Monitoring for Metabolic Side Effects Among Outpatients With Dementia Receiving Antipsychotics

The most common use of second-generation antipsychotics for older adults is to ameliorate behavioral disturbances associated with dementia (1). In 2004, the Expert Consensus Panel for Using Antipsychotic Drugs in Older Patients recommended second-generation antipsychotics as an option for patients with dementia (1,2). However, the efficacy and effectiveness of second-generation antipsychotics in treating neurobehavioral symptoms of dementia are at best modest and not well established (3–6).

Nearly all research and guidelines regarding antipsychotics and neuropsychiatric symptoms of dementia (7) recommend caution when using these agents because of significantly greater risks of cerebrovascular events (CVEs) and mortality compared with a placebo (8–10). Because of these concerns, the U.S. Food and Drug Administration (FDA) issued a black-box warning in 2005 identifying the association between second-generation antipsychotic use among elderly patients with dementia and increased mortality (8); a similar warning was issued for first-generation antipsychotics in 2008 (9).

Antipsychotics, especially second-generation antipsychotics, are associated with increased risk of metabolic side effects, such as weight gain, diabetes, and hyperlipidemia (11). In 2003, the FDA required that antipsychotic product labeling include a warning about hyperglycemia and diabetes and recommended monitoring of fasting blood glucose among patients with diabetes, diabetes risk factors, or symptoms of hyperglycemia. In 2004, the American Diabetes Association (ADA) and American Psychiatric Association (APA) published expert consensus recommendations that providers should consider the relative risk of various antipsychotic agents; monitor weight, glucose or glycosylated hemoglobin (HBA1c), and lipids when a patient begins a new antipsychotic medication; and should continue to monitor these metabolic parameters, periodically assessing and treating any abnormalities identified (12). Despite the expert recommendations and the FDA warning, the rates of metabolic monitoring for patients with schizophrenia and related diagnoses who are prescribed antipsychotics remained low among Medicaid enrollees (13), veterans (14), and commercially insured patients (15).

Although some authors have suggested that metabolic side effects of antipsychotics in the elderly population may be somewhat attenuated compared with younger populations (16), patients with dementia are still at risk of developing metabolic side effects. In a small study (N=59) using data from the U.S. Department of Veterans Affairs (VA), Mathys and colleagues (17) investigated incidence of metabolic abnormalities among patients with dementia after they began taking antipsychotics and reported that 10% developed impaired glucose tolerance, 8.9% gained weight, and 14.5% developed lipid abnormalities during the one-year follow-up. In the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE study), second-generation antipsychotics did not appear to affect glucose or triglyceride levels, but clinically significant weight gain (>7%) was observed in the treatment groups compared with the placebo groups (18).

In addition, Mathys and colleagues (17) found that 51.8% of patients treated with antipsychotics for dementia-related behavioral disturbances were not monitored for glucose at baseline or follow-up. Other studies have suggested that the presence of dementia is associated with lower monitoring for diabetes mellitus (19). However, we are not aware of any studies that have systematically examined monitoring for metabolic side effects among patients with dementia for whom antipsychotics are prescribed.

This study helped fill the gaps in the existing literature by examining recent practices for guideline-concordant (20) monitoring of metabolic side effects (weight gain, hyperglycemia, and hyperlipidemia) among older VA outpatients with dementia at the time a new antipsychotic medication is prescribed for any reason (baseline) and at a three-month follow-up. We compared the frequency of metabolic monitoring among patients with dementia and no psychotic disorder diagnosis (not an FDA-approved indication for antipsychotic use) and patients with psychotic disorders and no dementia (for whom antipsychotics are FDA approved). We expected that the frequency of monitoring for antipsychotic side effects for patients with dementia without psychosis would be higher because of a lack of FDA indication for the use of antipsychotics in this group compared with patients with psychotic disorders.

Methods

Study design and data source

This study was a retrospective cohort analysis that included patients prescribed antipsychotics in 32 VA medical centers within the Veterans Integrated Service Networks 18–22. Data on service utilization, diagnosis, prescribed medications, and laboratory tests were extracted from Veterans Health Administration medical SAS data sets, and vital signs were extracted from the Region 1 Corporate Data Warehouse.

This project was approved by the Central Arkansas Veterans Healthcare System Institutional Review Board and Research and Development Committee.

