Search.

We searched Pubmed through October 9, 2012 for epidemiologic studies reporting the risk of mortality, stroke, ventricular arrhythmias, sudden cardiac death, venous thromboembolism, myocardial infarction, pneumonia, and hip fracture in older antipsychotic users. These events were chosen apriori through an initial review of the literature. Our final search strategy included free-text and controlled vocabulary terms (e.g. Medical Subject Headings) for these topics, their synonyms, abbreviations, and alternate spellings (see File S1).

Inclusion and exclusion criteria.

For this review, our inclusion criteria applied to studies evaluating the risk for medical events in FGA and SGA users. Our search strategy included finding comparative randomized trials, but we did not find any that reported on the medical events of interest as an outcome (listed in criterion #1). Thus, we formulated criteria to reduce the potential for bias in observational studies' estimates of the relationship between antipsychotic type (FGA versus SGA) and mortality or medical events, and included observational studies with the following characteristics: (1) evaluated antipsychotic users' risk of mortality or the following medical events: stroke, ventricular arrhythmia or sudden cardiac death, venous thromboembolism, pneumonia, or hip fracture (2) directly compared FGAs to SGAs or compared both to a non-user reference group; (3) the mean age of the study population was 65 or greater or age-stratified results—absolute rates or relative risks—were provided for adults over age 65; (4) the study sample was restricted to “new users” of antipsychotic medications or required a washout-period of no use prior to cohort entry; (5) adjusted for potential confounders that were assessed prior to antipsychotic initiation; (6) and did not require a minimum period of survival after antipsychotic initiation for inclusion in the analysis. We required new-user designs because prior studies demonstrate that the mortality hazard is highest immediately after antipsychotic initiation and decreases thereafter [29]. For the same reason, we excluded studies that required a minimum period of survival after antipsychotic initiation for cohort entry. We only considered studies where covariates were assessed prior to antipsychotic initiation because adjusting for subsequent changes in health does not control for confounding and may increase bias [30]. The rationale for these criteria is discussed further in the appendix (see File S3).

Selection process.

We reviewed the titles and abstracts of articles retrieved from Pubmed and selected primary research articles that met criterion 1. From those articles that then met criterion 2 and were published in English, we extracted information about their study population, design, outcome assessment and occurrence, and adjusted results. To locate additional articles for data extraction, Science Citation Index was used to “hand search” the titles and abstracts of references cited by these articles (and those cited by relevant review articles identified earlier during the selection process). Multiple studies reporting results from the same administrative records were included when differences in study design had potential for describing mortality or medical event risk in clinically relevant subgroups (e.g. nursing home and dementia), populations (e.g. veterans), for use over different calendar periods or length of follow-up, or when the analyses adjusted for different sets of potential confounders. The studies that met criteria 3 through 6 were included in our review (see Figure 1 for flowchart). A list of the excluded articles is provided for each medical event in the appendix (see File S7). The article selection process and data extraction were carried out by JWJ.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flowchart for search execution.

https://doi.org/10.1371/journal.pone.0105376.g001

Epidemiology.

Of the 20 included studies, 12 investigated mortality among antipsychotic users. Of these, 11 reported higher all-cause or non-cancer mortality for FGAs than SGAs, appearing as early as 30 days after antipsychotic initiation and lasting at least six months. In retrospective cohort studies of older community-dwelling adults, the hazard ratio was approximately 1.55 in the first 30 days after antipsychotic initiation, and ranged from 1.27 to 1.37 at six months [29], [34], [35] (Table 2). The relationship was more pronounced for doses above the median (HR = 1.67 to 1.73; median doses were unreported) and was attenuated for doses below the median (HR = 1.14 to 1.23). The corresponding absolute differences in six-month mortality varied from 7.3% (a statewide study in the mid-Atlantic U.S. [34]) to 2.5% (two province-wide studies in Canada [29], [35]), and a nationwide study of Australian Veterans Affairs beneficiaries reported a 10.6% difference in one year mortality [36]. While population differences in SGA-related mortality may explain these heterogeneous risk differences and relative risks [37], these studies consistently demonstrated a dose-dependent increase in mortality for FGAs early after initiation, lasting for at least six months in older community-dwelling adults.

Some analyses were restricted to dementia patients in the U.S. [34] or Canada [29], [35] and reported similar elevations in six-month mortality for FGAs compared to SGAs (HR = 1.23 to 1.29). One study of U.S. nursing home residents found elevations with FGAs only among patients with mostly mixed and vascular dementia [38], suggesting that the effect may be limited to a group of patients at elevated risk for cerebrovascular and cardiovascular disorders. Studies of U.S. veterans found higher mortality for haloperidol at 30 days [39] and 180 days [40], as compared to specific SGAs, while another study of U.S. veterans found no difference in mortality at one year follow up [41]. These results suggest that the excess mortality among dementia patients may be limited to the first six months after antipsychotic initiation.

