Table 3.

Performance comparison of the baseline and machine learning–optimized matching configurations in SantéMPI in the held-out evaluation sets, for the detection of definite linkages not needing manual review.

Data set	Sensitivity (%; 95% CI)	Patients, n	Correctly predicted linkages (ground-truth linkages)
FEBRL1a	98.0 (95.0-100.0)	100	98 (100)
FEBRL2	96.6 (93.9-98.6)	800	196 (203)
FEBRL3	94.9 (93.0-96.5)	400	558 (588)
FEBRL4	98.3 (97.5-99.1)	1000	983 (1000)
Hawaii	96.6 (93.7-98.9)	200	168 (174)

^aFEBRL: Freely Extensible Biomedical Record Linkage.