Whether the development of tardive dyskinesia is associated with a shortened survival time has been open to question. A recent meta-analysis Reference Ballesteros, Gonzalez-Pinto and Bulbena1 of seven studies found that tardive dyskinesia is a weak risk factor for increased mortality, with an odds ratio (OR) in a fixed-effects model of 1.4 (95% CI 1.2–1.7, P<0.005). However, the authors described recurrent methodological problems, including small sample sizes, a follow-up time of less than 5 years in three studies and varying definitions of tardive dyskinesia. We further note that six studies lacked data on interrater reliability. Despite the advent of atypical antipsychotics, the continuing use of conventional antipsychotics in poorer countries Reference Janno, Holi and Tuisko2 and in some segments of more economically advantaged areas Reference Owen, Feng, Thrush, Hudson and Austen3 makes this question still important, particularly in view of the metabolic risks associated with atypical antipsychotics. Reference Allison, Mentore, Heo, Chandler, Cappelleri and Infante4–Reference Koro, Fedder, L'Italien, Weiss, Magder and Kreyenbuhl6
Method
The Minneapolis Veterans Administration Medical Center requires that each patient exposed to antipsychotics be evaluated at least yearly for tardive dyskinesia by raters repeatedly trained to reliability in using the Dyskinesia Identification System-Coldwater version (DIS-CO), Reference Sprague, Kalachnik, Breuning, Davis, Ullmann and Cullari7 a scale which is equivalent in sensitivity Reference Dean, Russell, Kuskowski, Caligiuri and Nugent8 to the Abnormal Involuntary Movement Scale (AIMS). Reference Guy9 Dyskinesia Identification System scores were reviewed by the attending physician who had to reach a conclusion regarding the presence or absence of tardive dyskinesia based on the Schooler–Kane criteria. Reference Schooler and Kane10 Depending on staffing, only two to three trained raters were used during this study, with κ coefficients ranging from 0.80 to 0.85. Demographic data, smoking history, Axis I and Axis III diagnoses, 11 names and doses of medications extant at the time of each rating, and previous DIS-CO scores were entered in a computerised database. A waiver of authorisation was obtained from the Minneapolis Veterans Administration Medical Center Human Studies Subcommittee to submit detailed identification data on 1626 patients to the National Death Index. 12 Observation times ranged from 3 to 5074 days, with a mean observation time of 1446 days (s.d.=1179.3). The National Death Index, after ensuring that this study met their standards with regard to confidentiality and appropriate use of death certificate data, supplied us with certificates noting the causes of death in 1200 people. Pilot data had indicated that 24 of 386 people had died over a 10-year period. A power analysis indicated that utilisation of the complete database would give an 87% likelihood of detecting differences in mortality at a moderate effect size.
Data analysis
With survival time as the censored, failure-time, dependent variable, we used the Cox proportional hazards regression, a multivariate extension of the Kaplan–Meier method, to estimate the relative risk of mortality associated with several predictor variables, including time since last examination, age, the diagnosis of tardive dyskinesia (ever or current), course of antipsychotic use (conventional antipsychotics only, atypical antipsychotics only or a mix), and the interaction of tardive dyskinesia diagnosis with drug course. For these analyses the observation period was from the date of the last examination to death or the censor period of 31 December 2001.
Results
The study population (Table 1) was predominantly male (95%), White (91%) and African–American (6.6%). Overall, 60% had received only conventional antipsychotics and 23% received both atypical antipsychotics and conventional antipsychotics or switched use. A third developed tardive dyskinesia at some point during follow-up. Of the 1208 individuals with definitive mortality data from the National Death Index, 205 (17.1%) had died during the observation period. The mean time to death was 1030.1 days (s.d.=976.9) from the initial observation v. 1551 days (s.d.=1202.9) for those alive during the same period. The mortality rate was 5.07 per 100 patient years.
