To examine the association between atypical antipsychotic medications and incident treatment for diabetes mellitus or hyperlipidemia in elderly adults without diagnoses of schizophrenia or bipolar disorder.
To examine the association between atypical antipsychotic medications and incident treatment for diabetes mellitus or hyperlipidemia in elderly adults without diagnoses of schizophrenia or bipolar disorder.
Two case–control studies using medical and pharmacy claims data.
United States managed care population from multiple insurance plans.
Individuals aged 65 and older enrolled in a Medicare Advantage or commercial (health maintenance organization) managed care health plan in the western United States with no claims indicating diagnosis of schizophrenia or bipolar disorder in the 1 year pre-index period. Cases were defined as persons newly initiated on an antidiabetic (n = 13,075) or antihyperlipidemic (n = 63,829) medication on the index date. For the new diabetes mellitus analysis, 65,375 controls were matched to cases based on age, sex, health-plan type, and index date year. In the new hyperlipidemia analysis, 63,829 controls were matched to cases based on the same variables.
Conditional logistic regressions were performed to determine the odds of initiated antidiabetic or antihyperlipidemic medication for participants exposed to atypical antipsychotics compared with those with no exposure. The models included comorbidities possibly associated with the outcome.
Exposure to atypical antipsychotics was associated with significantly greater adjusted odds of starting an antidiabetic medication (1.32, 95% confidence interval (CI) = 1.10–1.59) but significantly lower odds of starting an antihyperlipidemic medication (0.76, 95% CI = 0.67–0.87).
Use of atypical antipsychotics in older adults for conditions other than schizophrenia and bipolar disorder was associated with incident treatment of diabetes mellitus but not of hyperlipidemia, suggesting that older adults may be susceptible to the adverse metabolic consequences of these agents.
Atypical antipsychotics are associated with undesirable metabolic effects in adults and children with schizophrenia and bipolar disorder, the two conditions for which all atypical antipsychotics have received Food and Drug Administration (FDA) indications. The rates of metabolic syndrome, dyslipidemia, and diabetes mellitus are significantly higher in people taking atypical antipsychotics than in those taking conventional antipsychotics or mood stabilizers to treat schizophrenia or bipolar disorder.[1-4] The metabolic risks of atypical antipsychotic treatment in elderly adults has been evaluated in few studies and may be different from those in younger individuals because of the different indications for use and the lower doses that are typically used. In a retrospective analysis of individuals filling prescriptions for atypical antipsychotics from May 2007 through April 2008, 71% of those younger than 65 had schizophrenia or bipolar disorder, whereas only 38% of those aged 65 and older had these indications.
There is some evidence that atypical antipsychotics do not confer significant metabolic risk in the elderly population,[7-12] but previous studies may not have been adequately powered to detect such events. To better understand the risk of metabolic adverse effects with atypical antipsychotic treatment in elderly adults, two case–control studies were conducted in a large group of individuals aged 65 and older using electronic pharmacy and medical claims. The aim was to explore the association between atypical antipsychotic use and the initiation of treatment for diabetes mellitus or hyperlipidemia. It was hypothesized that individuals with exposure to atypical antipsychotics would be more likely to initiate pharmacological treatment for diabetes mellitus or hyperlipidemia than similar individuals without such exposure.
Two independent case–control studies were performed using electronic pharmacy and medical claims and enrollment data within a large electronic administrative claims database for multiple managed care health plans with Medicare Advantage Prescription Drug (MAPD) and commercial patients (health maintenance organization) residing in the western United States. An external institutional review board reviewed these studies and gave a determination that the research did not involve human subjects because limited data sets without patient identifiers were used for the analysis.
The objective of the first case–control study was to determine whether atypical antipsychotic exposure is associated with greater odds of new treatment-dependent diabetes mellitus. Cases were defined as having newly started an antidiabetic medication during January 1, 2004, to December 31, 2008 (the identification period). The index date was defined as the date of the first fill of an antidiabetic medication. Controls had no treatment with an antidiabetic medication and had index dates randomly selected during the identification period. To be considered as a case or control for the new-onset diabetes study, the individual must have been continually enrolled in pharmacy and medical benefits for 1 year before the index date (the pre-index period) through the index date and must have been aged 65 and older on the index date. Controls were matched 5:1 to cases based on age at index date, sex, health plan type (commercial or Medicare Advantage), and index date year. The 5:1 matching was based on the number of controls available to maximize the power of the study.
