Objective: To determine a meta-analytical definition of the discriminant power of Ranson's signs in the prediction of acute pancreatitis severity and outcome, and of their information content, also compared with clinical judgment.
Data Sources: Two hundred eleven studies since 1974, reporting any predictive system for acute pancreatitis (MEDLINE by various Medical Subject Headings in MEDLARS, Current Contents, Medscape, Virtual Hospital, and other on-line medical services).
Study Selection: One hundred ten studies reporting clinical use of Ranson's signs were retained. A quality index was calculated for each study. A selection was made according to inclusion criteria, separately for prediction of severity (19 studies; group S) and prognosis (10 studies; group P). Six other studies reporting clinical judgment results were also selected (group C).
Data Extraction: Sensitivity and specificity values were extracted. Effect sizes were calculated and summarized by the inverse variance-weighted method. Categorical models were studied by analysis of variance. Publication bias was sought by correlation test and analysis of variance. Summary receiver operating characteristic curves were drawn, and the corresponding false-positive rate (FPR) and true-positive rate were calculated for each group. From the total true-positive rate and FPRs, the probabilities of illness for positive and negative results were calculated, for severe pancreatitis prevalence from 0 to 1. Last, the area below the curve and the ratio between this and that of the "perfect test" were calculated as a measurement of information content.
Data Synthesis: Ranson's signs demonstrated poor discriminant power in both predictions: "d" values were 1.200 (95% confidence interval, 1.083-1.318) and 1.302 (95% confidence interval, 1.046-1.559), respectively. The lack of homogeneity in group S (Q = 58.737; p = .0000032) can be explained by the presence of three outliers. The summary curves showed, for low FPRs, a higher sensitivity of clinical judgment; Ranson's signs reached useful sensitivity only for high FPRs. No differences between groups in the area below the information content curves were found.
Conclusions: Ranson's signs showed a poor predictive power. The information content did not differ from that of clinical judgment.