Next generation phenotyping using narrative reports in a rare disease clinical data warehouse

Table 4 Description and evaluation of the 6 sets of patients

Sets	RETT	DOCK8 deficiency	LOWE	SILVER RUSSELL	BARDET BIEDL	APDS 1 and 2
Median age at visit (years)	8.2 [4.8–12.6]	11.4 [9.3–14.1]	12.8 [5.8–20.3]	2.4 [0.8–5.4]	15.7 [10.1–41.5]	12.8 [7.7–18.6]
Median follow up (years)	2.6 [0–4.9]	3.1 [0.3–9]	6.6 [3–10.3]	2 [0.8–4.7]	2 [0.1–6.6]	7.5 [4.8–8.6]
# Patients	209	15	23	50	53	23
# Documents	5034	3296	1325	1133	1317	2337
Phenotypes extracted, not negated and in patient context
# Phenotypes	18,538	6886	5281	6563	6345	9716
# distinct Phenotypes	1022	706	577	738	801	710
Evaluation by experts in the Top50 phenotypes
Medical Experts	NBB	CP	RS	JA	RS	NM
# Phenotypes ranked by Freq	31	36	36	16	17	39
# Phenotypes ranked by TF-IDF	38	37	41	11	12	37
# Phenotypes Freq union TF-IDF	42	52	50	16	19	52
# Phenotypes Freq intersect TF-IDF	28	22	28	11	11	25
Average Precision, ranked by Freq	0.86	0.91	0.88	0.55	0.66	0.83
Average Precision, ranked by TF-IDF	0.91	0.84	0.90	0.49	0.52	0.83

ISSN: 1750-1172