Evaluation of methods to identify chemotherapy‐related hematologic toxicities in real‐world settings

Abstract Hematologic toxicities are common in those receiving chemotherapy and can affect subsequent treatment delivery and patient outcomes. Laboratory results and International Classification of Diseases (ICD) codes are often used to define toxicities in real‐world settings; optimal use of this information for research remains unclear. In 854 women receiving adjuvant chemotherapy for stage I–IIIA breast cancer with abstracted medical record data at Kaiser Permanente Washington, we defined three chemotherapy‐related hematologic toxicities (anemia, neutropenia, and thrombocytopenia) using laboratory data, ICD codes, and a combination of both during and up to 1 month after chemotherapy. Cutpoints for laboratory values corresponding to grades 1–3 were derived from Common Terminology Criteria for Adverse Events. We examined sensitivity, specificity, and positive predictive value (PPV), comparing laboratory and ICD definitions to chart review. Chart review identified 180 women with anemia, 168 with neutropenia, and 21 with thrombocytopenia. Grade‐1+ showed the highest sensitivity for all toxicities (range: 0.84–0.95) but lowest specificity (range: 0.45–0.86) and PPV (range: 0.14–0.47). Grade‐3+ showed the lowest sensitivity for all toxicities (range: 0.13–0.59) and highest specificity (range: 0.85–1.00) and PPV (range: 0.63–0.82). ICD and grade‐2+ each showed balanced performance across toxicities, with ICD having slightly higher sensitivity (range: 0.62–0.76) than grade‐2+ (range: 0.48–0.77). Combining grade‐2+ and ICD improved sensitivity (range: 0.76–0.93) for all toxicities without much loss of specificity (range: 0.64–0.98), and no change to PPV. In conclusion, ICD and grade‐2+ showed moderate‐to‐good performance for defining hematologic toxicities; the combination of laboratory data (especially grade‐2+) and ICD showed better performance than either source alone.