Individuals were excluded from the study if a medical claim for a hospitalization, emergency department (ED) visit or outpatient office visit was found with an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis code for diabetes mellitus, schizophrenia, or bipolar disorder in any field (Supplementary Appendix Table S1) during the pre-index period or if a prescription claim for an antidiabetic agent was identified in the pre-index period. Individuals with diagnoses of schizophrenia or bipolar disorder were excluded from the analyses, because it could not be assumed that they had no prior exposure to atypical antipsychotics.
The objective of the second study was to determine whether atypical antipsychotic exposure was associated with greater odds of treatment for hyperlipidemia. The cases for this study were defined as having newly started an antihyperlipidemic medication. The same identification period was used as the diabetes mellitus case–control study. The index date was the date of the first fill of an antihyperlipidemic medication for cases and was randomly selected for controls during the identification period. For the hyperlipidemia study, cases and controls must have been continually enrolled in pharmacy and medical benefits during the pre-index period (1 year before index date) through the index date and must have been aged 65 and older on the index date. Individuals were excluded from the study if they had a medical claim for a hospitalization, ED visit, or outpatient office visit for hyperlipidemia, schizophrenia, or bipolar disorder (Table 1) during the pre-index period or if a prescription claim for an antihyperlipidemic drug was identified during the pre-index period. Controls were matched 1:1 to cases based on the same variables used for matching in the diabetes mellitus case–control study and the available number of potential controls while optimizing the number of case cohort members to maximize the power of the study.
|Characteristic||Diabetes Mellitus Analysis||Hyperlipidemia Analysis|
|Cases (n = 13,075)||Controls (n = 65,375)||Cases (n = 63,829)||Controls (n = 63,829)|
|Age, mean ± SD||75.5 ± 6.8||75.5 ± 6.8||76.5 ± 6.8||76.5 ± 6.8|
|Age, n (%)|
|65–74||6,379 (48.8)||31,895 (48.8)||27,120 (42.5)||27,120 (42.5)|
|75–84||5,204 (39.8)||26,020 (39.8)||28,171 (44.1)||28,171 (44.1)|
|≥85||1,492 (11.4)||7,460 (11.4)||8,538 (13.4)||8,538 (13.4)|
|Sex, n (%)|
|Female||7,158 (54.8)||35,790 (54.8)||36,487 (57.2)||36,487 (57.2)|
|Male||5,917 (45.3)||29,585 (45.3)||27,342 (42.8)||27,342 (42.8)|
|Health plan, n (%)|
|Commercial||1,116 (8.5)||5,580 (8.5)||4,875 (7.6)||4,875 (7.6)|
|Medicare||11,959 (91.5)||59,795 (91.5)||58,954 (92.4)||58,954 (92.4)|
|Charlson Comorbidity Index, mean ± SD||1.5 ± 2.2||1.0 ± 1.7||1.8 ± 2.2||1.1 ± 1.9|
|Comorbidities, n (%)|
|Diabetes mellitus medical or pharmacy claim||N/A||N/A||17,917 (28.1)||7,837 (12.3)|
|Hyperlipidemia medical or pharmacy claim||7,436 (56.9)||32,389 (49.5)||N/A||N/A|
|Hypertension medical or pharmacy claim||9,865 (75.5)||41,429 (63.4)||N/A||N/A|
|Stroke, ischemic heart disease, coronary heart disease medical claim||N/A||N/A||22,318 (35.0)||7,506 (11.8)|
|Obesity medical claim||773 (5.9)||1,684 (2.6)||2,233 (3.5)||1,400 (2.2)|
|Dementia medical claim||769 (5.9)||2,802 (4.3)||3,539 (5.5)||3,528 (5.5)|
|Anxiety disorder medical claim||722 (5.5)||3,468 (5.3)||3,583 (5.6)||3,127 (4.9)|
|Adjustment disorder medical claim||73 (0.6)||358 (0.6)||346 (0.5)||346 (0.5)|
|Depression medical or pharmacy claim||1,643 (12.6)||7,052 (10.8)||7,184 (11.3)||6,065 (9.5)|
