Jeffrey B. Weilburg, MD

Use of generic drugs is an important tool in controlling healthcare costs. Switching patients from branded agents to generic therapeutic alternatives reduces pharmacy costs and has become a common cost control method. Lipid-lowering agents have been targeted for therapeutic substitution due both to their high cost to patients and payers

 and to the similarity of action among drugs within the statin class. However, questions have been raised about whether a switch from brand Lipitor (atorvastatin) to generic simvastatin would provide equal level of control of low-density lipoprotein (LDL) cholesterol. Therapeutic substitution programs for these agents are only appropriate if outcomes are comparable and clinical objectives are met.

Lipid levels and cardiovascular outcomes are directly related to statin dosing.

 The potency (lipid-lowering capacity/mg) of statins varies, so comparisons between agents requires the use of equipotent doses. Dose-response trials support the inference of equivalence between low-dose (10 mg to 20 mg) atorvastatin and mid-dose (20 mg to 40 mg) simvastatin to reduce LDL cholesterol levels.

Previous studies of statin switches have demonstrated LDL outcomes consistent with this dose-response relationship. Statin interchange programs that decreased the average equipotent dose found unfavorable patient outcomes.

 From an analysis of prescription claims, Hess and colleagues observed that 38% of atorvastatin-to-simvastatin changes were to lower-potency doses, suggesting that greater caution is necessary for these interchanges.

 Several switch programs resulting in higher average equipotent doses documented superior lipid control, but applicability of these studies is limited because they involved switching to or from cerivastatin (Baycol), which was withdrawn from the market in August 2001.

The studies that have investigated equipotent switches or comparisons have been insufficient to resolve outstanding questions. An observational trial described differences in outcomes between equipotent doses of atorvastatin and simvastatin but did not measure or control for lipid levels.

 A small audit of atorvastatin-to-simvastatin switches (n = 70) in the United Kingdom demonstrated substantial cost savings and did not demonstrate a significant impact on lipid control.

Our research was conducted to evaluate the equivalence of lipid control (as measured by LDL serum levels) through retrospective observation of a commercially insured population that chose, for financial reasons, to interchange medications through a therapeutic substitution program. This study observed and evaluated the impact of a therapeutic substitution program from low-dose atorvastatin to simvastatin on lipid control. We hypothesized that patients switching to simvastatin would exhibit lipid control comparable to that achieved by equipotent atorvastatin therapy.

This study was reviewed and approved by Massachusetts General Hospital’s Institutional Review Board under protocol number 2008P001583. Subject records were deidentified and the study qualified for and was granted a waiver to informed consent.

The Massachusetts General Physicians Organization (MGPO) is a multi-specialty medical group consisting of more than 2000 physicians practicing at a large tertiary urban academic medical center.

The MGPO implemented a therapeutic substitution program in 2007 as part of its efforts to address rising healthcare costs. This program, known as “Switch To,” assisted physicians in offering their patients access to lower-cost medications. The program was conducted as a standard-of-care medical management intervention, so design choices prioritized operational concerns over considerations related to subsequent research questions. The MGPO—in consultation with clinical experts&mdash;selected drugs with therapeutic alternatives that could be used to achieve comparable clinical outcomes at lower cost to patients and the health system. An interchange from atorvastatin 10 mg to simvastatin 20 mg was among the first suitable interchanges identified for the Switch To program, and is the focus of the current analysis. The program aimed to address patients prescribed atorvastatin before the introduction of generic mid-potency statins (pravastatin and simvastatin) to the US market in 2006.

Patients taking brand atorvastatin were identified from pharmacy claims from 3 large commercial payers made available to the program through existing pay-for-performance agreements. Contractual limitations prevented investigators from using claims to assess related issues such as drug costs or adherence. Patients who had claims for simvastatin before their current

atorvastatin therapy were excluded out of concern that it might signify a prior treatment failure. Pharmacy claims and health records were used to identify the predominant prescriber for each patient’s atorvastatin.

