Ann Intern Med. 2026 Jan 27. doi: 10.7326/ANNALS-25-01344. Online ahead of print.
ABSTRACT
BACKGROUND: In embarking on randomized clinical trials (RCTs), researchers can hypothesize that a more intensive treatment is better than a less intensive treatment (positive hypothesis) or that a less intensive treatment is similar or noninferior to a more intensive treatment (negative hypothesis). Researchers may design noninferiority RCTs (NI-RCTs) to support negative hypotheses and standard RCTs (S-RCTs) to support negative or positive hypotheses. Regardless of hypotheses, S-RCTs and NI-RCTs should produce consistent results when assessing similar participants, interventions, control, and outcomes.
OBJECTIVE: To compare effect estimates in S-RCTs with positive hypotheses versus NI-RCTs and in S-RCTs with negative hypotheses versus NI-RCTs.
DESIGN: Meta-research.
SETTING: 98 meta-analyses.
PARTICIPANTS: 468 RCTs, including 153 NI-RCTs and 315 S-RCTs (149 positive and 166 negative hypotheses).
INTERVENTION: S-RCTs as the exposure and NI-RCTs as the control.
MEASUREMENTS: The ratio of effect estimates between S-RCTs and NI-RCTs in each meta-analysis was combined across meta-analyses.
RESULTS: Standard RCTs with positive hypotheses produced effect estimates 1.47 (95% CI, 1.27 to 1.70) times larger than NI-RCTs; among RCTs rated as having low risk of bias for blinding, the ratio was 1.01 (CI, 0.70 to 1.45), whereas among those rated as having high or unclear risk of bias for blinding, the ratio was 1.81 (CI, 1.41 to 2.33). Standard RCTs with negative hypotheses did not produce statistically different effect estimates from NI-RCTs (ratio, 0.93 [CI, 0.84 to 1.03]).
LIMITATION: Findings may be limited by residual differences between S-RCTs and NI-RCTs in the same meta-analysis.
CONCLUSION: The researchers' hypotheses may bias the results of published RCTs, especially those with high or unclear risk of bias for blinding. The effect of researchers' hypotheses should be assessed in systematic reviews and clinical practice guidelines when RCTs addressing the same clinical question report conflicting hypotheses.
PRIMARY FUNDING SOURCE: The Shenzhen Municipal Government, Guangdong Province, China, and the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences.
PMID:41587480 | DOI:10.7326/ANNALS-25-01344