Document Type

Conference Proceeding

Publication Date

9-1-2022

Publication Title

Ann Oncol

Abstract

Background: Patients with cancer are at higher risk of dying of COVID-19. Known risk factors for 30-day all-cause mortality (ACM-30) in patients with cancer are older age, sex, smoking status, performance status, obesity, and co-morbidities. We hypothesized that common clinical and laboratory parameters would be predictive of a higher risk of 30-day ACM, and that a machine learning approach (random forest) could produce high accuracy.

Methods: In this multi-institutional COVID-19 and Cancer Consortium (CCC19) registry study, 12,661 patients enrolled between March 17, 2020 and December 31, 2021 were utilized to develop and validate a model of ACM-30. ACM-30 was defined as death from any cause within 30 days of COVID-19 diagnosis. Pre-specified variables were: age, sex, race, smoking status, ECOG performance status (PS), timing of cancer treatment relative to COVID19 diagnosis, severity of COVID19, type of cancer, and other laboratory measurements. Missing variables were imputed using random forest proximity. Random forest was utilized to model ACM-30. The area under the curve (AUC) was computed as a measure of predictive accuracy with out-of-bag prediction. One hundred bootstrapped samples were used to obtain the standard error of the AUC.

Results: The median age at COVID-19 diagnosis was 65 years, 53% were female, 18% were Hispanic, and 16.7% were Black. Over half were never smokers and the median body mass index was 28.2. Random forest with under sampling selected 20 factors prognostic of ACM-30. The AUC was 88.9 (95% CI 88.5-89.2). Highly informative parameters included: COVID-19 severity at presentation, cancer status, age, troponin level, ECOG PS and body mass index.

Conclusions: This prognostic model based on readily available clinical and laboratory values can be used to estimate individual survival probability within 30-days for COVID-19. In addition, this model can be used to select or classify patients with cancer and COVID-19 into risk groups based on validated cut points, for treatment selection, prophylaxis prioritization, and/or enrollment in clinical trials. Future work includes external validation using other large datasets of patients with COVID-19 and cancer.

Comments

Clinical trial identification: NCT04354701.

Volume

33

First Page

S771

Last Page

S772

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.