To the Editor:

I have always read with interest the research coming out of the Barcelona Clinic Liver Cancer (BCLC) Group, which is usually on the cutting edge of our understanding of the pathogenesis and treatment of hepatocellular carcinoma.

In the December issue of HEPATOLOGY, Sala et al. from the BCLC aimed to investigate the effect of percutaneous ablation for hepatocellular carcinoma on survival.1 They abandoned the use of any untreated control group and decided instead to compare the survival of persons in whom percutaneous ablation was successful to the survival of persons in whom percutaneous ablation was unsuccessful. They found that survival was better in patients with successful ablation and concluded that “our data show that if the tumor is effectively ablated this translates into an improved outcome” and that “initial complete tumor necrosis should be considered a relevant therapeutic target.”

However, persons who did not respond to percutaneous ablation were more likely to have larger tumors or multiple tumors (as shown in Table 2 of Sala et al.). Also, tumor size and stage were very strong predictors of survival (as shown in Table 3 of Sala et al.). Despite clearly demonstrating that tumor size and stage were associated with both success of percutaneous ablation and with survival, the authors fail to adjust for these factors in their multivariate model when investigating the potential “independent” effect of percutaneous ablation on survival. Their justification appears to be that in their multivariate model they included only variables significant (P < .05) in the stepwise Cox regression analysis. This is an entirely erroneous use of these statistical techniques. If the authors wanted to identify an unbiased estimate of the independent effect (if any) on survival of successful ablation compared to unsuccessful ablation, then all potential confounders should be simultaneously adjusted for, including every single factor in Table 3 (tumor stage, tumor size, ascites) not just blood urea nitrogen and Child-Turcotte-Pugh class. It is entirely possible that if the authors had a sufficiently large sample size to be able to simultaneously adjust for all important confounders, no survival benefit would be associated with initial complete response. In nonstatistical terms, the reason that patients with “successful” ablation have better survival may be that they are more likely to have smaller tumors, single tumors, and better liver function than the patients with unsuccessful ablation and their survival may have nothing to do with the ablation itself.

Furthermore, the study design is fundamentally flawed because instead of comparing the survival in treated versus untreated patients, the authors chose to compare survival in “successfully treated” versus “unsuccessfully treated” patients. Clearly, even if a treatment is overall completely ineffective, those who happen to have a good “response” to the treatment will always have a better survival than those who happen to have a bad “response” to the treatment.

Incidentally, the authors describe the results of their multivariate analyses in terms of odds ratios. However, in the Methods section they reported that they performed a Cox proportional hazards analysis, which yields hazard ratios, not odds ratios. The authors should clarify this discrepancy.


  1. Top of page
  • 1
    Sala M, Llovet JM, Vilana R, Bianchi L, Sole M, Ayuso C, et al. Initial response to percutaneous ablation predicts survival in patients with hepatocellular carcinoma. HEPATOLOGY 2004; 40: 13521360.

George N. Ioannou M.D., M.S.*, * Division of Gastroenterology, Veterans' Affairs Puget Sound Health Care System and University of Washington, Seattle, WA.