Preliminary Development and Evaluation of an Algorithm to Identify Breast Cancer Chemotherapy Toxicities Using Electronic Medical Records and Administrative Data

Abstract

Purpose: Breast cancer chemotherapy toxicity is not well documented outside of randomized trials. We developed and conducted preliminary evaluation of an algorithm to detect grade 3 and 4 toxicities using electronic data from a large integrated managed care organization.

Methods: The algorithm used administrative, pharmacy, and electronic data from outpatient, emergency room, and inpatient records of 99 women diagnosed with breast cancer from 2006 to 2009 who underwent chemotherapy. Data were abstracted for 12 months post-treatment initiation (24 months for trastuzumab recipients). An oncology nurse independently blindly reviewed records; these results were the "gold standard." Sensitivity and specificity were calculated for overall toxicity, categories of toxicities, and toxicity by age or regimen. The algorithm was applied to an independent sample of 1,575 patients with breast cancer diagnosed during the study period to estimate prevalence rates.

Results: The overall sensitivity for detecting chemotherapy-related toxicity was 89% (95% CI, 77% to 95%). The highest sensitivity was for identification of hematologic toxicities (97%; 95% CI, 84% to 99%). There were good sensitivities for infectious toxicity, but rates dropped for GI and neurological toxicities. Specificity was high within each category (89% to 99%), but when combined to measure any toxicity, it was lower (70%; 95% CI, 57% to 81%). When applied to an independent chemotherapy sample, the algorithm estimates a 26% rate of hematologic toxicity; rates were higher among patients age ≥ 65 years versus less than 65 years.

Conclusions: If validated in other samples and health care settings, algorithms to capture toxicity could be useful in comparative and cost-effectiveness evaluations of community practice-delivered treatment.