Conventional vs. Atypical Antipsychotic Medications
http://www.100md.com
《新英格兰医药杂志》
To the Editor: Wang et al. (Dec. 1 issue)1 report that, as compared with atypical antipsychotic medications, the use of conventional antipsychotic agents increased the death rate among elderly users. Although we appreciate the instrumental-variable analysis, we have two concerns. First, it seems as though confounding would still be possible if the instrumental variable (a physician's choice of a conventional antipsychotic medication as his or her most recent prescription) was independently associated with the outcome (risk of death of the index patient). This confounding could happen if physicians who choose conventional antipsychotic medications also tend to care for very sick patients or to be less aggressive than other physicians in prolonging those patients' lives. Second, in the instrumental-variable analysis, the authors report an increase of 7.3 percent in the absolute risk of death within six months with conventional antipsychotic medications, whereas for the primary analysis, they report only a 3.3 percent increase in this risk. Since the instrumental variable (the type of antipsychotic agent the physician most recently prescribed before the index prescription) is an imperfect measure of the exposure of interest (the type of antipsychotic agent the index patient received), why isn't the difference in risk in the instrumental-variable analysis biased toward 0 — that is, lower than 3.3 percent?
Sei J. Lee, M.D.
Veterans Affairs Medical Center
San Francisco, CA 94121
sei.lee@med.va.gov
Thomas B. Newman, M.D., M.P.H.
University of California at San Francisco
San Francisco, CA 94143
References
Wang PS, Schneeweiss S, Avorn J, et al. Risk of death in elderly users of conventional vs. atypical antipsychotic medications. N Engl J Med 2005;353:2335-2341.
To the Editor: The article by Wang et al. suggests that the use of conventional antipsychotic medications, as compared with the use of atypical agents, is associated with an increased risk of death. However, confounding according to indication might have biased the analysis, because in real-life conditions, conventional antipsychotic agents are more often prescribed than are atypical agents as add-on medications for severely ill patients.1 In addition, Table 1 of the article shows that atypical agents were prescribed more often than were conventional agents for persons with affective disorders, who received more prescriptions for other psychotropic medications that might have had a protective role.2
The authors adjusted for the presence of a number of physical conditions associated with death, but not for the severity of medical illness. In addition, we do not know whether exposure variables were equally distributed in terms of the number of conventional and atypical agents actually prescribed during the follow-up period, the concomitant use of two or more agents, and history of use of antipsychotic agents. Given that the baseline risk of death in this cohort of subjects was very high, we suspect that these limitations might have hampered the validity of the conclusions.
Corrado Barbui, M.D.
Andrea Cipriani, M.D.
Michele Tansella, M.D.
University of Verona
37134 Verona, Italy
corrado.barbui@univr.it
References
Mirandola M, Andretta M, Corbari L, Sorio A, Nose M, Barbui C. Prevalence, incidence and persistence of antipsychotic drug prescribing in the Italian general population: retrospective database analysis, 1999-2002. Pharmacoepidemiol Drug Saf (in press).
Cipriani A, Pretty H, Hawton K, Geddes JR. Lithium in the prevention of suicidal behavior and all-cause mortality in patients with mood disorders: a systematic review of randomized trials. Am J Psychiatry 2005;162:1805-1819.
To the Editor: Wang et al. attempted to estimate the risk of death from the use of conventional antipsychotic medications. Their adjustment methods reduced the estimate from 1.51 to 1.37, confirming confounding; this adjustment is probably insufficient. The authors were severely constrained in their ability to correct for confounding. Diagnoses obtained or inferred from claims data are variably reliable, and risk factors such as cigarette smoking and alcohol consumption are not included. Although users of conventional antipsychotic agents had poorer prognoses, users of atypical agents were more likely to have been in hospitals or nursing homes, giving more opportunities for diagnoses to be reported. The authors also did not account for differences in severity within diagnostic categories.
The ability to remove confounding can be tested by showing how much of the observed mortality within each treatment group can be explained. If the data cannot fully explain within-group mortality, it is likely that substantial confounding remains. Improved methods of estimating mortality could narrow, eliminate, or reverse differences in mortality.
