Home › Survival › Survival estimators

Each subject's median survival time: lifelines' predict_median.

public double[] PredictMedian(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 time per subject.

Exceptions — ArgumentException when design is not whole rows of finite covariates. ArgumentException naming design also when it is more cells than one array holds (#1614).

Example — no dose, and two units of it, in the fit of the AftSummary page.

using Lodestar.Survival;

double[] dose = [1.0, 2.0, 0.5, 3.0, 0.0, 1.5, 0.0, 2.5, 1.0, 0.5];
double[] months = [5, 8, 12, 3, 15, 9, 20, 6, 11, 14];
bool[] died = [true, true, false, true, true, true, false, true, true, false];

AftSummary fit = AcceleratedFailureTime.Fit(AftModel.Weibull, dose, months, died, featureCount: 1);

double[] medians = fit.PredictMedian([0.0, 2.0]);
double untreated = Math.Round(medians[0], 6);   // => 19.598539
double twoUnits = Math.Round(medians[1], 6);    // => 6.534881

Remarks — PredictPercentile at one half. The ratio of the two medians above, 0.333, is ExpCoefficients[0] squared: under an accelerated failure time model every percentile moves by the same factor.

Applies to — net10.0, netstandard2.0.

See also — AftSummary, AftSummary.PredictExpectation.