metaROC: Non-parametric ROC curve estimate for meta-analysis In nsROC: Non-Standard ROC Curve Analysis

Description

This function performs meta-analytic studies of diagnostic tests for both the fixed and random-effects models. In particular it reports a fully non-parametric ROC curve estimate when data come from a meta-analysis study using the information of all cut-off points available in the selected original studies. The approach considered is the one proposed by Martinez-Camblor et al. (2017) based on weighting each individual interpolated ROC curve. See References below.

Usage

 1 2 3 4 5 6 metaROC(data, ...) ## Default S3 method: metaROC(data, Ni=1000, model=c("fixed-effects","random-effects"), plot.Author=FALSE, plot.bands=TRUE, plot.inter.var=FALSE, cex.Author=0.7, lwd.Author=12, col.curve='blue', col.bands='light blue', alpha.trans=0.5, col.border='blue', ...) 

Arguments

 data a data frame containing at least the following variables (with these names): Author: a vector assigning different numbers to each paper/author. TP: true positives. FP: false positives. TN: true negatives. FN: false negatives. Ni number of points of the unit interval (FPR values) considered to calculate the curve. Default: 1000. model the meta-analysis model used to estimate the ROC curve. One of "fixed-effects" (it only considers the within-study variability) or "random-effects" (it takes into account the variability between the studies). plot.Author if TRUE, a plot including ROC curve estimates (by linear interpolation) for each paper under study is displayed. plot.bands if TRUE, confidence interval estimate for the curve is added to the plot of the ROC curve estimate. plot.inter.var if TRUE, a plot including inter-study variability estimate is displayed on an additional window. cex.Author the magnification to be used to display the paper/author points labels relative to the current setting of cex. lwd.Author the size to be used for the paper/author points. col.curve the color to be used for the (summary) ROC curve estimate. Default: blue. col.bands the color to be used for the confidence interval of ROC curve estimate. Default: light blue. alpha.trans proportion of opacity to be used for the confidence interval of ROC curve estimate. A number in the unit interval where 0 means transparent. Default: 0.5. col.border the color to be used for the border of confidence interval of ROC curve estimate. Default: blue. ... another graphical parameters to be passed.

Details

The slight modification considered to ensure the monotonicity of the summary ROC curve estimate is the following sRA(t) = max(sup_{z \in [0,t]} sRA(z), RA(t)).

Some basic information about the model used and the results obtained are printed.

Value

 data the data-frame considered ordered by Author-FPR-TPR and including the following variables: n: positive subjects sample size. m: negative subjects sample size. FPR: false positive rate. TPR: true positive rate. t values of the unit interval (FPR values) considered to calculate the curve. model the meta-analysis model used to estimate the ROC curve. One of "fixed-effects" (it only considers the within-study variability) or "random-effects" (it takes into account the variability between the studies). sRA non-parametric summary ROC curve estimate following the model considered with a slight modification to ensure the monotonicity. This is the one reported in graphics. RA non-parametric summary ROC curve estimate following the model without modifications. se.RA standard-error of summary ROC curve estimate. area area under the summary ROC curve estimate by trapezoidal rule. youden.index the optimal specificity and sensitivity (in the Youden index sense). roc.j a matrix whose column j contains the estimated ROC curve for the j-th study in each point t considered. w.j a matrix whose column j contains the weights in fixed-effects model for the j-th study in each point t considered. w.j.rem a matrix whose column j contains the weights in random-effects model for the j-th study in each point t considered. inter.var inter-study variability estimate in each point t considered. Only computed if model is "random-effects".

References

Martinez-Camblor P., 2017, Fully non-parametric receiver operating characteristic curve estimation for random-effects meta-analysis, Statistical Methods in Medical Research, 26(1), 5-20.

Examples

 1 2 3 4 5 6 7 8 data(interleukin6) # Fixed-effects meta-analysis showing linear interpolations of the papers considered in the graphic output1 <- metaROC(interleukin6, plot.Author=TRUE) # Random-effects meta-analysis displaying also a window with a plot of the inter-study # variability estimate output2 <- metaROC(interleukin6, model="random-effects", plot.Author=TRUE) 

Example output

Number of papers included in meta-analysis: 9
Model considered: fixed-effects
The area under the summary ROC curve (AUC) is 0.7721521.
The optimal specificity and sensibility (in the Youden index sense) for summary ROC curve are 0.6996997 and 0.7597871, respectively.
Number of papers included in meta-analysis: 9
Model considered: random-effects
The area under the summary ROC curve (AUC) is 0.7881458.
The optimal specificity and sensibility (in the Youden index sense) for summary ROC curve are 0.7007007 and 0.7625716, respectively.


nsROC documentation built on May 2, 2019, 2:31 p.m.