A new statistical method called the Causal Pivot has been developed by researchers from Rice University, Baylor College of Medicine, and Texas Children’s Hospital’s Jan and Dan Duncan Neurological Research Institute. This approach allows scientists to detect hidden genetic drivers in complex diseases and group patients according to the true biological causes of their illnesses. The findings were published in the American Journal of Human Genetics.
Complex diseases such as Parkinson’s, breast cancer, and high cholesterol can have different underlying causes among patients with the same diagnosis. Traditional large-scale genetic studies often average out these differences, making it difficult to identify specific genetic factors.
“Not everyone with a complex disease gets there the same way,” said Dr. Chad Shaw, lead author and professor of molecular and human genetics at Baylor College of Medicine, who also holds appointments at Rice University and the NRI. “The Causal Pivot is designed to detect those differences and sort patients into more precise, biologically meaningful subgroups. This is a foundational step toward truly personalized genetic medicine.”
The method builds on existing approaches in Causal Analysis, a field within statistical science taught by Shaw at Rice University.
The Causal Pivot uses polygenic risk scores (PRS) — summaries that reflect the combined effect of many common genetic variants — as a reference point for testing other possible disease causes such as rare DNA mutations. When a rare variant drives disease in some individuals, those carrying it typically have lower polygenic risk scores compared to others with the disease who do not carry it. The Causal Pivot formalizes this observation into a statistical test that can identify subgroups driven by rare variants.
Unlike standard genome-wide association studies that require both patient and control groups, this new method works even when only patient data are available. This makes it useful for studying rare subtypes or situations where healthy controls are unavailable. The team also included measures to reduce confounding from ancestry differences so results remain reliable across diverse populations.
To validate their approach, researchers used data from the UK Biobank — which includes information from over half a million participants — focusing on well-known gene-disease pairs: LDLR variants with high cholesterol (hypercholesterolemia), BRCA1 variants with breast cancer, and GBA1 variants with Parkinson’s disease. In each case, they found signals consistent with established biology while tests across unrelated diseases produced no false positives.
Further analysis in Parkinson’s disease revealed that people carrying multiple rare mutations in lysosomal storage pathway genes had lower polygenic risk scores than others with Parkinson’s. This suggests several rare mutations can combine to cause an alternative route into illness.
“This kind of subgroup detection changes the game,” Shaw said. “It opens the door to clinical testing and tools that can match patients to therapies based on the actual mechanism driving their disease, not just the name of the condition.”
The potential impact on personalized medicine could be significant. By identifying different genetic pathways leading to disease, doctors may improve how they select patients for certain tests or treatments. It could also allow clinical trials to focus on those most likely to benefit from specific therapies.
“Personalized medicine will be grounded in comprehensive genome sequencing for each patient, but it's hard for health care providers to understand because there are more than 5 million variants in each person's genome,” said Dr. John Belmont, adjunct professor at Baylor College of Medicine and co-author of the study. “Current interpretation workflows are rule-based, uncalibrated and don’t produce patient-level causal probabilities. Causal analysis, by placing the emphasis on the effects of interventions, is ideal for medical decision making. Dr. Shaw's new work is a step along the road to formally integrating information from both clinical and population cohorts. It introduces a new framework for modeling the competing factors in disease and will help doctors accurately weigh all the possibilities.”
“Most genetic studies of human disease have focused exclusively on either common or rare genetic variations –changes in the DNA code,” said Dr. Joshua Shulman, professor at Baylor College of Medicine and co-director of Duncan NRI.“Dr. Shaw’s team has created an elegant statistical model that helps unify these approaches, allowing for a more powerful, integrated analysis.We are excited to use this strategy to discover new genetic risk factors for Alzheimer’s and Parkinson’s disease.”
Researchers say future applications may extend beyond genetics; any known driver—such as environmental exposures or biomarkers—could serve as pivots within this analytical framework.
“Personalized medicine needs structure,” Shaw said.“We’ve created a clear,testable way to account for genetic diversity within a disease.It’s a tool we can use on large datasets today,and one we can adapt to the clinic tomorrow.”
Other contributors include C.J.Williams (Genetics & Genomics Services Inc.), Tao Tao Tan (Baylor College of Medicine), Daniel Illera,and Nicholas Di (Rice University). Funding was provided by several organizations including Genetics & Genomics Services Inc.,the Jan and Dan Duncan Neurological Research Institute,the Ting Tsungand Wei Fong Chao Foundation,and The Huffington Foundation.
