Machine learning for optimized individual survival prediction in resectable upper gastrointestinal cancer
Problem
Surgical oncologists need reliable individual long-term survival estimates for patients after resection of gastric or distal esophageal cancer.
Why it matters
Accurate prognosis informs life-altering treatment decisions for cancer patients.
What the AI did
Eight machine learning algorithms (including multi-task logistic regression and survival forests) were compared against usual methods to build individual survival estimations from 117 variables in 1360 patients.
Results
Random survival forest: concordance index 0.736, integrated Brier score 0.166, mean AUC 0.814 over 10 years post-diagnosis.
Impact
Random survival forest outperformed usual algorithms for individual survival prediction in a single-center retrospective cohort.
Record
| Venue | Journal of Cancer Research and Clinical Oncology |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Jin‐On Jung, Nerma Crnovrsanin, Naita Maren Wirsik, Henrik Nienhüser, Leila Peters, Felix Popp, A. Schulze, Martin Wagner et al. |
| Citations | 45 (OpenAlex, at import time) |
| References | 26 |
| Model or method | random survival forest (best performing of eight algorithms) |
| Organization | Heidelberg University Hospital, Department of General Surgery |
| Primary source | https://doi.org/10.1007/s00432-022-04063-5 |
| Open access | hybrid |
| Keywords | machine learning, survival prediction, gastroesophageal cancer, random survival forest, prognosis |
Importance rating
significant — thousands to millions affected
Cancer prognosis informs life-altering treatment decisions; single-center retrospective study of 1360 patients, no broader scale stated in abstract.