Back

Manual Validation of an Algorithm for Identifying Osteoarthritis within the Centralized Interactive Phenomics Resource of the Million Veteran Program

Nguyen, C. T.; Richman, J. S.; Chiles, J. W.; Singh, J. A.; McDonald, M.-L. N.

2025-05-13 rheumatology
10.1101/2025.05.12.25327449 medRxiv
Show abstract

ObjectiveThis study aimed to validate an osteoarthritis (OA) phenotyping algorithm within the Million Veteran Program (MVP) using the United States (US) Department of Veterans Affairs (VA) Centralized Interactive Phenomics Resource (CIPHER). MethodsA random sample of 213 veterans was analyzed sing ICD-9-CM/ICD-10-CM codes from a previously published algorithm (PMID:29559693). OA cases required two OA codes at least 30 days apart, while controls were excluded based on codes for conditions more common in OA patients, such as chondrocalcinosis and crystal arthropathies. Manual chart reviews identified documented OA mentions and joint replacements. Cohens kappa statistic assessed agreement. Discrepancies between chart data and coding were re-evaluated through re-abstraction. ResultsAmong 213 veterans, 174 (82%) had chart-documented OA. Agreement between chart review and code-based general OA identification was moderate (kappa = 0.47). Joint-specific agreement was substantial for knee (kappa = 0.63) and hip OA (kappa = 0.59), but lower for spine (kappa = 0.16) and hand (kappa = 0.34). Agreement was high for hip (kappa = 0.86) and knee replacements (kappa = 0.69). The McNemar test showed significant asymmetry for general OA, hand OA, and thumb OA, indicating discrepancies between chart and coded data. No significant asymmetry was found for knee and hip OA, supporting better alignment. ConclusionsThis study supports the validity of the OA phenotyping algorithm using the VA database for identifying OA. The variability in identifying milder cases highlights the need for refined phenotyping algorithms and standardized diagnostic protocols to improve OA detection and personalized care for veterans.

Matching journals

The top 2 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.