A hospital discharge summary tells you what happened. A predictive model tells you what is about to happen. That distinction is the difference between reactive care and proactive intervention.
Health care organizations sit on enormous volumes of clinical, claims, and operational data. The challenge has never been collecting it. The challenge is turning it into something a care manager can act on before a patient’s condition deteriorates, before a preventable ER visit occurs, or before a high-risk pregnancy goes unmonitored.
That’s the problem SAS Health was built to solve, not by replacing clinical judgment, but by giving clinicians and analysts the signals they need at the moments they need them.
Value-based care models are pushing payers and providers to do more than report on what already happened. CMS quality measures, HEDIS scores, and state Medicaid program requirements all demand that organizations identify risk early and intervene before costly outcomes materialize.
But early identification at scale requires more than dashboards and descriptive statistics. It requires models that learn from historical patterns, including demographic, clinical, behavioral, and socioeconomic factors, to generate individual-level risk scores that drive targeted outreach.
SAS Health, built on the SAS Viya platform, provides the data foundation for this work, including an industry standard health data model to simplify health data management and accelerate analytic discovery to make decisions confidently.
On top of that foundation, SAS Viya delivers a full suite of machine learning and AI tools accessible to both coders and non-coders alike.
Let’s walk through the end-to-end workflow that makes predictive modeling on SAS Health and SAS Viya practical, not theoretical.
1. Data integration and preparation (SAS Health)
SAS Health provides out-of-the-box (OOB) industry standard data models that help organizations unify clinical, claims, eligibility, and enrollment data in a single, governed data and AI environment. This gives teams a strong foundation for analytics without requiring them to build every data structure from scratch.
Just as important, the platform supports the ongoing preparation of new analytic-ready data assets beyond the initial ingestion layer. Using open source programming, SAS and SQL-based transformation, or no-code/low-code data preparation tools, teams can create additional curated datasets for ad hoc analytics, investigation, reporting, and model development. This gives organizations flexibility to extend the core data models with business-specific data products while preserving governance, repeatability, and alignment across analytic workflows.
And not every user preparing cohorts or exploring data has to be a data scientist. SAS Viya Copilot for Clinical Data Discovery uses AI-powered natural language search to let clinicians and researchers query health care data conversationally. Ask a question in plain language, get a report back, no code required. This accelerates the feedback loop between insight generation and clinical decision-making.
2. Predictive modeling: build your own or deploy industry accelerators (SAS Viya)
With the data loaded and available to analytic teams, SAS Viya gives you two paths to predictive modeling. You can either build custom models from scratch that are tailored to your population or deploy industry accelerators that are purpose-built for health care. Most organizations will use both.
Build your own with Model Studio
The SAS Viya platform provides a drag-and-drop interface for constructing full model pipelines, from preprocessing through supervised learning to post-processing, without requiring a single line of code.
A typical pipeline might look like this:
- Preprocessing: Imputation of missing values, transformation of skewed variables (e.g., BMI, weight gain), variable selection, and feature engineering.
- Supervised learning: Choose from gradient boosting, random forests, neural networks, logistic regression, support vector machines, and more.
- Post-processing: Ensemble methods that take a weighted vote or weighted average across multiple models for a final, more robust prediction.
You can build multiple pipelines side by side. For example, one pipeline might focus on tree-based models with variable transformations, while another uses regression models that require imputation and stepwise variable selection. SAS Viya lets you compare them head-to-head using metrics like the K-S statistic, ROC curves, or your own custom assessment criteria.
And if you are a coder? You can bring your existing Python or R models directly into the workflow and score them against SAS-native models in the same project.
Deploy industry accelerators
For common health care use cases, SAS also offers industry accelerators that can be used by customers to train models more efficiently:
- Medication Adherence Risk: Medication adherence has long been a measure of patient health outcomes and a key factor in regulatory quality assessments — including quality ratings for Medicare Advantage, Medicaid, and Exchange plans. SAS Medication Adherence Risk enables managed care organizations to identify where resources are needed for timely and targeted intervention, resulting in enhanced patient engagement, better health outcomes, improved quality metrics, lower health care costs, and a significant ROI.
