Article type
Year
Abstract
Background: Industrial funding of the biomedical studies is associated with practices which increase the possibility of bias, including subgroup and per protocol analyses, and testing of multiple hypotheses or multiple variables.
Objective: To find out if industrial funding associated with the use of survival analysis (SA) to estimate the cause-specific survival (CSS).Methods: We searched MEDLINE by 'survival analysis'. CSS studies were found by 'specific survival’. Only human-related studies were selected. To evaluate the source of financial support, the full texts of the articles were checked. The primary analysis was limited to year 2009.
Results: In 2009,12787 reports were classified as using SA, and only 423 (3%) employed analysis of CSS. All 104 articles using CSS and available in full text were evaluated, as well as 104 of the first articles of 3813 using SA but not CSS and available in full text. Evaluation of the sources of funding were done by two authors independently with 30% overlap to estimate the agreement (K = 0.67). Of 104 studies using CSS, 16 (15%) were funded by industry, while of 104 studies using SA without estimates of CSS, 7 (6.7%) reported industrial funding. The 95% confidence interval for the difference is 0.001 to 0.175.
Conclusions: The possibility of finding significant differences in specific-survival when total survival is not different between groups may be an attractive perspective. This possibility is used by scientists 'torturing’ the data in search of statistical significance. Authors of systematic reviews need to know about this possible source of bias in estimates of survival. The full presentation of this study will include the data on use of specfic-survival in clinical trials.
Objective: To find out if industrial funding associated with the use of survival analysis (SA) to estimate the cause-specific survival (CSS).Methods: We searched MEDLINE by 'survival analysis'. CSS studies were found by 'specific survival’. Only human-related studies were selected. To evaluate the source of financial support, the full texts of the articles were checked. The primary analysis was limited to year 2009.
Results: In 2009,12787 reports were classified as using SA, and only 423 (3%) employed analysis of CSS. All 104 articles using CSS and available in full text were evaluated, as well as 104 of the first articles of 3813 using SA but not CSS and available in full text. Evaluation of the sources of funding were done by two authors independently with 30% overlap to estimate the agreement (K = 0.67). Of 104 studies using CSS, 16 (15%) were funded by industry, while of 104 studies using SA without estimates of CSS, 7 (6.7%) reported industrial funding. The 95% confidence interval for the difference is 0.001 to 0.175.
Conclusions: The possibility of finding significant differences in specific-survival when total survival is not different between groups may be an attractive perspective. This possibility is used by scientists 'torturing’ the data in search of statistical significance. Authors of systematic reviews need to know about this possible source of bias in estimates of survival. The full presentation of this study will include the data on use of specfic-survival in clinical trials.