Clinical prediction rules. A review and suggested modifications of methodological standards

Clinical prediction rules are decision-making tools for clinicians, containing variables from the history, physical examination, or simple diagnostic tests.

To review the quality of recently published clinical prediction rules and to suggest methodological standards for their development and evaluation.

Four general medical journals were manually searched for clinical prediction rules published from 1991 through 1994.

Four hundred sixty potentially eligible reports were identified, of which 30 were clinical prediction rules eligible for study. Most methodological standards could only be evaluated in 29 studies.

Two investigators independently evaluated the quality of each report using a standard data sheet. Disagreements were resolved by consensus.

The mathematical technique was used to develop the rule, and the results of the rule were described in 100% (29/29) of the reports. All the rules but 1 (97% [28/29]) were felt to be clinically sensible. The outcomes and predictive variables were clearly defined in 83% (24/29) and 59% (17/29) of the reports, respectively. Blind assessment of outcomes and predictive variables occurred in 41% (12/29) and 79% (23/29) of the reports, respectively, and the rules were prospectively validated in 79% (11/14). Reproducibility of predictive variables was assessed in only 3% (1/29) of the reports, and the effect of the rule on clinical use was prospectively measured in only 3% (1/30). Forty-one percent (12/29) of the rules were felt to be easy to use.

Although clinical prediction rules comply with some methodological criteria, for other criteria, better compliance is needed.