Starting postoperative radiotherapy within six weeks was not predicted to improve survival

Design

Retrospective cohort with inverse probability weighted counterfactual modelling (National Cancer Database)

Population

58,184 patients with HPV-negative head and neck squamous cell carcinoma given postoperative radiotherapy

Primary outcome

5-year overall survival

Effect

Observed 0.58 vs predicted 0.58 with universal start within 42 days

Using the US National Cancer Database from 2004 to 2020, researchers studied 58,184 patients with HPV-negative head and neck squamous cell carcinoma who had postoperative radiotherapy. They modelled what five-year survival would have been if every patient had started within 42 days of surgery.

Observed five-year survival was 58%, and the predicted figure under universal timely radiotherapy was also 58%. Some subgroups, such as stage IV and hypopharyngeal cancers, showed a predicted benefit, but the intervals included no effect. One subgroup showed a predicted decline, which is hard to explain biologically and is a reminder that models of this kind carry their own biases.

The six-week target remains a quality standard in many guidelines, and this analysis does not change that. Earlier studies have linked delay to worse outcomes, and the model cannot capture every reason radiotherapy was late.

What it does suggest is that, across a whole population, closing the timing gap alone may not move survival much, and departments should not neglect other drivers such as margin status, adequate dose and treatment completion.

• Keep aiming to start postoperative radiotherapy within six weeks; guidelines have not changed.
• Do not assume timing alone explains outcome differences between units.
• Track margins, treatment completion and dose alongside time to radiotherapy.
• Plan radiotherapy referral before surgery to avoid avoidable delay.

Why it matters

It questions how much survival a timing-focused quality programme can buy on its own.

Don't overread it

A modelled, observational analysis; it does not show delay is harmless or that the six-week target should change.

The statistics, in plain English

The observed and predicted survival were both 0.58, so the model found no gain. Subgroup estimates had wide intervals including zero effect, and modelled 'what if' survival depends heavily on how well confounding was adjusted.

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.