Venoocclusive disease of the liver: Development of a model for predicting fatal outcome after marrow transplantation

Purpose: Hepatic venoocclusive disease (VOD) is a common complication of cytoreductive therapy for marrow transplantation. Only 25% of patients who develop VOD have severe disease. We tested the hypothesis that early clinical signs of VOD would predict which patients would recover and which would die. Patients and Methods: We evaluated 355 consecutive patients who had transplants between August 6, 1987 and July 21, 1988 for occurrence of VOD and whether it was reversible within 100 days of transplant. Total serum bilirubin and weight gain from day -7 through day +16 posttransplant were compared among patients with no, severe, or nonsevere VOD. Logistic regression models were developed to estimate probabilities of severe VOD at each of six time intervals. The accuracy of these models was tested by applying them to 392 consecutive patients who underwent transplantation between July 22, 1988 and July 20, 1989. As early as day -1, bilirubin and weight gain were significantly different between patients whose VOD proved to be severe and patients with reversible VOD or no disease. Regression models were used to generate coefficients (β₀, β₁, β₂) for the equation P = 1/(1 + e^-z), where P is the probability of severe VOD and z = β₀ + β₁ (In total serum bilirubin [mg/dL]) + β₂ (percent weight gain). Application of this equation to the next 392 patients allowed us to calculate sensitivity, specificity, and positive predictive value for a range of probabilities. The course of VOD after cytoreductive therapy can be predicted by knowing the serum bilirubin and weight gained within 1 to 2 weeks of transplantation. Probability estimates derived from patient data are highly specific and moderately sensitive. Such probability estimates may be useful when considering potentially risky interventions to treat VOD, such as recombinant human tissue plasminogen activator.