13 Oct 2026
Independent validation can strengthen confidence in research findings. ADSI uses different validation approaches across analyses, depending on the data and research questions.
For analyses involving linked UW Health electronic health records and Wisconsin birth records, an additional challenge arises: how can an external team independently evaluate the analysis without direct access to protected health data?
How does it work?
Our approach is designed around two separate secure computing environments.
The external validation team will be able to:
- Review the analysis code
- Initiate a validation run
- Review approved, non-sensitive results
The protected data:
- Remain in a separate secure environment
- Are accessible only to authorized researchers
- Are not directly accessible to the external validation team
Separate secure computing environments allow independent validation of the analytic workflow while protecting access to linked health data.
Why separate the environments?
The separation allows independent researchers to evaluate whether the analytic workflow performs as intended without needing access to the protected linked health data themselves.
Together with synthetic-data testing and cybersecurity review, this approach is designed to balance:
Independent Validation • Reproducibility • Data Privacy
23 Sep 2026
Linked UW Health electronic health records and Wisconsin birth records allow ADSI to examine our research questions in a different, real-world data environment. SEED provides detailed research and clinical information not normally found in health records, while the linked records allow us to examine similar questions using information collected through birth records and routine health care.
Linking Wisconsin birth records with UW Health electronic health records provides a different data setting for examining ADSI research questions.
Together, the linked data allow us to:
- Study autism in a different data setting by linking Wisconsin birth records with health information collected through routine care over time.
- Examine early-life factors alongside health information collected across different stages of life.
- See whether patterns observed in detailed research data like SEED are also reflected in large-scale clinical and administrative records.
- Examine changes over time, including how changes in early-life factors may relate to changes in how common autism is.
18 Sep 2026
CDC’s Study to Explore Early Development (SEED) is particularly well positioned to support causal inference research in autism. Conducted across multiple phases, SEED has built a rich and deeply characterized research resource that brings together a large and diverse sample, detailed information on prenatal and early-life factors, medical histories, rigorous clinical assessments, and well-characterized comparison groups.
These features provide an unusually strong foundation for moving beyond observed associations toward causal questions: under clearly stated assumptions, what can relationships in the data tell us about factors that may contribute to autism and its heterogeneity?
Through ADSI, we are building on the depth of SEED by applying modern causal inference methods to examine these questions and strengthen how observational autism data are interpreted.
SEED brings together detailed information on early-life factors, medical histories, clinical assessments, and comparison groups, providing a rich resource for examining causal questions in autism research.
12 Sep 2026
Understanding what may contribute to autism is challenging because early-life factors do not occur in isolation. Health conditions, medications, environmental exposures, family characteristics, and social conditions can be interconnected, making it difficult to determine what an observed relationship actually means. Finding that a factor is associated with autism is therefore only the beginning of the scientific question.
We are using causal inference to examine these relationships more carefully. Rather than asking only whether two things occur together, we ask what the data can tell us about the potential contribution of a particular factor when other relevant explanations are considered. This requires researchers to think carefully about which factors should be included, how they may be connected, and what assumptions are necessary to draw conclusions from observational data.
Causal inference does not turn observational research into a randomized experiment or provide simple proof of cause and effect. Instead, it provides a more structured and transparent way to investigate causal questions. By making assumptions explicit and open to examination, we hope to strengthen both the evidence we produce and the conversation about what that evidence can—and cannot—tell us.
Causal inference helps researchers move beyond observed associations by examining how factors may be connected, what alternative explanations should be considered, and what assumptions are needed to interpret the evidence.
