Back

Tools for categorization of diagnostic codes in hospital data: Operationalizing CCSR into a patient data repository

Malecki, S.; Loffler, A.; Tamming, D.; Fralick, M.; Sohail, S.; Shi, J.; Roberts, S.; Colacci, M.; Razak, F.; Verma, A.

2022-11-30 health informatics
10.1101/2022.11.29.22282888 medRxiv
Show abstract

BackgroundThe Clinical Classification Software refined version (CCSR) is a tool to aggregate International Classification of Diseases, 10th Revision, Clinical Modification/Procedure Coding System (ICD-10-CM/PCS) diagnosis codes into clinically meaningful categories. ICD-10-CM/PCS codes are primarily used in the United States and the tool has not been optimized for use with other country-specific ICD-10 coding systems. MethodWe developed an automated procedure for mapping Canadian ICD-10 codes (ICD-10-CA) to CCSR categories using discharge diagnosis data from adult medical hospitalizations at 7 hospitals between Apr 1 2010 and Dec 31 2020, and manually validated the results. ResultsThere were 383,972 Canadian hospital admissions with 5,186 distinct ICD-10 discharge diagnosis codes. Only 46.6% of ICD-10-CA codes could be mapped directly to CCSR categories. Our algorithm improved mapping of hospital codes to CCSR categories to 98.2%. Validation of the algorithm demonstrated a high degree of accuracy with strong interrater agreement (observed proportionate agreement of 0.98). The algorithm was critical for mapping the majority of diagnosis codes associated with heart failure (96.6%), neurocognitive disorders (96.0%), skin and subcutaneous tissue infections (97.2%), and epilepsy (92.5%). ConclusionOur algorithm for operationalizing CCSR into a patient data repository (https://github.com/GEMINI-Medicine/gemini-ccsr) has been validated for use with Canadian ICD-10 codes and may be useful to clinicians and researchers from diverse geographic locations.

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.