Projecting COVID-19 disease severity in cancer patients using purposefully-designed machine learning (preprint)

BackgroundAccurately predicting outcomes for cancer patients with COVID-19 has been clinically challenging. Numerous clinical variables have been retrospectively associated with disease severity, but the predictive value of these variables, and how multiple variables interact to increase risk, remains unclear. MethodsWe used machine learning algorithms to predict COVID-19 severity in 354 cancer patients at Memorial Sloan Kettering Cancer Center in New York City. Using clinical variables only collected on or before a patients COVID-19 positive date (time zero), we sought to classify patients into one of three possible future outcomes: Severe-early (the patient required high levels of oxygen support within 3 days of being tested positive for COVID-19), Severe-late (the patient required high levels of oxygen after 3 days), and Non-severe (the patient never required oxygen support). ResultsOur algorithm classified patients into these classes with an AUROC ranging from 70-85%, significantly outperforming prior methods and univariate analyses. Critically, classification accuracy is highest when using a potpourri of clinical variables -- including patient demographics, pre-existing diagnoses, laboratory and radiological work, and underlying cancer type -- suggesting that COVID-19 in cancer patients comes with numerous, combinatorial risk factors. ConclusionsOverall, we provide a computational tool that can identify high-risk patients early in their disease progression, which could aid in clinical decision-making and selecting treatment options.

The Effect of Neutropenia and Filgrastim (G-CSF) in Cancer Patients With COVID-19 Infection (preprint)

BackgroundNeutropenia is commonly encountered in cancer patients, and recombinant human granulocyte colony-stimulating factor (G-CSF, filgrastim) is widely given to oncology patients to counteract neutropenia and prevent infection. G-CSF is both a growth factor and cytokine that initiates proliferation and differentiation of mature granulocytes. However, the clinical impact of neutropenia and G-CSF use in cancer patients, who are also afflicted with coronavirus disease 2019 (COVID-19), remains unknown. MethodsAn observational cohort of 304 hospitalized patients with COVID-19 at Memorial Sloan Kettering Cancer Center was assembled to investigate links between concurrent neutropenia (N=55) and G-CSF administration (N=16) on COVID-19-associated respiratory failure and death. These factors were assessed as time-dependent predictors using an extended Cox model, controlling for age and underlying cancer diagnosis. To determine whether the degree of granulocyte response to G-CSF affected outcomes, a similar model was constructed with patients that received G-CSF, categorized into "high"- and "low"- response, based on the level of absolute neutrophil count (ANC) rise 24 hours after growth factor administration. ResultsNeutropenia (ANC < 1 K/mcL) during COVID-19 course was not independently associated with severe respiratory failure or death (HR: 0.71, 95% Cl: 0.34-1.50, P value: 0.367) in hospitalized COVID-19 patients. When controlling for neutropenia, G-CSF administration was associated with increased need for high oxygen supplementation and death (HR: 2.97, 95% CI: 1.06-8.28, P value: 0.038). This effect was predominantly seen in patients that exhibited a "high" response to G-CSF based on their ANC increase post-G-CSF administration (HR: 5.18, 95% CI: 1.61-16.64, P value: 0.006). ConclusionPossible risks versus benefits of G-CSF administration should be weighed in neutropenic cancer patients with COVID-19 infection, as G-CSF may lead to worsening clinical and respiratory status in this setting.