adherence. The logistic regression estimates were pooled across five multiply imputed data sets by using Rubin’s method (24).
To assess for concern that the time lag between treatment recommendation and initial interview might have affected the results, we also examined the relationship between initial PHQ-9 score and length of time between treatment recommendation and initial interview.
Results
A total of 834 patients were potentially eligible for the study on the basis of age and a new antidepressant prescription in a participating clinic. We were able to contact 448 (54%) of these potential participants. A total of 379 patients completed screening, and of these, 318 were found to be eligible for study participation. Of eligible patients, 98% (N=311) agreed to study participation, provided informed consent, and completed initial assessments. No participants were lost to follow-up in the four-month study period. [A CONSORT diagram illustrating sample recruitment is included in an online supplement to this article.]
Table 1 summarizes data on demographic and health characteristics of the sample (N=311). Most participants were male (97%). The sample ranged in age from 60 to 86, with a mean age of 64.9. Seventy-nine percent of participants identified as white, 17% as African American, and 3% as other. Forty-two percent reported that they were not married or partnered, and 59% reported at least some college education. Sixty-one percent reported receiving prior treatment for depression, and 38% were currently seeing a therapist. Prevalence of comorbid psychiatric diagnoses was as follows: posttraumatic stress disorder, 38%; other anxiety disorder, 20%; and any substance use disorder, 17% (alcohol use disorder, 14%; and drug use disorder, 8%). Table 1 also presents mean scores on the clinical assessments conducted at the initial interview. Initial PHQ-9 scores ranged from 5 to 26. The number of patients whose new antidepressant augmented an existing antidepressant (instead of as a primary treatment) was small (N=22 participants, or 7%).
Characteristic | Total (N=311) | Adherent (N=220) | Nonadherent (N=91) | |||
---|---|---|---|---|---|---|
N | % | N | % | N | % | |
Site of recruitment | ||||||
Ann Arbor | 133 | 43 | 95 | 43 | 38 | 42 |
Battle Creek | 96 | 31 | 70 | 32 | 26 | 29 |
Detroit | 82 | 26 | 55 | 25 | 27 | 30 |
Antidepressant prescribed in primary care settinga | 143 | 46 | 93 | 42 | 50 | 55 |
Male | 303 | 97 | 215 | 98 | 88 | 97 |
Raceb | ||||||
African American | 54 | 17 | 27 | 12 | 27 | 30 |
White | 247 | 79 | 183 | 83 | 64 | 70 |
Otherc | 10 | 3 | 10 | 5 | 0 | — |
Aged | ||||||
60–64 | 216 | 69 | 158 | 72 | 58 | 64 |
65–74 | 60 | 19 | 37 | 17 | 23 | 25 |
75–86 | 35 | 11 | 25 | 11 | 10 | 11 |
High school education or below | 128 | 41 | 84 | 38 | 44 | 48 |
Without a partner or spouseb | 131 | 42 | 80 | 36 | 51 | 56 |
Antidepressant use | ||||||
Citalopram or sertraline | 190 | 61 | 127 | 58 | 63 | 69 |
Other antidepressant | 121 | 39 | 93 | 42 | 28 | 31 |
Prior depression treatment | 191 | 61 | 137 | 62 | 54 | 59 |
Seeing a therapist | 117 | 38 | 89 | 40 | 28 | 31 |
ICD-9 psychiatric diagnosis | ||||||
Delirium | 12 | 4 | 10 | 5 | 2 | 2 |
Schizophrenia spectrum disorders | 1 | <1 | 0 | — | 1 | 1 |
Bipolar I disorder | 2 | 1 | 1 | <1 | 1 | 1 |
Bipolar II disorder | 1 | <1 | 1 | <1 | 0 | — |
Parkinson's disease | 3 | 1 | 2 | 1 | 1 | 1 |
Any substance use disorder | 53 | 17 | 35 | 16 | 18 | 20 |
Alcohol use disorder | 42 | 14 | 27 | 12 | 15 | 16 |
Drug use disorder | 24 | 8 | 15 | 7 | 9 | 10 |
Posttraumatic stress disorder | 117 | 38 | 90 | 41 | 27 | 30 |
