Ten derivation studies (10,359 patients) and 14 validation studies (26,488 patients) were included in the review. One clinical prediction rule was validated in six new patient series and two were validated in three new patient series; all other rules were validated in one or two studies. QUADAS scores ranged from 7 to 11. All studies used an appropriate reference standard and avoided partial verification bias. None of the studies provided information on inter- or intra-observer reliability.
Sensitivity of the clinical prediction rules in the derivation studies ranged from 81% to 100% and specificity ranged from 17% to 93%. Four derivation studies reported sensitivity of greater than 95%. Validation studies all showed lower estimates of sensitivity than those in the corresponding derivation studies. Negative likelihood ratios ranged from <0.01 (excellent ability to rule out acute coronary syndrome) to 0.9 (limited utility to rule out acute coronary syndrome) across all studies.
For one clinical prediction rule (Selker), five out of six validation studies reported estimates of sensitivity of 95% or more (range 95% to 100%) and one reported a sensitivity of 84%. This corresponded to negative likelihood ratios that ranged from <0.01 to 0.4. For all other clinical prediction rules, validation studies reported estimates of sensitivity of less than 95%. The threshold that the authors stated was required for a clinical prediction rule to be acceptable to clinicians was a sensitivity of 95%.