Atypical antipsychotics included aripiprazole, clozapine, olanzapine, paliperidone, quetiapine, risperidone, and ziprasidone. Patients taking atypical antipsychotics were stratified according to the individual drug, the number of days of exposure in the pre-index period, and dosage of therapy as defined in the American Psychiatric Association (APA) guideline for the treatment of dementia. Atypical antipsychotic usage was considered low dose if the daily dose was less than or equal to the maximum daily doses recommended in the APA guideline (olanzapine 10 mg/d, clozapine 100 mg/d, risperidone 2 mg/d, quetiapine 300 mg/d, aripiprazole 15 mg/d). High-dose usage was defined as exceeding these recommended upper limits. Because the guideline did not include ziprasidone, the upper limit of the low-dose ziprasidone (80 mg/d) was taken from an exploratory trial in individuals with dementia. Comorbidities identified during the pre-index period for all cases and controls were diabetes mellitus, hyperlipidemia, obesity, hypertension, dementia, anxiety, adjustment disorders, and depression (Table 1). The hyperlipidemia case–control study examined the number of individuals in either cohort with a composite variable of diagnosis of stroke, ischemic heart disease, or coronary heart disease during the pre-index period, which was intended as a proxy for cardiovascular disease. Prescription claims for antidiabetic, antihyperlipidemic, antihypertensive, and antidepressant medications were also identified during the pre-index period.
A modified Charlson Comorbidity Index score was determined for each individual. The adapted index contains 17 diagnostic categories and was developed specifically for use with administrative claims databases.
Statistical analyses were conducted using SAS version 9.1 (SAS Institute Inc., Cary, NC). Descriptive statistics are reported for each cohort. Means were compared using t-tests, and percentages were compared using Pearson chi-square tests. All comparisons were two-sided, using P < .05 to define statistical significance. Conditional logistic regressions were performed for each case–control analysis with yes or no dependent variables for new-onset treatment-dependent diabetes mellitus for the first study and new-onset treatment for hyperlipidemia for the second study. Unadjusted and adjusted odds ratios were reported. The effect of possible confounding variables on the outcome was also assessed in separate conditional logistic regression models. Covariates for the separate conditional logistic regression models included Charlson Comorbidity Index and the following comorbidities that could contribute to the outcome: diabetes mellitus (hyperlipidemia study only); hyperlipidemia (diabetes mellitus study only); hypertension (diabetes mellitus study only); composite of stroke, ischemic heart disease, or coronary heart disease (hyperlipidemia study only); and obesity. Psychiatric comorbidities that atypical antipsychotics may possibly be used to treat on an outpatient basis (anxiety, adjustment disorders, depression, and dementia) were also included in the logistic regression models because there may be interactions between psychiatric indications and medical conditions.
For new-onset diabetes mellitus, 13,075 individuals met case criteria, and 385,512 met control criteria. Five controls were matched to every case identified for a total analytical sample of 13,075 cases and 65,375 controls. The matched individuals had a mean age of 75.5 years and were 54.8% female (Table 1). Most individuals were enrolled in a Medicare Advantage plan (91.5%). Hyperlipidemia and hypertension diagnoses were common during the pre-index period in cases and controls.
