Table 2.

Results of the HF subtype validation and calculated epidemiological measures

Category	HF based on clinician’s diagnosis					Calculated epidemiological measures
Algorithm-based HF	HF, all	HFrEF	HFmrEF	HFpEF	No HF	Sensitivity	Specificity	PPV	NPV	Accuracy
HF, all	63	18	17	28	25	100%	32%	72%	100%	75%
HFrEF	17	17	0	0	0	94%	100%	100%	99%	99%
HFmrEF	19	1	17	1	1	100%	96%	85%	100%	97%
HFpEF	27	0	0	27	24	96%	67%	53%	98%	75%
No HF	12	0	0	0	12	32%	100%	100%	72%	75%

The algorithm defined ‘No HF’ as EF ≥ 50% and proBNP ≤ 125 ng/ml, ‘HF’ as any HF subtype present, ‘HFrEF’ as EF < 40%, ‘HFmrEF’ as EF 40–49% and ‘HFpEF’ as EF ≥ 50% and proBNP ≥ 125 ng/ml, using EF values mined by the algorithm. The clinician made the diagnosis of HF subtype based on EHR text, EF report and proBNP values

Abbreviations: EF, ejection fraction; HF, heart failure; HFrEF, HF with reduced EF; HFmrEF, HF with mildly reduced EF; HFpEF, HF with preserved EF; PPV, positive predictive value; NPV, negative predictive value; proBNP, N-terminal pro b-type natriuretic peptide