Claims-based studies of antipsychotics have been criticized for their inability to adjust for important determinants of antipsychotic use that predict mortality in older adults—namely dementia severity, hallucinations, and delirium. Studies of U.S. nursing home residents also adjusted for measures of cognitive and functional decline as captured by federally mandated clinical assessments. Results from these studies [17], [38], [42], [43]—and others that further adjusted for facility-level characteristics [6], [17]—were similar to those conducted in community-based populations in terms of magnitude, timing, and dose-dependency [44]. A clinical study using validated measures of cognitive function and dementia symptoms found higher mortality with FGA use among nursing home residents, but not community-dwelling residents [45]. Psychoses, though, were found to strongly predict mortality in both samples. Confounding by delirium could be driven by the predominant use of haloperidol to treat agitation among intensive care unit patients with delirium [46], who have elevated six-month mortality rates of six-month mortality [47]. Hospital medication use is often not captured in claims data so some apparent new users may have in fact initiated haloperidol for delirium during an inpatient encounter. Because delirium is not well recorded in clinical or claims data, many of the studies reviewed here likely suffer from some degree of residual bias.

In sum, the evidence for elevated mortality in FGA users compared to SGA users is consistent across several national and clinical populations. Several strategies have been used to adjust for potential confounders and most of these results signal higher risk among FGA users, yet residual and unmeasured confounding cannot be completely excluded as alternate explanations. We now discuss the epidemiologic evidence for explaining this observed difference in mortality in terms of potential intermediate medical events.

Overview.

From 2004 to 2012, 16 studies investigated the risk of stroke in FGA and SGA users, but only six met our inclusion criteria. Their results support higher stroke risk with FGAs than SGAs, but only within the first months after antipsychotic initiation (Table 2; HR = 1.03 to 1.91). The average incidence rate per 100 persons over six months among SGA users was 2.5, and among FGA users it was calculated to be 3.5 after applying the average relative risk for stroke (RR = 1.4). Assuming that the excess mortality for those who experience stroke after antipsychotic use is 17.8%, stroke may explain up to 6.7% of the mortality difference between FGAs and SGAs (Figure 2). The proportion mediated would be 7.4% if we expected poor sensitivity (Sn = 0.5) and a lower excess mortality of 10.0%; it would be 18.9% for the same sensitivity but a higher excess mortality of 26.0%.

Epidemiology.

Six studies compared the risk for cerebrovascular events in populations with varying degrees of pre-existing cerebrovascular disease and duration of follow-up. Wang et al. followed low-income, community-dwelling older adults for 120 days after antipsychotic initiation [13] and reported a slightly higher risk of cerebrovascular events for FGAs than SGAs after 30 days (HR = 1.08; 95%CI 0.99 to 1.18) and through the end of follow-up (HR = 1.09; 95%CI 1.02 to 1.16). Finkel et al. followed Medicaid beneficiaries with dementia for three months and found higher risk for haloperidol compared to risperidone [10] (HR = 1.91; 95%CI 1.02 to 3.60). Another study followed older Medicaid beneficiaries for an average of three months and found higher risk of stroke with some FGAs as compared to SGAs; the risk was higher for phenothiazines (HR = 2.34; 95%CI 1.01 to 5.41) but not butyrophenones (HR = 1.44; 95%CI 0.55 to 3.76) [14]. Other studies followed patients for six months or longer and reported contrasting results. A study of nursing home residents found lower risk of stroke or transient ischemic attack for FGAs compared to SGAs (HR = 0.81; 95%CI 0.65 to 1.01) [18]. Gill et al. followed patients with dementia for an average of eight months and found no difference in risk [11] (HR = 0.99; 95%CI 0.79 to 1.23) even among high-risk subgroups defined by history of stroke or long-term care residence. A study based in Hong Kong followed dementia patients for an average of 2.4 years and also found no difference in risk [16] (HR = 0.93; 95%CI 0.52 to 1.67). These results may be explained by these studies' longer periods of follow-up, particularly if high-risk FGA patients experience stroke only during the first few months after antipsychotic initiation or if such patients were more likely to discontinue or switch their antipsychotic medication, which were both treated as censoring events in these analyses.

Biological plausibility.

