Article type
Year
Abstract
Background: Increasingly, systematic reviews and meta-analyses of observational studies are used to address aetiological research questions. These meta-analyses often present challenges additional to those encountered in reviews of randomised trials, since exposure effect estimates derived from observational studies may be affected by confounding and by information and selection biases. One additional criterion for observational studies that supports a causal relationship is a dose-response association.
Objectives: To quantify the proportion of results of observational studies of the association between diet and cancer that can be used for meta-analysis.
Methods: Data from systematic reviews on the association of diet, nutrition and physical activity with the risk of bladder and prostate cancer were used. For each exposure, we recorded the OR per unit increase. Where this was not reported, we estimated it using the methods of Greenland and Longnecker (Am J Epidemiol 1992) and Chene and Thompson (Am J Epidemiol 1996). Reasons for not being able to estimate this measure were documented.
Results: The reviews on prostate and bladder cancer included 468 and 303 papers, respectively. In total, there were 274 cohort and 350 case-control studies that reported on 2626 exposures. 1106 results (42%) could not be used for the meta-analysis.
In cohorts, 887 results (72%) were reported as quantiles / categories. 365 (41%) could not be meta-analysed because the numbers of cases or denominators were missing (82%), the range of the exposure was poorly defined (54%) and/or the units or frequencies were missing (54%). 187 results (15%) were presented as mean data. 95 (51%) could not be used because no SD was given (96%). 158 results (13%) were presented as continuous data. 20 (12%) were not usable, mainly because a SD was missing (65%). In case-control studies, the proportion unusable results is higher. However, the reasons for excluding results were very similar.
Conclusions: The proportion of studies reporting usable results is small: this is an important threat to the validity and precision of systematic reviews of observational studies. Results presented as mean data were the least usable for meta-analysis. Standardised methods of presenting results of aetiological studies should be developed and implemented.
Objectives: To quantify the proportion of results of observational studies of the association between diet and cancer that can be used for meta-analysis.
Methods: Data from systematic reviews on the association of diet, nutrition and physical activity with the risk of bladder and prostate cancer were used. For each exposure, we recorded the OR per unit increase. Where this was not reported, we estimated it using the methods of Greenland and Longnecker (Am J Epidemiol 1992) and Chene and Thompson (Am J Epidemiol 1996). Reasons for not being able to estimate this measure were documented.
Results: The reviews on prostate and bladder cancer included 468 and 303 papers, respectively. In total, there were 274 cohort and 350 case-control studies that reported on 2626 exposures. 1106 results (42%) could not be used for the meta-analysis.
In cohorts, 887 results (72%) were reported as quantiles / categories. 365 (41%) could not be meta-analysed because the numbers of cases or denominators were missing (82%), the range of the exposure was poorly defined (54%) and/or the units or frequencies were missing (54%). 187 results (15%) were presented as mean data. 95 (51%) could not be used because no SD was given (96%). 158 results (13%) were presented as continuous data. 20 (12%) were not usable, mainly because a SD was missing (65%). In case-control studies, the proportion unusable results is higher. However, the reasons for excluding results were very similar.
Conclusions: The proportion of studies reporting usable results is small: this is an important threat to the validity and precision of systematic reviews of observational studies. Results presented as mean data were the least usable for meta-analysis. Standardised methods of presenting results of aetiological studies should be developed and implemented.