The predominant prescriber, most frequently the patient’s primary care physician, was invited by the program manager to review their patients in a secure Web tool and indicate which patients should be offered a switch to simvastatin. If the prescriber approved the change, 3 attempts were made to contact patients by phone, inviting them to switch. Letters were sent to patients who were not reached by phone, with instructions on how to opt out of switching. If the patient agreed to the switch by phone or did not opt out of the switch after a mailing, the program worked with the prescriber, the electronic medical record (EMR), and the patient’s pharmacy to effect the switch. After a switch, the program sent confirmation notices to patients and prescribers to support follow-up labs, and performed population-level surveillance for quality assurance purposes.

The study is a retrospective review of patient participation in the Switch To program and their subsequent lipid control. Patients identified by the Switch To program for having pharmacy claims for atorvastatin 10 mg and having at least 1 recorded LDL result before and after the intervention period met the inclusion criteria. Patients were excluded from the analysis if their physician requested an alternate simvastatin dosing regimen. Also excluded were patients whom the program recorded as switching to simvastatin, but whose medical records indicated that they did not switch, or switched back to atorvastatin.

Demographics and lipid lab results were obtained from the EMR, and the EMR information was combined with pharmacy insurance claims information to evaluate previous and current medication use. The Switch To program recorded outreach efforts and physician and patient responses, though neither physicians nor patients were asked to explain the reasoning underlying their decisions.

The change between average pre-intervention and average post intervention LDL levels was the primary outcome of interest. Average LDL levels were calculated from the subjects’ medical records, during periods when the patients were taking atorvastatin or simvastatin. We were also interested in the participation rates at each stage of the program to quantify the effectiveness of the Switch To program.

The pre- and post intervention periods were defined relative to the date that the patient switched medications, or by the last contact with the program for patients who did not switch. Pre-intervention LDL results were included in the patient’s average if they were: 1) within 2 years of this date, and 2) at least 6 weeks after the first recorded prescription in the EMR for atorvastatin. The long potential evaluation period was used because the program reached patients at various times in relation to their regular office visits—there was some reasonable likelihood that very recent lab values were not available. Patients without a recorded atorvastatin start date in the EMR, but with atorvastatin claims, were evaluated based on lab results occurring within 180 days before the switch or last contact date. The post intervention LDL average was based on results obtained at least 6 weeks after a switch, or immediately after the last contact date until the last available data. LDL values were included without regard to fasting since the patient’s fasting status was not reliably recorded, and studies demonstrate that fasting and nonfasting LDL levels are similar.

 We described the Switch To program performance characteristics using descriptive statistics.

All statistical analyses were performed using SAS statistical software, version 9.3 (SAS Institute, Cary, North Carolina). The hypothesis and normally distributed predictors were evaluated using a 

test for significance, with alpha 0.05. Median income by zip code was non-normally distributed and was therefore evaluated for differences using a Wilcoxon rank-sum test. Switchers and nonswitchers did not differ with respect to observed pretreatment factors, nor did those factors predict participation; therefore, propensity scoring was not appropriate.

We identified 1687 patients from commercial pharmacy claims who filled at least 1 prescription for 10 mg atorvastatin. Of the 1687 patients, 1312 patients had lipid lab results available before and after the program intervention. Four patients who were switched by the program to nonequivalent doses were removed from this analysis, and 28 patients for whom EMR medication records disagreed with program records were also removed, leaving 1280 evaluable patients.

Of the 1280 cases meeting inclusion criteria, physicians responded regarding 1261 patients, encompassing 98.5% of the identified patients (

). Physicians chose to enroll 68% of the patients they reviewed. Among enrolled patients, the program reached 533 patients by phone, with 115 (21.6%) of those electing to switch. Letters were sent to 304 patients not reached by phone; these letters informed patients that they would be switched and provided instructions on how to avoid switching by opting out. Nonresponse to the opt-out letters led to 101 (33.2%) patients being switched.

