ObjectivePatients with dementia may be at high risk for psychiatric adverse drug reactions (ADRs). A Bayesian approach was developed for the assessment of changes in cognition and behaviour suspected as being related to psychotropic medications.DesignThis was a prospective observational study. A Bayesian model calculated the posterior probability (PsP) in favour of psychotropic drugs being the cause of the psychiatric adverse events detected. Bayesian results were cross-validated by comparison with an expert clinician and 2 other clinicians using a causality visual analogue scale (VAS).Setting and PatientsThe setting was a cognitive support unit in a long term care facility associated with a teaching hospital. 107 (97 males and 10 females) cognitively impaired institutionalised elderly patients (mean age ± SD = 78 ± 7, range 59 to 98 years) were monitored for 3 months for changes in cognition and behaviour.Results13 behavioural events (6 agitation, 4 confusion and 3 cognitive decline) were detected. Bayesian analysis implicated lorazepam, levomepromazine (methotrimeprazine) plus oxazepam, perphenazine and zopiclone as causing 4 of 13 behavioural events. Bayesian results were correlated to the VAS results of the expert rater (r = 0.78, p = 0.002) and less so to the other clinicians (r = 0.68, p = 0.01; r = 0.49, p = not significant). The expert rater identified the same 4 cases as being most likely to have been drug induced. Bayesian results indicated that drug-induced psychiatric events occurred in 3.7% of patients and the drug-attributable risk was 5.4% over 3 months.ConclusionsThe Bayesian approach identified psychiatric ADRs in high agreement with an expert rating. These results suggest that drug-induced cognitive and behavioural symptoms are an important source of reversible morbidity in patients with dementia.