- Document Analysis for Health Records: Automates the ingestion and analysis of unstructured health records to extract clinically relevant information quickly and accurately. Advanced OCR converts scanned documents and digitized medical records into structured, tabular data with optional annotated PDFs for review. An intuitive workflow standardizes the identification of key medical findings, improving consistency across reviews and reducing manual effort and variability.
- AI-Driven Entity Resolution: Streamlines decision-making and boosts operational efficiency by accurately identifying and consolidating entities across various datasets — including in health care, insurance, and public sector. This model pipeline offers robust data preparation, flexible fuzzy matching, and scoring to eliminate duplicates and inconsistencies, ensuring a single, accurate view of each entity.
3. Bias detection and explainability (SAS Viya)
Health care data carries inherent biases. Historical disparities in access, diagnosis patterns, and social factors are all embedded in the data your models learn from. SAS Viya helps you gain understandable and defensible insights with AI that deliver plain-language explanations of data, models and predictions, built-in bias monitoring and full auditability. The built-in bias detection capabilities let you flag variables for monitoring throughout the modeling lifecycle. This helps your models be accurate as well as equitable across all patient populations. SAS Viya helps you gain understandable and defensible insights with AI that deliver plain-language explanations of data, models and predictions, built-in bias monitoring and full auditability.
4. Model deployment and monitoring (SAS Viya)
Once you’ve identified your champion model, SAS Viya makes the path to production straightforward. You can publish and register a model at the push of a button, then Viya will help you monitor its performance over time and help you retrain it when you detect drift or bias. This operationalization step, which traditionally takes weeks of custom engineering, is built directly into the platform.
5. From predictive models to agentic AI
Predictive models generate risk scores, but a risk score sitting in a database doesn’t improve a patient’s outcome. The real impact comes from operationalizing those scores into automated, real-time decisions at scale.
SAS Viya provides the platform to make that possible. It enables organizations to combine predictive models, business rules, and large language model (LLM) capabilities into governed analytic and decisioning workflows that can execute at enterprise scale. Instead of handing a care manager a static list of high-risk patients, organizations can build logic that dynamically determines the right intervention, for the right patient, through the right channel, triggered automatically when a risk threshold is crossed.
With SAS Viya’s agentic AI capabilities, organizations can go even further by designing purpose-built AI agents that do more than score and flag. These agents can analyze context, recommend or initiate next steps, and support action across operational workflows by combining the rigor of deterministic analytics with the flexibility of large language models.
For health care, this means moving beyond dashboards and risk lists toward more autonomous, intelligent workflows, where the platform can help identify a high-risk patient, evaluate the appropriate intervention based on clinical rules and predictive scores, and support outreach through the right channel, all with built-in governance, explainability, and human oversight where needed.
SAS Viya’s approach to agentic AI is built on three core principles:
- Decisioning: Combine predictive models, business rules, and LLM reasoning into hybrid workflows that balance precision with adaptability.
- Human/AI balance: Configure the right level of autonomy for each task, from fully automated low-risk decisions to human-reviewed interventions where appropriate.
- Governance: Ensure outputs are explainable, auditable, and compliant, because in health care, trust is not optional.
This is what closes the loop. Data is integrated, prepared, modeled, deployed, and operationalized within SAS Viya allowing organizations to move from insight to action at scale, in real time, with the governance health care demands.
Predictive modeling and AI in health care isn’t about replacing the expertise of clinicians and care managers. It’s about equipping them with earlier, more precise signals so they can focus their attention where it matters most.
SAS Viya brings together the full analytics lifecycle on a single platform, from data ingestion and standardization, to model building and deployment, to real-time decisioning and agentic AI workflows that act at the point of care. It’s a production-grade data and AI platform with the governance, scalability, and interoperability that health care demands.
If you’re looking to move from retrospective reporting to prospective, model-driven care management, the platform is here. The data is waiting.