The patient group that switched and the group that did not switch had similar demographics by age, race, and median income by zip code (

). Patients who switched were disproportionately male (

= .003). Patients who switched had an average of 2.04 (SD = 1.02) LDL values before the switch, 0.16 (SD = 0.37) LDL values excluded during the switch, and 2.12 (SD = 1.07) LDL values after the switch. Nonswitchers averaged LDL results of 2.33 (SD = 1.27) before contact and 2.55 (SD = 1.36) after. On average, pre-treatment labs among switchers occurred 363.4 (SD = 207.0) days before the intervention date, and 332.4 (SD = 205.4) days after the switch. Switchers’ labs occurred an average of 349.1 (SD = 210.1) days before intervention date, and 382.3 (SD = 225.5) days after.

Over the evaluation period, LDL control declined, illustrated by rising LDL levels in both groups (

). The small increase in LDL in both groups was not statistically significantly different for switchers and nonswitchers (

The purpose of this study was to investigate the impact on lipid control of a therapeutic substitution from atorvastatin 10 mg to equipotent simvastatin 20 mg. While the average LDL levels of both switchers and nonswitchers increased, there was no statistically significant difference between patients who switched to simvastatin 20 mg and those who remained on atorvastatin 10 mg. We observed a 16.9% switch rate using our therapeutic substitution program.

The finding that use of equipotent doses of generic and brand statin did not produce statistically significant differences in lipid control suggests that statin therapeutic substitution programs are a clinically appropriate option to control pharmacy costs. The difference in mean LDL measurements between switchers and nonswitchers of 2.36 mg/dL was judged to be clinically insignificant.

The decline in the LDL control of both groups was concerning. A partial explanation for the decline in LDL control may relate to the long evaluation period. The unfavorable trend may indicate that patients require escalating doses to maintain control, or that patients are not adequately continuing with the prescribed doses. We did not have access to pharmacy claims data post intervention, so we could not evaluate the impacts of adherence and persistence on LDL control. Thus, even if patients were fully adherent, it is possible that their doctors did not adjust the medications in response to persistently elevated LDL levels. This study was not designed to address this issue, so no data were generated that would explain the upward drift in both groups.

Our therapeutic substitution program resulted in a 16.9% effective switch rate. In the context of an optional program for prescribers and patients, we consider these results to be moderately effective. The program achieved a high physician response rate, which demonstrates a potential benefit to conducting pharmacy management within the prescribers’ organization. Working within the prescribers’ institution enabled the program to integrate multiple insurers’ patients into a single effort and streamline messaging and response with frontline providers. Operating within the prescribers’ organization also helped ensure that the clinical experts involved in creating the program had credibility with the participating providers.

The program experienced considerable difficulties reaching patients by phone, despite up to 3 attempts for each patient. There was a substantially different patient participation rate depending on whether they were reached by phone or by letter. Calls and mailings also used different default choices: phone calls invited patients to opt in to switching, while letters required patients to opt out if they wanted to avoid the interchange. Unfortunately, we could not distinguish between the impact of the communication medium and the impact of differing default choices.

The differing participation rates between patients who had to act to opt in by phone and those who were automatically enrolled by letter and had to act to opt out illustrates choice architecture, which theorizes that the way choices are presented to decision makers influences the choices they select.

 There were 2 aspects of choice architecture at work: first, the media used were letter versus phone; second, the default choices were opt in versus opt out. Since both properties were changed together, we could not distinguish between the 2 effects, but this aspect of choice architecture merits further investigation. Amid increasing pressure to control costs while attempting to preserve patient autonomy, healthcare management interventions may increasingly base their intervention designs on principles of choice architecture. Measuring these influences will be necessary to ensure that healthcare providers use choice architecture in an ethical manner, thus empowering patients with meaningful choices.

Since the Switch To program, the American College of Cardiology (ACC) and American Heart Association (AHA) released updated guidelines.

 Within the ACC / AHA guidelines, atorvastatin 10 mg and simvastatin 20 mg are classified as moderate-intensity therapy. The updated guidelines preferentially recommend considering high-intensity statin treatment, but also describe a role for moderate-intensity therapy. To contrast the guidelines with this study, the guidelines recommend when to use moderate-intensity statins, while this study examines agent selection after a decision to treat with moderate intensity statins is made.

