G-CODE turns bedside frailty data into a mortality forecast for older adults with cancer
Older adults with cancer rarely fit neatly into the assumptions of standard oncology trials. Chronological age alone does not capture the difference between a robust 74-year-old and an 84-year-old with weight loss, poor mobility, and limited social support, yet treatment decisions often still hinge on age bands and tumor factors. The Geriatric Core Dataset, or G-CODE, was designed to make geriatric assessment more usable in research, and this study asked whether it could do more than organize data — whether it could predict survival in a clinically meaningful way.
It did.
Using two prospective French cohorts of patients aged 70 and older referred for geriatric assessment, investigators developed and externally validated a mortality prediction score based on G-CODE items. The goal was survival prediction at 6 months, the main endpoint, with 12- and 24-month mortality as secondary endpoints. The work linked geriatric deficits to near-term prognosis in a way that is immediately relevant to oncology decision-making, where the balance between benefit and harm can change quickly once frailty, comorbidity, and functional decline enter the picture.
Two French cohorts, one practical risk score
The derivation set came from the Elderly Cancer Patients cohort and included 950 patients. The external validation set came from the Geriatric Oncology Marseille cohort and included 766 patients. Mean age was 83 ± 6 years in the training set and 81 ± 6 years in the validation set; men accounted for 55% and 59% of the two cohorts, respectively. Six-month mortality reached 27% in the training group and 22% in the validation group, a reminder that this was a genuinely vulnerable population rather than a selected, relatively fit subgroup.
The score was built using multivariate Cox regression with Akaike information criterion selection. In practical terms, that approach asks which combination of geriatric variables best tracks with mortality without overcomplicating the model. The final G-CODE score ranges from 0 to 17, a scale compact enough to imagine at the bedside and structured enough to support research use across studies and settings.
Five geriatric deficits carried the strongest short-term signal
Five factors emerged as strong predictors of 6-month mortality: not having a caregiver, weight loss, comorbidities, an impaired 4-instrumental activities of daily living score, and an impaired timed-up-and-go test. Each one points to a different layer of physiologic reserve. Weight loss captures nutritional depletion and often deeper catabolic stress. Comorbidity reflects competing organ dysfunction and the burden of chronic disease. The instrumental activities score and timed-up-and-go test map directly to autonomy and mobility, two proxies for resilience when chemotherapy, surgery, or radiation begin to tax an already strained system. Lack of a caregiver adds a social vulnerability component that can determine whether treatment plans are actually executable outside the hospital.
This is not abstract geriatric scoring. It is a clinical picture of vulnerability rendered in measurable terms.
The biology behind the score is straightforward, even if the bedside implications are hard. Older adults with multiple comorbidities and impaired physical performance have less reserve to tolerate inflammatory stress, treatment toxicity, immobility, and nutritional decline. When a patient is already losing weight and walking slowly, the physiologic margin for surviving cancer treatment narrows quickly. Social support matters because frail patients often fail not only from disease progression, but from missed medications, delayed symptom recognition, poor intake, and difficulty reaching care at the right time.
Discrimination stayed strong through 24 months
The model performed well in both cohorts. Time-dependent area under the receiver operating characteristic curve values were 82%, 78%, and 80% for 6-, 12-, and 24-month mortality. Those are solid discrimination values for a clinical prediction tool in an older cancer population, especially one built from geriatric variables rather than tumor burden alone. The inclusion of G-CODE, age, cancer type, metastases, and treatment in the model helped it capture both the cancer-specific and host-specific dimensions of risk.
For clinicians, that matters because prognosis in geriatric oncology is rarely driven by the tumor in isolation. Two patients can have the same stage and same diagnosis, yet diverge sharply because one is independent, nourished, and mobile while the other is losing weight, has multiple comorbidities, and cannot rise from a chair without effort. This score formalizes those differences rather than leaving them to impressionistic judgment.
Why geriatric assessment needs a score like this
Comprehensive geriatric assessment already has a strong place in oncology because it helps identify vulnerabilities in function, cognition, comorbidity, nutrition, medication use, and social support. It also informs treatment modification — dosing, intensity, regimen choice, and supportive care can all change once the assessment is in hand. The problem has never been clinical relevance. The problem has been implementation. Geriatric assessment is rarely used in oncology trials, even though older adults make up a large and growing share of cancer patients.
A score built from a standardized geriatric dataset offers a way to make this information easier to incorporate into research and, eventually, into routine care. By reducing the assessment to a manageable set of predictors, G-CODE may help investigators stratify risk, compare populations, and select treatment approaches with more realistic expectations of survival. It may also help separate a patient whose limited life expectancy reflects reversible vulnerabilities from one whose prognosis is dominated by advanced disease.
What this score can and cannot answer today
The study establishes G-CODE as a mortality prediction tool with external validation, but it does not replace clinical judgment. It predicts survival, not treatment benefit, toxicity, quality of life, or patient preference. A high-risk score does not automatically mean treatment should be withheld, and a low-risk score does not guarantee tolerance of intensive therapy. Oncology decisions in older adults remain a negotiation among prognosis, tumor biology, functional reserve, and the patient’s goals.
Still, the score’s structure is clinically intuitive. A patient without a caregiver, with recent weight loss, chronic comorbidity, functional impairment, and slow mobility is the type of patient whose survival can change rapidly once treatment starts or a complication occurs. In that setting, a tool that translates geriatric deficits into a validated mortality estimate can sharpen conversations about feasibility, support needs, and whether the likely burden of therapy aligns with the likely time to benefit.
The main limitation is embedded in the question itself: a prediction model built in referred geriatric assessment cohorts may not behave identically in every oncology setting, especially where baseline frailty or support systems differ. Even so, the validation across two prospective French cohorts gives the score credibility as more than a statistical exercise. The unanswered issue is how often the score will change real-world management once embedded into oncology workflows and whether that change improves outcomes beyond prediction alone.