Article Text

Download PDFPDF

Two models better than one
  1. C Hur,
  2. N S Nishioka,
  3. G S Gazelle
  1. Massachusetts General Hospital, Gastrointestinal Unit and Institute for Technology Assessment, Boston, Massachusetts, USA
  1. Correspondence to:
    Dr C Hur
    Massachusetts General Hospital, Gastrointestinal Unit and Institute for Technology Assessment, 101 Merrimac Street, 10th Floor, Boston, MA 02114, USA;
  1. N J Shaheen

Statistics from

The study by Shaheen and colleagues (Gut 2004;53:1736–44) is the results of a decision analysis model which determined the cost effectiveness of various management strategies for high grade dysplasia in Barrett’s oesophagus. We were surprised to note that the authors of this article did not reference our analysis which was published in July 2003.1 Our model and analysis had conclusions that were identical to those published by Shaheen et al. Similarities included the finding that endoscopic ablation (photodynamic therapy in our model) results in the greatest number of quality adjusted life years with similar incremental cost effectiveness ratios (ICER) compared with endoscopic surveillance. Also, both of our analyses found that endoscopic surveillance was less expensive than endoscopic ablation but associated with shorter survival.

The authors state in their discussion that their model has several strengths that distinguish it from previously published decision models of Barrett’s oesophagus, including the possibility of histological misdiagnosis of specimens as well as a non-linear progression to cancer, including the possibility of pathological regression. Our model also incorporated these strengths.

This congruency in the results of two independently constructed models only serves to strengthen and validate the findings of both models.


Author’s reply

I thank Hur et al for their interest in our article. I agree that his article,1 which appeared after the initial iterations of our article (Gut 2004;53:1736–44) had been written but prior to the acceptance of our revised manuscript, is highly pertinent to our work as it models the same clinical scenario.

There are clearly some differences in the models, which are likely due in part to the estimates used to construct it. For instance, average quality adjusted life expectancy when going from surgery to photodynamic therapy (PDT) in our model was increased by approximately 0.5 years whereas in the model by Hur et al the increase was 2.2 years, or four times our estimate. Also, some of our estimated lifetime costs for various therapies varied by as much as 25% from those estimated by Hur et al.

However, considering the number of assumptions and estimates inherent in modelling a complex clinical decision such as Barrett’s with high grade dysplasia (HGD), the model of Hur et al reports remarkably similar results to ours. An ablative approach with PDT yielded an increased quality adjusted life expectancy at a reasonable cost.

I agree with Hur et al that the similar findings of the models strengthens and validates the findings. More generally speaking, I feel that any model that features an intervention with some efficacy in the setting of HGD is likely to demonstrate that this intervention will be cost effective. The frequent progression of HGD to cancer, the high cost associated with caring for subjects with cancer, and the poor prognosis associated with cancer all suggest that any intervention keeping even a small fraction of patients with HGD from developing cancer is likely to be cost effective. This is true even if the intervention itself is costly (such as PDT). It probably does not matter whether the intervention is chemoprevention (as elegantly modelled recently by Sonnenberg and colleagues2) or ablative therapy, as modelled by Hur et al and ourselves.

Of course, there is a possibility that both models share the same flaws, leading them to come to similar, but erroneous, conclusions. After all, these models are only as good as the data used to create them, and good data on the natural history of various subsets of Barrett’s patients are hard to obtain, given the current state of the literature. However, until good randomised data comparing the treatment modalities for HGD are available with which to make these clinical decisions, the models are superior to expert opinion, intuition, or just plain guessing, as to the most appropriate path.


View Abstract


  • Competing interest: none declared


  • Conflict of interest: None declared.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.