When the incidence trait is binary, we have mainly considered a case-only analysis of prognosis. Other approaches are possible, such as setting the prognosis to a degenerate value for controls and then analysing cases and controls together, with adjustment for case/control status18. In our simulations, we found no systematic difference between the case/control and case-only analysis. Note that our approach could be applied in conjunction with the case/control analysis, and possibly with further adjustment for measured confounders of incidence and prognosis. This would have the desirable effect of reducing index event bias through several complementary approaches at once.
