Modelling suggests that guaranteeing radiotherapy within six weeks would not change overall survival

Design

Retrospective cohort from the National Cancer Database with inverse probability weighting

Population

58,184 patients with HPV-negative head and neck squamous cell carcinoma who received adjuvant radiotherapy

Primary outcome

Five-year overall survival

Effect

Observed 0.58 vs predicted 0.58 with universal radiotherapy within 42 days

This National Cancer Database study from 2004 to 2020 included 58,184 patients with HPV-negative head and neck squamous cell carcinoma who received adjuvant radiotherapy. Timely radiotherapy was defined as starting within 42 days of surgery. Inverse probability weighted models estimated survival if all patients had started within six weeks.

Observed five-year overall survival was 0.58 and predicted survival with universal timely radiotherapy was also 0.58. Subgroup predictions varied: the largest predicted gains were in privately insured, white, rural men and in hypopharyngeal and stage IV disease, though their confidence intervals included no effect. Black patients were predicted to do worse with timely radiotherapy (observed 0.54, predicted 0.50).

The authors conclude that even optimal implementation would not be predicted to improve survival overall, and suggest modelling can guide resource allocation. This is a retrospective model with unmeasured confounding and it cannot show that delay is harmless for an individual patient, so established targets for starting radiotherapy are not overturned.

• Predicted 5-year survival with universal timely radiotherapy matched the observed 0.58.
• Subgroup predictions had wide intervals that included no effect.
• The model predicted a lower survival for one subgroup, which may reflect confounding rather than harm.
• Existing guidance on timing has not changed; keep to local targets.
• Use the findings to discuss where service effort is best spent.

Why it matters

It questions whether time to radiotherapy is a lever that moves survival, which affects where services invest effort.

Don't overread it

A modelled counterfactual from observational data does not show that delay is safe, and no guideline on timing has changed.

The statistics, in plain English

A predicted survival of 0.58 against an observed 0.58 means the model found no overall gain from forcing everyone into the six-week window. Inverse probability weighting adjusts for measured factors only, so unmeasured differences between early and late starters can still bias the estimate.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.