Systematic review of clinical prediction rules for neuroimaging in the evaluation of dementia.

BACKGROUND Clinical practice guidelines for dementia do not recommend routine neuroimaging but vary in their recommended clinical prediction rules to identify patients who should undergo neuroimaging for potentially reversible causes of dementia. METHODS Using a MEDLINE search supplemented by other strategies, we identified studies from January 1, 1983, through December 31, 1998, that evaluated the diagnostic performance of a clinical prediction rule. We calculated the sensitivity and specificity of each rule, then evaluated their diagnostic performance in a hypothetical cohort of 1000 patients with dementia, varying the prevalence of potentially reversible dementia from 1% to 15%. RESULTS We identified 7 studies that evaluated at least 1 of 6 different clinical prediction rules. Only one rule consistently had high sensitivity (>85%) across all studies; none consistently had high specificity (>85%). Six of the 7 studies included less than 15 cases of potentially reversible dementia; thus the sensitivity and specificity for each rule had relatively wide confidence intervals. At a 5% prevalence of potentially reversible dementia, all rules had low positive predictive value (<15%) in our hypothetical cohort. Depending on the rule, our analysis predicts 6 to 44 of the 50 patients with potentially reversible dementia (5% prevalence in cohort of 1000 patients) would not undergo imaging. CONCLUSIONS There is considerable uncertainty in the evidence underlying clinical prediction rules to identify which patients with dementia should undergo neuroimaging. Application of these rules may miss patients with potentially reversible causes of dementia.