KEY POINTS
- The Danish multi-endpoint prediction model was externally validated using patients treated with curative-intent intensity-modulated radiotherapy at six Australian centres. Of 781 eligible patients, 502 had complete model data, including 129 patients aged ≥70 years.
- The validated “basis model” used routinely available variables rather than tumour volume. Depending on the endpoint, inputs included smoking status, T and N stage, tumour subsite and p16 status, age, WHO performance status, and whether concurrent cisplatin was prescribed.
- At three years, observed cumulative incidences in the full cohort were 17.8% for locoregional failure, 10.4% for distant metastasis, and 9.1% for death without evidence of cancer. Among older patients, the corresponding rates were 20.9%, 11.2%, and 16.1%.
- Three-year discrimination was moderate: AUC 0.66 (95% CI 0.60–0.73) for locoregional failure, 0.73 (0.65–0.80) for distant metastasis, and 0.73 (0.66–0.79) for death without evidence of disease. Corresponding Brier scores were 0.14, 0.09, and 0.08.
- The predefined low-, intermediate-, and high-risk groups significantly separated locoregional failure and distant metastasis in the full cohort (both p<0.001) and in patients aged ≥70 years (p=0.04 and p=0.002, respectively).
- Prediction of death without evidence of cancer was less convincing. In the full cohort, separation was driven mainly by the low- versus high-risk comparison (HR 2.40, 95% CI 1.15–5.02), while low- and intermediate-risk groups were not distinguished; no meaningful risk-group separation was seen in older patients (p=0.41).
- Calibration was generally acceptable for recurrence and metastasis, although locoregional failure was overestimated at the highest predicted risks. Death without evidence of disease was underestimated above approximately 25% predicted risk, and the model omits direct measures of frailty, comorbidity, cardiovascular disease, and competing mortality.
CLINICAL TAKEAWAY
The model appears transportable enough to support prognostic discussions about recurrence and distant failure across different healthcare systems, including in older patients. It should not yet be used to omit cisplatin, de-escalate radiotherapy, or otherwise modify treatment: its non-cancer mortality component requires improvement, and prospective impact testing is still needed.