Lipid-Lowering Drugs and the Risk of Depression and Suicidal Behavior

Background  A possible association between lipid-lowering drug therapy and psychological well-being remains an issue of debate. To provide more information, we performed a nested case-control study to evaluate the effect of lipid-lowering drugs on depression and suicidal behavior.

Methods  Within the United Kingdom General Practice Research Database, we identified all cases with newly treated depression needing a referral or hospitalization and all cases with first-recorded diagnosis of suicidal behavior between January 1, 1991, and December 31, 1999, from a study base that comprised all patients who were aged between 40 and 79 years and who had various exposures of interest. Each case was matched with up to 4 controls, randomly selected from the study base, on age, sex, medical practice, calendar time, and years since enrollment in the General Practice Research Database.

Results  A nested case-control analysis comprised 458 newly diagnosed cases of depression with 1830 controls, and 105 cases of suicidal behavior with 420 controls. The adjusted odds ratio of depression was 0.4 (95% confidence interval, 0.2-0.9) for current statin use, compared with hyperlipidemic nonuse. The adjusted odds ratios for other exposures were all around 1.0. None of the adjusted odds ratios for suicidal behavior were significantly different from unity.

Conclusions  The use of statins and other lipid-lowering drugs is not associated with an increased risk of depression or suicide. On the contrary, individuals with current statin use may have a lower risk of developing depression, an effect that could be explained by improved quality of life due to decreased risk of cardiovascular events or more health consciousness in patients receiving long-term treatment.

A POSSIBLE LINK between lowered serum cholesterol and psychological well-being remains an issue of debate. In 1990, Muldoon et al¹ in a review of 6 randomized trials found that the reductions in coronary heart disease mortality by lipid-lowering therapy were offset by a rise in noncardiovascular mortality, especially by a significant increase in deaths due to unintentional injury, suicide, or violence. Although there has been concern about the data analyzed in some trials,² this article was followed by several observational studies^3,4 and meta-analyses^5,6 that indicated that lowered serum cholesterol might be associated with depression or suicide. Experiments in primates also suggested that altered dietary fat could sufficiently change brain content and lead to an increase in aggressive behavior.⁷ Such findings raised concern about the potential adverse psychiatric effects of lipid-lowering drugs (LLDs), because these drugs, especially 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors (statins), are widely prescribed in daily practice. Several possible mechanisms were hypothesized to explain such an association.⁸

Not all studies^9-12 supported the link between lipid-lowering therapy and increased risk of depression or suicide. For example, Downs et al⁹ found that lovastatin treatment did not have any adverse effect on the quality of life among 1126 hyperlipidemic subjects completing 1 year of follow-up. In a trial of simvastatin, Wardle et al¹⁰ followed up 491 subjects for nearly 3 years and did not find significant differences in the use of psychotropic medication or in reports of mood-related symptoms between treatment groups. Muldoon et al¹³ in a recent meta-analysis also concluded that, in randomized trials using statins, noncardiovascular mortality was not significantly increased among subjects receiving treatment (odds ratio [OR], 0.84; 95% confidence interval [CI], 0.50-1.41), whereas a trend toward increased deaths from suicide and violence was observed in dietary and nonstatin LLD trials (OR, 1.32; 95% CI, 0.98-1.77).

Although results from recent clinical trials on statins did not suggest a higher risk of depression or suicide among patients receiving LLDs, the debated issue between LLD therapy and psychological well-being has not been completely resolved for the following reasons. First, the number of depressed subjects or number of suicides is usually small in individual clinical trials and the follow-up may not be long enough, which could result in inadequate statistical power to detect specific hazards. Second, it has been suggested that the rigorous eligibility criteria of clinical trials may have affected the study results. Third, the association between LLD use and suicidal ideation or attempts has not been assessed in clinical trials. Finally, although there is accumulating evidence that ensures the safety of statins, this is not true for nonstatin LLDs. Because the potential adverse psychiatric effects of LLDs are of great importance clinically and to the public health, we decided to study the association between LLDs and the risk of depression or suicidal behavior in a population-based research database to provide further information.