Of cases for the diabetes mellitus study, 1.3% had atypical antipsychotic exposure in the pre-index period, compared with 0.8% of controls. The unadjusted odds ratio (OR) for new-onset treatment-dependent diabetes mellitus was 1.63 (95% confidence interval (CI) = 1.37–1.94) for individuals with any versus no atypical antipsychotic exposure (P < .001) (Table 2). When the Charlson Comorbidity Index and specific comorbidities were added as covariates in the conditional logistic regression, the adjusted OR was 1.32 (95% CI = 1.10–1.59) for patients with versus without atypical antipsychotic exposure (P = .004) (Table 3). Covariates indicating greater odds of new-onset diabetes mellitus included greater overall burden of comorbidity (Charlson Comorbidity Index), diagnosis or treatment of hyperlipidemia or hypertension, diagnosis of obesity, and diagnosis of dementia. Diagnosis of anxiety during the pre-index period was associated with lower risk of initiation of antidiabetic medication.
|Analysis||n (%)||Odds Ratio (95% Confidence Interval)||P-Value|
|Diabetes mellitus study: atypical antipsychotic exposure &!ensp;versus no atypical antipsychotic exposure &!ensp;(13,075 cases; 65,375 controls)||174 (1.3)||542 (0.8)||1.63 (1.37–1.94)||<.001|
|Hyperlipidemia study: atypical antipsychotic exposure &!ensp;versus no atypical antipsychotic exposure &!ensp;(63,829 cases; 63,829 controls)||505 (0.8)||630 (1.0)||0.80 (0.71–0.90)||<.001|
|Parameter (Measured in Pre-Index Period)||Adjusted Odds Ratio (95% Confidence Interval) P-Value|
|Diabetes Mellitus Analysis (13,075 cases; 65,375 controls)||Hyperlipidemia Analysis (63,829 cases; 63,829 controls)|
|Atypical antipsychotic exposure &!ensp;versus no exposure||1.32 (1.10–1.59) .004||0.76 (0.67–0.87) <.001|
|Charlson Comorbidity Index||1.12 (1.11–1.13) <.001||1.04 (1.03–1.05) <.001|
|Diabetes mellitus medical or pharmacy claim||N/A||2.48 (2.40–2.56) <.001|
|Hyperlipidemia medical or pharmacy claim||1.17 (1.12–1.22) <.001||N/A|
|Hypertension medical or pharmacy claim||1.58 (1.51–1.66) <.001||N/A|
|Stroke, ischemic heart disease, &!ensp;coronary heart disease medical claim||N/A||3.97 (3.85–4.10) <.001|
|Obesity medical claim||2.12 (1.94–2.32) <.001||1.22 (1.13–1.31) <.001|
|Dementia medical claim||1.24 (1.14–1.36) <.001||0.80 (0.76–0.85) <.001|
|Anxiety disorder medical claim||0.90 (0.83–0.98) .02||1.01 (0.96–1.07) .75|
|Adjustment disorder medical claim||0.92 (0.71–1.20) .55||0.83 (0.71–0.98) .03|
|Depression medical or pharmacy claim||1.04 (0.97–1.10) .28||1.08 (1.04–1.12) <.001|
For the hyperlipidemia case–control study, there were 63,832 individuals who met the case criteria and 216,456 who met control criteria. After matching, there were 63,829 cases and 63,829 controls. The matching ratio was 1:1, based on the number of available controls. The mean age was 76.5, and 57.2% were female. Most individuals (92.4%) were enrolled in a Medicare Advantage plan. Of identified cases for the hyperlipidemia study, 0.8% had atypical antipsychotic exposure in the pre-index period, compared with 1.0% of controls. The unadjusted OR for new-onset treatment-dependent hyperlipidemia was 0.80 (95% CI = 0.71–0.90) for exposed versus unexposed individuals (P < .001). The adjusted OR was 0.76 (95% CI = 0.67–0.87) (Table 3). Covariates indicating greater odds of antihyperlipidemic initiation included greater burden of comorbidity, diagnosis or treatment of diabetes mellitus, diagnosis of obesity, and diagnosis or treatment of depression. A composite of diagnosis of stroke, ischemic heart disease, or coronary heart disease had an OR of 3.97 (95% CI = 3.85–4.10). Diagnosis of dementia and diagnosis of adjustment disorders were associated with lower odds of antihyperlipidemic mediation initiation.
