Home › Survival › Survival estimators

Each subject's expected lifetime: lifelines' predict_expectation, the trapezoid under its survival.

public double[] PredictExpectation(ReadOnlySpan<double> design)

Parameters — design holds the subjects' covariates row-major, FeatureCount per row; empty for the one subject of a fit with no covariate.

Returns — one value per subject.

Exceptions — ArgumentException when design is not whole rows of finite covariates. ArgumentException naming design also when it, with any intercept column, or its rows times the event times, is more cells than one array holds (#1614, #1616).

Example — an untreated and a treated patient.

using Lodestar.Survival;

double[] treated = [0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1];
double[] months = [3, 10, 5, 4, 8, 18, 12, 6, 9, 20, 7, 15];
bool[] died = [true, true, true, false, true, true, true, true, true, false, true, true];

AalenSummary fit = AalenAdditive.Fit(treated, months, died, featureCount: 1);

double[] means = fit.PredictExpectation([0.0, 1.0]);
double untreated = Math.Round(means[0], 6);     // => 4.696046
double onTreatment = Math.Round(means[1], 6);   // => 11.776772

Remarks — the trapezoid runs over EventTimes, from the first event time, not from zero, as lifelines integrates it. The area before the first event and after the last is left out, so the value is not a mean lifetime: above, the untreated patient's 4.67 months omits the three before the first death, and the treated patient's stops at month 18 with nearly half still alive. Read it as a comparison between subjects of one fit.

Applies to — net10.0, netstandard2.0.

See also — AalenSummary, AalenSummary.PredictMedian.