Serial laboratory determinations are now routinely performed on patients admitted to intensive-care units. Adequate interpretation of such cumulative information for clinical decision-making purposes is a challenging problem. We describe a statistical method for predicting--sequentially as the data become available--the patient's outcome, death or survival. Thus, the method goes beyond previously reported techniques that base such prediction on only a single multivariate observation. The method has been applied to daily measurements of serum urea and lactate dehydrogenase, performed during one week on patients hospitalized in the coronary-care unit with acute myocardial infarction. Two baseline variables were also included in the dynamic risk index so derived: the age of the patient and the number of previous myocardial infarctions recorded on admission. We also discuss the problems of selecting the most-predictive laboratory tests and of determining for each test the amount of past data needed to achieve satisfactory prediction. We distinguish between global evaluation of the dynamic risk index obtained (in terms of specificity and sensitivity) and individual interpretation (in terms of posterior/prior probability ratio) of a given risk score for a particular patient. The approach described may contribute to more effective use of results of repeated laboratory tests on critically ill patients.