Other anxiety disorder | 61 | 20 | 44 | 20 | 17 | 19 |
Personality disorder | 3 | 1 | 3 | 1 | 0 | — |
Charlson Comorbidity Indexb,e | ||||||
0 | 116 | 37 | 86 | 39 | 30 | 33 |
1 | 84 | 27 | 69 | 31 | 15 | 16 |
>1 | 111 | 36 | 65 | 30 | 46 | 51 |
SF-12 physical component summary score (M±SD)f | 36.8±11.4 | 36.9±11.4 | 36.7±11.5 | |||
SF-12 mental component summary score (M±SD)f | 37.8±11.4 | 37.8±11.5 | 37.7±11.2 | |||
PHQ-9g | 12.5±6.2 | 12.5±6.4 | 12.6±5.8 | |||
HADS anxiety subscale score (M±SD)h | 8.8±5.1 | 9.1±5.0 | 8.2±5.2 | |||
ASI-R subscales score (M±SD)i | 53.1±26.3 | 53.1±26.0 | 53.1±27.1 | |||
IADL Scale total score (M±SD)j | 7.5±1.3 | 7.4±1.3 | 7.5±1.4 | |||
Wechsler LNS subscalek | 7.9±3.6 | 7.9±3.5 | 7.9±3.7 |
Characteristics at baseline of total sample and by four-month adherence status
At the four-month follow-up, overall self-reported nonadherence to antidepressant treatment was 29% (N=91). There was no association between initial PHQ-9 score and time between treatment recommendation and initial interview. Initial PHQ-9 scores did not differ between nonadherent and adherent participants in adjusted analyses.
Nonadherent individuals were significantly more likely than adherent participants to be African American; African Americans accounted for 30% of the nonadherent group. Nonadherent participants were also significantly more likely than adherent participants to have no spouse or significant other (56% versus 36%; χ2=10.23, df=1, p=.001) and to have significantly greater general medical comorbidity, as indicated by a Charlson score of >1 (51% versus 30%; χ2=12.37, df=1, p<.001). The proportion of participants currently seeing a therapist was smaller in the nonadherent group than in the adherent group (31% and 40%), but the difference was not statistically significant. Finally, nonadherent participants were significantly more likely than adherent participants to have been prescribed the antidepressant in a primary care setting rather than a mental health setting (55% versus 42%; χ2=4.16, df=1, p=.04).
Adherent patients were more likely than nonadherent patients to show improvement as measured by the PHQ-9 at four-month follow-up (55%, N=122, versus 45%, N=41), but the difference was not statistically significant. However, adherence was significantly associated with depression symptom improvement among African Americans: 63% (N=17) of adherent African-American participants showed improvement at follow-up as measured by the PHQ-9, compared with only 33% (N=9) of nonadherent African-American participants (χ2=8.31, df=2, p=.016).
Logistic regression models were run sequentially to assess for predictors of nonadherence (Table 2). Among demographic characteristics, race and marital status were each significantly associated with nonadherence. When illness burden variables were added to the model, the Charlson Comorbidity Index was also significantly associated with nonadherence. Race and comorbidity remained significant when functional variables were added and when site of care and medical setting were entered. After adjusting for demographic variables and comorbidity, we did not find medical setting to be associated with nonadherence. Relationship status was not significant when illness burden or functional variables were added to the model but became significant when site of care and medical setting were entered. [The online supplement has a more extensive version of this table, which includes all the variables assessed in each block, not just the ones found statistically significant, with their findings.]
