Table 3. Logistic regression results for the Cleveland Clinic data when all four datasets were used for estimation (Cleveland Clinic, Swiss, Hungarian and Veterans).

Variable	Original	modgo
	Validation datasets	As single dataset	Multicenter approach
Exercise-induced angina (Yes)	3.80	4.37 (3.13–6.10)	3.83 (2.66–5.73)
Age ( y ears)	1.03	1.03 (1.02–1.05)	1.03 (1.01–1.05)
Max heart rate (bpm)	0.98	0.98 (0.97–0.98)	0.98 (0.97–0.98)
ST depression (mm)	1.70	1.52 (1.29–1.78)	1.78 (1.52–2.12)

Displayed are odds ratios for the original datasets (column 2). Two simulation approaches were used with modgo, and medians of odds ratios and empirical 2.5 and 97.5% quantiles (in parenthesis) are displayed in the last two columns. Column 3 the multicenter nature of the data was ignored, and data from all centers were pooled for simulating new data; column 4: the multicenter nature of the data was taken into account. Correlation matrices were estimated per center, data were simulated per center and pooled before running logistic regression.