We have read with interest the letter by Ioannou where he raises concerns about our manuscript describing the efficacy of percutaneous ablation for patients with hepatocellular carcinoma.1 Ioannou states that the study should have been designed as a randomised controlled trial (RCT) comparing active treatment vs best patient care. We agree that the benefits of treatment in oncology should be assessed through RCTs, as we have done for options such as chemoembolization, tamoxifen and interferon.2 RCTs should be conducted, however, when there is uncertainty on the final outcome. No such approach is justified if the results obtained in cohort studies are clearly better than the natural course of the disease. This is the case for resection and liver transplantation for early tumors, as they provide survival rates clearly better than their untreated counterparts (5-yr survival rates: 40%-70% vs. <20%) (3). Percutaneous ablation is currently accepted by most groups as an option that provides a high rate of complete response to therapy and thus, may provide long term cure of cancer.4 This therapeutic efficacy has led some authors to suggest that percutaneous ablation could have the same efficacy as surgical resection.5 Based on this high efficacy, it is widely agreed that the comparison of percutaneous ablation versus no treatment would not be ethically acceptable by most authors, Ethical Review Boards and regulatory authorities. To our knowledge no attempt to follow the Ioannou proposal has ever been conducted, and it is very unlikely that this might happen in the future.
In view of the unfeasible nature of such a trial, we performed an analysis similar to those done to evaluate all potentially curative treatments. All studies describe the treatment efficacy in a cohort of patients and identify the predictors of therapeutic success and better survival. This defines the best candidates for therapy and also provides the tool to predict outcome after therapy. Such an analysis requires a large sample size to derive robust conclusions, and we have to stress that our cohort of 282 patients represents one of the largest series ever reported in the West.
In the statistical analysis, we screened all relevant parameters by univariate analysis and, entered those with statistical significance (see Table 3) into the multivariate analysis. Twelve of the 24 assessed variables were significantly related to survival in the univariate analysis and were entered in the Cox-regression analysis, the ratios given reflecting hazard ratio. Only three variables had independent predictive power and were retained in the final model: CTP, BUN and response to treatment. These are the parameters listed in the bottom of Table 3, not implying they were the only ones entered into the multivariate approach. Thus, we adjusted the predictive power of treatment efficacy for all the potential confounding factors. As a consequence, we provided an unbiased estimate of the independent effect of initial treatment response.