Reducing bias in research: Building better tools to combine study results

When researchers want to know whether something causes a health outcome, like whether a vitamin reduces the risk of heart disease, they often look at results from many different studies. This process, known as ‘triangulation’, is a way of combining evidence from multiple sources to make stronger conclusions.

A common method for combining results is meta-analysis, but this process usually assumes that all studies are very similar. But studies differ by the people included, the measures used and how reliable they are, and how data is analysed.

Some studies are more trustworthy than others because they have less ‘bias’ – errors in their method or data that can distort the truth. Traditional meta-analysis handles this by excluding studies, with very high bias, but more advanced methods try to adjust for this bias instead.

Project aims

We want to improve how bias is accounted for when combining results from different studies.

Current tools, such as the R package triangulate, adjust for different types of bias in a step-by-step order: first one type of bias, then another. While this may work for clinical trials, research suggests it may not be appropriate for observational studies, where multiple biases interact in more complex ways.

Our main aim is to test whether changing the order of bias adjustments leads to different conclusions and whether a new approach, called ‘simultaneous adjustment’, provides more reliable answers. We will use computer simulations and real-world examples (beta-carotene intake and heart disease, Mediterranean diet on breast cancer survival) to compare the 2 approaches.

What we hope to achieve

By developing a more accurate and user-friendly tool for bias adjustment, we aim to:

1. Help researchers combine high-quality evidence more effectively
2. Provide guidance on how future studies can be designed to reduce bias
3. Support new methodological advances in combining evidence across studies, so that decisions in health policy and clinical practice are based on the most reliable evidence