Table 2.

Best performing medication scores evaluated in three separate validation datasets by C-index and Negelkerke R².

Score derivation method	Number of medications included	Main cohort FINRISK validation subset		Replication cohorts
				R2		C-index ∆C baseline
		R2	C-index (95% CI) ∆C baseline (95% CI)	H2000	Estonia	H2000	Estonia
Baseline model (age and sex)	–	0.134	0.766 (0.745–0.788)	0.309	0.098	0.823	0.775
Expert consensus panel	166	0.149	0.779 (0.758–0.800)	0.316	0.107	0.827	0.786
			+ 0.013 (0.006–0.019)			+ 0.004	+ 0.011
Number of medications	301	0.148	0.777 (0.756–0.798)	0.317	0.104	0.828	0.785
			+ 0.011 (0.005–0.017)			+ 0.005	+ 0.010
Shotgun stochastic search	8	0.158	0.785 (0.764–0.806)	0.317	0.102	0.827	0.781
			+ 0.019 (0.01–0.027)			+ 0.004	+ 0.006
Cox LASSO	65	0.164	0.791 (0.771–0.812)	0.324	0.109	0.830	0.793
			+ 0.025 (0.017–0.033)			+ 0.007	+ 0.018
Cox LASSO signs only	65	0.162	0.791 (0.770–0.811)	0.320	0.110	0.829	0.791
			+ 0.025 (0.017–0.032)			+ 0.006	+ 0.016
Ridge regression	301	0.164	0.793 (0.772–0.813)	0.322	0.113	0.830	0.795
			+ 0.026 (0.018–0.034)			+ 0.007	+ 0.020
Elastic net	214	0.166	0.794 (0.773–0.814)	0.324	0.108	0.830	0.784
			+ 0.028 (0.019–0.036)			+ 0.007	+ 0.009

H2000 does not include all the ATC codes of the derivation set.