Limits...
Is the relationship among outcome variables shown in randomized trials?

Schriger DL, Cooper RJ, Lopez-O'Sullivan A, Wystrach C, Altman DG - Trials (2015)

Bottom Line: Readers are rarely shown the relation between outcomes.Mandatory posting of datasets or requirements for detailed appendices would allow readers to see these cross-tabulations, helping future investigators know which outcomes are redundant, which provide unique information and which are most responsive to changes in the independent variables.While not every relationship between outcomes requires depiction, at present such information is seldom portrayed.

View Article: PubMed Central - PubMed

Affiliation: Department of Emergency Medicine, University of California, 924 Westwood Blvd, Suite 300, Los Angeles, CA, 90024, USA. schriger@ucla.edu.

ABSTRACT

Background: Randomized controlled trials (RCTs) often have more than one primary outcome and frequently have secondary and harm outcomes. Comparison of outcomes between study arms is the primary focus of RCTs, but there are times when the relation between outcomes is important, such as determining whether an intermediate outcome and a clinical outcome have a strong association. We sought to determine how often reports of RCTs depict the relations among outcomes at the individual patient level and, for those studies that use composite outcomes, how often the relations between component elements are depicted.

Methods: We selected 20 general, specialty and subspecialty medical journals with high impact factors that publish original clinical research. We identified every RCT in the 2011 and 2012 issues and randomly selected 10 articles per journal. For each article we recorded the number of outcomes, the number of composite outcomes and how often the relations between outcomes or elements of composite outcomes were portrayed.

Results: All but 16 of the 200 RCTs had more than one outcome. Thus, outcomes could have been related in 92% of studies, but such relations were only reported in 2 (1%). A total of 33 (17%) investigations measured a composite outcome, 32 of which showed data for each component. None, however, showed cross-tabulation of the components.

Conclusions: Readers are rarely shown the relation between outcomes. Mandatory posting of datasets or requirements for detailed appendices would allow readers to see these cross-tabulations, helping future investigators know which outcomes are redundant, which provide unique information and which are most responsive to changes in the independent variables. While not every relationship between outcomes requires depiction, at present such information is seldom portrayed.

Show MeSH
Histogram of the number of outcomes reported in the 200 randomized controlled trials. (A) Primary outcomes; (B) Secondary outcomes; (C) Harm outcomes. Solid red line = median. Dashed red line = mean.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4374175&req=5

Fig1: Histogram of the number of outcomes reported in the 200 randomized controlled trials. (A) Primary outcomes; (B) Secondary outcomes; (C) Harm outcomes. Solid red line = median. Dashed red line = mean.

Mentions: All but two studies identified a primary outcome; in one the authors stated that they intentionally avoided considering any of the 11 outcomes primary, in the other there were six outcomes and no comment regarding which was primary [9,10]. While the word primary implies that there would be a single most important outcome, the CONSORT statement recommends a single primary outcome and study size calculations are almost always based on a single outcome, 43% (86 out of 200) of articles had more than one primary outcome [11]. The mean number of primary outcomes was 2.2 (median: 1, range: 1 to 19) (Table 1, Figure 1A). A total of 134 (67%) studies reported secondary outcomes and 64 (32%) reported harms. Across all studies, a mean of 3.8 secondary outcomes and 2.4 harms were reported. (Figure 1B, C). Studies that reported at least one secondary outcome reported a mean of 5.7 (median: 4, range: 1 to 25). For harms these numbers were 7.3 (median: 6, range: 1 to 21).Table 1


Is the relationship among outcome variables shown in randomized trials?

Schriger DL, Cooper RJ, Lopez-O'Sullivan A, Wystrach C, Altman DG - Trials (2015)

Histogram of the number of outcomes reported in the 200 randomized controlled trials. (A) Primary outcomes; (B) Secondary outcomes; (C) Harm outcomes. Solid red line = median. Dashed red line = mean.
© Copyright Policy - open-access
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4374175&req=5

Fig1: Histogram of the number of outcomes reported in the 200 randomized controlled trials. (A) Primary outcomes; (B) Secondary outcomes; (C) Harm outcomes. Solid red line = median. Dashed red line = mean.
Mentions: All but two studies identified a primary outcome; in one the authors stated that they intentionally avoided considering any of the 11 outcomes primary, in the other there were six outcomes and no comment regarding which was primary [9,10]. While the word primary implies that there would be a single most important outcome, the CONSORT statement recommends a single primary outcome and study size calculations are almost always based on a single outcome, 43% (86 out of 200) of articles had more than one primary outcome [11]. The mean number of primary outcomes was 2.2 (median: 1, range: 1 to 19) (Table 1, Figure 1A). A total of 134 (67%) studies reported secondary outcomes and 64 (32%) reported harms. Across all studies, a mean of 3.8 secondary outcomes and 2.4 harms were reported. (Figure 1B, C). Studies that reported at least one secondary outcome reported a mean of 5.7 (median: 4, range: 1 to 25). For harms these numbers were 7.3 (median: 6, range: 1 to 21).Table 1

Bottom Line: Readers are rarely shown the relation between outcomes.Mandatory posting of datasets or requirements for detailed appendices would allow readers to see these cross-tabulations, helping future investigators know which outcomes are redundant, which provide unique information and which are most responsive to changes in the independent variables.While not every relationship between outcomes requires depiction, at present such information is seldom portrayed.

View Article: PubMed Central - PubMed

Affiliation: Department of Emergency Medicine, University of California, 924 Westwood Blvd, Suite 300, Los Angeles, CA, 90024, USA. schriger@ucla.edu.

ABSTRACT

Background: Randomized controlled trials (RCTs) often have more than one primary outcome and frequently have secondary and harm outcomes. Comparison of outcomes between study arms is the primary focus of RCTs, but there are times when the relation between outcomes is important, such as determining whether an intermediate outcome and a clinical outcome have a strong association. We sought to determine how often reports of RCTs depict the relations among outcomes at the individual patient level and, for those studies that use composite outcomes, how often the relations between component elements are depicted.

Methods: We selected 20 general, specialty and subspecialty medical journals with high impact factors that publish original clinical research. We identified every RCT in the 2011 and 2012 issues and randomly selected 10 articles per journal. For each article we recorded the number of outcomes, the number of composite outcomes and how often the relations between outcomes or elements of composite outcomes were portrayed.

Results: All but 16 of the 200 RCTs had more than one outcome. Thus, outcomes could have been related in 92% of studies, but such relations were only reported in 2 (1%). A total of 33 (17%) investigations measured a composite outcome, 32 of which showed data for each component. None, however, showed cross-tabulation of the components.

Conclusions: Readers are rarely shown the relation between outcomes. Mandatory posting of datasets or requirements for detailed appendices would allow readers to see these cross-tabulations, helping future investigators know which outcomes are redundant, which provide unique information and which are most responsive to changes in the independent variables. While not every relationship between outcomes requires depiction, at present such information is seldom portrayed.

Show MeSH