Atypical antipsychotic exposure was primarily to risperidone (30–42% of exposed individuals in each cohort) and quetiapine (31–40% of exposed individuals in each cohort). Olanzapine was the third most prevalent atypical antipsychotic (21–26% of exposed individuals in each cohort). Exposure to aripiprazole, clozapine, and ziprasidone was minimal. No claims for paliperidone were identified. The high-dose vs low-dose atypical antipsychotic dichotomous variable was not included in the regression models because 97.0% to 97.8% of those with atypical antipsychotic exposure in the two studies were within the daily dose recommended by the APA for the treatment of individuals with Alzheimer's disease and other dementias.
The median gap between the end of days supply of atypical antipsychotics and the index date for cases was 0 days for those with atypical antipsychotic exposure in both studies, indicating current exposure at the time of event. Of diabetes mellitus cases with atypical antipsychotic exposure, 76.4% had exposure in the 30-day period before the index date, compared with 71.8% of controls. Of hyperlipidemia cases with atypical antipsychotic exposure, 70.1% had exposure in the 30 days before the index date, compared with 72.8% of controls.
The development of diabetes mellitus, as indicated by an antidiabetic prescription fill, may be of concern in elderly adults taking atypical antipsychotics, perhaps even at the low doses usually recommended for treatment of neurobehavioral complications of dementia, although rates of initiation of medication for lipid management were lower after antipsychotic exposure.
Although the finding of greater odds of starting an antidiabetic agent after atypical antipsychotic exposure in elderly adults may appear to conflict with past research,[6-11] previous studies may not have been adequately powered to detect the incidence or risk of diabetes mellitus. Prior retrospective chart reviews and a subgroup analysis of a cross-sectional population study[6-8] did not show an association between development of diabetes mellitus and use of atypical antipsychotics in elderly adults, however, their small sample sizes limited the ability to draw firm conclusions. The current study was able to study more events of new-onset diabetes mellitus by using claims data from a large managed care organization together with the efficiency of a case–control study design.
One of the larger prior studies was a retrospective cohort study that compared 3,250 elderly residents of long-term care taking olanzapine, quetiapine, or risperidone with 5,326 similar residents taking benzodiazepines. The diabetes mellitus incidence rate was found to be 31/1,000 person-years for the atypical antipsychotic cohort and 40/1,000 person-years for the benzodiazepine cohort. The authors concluded that atypical antipsychotic agents in older adults did not increase the risk of developing diabetes mellitus but did not comment on the number of individuals required to provide adequate power for detecting new-onset diabetes mellitus in this population. Although the prevalence of diagnosed diabetes mellitus in persons aged 65 to 74 in the United States was reported at 19.1% in 2007 and 23.1% of all those aged 60 and older, incident cases of diabetes mellitus in elderly adults are relatively rare. The Canadian Study of Health and Aging estimated the annual incidence of diabetes mellitus to be 8.6/1,000 for older adults, with the incidence decreasing with age. The Personnes âgées Quid Epidemiological Survey in France found the incidence of drug-treated diabetes mellitus to be 3.8/1,000 person-years in 3,777 individuals aged 65 and older. These figures suggest that the sample size needed for a cohort study or a prospective trial examining differences in new-onset diabetes mellitus in groups of elderly adults would need to be large.
The Clinical Antipsychotic Trials of Intervention Effectiveness in Alzheimer's Disease is the only prospective study to examine the effect of atypical antipsychotics on metabolic outcomes in a population of elderly adults. Outpatients with Alzheimer's disease were randomized to olanzapine, quetiapine, or risperidone, and outcomes were measured over 36 weeks. Serum glucose and plasma lipid assays were collected, but data were excluded because blood samples were taken when participants had not been fasting. Olanzapine use was associated with significantly lower high-density lipoprotein cholesterol (HDL-C). No treatment effects were seen on glucose or triglycerides. The authors stated that these outcomes were provisional because of power limitations, but women had significant increases in weight and body mass index during atypical antipsychotic use, an effect not seen in men. The metabolic or other health significance of this weight gain is unknown.
