How causal perspectives can inform problems in neuroimaging
Poster No:
1572
Submission Type:
Abstract Submission
Authors:
Eric Bridgeford1, Maya Mathur1, Brian Caffo2, Russell Poldrack1
Institutions:
1Stanford University, Stanford, CA, 2Johns Hopkins University, Baltimore, MD
First Author:
Co-Author(s):
Introduction:
The advancement of neuroscience has long relied on statistical methods to uncover patterns in complex biological systems and draw meaningful conclusions about brain function. Linear statistical models, particularly ANOVA and regression, form the backbone of this analytical approach, with a recent issue of Nature Neuroscience demonstrating their near-universal adoption-14 out of 15 articles employed these methods. While these tools have proven invaluable, there exists a fundamental tension between statistical correlation and causal inference, captured by the axiom "correlation does not imply causation."
Methods:
Our first methodological contribution is introducing causal graphs as a tool for analyzing neuroimaging variables and their relationships. We emphasize how causal identification-determining when we can validly infer cause-and-effect relationships from data-relies on key assumptions including conditional ignorability, positivity, consistency, and SUTVA (Hernan 2023) which can be discerned via these causal graphs (Pearl 2009). Using batch effect estimation as an example, we demonstrate how failing to account for certain variables (like socioeconomic factors) in these causal graphs can prevent valid identification of batch effects (Bridgeford 2024), as it violates these assumptions needed for causal inference (Figure 1A).
Our second methodological contribution is to delineate flaws in brain-behavioral studies for causal inference. We illustrate this via the head motion issue in neuroimaging, in which individuals with high head motion tend to have derivatives artificially biased with distance-dependent correlations. Through a causal lens, this materializes as head motion being a common outcome of the connectome and the behavioral phenotype of interest (Figure 1B). Strategies such as excluding high head motion individuals (Power 2014) yield collider biases, which arise from conditioning on common outcomes of the exposure and the outcome. Attempts to address this challenge instead explore reverse-causal relationships (Nebel 2022), which are not generally informative of causal relationships.
Our second methodological contribution is to delineate flaws in brain-behavioral studies for causal inference. We illustrate this via the head motion issue in neuroimaging, in which individuals with high head motion tend to have derivatives artificially biased with distance-dependent correlations. Through a causal lens, this materializes as head motion being a common outcome of the connectome and the behavioral phenotype of interest (Figure 1B). Strategies such as excluding high head motion individuals (Power 2014) yield collider biases, which arise from conditioning on common outcomes of the exposure and the outcome. Attempts to address this challenge instead explore reverse-causal relationships (Nebel 2022), which are not generally informative of causal relationships.
Results:
We explore the potential prominence of these biases using the ABCD study (Karcher 2021). For investigating differences across sites, we use the Fisher-Freeman-Halton test for binary covariates, and the Kruskal-Wallis test for continuous covariates. We find that many variables, such as racial or environmental factors, tend to be different across sites (Figure 2A). This has the implication that failures to incorporate these variables into batch effect correction models may yield identification violations and could not identify batch effects. We support these claims with simulation benchmarks (found in Bridgeford 2024) which illustrate the value of instead considering causal statistical methods.
For investigating the disproportionate inclusion/exclusion of particular demographic or behavioral groups introduced by conditioning on head motion for brain-behavioral studies, we use the Fisher exact test for binary covariates, and the Mann-Whitney U test for continuous covariates. We find that excluded individuals tend to overwhelmingly feature higher ADHD diagonstic indicators, and tend to disproportionately be members of traditionally underserved socioeconomic groups (Figure 2B + C). We illustrate via simulation that current efforts to estimate effects under these types of scenarios yield substantial biases that are unaddressed by current analytical techniques.
For investigating the disproportionate inclusion/exclusion of particular demographic or behavioral groups introduced by conditioning on head motion for brain-behavioral studies, we use the Fisher exact test for binary covariates, and the Mann-Whitney U test for continuous covariates. We find that excluded individuals tend to overwhelmingly feature higher ADHD diagonstic indicators, and tend to disproportionately be members of traditionally underserved socioeconomic groups (Figure 2B + C). We illustrate via simulation that current efforts to estimate effects under these types of scenarios yield substantial biases that are unaddressed by current analytical techniques.
Conclusions:
This work addresses a fundamental issue in neuroimaging-attributing causality-and advocates for the development of robust causal frameworks for more widespread adoption of neuroimaging methodologies. We conclude by making recommendations, including mega-study design considerations and potential methodological approaches from causal inference (supported by our simulations), for potential future avenues whose adoption may prove fruitful.
Modeling and Analysis Methods:
Methods Development 1
Multivariate Approaches
Other Methods 2
Keywords:
Data analysis
Modeling
Statistical Methods
Other - causal inference
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Hernán, M. A. (2023). What If: Causal Inference and Counterfactuals in Epidemiology. Chapman and Hall/CRC.
Karcher, N. R. (2021). The ABCD study: understanding the development of risk for mental and physical health outcomes. Neuropsychopharmacology, 46(1), 131-142. https://doi.org/10.1038/s41386-020-0736-6
Nebel, M. B. (2022). Motion-related artifacts in structural brain images revealed with independent estimates of in-scanner head motion. NeuroImage, 247, 118849.
Pearl, J. (2009). Causal inference in statistics: An overview. Statistics Surveys, 3, 96-146.
Power, J. D. (2014). Studying brain organization via spontaneous fMRI signal. Neuron, 84(4), 681-696.