About the Project
Randomised controlled trials (RCTs) are the gold standard for evaluating new interventions. This is due to randomisation preventing selection bias and balancing out any known/unknown confounders that may cause systematic differences between trial arms. RCTs are not always feasible however, and it is increasingly common for researchers to access the growing accessibility routine datasets to evaluate the effect of non-randomised exposures to treatments. Such approaches are therefore subject to…
selection bias and confounding. One approach to address this is using propensity score weighting (PSW) which aims to create a weighted population where the covariate distributions are similar between exposure groups being compared.
In cluster randomised trials (CRTs), groups of individuals are randomised (e.g. hospitals). With this design, statistical complexities may arise due to the presence of clustering violating the traditional assumption of independence. Moreover, it is not uncommon to design CRTs with a small number of clusters where both unbiased and efficient estimations may be at risk.
The proposed project will investigate how propensity scores should be estimated in CRTs with a small number of clusters in the presence of selection bias, and when and how propensity score weighting (PSW) should be used to estimate average causal effects in such settings. This project will provide data analysis guidelines to researchers and contribute to improving the quality of evidence drawn from small CRTs with selection bias.
BACKGROUND: Using RCT data, researchers may want to investigate the effect of exposures that are not themselves randomised and therefore subject to selection bias and confounding. An example of such exposures is the treatment or intervention actually received by participants if there is non-adherence to treatment following randomisation. To address causal questions in the presence of selection bias, causal methods such as PSW have been developed. PSW helps create a weighted population where the covariate distributions are similar between the groups of the exposure being compared. PSW can be used to estimate the average causal effect of a treatment.
CRTs have been increasingly used to evaluate complex interventions. CRTs offer some practical advantages such as administrative convenience and improved adherence to treatment. In spite of these advantages, statistical complexities may arise due to the presence of clustering violating the traditional assumption of independence. Out of 123 CRTs published in peer-reviewed journals in 2011, half are made of 12 clusters or less per trial arm. The literature is sparse on how to use PSW in small CRTs settings to estimate causal effects either at the individual or cluster level.
OBJECTIVES: To provide guidelines as to when and how PSW may be used to estimate average causal effects in small CRTs where there is selection bias.
NOVELTY: In the presence of a small number of clusters, some advocate the use of cluster-level analysis where inferences target the clusters. However, researchers may be interested in making inferences at the individual level. The proposed research will help estimate causal effects at both cluster and individual levels and guide researchers in their causal analysis steps in small CRTs settings.
EXPERIMENTAL APPROACH: First, a systematic review of causal inference methods implemented in CRTs with a small number of clusters to address selection bias will be conducted. Next, methods for estimating the average causal effect by PSW in CRTs with a small number of clusters where there is selection bias will be developed and tested via extensive simulation studies. Both continuous and binary outcome variables will be considered. Inferences at both cluster and individual levels will be investigated. The developed methods will then be applied to real CRTs data on improving quality of care for obstetric emergencies in West Africa. Finally, a statistical methodology framework will be proposed as to when and how a PSW approach may be used to estimate average causal effects in small CRTs where there is selection bias.
POTENTIAL IMPACT: This project will facilitate data analysis of small CRTs by researchers interested in the causal effects of exposures other than the randomised interventions and thus improve the data analysis practice for small CRTs with selection bias. Subsequently, this may contribute to a proposed standardised approach for analysis of small CRTs with selection bias that might result in an update of the CONSORT guidelines. Improving the quality of data analysis is critical for the benefit of participants, to allow reliable evidence to be available for informed decisions making.
Interested students should send a copy of the CV and cover letter to Dr Schadrac Agbla schadrac.agbla@liv.ac.uk. The post will close as soon as a suitable candidate is identified.
Funding Notes
This is a fully funded PhD, covering tuition fees, bench fees and living expenses.
References
- Altman DG, Bland JM. Treatment allocation in controlled trials: why randomise? British Medical Journal. 1999;318(7192):1209-1209.
- Hirano K, Imbens GW. Estimation of causal effects using propensity score weighting: An application to data on right heart catheterization. Health Serv Outcomes Res Methodol. 2001; 2:259–78.
- Hayes R, Moulton L. Cluster randomised trials. Chapman and Hall/CRC. 2009.
- Pirracchio R, Resche-Rigon M, Chevret, S. Evaluation of the propensity score methods for estimating marginal odds ratios in case of small sample size. BMC medical research methodology. 2012;12:1-0.
- Agbla SC, DiazOrdaz K. Reporting non-adherence in cluster randomised trials: a systematic review. Clinical Trials. 2018;15(3):294-304.
- Chang TH, Nguyen TQ, Lee Y, Jackson JW, Stuart EA. Flexible propensity score estimation strategies for clustered data in observational studies. Statistics in Medicine. 2022; 10;41(25):5016-32.