Characteristics of the sample
| Characteristics | Sample (n=1208) |
|---|---|
| Men, % | 95.5 |
| Age at first observation, years: mean (s.d.) | 49.2 (14.0) |
| Age at last observation, years: mean (s.d.) | 53.7 (13.6) |
| Ethnicity, % | |
| White | 91.0 |
| African–American | 6.6 |
| Latino | 1.0 |
| Asian | 0.3 |
| Native American | 0.1 |
| Drug course, % | |
| Conventional antipsychotic | 60.8 |
| Atypical antipsychotic | 16.1 |
| Dual or switched use | 23.0 |
| Tardive dyskinesia | |
| Any, % | 33.4 |
| DIS-CO score, mean (s.d.) | 3.84 (3.63) |
| Maximum DIS-CO score, mean (s.d.) | 5.29 (4.66) |
Individual factors associated with mortality
Risk ratios and associated statistics for individual predictors are shown in Table 2. On the initial analysis, without covariates, the presence of tardive dyskinesia, either current or remitted, was significantly associated with an increase in mortality, with a risk ratio (RR) of 1.57 (95% CI 1.23–1.99, P<0.001). A parallel analysis was conducted in which the maximum DIS-CO score was used, rather than the tardive dyskinesia diagnosis. We found that for every point increase in the DIS-CO score, there was a 5% increase in the likelihood of death (RR=1.05, 95% CI 1.03–1.07, P<0.001).
Unadjusted predictors
| Variable | Coefficient | s.e. | Wald | P | RR | 95% CI |
|---|---|---|---|---|---|---|
| Diagnosis of tardive dyskinesia (present or remitted) | 0.45 | 0.12 | 13.26 | <0.001 | 1.57 | 1.23-1.99 |
| DIS-CO score (maximum) | 0.05 | 0.01 | 20.72 | <0.001 | 1.05 | 1.03-1.07 |
| Use of conventional antipsychotics | 0.7 | 0.29 | 5.8 | 0.02 | 2 | 1.13-3.53 |
| Age, years | 138.1 | <0.001 | ||||
| <44 | 1 | |||||
| 45-52 | 0.53 | 0.28 | 3.63 | 0.06 | 1.7 | 0.99-2.93 |
| 53-65 | 1.4 | 0.18 | 64.41 | <0.001 | 4.55 | 2.84-7.30 |
| 66+ | 2.05 | 0.17 | 152.02 | <0.001 | 9.06 | 5.82-14.09 |
Age
We grouped the cohort into four categories: age <44 (the reference category), 44–52, 53–65, and 66 years and older. We found that age at the time of the last rating was significantly associated with mortality. The risk ratios were 1.70 (95% CI 0.99–2.93), 4.55 (95% CI 2.84–7.30) and 9.06 (95% CI 5.82–14.09) respectively.
Antipsychotic medication
For drug course, those who had taken conventional antipsychotics or conventional antipsychotics plus atypical antipsychotics were compared with those who had taken only atypical antipsychotics (the reference category). Those who were exposed only to conventional antipsychotics were over twice as likely to die during the observation period as those who had taken only atypical antipsychotics (RR=2.00, 95% CI=1.13–3.53, P<0.02).
Given that atypical antipsychotics had been less widely used during the early period of this study, we examined the possibility of an interaction between age and drug course, but this was not significant. In a further analysis of drug course and mortality across age strata, we noted that in the age group >44 years, the risk ratio for those taking conventional antipsychotics v. atypical antipsychotics was 2.2 (95% CI 0.3–17.4), for the group aged 44–52 years the RR=2.1 (95% CI 0.5-9.0), for those aged 53–65 the RR=2.9 (95% CI 0.71–12.1), and in those 66 years and older the RR=1.8 (95% CI 0.74–4.5).
Cox proportional hazards regression analysis
We entered the presence of tardive dyskinesia, drug course and age into a regression model (final regression model shown in Table 3). Age and drug course continued to be significantly associated with a shortened survival time, but tardive dyskinesia was no longer significant. A parallel analysis using the maximum DIS-CO score yielded similar results. An interaction term with tardive dyskinesia status and drug course was not significant.