The United Kingdom General Practice Research Database (GPRD) has been described elsewhere.^14,15 The GPRD contains computerized records for some 3 million residents in the United Kingdom enrolled with selected general practitioners (GPs) who have agreed to supply anonymous, computerized data for research purposes. The recorded information includes patient demographics and characteristics (eg, height, weight, and smoking status), symptoms, medical diagnoses, referrals, hospitalizations, and drug prescriptions (including specific preparation, route of administration, dosage, and number of tablets for each prescription). On request, referral letters and hospital discharge summaries are available for review to validate the diagnoses. Previous studies^14,15 have documented the accuracy and completeness of the information recorded in the database.

Within the GPRD, we selected 3 groups of people who were aged 40 to 79 years between January 1, 1991, and December 31, 1999. Group 1 included patients who received 1 or more prescriptions for a statin (atorvastatin calcium, cerivastatin sodium, fluvastatin sodium, pravastatin sodium, or simvastatin), a fibrate (bezafibrate, ciprofibrate, clofibrate, fenofibrate, or gemfibrozil), or an LLD other than statins or fibrates (colestipol hydrochloride, cholestyramine, acipimox, nicotinic acid, or probucol). Group 2 comprised patients with a computer-recorded hyperlipidemia diagnosis who did not receive LLDs. Group 3 was a random sample of 50 000 patients who had neither a prescription for an LLD nor a diagnosis of hyperlipidemia at any time.

All subjects were included in the study population from January 1, 1991, or the date at which the first prescription records began (whichever came later) until they developed the outcomes of interest or any exclusion criteria, left the practice, died, or became 80 years of age, or the study period ended (whichever came first). We excluded individuals with a history of depression, suicidal behavior, mania, psychosis, schizophrenia, paranoid illnesses, degenerative neurologic diseases manifesting dementia (eg, Alzheimer disease), multiple sclerosis, chronic liver or renal diseases, human immunodeficiency virus infection, and alcohol or other drug abuse before January 1, 1991.

Within the study base, we identified 6765 subjects with a first-time diagnosis of treated depression (ie, depression with antidepressant treatment). Among them, 736 patients were referred to a consultant or hospitalized for the depression. We reviewed the computerized medical records for these 736 patients, without knowledge of exposure to LLDs, and considered a subject as depressed only if he or she was free of all of the exclusion criteria before the diagnosis of depression. In addition, subjects who had a diagnosis of cancer (excluding nonmelanoma skin cancer), treated epilepsy, acute hepatitis, chronic painful physical illnesses (eg, osteoarthritis) requiring continuous analgesic treatment, personality disorders, or other treated psychiatric illnesses (eg, anxiety) at the time of depression diagnosis were excluded. Subjects with specific causes for depression (eg, postpartum depression) were also excluded. The index date was the date of the first diagnosis for depression.

To validate computer-recorded diagnoses of depression, we requested a copy of any manual medical records (eg, referral letters) relating to the diagnoses from the GPs for a sample of randomly selected patients with a diagnosis of depression. Two investigators (C.-C.Y. and S.S.J.) reviewed these records without knowledge of exposures, and a final case status was determined. Any disagreement about case status between the 2 reviewers was discussed and resolved by consensus.

For patients with suicidal behavior, we identified all patients who had a computer-recorded diagnosis of suicide risk (Oxford Myocardial Infarction Incidence Study [OXMIS] code 3009B), ideation suicidal (code 3009BN), suicidal plan (code 3009BP), suicidal thoughts (code 3009BT), attempted suicide (codes 3009C and 9947TC), threat suicide (code 3009CT), suicidal drug overdose (codes 9779DL, 9779L, and 9779NA), drug overdose (code 9779A), parasuicide (code L3009P), or suicide (code 3009D). There were 14 potential cases of completed suicide and 204 potential incident cases of attempted suicide or suicidal ideation. The index date was the date of first relevant diagnosis. A subject was defined as a case if he or she was free of all of the exclusion criteria before the index date. In addition, a subject could not have a diagnosis of cancer, treated epilepsy, personality disorder, or chronic painful physical illnesses at the index date. We did not request manual records to validate the diagnoses of suicidal behavior, because it has previously been shown that the computer-recorded diagnoses for suicidal behavior were of satisfactory quality and completeness.¹⁶

