Table 3.

This table shows AUROC scores from the various models.

	Baselines		Statistical Models				Causal Models
Adverse event type [+ ctrls, − ctrls]	ROR	χ 2	${\beta }_{sql}^5$	${\beta }_{psi}^5$	${\beta }_{sql}^{10}$	${\beta }_{psi}^{10}$	${\mathrm{\Delta }}_{sql}^5$	${\mathrm{\Delta }}_{psi}^5$	${\mathrm{\Delta }}_{sql}^{10}$	${\mathrm{\Delta }}_{psi}^{10}$
Acute kidney failure [10+, 9−]	0.6222	0.4667	0.6	0.7444	0.5667	0.6222	0.5	0.8111 †	0.5111	0.5
Acute liver failure [43+, 11−]	0.5835	0.5814	0.6004	0.6723	0.4905	0.6765 †	0.4926	0.5243	0.5433	0.5624
Acute myocardial infarction [15+, 23−]	0.6261	0.6754	0.658	0.5159	0.4783	0.5217	0.6	0.6725	0.6667	0.6464
Gastrointestinal hemorrhage [20+, 32−]	0.5688	0.6	0.5672	0.7078	0.5688	0.7156 †	0.5484	0.55	0.5203	0.4891
Weighted overall AUROC [88+, 75−]	0.5933	0.5959	0.6032	0.6556 †	0.5363	0.6005	0.5215	0.6466	0.561	0.5513

χ² = chi squared. β = coefficient from multiple variable logistic regression. Δ = average treatment effect using graphical causal models. Results exceeding best baseline performance are in boldface.

^†indicates best performance for a side effect.