Published randomized, controlled trials of conventional antipsychotic medications in patients with dementia have not included data on mortality; such information would be welcome.
Marc Stone, M.D.
Judith A. Racoosin, M.D., M.P.H.
Thomas Laughren, M.D.
Food and Drug Administration
Silver Spring, MD 20993-0002
To the Editor: Although we agree with the conclusions of Wang et al., we recommend using caution in the interpretation of the results of a retrospective cohort study. The method of looking at prescriptions written and filled can be misleading, because it can extract erroneous conclusions from static data obtained from a decision tree unknown to the investigators.
Although the authors cite concern regarding confounding variables, we have several additional questions that could alter the interpretation of the results. First, did the group receiving atypical antipsychotic agents have a built-in protective — and possibly a selection — bias, since they were more likely than users of conventional agents to be white and female and less likely to have certain life-threatening diseases (congestive heart failure, ischemic heart disease, or cancer)? Second, did the higher rate of use of antidepressants (P<0.001) in the group receiving atypical agents confer a protective effect with regard to the outcome? Third, did the fact that persons with neuropsychiatric disorders preferentially received more atypical agents (P<0.001) have a salutary effect on the outcome?
We would welcome input from the authors with regard to the elaboration and clarification of other factors that have been considered, or that could be considered, in a follow-up analysis of their original data.
Stefan P. Kruszewski, M.D.
732 Forest Rd.
Harrisburg, PA 17112
joeysdogma@comcast.net
Steven G. Klotz, M.D.
2461 Lititz Pike
Lancaster, PA 17601
The authors reply: Each letter writer justifiably suggests the possibility of residual confounding by factors such as severity of illness. The analyses presented did control for age, sex, and coexisting medical conditions. We also conducted additional analyses in which we controlled for a commonly used measure of the severity of coexisting illness1 and found no change in the 180-day mortality associated with the use of conventional as compared with atypical antipsychotic medications (adjusted hazard ratio, 1.37; 95 percent confidence interval, 1.26 to 1.48). The authors of two of the letters also inquire as to whether the use of other psychiatric drugs may have played a role. However, neither the use of antidepressants (adjusted hazard ratio, 0.94; 95 percent confidence interval, 0.87 to 1.09) nor the use of other psychotropic drugs (adjusted hazard ratio, 0.99; 95 percent confidence interval, 0.89 to 1.09) conferred significant protective effects, and we controlled for both variables in adjusted models.
Lee and Newman expected the instrumental-variable estimate to be biased toward the null. But because instrumental-variable methods correct for an omitted variable bias, the point estimate could move either up or down, depending on the nature of the unobserved confounding. The validity of instrumental-variable methods does not depend on a perfect concordance between the treatment and the instrument, although instruments that are more strongly related to treatment yield estimates with smaller standard errors.2,3 Although we cannot completely eliminate the possibility of a residual confounding by the other variables that the letter writers mention, a recent meta-analysis of randomized trials conducted among elderly persons with dementia showed that the conventional agent haloperidol increased the risk of death in the short term, as compared with placebo, by 107 percent.4 This risk was higher than that seen for atypical agents and remarkably close to the sum of the 60 to 70 percent increase in the risk associated with atypical agents, as compared with placebo, in the Food and Drug Administration's analyses plus the 37 percent increase in the risk with conventional agents, as compared with atypical agents, that we observed in our study.
In spite of the robustness of our findings to various analyses we conducted and corroboration from the limited data that have emerged from randomized trials, we agree with all the correspondents that observational findings such as these need to be interpreted cautiously. We continue to conclude conservatively only that conventional antipsychotic medications may not be safer than atypical agents in elderly populations. Finally, we completely concur with Stone and colleagues that additional data from randomized, controlled trials, particularly those involving elderly persons, are sorely needed.
Philip S. Wang, M.D., Dr.P.H.
Sebastian Schneeweiss, M.D., Sc.D.
M. Alan Brookhart, Ph.D.
Harvard Medical School
Boston, MA 02115
pwang@rics.bwh.harvard.edu
References
Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613-619.
