Publications

Abstract : The leading cause of mortality around the world is cancer. World Health Organization (WHO) reported around 10 million deaths in 2020 due to cancer. Neutropenic sepsis is a life-threatening condition and an immense cause of cancer chemotherapy related mortality, which is incited by the breach in the first line of defense against microbes constituted by decrease in the absolute neutrophils count. The commonest source of infection is the alimentary system whilst bloodstream infections carry a high mortality. Antibiotics have been the mainstay of treatment in neutropenic sepsis however with rising rates of antimicrobial resistance, the effectiveness of antibiotics has decreased. Broad-spectrum antibiotics introduce Antimicrobial resistance (AMR) are commenced empirically for neutropenic sepsis while awaiting culture results and this invariably incites misuse of antimicrobials and kindles selection of resistant microbes. In the current setting, appropriate antimicrobial therapy for a bloodstream infection requires at least 2-5 days for culture results. We hypothesized that early prediction of the presence of bacterial growth in blood cultures and its characterization before culture results would be enabled by machine learning using hematological parameters. We present a machine learning approach to predict bacterial growth at least 2-5 days before actual culture results. Our study involves 231 cancer patients with their blood lab test values, and inflammatory markers included as features for the prediction of bacterial growth in neutropenic patients. Different machine learning models were used for prediction in which decision tree and random forest demonstrate 66 percent accuracy in the prediction of bacterial growth in neutropenic patients.