Background: In 2005, a meta-analysis of 15 randomized placebo-controlled trials showed that atypical antipsychotics were associated with an increased risk of mortality in patients with dementia. Consequently, health authorities warned against their use. However, many trials had baseline differences that might explain the meta-analytic findings.
Objective: To explore whether the association between baseline differences and mortality could be estimated with a meta-regression analysis.
Methods: We abstracted from the articles: 1) randomization features; 2) trial characteristics; and 3) five baseline characteristics that predict death, that is higher mean age, neuropsychiatric symptom (NPS) score, and higher percentage of male, non-white and vascular/mixed dementia patients per treatment group. Each characteristic was scored as high risk if higher or unclear in active treatment group, and otherwise as low.
Results: Randomization procedures were poorly described in all trials. Two reports did not present a baseline table. We found not high percentage of males in the active treatment group but studies with a high percentage of females yielded a pooled increased risk of death. The other variables predicted an increased risk of mortality as expected. For each of the five baseline characteristics there were more studies that scored high risk. We then calculated a sum-variable ‘baseline differences’ per study that indicated the number of variables that predicted death (0-5). Both studies that had a haloperidol group in addition to a atypical antipsychotic group scored 0, and nine of the other studies had a score 3 , 4 of 5. In a meta-regression analysis of 12 trials, this sum-variable predicted the risk of mortality (beta 0.32; 95% CI -0.03 to 0.7).
Conclusion: Unfavorable baseline characteristics in the atypical antipsychotic group might have mistakenly suggested an increased mortality rate in randomized placebo-controlled trials.