Our analysis was limited by data availability and the observational nature of the inquiry. We were unable to obtain additional pharmacy claims data that would have allowed evaluation of adherence and persistence metrics. We only included analysis switches that were facilitated by the therapeutic substitution program, but some patients switched therapies with their physician outside the formal program. As patients were not randomly assigned to treatment groups, our analysis was limited to observational methods. While switchers and nonswitchers were similar with respect to observed factors, we cannot rule out the possibility of unobserved confounders. This study found no evidence to support the proposition that observable factors meaningfully predict program participation, but such factors would be of interest if they could be identified.

A greater understanding of why physicians declined to enroll patients or why patients declined to switch is also of interest, but this information was not collected by the program. Some physicians informally reported that they didn’t enroll patients due to prior use of simvastatin that the available data failed to identify. A few prescribers shared concerns about having staff reaching out to their patients, or about the therapeutic equivalence of the switch. The program’s callers reported that the most common reason patients cited for not switching was interest in discussing the interchange with their doctor first.

The Switch To program continued, addressing other brand-to-generic therapeutic substitution opportunities. The program subsequently offered a switch from atorvastatin 20 mg to simvastatin 40 mg, but did not address higher doses. Further large efforts included a switch from brand proton pump inhibitors to generic omeprazole and from brand nasal steroids to generic fluticasone.

This analysis supports the assertion that there is an appropriate role for therapeutic substitution programs to contain costs while achieving equivalent patient management. Of interest in the current drive toward accountable care organizations, the findings also demonstrate that very high provider participation rates can be achieved by programs initiated by the provider’s own system and leadership. These results raise questions about decision inertia among providers and patients and the power of default choices, which warrants further study. Finally, the unfavorable trend in LDL control suggests further investigation into prescription adherence and persistence.

Articles

Evaluation of a therapeutic substitution from atorvastatin to simvastatin on low-density lipoprotein control, and assessment of program participation rates for a commercially insured population.

Therapeutic Substitution From Atorvastatin to Simvastatin: Impact on LDL Control

Depression and hyperlipidemia are common conditions that often coexist and may clinically interact with each other. Depression has a negative impact on medication adherence.

 In addition, depressed hyperlipidemic diabetes patients tend to refill their statin prescriptions less often than those without depression.

Many patients in primary care receive inadequate dose trials of antidepressant (AD) therapy.

 Prior studies have shown that patients receiving adequate AD treatment are twice as likely to adhere to chronic disease medications (including statins) than patients receiving inadequate AD treatment.

 The reason for the association between adequate AD treatment and adherence to chronic disease medications is not fully understood. Disease-related factors (resolution of depression) may play a role insofar as adequate AD treatment produces resolution of depression. However, patient-related factors (eg, baseline tendency toward adherence to all medications, medication-taking “style”) also may be involved. Patients likely to adhere to ADs also may be likely to adhere to other medications (statins). The relative impact of disease and patient-related factors in the association between AD treatment adequacy and adherence to other medications needs to be clarified.

We addressed the impact of disease- and patient-related factors among statin and AD users by focusing on statin adherence before and after AD initiation in patients with poor statin adherence at baseline. These patients may have poor statin adherence because of disease factors (the tendency of depressed patients to show poor medication adherence

) and/or patient factors (a general tendency to show poor adherence to any medication). Patient factors are likely to be stable, whereas disease factors may change if the disease that drives them—in this case, depression&mdash;resolves. Adequate AD treatment is more likely than inadequate AD treatment to produce improvement in depression.

 We hypothesize that if poorly adherent statin patients developed good statin adherence after an adequate trial of ADs significantly more often than after an inadequate trial of ADs, the change most likely would be due to the impact of the AD on the depression (eg, on medical factors).

