Survival in STATA

2
Survival in STATA If you want to… then use command…. useful options (things you put after the comma) Tell STATA you have a censored outcome with observation time obstime and event indicator status stset obstime, failure(status) some exist in STATA, but you won’t need them currently Graph a Kaplan-Meier Survival curve sts graph by(varname) allows you to graph separate survival curves on same plot by level of varname gwood adds Greenwood 95% CIs around the survival curve Graph a smoothed curve of the baseline hazard function using kernel density estimation (based on the data, not any model) sts graph hazard makes a graph of baseline hazard instead of K-M survival curve cihazard adds confidence bands around hazard function List Kaplan-Meier survival estimates sts list by(varname) lists K-M survival estimates by levels of varname Perform a logrank test of equality of survival functions sts test varname wilcoxon allows you to do a Wilcoxon- Breslow test instead of a logrank test Fit a Cox Proportional Hazards model stcox predictor list nohr gives estimated beta coefficients instead of hazard ratios robust provides robust standard error estimates cluster(id) lets STATA know you have dependent data by id schoenfeld(varname*) stores Schoenfeld residuals for each observation in a variable for each predictor in predictor list scaledsch(varname*) stores scaled Schoenfeld residuals mgale(varname) stores Martingale residuals for each observation basesurv(varname) stores estimates of baseline survival in varname basehc(varname) stores estimates of baseline hazard in varname

description

Survival in STATA

Transcript of Survival in STATA

  • Survival in STATA If you want to then use command. useful options (things you put after

    the comma) Tell STATA you have a censored outcome with observation time obstime and event indicator

    status

    stset obstime, failure(status)

    some exist in STATA, but you wont

    need them currently

    Graph a Kaplan-Meier Survival curve sts graph by(varname) allows you to graph separate survival curves on same plot

    by level of varname gwood adds Greenwood 95% CIs around

    the survival curve Graph a smoothed curve of the baseline

    hazard function using kernel density estimation (based on the data, not any

    model)

    sts graph hazard makes a graph of baseline hazard instead of K-M survival curve cihazard adds confidence bands around

    hazard function

    List Kaplan-Meier survival estimates sts list by(varname) lists K-M survival estimates by levels of varname

    Perform a logrank test of equality of survival functions

    sts test varname wilcoxon allows you to do a Wilcoxon-Breslow test instead of a logrank test

    Fit a Cox Proportional Hazards model stcox predictor list nohr gives estimated beta coefficients instead of hazard ratios

    robust provides robust standard error estimates

    cluster(id) lets STATA know you have dependent data by id

    schoenfeld(varname*) stores Schoenfeld residuals for each observation in a

    variable for each predictor in predictor list

    scaledsch(varname*) stores scaled Schoenfeld residuals

    mgale(varname) stores Martingale residuals for each observation

    basesurv(varname) stores estimates of baseline survival in varname

    basehc(varname) stores estimates of baseline hazard in varname

  • Test the Proportional Hazards assumption (after fitting a Cox model)

    stphtest global test requires having specified schoenfeld() in stcox

    individual tests require having specified scaledsch() in stcox

    Plot a ln(-ln(S(t)) plot (after fitting a Cox model)

    stphplot by(varname) specifies plotting by levels of varname

    Plot survival predicted by Cox model with observed survival by Kaplan-Meier

    stcoxkm by(varname) specifies plotting by levels of varname

    Plot estimates of the hazard (survival) function at mean values of covariates (after

    fitting a Cox model)

    stcurve hazard is for plotting hazard function (requires basehc() option to have been

    specified in stcox) survival is for plotting survival

    function (requires basesurv() to have been specified in stcox)

    at(varname=#) allows plotting of curve at covariate values specified by

    varname=# at1(varname=#) allows you to plot

    curve for another set of covariates on same plot as at()

    range(startnum endnum) allows you to plot in a range between startnum and

    endnum (useful for zooming in on plots to get survival estimates)

    Plot Scaled Schoenfeld residuals vs. time and look for flatness of the smooth (flat smooth

    implies PH is okay for that covariate, increasing or decreasing smooth implies

    problems with PH)

    lowess varname _t (here, varname is the stored scaled

    Schoenfeld residuals corresponding to covariate you are checking for PH)

    bwidth(#) allows bandwidth specification by a number #

    (lower bandwidth makes curve try and fit each data point more)

    Plot Martingale residuals for continuous covariates and look for shape of smooth (flat smooth implies functional form for covariate is okay, curved smooth implies transforming

    covariate in the model)

    lowess varname covariate (here, varname is the stored

    Martingale residuals, covariate is a continuous covariate)

    bwidth(#) allows bandwidth specification by a number #

    mean uses running-mean smoothing instead of least-squares smoothing

    (might want to do this when plotting Martingale residuals)