While the mechanism of antipsychotic-induced stroke is unclear, a recently proposed model postulated that antipsychotics could trigger cerebrovascular events in older adults with pre-existing medical comorbidities and concomitant medication use [21]. This is consistent with results from a recent case-case-time-control study of antipsychotic use and stroke in older adults [48], where cases essentially served as their own controls. FGAs, particularly haloperidol, the most widely used FGA, were associated with acute movement disorders that began soon after initiation. These extrapyramidal side-effects, as well as sedation by low-potency FGAs such as chlorpromazine, increase immobility, which can lead to or might exacerbate pre-existing cases of venous thromboembolism. Low-potency FGAs can also induce orthostatic hypotension, although it can occur with SGAs as well. Both of these complications can cause stroke, which is a leading cause of death.

Mortality associated with stroke.

Based on the following data, we assumed that 20% of older adults who experience a stroke die within six months based on the following data: 80 to 90% of strokes are ischemic, and a study of patients hospitalized for ischemic stroke estimated their mortality as 7.4% within 30 days, 11.4% within 90 days, and 19.1% within one year. The corresponding proportions for hemorrhagic stroke were 18.8%, 24.6% and 31.8% [49]. Very little data on sudden death in ischemic stroke exists [50], but in the case of hemorrhagic stroke, 12% die before admission.

Overview.

Six studies investigated the risk for sudden cardiac death in FGA and SGA users. Of these, two met our inclusion criteria. These studies suggest that the excess risk for ventricular arrhythmias for FGAs compared to SGAs is limited to older, sicker populations during the first weeks of follow-up (Table 2; HR = 1.20). Similarly, FGAs and SGAs appear to increase the risk for ventricular arrhythmia in younger, healthier community dwelling patients followed for longer periods. The average incidence rate per 100 persons over six months among SGA users was 0.24, and among FGA users it was calculated to be 0.26 after applying the average relative risk for ventricular arrhythmia (RR = 1.1). Assuming that the excess mortality for those who experience ventricular arrhythmia after antipsychotic use is 87.8%, ventricular arrhythmia may explain up to 0.9% of the mortality difference between FGAs and SGAs (Figure 2). The proportion mediated would be 3.9% if we expected very poor sensitivity (Sn = 0.2) and a lower excess mortality of 81%; it would be 4.8% for the same sensitivity but a much higher excess mortality of 99.6%.

Epidemiology.

The risk of out-of-hospital sudden cardiac death was investigated by Ray et al. in a retrospective cohort of non-psychotic, low-income persons with no history of prior cardiovascular disease [15]. This younger cohort (mean age 45.7) was followed for an average of two years and the results showed similar dose-dependent elevations in risk for current use of both FGAs (HR = 1.74; 95%CI 1.14 to 2.67) and SGAs (HR = 1.86; 95%CI 1.35 to 2.57) compared to non-use. Wang et al. investigated the risk of hospitalization for ventricular arrhythmia in a cohort of older Medicare beneficiaries and found higher risk for FGAs than SGAs but only within the first 30 days of follow up [13] (HR = 1.20; 95%CI 1.03 to 1.39). By 120 days of follow up, the effect had decreased and was no longer statistically significant (HR = 1.06; 95%CI 0.96 to 1.17).

Biological plausibility.

FGAs such as thioridazine, pimozide, haloperidol, and droperidol are known for their dose-dependent risk of QT prolongation through blockage of potassium ion channels in cardiac tissue; ziprasidone is the only SGA associated with QT prolongation. In rare cases, QT prolongation above 450 msec increases the likelihood for a polymorphic ventricular arrhythmia known as Torsades de Pointes (TdP) [51], which commonly leads to ventricular arrhythmia and sudden cardiac death. The risk is higher among older patients with pre-existing cardiovascular disease [52].

Mortality associated with ventricular arrhythmia.

Based on the following data, we assumed that 90% of older adults who experience a ventricular arrhythmia die within six months based on the following data: 80% percent of cardiac arrest cases occur in the community, where chances for survival are extremely low; one study estimated an overall survival of 6% [53]; for inpatients, the survival ranges from 5 to 35% [54].

Overview.

From 2005 to 2012, seven studies investigated the risk of venous thromboembolism (VTE) or pulmonary embolism (PE) in FGAs and SGAs. Of these, one met our inclusion criteria and reported higher VTE/PE risk for SGAs in the first six months after antipsychotic initiation (Table 2; HR = 0.5). The average incidence rate per 100 persons over six months among SGA users was 0.62, and among FGA users it was calculated to be 0.31, given this reported relative risk. Assuming that the excess mortality for those who experience VTE/PE after antipsychotic use is 13.8%, then the amount of the mortality difference between FGAs and SGAs explained by VTE/PE may be as high as −2.2%. Thus, VTE/PE does not appear to explain any of the higher mortality among FGAs as compared to SGAs (Figure 2).

