Skip to contents

Fit a Gaussian or multinomial random-effects model for the sensitivity and specificity of two tests applied to the same participants in each study.

Usage

mars_dta_paired(
  data,
  studyID = "study_id",
  stratum = "stratum",
  pp = "pp",
  pn = "pn",
  np = "np",
  nn = "nn",
  tests = c("test1", "test2"),
  zero_cell = c("none", "add_half"),
  estimation_method = c("REML", "MLE"),
  structure = c("UN", "DIAG1"),
  control = list(),
  family = c("gaussian", "multinomial"),
  quadrature = 11L
)

Arguments

data

Data frame with two rows per study, one for each reference-standard stratum.

studyID, stratum

Names of the study identifier and stratum columns. Strata must be labeled "diseased" and "non_diseased".

pp, pn, np, nn

Names of the joint count columns. The letters indicate positive (p) or negative (n) results in the order test 1, test 2. For example, pn counts participants positive on test 1 and negative on test 2. Counts must be finite nonnegative integers, and both strata must have positive totals.

tests

Two distinct test labels, in the order used by the count columns.

zero_cell

Use "none" (the default) to require finite logits. With "add_half", add 0.5 to all four joint cells in a stratum when either test has a zero marginal success or failure count. Empty joint cells alone do not trigger a correction.

estimation_method

Restricted maximum likelihood ("REML", the Gaussian default) or maximum likelihood ("MLE", required for multinomial).

structure

Between-study covariance structure. Use "UN" (default) for an unstructured matrix or "DIAG1" for separate variances with zero covariances. The matrix is 4-by-4 for Gaussian fits and 2-by-2 for multinomial fits.

control

Named optimizer controls passed to mars() or to stats::optim() for multinomial fits.

family

Sampling model, either "gaussian" (default) or "multinomial".

quadrature

Number of adaptive Gauss–Hermite nodes per dimension for multinomial fits, an integer from 3 to 35.

Value

A mars_dta_paired object containing the underlying Gaussian mars fit (fit), the prepared effects, joint_counts and sampling_covariance, the correction policy (zero_cell), test labels (tests), and the labeled between-study covariance (tau). Multinomial fits contain the optimizer result in fit, raw joint_counts, stratum-specific association odds ratios (association), and the between-study covariance of the two common shifts (tau).

Details

The Gaussian model uses dta_paired_effects() to construct four study-level logits and their sampling covariance. Both tests must be represented in every study. At least two studies are required. Both heterogeneity structures use the full paired sampling covariance. The Gaussian engine needs invertible sampling blocks. Singular or numerically singular blocks, such as those from perfectly concordant tests within a stratum, raise an error identifying the affected studies.

The multinomial model uses the raw four-cell tables and requires zero_cell = "none". Within each stratum, the log-linear model has one main effect per test and a fixed interaction. A common normal study effect shifts both test main effects in that stratum. The study effects can covary across strata. The interaction is the log odds ratio of association between the two test results, shared across studies within that stratum. Empty cells are valid within individual studies, with each joint outcome observed somewhere in each stratum.

References

Nikoloulopoulos, A. K. (2024). Joint meta-analysis of two diagnostic tests accounting for within and between studies dependence. Statistical Methods in Medical Research, 33(10), 1800–1817. doi:10.1177/09622802241269645 .

Examples

i <- seq_len(24)
make_stratum <- function(stratum, p_a, p_b) {
  # Mix independent and concordant positive results in these fictional data.
  both <- .7 * p_a * p_b + .3 * pmin(p_a, p_b)
  pp <- round(200 * both)
  pn <- round(200 * (p_a - both))
  np <- round(200 * (p_b - both))
  data.frame(study_id = sprintf("s%02d", i), stratum, pp, pn, np,
             nn = 200 - pp - pn - np)
}
paired_studies <- rbind(
  make_stratum("diseased", plogis(1 + .8 * sin(i)),
               plogis(1.4 + .5 * sin(i) + .7 * cos(2 * i))),
  make_stratum("non_diseased", plogis(-1.1 + .65 * cos(i)),
               plogis(-.8 + .45 * cos(i) + .65 * sin(2 * i)))
)

paired_fit <- mars_dta_paired(paired_studies, tests = c("A", "B"))
summary(paired_fit)$contrasts
#>      endpoint test reference  difference difference_lower difference_upper
#> 1 sensitivity    B         A  0.06592716       0.02756036       0.10429397
#> 2 specificity    B         A -0.06109458      -0.10218769      -0.02000147
#>   odds_ratio odds_ratio_lower odds_ratio_upper
#> 1  1.4393821         1.164742        1.7787809
#> 2  0.7366104         0.604432        0.8976939