Do decision-analytic models identify cost-effective treatments? A retrospective look at helicobacter pylori eradication.

BACKGROUND Pharmacoeconomic models of Helicobacter (H) pylori eradication have been frequently cited but never validated. OBJECTIVE Examine retrospectively whether H pylori pharmacoeconomic models direct decision makers to cost-effective therapeutic choices. METHODS We first replicated and then validated 2 models, replacing model assumptions with empirical data from a multipayer claims database. Database subjects were 435 commercially insured U.S. patients treated with bismuthmetronidazole- tetracycline (BMT), proton pump inhibitor (PPI)-clarithromycin, or PPI-amoxicillin. Patients met >1 clinical requirement (ulcer disease, gastritis/duodenitis, stomach function disorder, abdominal pain, H pylori infection, endoscopy, or H pylori assay). Sensitivity analyses included only patients with ulcer diagnosis or gastrointestinal specialist care. Outcome measures were: (1) rates of eradication retreatment; (2) use of office visits, hospitalizations, endoscopies, and antisecretory medication; and (3) cost per effectively treated (nonretreated) patient. RESULTS Model results overstated the cost-effectiveness of PPI-clarithromycin and underestimated the cost-effectiveness of BMT. Prior to empirical adjustment, costs per effectively treated patient were 1,001 US dollars, 980 US dollars, and 1,730 US dollars for BMT, PPIclarithromycin, and PPI-amoxicillin, respectively. Estimates after adjustment were US dollars for BMT, 1,118 US dollars for PPI-clarithromycin, and 1,131 US dollars for PPI-amoxicillin. Key model assumptions that proved retrospectively incorrect were largely unsupported by either empirical evidence or systematic assessment of expert opinion. CONCLUSIONS Organizations with access to medical and pharmacy claims databases should test key assumptions of influential models to determine their validity. Journal peer-review processes should pay particular attention to the basis of model assumptions.