Brookhart MA, Wang P, Solomon DH, Schneeweiss S. Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable. Epidemiology (Web, EpiFast-Track). (Accessed February 11, 2006, at http://www.epidem.com.)
Angrist JD, Imbens GW, Rubin DB. Identification of causal effects using instrumental variables. J Am Stat Assoc 1996;91:444-472.
Schneider LS, Dagerman KS, Insel P. Risk of death with atypical antipsychotic drug treatment for dementia: meta-analysis of randomized placebo-controlled trials. JAMA 2005;294:1934-1943.
Sei J. Lee, M.D.
Veterans Affairs Medical Center
San Francisco, CA 94121
sei.lee@med.va.gov
Thomas B. Newman, M.D., M.P.H.
University of California at San Francisco
San Francisco, CA 94143
References
Wang PS, Schneeweiss S, Avorn J, et al. Risk of death in elderly users of conventional vs. atypical antipsychotic medications. N Engl J Med 2005;353:2335-2341.
To the Editor: The article by Wang et al. suggests that the use of conventional antipsychotic medications, as compared with the use of atypical agents, is associated with an increased risk of death. However, confounding according to indication might have biased the analysis, because in real-life conditions, conventional antipsychotic agents are more often prescribed than are atypical agents as add-on medications for severely ill patients.1 In addition, Table 1 of the article shows that atypical agents were prescribed more often than were conventional agents for persons with affective disorders, who received more prescriptions for other psychotropic medications that might have had a protective role.2
The authors adjusted for the presence of a number of physical conditions associated with death, but not for the severity of medical illness. In addition, we do not know whether exposure variables were equally distributed in terms of the number of conventional and atypical agents actually prescribed during the follow-up period, the concomitant use of two or more agents, and history of use of antipsychotic agents. Given that the baseline risk of death in this cohort of subjects was very high, we suspect that these limitations might have hampered the validity of the conclusions.
Corrado Barbui, M.D.
Andrea Cipriani, M.D.
Michele Tansella, M.D.
University of Verona
37134 Verona, Italy
corrado.barbui@univr.it
References
Mirandola M, Andretta M, Corbari L, Sorio A, Nose M, Barbui C. Prevalence, incidence and persistence of antipsychotic drug prescribing in the Italian general population: retrospective database analysis, 1999-2002. Pharmacoepidemiol Drug Saf (in press).
Cipriani A, Pretty H, Hawton K, Geddes JR. Lithium in the prevention of suicidal behavior and all-cause mortality in patients with mood disorders: a systematic review of randomized trials. Am J Psychiatry 2005;162:1805-1819.
To the Editor: Wang et al. attempted to estimate the risk of death from the use of conventional antipsychotic medications. Their adjustment methods reduced the estimate from 1.51 to 1.37, confirming confounding; this adjustment is probably insufficient. The authors were severely constrained in their ability to correct for confounding. Diagnoses obtained or inferred from claims data are variably reliable, and risk factors such as cigarette smoking and alcohol consumption are not included. Although users of conventional antipsychotic agents had poorer prognoses, users of atypical agents were more likely to have been in hospitals or nursing homes, giving more opportunities for diagnoses to be reported. The authors also did not account for differences in severity within diagnostic categories.
The ability to remove confounding can be tested by showing how much of the observed mortality within each treatment group can be explained. If the data cannot fully explain within-group mortality, it is likely that substantial confounding remains. Improved methods of estimating mortality could narrow, eliminate, or reverse differences in mortality.
Published randomized, controlled trials of conventional antipsychotic medications in patients with dementia have not included data on mortality; such information would be welcome.
Marc Stone, M.D.
Judith A. Racoosin, M.D., M.P.H.
Thomas Laughren, M.D.
Food and Drug Administration
Silver Spring, MD 20993-0002
To the Editor: Although we agree with the conclusions of Wang et al., we recommend using caution in the interpretation of the results of a retrospective cohort study. The method of looking at prescriptions written and filled can be misleading, because it can extract erroneous conclusions from static data obtained from a decision tree unknown to the investigators.
