Increasing the efficiency of trial-patient matching: automated clinical trial eligibility Pre-screening for pediatric oncology patients

Table 1 The performance of the demographics-based filter (baseline) and the EHR-based ES algorithms

Trial-centered Patient Cohort Identification
Algorithm	Evaluation Metrics
Algorithm	WL	95% CI	P[%]	Sp[%]	PV*
Demographics-based Filter	163	149-179	1.9	24.3	8.30E-21
DX/ICD-9	50	35-64	6.20	78.1	5.27E-4
NOTE	28	16-41	10.7	87.9	7.75E-2
DX/ICD-9+NOTE	24	14-35	12.6	89.9	N/A
Patient-centered Trial Recommendation
Algorithm	Evaluation Metrics
	Sub-population case (127 patients)					Full-population case (215 patients)
	WL	95% CI	P[%]	Sp[%]	PV*	WL	95% CI	P[%]	Sp[%]	PV*
Demographics-based Filter	42	40-43	3.2	25.5	1.7E-143	42	40-43	1.9	24.3	1.5E-39
DX/ICD-9	8	6-10	16.8	87.8	2.36E-7	22	19-25	3.6	60.7	3.85E-7
NOTE	4	3-5	33.1	95.0	2.54E-2	20	17-23	3.9	64.9	2.54E-2
DX/ICD-9+NOTE	3	3-4	35.7	95.5	N/A	19	17-22	4.0	65.5	N/A

DX/ICD9 indicates ES algorithm using only structured diagnoses and ICD-9 codes; NOTE, ES algorithm using only clinical notes; DX/ICD-9+NOTE, ES algorithm using both structured data and clinical notes.
WL indicates workload; CI, confidence interval; P, precision and Sp, specificity, PV, p-value.
*P-values were calculated by comparing the workload between DX/ICD-9+NOTE with the other algorithms.
N/A indicates that the performances between the two algorithms are identical and no p-value is returned.

ISSN: 1472-6947