Final Cox proportional hazards regression model
| Variable | Coefficient | s.e. | Wald | P | RR | 95% CI |
|---|---|---|---|---|---|---|
| Diagnosis of tardive dyskinesia | 0.11 | 0.134 | 0.657 | 0.418 | 1.12 | 0.86-1.45 |
| Use of conventional antipsychotics | 0.73 | 0.33 | 5 | 0.025 | 2.08 | 1.1-3.96 |
| Age, years | 131.58 | <0.001 | ||||
| <44 | 1 | |||||
| 45-52 | 0.54 | 0.28 | 3.75 | 0.053 | 1.71 | 1.00-2.95 |
| 53-65 | 1.49 | 0.24 | 38.1 | <0.001 | 4.44 | 2.77-7.13 |
| 66+ | 2.17 | 0.23 | 91.69 | <0.001 | 8.77 | 5.63-13.68 |
Tardive dyskinesia and causes of death
Using a broad classification of causes of death, we found that 23.9% had died of cardiac disease/myocardial infarction, 20.5% from cancer and 8.3% from stroke/cardiovascular disease. Death by suicide was found in 6.8% and 4.9% had died in accidents or from external causes. There were no significant differences in causes of death based on tardive dyskinesia status. Mortality rates per 100 000 per year were calculated for each category of disease and are reported in Table 4.
Causes of death and mortality rates per 100 000
| Antipsychotic | ||||||||
|---|---|---|---|---|---|---|---|---|
| Overall | Atypical only | Conventional only | Atypical and conventional | |||||
| Cause of death | n | Rate per 100 000 per year | n | Rate per 100 000 per year | n | Rate per 100 000 per year | n | Rate per 100 000 per year |
| Infection | 4 | 84.17 | 0 | 3 | 80.32 | 1 | 158.37 | |
| Cancer | 42 | 883.84 | 0 | 40 | 1070.96 | 2 | 316.73 | |
| Diabetes | 9 | 189.39 | 0 | 9 | 240.97 | 0 | ||
| Heart disease/myocardial infarction | 49 | 1031.15 | 2 | 518.76 | 44 | 1178.06 | 3 | 475.11 |
| Stroke/cerebrovascular disease | 17 | 357.74 | 0 | 15 | 401.61 | 2 | 316.73 | |
| Influenza/pneumonia | 8 | 168.35 | 0 | 8 | 214.19 | 0 | ||
| Chronic obstructive pulmonary disease | 12 | 252.53 | 0 | 8 | 214.19 | 4 | 633.48 | |
| Liver disease | 2 | 42.09 | 1 | 259.38 | 1 | 26.77 | 0 | |
| Accidents/external causes | 10 | 210.44 | 0 | 9 | 240.97 | 1 | 158.37 | |
| Suicide | 14 | 294.61 | 0 | 11 | 294.51 | 3 | 475.11 | |
| Other disease | 38 | 799.66 | 2 | 518.76 | 32 | 856.77 | 4 | 633.48 |
| All causes | 205 | 4313.97 | 5 | 1296.91 | 180 | 4819.32 | 20 | 3167.4 |
Discussion
Despite the advent of atypical antipsychotics, the debate over the risk:benefit ratio of antipsychotics has become increasingly complex. Reference Abidi and Bhaskara13 This is due to the well-known neurological side-effects of conventional antipsychotics and the growing concerns over the metabolic side-effects of atypical antipsychotics, Reference McIntyre, McCann and Kennedy5,Reference Koro, Fedder, L'Italien, Weiss, Magder and Kreyenbuhl6 issues surrounding cost-effectiveness, Reference Rosenheck, Perlick, Bingham, Liu-Mares, Collins and Warren14 and evidence indicating that atypical antipsychotics have little advantage over reasonably dosed conventional antipsychotics in drop-out rates as a result of adverse events, Reference Lieberman, Stroup and McEvoy15 quality of life, Reference Rosenheck, Perlick, Bingham, Liu-Mares, Collins and Warren14 efficacy, and rates of parkinsonism and tardive dyskinesia. Reference Rosenheck, Perlick, Bingham, Liu-Mares, Collins and Warren14,Reference Lieberman, Stroup and McEvoy15
The range of problems associated with both classes of drugs leaves even well-informed clinicians, patients and families still struggling with the risk: benefit ratio, and indeed facing an even greater degree of complexity than in the era preceding the widespread use of atypical antipsychotics. Adding to this debate has been the question of shortened survival time in those who develop tardive dyskinesia, but previous studies have been inconclusive. Reference Ballesteros, Gonzalez-Pinto and Bulbena1 In this study, the initial increase in mortality found in those with tardive dyskinesia was overshadowed by the exposure to conventional antipsychotics only, which resulted in a significantly increased mortality rate compared with those taking only atypical antipsychotics. Not surprisingly, older age was a significant predictor of increased mortality, but there was no interaction between age and drug course.
