Thirty-six studies were included in the review (n=59,978 participants). All studies provided details on the reliability and validity of the reference standard. All studies indicated sampling methods. Eleven studies did not report blinding of physicians to the reference standard diagnosis.
Sensitivity ranged from 0% to 79.4% and specificity from 48% to 99%. Pooled sensitivity was 36.4% (95% CI 27.9 to 44.8). Pooled specificity was 83.7% (95% CI 77.5 to 90.0). The pooled diagnostic odds ratio was 4.0 (95% CI 3.2 to 4.9). Overall sensitivity based on summary ROC curves was 42.3%. Results were reported as heterogeneous (no data provided).
Meta-regression showed that method of documentation, age of sample and date of publication showed a statistically significant effect on summary sensitivity. Further analysis demonstrated that use of physician diagnosis as a method of documentation had a higher sensitivity than chart review when controlling for sample age and date of publication.