• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

28-Day Mortality Predicted by Machine Learning in Patients With CRC

News
Article

Critically ill older patients with colorectal cancer (CRC) had their 28-day mortality predicted using machine learning algorithms.

Machine learning algorithms were able to be used to predict the 28-day mortality of critical ill older patients with colorectal cancer (CRC). The study published in Journal of International Medical Research found that the models had superior predictive performance compared with conventional logistic regression models.

Cancer is the second most common cause of death around the world, with many at risk of life-threatening complications. CRC is a common malignant cancer in older patients, who are contributing to the increased number of cases. In addition, with more comorbidities in older patients, it can be harder to manage the cancer compared with younger patients. This study aimed to use machine learning methods to get clinical data that would help accurately predict mortality.

Healthy large intestine anatomy on doctor hands | Image credit: Orawan - stock.adobe.com

Healthy large intestine anatomy on doctor hands | Image credit: Orawan - stock.adobe.com

The eICU Collaborative Research Database was used to extract data for the training cohort from 200,000 intensive care unit (ICU) admissions across the United States. Data on the validation cohort came from the Medica Information Mart for Intensive Care (MIMIC) database on 50,000 patients in the ICU between 2008 and 2019 at Beth Israel Deaconess Medical Center in Boston, Massachusetts, and a cohort of 95 older patients from Wuhan Union Hospital.

Patients were excluded if they were younger than 60 years, had repeated admissions to the ICU, had missing data, or had an ICU stay of less than 24 hours. Demographic information was collected for every participant. There were 5 machine learning approaches and a conventional logistic model that were used to get the 28-day mortality prediction.

There were 693 patients who were from the eICU cohort, 181 patients from the MIMIC cohort, and 95 from the Wuhan cohort included in the study. The Wuhan cohort had a significantly lower mean (SD) age (66.9 [5.9] vs 75.1 [9.0] in eICU and 74.4 [9.0] years in MIMIC). The Wuhan cohort was 46% male compared with 60% male in the eICU and 64% male in the MIMIC cohorts. Corresponding proportions of patients on mechanical ventilation were 64% vs 30% vs 33%. Thirteen percent in the eICU cohort, 16% in the MIMIC cohort, and 6% in the Wuhan cohort died.

Of 6 machine learning models used, the best predictive ability came from the ensemble model (area under curve [AUC], 0.86) in the eICU cohort. This was followed by the random forest (AUC, 0.83) and LightGBM (AUC, 0.82). Predictive performance was worst in the logistic regression model (AUC, 0.68). The MIMIC cohort had similar results, with the ensemble model having an AUC of 0.73. The decision tree model had the worst predictive performance (AUC, 0.57). The Wuhan cohort was used to verify the prediction models, and this showed that the ensemble model still had the highest predictive ability (AUC, 0.81) with the logistic regression model having the worst (AUC, 0.65).

The top most influential features in 28-day mortality were blood urea nitrogen, vasopressors, and Charlson comorbidity index in the random forest model; serum albumin, hemoglobin, and alkaline phosphatase in the LightGBM model; and vasopressors, serum albumin, and blood urea nitrogen in the XGBoost model.

There were some limitations to the study. Although the predictive accuracy of machine learning was good, it was not excellent due to the retrospective design. Also, the external validation set had a small number of participants and pathology data were not collected.

Machine learning algorithms could be accurately used for the prediction of survival in patients aged 60 years and older who are critically ill and have CRC, due to its superior performance compared with the logistic regression model, the study investigators concluded.

Reference

Guo C, Pan J, Tian S, Gao Y. Using machine learning algorithms to predict 28-day mortality in critically ill elderly patients with colorectal cancer. J Int Med Res. 2023;51(11):1-13. doi:10.1177/03000605231198725

Related Videos
Milind Desai, MD
Masanori Aikawa, MD
Screenshot of Susan Wescott, RPh, MBA
Glenn Balasky, executive director of the Rocky Mountain Cancer Center.
Cesar Davila-Chapa, MD
Female doctor in coat with stethoscope on blue background - Pixel-Shot - stock.adobe.com
Krunal Patel, MD
Juan Carlos Martinez, MD
Benjamin Scirica, MD, MPH, associate professor of medicine at Harvard Medical School and director of quality initiatives at Brigham and Women’s Hospital’s Cardiovascular Division
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.