Variable | Model 1a | Model 2b | Model 3c | Model 4d | Model 5e | Model 6f | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
AOR | 95% CI | p | AOR | 95% CI | p | AOR | 95% CI | p | AOR | 95% CI | p | AOR | 95% CI | p | AOR | 95% CI | p | |
African American (reference: white) | 3.20 | 1.71–6.01 | <.001 | 3.54 | 1.76–7.11 | <.001 | 3.84 | 1.88–7.84 | <.001 | 4.25 | 2.03–8.91 | <.001 | 4.24 | 2.02–8.87 | <.001 | 4.23 | 2.01–8.88 | <.001 |
Without partner or spouse (reference: with a partner or spouse | 2.02 | 1.20–3.38 | .008 | 1.67 | .96–2.92 | .070 | 1.73 | .99–3.03 | .056 | 1.79 | 1.01–3.17 | .045 | 1.82 | 1.03–3.22 | .041 | 1.82 | 1.03–3.23 | .041 |
Charlson Comorbidity Index | 1.31 | 1.11–1.54 | .002 | 1.33 | 1.12–1.58 | .001 | 1.33 | 1.11–1.59 | .002 | 1.33 | 1.11–1.58 | .002 | 1.33 | 1.11–1.58 | .002 | |||
Antidepressant prescribed in primary care (reference: prescribed in a mental health setting) | 1.40 | .76–2.55 | .277 | 1.33 | .72–2.44 | .363 | 1.32 | .71–2.48 | .384 | |||||||||
Citalopram or sertraline (reference: other antidepressant) | 1.50 | .81–2.77 | .198 | 1.50 | .81–2.78 | .203 | ||||||||||||
Seeing a therapist (reference: not seeing a therapist) | .98 | .51–1.88 | .952 |
Adjusted odds ratio (AOR) estimates from logistic regression models of predictors of antidepressant nonadherence at four-month follow-up among older veterans
In the final model (model 6 in Table 2), African-American participants were significantly more likely than participants of white or other races to be nonadherent to antidepressant medication (OR=4.23, p<.001). Patients with higher medical burden, as measured by the Charlson Comorbidity Index, were significantly more likely than those with less medical burden to be nonadherent (OR=1.33, p=.002); a 1-point increase in Charlson score resulted in a 33% increase in nonadherence.
Most participants were taking selective serotonin reuptake inhibitors (N=208, 67%), with the remainder taking bupropion (N=47, 15%), mirtazapine (N=30, 10%), serotonin-norepinephrine reuptake inhibitors (N=12, 4%), trazodone (N=9, 3%), or tricyclic antidepressants (N=10, 1%). (The numbers sum to slightly more than 311 because of a small number of patients taking more than one antidepressant.) In model 5 (Table 2), we examined the role of antidepressant type: citalopram or sertraline versus all other antidepressants (citalopram and sertraline were chosen as comparators because of frequent first-line use for tolerability), but we did not find a significant impact of antidepressant type on adherence.
VA pharmacy data were available for most participants (298 of 311) to calculate four-month MPRs. Nonadherence as measured by the MPR was higher than as measured by BMQ self-report; 53% (N=158) of the 298 participants were nonadherent as measured by the MPR, compared with 29% (N=91) of the 311 participants who reported nonadherence on the BMQ. As with the BMQ, MPR results indicated poorer adherence by race: as measured by the MPR, 77% (N=41) of African Americans were nonadherent, compared with 49% (N=114) of whites (χ2=16.65, df=2, p<.001). African Americans were 5.48 (95% confidence interval [CI]=2.44–12.31, p<.001) times more likely to be nonadherent on the basis of the MPR in the final adjusted model. Other than race, the only other potential predictor of nonadherence significant in the MPR model was initial prescription in primary care; those participants were 1.78 (CI=1.01–3.15, p<.05) times more likely to be nonadherent than participants whose prescriptions were received in mental health settings.
Discussion
Four-month medication nonadherence among older veterans with depression who received a new prescription for antidepressants in outpatient settings was examined. Overall self-reported nonadherence was 29%. Two important predictors of nonadherence (race and general medical comorbidity) were identified, indicating that significant barriers remain in important patient subgroups. These predictors remained significant even after the analysis controlled for an array of initial differences in demographic, illness, and function variables. Illness burden and functional status were not predictive of nonadherence.
The overall self-reported nonadherence rate in this study is somewhat lower than the rate reported in a non-VA sample (2). Sample differences may have been a contributing factor (African-Americans were oversampled in the university study sample). However, it should also be noted that in the study reported here, participant data were linked to VA pharmacy administrative data to calculate an additional measure of adherence. As in prior studies, MPR values indicated that the level of nonadherence was significantly greater than participants disclosed on self-report (23,25). The discordance between MPR and self-report is likely a result of limitations in both measures and of different time frames between the measures (MPR measured a four-month time interval and BMQ a one-week interval).
