Table 2.

Simulated deployment results.

Model	Risk level	Sensitivity	Specificity	PPV (TrxPop. Est.)	SIR (TrxPop. Est.)
PrismNN	1	54.6% (53.4–55.8)	85.3% (85.1–85.5)	0.28% (0.27–0.29)	2.38 (2.34–2.41)
	2	48.7% (47.5–49.9)	89.1% (89.0–89.3)	0.34% (0.33–0.35)	2.87 (2.82–2.91)
	3	35.9% (34.8–37.1)	95.3% (95.2–95.4)	0.58% (0.56–0.60)	5.10 (5.02–5.18)
	4	31.4% (30.4–32.5)	96.9% (96.9–97.0)	0.77% (0.74–0.80)	7.07 (6.96–7.17)
	5	17.3% (16.4–18.2)	99.5% (99.5–99.6)	2.81% (2.59–3.06)	29.0 (28.6–29.5)
	6	11.9% (11.2–12.7)	99.9% (99.9–99.9)	8.62% (7.42–10.0)	96.0 (94.3–97.6)
PrismLR	1	52.3% (51.1–53.4)	86.2% (86.1–86.4)	0.29% (0.28–0.29)	2.22 (2.19–2.25)
	2	46.4% (45.2–47.6)	89.9% (89.7–90.0)	0.35% (0.34–0.36)	2.66 (2.61–2.70)
	3	31.8% (30.7–32.9)	95.4% (95.3–95.5)	0.53% (0.50–0.55)	4.06 (3.99–4.12)
	4	26.4% (25.3–27.4)	97.0% (96.9–97.1)	0.66% (0.63–0.69)	5.17 (5.09–5.25)
	5	8.95% (8.30–9.64)	99.6% (99.5–99.6)	1.51% (1.36–1.67)	13.1 (12.9–13.3)
	6	2.93% (2.55–3.35)	99.9% (99.9–99.9)	2.88% (2.32–3.56)	24.2 (23.7–24.7)

Numbers in brackets are 95% CI.

PPV: Positive Predictive Value.

SIR: Standardised Incidence Ratio.

TrxPop. Est.: Estimation on the whole TriNetX population that accounts for unbalanced sampling.

All calculations were based on the outcome during the followup period of individual patients. Followup period was determined by the age and EHR data availability of each patient. More details can be found in Section 2.5 and in the Supplemental Material.