Immortal Time Bias: the hidden confounder in assessing cardiosurgical treatment effects.

Abstract

(A) Schematic illustration of the principle of immortal time bias. The grey mortality curve illustrates an assumption of mortality risk in patients with infective endocarditis while waiting for surgery based on data from four studies reporting on the survival of patients with infective endocarditis who had a surgical indication but did not receive surgery.1–4 The columns schematically reflect an assumption of mortality risk associated with waiting for (brown columns) and with receiving surgery (blue columns) at a given time (early or late surgery), as an example. The vertical red line then represents the difference in mortality associated with waiting for surgery, which reflects the ‘immortal time bias’ in this example, if a retrospective assessment of surgical mortality is performed. (B) Postoperative mortality (left panel) and postoperative neurological deterioration (right panel) associated with surgery or conservative treatment, depending on the time of surgery. Data were obtained from a meta-analysis involving nine studies on infective endocarditis with intracranial haemorrhage.5 The times on the x-axis reflect the cut-off for surgery performed before or after this time.