Bias in artificial intelligence algorithms and recommendations for mitigation.

The adoption of artificial intelligence (AI) algorithms is rapidly increasing in healthcare. Such algorithms may be shaped by various factors such as social determinants of health that can influence health outcomes. While AI algorithms have been proposed as a tool to expand the reach of quality healthcare to underserved communities and improve health equity, recent literature has raised concerns about the propagation of biases and healthcare disparities through implementation of these algorithms. Thus, it is critical to understand the sources of bias inherent in AI-based algorithms. This review aims to highlight the potential sources of bias within each step of developing AI algorithms in healthcare, starting from framing the problem, data collection, preprocessing, development, and validation, as well as their full implementation. For each of these steps, we also discuss strategies to mitigate the bias and disparities. A checklist was developed with recommendations for reducing bias during the development and implementation stages. It is important for developers and users of AI-based algorithms to keep these important considerations in mind to advance health equity for all populations.

Validation of the CSRFENCE score for prediction of febrile neutropenia during chemotherapy cycles 2-6.

PURPOSE: Though febrile neutropenia (FN) risk prediction models are important in clinical practice, their external validation is limited. In this study, we validated the Cycle-Specific Risk of FEbrile Neutropenia after ChEmotherapy (CSRFENCE) score for predicting FN. METHODS: We reviewed the medical records of patients with solid malignancies and diffuse large B-cell lymphoma during chemotherapy cycles 2-6 and recorded if patients developed FN, defined as absolute neutrophil counts less than 500 cells/microL with fever more than or equal to 38.2 â„ƒ. The CSRFENCE score was determined by adding the risk factors' coefficients described by the original study; subsequently, the score was used to classify chemotherapy cycles into the following risk groups for developing FN: low, intermediate, high, and very high risk. The discriminatory ability of the score was assessed using area under the receiver operating characteristics curve (AUROCC) and incidence rate ratios (IRR) within each CSRFENCE risk group. RESULTS: We analyzed 2870 chemotherapy cycles, of which 42 (1.5%) were associated with FN. Among those, 3 (7.1%), 14 (33.3%), 5 (12%), and 20 (47.6%) were classified as low, intermediate, high, and very high risk for developing FN, respectively. The AUROCC was 0.72 (95% CI 0.64-0.81). Compared with the low risk group (n = 666), the IRR of developing FN was 1.01 (95% CI 0.15-43.37), 0.69 (95% CI 0.08-32.46) and 1.17 (95% CI 0.17-49.49) in the intermediate (n = 1431), high (n = 498) and very high (n = 275) risk groups, respectively. CONCLUSION: The CSRFENCE model can moderately stratify patients into four risk groups for predicting FN prior to chemotherapy cycles 2-6.

Validation of the FENCE Risk Groups for Prediction of Febrile Neutropenia With First-Cycle Chemotherapy.

BACKGROUND: The FEbrile Neutropenia after ChEmotherapy (FENCE) score was developed to estimate the risk of febrile neutropenia (FN) at first cycle of chemotherapy but has not been externally validated. We aimed to validate the FENCE score based on its risk groups in patients treated at a comprehensive cancer center. METHODS: We conducted a retrospective study of treatment-naïve adult patients with solid tumors and diffuse large B-cell lymphoma who received first-cycle chemotherapy between January and November 2019. Patients were followed until the second cycle of chemotherapy to identify any FN events (neutrophil count <0.5 × 109/L with fever ≥38.2°C). The FENCE score was determined and patients classified as low, intermediate, high, and very high risk. The discriminatory ability of classifying patients into FENCE risk groups was calculated as the area under the receiver operating characteristics curve and incidence rate ratios within each FENCE risk group. RESULTS: FN was documented during the first cycle of chemotherapy in 45 of the 918 patients included (5%). The area under the receiver operating characteristics curve was 0.66 (95% confidence interval [CI] = 0.58 to 0.73). Compared with the low-risk group (n = 285), the incidence rate ratio of developing FN was 1.58 (95% CI = 0.54 to 5.21), 3.16 (95% CI = 1.09 to 10.25), and 3.93 (95% CI = 1.46 to 12.27) in the intermediate (n = 293), high (n = 162), and very high (n = 178) risk groups, respectively. CONCLUSIONS: In this study, classifying patients into FENCE risk groups demonstrated moderate discriminatory ability for predicting FN. Further validation in multicenter studies is necessary to determine its generalizability.