Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Covariate balancing & weighting web app (COBWEB): an online tool simplifying robust causal inference in observational studies

Markoulidakis, Andreas, Holmans, Peter ORCID: https://orcid.org/0000-0003-0870-9412, Pallmann, Philip ORCID: https://orcid.org/0000-0001-8274-9696, Busse, Monica ORCID: https://orcid.org/0000-0002-5331-5909 and Griffin, Beth Ann 2022. Covariate balancing & weighting web app (COBWEB): an online tool simplifying robust causal inference in observational studies. Journal of Neurology, Neurosurgery and Psychiatry 93 (S1) , A44. 10.1136/jnnp-2022-ehdn.114

Full text not available from this repository.

Abstract

Background Observational study impose challenges to make conclusions about causal relationships, requiring the use of statistical techniques to account for imbalance of confounders between treatment groups. Propensity score and balance weighting (PSBW) are useful techniques that aim to reduce these imbalances by weighting the groups to be as similar as possible on the observed confounders. Aims Although here are many methods available to perform PSBW, there is little guidance on their implementation on small sample sizes, which are a common limiting factor in HD research. Motivated by the Physical Activity and Exercise Outcomes in Huntington’s Disease (PACE-HD) study, which evaluated the impact of enhanced physical activity on the progression and severity of the disease, we explored the challenges of performing PSBW analysis with small sample sizes. Methods We have designed a user-friendly online tool, called the Covariate Balancing & Weighting Web App (CoBWeb), to enable non-specialist researchers to estimate the causal effect of treatment from observational data while minimising confounding bias using PSBW. Outcome The app implements the following five key steps: 1) evaluate overlap of the treatment groups, 2) obtain estimates of PSBW using multiple methods, 3) check for covariate balance and select the best performing method, 4) estimate the causal treatment effect, and 5) assess sensitivity to unobserved confounding, and comes with a tutorial using simulated data based on PACE-HD.

Item Type: Article
Date Type: Publication
Status: Published
Schools: Medicine
Centre for Trials Research (CNTRR)
Publisher: BMJ Publishing Group
ISSN: 1468-330X
Last Modified: 30 Nov 2022 09:05
URI: https://orca.cardiff.ac.uk/id/eprint/152638

Actions (repository staff only)

Edit Item Edit Item