get_cox_calibration: Calculate D'Agostino - Nam calibration stats.

In PiotrTymoszuk/coxExtensions: Extended Inference and Goodness of Fit for Cox Models

Calculate D'Agostino - Nam calibration stats.

Description

Calculates the D'Agostino - Nam calibration for a Coxph model, a variant of the Hosmer - Lemeshov test (DOI: 10.1016/S0169-7161(03)23001-7), and computes square distances between the predicted survival probability and the outcome as proposed by Graf et al. For the D'Agostino - Nam calibration the linear predictor score of the Cox model is cut into n quantile strata by cut_quantiles. The 'observed' survival odds are derived from a Kaplan-Meier survival estimate calculated by surv_fit. The 'fitted' survival obs are derived from the Cox proportional hazard models of survival as a function of the linear predictor score quantile intervals.

Usage

get_cox_calibration(
  cox_model,
  n = 3,
  labels = NULL,
  right = FALSE,
  use_unique = FALSE,
  ...
)

## S3 method for class 'coxex'
calibrate(fit, n = 3, labels = NULL, right = FALSE, use_unique = FALSE, ...)

Arguments

cox_model	a CoxpPH model or a coxex object.
n	a single numeric defining the number of quantile intervals of the Cox model's linear predictor score.
labels	an optional user-provided vector of labels for the quantile intervals.
right	logical, indicating if the quantile intervals should be closed on the right (and open on the left) or vice versa.
use_unique	logical, should unique values of the Cox model's linear predictor scores be used instead of quantile strata? This option is non-canonical and experimental, may be utilized to check an association between an ordinal variable and survival.
...	additional arguments, currently none.
fit	a CoxPH model or a coxex object.

Value

an object of class 'calibrator', whose statistic and graphical summary can be accessed by S3 'summary' and 'plot' methods. The object consists of the following elements:

• 'lp_scores': a data frame with the linear predictor scores and its quantile intervals;

• 'surv_fit' an object of the 'surv_fit' class (surv_fit), which stores the Kaplan - Meier survival estimates;

• 'cox_fit': a coxph model of the survival as a function of the linear predictor score strata;

• 'km_estimates': a data frame with the Kaplan-Meier survival estimates for each observation, see: surv_summary for details;

• 'cox_estimates': a data frame storing the number of events and survival probability for each observation in the linear predictor strata Cox model;

• 'strata_calibration': a data frame storing the numbers of events and survivors, survival probabilities, relative survival probability (Kaplan - Meier to Cox ratio), relative risk (rr: Kaplan-Meier to Cox ratio) and the D'Agostino-Nam chi-squared statistic (x2_dn) for each strata;

• 'global_calibration': a data frame with the mean relative risk, the sum D'Agostino - Nam chi-squared statistic, its degrees of freedom (df = strata number - 2 or df = 1 for two strata) and p value for the global linear predictor score.

• 'squares': a data frame with times, observed event, model-predicted survival probability and squared distance between the prediction and observed survival.

References

• D’Agostino, R. B. & Nam, B. H. Evaluation of the Performance of Survival Analysis Models: Discrimination and Calibration Measures. Handb. Stat. 23, 1–25 (2003).

• Royston, P. Tools for checking calibration of a Cox model in external validation: Approach based on individual event probabilities. Stata J. 14, 738–755 (2014).

• Crowson, C. S. et al. Assessing calibration of prognostic risk scores. Stat. Methods Med. Res. 25, 1692–1706 (2016).

• Graf, E., Schmoor, C., Sauerbrei, W. & Schumacher, M. Assessment and comparison of prognostic classification schemes for survival data. Stat. Med. 18, 2529–2545 (1999).

PiotrTymoszuk/coxExtensions documentation built on Feb. 6, 2024, 10:58 p.m.