Introduction: Vein bypass grafting is standard surgical therapy in the treatment for peripheral arterial disease (PAD), especially for those with limb threatening ischemia. However, the durability of vein grafts remains problematic with recent studies reporting disappointing 1-year primary patency rates as low as 61%. We hypothesized that an individual’s systemic inflammatory state, characterized by genome-wide inflammatory gene expression, would be predictive of ultimate clinical success or failure of these grafts. Methods: 41 patients undergoing vein bypass grafting for symptomatic PAD (85% critical limb ischemia; 25% claudication) had complete pre-operative gene expression data for analysis and one-year follow-up. There were 16 failures (39%) by one-year. Gene expression was analyzed for the entire genome using the novel Affymetrix GGH2 6.9 million feature oligonucleotide array system. Expression levels for the 35,123 gene transcripts and initial unsupervised and supervised (success/failure outcome) clustering was performed using BRB array tools. To then determine the additional predictive influence of specific genes or groups of genes, a novel objective Bayes method for dichotomous outcomes using stochastic search algorithms and probit regression models was employed. A series of increasingly predictive candidate models of gene combinations were derived. Clinical parameters including Rutherford disease severity score at presentation, vein conduit (GSV vs. composite vein), diabetes, smoking, statin use, and antiplatelet/anticoagulant use were standardized in the predictive model. Results: A random search through the space of possible models including the six clinical covariates and different combinations of gene products was executed, resulting in a list of 20 models ranked based on their posterior probabilities. The additional predictive power of the top gene products are listed in Table 1 and genes with the highest marginal posterior probabilities are listed in Table 2. This modeling approach suggests that the genes identified, independent of any relationship functionally, exhibit class prediction potential for bypass outcome. Conclusions: Using both a powerful and novel gene array system and a novel probit model selection method, we were able to identify a small number of gene clusters with predictive capability for bypass success or failure. This may both be prognostic and feasible for use in clinical practice as a point-of-care tool moving forward. Validation of these models is currently underway.
Read full abstract