A numerical similarity approach for using retired Current Procedural Terminology (CPT) codes for electronic phenotyping in the Scalable Collaborative Infrastructure for a Learning Health System (SCILHS)

Table 3 Analysis of the miscategorization of our method for placing deprecated CPT codes in the hierarchy

Miscategorization Analysis of Retired CPT Code Placement
	Validation set		Remaining codes
Code movement (correction) by human reviewer	Moved for optimal placement	Moved due to miscategorization	Moved for optimal placement	Moved due to miscategorization
Deeper	51	0	0	0
Higher	4	4	2	2
Parallel folder	17	5	2	2
Distant	10	5	7	3
Total codes	171		402
Precision	52 %	92 %	97 %	98 %
Avg. precision	52.0 % (optimal placement)/96.4 % (due to miscategorization)

The table body shows how the human reviewer moved codes to correct miscategorizations. The first pair of columns is for the validation set, and the second pair covers all remaining codes. (These are shown separately because the validation set was subjected to a more rigorous analysis of optimal placement.) Moved for Optimal Placement includes all codes moved by the reviewer. Moved due to Miscategorization includes codes moved due to a true miscategorization, according to the secondary analysis. Total codes is all codes in each set (validation and remaining). Precision is the precision of the automated method for each subset of data. The final line shows the average precision in two cases: considering all moved codes (optimal placement) and considering errors only. The precision for optimal placement of remaining codes is not included in the averages; the second reviewer was not told to move codes for optimal placement, so this result would be invalid

ISSN: 1472-6947