Patient selection

The dementia group (N=3,903) included patients with a diagnosis of dementia but no psychotic disorder (schizophrenia, bipolar disorder, or other psychotic disorders) who were prescribed a new antipsychotic prescription (the “index medication”) between October 1, 2005, and September 30, 2011, with at least a 60-day supply in the subsequent 90 days. The psychosis group (N=5,779) included patients with a psychotic disorder but without dementia who met the same inclusion criteria regarding a new antipsychotic prescription. ICD-9-CM codes were used for selecting the groups: schizophrenia (295.0X−295.4X and 295.6X−295.9X), bipolar disorder (296.0X, 296.1X and 296.4X−296.8X), other psychotic disorders (293.81, 293.82, 297.0X−297.3X, 297.8X, 297.9X, 298.0X−298.4X, 298.8X, and 298.9X), and dementia (290.0, 290.1X–290.4X, 291.2, 294.10, 294.11, 294.20, 294.21, 331.0, 331.1X, and 331.82).

We identified patients clearly in need of monitoring for a new episode of treatment by including patients who had a new antipsychotic medication (that is, one that had not been prescribed in the previous 180 days) and who also had a stable medication regimen (that is, no other new antipsychotic medications had been prescribed in the prior six months). Patients were included whether the index prescription was a new antipsychotic start (that is, for a patient not prescribed any antipsychotic agent in the previous 180 days), a switch to a different antipsychotic agent, or the addition of a different antipsychotic medication to ongoing antipsychotic treatment. If more than one new prescription was identified within the study period, we selected the most recent occurrence. We excluded patients who had hospital or extended-care stays (for example, a nursing home or residential program) from 30 days before through 120 days after the index prescription date to focus on outpatient monitoring practices. In analyses not reported here, hospital or extended-care stays were strongly associated with completion of metabolic monitoring. Because dementia diagnoses are rare for persons younger than 60, only patients aged ≥60 years were included for both groups. In addition, given that elderly patients (≥65 years) may use Medicare-covered services, which may not be detected in VA databases, only ongoing VA service users with at least two VA outpatient visits on two different days in the previous 360 days (including the index date), were included. [A flow diagram of patient selection is available in an online data supplement to this article.]

Assessment of monitoring

Observed metabolic monitoring rates vary depending on the definition of monitoring periods (21). On the basis of published recommendations (12), we examined baseline monitoring of weight, plasma glucose or hemoglobin A1c (HBA1c), and serum low-density lipoprotein cholesterol (LDL) within the 30 days before or after the index medication date. Similarly, three-month follow-up monitoring was operationalized as monitoring these parameters between 60 and 120 days after the index date (14). Because VA data often do not indicate whether the patient was fasting, any glucose or LDL test was included. Although the consensus statements specifically recommend fasting laboratory tests, most studies of routine monitoring practices have examined whether any glucose or lipid tests were obtained (22).

Antipsychotic side-effect risk classification

The risk of the index medication to cause metabolic side effects was classified as high (clozapine and olanzapine), medium (quetiapine, risperidone, chlorpromazine, thioridazine, loxapine, perphenazine, paliperidone, thiothixene, and trifluoperazine), or low (aripiprazole, ziprasidone, haloperidol, fluphenazine, molindone, pimozide, and mesoridazine) on the basis of the consensus ADA-APA statement (12), comprehensive review articles (23,24), and CATIE study results (25).

Statistical analysis

For each metabolic parameter (weight, glucose or HBA1c, and LDL), we compared monitoring frequency between the dementia and psychosis groups with a chi square test for equal proportions. We did this for baseline and follow-up monitoring separately. Changes in the monitoring rate from baseline to follow-up within each diagnostic group were compared using McNemar’s tests.