Epidemiology.

A retrospective cohort study of older nursing home residents assessed VTE risk over six months of follow-up after antipsychotic initiation [12]. Compared to non-users, SGAs showed nearly double the VTE risk in non-users (HR = 2.01; 95%CI 1.50 to 2.70) while FGAs showed no difference (HR = 1.02; 95%CI 0.67 to 1.55), even among low- and high-risk subgroups. While residual confounding cannot be ruled out as a possible explanation, these subgroup analyses suggest that strong confounding by disease severity or indication are unlikely.

Biological plausibility.

The mechanism for antipsychotic-induced VTE—in particular why the risk may differ between FGAs and SGAs—is not known, but it may involve sedation and immobility in low-potency FGA users, increased weight gain and its sequale in SGA users, or other risk factors such as altered anticardiolipin antibody levels, fibrinolytic activity, or platelet aggregation [55]. VTE often begins as a deep vein thrombosis (DVT), which can be lethal when it leads to PE. Among older adults, PE occurs in a sizable number of VTE cases in both hospital and community populations [56].

Mortality associated with venous thromboembolism.

Based on the following data, we assumed that 15% of older adults who experience a VTE die within six months based on the following data: a study of patients hospitalized for their first VTE estimated the case mortality as 3.6% by 30 days and 12.6% by one year among those without cancer [57]. Nearly one-quarter of those who develop PE present with sudden death and another quarter die within a year after discharge [58]. In the study reporting on VTE risk among antipsychotic users, 78% were diagnosed with DVT and 22% with PE.

Overview.

We found five articles published from 2006 to 2012 that investigated myocardial infarction (MI) risk for FGAs and SGAs, two of which met our inclusion criteria. These studies are consistent with higher risk for FGAs than SGAs (Table 2; HR = 1.16 to 1.23). The average incidence rate per 100 persons over six months among SGA users was 1.0, and among FGA users it was calculated to be 1.2 after applying the average relative risk for MI (RR = 1.2). Assuming that the excess mortality for those who experience MI after antipsychotic use is 42.8%, then the proportion of the mortality difference between FGAs and SGAs explained by MI may be as high as 3.5% (Figure 2). The proportion mediated would be 4.2% if we expected poor sensitivity (Sn = 0.5) and a much lower excess mortality of 26.0%; it would be 9.5% for the same sensitivity but a much higher excess mortality of 59.6%.

Epidemiology.

Two cohort studies have evaluated the risk of MI in both FGAs and SGAs. The Wang et al. study described earlier found a higher risk for FGA users at 120 days after antipsychotic initiation (HR = 1.16; 95%CI 0.91 to 1.48), but no difference in risk at 30 or 60 days [13]. The Huybrechts et al. study of nursing home residents also assessed the risk of hospitalization for MI at six months after initiation and found higher risk for FGAs compared to SGAs (HR = 1.23; 95%CI 0.82 to 1.82) [18]. This was the only study that reported an absolute rate of MI [18], which was low (2.0 cases per 100 person-years), so there may have been insufficient power to provide a more precise estimate.

Biological plausibility.

The mechanisms by which antipsychotic use could lead to MI may involve effects on blood lipid levels, which may exacerbate atherosclerosis, or effects on platelets and other clotting factors leading to increased coagulability. MI is a leading cause of death.

Mortality associated with myocardial infarction.

Based on the following data, we assumed that 45% of older adults who experience a myocardial infarction die within six months: in a study of older Medicare patients admitted for MI, 26.3% died within 30 days after discharge and an additional 20% died within the remaining year [5].

Overview.

From 2004 to 2012, 12 studies investigated the association between falls or fracture risk in both FGA and SGA users, two of which met our inclusion criteria. These studies suggested that FGAs carry higher risk for hip fracture in nursing home residents soon after antipsychotic initiation (Table 2; HR = 1.27 to 1.61). Because falls and fractures occur more frequently among nursing home patients, this estimate may not apply to the general population. Nevertheless, the average incidence rate per 100 persons over six months among SGA users was 2.38, and among FGA users it was 3.33 after applying the average relative risk for hip fracture (RR = 1.4) (we averaged as-treated and intent-to-treat HR estimates reported in the same study; see File S2). Assuming that after antipsychotic use, the excess mortality for those who experience hip fracture is 17.8%, then hip fracture may explain up to 6.5% of the mortality difference between FGAs and SGAs (Figure 2). The proportion mediated would be 1.3% if we expected poor sensitivity (Sn = 0.5) and a much lower excess mortality of 3.1%; it would be 9.2% for the same sensitivity but a much higher excess mortality of 22.6%.