The lower likelihood of initiating hyperlipidemia treatment observed in the current study may not correspond to a lower likelihood of developing hyperlipidemia but may instead be reflective of lower enthusiasm of healthcare providers for initiating a treatment with primarily long-term benefits if the individual's health is already in decline. Support for this possibility was found in the present study; a diagnosis of dementia was, by itself, associated with similarly lower odds of initiating antihyperlipidemic treatment when examined in relation to the other variables in the model. Although initiation of antihyperlipidemic medication is common in elderly adults, new-onset hyperlipidemia independent of iatrogenic causes may be rare. Total and low-density lipoprotein cholesterol levels increase from puberty until approximately age 65 in men and 75 in women, at which point levels begin to fall. In the current study, individuals with an ICD-9-CM code for stroke, ischemic heart disease, or coronary heart disease were four times as likely to initiate hyperlipidemia treatment as those without one of those diagnoses. Initiating treatment for hyperlipidemia may therefore be primarily driven by cardiovascular events rather than new-onset dyslipidemias. Cardiovascular risk without a prior event is likely to be of secondary importance for individuals experiencing agitation or psychosis associated with dementia. There is also the possibility of some other unknown and therefore unmeasured confounding variable.
This study was not able to account for differential effects of atypical antipsychotics because of small numbers of individuals for each individual medication. In 2003, the FDA required labeling changes for all atypical antipsychotics regarding the risk of diabetes mellitus.[21, 22] Since that time, atypical antipsychotics have been shown to differ with regard to this risk. The risks for dyslipidemias are also not evenly distributed between the atypical antipsychotic class. Olanzapine is associated with more weight gain and increases in glucose and cholesterol levels than all other atypical antipsychotics except for clozapine. Aripiprazole does not seem to be associated with the same increases in cholesterol levels as risperidone and olanzapine, and ziprasidone has shown an absence of adverse effects on lipid profile and glycosylated hemoglobin, unlike the undesirable effects seen with olanzapine.[25, 26] Quetiapine appears to have significant adverse effects on HDL-C and fasting triglycerides not seen with risperidone.
Other limitations of the current studies include those inherent in database analyses, such as lack of information on disease severity and possible miscoding of diagnoses or events. It was not possible to validate diagnoses identified in medical claims, and not all cases of diabetes mellitus or hyperlipidemia may have been identified, because the claims databases do not include laboratory results. Pharmacy and medical claims also lack information about ethnicity and family history, important risk factors for metabolic dysfunction. It is possible that the potential confounding factor of obesity was not entirely taken into account, because obesity is unreliably coded in medical claims.
The greater odds of new-onset diabetes mellitus associated with atypical antipsychotic treatment does not imply causation, and the odds of developing diabetes mellitus or hyperlipidemia with atypical antipsychotic exposure could have been underestimated in this study because many individuals may try lifestyle modification before pharmacological treatment. Moreover, there may be other contributing factors not accounted for in the analyses, such as total duration of exposure and higher-dose treatment, both of which may affect metabolic risks, and effects of other psychiatric or nonpsychiatric conditions not included in the analyses. The data cannot explain the lower odds of initiation of antihyperlipidemic therapy in elderly adults treated with atypical antipsychotics, but this finding could indicate that physicians elect not to treat or monitor lipids as aggressively in this group as they might in other patients.
Despite these limitations, the analysis suggests that elderly adults started on atypical antipsychotics, even in low doses and for conditions other than schizophrenia or bipolar disorder, are at risk for new-onset diabetes mellitus. Clinicians need to be aware that older persons may not be protected from the adverse metabolic effects of atypical antipsychotic agents. Although the magnitude of this risk is not great, it will be important for some individuals and for population-based approaches to health care of older adults.
Conflict of Interest: The editor in chief has reviewed the conflict of interest checklist provided by the authors and has determined that the authors have no financial or any other kind of personal conflicts with this paper.
Author Contributions: Dr. Erickson: Primarily responsible for design, interpretation of results, and preparation of the manuscript. Ms. Le: Primarily responsible for the data extraction and analysis. Dr. Zakharyan: Contributed to the statistical design. Drs. Stockl, Harada, Borson, and Ramsey: Contributed to the study design and manuscript preparation. Dr. Curtis: Reviewed the manuscript.
Sponsor's Role: None.