Prognostic modeling of clinical outcomes: an illustration with data from patients with membranous nephropathy.

Probabilities that a patient will occupy any of five clinically defined compartments at different future times are generated and graphed by a personal computer. The probabilities are functions of a patient's relevant baseline characteristics (treated or control group), clinical status, and follow-up time at which the prognosis is made. The illustrative prognostic model is based on a reanalysis of detailed individual records for 81 patients with idiopathic membranous nephropathy (42 treated with methylprednisolone and chlorambucil; 39 controls) in a randomized clinical trial. The compartments to and from which patients may pass are identified as (1) complete remission, (2) partial remission, (3) the nephrotic syndrome, (4) renal failure, and (5) death. Estimated risk functions for transitions between compartments involve baseline treatment, and intermediate and temporal variables, together with their relevant interactions. The model illustrates how, despite the overall advantage of treated over control patients, the comparative prognoses can change greatly and can even sometimes be reversed, depending on a variety of follow-up experiences.