We retrospectively analyzed a cohort of commercially insured patients receiving statin therapy. We used the Integrated Healthcare Information Services database system, a comprehensive, deidentified US healthcare claims database that is representative of the nonelderly, insurancecarrying population in the United States from more than 35 different healthcare plans serving members across 9 census regions. The database is compliant with the Health Insurance Portability and Accountability Act and features encrypted member and provider identification numbers. The average enrollment was 20 months; the annual attrition rate was roughly 15% to 25%. We identified all patients with (1) continuous plan enrollment; (2) more than 1 pharmacy claim for statin therapy (atorvastatin calcium, fluvastatin sodium, pravastatin sodium, lovastatin sodium, rosuvastatin calcium, or simvastatin calcium) in the first enrollment year; and (3) total cholesterol results measured in each of 2 consecutive enrollment years. 

From the 1.24 million patients receiving statin prescriptions during the time period studied, we were left with 49,943 patients after excluding patients who were not enrolled continuously, who did not have at least 2 statin prescriptions prior to either the onset of AD use or the midway point of their enrollment, or who did not have cholesterol levels available within the (mostly claims-based) database. Within our cohort of statin-treated patients, we identified all patients coprescribed AD therapy (AD cohort). For the purposes of our analysis, we limited our AD cohort to patients whose initial statin prescription date preceded their AD initiation date to allow the measurement of pre-AD treatment statin adherence. In addition, patients whose AD trial(s) consisted solely of low-dose tricyclic ADs or trazodone were excluded because of the likelihood that these prescriptions were used to treat chronic pain or insomnia. Approximately one-quarter of the patients in our statin cohort received AD therapy. After excluding patients whose first AD prescription preceded their initial statin prescription, our final cohort consisted of 42,227 eligible patients with complete data, including 4781 AD-treated patients (11% of the final dataset).

Pharmacy claims included the patient’s age, a unique patient identification number, the date the prescription was filled, and the drug name, dose, and number of pills supplied. Patient sex was determined from the health plan enrollment records. Using pharmacy claims for the overall study period, we defined 3 comorbid conditions related to cardiovascular disease based on prescription of condition-specific medications. Patients were categorized as having hypertension if they were prescribed any blood pressure—lowering medicines; diabetes if they were prescribed insulin or oral hypoglycemic agents (including sulfonylureas, metformin, thiazolidinediones, and alpha-glucosidase inhibitors); or ischemic heart disease if they were prescribed any nitrate-containing antianginal agents. This method of categorization does not include patients who may have been diagnosed with 1 of the above conditions but were not medically treated or, in the case of ischemic heart disease, treated with other agents.

Our primary outcome measures were statin adherence and statin refill persistence before and after each patient’s index date. The index date for AD-treated patients was the date of the first AD prescription following their initial statin prescription. For comparison purposes, the index date for non-AD patients was the midpoint of their enrollment period. To determine statin adherence, we calculated each patient’s statin medication possession ratio (MPR) in the years before and after the index date. The MPR was calculated by dividing the total number of pills received during each period (excluding the last prescription) by the days between the first and last fill dates during the period.

We also calculated statin refill persistence. Patients were determined to be persistent for 1 year if they had prescriptions filled more than 1 year after the index date. Patients whose enrollment period ended before the 1-year time frame were censored from the refill persistence calculation if they received enough pills at their last fill date to bring them within 30 days of their enrollment ending period. We calculated second year persistence in a similar manner. Our analysis was weighted more heavily with patients who persisted in their statin therapy because we required at least 2 post—index date statin prescriptions in order to determine MPR.

The AD cohort was further divided into those who received adequate AD treatment and those who did not. Adequacy of AD treatment was defined as the continuous use of an AD for 90 or more days at a minimum likely effective dose as determined by prescription refills.

