Table 6.

Meta‐analysis results for calibration of either PPV or NPV, when using meta‐analysis model (17) as estimated in a frequentist or Bayesian framework.¹

Example	Meta‐analysis method	Statistical framework	Summary calibration, a (95% CI)	${\widehat{\tau }}_a$	95% prediction interval for O/E in a new population∗	Probability 0.9 < O/E < 1.1 in a new population∗
Calibration for just NPV
PTH data 1–2 h	Option B	Bayesian	0.24 (−0.97 to 1.81)	0.51 (0.03 to 0.98)	0.86 to 1.05	0.95
	Option B	Frequentist	0.093 (−1.06 to 1.25)	0	0.87 to 1.03	—
PTH data 0–20 min	Option B	Bayesian	0.021 (−0.82 to 1.02)	0.51 (0.03 to 0.97)	0.86 to 1.04	0.95
	Option B	Frequentist	−0.044 (−0.83 to 0.74)	0.34	0.80 to 1.03	—
Calibration for just PPV
Temperature data	Option A	Bayesian	−0.017 (−0.63 to 0.77)	0.65 (0.11 to 0.99)	0.90 to 1.03	0.98
	Option A	Frequentist	0.015 (−0.75 to 0.78)	0.70	0.87 to 1.03	—

¹All frequentist analyses used maximum likelihood estimation of model (17). All Bayesian analyses used a prior distribution of N(0, 1 000 000) for a, and a prior distribution of uniform(0, 1) for τ, with a 10 000 burn‐in followed by 100 000 samples for posterior inferences. Median values of the posterior distribution are shown for a and τ.

^∗Based on a predicted positive predictive value (PPV) of 0.97 in the temperature analysis, and a negative predictive value (NPV) of 0.95 in the parathyroid (PTH) analysis.

O/E, observed/expected.