Artificial intelligence methods to predict chemotherapy-induced neutropenia in breast cancer patients.

Abstract

6555 Background: While chemotherapy improves outcomes in breast cancer patients, it also increases the risk of neutropenia. There is a need for improved risk prediction models of chemotherapy-induced neutropenia. This study developed an artificial intelligence (AI) model to predict neutropenia risk within six months of chemotherapy and compared it to traditional logistic regression models. Methods: We obtained a cohort of 10,288 breast cancer patients from the ASCO CancerLinQ Discovery™ dataset, who were treated with doxorubicin and cyclophosphamide (AC) followed by paclitaxel (T) with or without pre-chemotherapy WBC growth factor prophylaxis. We created a hierarchy of predictors, then trained and evaluated a neural network algorithm. Results: Using only the relevant predictors for this more specific study, we demonstrate an average 27% increase across the scenarios in AUC ROC (p < 0.001) compared to existing studies (Lyman, 2011). We improve prediction by an average of 45%, achieving 0.56 positive predictive value (PPV) and 0.92 negative predictive value (NPV) prior to chemotherapy. Conclusions: These results demonstrate that this larger dataset combined with an AI algorithm enabled substantial improvement in prediction of chemotherapy-induced neutropenia. In the clinical setting, this would improve decisions, and enable early intervention for patients that would benefit most from prophylaxis, thus reducing neutropenic fever and infections requiring hospitalization. Clinical Decision Point Cohort Predictive Models Statistics Logistic Regression (Lyman, 2011) Logistic Regression (Current Study) PH.AI Neural Net (Current Study) N Neutropenia incidence within 6 mo. Stage Distribution (I,II,III,IV) Median time to Neutropenia (days) AUC-ROC # Predictors AUC-ROC # Predictors AUC-ROC # Predictors Prior to Starting AC with Prophylaxis 7097 .017 .14, .50, .31, .04 18 .58 ± .03 7 .73 ± .02 19 .78 ± .04 704 no Prophylaxis 1075 .093 .13, .51, .31, .05 22 .59 ± .03 7 .69 ± .02 15 .74 ± .01 2169 Prior to Starting T with Prophylaxis 6753 .084 .14, .50, .31, .04 31 .59 ± .03 7 .76 ± .02 21 .74 ± .01 2130 no Prophylaxis 1078 .074 .13, .53, .29, .05 15 .57 ± .02 7 .74 ± .02 16 .89 ± .02 873