Conflict of interest in economic evaluations of pharmaceuticals
Objective. The objective of this study was to investigate conflict of interest attributed to the affiliation (non-profit organization vs pharmaceutical companies) of the sponsors of economic evaluations of new drugs. ^ Methods. Regression models consisting of Sponsor alone were compared with models including other explanatory variables. Primary analysis used logistic regression to assess the binary dependent variable, Qualitative Conclusion of economic evaluations. This was supplemented with secondary analysis using multiple linear regression models to assess the dependent variable, Numerical Results of evaluations. Coefficients on Sponsor in all models were evaluated for relevance and were tested for statistical significance using the t statistic. Global fit of models was assessed with the pseudo R 2 statistic for logistic models and the R2 statistic for linear models. Models were compared for best fit using the ANOVA F-test. Statistical tests were conducted at the traditional significance level of p < 0.05; however, precise p values are reported should readers prefer to choose a different significance level when forming their own conclusions based on the evidence. ^ Results. In primary analysis using logistic regression, the impact of a change in Sponsor from nonprofit organizations to pharmaceutical companies increased the probability of a favorable evaluation outcome by only ~ 8% (this was in the model with the greatest effect, which was the model regressed on Sponsor alone). However, no coefficient on Sponsor was statistically significant. The global fit statistic (pseudo R2) of the model with Sponsor alone was poor compared with models containing additional variables (0.05 vs 0.20 to 0.531; p < 0.05). The model with the best global fit (0.53) contained 4 single-variable terms plus 3 sponsor-interaction terms, but this model was not statistically significantly better than the global fit of a simpler model with 3 variables and no interaction terms (0.37; p = 0.20). Secondary (confirmatory) analysis using linear regression revealed coefficients on Sponsor ranging from about $1500 to $3000; the coefficient on Sponsor was statistically significant in one model. Global fit of the linear model with Sponsor alone was poor, with an R2 of 0.06 compared with 0.28 to 0.58 for other models. ^ Conclusions. The collective evidence of this research fails to argue convincingly for the existence of systematic sponsor bias in economic evaluations of pharmaceuticals. The evidence also fails to support two-way analysis of research outcomes and sponsor affiliation as an adequate method to detect bias, because other variables were found to contribute important additional information to the analysis. Structural issues relevant to the economic evaluations themselves—including such factors as illness, drug class, and therapeutic approach—must be considered in addition to sponsor affiliation to accurately analyze research outcomes for purposes of detecting putative sponsor bias. ^
Economics, General|Economics, Commerce-Business
Thomas J Hogan,
"Conflict of interest in economic evaluations of pharmaceuticals"
(January 1, 2007).
ETD Collection for Fordham University.