Our data are supportive of a retrospective cohort study Reference Wang, Schneeweiss, Avorn, Fischer, Mogun and Solomon16 of first-time users of antipsychotics (n=22 890), aged 65 and older, in which those exposed to conventional antipsychotics had a 37% increase in mortality compared with those using atypical antipsychotics (P=0.001). Perhaps most strikingly, the time to follow-up was 180 days or less, with the risk of death being significant at <40 days and at higher doses. Cancer, HIV infection and congestive heart failure were the only conditions conferring a greater adjusted risk than did conventional antipsychotics; however, the presence of antipsychotic-induced movement disorders was not described. The presence or absence of dementia was not associated with the increase in mortality.
Whether the use of conventional antipsychotics is a consistent predictor of shortened survival time is doubtful, since the use of either conventional antipsychotics or atypical antipsychotics in people with schizophrenia was associated with a significant, fourfold increase in mortality. Reference Enger, Weatherby, Reynolds, Glasser and Walker17 In a cohort study of three US Medicaid programmes, Reference Hennessy, Bilker and Knauss18 the adjusted rate ratios for mortality, cardiac arrest and ventricular arrhythmias were higher for risperidone than haloperidol. However, the highest rate occurred in those receiving the lowest dose of risperidone, suggesting that this group may have been at higher risk and thus treated conservatively. Nevertheless, Saha et al Reference Saha, Chant and McGrath19 in their systematic review of mortality in schizophrenia have emphasised the possibility that the increasing use of atypical antipsychotics and the development of the metabolic syndrome may be in part responsible for a striking increase in the median standardised mortality ratio (SMR) from 1.84 in the 1970s to 3.20 in the 1990s, despite the fact that in most countries age-standardised SMRs have fallen. Unfortunately, the authors found only three studies examining SMRs from lower income countries, but, given this cautionary note, they found no difference in SMRs in schizophrenia across sites.
Complicating factors
Since the population in this study is composed primarily of individuals with schizophrenia, we note that an excess mortality rate in this disorder was described as early as the 19th century, Reference Brown20 long before the introduction of biologically oriented therapies. This has continued, with the most recent analysis Reference Saha, Chant and McGrath19 finding a median SMR of 2.58. Excessive deaths from both natural and unnatural causes have been widely recognised, with suicide being a prominent factor; Reference Saha, Chant and McGrath19,Reference Allebeck and Wisted21 however, others Reference Laursen, Munk-Olsen, Nordentoft and Mortensen22 have found that death from unnatural causes was lower in schizophrenia compared with unipolar, bipolar and schizoaffective disorders but only in those aged 40–79 years. None the less, these authors found excessive mortality in all of the four diagnostic groups. In a meta-analytic study, Reference Brown20 12% of all deaths and 28% of excess deaths were as a result of suicide, although deaths from unnatural causes were also prominent. Similarly, a study Reference Mortensen and Juel23 of first-admission patients admitted between 1970 and 1986 noted that of the 1100 people who died, half died by suicide.