Although the authors cite concern regarding confounding variables, we have several additional questions that could alter the interpretation of the results. First, did the group receiving atypical antipsychotic agents have a built-in protective — and possibly a selection — bias, since they were more likely than users of conventional agents to be white and female and less likely to have certain life-threatening diseases (congestive heart failure, ischemic heart disease, or cancer)? Second, did the higher rate of use of antidepressants (P<0.001) in the group receiving atypical agents confer a protective effect with regard to the outcome? Third, did the fact that persons with neuropsychiatric disorders preferentially received more atypical agents (P<0.001) have a salutary effect on the outcome?
We would welcome input from the authors with regard to the elaboration and clarification of other factors that have been considered, or that could be considered, in a follow-up analysis of their original data.
Stefan P. Kruszewski, M.D.
732 Forest Rd.
Harrisburg, PA 17112
joeysdogma@comcast.net
Steven G. Klotz, M.D.
2461 Lititz Pike
Lancaster, PA 17601
The authors reply: Each letter writer justifiably suggests the possibility of residual confounding by factors such as severity of illness. The analyses presented did control for age, sex, and coexisting medical conditions. We also conducted additional analyses in which we controlled for a commonly used measure of the severity of coexisting illness1 and found no change in the 180-day mortality associated with the use of conventional as compared with atypical antipsychotic medications (adjusted hazard ratio, 1.37; 95 percent confidence interval, 1.26 to 1.48). The authors of two of the letters also inquire as to whether the use of other psychiatric drugs may have played a role. However, neither the use of antidepressants (adjusted hazard ratio, 0.94; 95 percent confidence interval, 0.87 to 1.09) nor the use of other psychotropic drugs (adjusted hazard ratio, 0.99; 95 percent confidence interval, 0.89 to 1.09) conferred significant protective effects, and we controlled for both variables in adjusted models.
Lee and Newman expected the instrumental-variable estimate to be biased toward the null. But because instrumental-variable methods correct for an omitted variable bias, the point estimate could move either up or down, depending on the nature of the unobserved confounding. The validity of instrumental-variable methods does not depend on a perfect concordance between the treatment and the instrument, although instruments that are more strongly related to treatment yield estimates with smaller standard errors.2,3 Although we cannot completely eliminate the possibility of a residual confounding by the other variables that the letter writers mention, a recent meta-analysis of randomized trials conducted among elderly persons with dementia showed that the conventional agent haloperidol increased the risk of death in the short term, as compared with placebo, by 107 percent.4 This risk was higher than that seen for atypical agents and remarkably close to the sum of the 60 to 70 percent increase in the risk associated with atypical agents, as compared with placebo, in the Food and Drug Administration's analyses plus the 37 percent increase in the risk with conventional agents, as compared with atypical agents, that we observed in our study.
In spite of the robustness of our findings to various analyses we conducted and corroboration from the limited data that have emerged from randomized trials, we agree with all the correspondents that observational findings such as these need to be interpreted cautiously. We continue to conclude conservatively only that conventional antipsychotic medications may not be safer than atypical agents in elderly populations. Finally, we completely concur with Stone and colleagues that additional data from randomized, controlled trials, particularly those involving elderly persons, are sorely needed.
Philip S. Wang, M.D., Dr.P.H.
Sebastian Schneeweiss, M.D., Sc.D.
M. Alan Brookhart, Ph.D.
Harvard Medical School
Boston, MA 02115
pwang@rics.bwh.harvard.edu
References
Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613-619.
Brookhart MA, Wang P, Solomon DH, Schneeweiss S. Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable. Epidemiology (Web, EpiFast-Track). (Accessed February 11, 2006, at http://www.epidem.com.)
Angrist JD, Imbens GW, Rubin DB. Identification of causal effects using instrumental variables. J Am Stat Assoc 1996;91:444-472.
Schneider LS, Dagerman KS, Insel P. Risk of death with atypical antipsychotic drug treatment for dementia: meta-analysis of randomized placebo-controlled trials. JAMA 2005;294:1934-1943.