reitsma-class function

Methods for reitsma objects.

Objects of the class reitsma are output by the function with the same name. Apart from standard methods the functions sroc, mcsroc and ROCellipse provide SROC curves and confidence regions for fits.

## S3 method for class 'reitsma'
print(x, digits = 4, ...)
## S3 method for class 'reitsma'
summary(object, level = 0.95, sroc.type = "ruttergatsonis", ...)
## S3 method for class 'reitsma'
sroc(fit, fpr = 1:99/100, type = "ruttergatsonis", return_function = FALSE, ...)
## S3 method for class 'reitsma'
mcsroc(fit, fpr = 1:99/100, replications = 10000, lambda = 100, ...)
## S3 method for class 'reitsma'
ROCellipse(x, level = 0.95, add = FALSE, pch = 1, ...)
## S3 method for class 'reitsma'
crosshair(x, level = 0.95, length = 0.1, pch = 1, ...)
## S3 method for class 'reitsma'
plot(x, extrapolate = FALSE, plotsumm = TRUE, level = 0.95, 
     ylim = c(0,1), xlim = c(0,1), pch = 1, sroclty = 1, sroclwd = 1, 
     predict = FALSE, predlty = 3, predlwd = 1, type = "ruttergatsonis", ...)
## S3 method for class 'reitsma'
anova(object, fit2, ...)
## S3 method for class 'anova.reitsma'
print(x, digits = 4, ...)

Arguments

• x: a reitsma object.
• object: a reitsma object.
• fit: a reitsma object.
• fit2: a reitsma object.
• digits: number of decimal digits to print.
• level: numeric, the level for calculations of confidence intervals (summary) or regions (ROCellipse)
• sroc.type: character, which SROC curve should be used to calculate the AUC in the summary? Besides the default ruttergatsonis the option naive is available.
• return_function: logical. Should a function on ROC space be returned or the values at the points given by fpr?
• fpr: numeric, the false positives rates for which to calculate the predicted sensitivities
• replications: integer, the number of replications for the Monte-Carlo SROC curve
• lambda: numeric, the parameter lambda of the Monte-Carlo run, see details
• add: logical, should the confidence region be added to the current plot? If set to FALSE a matrix of points of the ellipse is returned
• extrapolate: logical, should the SROC curve be plotted beyond the observed false positive rates?
• plotsumm: logical, should the summary pair of sensitivity and false positive rate together with its confidence region be plotted?
• length: positve numeric, length of the "whiskers" of the crosshairs.
• ylim: numeric of length 2, which section of the sensitivities to plot?
• xlim: numeric of length 2, which section of the false positive rates to plot?
• pch: integer, symbol for the pair of mean sensitivity and false positive rate
• sroclty: integer, line type of the SROC curve
• sroclwd: integer, line width of the SROC curve
• predict: logical, draw prediction region?
• predlty: integer, line type of prediction region
• predlwd: integer, line width of prediction region
• type: character, type of SROC curve to plot. Can be either the generalization of the Rutter & Gatsonis (2001) SROC curve (see below) or the naive curve implied the bivariate model.
• ...: arguments to be passed on to other functions

Details

The confidence regions of ROCellipse are first calculated as ellipses on logit-ROC space, so the back-transformed regions that are output are not necessarily ellipses. The Monte-Carlo SROC curves are generated from random samples from the fitted model and a lowess smooth through them is output. Many computational details are to be found in Doebler et al. (2012).

The summary function for reitsma objects also contains the five parameters of the HSROC model by Rutter & Gatsonis (2001) if no regression is performed. These values are calculated by using the formulae from Harbord et al. (2007).

The plot method for reitsma objects will plot the generalization of the Rutter-Gatsonis curve.

If you require positive or negative likelihood ratios, you should use SummaryPts. If you require positive or negative predictive values, see predv_r and predv_d.

Returns

sroc returns a matrix ready for plotting. Each row corresponds to one point in ROC space. mcsroc returns a lowess smooth. ROCellipse returns a list, the first element being a matrix of points in ROC space that delimit the confidence region and the second is the point estimate of the pair of sensitivity and false positive rate in ROC space.

References

Doebler, P., Holling, H., Boehning, D. (2012) A Mixed Model Approach to Meta-Analysis of Diagnostic Studies withBinary Test Outcome. Psychological Methods, to appear

Author(s)

Philipp Doebler philipp.doebler@googlemail.com

See Also

reitsma, SummaryPts

Examples

# load data
data(Dementia)
# fit model
fit <- reitsma(Dementia)
# calculate a confidence region but do not plot it
cr.Dementia <- ROCellipse(fit)
#calculate a SROC curve
sroc.Dementia <- sroc(fit)
# plot the confidence region in ROC space as a line
plot(cr.Dementia$ROCellipse, type = "l", xlim = c(0,1), ylim = c(0,1))
# add the point estimate of the mean
points(cr.Dementia$fprsens)
# add the SROC curve
lines(sroc.Dementia)