Table 3.

Performance of different SNOMED CT-based phenotype definitions, compared to the original manually curated Read V2 codelists as a gold standard

Phenotype	Method	N concepts	N CPRD medcodes	N patients	Precision (95% CI)	Recall (95% CI)	F1 score
Diabetes mellitus	Read	216	784	28 233
	Primary	152	593	24 882	0.999 (0.998–0.999)	0.880 (0.876–0.884)	0.936
	Extended	204	779	25 214	0.988 (0.987–0.990)	0.883 (0.879–0.886)	0.933
	Value Set	258	739	28 849	0.921 (0.918–0.924)	0.941 (0.938–0.944)	0.931
Asthma	Read	25	69	51 020
	Primary	34	99	52 593	0.969 (0.967–0.970)	0.999 (0.998–0.999)	0.984
	Extended	36	102	55 552	0.918 (0.916–0.920)	1.000 (1.000–1.000)	0.957
	Value Set	53	107	57 326	0.890 (0.887–0.892)	1.000 (0.999–1.000)	0.941
Heart failure	Read	38	78	9454
	Primary	27	54	8474	0.995 (0.993–0.996)	0.892 (0.885–0.898)	0.941
	Extended	30	59	8546	0.990 (0.988–0.992)	0.895 (0.888–0.901)	0.940
	Value Set	44	76	9521	0.982 (0.980–0.985)	0.989 (0.987–0.991)	0.986