https://meetinglibrary.asco.org/record/113232/abstract
Authors:
Liede, A, Hernandez, RK, Roth, M, Calkins, G, Larrabee, K, Nicacio, LV
Background:The accuracy of bone metastases coding is unknown for most large healthcare databases commonly used for epidemiologic research. EMR hold promise in cancer research in richness of data, but often clinically-relevant data elements are found in free-text unstructured data fields. We examined validity of bone metastases coding in structured EMR and claims data, relative to a gold standard data set based on the complete (structured and unstructured) EMR using a technology-enabled abstraction process.Methods:Breast cancer patients with ≥ 1 visit on or after Nov 18, 2010 were identified from 3 community oncology practices in the US. We calculated sensitivity (Se), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV) of bone metastases ICD-9 code 198.5. We evaluated effects of misclassification when studying the proportion of patients treated with bone-targeting agents (BTA: zoledronic acid, pamidronate, denosumab), BTA timing relative to diagnosis, and skeletal complications.Results:Among 8796 breast cancer patients, 524 were confirmed to have bone metastases using the technology-enabled abstraction process. Structured data and claims missed 17% of bone metastases (89 of 524); Se was 67% based on structured EMR data fields (Sp 98%, PPV 71%, NPV 98%), and 77% based on claims (Sp 98%, PPV 72%, NPV 99%). Combining structured EMR and claims increased Se to 83% (Sp 98%, PPV 70%, NPV 99%). False negatives led to an overestimation of the proportion treated with a BTA or with a skeletal complication (Table). Date of diagnosis was delayed in structured data (median 32 days) and claims (median 43 days) compared to technology-assisted EMR.Conclusions:Misclassification and low sensitivity of bone metastases can lead to underestimation of absolute risks, dilution of associations, or residual confounding. Technology-enabled chart abstraction of the structured and unstructured EMR can greatly improve data quality and minimize bias.
Sources:
ASCO Annual Meeting