A small practice can run population health analytics without a data team. Start with five questions: risk distribution, utilization, open care gaps, medication adherence and unengaged patients. Answer them from EHR exports, payer files, claims and ADT feeds in an AI workspace covered by a business associate agreement. Then sanity-check every total against its source.
Takeaways
Five questions cover most of it
Who is high risk, who uses the ER, which gaps are open, who isn't taking their meds, and who hasn't been seen.
You already have the data
EHR exports, payer rosters and gap reports, claims files and hospital ADT alerts.
Check before you share
Reconcile every count against the source file or payer portal.
Know what it isn't
An AI workspace is not a certified quality reporting tool. Official submissions still go through your EHR, registry or payer.
The questions that matter
Population health management means looking at your patients as a group, then acting on the ones who need it. In value-based care contracts, your results depend on it. You don't need a dashboard with fifty charts. You need answers to five questions, refreshed monthly.
1. Risk distribution
How many of your attributed patients are high, medium or low risk? Which PCP panels carry the most high-risk patients?
2. Utilization
Who went to the ER or was admitted in the last 90 days? Who has repeat ER visits?
3. Care gaps
Which quality measures have the most open gaps? Which patients have two or more?
4. Medication adherence
Who is below 80% PDC on diabetes, blood pressure or statin medications? Who stopped refilling?
5. Unengaged patients
Which attributed patients haven't been seen in 12 months or more? Who has never had an annual wellness visit?
PDC means proportion of days covered: the share of days a patient had medication on hand. The Pharmacy Quality Alliance uses 80% as the adherence threshold. See PDC in our glossary.
The data you already have
Most practices have more data than they think. It just lives in different places, in different formats.
- EHR exports. Patient demographics, problem lists, visit history, last A1c or blood pressure. Most EHRs export these as CSV or Excel reports.
- Payer files. Attribution rosters, care gap reports, and sometimes risk scores and pharmacy adherence lists. These usually arrive monthly by portal download or email.
- Claims files. Some payers and ACOs share claims-level data showing services patients got outside your practice, like specialist visits and ER use.
- ADT feeds. ADT stands for admission, discharge and transfer. Hospitals and health information exchanges send these alerts when your patient is admitted or discharged. See ADT notifications.
You won't have all four. Start with what you have. A roster plus an EHR visit export answers question 5 by itself.
How to answer them in an AI workspace
An AI workspace lets you upload these files and ask questions in plain English. It does the joins, counts and sorting that used to need a database person.
Because these files hold PHI, the tool must have signed a business associate agreement (BAA) with you. OpenAI says ChatGPT Free, Plus, Pro and Business aren't eligible services for PHI. Anthropic says its Team and individual plans can't enable HIPAA. Rivvi includes a HIPAA BAA on the free plan and logs PHI access. Patient data isn't used to train third-party models.
Here's how one question flows:
- 1
Upload the files
Upload the attribution roster, an EHR visit export and the payer gap report as-is. Rivvi enriches uploads against CMS and NPI registry data.
- 2
Ask the question
Which attributed patients haven't been seen in 12 months, grouped by PCP, with their open gaps and risk score?
- 3
Ask how it got there
How many patients were on the roster? How many matched to the visit export? How many didn't match, and why?
- 4
Save the logic
Ask the workspace to remember how you joined the files and defined each group. In Rivvi, that memory can be viewed, edited or removed by your team.
Good prompts for the five questions:
- "Split the roster into risk tiers using the payer's risk score. Show counts by PCP."
- "From this ADT file, list patients with two or more ER visits in the last 90 days, with last PCP visit date."
- "Which quality measures have the most open gaps? Which patients have two or more?"
- "From the payer's adherence file, list patients below 80% PDC by drug class."
- "Which attributed patients have no visit in 12 months and no AWV on record?"
Answer the five questions on your own data
Free for you and two colleagues, with a HIPAA BAA. No card. No clock.
A monthly routine one person can run
You don't need a full-time analyst. You need one person, a calendar reminder and about the same steps each month.
- Collect the files. Download the new roster and gap report from each payer portal. Run your standard EHR visit export. Save any ADT alerts from the month.
- Upload and reuse last month's logic. Ask the workspace to apply the same joins and definitions it used last time.
- Run the five questions. Same prompts, same order, every month. Consistency is what makes month-over-month comparisons mean something.
- Compare to last month. Which numbers moved? Did unengaged patients go down? Did open gaps on your worst measure close?
- Pick three actions. One outreach list, one workflow fix, one item for the provider meeting.
Keep a short running log of what you asked and what you found. When a payer changes its file format or a number jumps, that log helps you tell a data problem from a real change.
How to sanity-check the answers
AI can be confidently wrong. Build a habit of checking before anyone acts on a number.
- Row counts. Total patients in should equal total patients out, plus any it dropped. Ask it to list the dropped rows.
- Match rate. If only half the roster matched your EHR export, the join key is probably wrong. Member IDs and MRNs rarely match. Try name plus date of birth.
- Portal totals. Compare open gaps by measure against the payer portal. Small gaps in timing are normal. Big ones mean a mapping error.
- Spot checks. Pick five named patients and confirm them in the EHR.
- Date windows. Confirm "last 12 months" means the dates you think it means.
How to share the results
Analysis nobody sees doesn't change anything. Pick the format your team already uses.
- Excel. Ask for a clean export with one tab per question. Rivvi connects to Excel, SharePoint and OneDrive.
- Power BI. If your group already runs dashboards in Power BI, Rivvi has a Power BI connector through its Microsoft integration.
- Microsoft Teams. Staff can tag @Rivvi in a Teams channel to ask a follow-up question.
- A one-page summary. Ask for five bullets for your provider meeting: what changed, what's worst, what to do this month.
Then act. Unengaged patients and patients with multiple gaps go on an outreach list. Our guide to building a care gap outreach list walks through it. On Rivvi's Team plan, the AI can also call and text those patients from the same workspace.
What an AI workspace is not
Be honest with your team and your board about the limits.
- Not a certified quality reporting tool. MIPS, HEDIS and ACO quality submissions follow specific technical specs and certified pathways. Use your EHR, registry or payer for official reporting.
- Not the payer's official record. Only the payer closes a gap or sets a risk score.
- Not a replacement for clinical judgment. A risk tier is a starting point for a care team's review, not a clinical decision.
- Only as good as the files. Payer files lag. ADT feeds miss hospitals that don't send them.
What it does well: turning a pile of exports into answers your team can act on this week. For a full product view, see patient analytics.
Start with one roster and one question
Upload your files and ask. Free for you and two colleagues, with a HIPAA BAA.