Existing predictor systems of severe pancreatitis are cumbersome and can require up to 48 hr to complete. This study aimed to determine whether useful likelihood ratios exist for the prediction of severe pancreatitis, corresponding to various ranges of admission hematocrit. A retrospective cohort of 200 patients admitted with acute pancreatitis was identified. Likelihood ratios were calculated for a priori defined hematocrit ranges. Using multivariate logistic regression, initial hematocrit was evaluated as a predictor of severe pancreatitis as defined a priori by local and/or systemic complications (Atlanta criteria, 1992). Planned subgroup analysis was performed on those with a hematocrit >50%, stratified by 24-hr hematocrit. Fourteen patients (7%) developed severe pancreatitis. Likelihood ratios were 0.45, 0.70, and 7.5 for hematocrit ranges of < or =45, 45.1-49.9, and > or =50%, respectively. Hematocrit (as increases by 5%) was a significant predictor of severe pancreatitis (odds ratio [OR] = 2.8; P = 0.001), length of stay (P < 0.0001), necrosis (OR = 3.9; P = 0.001), and need for intensive care (OR 4.5; P = 0.002). The negative predictive value of the lowest range and positive predictive value of the highest range were 97% (95% CI: 92-99%) and 37% (16-62%), respectively. Lack of normalization of hematocrit by 24 hr did not predict severe pancreatitis. Initial hematocrit appears to be an early, simple, and useful predictor of severe pancreatitis. A normal 24-hr hematocrit does not appear to alter the prediction made by the initial hematocrit.