TABLE 6.

Estimated regression coefficients of seven models for predicting DVT estimated with (S)IECV

Predictor	None	A M	$A^{{\text{RE}}_1}$	$A^{{\text{RE}}_{1/2}}$	$A^{{\text{RE}}_0}$	A SD	A Gini
Intercept	−2.17	−3.54	−3.54	−5.13	−5.00	−5.13	−3.99
Malignancy	0.98	0.76	0.76	1.64	1.68	1.64	2.05
Calf difference	1.26	1.13	1.13	1.38	1.34	1.38	1.07
Surgery	0.55	−0.04	−0.04	0.25	0.25	0.25	0.34
D‐dimer positive		2.76	2.76	2.99	2.94	2.99	2.81
Age/25		−0.22	−0.22
Vein distension		0.46	0.46
Surgery × no leg trauma		0.68	0.68
No leg trauma				0.95	0.96	0.95
(Age/25)2				−0.02		−0.02
Male				0.32	0.36	0.32	0.52
D‐dimer positive × male				−0.20	−0.24	−0.2	−0.28
Malignancy × age/25				−0.32	−0.35	−0.32	−0.50

Note: Malignancy: history of malignancy; Calf difference: difference in circumference of calves ≥ 3 cm; Surgery: recent surgery; Age/25: age divided by 25; Duration: duration of symptoms; None: model with no predictor selection. Empty cells indicate the predictor was not selected for inclusion in the corresponding model. The predictor functions that were not selected in any final model are omitted from this table. Summary predictor effects were estimated by the Dersimonian and Laird method, as REML did not converge for the estimation of some models. Although REML has better theoretical properties for the heterogeneity estimate, the difference for the summary effects (presented here) is limited.

Abbreviations: A ^Gini, Gini's mean difference of MSE; A ^M, average of MSE; $A^{{\text{RE}}_1}$, random effects meta‐analytic estimate of mean of MSE; $A^{{\text{RE}}_0}$, random effects meta‐analytic estimate of heterogeneity of distribution of MSE; $A^{{\text{RE}}_{1/2}}$, sum of random effects meta‐analytic estimates of mean and heterogeneity of distribution of MSE; A ^SD, standard deviation of MSE; DVT, deep vein thrombosis; MSE, mean square error; (S)IECV, (stepwise) internal‐external cross‐validation.