Using the ICD-9 codes (140–165, 170–175, 179–184, 185–208, and 284), all health economic publications in the field of oncology (n=1381) were identified in the Health Economic Evaluations Database (HEED, Version 1995–2000). Databases covered in HEED are Medline and Embase. After eliminating redundant publications (n=161), letters-to-editors, review articles, editorials and methodological studies (n=329), a total of 891 applied studies remained. We excluded all studies (n=1603) not disclosing a sponsor, so as to exclude studies that were not published or did not reveal the sponsor to the editors in order to have indisputable data for our study question. The reason that we did not send letters to the studies where no sponsor was indicated to see who actually sponsored the study was that if someone has not disclosed any sponsor for a sponsored study in his publication he has done a scientific misconduct and it is unlikely that he will admit this.
For 288 applied studies, a sponsor was named. After the exclusion of studies that did not compare alternative health technologies (cost-analysis (n=24), cost-consequence (n=98), and cost-of-illness (n=16) studies), a total of 150 sponsored studies involving comparisons remained (). They were evaluated on the basis of the criteria type sponsorship, type of economic analysis, health technology assessed, sensitivity analysis performed, publication status, and qualitative conclusions about costs.
An investigation was compared to have been industry-sponsored or to have been sponsored by nonprofit organisations if, in the publication, it was disclosed that all or part of the support provided originated from industry, respectively, from nonprofit organisations. For assessing the type of economic analysis used, distinctions were made among cost-effectiveness analysis (measures benefits of different medical treatments having a common health outcome (e.g., life years saved) and computes a cost-effectiveness measure ratio for comparison), cost-benefit analysis (measures costs and benefits of a medical treatment in financial units and computes a net monetary gain/loss or a cost–benefit ratio), cost-utility analysis (measures benefits of different medical treatments in utility-weighted life years (quality-adjusted life years, QALYs) and computes a cost–utility ratio for comparison), and cost-minimisation analyses (finds the least costly medical treatment among those shown or assumed to be of equal medical benefit).
Studies were also differentiated into categories for health technology assessed: pharmacokinetic monitoring, supportive care, medical devices, diagnostic, drugs, screening, and surgery. Publication status as a marker for the publication rated studies as published in peer-reviewed or non-peer-reviewed journals. Since even within the peer-reviewed literature, the quality of publications is highly variable, a sensitivity analysis (one-way or multiple-way) was chosen from the checklist for this subject, developed by Drummond (Drummond and Jefferson, 1996
) as an additional marker for the study quality. Sensitivity analysis (one-way or multiple-way) was chosen, because it is a process through which the robustness of an economic model is assessed by examining the changes in results of the analysis when key variables are varied over a specified range. Therefore, a yes/no decision was made as to whether a sensitivity analysis had been performed. Qualitative conclusions about costs were evaluated as positive, negative or neutral. Conclusions about costs were assumed to be positive if there was a statement in the publication that the health technology assessed reduced costs, was cost effective, cost beneficial, or cost utilitarian, and negative if the technology assessed had higher cost, and was not cost effective, cost beneficial, or cost utilitarian. When there was neither a positive nor a negative statement about the outcome of cost assessment, it was assumed to be neutral for the purpose of the present analyses. All evaluations were made independently by two investigators (MH and HK). If the investigators disagreed over the evaluation of any article, a third investigator (DS) made a deciding evaluation.
Relationships between the type of sponsorship and type of economic evaluation, health technology assessed, sensitivity analysis performed, publication status, and qualitative conclusions about costs were analysed using Fisher's exact test (the modifications by Metha for tables larger than 2 × 2) to provide a two-sided probability (P<0.05, against the null hypothesis of no relationship was considered significant).
In a second analysis, large tables (analysis type, health technology, and conclusions about costs) were recombined to provide 2 × 2 tables. That is, for dealing with the relative risk (industry vs nonprofit) for ‘cost-minimisation’ analyses, all other types were collected in one category called ‘other’. The reason behind the collapsing into 2 × 2 tables was not the (two-sided) P-value itself, but the one-sided P-value, because Fisher's exact test is inherently two-sided (at least for tables larger than 2 × 2); furthermore by collapsing into 2 × 2 tables, we could construct confidence intervals (CIs) for the relative risk for ‘cost minimisation,’ which is not possible for larger tables. Hence, by using these 2 × 2 tables, one-sided 99% CIs for the relative risks were calculated by computing P-values. A (small) one-sided P-value allows to conclude that one sponsoring type provides significant larger (or smaller) numbers of, for example, positive outcomes. The higher confidence level of 99% was chosen so as to insure that simultaneous computing of two (ore more) intervals provides reasonable small overall error probabilities and does not lead to high error probabilities. No formal adjustment was made for multiple comparisons and, therefore, the 99% CIs have descriptive interpretation rather than being used for hypothesis testing. All computations were executed by SAS 8.2 Software (SAS Institute Inc., Cary, NC, USA).