Bayesian econometric modelling of observational data for cost-effectiveness analysis:establishing the value of Negative Pressure Wound Therapy in the healing of open surgical wounds

Background/Introduction In the absence of evidence from randomised controlled trials on the relative effectiveness of treatments, cost-effectiveness analyses increasingly use observational data instead. Treatment assignment is not, however, randomised, and naïve estimates of treatment effect may be biased. To deal with this bias, one may need to adjust for observed and unobserved confounders. In this work we aim to explore and discuss the challenges of these adjustment strategies using a case study of negative pressure wound therapy (NPWT) versus standard dressings for the treatment of open surgical wounds. Methods Time to wound healing, was estimated using Bayesian inference methods: i) OLS models, ii) OLS model adjusting for potential observed confounders and iii) two-stage instrumental variable (IV) models. A panel data regression approach was used to model health-related quality of life weights and costs. Cost-effectiveness estimates were obtained for selected models. Results The case study was a longitudinal cohort study of 393 participants followed up by on average 500 days. In all the modelling approaches we implemented, the treatment NPWT was estimated to offer less benefit at higher costs than competing interventions. Conclusions This study shows how to use observational data to assess cost-effectiveness by adjusting for both observable and unobservable confounders. Within the case study, we could not demonstrate that existing uncontrolled confounding affects the effectiveness of NPWT. There was no evidence that NPWT was effective or cost-effective compared to standard dressings for the treatment of SWHSI.