You are here
Frequency and nature of changes in primary outcome measures.
Objective Prespecification of outcome measures forms the basis of most statistical analyses of clinical trials. Trial registration allows for the tracking of changes to outcome measures from study initiation to eventual publication. Changes may occur any time after study initiation, though there is no standard for distinguishing important vs unimportant changes. The objective of this study was to determine the frequency and type of changes in primary outcome measures (POMs) between entries in ClinicalTrials.gov and associated publications and between initial and current registry entries.
Design We identified 75 sequential MEDLINE citations with ClinicalTrials.gov entries, generating "registry-publication pairs." Study 1 (50 pairs) compared the registry POM to outcomes in the publication. Study 2 (25 pairs) compared the publication POM to the registered outcomes. Study 3 examined changes over time for all 75 registry POMs. Each outcome measure was coded for "principality" (primary, secondary, or unspecified), domain (eg, depression), specific measure (eg, HAM-D), and time frame. Pairs were considered matches if the domain was the same. If either specific measure or time frame were not consistent, pairs were "substantively different."
Results Sixty-two of the 75 registry-publication pairs were consistent, though differences in level of specificity within pairs were observed (Table 9). One publication POM had been registered as a secondary outcome measure. Eight of 50 Study 1 pairs and 3/25 Study 2 pairs had substantive differences. Six of 75 Study 3 POMs changed substantively after initial registration. Forty-eight of 75 of the ClinicalTrials.gov entries were initially registered more than 3 months after the start date, with some delayed by years.
Table 9. Results of 2 Studies Characterizing primary Outcome measures (pOms)
a Two POMs were considered a match if they had the same domain (eg, pain).
Conclusions The taxonomy enabled us to categorize POM pairs. Most POM pairs were consistent based on our criteria. Our ability to detect inconsistencies was limited at times by vague registry entries or substantially delayed initial registrations. This taxonomy could be used to develop consensus criteria for tracking and communicating outcome measure changes.