Time-to-event figures

The catalog opens here because this is most of what we publish. Cardiac-surgery outcomes are almost always time-to-event questions – freedom from reoperation, survival after a valve repair, the cumulative risk of a complication over years of follow-up – so the survival family earns the front of the book.

Each chapter takes one view of the same kind of data, a follow-up time and an event indicator, and answers a different question with it. The Kaplan-Meier curve, with its numbers-at-risk table, is the workhorse for showing how a group survives over time. Hazard curves read the same data as an instantaneous rate rather than a running total, and the temporal-hazard plot shows how that rate rises or falls across the follow-up window. The number-needed-to-treat chapter closes the part by turning an effect size into the single count a clinician acts on. Every recipe builds its own sample cohort, so you can run it before pointing it at your own data.

When a Cox model is too rigid for your data – when a predictor’s effect bends, or depends on the patient it acted on – a survival random forest can find that structure without you specifying it up front. The Random forests & model visualization part, later in the book, fits one and reads the same time-to-event question off it.