Results
There were 88 papers (39.5%) with a prospective study design while 43 (19.3%) and 92 (41.2%) papers had a case-control and cross-sectional design, respectively. Amongst all included papers, 45.4% fully reported all 22 STROBE items according to our conservative definition or by citation. This differed by study type with 47.8% for cohort studies, 45.7% for cross-sectional studies, and 41.4% for case-control studies (pChiSq=0.007 for the overall comparison of the three designs). This difference was brough about by the case-control studies which differed from cohort (pChiSq=0.0017) and cross-sectional studies (pChiSq=0.035); the difference between cohort and cross-sectional studies was not statistically significant. This difference by study design remains when using our liberal scoring of the 22 STROBE items or reporting by citation. Here, we observed 67.0% for cohort studies, 64.1% for cross-sectional studies, and 58.1% for case-control studies (pChiSq<0.0001 and pChiSq=0.0015 comparing case-control to cohort and to cross-sectional studies, respectively; pChiSq<0.0001 for the overall comparison of the three designs; no statistically significant difference between cohort and cross-sectional studies).
A clear difference between our conservative and liberal scoring definitions to the STROBE checklist was observed for the title and abstract sections. This was due to the low proportion of papers reporting their study design in the title or abstract (28.3%). Of note, this proportion was higher for cohort studies and cross-sectional studies (36.4% and 33.7%, respectively) than for case-control studies (18.6%). The percentages of full reporting by item, paper section, and study design are shown in figure 1.
The overall score was stable over time, irrespective of our conservative or liberal definition. Investigating sections of the STROBE, we observed that funding reporting increased from less than 60% to more than 90% while the scoring for other STROBE sections remained rather constant over the period from 2009 to 2018 (Fig 2).
Finally, results from the correspondence analyses show that its first two dimensions together explain 29.6% and 36.9% of the overall variance in our conservative and liberal STROBE scores, respectively. The first dimension separates full reporting from “not reported”, explaining 18.6% and 23.9% of the overall variance in our conservative and liberal scores, respectively. The positioning of study designs along this dimension shows that reporting in papers based on a cohort study was more complete than reporting in papers based on a case-control design; here, the cross-sectional studies had an intermediate position. For the conservative score, the second dimension separated the STROBE items 1, 4, 7-11, 14, 17, and 21 from the other STROBE items. Notably, those items investigate reporting of features of the study design in the title and methods, of types of variables and their measurement/assessment, of bias and confounding, of study size, and of grouping of variables. These items are also the more often poorly reported ones. In the correspondence analysis of our liberal score, this cluster is extended to the items 6, 9, 10, and 12 from the methods section, 16 and 19 from the results section, and item 22 regarding the funding source. In essence, the second dimension separates items from the methods and results sections with low proportion of full reporting from the other STROBE items. Again, positioning of the study designs along this dimension separates cohort and cross-sectional designs from case-control studies confirming the aforementioned poor reporting in papers based on a case-control design.