SNS 기사보내기
A predictive model has been developed to assist doctors in making treatment decisions for patients with metastatic cancer, particularly regarding radiation therapy, by providing an objective criterion to assess 30-day mortality risk.
Professors Kim Hae-young (left) and Lee Tae-hoon (Courtesy of Samsung Medical Center)
Professors Kim Hae-young (left) and Lee Tae-hoon (Courtesy of Samsung Medical Center)
Samsung Medical Center said Thursday that a research team led by Professors Kim Hae-young and Lee Tae-hoon of the Department of Radiation Oncology developed a model to predict the risk of death within 30 days by analyzing 3,756 patients who received radiotherapy for metastatic solid cancer at Samsung Medical Center and Samsung Changwon Hospital between 2018 and 2020.
Because radiation therapy often takes time to take effect, predicting the risk of death within 30 days can help patients avoid radiation therapy and give them more time to spend with their families.
The research team first trained the prediction model on data from 2,652 patients who received palliative radiation therapy at Samsung Medical Center's Department of Radiation Oncology and performed internal validation on data from 663 patients. External validation was also conducted on 441 patients at Samsung Changwon Hospital to confirm the model's reliability.
A machine learning algorithm—Gradient Boosting Model (GBM)—was applied to develop the model, and 12 clinical indicators, including patient age, gender, and treatment history, as well as seven blood test results, were referenced. A more generalized model (GBM-B) was also developed using only basic blood test results. The performance of the GBM model was compared to a conventional logistic regression model.
According to the researchers' analysis, the GBM model predicted the risk of death within 30 days more accurately than the conventional logistic regression model. When measuring the performance of the prediction model—Area Under the Curve (AUC)—the GBM model outperformed the traditional model (0.804) in external validation with an AUC of 0.833.
The model using only blood test results (GBM-B) outperformed the original model with an AUC of 0.830 in external validation. The results were similar when they checked the actual mortality rates according to the mortality risk classification. When patients were divided into four quartiles based on the mortality risk predicted by the GBM-B model, the first quartile, which had the lowest mortality risk, had an actual mortality rate of 0 percent.
That was followed by 3.4 percent in quartile 2, 12.9 percent in quartile 3, and 36.6 percent in quartile 4. This confirms that the actual 30-day mortality rate increases as the risk predicted by the model increases. The researchers hope the model can serve as a reference point for optimizing end-of-life care for cancer patients.
If the 30-day mortality risk predicted by the model is high, the number of radiotherapy treatments can be minimized, or other treatments can be considered based on the physical and economic burden, according to the team, researchers explained.
“What's best for the patient is a question that comes with the territory for doctors,” Professor Kim said. “It's not an easy decision to make, especially when faced with a patient with advanced cancer. Objective and scientific evidence is needed to help patients live a comfortable life. We hope this study will help both patients and healthcare providers.”
tweet
Kim Kyoung-Won kkw97@docdocdoc.co.kr
See Other Articles
Copyright © KBR Unauthorized reproduction, redistribution prohibited