We matched up to 4 randomly selected controls to each case on age (±2 years), sex, general practice, calendar time (by using the same index date as for the cases), and years of history in the GPRD (±1 year). Patients were eligible to become controls if they did not have the relevant outcome before the index date. The same exclusion criteria for the cases were applied to all controls. All cases and controls had 1 or more years of information in the GPRD before the index date.

We grouped exposure into statins, fibrates, other LLDs, mixed use (persons who took ≥2 drug classes concomitantly), hyperlipidemic nonuse, and normolipidemic nonuse. The exposure timing of various LLDs was further categorized as current (last prescription of an LLD ≤60 days before the index date), recent (LLD use 61-180 days prior), or past (LLD use ≥181 days prior). For current use, we classified the exposure duration as less than 3 months, 3 months to less than 1 year, 1 to less than 2 years, or 2 or more years of treatment. In the United Kingdom, a prescription for an LLD usually lasts for 30 days.

Potential confounders considered in this study were a record of stressful life events (eg, marital problem or bereavement) within 6 months, number of GP visits, referral to a psychiatrist in the prior year, smoking status, body mass index (BMI; calculated as weight in kilograms divided by the square of height in meters), use of corticosteroids, history of certain psychiatric diseases (treated or untreated, with exclusion of those disorders listed in the exclusion criteria) 6 months or longer before the index date, current benzodiazepine use, and various comorbidities (including diabetes mellitus, other endocrine disorders, cerebrovascular disease, ischemic heart disease, congestive heart failure, peripheral vascular disease, and hypertension). Corticosteroid use was classified as current (having had the last prescription of a corticosteroid within less than 3 months), recent (3-6 months), or nonuse (>6 months).

We used conditional logistic regression analysis to explore the association between the risk of various psychiatric outcomes and LLDs under study. All variables that had a P≤.20 in univariate analyses were entered into the multivariate model, and a stepwise procedure was then used to eliminate variables that became insignificant. Relative risks were estimated by ORs, calculated using hyperlipidemic nonuse as the reference group, and were given with 95% CIs. All analyses were performed using SAS software, version 6.12 (SAS Institute Inc, Cary, NC).

The study base comprised 94 441 individuals, including 30 703 subjects who received 1 or more prescriptions for LLDs, 13 738 individuals with a diagnosis of hyperlipidemia who did not take LLDs, and 50 000 normolipidemic subjects.

After review of the computerized records, 458 patients remained as cases of newly diagnosed depression. In the random sample of manual records received from the GPs, 25 (83.3%) of 30 patients were classified as having incident depression, defined as the first-time diagnosis of depression made by a psychiatrist, or the presence of 4 or more depressive symptoms and other diagnostic criteria listed in the DSM-IV-TR¹⁷ diagnostic criteria for major depressive episode when a referral letter from a psychiatrist was not available. Because the confirmation rate was high, we included all 458 patients in further analyses. The 1830 controls, comprised of 1590 controls from the same practice as their matched cases and 240 controls from other practices, were matched to the cases. The mean ± SD age was 55.4 ± 9.0 years for case and control patients, and male patients (53.0%) outnumbered female patients in the study. Cases and controls were different in many baseline characteristics, including number of GP visits, history of certain psychiatric diseases, preexisting peripheral vascular disease, smoking status, and BMI (Table 1). In addition, oral corticosteroid use, congestive heart failure, cerebrovascular disease, and prior referral to a psychiatrist were associated with depression (data not shown). The distribution of patient characteristics for cases with a diagnosis of depression and their controls is given in Table 1.

