Ignoring Bilateral Conditions Can Bias Orthopedic Research Outcomes

Research study background

Orthopedic specialists regularly treat bilateral musculoskeletal conditions, and interventions can create interdependence between limbs (within subject correlation). However, many clinical studies treat each limb as an independent observation and fail to account for this limb interdependence. This can lead to conclusions that underestimate standard errors, increase the rate of false positives and overstate treatment effects in orthopedic research.

Here, experts from the Orthopedics Institute at Children’s Hospital Colorado and the University of Colorado Anschutz Medical Campus evaluated three analytical strategies to determine how well each accounted for bilateral conditions in orthopedic clinical trials. They compared the naïve method, which treated both limbs as independent; the random method, which analyzed one randomly selected limb per patient; and the linear mixed model (LMM) method, which accounted for the correlation between an individual’s two limbs.

Using a hypothetical randomized trial for knee osteoarthritis, the team tested each method under different rates of bilateral involvement (prevalence of patients with bilateral conditions ranged from 10%–100%) and varying correlations between limbs. They also compared outcome scores at baseline and at two years across simulated treatment and untreated (control) groups. The team modeled two scenarios: in the first, there was no true difference among treatment groups (the intervention was ineffective), and in the second, the mean difference was 10 points (treatment effect was clinically meaningful). The simulations assessed for bias (the difference between true and observed), power, type 1 error (false positive) rate and 95% confidence interval (CI) coverage.

Bias was comparable across all three methods. Compared to the random and LMM methods, the naïve approach consistently produced inflated false positives (11.3% vs. 4.9% and 5.0%, respectively) in the first scenario and poorer CI coverage (87.8% vs. 95% and 95.1%, respectively) in the second scenario. Naïve model performance worsened as bilateral prevalence and within subject correlation increased. While both the random and LMM methods performed similarly well and appropriately accounted for correlated data, the LMM method was slightly more statistically powerful when differences were present, especially as the prevalence of bilateral cases increased.

Relevance to future research

This study, which aligned with previous findings, demonstrated that failing to account for bilateral conditions can meaningfully increase the risk of false-positive findings in orthopedic clinical trials and potentially misinform clinical decision-making.

“We believe strong, thoughtful methods are critical for translating evidence into clinical practice,” says lead study author Patrick Carry, PhD, assistant professor in the Department of Orthopedics at the University of Colorado Anschutz Medical Campus. “Our goal is to make this topic educational and practical, giving clinical researchers the tools they need to tackle a common challenge in orthopedic clinical research.”

The study authors shared recommendations for improving the accuracy and reliability of orthopedic clinical studies. Researchers should clearly report the prevalence of bilateral participants, describe how bilateral data are handled, and use the LMM method to analyze continuous data in studies that include subjects with bilateral involvement.