We compared patient characteristics on the day of the index antipsychotic prescription or in the 180 days before (age; gender; race-ethnicity; marital status; and preexisting medical comorbidities, such as diabetes, dyslipidemia, obesity, hypertension, and heart disease) and the metabolic side-effect risk of the index medication (high, medium, or low) between the two diagnostic groups using a t test for the continuous variable (age) and chi square tests of independence for categorical variables. Because dementia patients differed from psychosis patients in all observed patient characteristics listed above, especially age, a propensity score analysis was conducted. Propensity score analysis facilitates the comparison of two groups of patients with different characteristics (26,27). Propensity scores were estimated by using logistic regression to predict the diagnostic group (dementia versus psychosis) based on the aforementioned patient characteristics. In addition to the variables for patient characteristics listed above, the model included all statistically significant interaction terms and quadratic and cubic terms of age identified following the model development process described in Austin and Mamdani (28). Psychosis patients were matched to dementia patients by using a greedy matching algorithm (29) with a caliper of .1 standard deviation of the logit of the estimated propensity score to enhance comparability of the matched patients. Because age was the most significant difference between the two groups, we further required age differences between the matched pairs to be no more than one year. After matching, differences in observed patient characteristics were tested by using a paired t test for the continuous variable and McNemar’s test for binary variables to account for the matched nature of the sample. Standardized differences were also calculated as recommended, and any differences of 10 percentage points or larger were marked as “large” (28). The matching used an iterative process until there were no statistically significant differences in patient characteristics between diagnostic groups and all standardized differences were less than 10 percentage points. After matching, monitoring rates between diagnostic groups were compared at each time point and across time points with McNemar’s tests. Conditional logistic regressions for matched data were also conducted for each metabolic parameter at baseline and at follow-up with the diagnostic group as the only predictor to generate the odds ratios.

All analyses were conducted with SAS, version 9.1, and p values less than .05 were considered statistically significant.

Results

The final sample included 1,576 matched pairs. Table 1 summarizes data on patient characteristics before and after matching. Before matching, the differences between the two diagnostic groups were statistically significant for all characteristics, and all but six standardized differences were >10 percentage points. The most significant difference, clinically and statistically, was age, with dementia patients being much older than psychosis patients (79.1±8.4 versus 65.9±6.7, p<.001). After matching, the two groups did not differ significantly on any of the patient characteristics, and the largest standardized differences were <4 percentage points.

Table 2 presents data for the two groups on each of the three metabolic parameters at baseline and at follow-up. The results determined after matching were our primary results; we include before-matching results for completeness. After matching, all between-group differences in monitoring rates (except weight at baseline) were reduced from the comparisons made before matching. However, differences between the two groups in follow-up monitoring for all parameters remained statistically significant, even after matching patient baseline characteristics between diagnostic groups For both diagnostic groups, monitoring was greater at baseline than at follow-up for each metabolic parameter: weight, 68% at baseline and 46% at follow-up in the dementia group and 64% versus 51% in the psychosis group; glucose or HBA1c, 41% versus 26% in the dementia group and 44% versus 31% in the psychosis group; LDL, 24% versus 13% in the dementia group and 27% versus 18% in the psychosis group (p<.001 for each comparison).

Discussion

This work represents a systematic examination of metabolic monitoring of older outpatients with dementia who were prescribed off-label antipsychotics in a large VA sample. At baseline, metabolic monitoring of glucose or HBA1c and LDL for the dementia group was comparable to the propensity score–matched psychosis group, but weight was monitored significantly more often in the dementia group. At follow-up, monitoring of all three metabolic parameters was significantly lower in the dementia group compared with the matched psychosis group, although the absolute differences were small (<5%). Overall, monitoring rates for all metabolic parameters were low in both groups, both at baseline and follow-up, indicating suboptimal metabolic monitoring for patients in both dementia and psychosis groups on the basis of the consensus ADA-APA recommendations (20). Although suboptimal, the monitoring in our VA sample was higher than that reported previously in non-VA samples (15,30).

Suboptimal monitoring after prescription of antipsychotics, especially for psychotic disorders, has been documented in various settings (13), including the VA (14). However, because of the potential for serious side effects, including mortality and CVEs, we had expected that our sample of outpatients with dementia who were receiving off-label prescriptions for antipsychotics would be monitored for metabolic side effects with greater frequency compared with psychosis patients, for whom antipsychotics are FDA approved.