Epidemiology.

Huybrechts et al. followed newly admitted nursing home residents in British Columbia for six months and found higher risk for FGAs than SGAs [17] (HR = 1.61; 95%CI 1.03 to 2.51), following patients until they experienced a hip fracture, died, discontinued their antipsychotic or started using an antipsychotic from the comparison group. When analyzed using the intention-to-treat approach, the hazard ratio was considerably lower (HR = 1.16; 95%CI 0.82 to 1.63). Another study followed nursing home residents without major psychotic disorders or cancer for six months and also found higher risk for FGAs (HR = 1.27; 95%CI 0.94 to 1.72) [18].

Biological plausibility.

Acute extrapyramidal symptoms occur more frequently with FGAs [59]. Dystonia, parkinsonism, dyskinesia, and akathisia can manifest as early as the first few days after initiation or as late as three months. Such symptoms can cause gait disturbances and impair mobility and balance, which are risk factors for falls (and thus fractures) in older adults [60]. Fractures increase the risk for mortality in older adults.

Mortality associated with hip fracture.

Based on the following data, we assumed that 20% of older adults who experience a hip fracture die within six months based on the following data: hip fracture increases mortality and 70% of the deaths occurring in the year following fracture happen within the first six months after injury [61]; at six months, the mortality among hip fracture cases ranges from 7.1% to 23%. Some of this mortality may be attributable to underlying comorbidity that increases risk for both hip fracture and mortality.

Overview.

From 2004 to 2012, seven studies investigated the risk for pneumonia in both FGAs and SGAs. The three that met our inclusion criteria reported no differences in risk between FGAs and SGAs within the first six months after antipsychotic initiation. One study (Wang et al. 2007 [62]) reported lower risk in FGA users but only after 120 days follow-up, which could indicate that SGA users experience higher risk after three or four months following antipsychotic initiation but the average relative risk was null over follow-up. Taken together, these results suggest that pneumonia does not explain any of the higher mortality in FGAs as compared to SGAs (Figure 2).

Epidemiology.

Three retrospective cohort studies evaluated the risk of pneumonia in FGA and SGA users. The Wang et al. study [62] of low-income older, frail, community-dwelling adults found no difference in risk at 30 (HR = 1.11; 95%CI 0.76 to 1.63) or 60 days (HR = 1.03; 95%CI 0.76 to 1.38). By 120 days the pneumonia risk was lower for FGAs as compared to SGAs (HR = 0.84; 95%CI 0.66 to 1.05). A study of newly admitted nursing home residents in British Columbia [17] showed no difference in risk at six months after antipsychotic initiation (HR = 1.03; 95%CI 0.62 to 1.69). Another study followed U.S. nursing home residents without cancer, schizophrenia, or bipolar disorder for up to six months and also found no difference in risk [18] (HR = 1.28; 95%CI 0.87 to 1.88).

Biological plausibility.

The most plausible mechanism for possible differences in pneumonia risk for FGAs and SGAs is the higher frequency of extrapyramidal symptoms and sedation among FGA users. Extrapyramidal symptoms involving the pharyngeal musculature could lead to dysphagia, and excess sedation could lead to decreased cough-reflex, both of which are risk factors for aspiration pneumonia in older adults [63]. Another potential mechanism could involve altered cytokine profiles and immune response, but the supporting evidence for this pathway is incomplete as these studies have not comprehensively investigated how specific antipsychotics, FGAs or SGAs, influence inflammatory activity across in vivo, ex vivo, and in vitro systems [64].

Mortality associated with pneumonia.

Based on the following data, we assumed that 20% of older adults who experience die within six months based on several lines of evidence. An observational study of nursing home residents with advanced dementia reported a 90-day mortality rate, after pneumonia diagnosis, ranging from 36% to 67% depending on the type of care received [65]; 51% of these cases had a do-not-hospitalize order. A review of community-acquired pneumonia reported a mortality rate of less than 1% for cases managed in outpatient settings [66]; among the 22 to 42% of patients in population-based studies with severe enough pneumonia to warrant inpatient or intensive care, mortality ranged from 14 to 54%. A large study of Medicare beneficiaries hospitalized for community-acquired pneumonia reported 11% mortality during hospitalization and 36% mortality over six-months among those discharged alive; at one year the total mortality was 41% [67]. Differences in pneumonia management and medical comorbidity, especially considering end-of-life decision making for older adults, may contribute to the wide variability in mortality among studies of inpatient pneumonia cases.