As in the prior study (2), race was found to be a significant predictor of nonadherence. In the study reported here, nonadherence among African-American participants was significantly higher than among whites. Thirty percent of nonadherent participants were African American. Of note, African-American participants who did not adhere to treatment were significantly less likely to experience depression symptom reduction. This confirmatory finding of the contribution of race to nonadherence suggests the need for additional efforts to reach out to patients from minority populations who may find antidepressants less acceptable compared with white patients (26) and may have greater concerns about side effects and addiction. In previous work, we have shown that after analyses controlled for differences in beliefs (specifically worries about side effects and beliefs about the importance of antidepressants), racial differences were no longer significant (27).
The other predictor significantly correlated with nonadherence was general medical comorbidity. Nonadherence among patients with greater levels of comorbidity was significantly higher than among those with less comorbidity. Later-life depression frequently coexists with general medical illness that may exacerbate depression and complicate treatment. Medical comorbidity is related to polypharmacy, setting the stage for medication side effects and drug interactions. A prior study using claims data found that the presence of a side effect (for example, headache or dizziness) was associated with a 3.7-percentage point increase in discontinuation rates of antidepressants among older adults (28). In that study, side effects were common during the first four weeks of treatment, and only 45% of participants who had a documented side effect one month after treatment initiation filled a second prescription for the same antidepressant.
Although medical setting and relationship status were associated with adherence in unadjusted analyses, these variables were no longer significant after adjustment for demographic variables and general medical comorbidity. However, in the secondary MPR analyses, participants who received their initial prescription in primary care were significantly more likely to be nonadherent than those who received their prescription in a mental health setting. These results suggest that the PC-MHI program may be successful in reducing treatment barriers for a number of veterans in primary care; however, adherence is still an issue for a subgroup of vulnerable patients (as identified by the predictors of race and comorbidity burden). Such patients may require interventions more tailored to their specific needs and barriers. Although race is not a modifiable factor, the beliefs associated with nonadherence are amenable to intervention strategies, such as the Treatment Initiation Program, which blend elements of psychoeducation, motivational interviewing, and problem solving and have been successfully used with older adults (22). Similarly, PC-MHI providers could assist individuals with high levels of medical comorbidity linked to polypharmacy with strategies to enhance adherence, such as using pillboxes or electronic medication dispensers. Patients may also need encouragement to discuss concerns about side effects with their providers.
Several factors may limit generalizability. Because the sample consisted of older veterans, most were male. The sample was drawn from outpatient clinics affiliated with three VA medical centers in Michigan; results may have differed if the study were conducted in another geographical area. We were able to contact only 54% of potentially eligible participants, which may have affected our results. However, a substantial portion (40%) of the potential participants whom we were unable to contact was from the inner-city site, and 44% of those were African American; thus we suspect that nonadherence rates might have been even higher in this unreachable group. The absence of other racial-ethnic groups, notably Latinos, also potentially limits generalizability. Finally, largely because of required human subject protections procedures, there was a time lag of several weeks for most participants between treatment recommendation and initial interview, which may have affected our results; however, we did not find any association between this time lag and PHQ-9 scores.
Conclusions
This study is the first to our knowledge to examine nonadherence behaviors among older veterans with depression. Given the numbers of older patients treated with antidepressants yearly at substantial cost, it is important to further understand the extent to which these veterans adhere to treatment. Two main predictors associated with nonadherence during the acute treatment phase of later-life depression were identified. In programs like the PC-MHI, providers should consider the impact of race and general medical comorbidity and tailor interventions accordingly.
