Predictive model of chemotherapy-related toxicity in elderly Chinese cancer patients

Purpose: Older cancer patients are more likely to develop and die from chemotherapy-related toxicity. However, evidence on drug safety and optimal effective doses is relatively limited in this group. The aim of this study was to develop a tool to identify elderly patients vulnerable to chemotherapy toxicity. Patients and methods: Elderly cancer patients ≥60 years old who visited the oncology department of Peking Union Medical College Hospital between 2008 and 2012 were included. Each round of chemotherapy was regarded as a separate case. Clinical factors included age, gender, physical status, chemotherapy regimen and laboratory tests results were recorded. Severe (grade ≥3) chemotherapy-related toxicity of each case was captured according to the National Cancer Institute Common Terminology Criteria for Adverse Events, version 5.0. Univariate analysis was performed by chi-square statistics to determine which factors were significantly associated with severe chemotherapy toxicity. Logistic regression was used to build the predictive model. The prediction model was validated by calculating the area under the curve of receiver operating characteristic (ROC). Results: A total of 253 patients and 1,770 cases were included. The average age of the patients was 68.9 years. The incidence of grade 3-5 adverse events was 24.17%. Cancer type (non-GI cancers), BMI<20 kg/m², KPS<90%, severe comorbidity, polychemotherapy, standard dose chemotherapy, low white blood cells count, anemia, low platelet cells count, low creatine level and hypoalbuminemia were associated with severe chemotherapy-related toxicity. We used these factors to construct a chemotherapy toxicity prediction model and the area under the ROC curve was 0.723 (95% CI, 0.687-0.759). Risk of toxicity increased with higher risk score (11.98% low, 31.51% medium, 70.83% high risk; p < 0.001). Conclusion: We constructed a predictive model of chemotherapy toxicity in elderly cancer patients based on a Chinese population. The model can be used to guide clinicians to identify vulnerable population and adjust treatment regimens accordingly.