Patient Data to Actionable Cohorts with Snowflake Intelligence
Healthcare analytics has long struggled with a familiar problem: data exists, but insight is slow. Analysts write SQL. Clinicians wait for dashboards. Leaders make decisions with lagging indicators.
With Snowflake Intelligence, that dynamic changes. At ArisData, we are helping organizations move from static reporting to conversational, cohort-driven analytics, where insights emerge in seconds, not weeks.
This blog walks through a real example using patient health data and demonstrates how Snowflake Intelligence enables meaningful cohort analysis using natural language.
The Problem: Cohort Analysis Is Powerful but Hard
Cohort selection is at the heart of population health, clinical research, and preventive care. But traditionally, it requires:
- Complex SQL logic
- Deep knowledge of schemas and metrics
- Iterative back-and-forth between business and data teams
As a result, many valuable questions never get asked. What if a clinician, analyst, or healthcare leader could simply describe the cohort they want to study?
The Dataset: Patient Health and Diabetes Risk
Using a structured patient dataset containing attributes such as:
- Demographics like age, gender, ethnicity, education, income
- Lifestyle factors like smoking, physical activity, sleep, diet
- Clinical indicators like BMI, blood pressure, cholesterol, glucose, HbA1c
- Risk scores and diabetes diagnosis status
ArisData modeled this data in Snowflake and enabled it for Snowflake Intelligence.
The Prompt: A Real Cohort Study in Plain English:
Here is the exact prompt entered into Snowflake Intelligence:
Identify patients aged 35–60 who are not currently diagnosed with diabetes but have a diabetes risk score above 25 or HbA1c between 5.7 and 6.4. From this cohort, analyze the impact of lifestyle factors by grouping results by physical activity level (low, medium, high) and smoking status, and report average BMI, fasting glucose, and progression risk indicators.
I successfully identified a cohort of 19,061 patients aged 35–60 who are not currently diagnosed with diabetes but have either a diabetes risk score above 25 or HbA1c levels between 5.7 and 6.4. Here’s the comprehensive analysis of lifestyle factors and their impact on progression risk indicators.
What Snowflake Intelligence Does Behind the Scenes:
Snowflake Intelligence automatically:
- Understands clinical thresholds like HbA1c ranges
- Interprets cohort logic and exclusions
- Translates intent into optimized Snowflake SQL
- Applies semantic understanding of metrics and dimensions
- Returns structured, analyzable results
This turns Snowflake into more than a data warehouse. It becomes an insight engine.
Identifying High-Risk Diabetes Patients by Lifestyle Factors:
Here is the list of Female Asian Patients under age 50 with Graduate-level Education from the Cohort. The query returned 223 female Asian graduate patients aged 35–49 who are not currently diagnosed with diabetes. But have a diagnostic risk score above 25 or HbA1c between 5.7-6.4
Female Asian Graduate Patients Under Age 50
Gender | Ethnicity | Age | Education Level | Diagnosed Diabetes | Diabetes Risk Score |
Female | Asian | 37 | Graduate | 0 | 53 |
Female | Asian | 44 | Graduate | 0 | 47 |
Female | Asian | 49 | Graduate | 0 | 47 |
Female | Asian | 43 | Graduate | 0 | 46 |
Female | Asian | 43 | Graduate | 0 | 46 |
Female | Asian | 42 | Graduate | 0 | 46 |
Female | Asian | 46 | Graduate | 0 | 44 |
Female | Asian | 47 | Graduate | 0 | 44 |
Female | Asian | 49 | Graduate | 0 | 44 |
Female | Asian | 49 | Graduate | 0 | 43 |
Female | Asian | 48 | Graduate | 0 | 42 |
Female | Asian | 47 | Graduate | 0 | 42 |
Why This Matters for Healthcare and Life Sciences:
This type of cohort is especially powerful because it focuses on early intervention:
- Patients not yet diagnosed
- Elevated risk indicators
- Lifestyle-driven factors that can be modified
These insights can support:
- Preventive care programs
- Value-based care initiatives
- Population health analytics
- Clinical research and real-world evidence studies
Most importantly, it lowers the barrier to asking better questions.
ArisData’s Point of View:
At ArisData, we believe the future of analytics is:
- Conversational, not query-driven
- Semantic, not schema-bound
- Proactive, not reactive
Snowflake Intelligence allows healthcare organizations to unlock value from their data faster, while maintaining governance, performance, and scale. This is not about replacing analysts. It is about amplifying decision-making across the organization.
What’s Next
With this foundation, organizations can easily extend to:
- Longitudinal cohort tracking
- Risk stratification and segmentation
- Personalized intervention analysis
- AI-assisted clinical insights
And it all starts with a single question.
Want to see this in action with your own data?
ArisData helps healthcare and life sciences organizations design, model, and operationalize Snowflake Intelligence at scale. Let’s turn your data into decisions.
Comprehensive Analysis of Lifestyle Factors and Their Impact on Progression Risk Indicators
Cohort Overview by Lifestyle Factors:
The analysis reveals significant patterns across different combinations of physical activity levels and smoking status.
Physical Activity Categorization:
- Low: < 150 minutes/week
- Medium: 150–300 minutes/week
- High: > 300 minutes/week
Key Findings:
1. Physical Activity Impact:
High Activity Groups show the best metabolic profiles:
- Lowest average BMI: 24.4–24.7
- Lowest fasting glucose: 101.7–103.0 mg/dL
- Lowest diabetes risk scores: 21.8–23.2
Low Activity Groups show concerning risk markers:
- Highest average BMI: 25.6–25.8
- Highest fasting glucose: 104.5 mg/dL
- Highest diabetes risk scores: 29.9–30.2
2. Smoking Status Patterns
- Smoking status shows minimal impact compared to physical activity
- Current smokers tend to have slightly lower BMI but similar glucose levels
- Never smokers in high-activity groups show optimal profiles
3. Population Distribution
- Low activity groups represent the largest cohort: 13,871 patients (73%)
- Medium activity groups: 4,437 patients (23%)
- High activity groups: 753 patients (4%)