Univariate analysis of adherence predictors was performed using t tests for normally distributed continuous variables and χ2 tests for dichotomous variables. For MPR, linear regression models were constructed to assess the impact of AD therapy after controlling for significant covariables such as age and sex. Logistic regression equations were used to model predictors of the dichotomous outcomes of 1-year and 2-year statin persistence. Because depressed patients with poor baseline statin adherence would likely have the most potential to benefit from AD therapy, we also planned an a priori analysis of all patients whose baseline statin adherence (MPR) was less than 80%. SAS version 9 was used for all analyses (SAS Institute, Cary, NC), and P <.05 was taken to indicate statistical significance. The study was approved by our institutional review board.

Patients with AD treatment were younger (54.0 years vs 55.4 years for patients without AD treatment) and more likely to be women (45% vs 37%; 

). Prevalences of diabetes (18% in both groups), hypertension (65% vs 63%), and ischemic heart disease (13% vs 10%) were similar in the 2 groups. Baseline adherence to statin medication was not significantly different in the AD group versus the non-AD group (MPR = 87%). Patients who received ADs had higher baseline total cholesterol levels (196 vs 193 mg/dL; P <.01).

Among patients who received AD treatment, 1769 received adequate treatment regimens (37%) and 3012 received inadequate regimens (63%). Clinical characteristics were similar in the adequately and inadequately treated patients (Table 1). Patients with adequate AD therapy had better baseline statin adherence than the inadequately treated group (90% vs 85%; P <.01). 

For the overall cohort, statin adherence for AD-treated and non—AD-treated patients changed less than 1% in the year after the index date. Patients who received inadequate AD treatment slightly increased their statin adherence in the first year and experienced no change in the second year. Adequately treated AD patients had no change in statin adherence in either the first or second year. There were 11,687 patients with poor baseline adherence (MPR <80%, 28% of analysis cohort) (

). In this poorly adherent subset, patients receiving adequate doses of ADs had greater MPR improvement (+18%, n = 414) in the first follow-up year than patients with either inadequate AD therapy (+12%, n = 974) or no AD therapy (+12%, n = 10,299; P <.01 for both comparisons) (

). In multivariate models that controlled for baseline differences between groups, AD adequacy was associated with a 6% greater MPR than inadequate therapy (P <.01) and a 7% greater MPR than no AD therapy (P <.01).

For the overall cohort, statin refill persistence remained high in the first follow-up year (88% for both groups) before dropping in the second year (67% for AD-treated patients and 68% for patients not treated with ADs; P = .33).

Persistence with statin therapy also was greater for the group receiving adequate AD treatment compared with patients receiving inadequate AD treatment or not receiving any AD treatment. Ninety-three percent of patients receiving adequate AD treatment persisted with statin therapy in the first follow-up year versus 85% of the inadequate AD group and 88% of the non-AD group (P <.01 for both comparisons). Similarly, 77% of the adequate AD group continued refilling statin prescriptions in the second year versus 61% of the inadequate AD group and 68% of the non-AD group (P <.01 for both comparisons). Differences in persistence remained significant in multiple regression analyses after controlling for age, sex, medical comorbidities, and baseline statin adherence.

The pattern of greater refill persistence among patients with adequate AD treatment also was seen in the subset of patients poorly adherent to statins at baseline (MPR <80%). Patients who were treated adequately with ADs had an 87% persistence rate in the first year, compared with an 82% persistence rate for patients who were not treated with ADs (P = .02) and an 80% persistence rate for those who were inadequately treated (P <.01). Statin refill persistence in the second year was 66% among patients with adequate treatment, 56% among patients not treated with ADs, and 51% among patients inadequately treated with ADs (P <.01 for both comparisons).

