Background: The strength of randomised controlled trials (RCTs) is that they allow investigators to draw reliable inferences about treatment differences; physicians can then make a choice between different options. Their weakness is that they are conducted on a set of patients who cannot be regarded as a random sample from the population that will be treated outside the trial. Observational data collected in a prospective clinical database may provide more realistic estimates.

Methods: At the Istituto Nazionale de Tumori of Milan an RCT was started in 1973 (MI1) in which Halsted mastectomy was compared with quadrantectomy plus axillary dissection and radiotherapy (QUART), a breast conserving treatment, for "early-stage" primary breast cancer. In addition, a prospective clinical database was established to collect, in standard format, information on all women undergoing breast conservation thereafter and followed up at the Institute. 1760 women were treated with QUART, 350 of them being accrued in the years 1973-80 in the context of the MI1 trial. The remaining 1408 received QUART after the MI1 trial, up to December, 1984.

Results: In the years after the trial, the criteria for use of QUART became more liberal. However, after adjustment for baseline characteristics, out-trial patients fared no worse than trial patients in terms of survival or distant metastasis. Out-trial patients did have a slightly higher rate of local recurrence, probably because of subtle and multiple changes in the surgical procedures since the time of the trial.

Interpretation: The results of QUART in out-trial patients resembled those reported from research settings. In this instance, the results of clinical trials did not exaggerate the benefits obtainable in routine practice.