Table 2 presents the results of univariate and multivariate analysis for various exposures. In these analyses and in all subsequent analyses in this study, we included all cases and controls because the results were largely identical to those using only controls from the same practice as their matched cases. We also combined all nonstatin LLDs because there was no material difference between fibrates and other nonstatin LLDs on the risk of the outcomes. All mixed LLD users were combined as a group as well because there were few cases in this group. We further grouped recent and past LLD exposures and categorized treatment duration as less than 1 year or 1 or more years, as the within-group risks of the outcomes were similar.

Using hyperlipidemic nonuse as the reference group, the OR adjusted for smoking, BMI, number of GP visits, record of stressful life events, preexisting peripheral vascular disease, and history of psychiatric disorders was 0.4 (95% CI, 0.2-0.9) for current statin use. The adjusted ORs were not significant for other exposures and were 1.0 (95% CI, 0.5-2.1) for current nonstatin LLD use and 1.3 (95% CI, 0.9-1.9) for normolipidemic nonuse. Among current LLD users, the adjusted ORs for less than 1 year and 1 or more years of statin treatment were 0.7 (95% CI, 0.2-1.9) and 0.4 (95% CI, 0.1-0.9), respectively, while the relevant ORs for exposure to nonstatin LLDs were 1.0 (95% CI, 0.3-3.1) and 1.1 (95% CI, 0.5-2.3). In addition to statin use, current smoking, BMI of 30.0 or higher, stressful life events, more GP visits, history of certain psychiatric diseases, and diagnosis of peripheral vascular disease were independently associated with the risk of depression (Table 1).

We further stratified the analysis by age, sex, and certain comorbidities. The results of these subgroup analyses were similar. Current statin use, however, appeared to be associated with a somewhat lower risk of depression among patients with preexisting cardiovascular disease (mainly patients with ischemic heart disease [OR, 0.3; 95% CI, 0.1-1.0]) compared with patients without cardiovascular disease (OR, 0.5; 95% CI, 0.1-1.6). Patients with prior cardiovascular disease who had longer statin and nonstatin LLD use (≥1 year) also tended to have a lower risk of depression than subjects who received less than 1 year of treatment (data not shown).

For patients with suicidal behavior, 83 patients with attempted suicide, 14 with suicidal ideation, and 8 with completed suicide remained in the study after manual review. Four hundred twenty controls, including 355 controls from the same practice and 65 controls from other practices, were matched to the cases. The mean ± SD age was similar between cases (55.8 ± 8.5 years) and controls (55.8 ± 8.4 years), and there were more female patients (61.9%) with suicidal behavior in the study. Compared with controls, cases with suicidal behavior also had different baseline characteristics (Table 3). In addition to those characteristics listed in Table 3, prior referral to a psychiatrist, history of any endocrine disorder, systemic lupus erythematosus, and congestive heart failure were also more common among patients with suicidal behavior (data not shown).

The results of multivariate analysis for various exposures are given in Table 4. Using hyperlipidemic nonuse as the reference group, none of the adjusted ORs for LLD exposure or normolipidemia significantly differed from unity. The ORs adjusted for record of stressful life events, number of GP visits, current use of benzodiazepines, history of treated psychiatric disease, and diagnosis of peripheral vascular disease were 0.5 (95% CI, 0.1-1.5) for current statin use, 0.4 (95% CI, 0.1-1.8) for current nonstatin LLD use, and 0.7 (95% CI, 0.3-1.5) for normolipidemic nonuse, respectively. Unlike the finding among patients with depression, there was no effect of treatment duration for statins or nonstatin LLDs. Presence of stressful life events, more GP visits, current benzodiazepine use, history of treated psychiatric disease, and peripheral vascular disease diagnosis were associated with the risk of suicidal behavior (Table 3). This relation was independent of depression, because we excluded patients with a history of depression.

We further performed subgroup analyses among (1) patients with attempted suicide or suicide, (2) patients with only attempted suicide, and (3) patients with suicidal ideation. Although it was not possible to conduct multivariate analysis among patients with only suicidal ideation because of the small number of cases, all of the other available results were consistent with prior analyses and did not show an association between LLD use and the risk of suicidal behavior. Furthermore, there was no evidence of effect modification by age or sex.