1 : Adherence to depression treatment in older adults: a narrative review. Drugs and Aging 25:559–571, 2008Crossref, Medline, Google Scholar
2 : Racial differences in adherence to antidepressant treatment in later life. American Journal of Geriatric Psychiatry 21:999–1009, 2013Crossref, Medline, Google Scholar
3 : Treatment for new episodes of depression. Psychiatric Services 58:1035, 2007Link, Google Scholar
4 : VA primary care-mental health integration: patient characteristics and receipt of mental health services, 2008–2010. Psychiatric Services 63:1137–1141, 2012Link, Google Scholar
5 : Primary care–mental health integration programs in the Veterans Affairs health system serve a different patient population than specialty mental health clinics. Primary Care Companion for CNS Disorders 14:14, 2012Google Scholar
6 : Continuation of care following an initial primary care visit with a mental health diagnosis: differences by receipt of VHA Primary Care–Mental Health Integration services. General Hospital Psychiatry 35:66–70, 2013Crossref, Medline, Google Scholar
7 : The PHQ-9: a new depression diagnostic and severity measure. Psychiatric Annals 32:509–515, 2002Crossref, Google Scholar
8 : Six-item screener to identify cognitive impairment among potential subjects for clinical research. Medical Care 40:771–781, 2002Crossref, Medline, Google Scholar
9 : Stigma as a barrier to recovery: perceived stigma and patient-rated severity of illness as predictors of antidepressant drug adherence. Psychiatric Services 52:1615-1620, 2001Link, Google Scholar
10 : Perceived stigma as a predictor of treatment discontinuation in young and older outpatients with depression. American Journal of Psychiatry 158:479–481, 2001Link, Google Scholar
11 : Higher-risk periods for suicide among VA patients receiving depression treatment: prioritizing suicide prevention efforts. Journal of Affective Disorders 112:50–58, 2009Crossref, Medline, Google Scholar
12 : The Anxiety Sensitivity Index–Revised: psychometric properties and factor structure in two nonclinical samples. Behaviour Research and Therapy 41:1427–1449, 2003Crossref, Medline, Google Scholar
13 : The Hospital Anxiety and Depression Scale. Acta Psychiatrica Scandinavica 67:361–370, 1983Crossref, Medline, Google Scholar
14 : A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. Journal of Chronic Diseases 40:373–383, 1987Crossref, Medline, Google Scholar
15 : A research and service oriented multilevel assessment instrument. Journal of Gerontology 37:91–99, 1982Crossref, Medline, Google Scholar
16 : A 12-item Short-Form Health Survey: construction of scales and preliminary tests of reliability and validity. Medical Care 34:220–233, 1996Crossref, Medline, Google Scholar
17 : WMS-III: Wechsler Memory Scale Administration and Scoring Manual. New York, Psychological Corp, 1997Google Scholar
18 : The Brief Medication Questionnaire: a tool for screening patient adherence and barriers to adherence. Patient Education and Counseling 37:113–124, 1999Crossref, Medline, Google Scholar
19 : Pharmacy data identify poorly adherent patients with schizophrenia at increased risk for admission. Medical Care 40:630–639, 2002Crossref, Medline, Google Scholar
20 : Nontricyclic antidepressants: predictors of nonadherence. Journal of Clinical Psychopharmacology 26:643–647, 2006Medline, Google Scholar
21 : Adherence to antidepressant therapy and mortality rates in ischaemic heart disease: cohort study. British Journal of Psychiatry 206:297–301, 2015Crossref, Medline, Google Scholar
22 : Improving antidepressant adherence and depression outcomes in primary care: the treatment initiation and participation (TIP) program. American Journal of Geriatric Psychiatry 18:554–562, 2010Crossref, Medline, Google Scholar
23 : Reasons for antidepressant nonadherence among veterans treated in primary care clinics. Journal of Clinical Psychiatry 72:827–834, 2011Crossref, Medline, Google Scholar
24 : Multiple Imputation for Nonresponse in Surveys. New York, Wiley, 1987Crossref, Google Scholar
25 : Variables influencing antidepressant medication adherence for treating outpatients with depressive disorders. Journal of Affective Disorders 123:216–221, 2010Crossref, Medline, Google Scholar
26 : Patient-centered communication, ratings of care, and concordance of patient and physician race. Annals of Internal Medicine 139:907–915, 2003Crossref, Medline, Google Scholar
27 : The association between race and gender, treatment attitudes, and antidepressant treatment adherence. International Journal of Geriatric Psychiatry 29:169–177, 2014Crossref, Medline, Google Scholar
28 : Antidepressant use in geriatric populations: the burden of side effects and interactions and their impact on adherence and costs. American Journal of Geriatric Psychiatry 19:211–221, 2011Crossref, Medline, Google Scholar