These data are in marked contrast to our findings, wherein deaths by suicide and external causes were 6.8% and 4.9% respectively. However, the suicide rate in our study is 294.6 per 100 000 per year, similar to that described in other studies of schizophrenia. We also found substantial increases in mortality with age, in contrast to Brown, Reference Brown20 who noted a decline in mortality in schizophrenia with increasing age. Not surprisingly, others Reference Mortensen and Juel23,Reference Ösby, Correia, Ekbom, Brandt and Sparén24 have found an increase, especially in deaths from cardiovascular disease.
The assessment of mortality in schizophrenia is also complicated by increased mortality rates in other psychiatric disorders (substance misuse, ‘organic’ mental disorders, eating disorders, major depression and others), Reference Laursen, Munk-Olsen, Nordentoft and Mortensen22,Reference Harris and Barraclough25 many of which are comorbid with schizophrenia. We could not accurately assess these, since we did not use a structured diagnostic interview.
Specific factors in antipsychotic-associated mortality
In recent years there has been considerable interest in the effects of antipsychotics on cardiac status. One focus has been on rates of arrhythmia and arrest, Reference Hennessy, Bilker and Knauss18 with rate ratios of 1.3–3.2 in people with schizophrenia taking clozapine, haloperidol, risperidone or thioridazine. The rate ratios for death ranged from 2.6–5.8. However, the authors noted the difficulty in separating out treatment effects from the effects of schizophrenia. In another study, Reference Enger, Weatherby, Reynolds, Glasser and Walker17 those using conventional antipsychotics had a five times greater risk of myocardial infarction than controls, but the risk was lower with greater intensity of antipsychotic use. Cardiomyopathy has been reported with fluphenazine, risperidone and lithium, and myocarditis with clozapine, lithium and chlorpromazine. Reference Coulter, Bate, Meyboom, Lindquist and Edwards26
Several other risk factors are worth noting. In a small study, Reference Waddington, Youseff and Kinsella27 antipsychotic polypharmacy was a risk factor for shortened survival time, as was being edentulous, an older male and not being treated with anticholinergics. A recent study Reference Hippisley-Cox, Vinogradova, Coupland and Parker28 found a 308% increase in the risk of colon cancer in people with schizophrenia taking antipsychotics, even when adjusted for smoking, obesity, comorbidity, concurrent medications and socio-economic status.
Finally, antipsychotics have been shown to induce cellular toxicity Reference Dean29 via a number of mechanisms, including changes in proteins affecting cell survival, impairment of the mitochondrial respiratory chain, increases in DNA fragmentation and other changes in cell morphology. However, many studies have been short-term, and have often neglected important confounders such as polypharmacy and brain changes associated with comorbid disorders. It appears that we are badly in need of long-term studies of antipsychotic-induced neuronal changes and their influence on mortality. It does not appear, however, that the development of tardive dyskinesia carries an increased risk of mortality. Our study has several advantages over previous work, including a large population followed intensively over a decade by raters trained to reliability. This is the only study of mortality rates in tardive dyskinesia to utilise the National Death Index, which lent certainty to the fundamental question of whether a given individual had died during the observation period, in addition to giving us death certificate data on a large clinic population rated repeatedly for the presence of tardive dyskinesia. However, the demographics of the population may limit its applicability.
It appears that more attention needs to be given to discussions focusing on the risk:benefit ratio of antipsychotics. In years past, obtaining informed consent for the use of these agents has focused on the risks of incurring tardive dyskinesia and parkinsonism, but in recent years it has expanded to include the metabolic syndrome, with the associated possibility of a shortened lifespan. Is it now time to emphasise this possibility to an even greater degree, given the range of morbidity and mortality associated with these drugs?
Acknowledgements
The authors would like to thank Ms Jean M. Russell for her work in data collection, rater training and in rating examinations performed in the Tardive Dyskinesia Assessment Clinic. We also thank Robert Bilgrad, of the Division of Vital Statistics, National Center for Health Statistics, for his help in completing the National Death Index application and making the data accessible and understandable.
eLetters
No eLetters have been published for this article.