Samsung Seoul Hospital researchers, led by Professors Hong-Kwan Kim and Hyun-Ae Jeong, have developed an artificial intelligence (AI) model named “RADAR” to predict the recurrence risk of early-stage non-small cell lung cancer (NSCLC) up to one year in advance. This AI model evaluates clinical, pathological, and CT scan results to quantify the likelihood of cancer recurrence one year after initial treatment. It was found that patients classified in the high-risk group had a 10-fold increase in recurrence risk compared to those in the low-risk group.
NSCLC accounts for 85% of all lung cancers and typically progresses at a slower rate. However, in early-stage patients, who make up about 35% of cases, surgery is commonly considered. Currently, post-surgery follow-up is conducted every 3-6 months regardless of an individual’s recurrence risk, making it challenging to establish tailored treatment strategies due to variability in patient conditions and tumor characteristics.
In response, Professors Kim and Jeong developed “RADAR CARE,” a transformer-based deep learning model, which processes diverse data types from clinical, pathological, and CT scan results. Published in the “JCO Precision Oncology” journal, the study analyzed data from 14,177 early-stage NSCLC patients who underwent surgery from January 2008 to September 2022 at Samsung Seoul Hospital.
Upon inputting basic information at the time of surgery, the model achieved a performance score (AUC) of 0.823, which increased to 0.854 with additional follow-up data. Patients were categorized based on their RADAR scores into low-risk (0.3 or lower), intermediate-risk (over 0.3 to 0.6), and high-risk (over 0.6) groups, with the scores reflecting recurrence likelihood within a year.
The study findings show that recurrence rates are 10% in high-risk, 5% in intermediate-risk, and 1% in low-risk groups. Importantly, patients with higher RADAR scores faced a higher likelihood of recurrence, regardless of cancer stage. Even stage 1 patients with high RADAR scores could have higher recurrence risks than stage 3 patients with low scores.
With these findings, the research team proposed individualized treatment strategies based on RADAR scores. Patients with persistently high scores were recommended for aggressive treatment, while those whose scores decreased over time could consider shorter treatment durations.
Professor Hyun-Ae Jeong emphasized the potential benefit this model offers in determining more favorable treatment paths for patients, considering current challenges in accurately predicting prognosis using only traditional staging methods. Professor Hong-Kwan Kim expressed hope that this advancement would lead to greater peace of mind and health recovery for many patients through informed strategic planning.