Development of a machine learning model to predict mild cognitive impairment using natural language processing in the absence of screening

Table 6 Prediction model performance characteristics in each population at various cutoffs for probability of correct classification

Cohort	Cutoff^a	Sensitivity	Specificity	PPV	NPV	F1 Score
ACT + Gen. Pop. training	0.3	0.95	0.17	0.46	0.83	0.62
	0.4	0.63	0.69	0.6	0.71	0.61
	0.5	0.37	0.9	0.73	0.66	0.49
	0.6	0.24	0.96	0.81	0.63	0.37
ACT training	0.3	0.99	0.04	0.51	0.76	0.67
	0.4	0.75	0.52	0.61	0.68	0.67
	0.5	0.49	0.82	0.73	0.62	0.59
	0.6	0.34	0.92	0.80	0.58	0.48
Gen. Pop. training	0.3	0.88	0.31	0.38	0.84	0.53
	0.4	0.37	0.86	0.57	0.74	0.45
	0.5	0.11	0.98	0.75	0.70	0.19
	0.6	0.02	1.00	1.00	0.68	0.04
Gen. Pop. validation	0.3	0.87	0.32	0.35	0.85	0.50
	0.4	0.32	0.88	0.53	0.75	0.40
	0.5	0.09	0.97	0.56	0.71	0.16
	0.6	0.02	1.00	0.70	0.70	0.04

^aCutoffs are the various probabilities that the researcher or health system would choose as a threshold to classify someone as “positive” for MCI

ISSN: 1472-6947