Historical control bias: adjuvant chemotherapy in esophageal cancer

Conclusion based on historical controls are known to be fallible in assessing efficacy of treatment due to selection bias. Historical controls have been repeatedly used in investigating efficacy of newer treatment avenues in esophageal cancer. The aim of the study was to assess the efficacy of adjuvant chemotherapy in esophageal cancer after correction for an historical control bias.

The database of 433 patients treated surgically for cancer of the esophagus at our institute between 1980 and 1989 was analyzed. The analysis was carried out using the Cox model for known prognostic factors without any correction for historical bias. Database was reanalysed after introducing registration year as a variable in the model to correct for historical control bias, which was further examined by carrying out a case-control study comparing chemotherapy (n = 83) vs contemporary control (n = 164) matched for lymph-node status, age and sex randomly selected from the same database.

The analysis without correction for historical control bias showed lymph-node metastasis (P = 0.000), female sex (P = 0.002), depth of invasion (P = 0.001) and adjuvant chemotherapy (P = 0.03) as significant predictors of survival. On introduction of registration year as a variable, lymph-node metastasis, sex and depth of invasion continued to be significant factors but chemotherapy was replaced by registration year (P = 0.02). The database with contemporary control showed lymph-node metastasis (P = 0.000), depth of invasion (P = 0.008) and female sex (P = 0.001) as significant factors. Chemotherapy had no effect on survival.

Results from historical controls are unreliable in detecting modest treatment benefits. Adjuvant chemotherapy in esophageal cancer should be tested within the tenets of randomized controlled trials with adequate-sample size to ascertain its efficacy.