Table 1.

Stratified comparison of diagnostic odds ratio (DOR) estimates

Number of reviews	Comparison of DORs at mean thresholda
	E-ML model vs. HSROC model	W-ML model vs. HSROC model
	Median ROR [IQR]	Median ROR [IQR]
Overall (n = 26)	0.94 [0.68, 1.04]	0.59 [0.46, 0.78]
By size of DORb
DOR < 35, n = 9	0.99 [0.88, 1.16]	0.78 [0.60, 0.95]
DOR 35–100, n = 10	1.00 [0.83, 1.05]	0.53 [0.42, 0.69]
DOR > 100, n = 7	0.62 [0.30, 0.75]	0.42 [0.16, 0.59]
By % zero cellsc
<5%, n = 10	1.01 [0.88, 1.05]	0.79 [0.53, 0.88]
5–10%, n = 8	1.02 [0.94, 1.11]	0.63 [0.50, 0.74]
>10%, n = 8	0.62 [0.30, 0.72]	0.42 [0.16, 0.63]
By range in ‘S’d
3 to <6, n = 7	0.82 [0.48, 1.03]	0.75 [0.37, 0.79]
6 to <8, n = 14	0.90 [0.68, 1.00]	0.53 [0.49, 0.75]
≥8, n = 5	1.05 [0.99, 1.16]	0.57 [0.42, 0.73]

Abbreviations: E-ML, equal-weight Moses–Littenberg; W-ML, weighted Moses–Littenberg; HSROC, hierarchical summary receiver operating characteristic; ROR, ratio of DORs between models; median ROR, ROR at the median; IQR, interquartile range in ROR from 25th to 75th percentile; DOR, diagnostic odds ratio.

^aEach Moses–Littenberg model is compared to the HSROC model (denominator).

^bThe stratification by DOR is based on the HSROC overall pooled estimate at mean threshold.

^cNumber of zero false-positive and false-negative cells as a percentage of the total number of cells per analysis.

^dBased on values for ‘S’ from Moses–Littenberg model, where S = logit(sensitivity) + logit(1 − specificity).