We used pharmacy claims data for a large cohort of insured patients prescribed statins to examine the effect of AD therapy on hyperlipidemia management. Among patients with poor baseline statin adherence, an adequate (eg, guideline concordant) course of AD therapy was associated with a greater tendency to persist with statin therapy and an increase in statin adherence among the patients who persisted with statin therapy. In contrast, patients receiving inadequate AD regimens had poorer rates of persistence and adherence that were not significantly different from those of the poorly adherent control patients not prescribed ADs. We infer from these findings that an effective course of AD therapy can have a significant impact on adherence behaviors for unrelated chronic medical conditions.

 found a significant correlation between AD treatment adequacy and statin adherence, but did not determine whether and how patient or disease factors might contribute to this correlation. Our work addressed this issue. We posited that patient factors include a tendency to take medications. This “adherence style” should be consistent across medications and over time. The impact of adherence style may explain our observation, consistent with that of Katon et al,

 that patients adherent to statins also tended to adhere to ADs. In contrast, disease factors (eg, interference with medication adherence related to depression1) may change over time and may respond to treatment. We suggest that resolution of depression-mediated interference with adherence may explain why patients who received adequate AD treatment tended to shift their statin adherence pattern from poor to good in the period after the AD treatment, whereas patients with inadequate AD treatment did not. 

Persistence (the tendency to remain on a statin by the 12th month of each calendar year) also was significantly greater for patients receiving adequate AD treatment in both years of the study. The difference between persistence and adherence underscores the complexity of patient behavior regarding use of medications for chronic illness over extended time periods and suggests that medication-taking styles may have multiple determinants.

We anticipated that patients with adequate AD treatment and good or improved statin adherence would show a reduction in total cholesterol compared with patients who had inadequate AD treatment and poor statin adherence. Rather, we found that patients with inadequate or no AD treatment had relatively lower total cholesterol than those with adequate AD treatment. This may be explained at least in part by the tendency of some selective serotonin reuptake inhibitors to raise cholesterol.

 Patients with adequate AD treatment were more likely to be exposed to the cholesterol-raising impact of selective serotonin reuptake inhibitors than those with inadequate or no AD treatment.

Our results have limitations and must be interpreted in the context of the study design. Without direct patient assessment, we could not determine whether all patients receiving ADs had depression. Some patients may have received AD treatment for anxiety, or anxiety plus depression. Patients receiving low-dose ADs were removed from the analysis, so it is likely we excluded most, if not all, patients receiving ADs for insomnia, pain, or other conditions where AD treatment adequacy would not be relevant. We could not determine whether AD treatment resulted in resolution of depression-related symptoms. However, other studies have generally found that adequate AD treatment is more likely than inadequate AD treatment to produce resolution of depression.

 Likewise, we could not assess the severity of patients’ depression, hyperlipidemia, or comorbid medical problems, so we could not determine whether illness severity before or after treatment played a role.

The pharmacy claims refill method is an indirect measure of adherence. In contrast to more direct methods such as self-report, pill counts, or electronic monitoring, however, patients are not aware that adherence is being measured and therefore do not alter their usual adherence behavior during the period of measurement.

 Moreover, although filling a prescription does not confirm actual ingestion of the medicines, the opposite remains true: patients with an insufficient number of pills for a given time period cannot be fully adherent. Thus, our method is well suited for identifying suboptimal adherence over time. Recognizing the inherent limitations of administrative database analyses, our study benefits from the large sample size and from our ability to assess 2 different parameters of medication adherence: daily pill taking and long-term refill persistence.

In summary, we demonstrated that among a large cohort of commercially insured patients receiving statin therapy, most patients with poor statin adherence in the first study year tended to remain poorly adherent in subsequent study years. This continued tendency toward poor statin adherence was likely driven by patient factors (eg, medication adherence style). An exception to the pattern of chronic poor adherence was seen in the subset of patients who received adequate courses of AD treatment. Our results suggest that an intervention—in this case, adequate AD treatment&mdash;that impacts disease-related factors may affect adherence and persistence with a medication prescribed for an unrelated chronic medical condition.

Because of the limitations inherent in this retrospective data analysis, further work is required to establish a causal relationship between AD treatment adequacy, resolution of depression, and change in adherence. However, given the high prevalence of depression and the impact of depression on medication adherence, mortality, and costs, future efforts to improve the management of patients with chronic medical conditions should include screening for and effectively treating depression, particularly among patients with poor medication adherence.

Among patients with poor statin adherence, adequate antidepressant treatment significantly improved adherence and persistence compared with inadequate or no antidepressant treatment.