Why do patients with dementia not receive recommended monitoring? Although the answer is unclear, one possible reason is that we used the ADA-APA monitoring recommendations, which were developed to apply to any patient treated with a second-generation antipsychotic but which did not specifically consider recommended monitoring for patients with dementia. Other possible reasons include patient and provider factors that could not be measured in a medical records database. These factors may affect patients with dementia more than patients with psychosis. A variety of factors that affect metabolic monitoring have been addressed in the literature (31,32). Patient-related factors may include reluctance to travel to the clinic for frequent lab tests, transportation difficulties, and limited understanding about the need for frequent monitoring. In an exploratory analysis, we found that 95% of patients with dementia had at least one outpatient visit during the baseline period and 75% had one during the three-month follow-up period. This suggests that lack of outpatient contact did not explain lack of monitoring for most patients. Provider-related factors may include reliance on other providers to order lab tests and monitor results, an assumption that patients with dementia may not understand instructions to manage metabolic side effects if they are detected, and less emphasis on strictly managing side effects and focusing on providing comfort care. For example, weight gain from antipsychotics might be considered helpful because unexplained and untreatable weight loss is common among individuals with dementia (33). However, our sample included only patients with dementia who were being managed in the outpatient setting and were thus likely to have less advanced dementia than those in nursing homes. Conflicting findings in the literature about the association between hyperglycemia and use of antipsychotics may lead to decreased vigilance for hyperglycemia risk among patients with dementia (16,34). In addition, the ADA and the American Geriatrics Society guidelines recommend less stringent control for HBA1c (8% rather than 7%) for older adults with dementia and diabetes mellitus (35,36). The allowance in treatment guidelines for less stringent diabetes mellitus control may influence providers to monitor less for hyperglycemia (19) among older persons with dementia who receive antipsychotics, despite the fact that monitoring may be particularly important in this context (19), and hyperglycemia and ketoacidosis can develop rapidly (37) and cause other imbalances in blood chemistries in already frail elders. However, in the absence of specific guidelines for monitoring outpatients with dementia who receive antipsychotics, adopting the ADA-APA approved monitoring recommendations seems reasonable.

Although practice guidelines recommend that nonpharmacological treatment and psychosocial interventions should be the first line of treatment for neurobehavioral symptoms of dementia (38), antipsychotics are frequently prescribed despite their limited efficacy and substantial risks, particularly in circumstances in which the patient exhibits identifiable risk of harm to self or others or has significant distress and when nonpharmacological interventions have been unsuccessful (39). There is a paucity of evidence-based treatment alternatives to antipsychotics for this population, and currently no such alternative treatment has been approved by the FDA for these symptoms (40,41). Similarly, the data on efficacy of specific psychosocial treatments for patients with dementia are limited and inconclusive (42). Clinicians, patients, and caregivers are left with unclear choices of treatment for dementia patients with severe behavioral disturbance. It is important to note that although there is a risk of increased mortality with the use of antipsychotics among patients with dementia, the absolute increased risk, at least in the short term, is relatively small (approximately 1%−2%) (1,43).

The study had some limitations. The results are applicable only to patients with dementia who receive care as an outpatient in the VA and cannot be generalized to patients with dementia receiving antipsychotics in inpatient, residential care, or nursing home settings. Because of the large age differences between patients with dementia without psychosis and psychotic patients without dementia, a significant number of older patients with dementia, particularly those aged 75 or older, were not matched. Therefore, the monitoring rates estimated by using the matched sample may not generalize well to dementia patients over 75 years of age.

Although propensity score matching was able to balance all observed differences between the dementia and psychosis groups, there may have been unmeasured differences between the two groups that may have explained some differences in monitoring. For example, travel distance to the medical center or clinic or differences in living situations may account for some of the between-group differences in monitoring. Finally, in the administrative data available to us, we were not able to identify the provider type or the clinic that ordered the laboratory tests for monitoring. This information will be useful to appropriately target efforts to improve monitoring.

Conclusions

The most significant finding of our study is that similar to rates for patients with psychosis, the rates of monitoring for metabolic side-effects were low for outpatients with dementia who were prescribed antipsychotics. In the absence of alternative, safe, and efficacious treatments, the use of antipsychotics for managing neurobehavioral symptoms of dementia is likely to continue in some clinical situations where the risks of unmanaged neurobehavioral symptoms outweigh the risks of not using antipsychotics. However, it is concerning to find that such prescribing was not accompanied by greater monitoring for metabolic side effects, especially considering that our sample included only outpatients whose dementia was likely to be less severe than that of patients in residential programs and nursing homes. Although there are ongoing efforts in the VA to improve monitoring for metabolic side effects associated with antipsychotic prescribing, our findings suggest that more targeted efforts are needed to improve monitoring among patients with dementia for whom the off-label use of antipsychotics lacks efficacy and poses substantial risks.