Skip to contents

Calculate a Rutter–Gatsonis summary ROC curve from a Gaussian or binomial mars_dta() fit.

Usage

dta_sroc(object, newdata = NULL, fpr = seq(0, 1, length.out = 101L))

Arguments

object

A fitted mars_dta() object.

newdata

Data frame with one row per moderator profile. Required when the fit includes moderators. All moderators, including a measured threshold if supplied in the model, are held fixed within each curve.

fpr

Numeric vector of false-positive rates (one minus specificity), between zero and one inclusive.

Value

A data frame with columns profile (row of newdata, or 1 for an intercept-only fit without newdata), fpr, and sensitivity. The false-positive rates are returned in input order for each profile.

Details

The curve varies the latent threshold at the profile's mean accuracy in the Rutter–Gatsonis parameterization. Its logit-scale equation is $$\operatorname{logit}(\mathrm{sensitivity}) = \mu_{\mathrm{se}} + b[\operatorname{logit}(\mathrm{FPR}) + \mu_{\mathrm{sp}}],$$ where \(\mu_{\mathrm{se}}\) and \(\mu_{\mathrm{sp}}\) are the fitted mean logits of sensitivity and specificity, and \(b\) is the slope. The slope is the ratio of the between-study standard deviations for logit sensitivity and logit specificity. Each curve is increasing and passes through its profile's pooled operating point. Profiles share a slope because mars_dta() estimates a common between-study covariance.

Both between-study variances must exceed 1e-8 times the maximum of one and the two variances. The curve uses the marginal variances and is also defined for structure = "DIAG1". The between-study correlation still enters the joint regions in the ROC plot.

The curve summarizes the fitted model at fixed accuracy. To examine the association with a measured threshold, use predict.mars_dta() to compare pooled operating points at different threshold values. False-positive rates beyond the observed range are extrapolations.

References

Rutter, C. M., and Gatsonis, C. A. (2001). A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Statistics in Medicine, 20(19), 2865–2884. doi:10.1002/sim.942

Arends, L. R., et al. (2008). Bivariate random effects meta-analysis of ROC curves. Medical Decision Making, 28(5), 621–638. doi:10.1177/0272989X08319957

Harbord, R. M., Deeks, J. J., Egger, M., Whiting, P., and Sterne, J. A. C. (2007). A unification of models for meta-analysis of diagnostic accuracy studies. Biostatistics, 8(2), 239–251. doi:10.1093/biostatistics/kxl004 .

Examples

counts <- data.frame(study = 1:8, TP = c(44, 50, 56, 59, 64, 68, 69, 71),
                     TN = c(52, 42, 62, 74, 53, 65, 59, 72))
counts <- transform(counts, FN = 80 - TP, FP = 90 - TN)
fit <- mars_dta(counts, studyID = "study")
dta_sroc(fit, fpr = c(.1, .2, .5))
#>   profile fpr sensitivity
#> 1       1 0.1   0.3468969
#> 2       1 0.2   0.5867273
#> 3       1 0.5   0.8840316