measures: Evaluating Fitted Models

In nyiuab/BhGLM: Bayesian hierarchical GLMs and survival models, with applications to Genomics and Epidemiology

Description

These functions provide various measures to evaluate fitted models.

Usage

measure.bh(object, new.x, new.y, new.offset) 

measure.glm(y, y.fitted, family, dispersion=1)

measure.polr(y, y.fitted)

measure.cox(y, lp)

surv.curves(y, lp, probs=0.50, mark.time=FALSE, main=" ", lwd=2, lty=1, col=c("black", "red"), add=FALSE) 

aucCox(y, lp, main=" ", lwd=2, lty=1, col="black", add=FALSE) 

peCox(y, lp, FUN=c("Brier", "KL"), main="", lwd=2, lty=1, col="black", add=FALSE) 

Arguments

object	a fitted object.
new.x	data frame or matrix of new values for variables used in object. for object from bglm or bpolr or bcoxph, it is data frame; for object from glmNet or bmlasso, it is matrix.
new.y	vector of new response values corresponding to new.x. If new.x or new.y are omitted, the fitted linear predictors are used for prediction.
new.offset	data frame or vector of offset values for new data points. If new.x includes offset, do not need to set new.offset.
y	observed response values.
y.fitted	predicted (estimated) response values for GLMs or probabilties of response values for ordinal models from a fitted model or cross-validation.
family	family in GLMs.
dispersion	dispersion in GLMs. theta in negative binomial model.
lp	a vector of prognostic index (linear predictor) from a fitted Cox model or cross-validation.
probs	numeric value or vector of probabilities with values in [0,1] for grouping the patients based on lp. Default probs = 0.50: partition samples to two groups according to the percentile of 50
mark.time	controls the labeling of the curves. If set to FALSE, no labeling is done. If TRUE, then curves are marked at each censoring time.
FUN	The error function, either "KL" (default) for Kullback-Leibler or "Brier" for Brier score.
add	logical. if TRUE, plot over the existing plot. The default is FALSE.
main, lwd, lty, col	same as in plot.

Details

The functions, measure.bh, measure.glm, measure.polr, measure.cox, provide various measures to evaluate fitted models. For all GLMs and polr models, return: deviance: estimate of deviance, mse: estimate of mean squared error. For binomial and polr models, also return: auc: area under ROC curve, misclassification: estimate of misclassification. For Cox models, return: deviance: estimate of deviance, Cindex: concordance index.

The function surv.curves plots survival curves of groups of individuals and test the difference between curves using the log-rank method in survdiff.

aucCox calculates model-free curve of Area Under the Curve values over time.

peCox calculates prediction error curve. It is an alteration of dynpred, which should be installed.

Author(s)

Nengjun Yi, [e-mail]

References

Steyerberg, E. W., 2009 Clinical Prediction Models: A Practical Approch to Development, Validation, and Updates. Springer, New York.

van Houwelinggen, H.G. & Putter, H. Dynamic Prediction in Clinical Survival Analysis, (CRC Press, 2012).

See Also

cv.bh, predict.glm, predict.coxph

Examples

library(BhGLM)

N = 1000
K = 50
x = sim.x(n=N, m=K, corr=0.6) # simulate correlated continuous variables  
h = rep(0.1, 4) # assign four non-zero main effects to have the assumed heritabilty 
nz = as.integer(seq(5, K, by=K/length(h))); nz
yy = sim.y(x=x[, nz], mu=0, herit=h, p.neg=0.5, sigma=1.6) # simulate responses
yy$coefs

y = yy$y.ordinal; fam = binomial

# partition the data into two parts; 
# fit a model using the first part and evaluate the prediction using the second part 
x1 = x[1:(N/2),]; y1 = y[1:(N/2)]
x2 = x[(N/2):N,]; y2 = y[(N/2):N]

f1 = bglm(y1 ~ ., data=x1, family=fam, prior=De(0,0.5))   
plot.bh(f1, vars.rm=1, threshold=0.01, gap=10)  

measure.bh(f1, x2, y2)


y = yy$y.surv
y1 = y[1:(N/2)]
y2 = y[(N/2):N]
f1 = bcoxph(y1 ~ ., data=x1, prior=De(0,0.5))   
plot.bh(f1, threshold=0.01, gap=10)  

measure.bh(f1, x2, y2)
 

nyiuab/BhGLM documentation built on Jan. 9, 2022, 3:31 p.m.