A HEDIS chart request asks for the pages the measure needs. A risk coder needs every page in the record.
Each makes sense for its own program. A HEDIS request is built to satisfy the measure, so it names the documents, the members, and the months quality needs. Risk review has no such limit, because every condition in the record counts and specificity matters on all of them.
Two different jobs with two different definitions of a complete chart. Most plans have quietly merged them into one retrieval operation, and the merge did not happen on neutral terms.
Consolidation happens on quality’s terms, and nobody decides it
The reasons to combine chart retrieval are good ones. Providers are tired of three different requests for the same records. Retrieval is expensive. Two teams pulling from the same offices in the same year creates abrasion that costs a plan more than the retrieval fees do.
The two programs do not run on the same cycle or the same population. HEDIS hybrid review is a seasonal pull, January through May, for a sample of a few hundred members per measure. Retrospective risk review runs most of the year against much larger suspect lists. The real overlap is the provider office, where the same practices get requests from both programs at the same time.
So the plan combines the chase. The question is whose specification survives, and it is almost always quality’s. Quality owns the retrieval calendar because HEDIS season sets it. Quality owns most of the vendor relationships, because that is where chart chase volume has historically lived. Risk adjustment arrives as a second consumer of a process that already exists.
The retrieval spec then inherits limits written for a different purpose. It asks for specific pages or document types, its member list comes from the HEDIS sample, and it runs on the January through May HEDIS calendar. The vendor pulls what that spec asks for, and risk coders can read only what comes back.
Nobody makes a bad decision anywhere in this. The savings appear in the same budget year and are easy to point at. Any cost appears the following year in a different department, and it arrives with no name on it.
The retrieval savings land this year in quality. Any recapture loss lands next year in risk, and nothing connects the two.
What it looks like a year later
Recapture moves every year. V28 changed which conditions count, tighter coding ahead of RADV audits changes what gets submitted, and provider mix shifts. A combined chase is one more variable, and it is the one most plans never check.
If suspects that used to close are coming back unconfirmed, the first question is whether the chart that came back was the chart risk adjustment needed. No coding tool can find evidence in pages that were never retrieved.
This is what makes the shared data problem expensive rather than merely inconvenient. The two scorecards are downstream of one decision, and the decision was made by whoever wrote the spec.
The CY2027 measure removals raise the pressure
In the CY2027 final rule, CMS removed 11 measures, mostly administrative and process measures. Most of those removals take effect with the 2029 Star Ratings. Two start with the 2028 ratings: Call Center - Foreign Language Interpreter and TTY Availability, and Statin Therapy for Patients with Cardiovascular Disease.
What remains is weighted further toward clinical outcomes, intermediate outcomes, and adherence. The measures that remain lean on clinical results, and the ones still reported through the hybrid method still depend on charts.
The consolidation conversation is going to happen at more plans next year, and it will be a well-reasoned conversation every time. The part worth getting right is which specification wins.
The question to ask this month
Do not ask for the retrieval volume report. Ask for the retrieval specification, and ask two things about it.
First, what does the spec request, and who wrote it? If the document types, the member list, and the calendar all trace to HEDIS, risk adjustment is reviewing records that were scoped for quality before they arrived.
Second, for charts retrieved under the shared process last year, what share produced a risk-relevant finding beyond the quality target? A plan that cannot answer that has no way to know what the consolidation cost, and most plans cannot, because the two results are stored in systems that were never joined at the member level.
Both questions can be answered in a week by a team that owns its own data.
The length of time it takes to get an answer is itself the finding.
One retrieval, one full read, two consumers
The answer is not two chart chases. Providers will not tolerate it and the economics do not support it.
The answer is that the record gets read once and completely, and both programs draw from the same evidence. The measure value and the condition documentation come out of the same note, on the same date of service, from the same provider, and both are traceable back to it.
That only works if the evidence layer is shared rather than the retrieval event. Sharing the retrieval while keeping two separate extractions gives a plan the abrasion savings and none of the analytic benefit, which is the arrangement most plans have now.
How this works at Invent Health
Risk Analytics identifies and ranks the members and conditions worth pursuing, and holds the member-level record both programs write back to.
Coder Workbench validates the clinical evidence behind every code. Our NLP surfaces evidence in unstructured documentation at 85 percent detection, rising to 90 percent once the feedback loop runs on additional charts, and coding accuracy holds at 95 percent across more than 100,000 charts coded. The read is of the whole record rather than of a targeted section of it.
Encounter Submissions carries the accepted record through to CMS and reconciles what was sent against what came back.
Quality Analytics reads the same evidence layer the risk side reads. One retrieval, one complete extraction, two programs drawing from it, and a single member-level record underneath both scorecards.
We run this across Medicare Advantage, Medicaid, and commercial populations for health plans, health systems, IPAs, and provider groups, covering more than 5 million member lives.
Before the retrieval plan gets built
Next year’s chart chase is being scoped right now, in most plans by the team that has always scoped it. That is the moment the request gets set, and it is far easier to influence a specification while it is being written than to investigate recapture twelve months after it takes effect. Ask us what a single complete read of the record produces for both programs.
Frequently Asked Questions
What limits risk adjustment in a combined chart chase?
The request. A combined chase is usually requested on HEDIS terms: specific pages or document types, a member list built from the HEDIS sample, and the January through May calendar. Risk coders can read every page that comes back, but only the pages the request asked for.
Is combining chart retrieval across quality and risk a mistake?
No. Combining retrieval reduces provider abrasion and cost, and both are real problems. The issue is which specification governs the combined process. A shared retrieval running on a quality specification delivers the savings and quietly limits what risk adjustment can find.
Why does the quality specification usually win?
Quality generally owns the retrieval calendar because HEDIS season sets it, and owns most of the chart chase vendor relationships because that is where the volume has historically lived. Risk adjustment joins a process that already exists rather than designing one alongside it.
How long before any effect shows up?
If there is one, usually a full year, which is what makes it hard to diagnose. The retrieval savings appear in the current budget year. Any recapture change appears in the next one, in a different department, with no event connecting them.
What would the effect look like on a risk adjustment scorecard?
Suspects that closed reliably in prior years returning unconfirmed, with coder volume and accuracy unchanged. Recapture also moves with V28, coding ahead of RADV audits, and provider mix, so the first check is whether the chart that came back was the chart risk adjustment needed.
Can NLP compensate for it?
Only within the pages it receives. Detection rates apply to the documentation that arrives. If the retrieval specification never requested the rest of the record, there is nothing upstream for any extraction tool to read, however good it is.
Do the 2027 measure removals change this?
They intensify it. CMS removed 11 measures, mostly administrative and process measures, in the CY2027 final rule. Most take effect with the 2029 Star Ratings and two start with 2028. The measures that remain lean on clinical results, and the ones still reported through the hybrid method still depend on charts.
What should a plan ask for specifically?
The retrieval specification rather than the volume report. Two questions about it: what the spec requests, including document types, member list, and calendar, and who wrote it. Then, for shared retrieval last year, what share of charts produced a risk-relevant finding beyond the quality target.
Why can most plans not answer the second question?
The quality abstraction result and the risk coding result are stored in separate systems that were never joined at the member level. Answering requires matching them chart by chart, which is why the answer usually takes weeks and several people.
What does sharing the evidence layer mean in practice?
The record is read once and completely, and both programs draw from that single extraction. The measure value and the condition documentation trace to the same note and the same date of service, which is what makes the two scorecards comparable rather than merely coincidental.