The findings of this nested case-control analysis including 458 depressed cases and 105 patients with suicidal behavior indicate that exposure to LLDs or untreated hyperlipidemia does not increase the risk of depression or suicide. On the contrary, current statin use was associated with a substantially reduced risk of depression (OR, 0.4; 95% CI, 0.2-0.9), even after controlling for measured baseline differences between cases and controls. The negative association between current statin exposure and depression is especially prominent among patients who have 1 or more years of statin use and is somewhat more significant among patients with preexisting cardiovascular disease. A similar but weaker pattern of effect was observed among patients with preexisting cardiovascular disease who currently received nonstatin LLDs. The data in this study also show a relation between an increased risk of depression and current smoking, obesity, presence of stressful life events, more GP visits, history of certain psychiatric diseases, and preexisting peripheral vascular disease, as well as an association between the risk of suicidal behavior and presence of stressful life events, more GP visits, current benzodiazepine use, history of treated psychiatric disease, and peripheral vascular disease diagnosis. As expected, most of these covariates have previously been shown to be independent predictors of depression or suicide.^18-21 There was no evidence of effect modification by age or sex.

Although we find that current statin use is inversely associated with depression, such an association is not likely to be directly causal because there is no known pharmacological mechanism for this effect. In fact, despite the various proposed beneficial effects of statins and the slightly lower prevalence of depression among patients taking statins in certain clinical trials,^11,12 to our knowledge, neither statins nor other LLDs have been shown to be causally related to a lower risk of depression in humans. A possible explanation for such an inverse association could be an indirect effect of statins on the risk of depression through improved quality of life due to decreased incidence of cardiovascular events or more health consciousness and compliance among patients having longer LLD treatment.

Evidence from randomized clinical trials has shown the efficacy of LLDs in lowering serum cholesterol and reducing the incidence of coronary events or stroke.^22-25 The reduction in the risk of coronary events generally becomes prominent after 1 to 2 years of therapy, although the lipid-lowering effect of LLDs manifests rapidly.^22,23,26 Because patients with preexisting cardiovascular disease are likely to be less healthy and have a higher probability of developing depression due to the presence of cardiovascular and other comorbidities, it seems reasonable to speculate that by reducing the incidence of recurrent cardiovascular events, statins may increase the quality of life and subsequently decrease the risk of depression. Such an effect, if present, would also be greater after long-term therapy. Moreover, patients who receive prolonged lipid-lowering therapy may be more health conscious and adhere better to various medications and disease management programs,²⁷ which will further reduce their risk of developing depression. This could explain why we observed a similar pattern of effect among patients with preexisting cardiovascular disease who received nonstatin LLDs, although the effect is less prominent than that of statins, as it has previously been found that patients taking nonstatin LLDs are less adherent to LLD therapy compared with patients receiving statins.²⁸ Furthermore, statins are usually more effective than nonstatin LLDs in lowering serum cholesterol.²⁹

The validity of the results of any observational pharmacoepidemiologic studies is dependent on the quality of the study design and the completeness of the data elements that are used to derive the results. Although biases or unknown confounders cannot be completely eliminated in this study, we have tried to minimize their effects. First, the use of computerized prescription records to assess drug exposures is not subject to recall bias. The high quality of the exposure data recorded in the GPRD,¹⁴ and the fact that most LLDs are prescribed on an outpatient basis and are generally filled continually because of the chronic nature of hyperlipidemia, further lowers the possibility of exposure misclassification.

Second, case identification of computer-recorded depression requiring a referral or hospitalization and ascertainment of patients with suicidal behavior were virtually complete, and the process was not biased by exposures. We were also able to validate a random sample of depressed cases, and 83.3% of cases were classified as having depression. We did not validate all cases, because the confirmation rate was satisfactory. In addition, by matching cases and controls on practice and by controlling for other important confounders, the effect of variations in the GPs' diagnostic threshold for depression should be lower. Nondifferential misclassification, if present, would tend to diminish the ability to detect an apparent protective effect of statins. In the present study, we did not request manual records to validate the diagnoses of suicidal behavior, because these diagnoses have previously been shown to be accurate in the GPRD.¹⁶ Furthermore, we did not study depressed patients receiving nonstandard treatment or depressed patients taking antidepressants but whose diagnosis was difficult to validate, because in epidemiologic research it is generally important to include only well-validated cases in the analysis.

