Keyword: Bias
2 results found.
Methodological Paper
Epidemiology and Health Data Insights, 1(5), 2025, ehdi018, https://doi.org/10.63946/ehdi/17368
ABSTRACT:
Background. This study examined the reporting practice of subgroup effects of meta-analytic research published in the leading journal of the ILAE.
Methods. We selected studies that used ratio measures and employed subgroup analyses. Subgroup differences were calculated as the difference between the log-transformed estimates over the square root of the sum of the squared standard errors. For the calculated test scores, a corresponding two-tailed p-value was calculated using the standard normal cumulative distribution function. The authors also conducted additional analyses accounting for multiple comparisons.
Results. The literature search identified 55 publications, of which 14 (25 %) were included. Neither study used a formal test to compare the subgroups. The number of reported subgroup estimates ranged from 2 to 20, and the number of pairwise comparisons ranged from 1 to 53. Overall, there were 187 comparisons, resulting in a median log difference of 0 (IQR 0.12 – 0.17) and a range from -2.92 to 2.32. The median p-value was 0.54 (IQR 0.21-0.85) with 18 (9%) comparisons showing p-values lower than the conventional significance level, whereas 6 and 21 contrasts were 0.05 < p ≤ 0.10 and 0.10 < p ≤ 0.20, respectively. Seven (4%) comparisons resulted in a p-value lower than the corrected significance level when adjusted for multiple comparisons.
Conclusion. There was a lack of compliance with the reporting guidelines. The findings from the subgroup analyses were commonly interpreted without employing a formal test. There is need to emphasize the importance of adherence to established reporting standards when presenting the subgroup effects.
Methods. We selected studies that used ratio measures and employed subgroup analyses. Subgroup differences were calculated as the difference between the log-transformed estimates over the square root of the sum of the squared standard errors. For the calculated test scores, a corresponding two-tailed p-value was calculated using the standard normal cumulative distribution function. The authors also conducted additional analyses accounting for multiple comparisons.
Results. The literature search identified 55 publications, of which 14 (25 %) were included. Neither study used a formal test to compare the subgroups. The number of reported subgroup estimates ranged from 2 to 20, and the number of pairwise comparisons ranged from 1 to 53. Overall, there were 187 comparisons, resulting in a median log difference of 0 (IQR 0.12 – 0.17) and a range from -2.92 to 2.32. The median p-value was 0.54 (IQR 0.21-0.85) with 18 (9%) comparisons showing p-values lower than the conventional significance level, whereas 6 and 21 contrasts were 0.05 < p ≤ 0.10 and 0.10 < p ≤ 0.20, respectively. Seven (4%) comparisons resulted in a p-value lower than the corrected significance level when adjusted for multiple comparisons.
Conclusion. There was a lack of compliance with the reporting guidelines. The findings from the subgroup analyses were commonly interpreted without employing a formal test. There is need to emphasize the importance of adherence to established reporting standards when presenting the subgroup effects.
Methodological Paper
Epidemiology and Health Data Insights, 1(1), 2025, ehdi001, https://doi.org/10.63946/ehdi/16216
ABSTRACT:
The measure of association in observational studies is often prone to misinterpretation, further necessitating a discussion of methodological challenges. Understanding the underlying causes and proposing proper steps may help prevent such issues in the literature. This article provides straightforward explanations of terminology and a decision tree to select appropriate measures and accurate interpretations.