AI tool measuring ‘face age’ predicts survival in older patients with lung cancer

New research suggests that how old a person looks may say more about their survival than their actual age. In a cohort of older adults with early-stage non-small cell lung cancer (NSCLC), Harvard researchers from Mass General Brigham and Dana-Farber Cancer Institute found that an artificial intelligence-generated “face age” was associated with overall survival and early mortality, whereas chronological age was not. The findings were published April 8, 2026, in JAMA Network Open.

Lung cancer remains the leading cause of cancer-related death worldwide, with most cases classified as NSCLC. Although surgery is the standard treatment for early-stage disease, many patients are not candidates due to other health conditions. For these individuals, stereotactic body radiotherapy (SBRT) offers a safe and effective alternative, particularly among older adults with multiple comorbidities. However, determining which patients are most likely to benefit from SBRT remains a persistent clinical challenge.

Assessing biological age, rather than just years lived, has emerged as a potential way to better understand a patient’s overall health and resilience. While chronological age does not always reflect underlying fitness, biological age aims to capture how well the body is functioning. Artificial intelligence tools are increasingly being developed to estimate this using noninvasive data, including medical imaging and even facial photographs.

In the current retrospective cohort study, first author Grace Lee, M.D., of Mass General Brigham and Dana-Farber Cancer Institute in Boston, and colleagues evaluated two measures of biological age in patients with early-stage NSCLC. One measure, known as face age, was estimated from pretreatment photographs using a deep learning algorithm. The other, lung age, was calculated using lung function testing (spirometry) data. The study included patients aged 60 years and older who underwent SBRT between 2009 and 2023 across six affiliated radiation oncology clinics.

Among the 670 patients included in the study, the median chronological age was 77 years, while the median face age was slightly higher at 79 years. Notably, 17% of patients were 85 years or older based on their actual age, compared with 22% when age was estimated using facial features. Over a median follow-up of nearly four years, the median overall survival was 47 months, and about one in four patients died within two years of treatment.

When researchers adjusted for key clinical factors such as sex, performance status, cancer stage, smoking history, and tumor type, a higher face age was linked to worse overall survival. Specifically, each decade increase in face age was associated with a 39% higher risk of death. In contrast, chronological age on its own was not significantly associated with survival. A similar pattern was seen for early mortality, with higher face age associated with a greater likelihood of death within two years.

For a subset of 477 patients who also had lung function testing, the median lung age was 98 years. Lung age showed minimal correlation with face. In analyses that included both measures, face age remained independently associated with overall survival after adjusting for lung age.

The investigators also found that, on average, patients’ face age was older than their chronological age, reflecting what they described as cumulative biological aging. Larger gaps between the two measures were associated with outcomes, with older face age linked to worse survival and younger face age associated with better survival. Patients with a face age of 85 years or older had a higher risk of early mortality regardless of their actual age, the authors reported.

Smoking, a major risk factor for lung cancer, can also accelerate visible aging by contributing to wrinkles, skin thinning and changes in complexion. These features may influence AI-based facial age estimates, potentially linking external appearance with underlying health status.

The authors also emphasized that a photograph-based estimate of biological age could offer a practical and accessible tool in clinical settings. Because identification photographs are routinely obtained in radiation oncology, they wrote that integrating face age into existing risk assessment approaches may be feasible and could help refine prognostic estimates when used alongside established clinical factors such as performance status and comorbidities.

“Having a better understanding of a patient’s biological age and how much time they likely have remaining allows oncologists to better tailor treatments,” Raymond Mak, M.D., corresponding author, said in a news release last year regarding the FaceAge tool. Mak is a faculty member at the Artificial Intelligence in Medicine Program and Harvard Medical School associate professor of radiology oncology.

The researchers cautioned that prospective validation is needed before broader clinical use. They noted that further studies in more diverse populations will be important to confirm the findings and better understand how factors such as environmental exposures or cosmetic interventions may influence model performance.