Although the rate of incident-treated depression found in the GPRD was comparable to findings in other studies,³⁰ inadequate recognition of depression has been a problem in general practice.^20,30,31 It has been shown that missed patients with depressive disorders are more likely to have physical illnesses and the severity of the disease is likely to be milder.^20,30,31 Because we only studied depressed patients needing a referral to a consultant or hospitalization and excluded patients with significant physical illnesses from the study, the effects of such a misclassification should have been substantially reduced. Furthermore, by matching and controlling for various potential confounders, the misclassification is likely to be nondifferential, which again would diminish, but not increase, the ability to detect a beneficial effect of statins.

Many psychiatric and physical illnesses are known predictors of depression or suicide¹⁸ and may confound the relation between LLD use and the outcomes if they are also related to the exposures. Therefore, one of the major concerns in studying the association between LLDs and depression or suicide is the possibility that patient characteristics, which are predictors of the outcomes, may be different between patients with statin or other exposures. To control for these potential confounders, we not only restricted the study to subjects without diseases that are important predictors of the outcomes (eg, alcoholism) but also controlled for other confounders, such as general practice, number of GP visits, and comorbidities, through matching and regression modeling. Although we did not have complete information on smoking status and BMI, they affected outcomes. Furthermore, complete control of these 2 factors in the analysis should yield an even stronger protective effect of statins, because smoking and BMI are positively associated with cardiovascular disease³² and depression^18,19 and patients having these factors would be more likely to receive LLDs.³³

In the present study, we also did not have information on physical activity. Lack of exercise is a risk factor for coronary heart disease,^32,34 and physical activity has been found to be inversely associated with depression.³⁵ Because hyperlipidemic patients who had this risk factor would be more likely to receive LLDs than hyperlipidemic patients without this factor,³³ controlling for physical activity should result in a more apparent inverse association between statin use and depression. Lack of information on baseline serum cholesterol levels and dietary behavior was not likely to significantly change the main findings of this study as well. Although serum cholesterol levels and failure of dietary modification are likely to be determinants of LLD treatment, and low serum cholesterol has been shown to be possibly associated with depression,³⁶ to our knowledge, there is no established relation between hyperlipidemia or diet and depression.

In conclusion, our findings do not support the hypothesis that use of LLDs increases the risk of depression or suicide. On the contrary, we found that current statin use was associated with a reduced risk of depression. The association was much stronger among patients with long-term use of statins and was somewhat more significant among patients with preexisting cardiovascular disease.

Corresponding author and reprints: Susan S. Jick, DSc, Boston Collaborative Drug Surveillance Program, Boston University School of Medicine, 11 Muzzey St, Lexington, MA 02421 (e-mail: sjick@bu.edu).

Accepted for publication October 31, 2002.

The Boston Collaborative Drug Surveillance Program is supported in part by grants from Berlex Laboratories, Montville, NJ, GlaxoSmithKline, Research Triangle Park, NC, Ingenix Pharmaceutical Services, Newton, Mass, Johnson & Johnson Pharmaceutical Research & Development LLC, Titusville, NJ, Hoffmann-La Roche, Basil, Switzerland, Pharmacia Corporation, Peapack, NJ, and Novartis Farmacéutica, Barcelona, Spain. Dr Yang received support from the Harvard Pharmacoepidemiology Teaching and Research Fund, the Yin Xun-Ruo Educational Foundation, and the Foundation for Poison Control in Taiwan.

We thank Marcia A. Testa, PhD, MPH, K. Arnold Chan, MPH, MD, Kimberly M. Thompson, ScD, MS, and I-Chao Liu, ScD, MD, MS, for their valuable help and advice. We also thank the participating general practitioners for their excellent cooperation and generous help.