Risk Adjustment and HCC Coding Software Problems: The 7 That Cost Real Money, and How to Avoid Them

Why does chart retrieval scope get underestimated so often?

Because retrieval is described as a vendor relationship and quoted as an integration. In reality it is an orchestration problem whose difficulty is set by your provider mix, not by anyone's software. A plan whose membership sits with three large health systems on one electronic health record can pull most of what it needs through direct connections. A plan with a long tail of independent practices, federally qualified health centres and behavioural providers is dealing with fax, portal downloads, on site retrieval and offices that treat a records request as an imposition.

The visible symptom is a status report saying 41,000 charts were requested and 28,000 received, with the gap concentrated exactly where the documentation is hardest to get. The invisible symptom is that chase priority is being set by a vendor's generic suspecting rather than by your own knowledge of which providers document well.

Scope it as orchestration from the start. One record of every attempt against every chart, across every channel, with the source, the date and the outcome. Chase priority computed from expected yield, which is a function of the member's suspected conditions, that provider's historical documentation quality with you, and the cost of the channel. That last input is the one no vendor holds, because it is your history with your providers, and it is the reason retrieval strategy is worth owning even when the physical retrieval stays outsourced.

What goes wrong when you migrate historical abstraction and chart data?

Your own abstraction history is the best input to suspecting you will ever have, and it is usually the worst prepared dataset in the project. It lives in a vendor platform, keyed to that vendor's identifiers, with coder decisions recorded as accept or reject and little else. The reason a condition was accepted, the passage it rested on and the guidance version in force are often absent, because the original system never asked for them.

Migrate that as it stands and you inherit a pile of outcomes with no reasoning attached. It is still useful for suspecting, and it is useless for defence. Plans discover the difference during their first self audit, when a diagnosis from two years ago cannot be tied to a page.

Handle it in two tiers. Migrate the full history for analytics, keyed to your member and encounter identifiers rather than the vendor's, and accept that older records carry no evidence chain. Then set the cutover date after which every abstraction carries the complete record: coder, decision, the documentation criteria met, the page image reference, the passage and the guidance version. Be explicit with your compliance team about which years are defensible and which are not, because that boundary will be tested and it is far better as a known fact than a discovery.

Why do EHR, health information exchange and submission integrations break after launch?

Clinical connections break because each one is a separate legal, technical and data quality arrangement that moves at the provider's pace. A FHIR bulk export that worked in testing changes when the practice upgrades. A health information exchange changes its participation terms. A provider who supplied structured documents starts supplying PDFs because their scanning vendor changed. None of that is a defect in your system and all of it stops your pipeline.

The submission side breaks differently. Encounter data submission has its own validation, and rejections arrive with reason codes that are meaningful only if somebody works them as a queue. Plans that summarise rejections as a rate rather than working them individually leave accepted revenue on the table and, worse, leave deletes unsubmitted.

Design for both. Treat a provider who can only supply PDFs as a supported path rather than a workaround, because that path will always exist somewhere in your network. Version every connection's expected shape and alert on drift rather than on failure, since silent partial delivery is more common than an outright outage. And carry one lineage identifier from chart to abstraction to submission to response, so reconciliation is a query rather than a quarterly project in a spreadsheet.

What happens when RADV evidence packaging is not covered?

It gets deferred because it is not needed this year, and then it is needed with a deadline. The task is to produce, for each sampled member and hierarchical condition category, the medical record page supporting it, signed and dated by a valid provider, from a valid encounter type, inside the correct service window. If charts sit in a retrieval vendor's repository, abstractions in a coding platform, submissions in an electronic data interchange system and provider validation in a credentialing database, assembling that package takes weeks per audit and the assembly itself introduces errors.

The deeper problem is that assembling under deadline tells you nothing you could have acted on. You find out your error rate at the moment you can no longer change it.

Build it as generation, not assembly. Every accepted diagnosis already points at the page image, the passage, the signature block, the provider's validated credential at the date of service and the encounter type, so a package is a query. Prepare a one best record selection in advance for your high value conditions rather than under pressure. Then run self audits quarterly on your own samples using the same rules an auditor would use. That produces your real error rate rather than your coder accuracy rate, and the gap between those two numbers is where repayment risk lives.

Should you build custom or configure what you already own?

Under roughly 20,000 risk adjusted lives, stay with Reveleer or Inovalon and spend the money on certified coders and on getting clinicians to document properly at the point of care. A build at that scale is a distraction from the two things that actually move your accuracy.

Keep buying retrieval even if you build everything else. Physical and on site chart retrieval is a logistics business with field staff and provider relationships, and Inovalon, Cotiviti and Reveleer are genuinely good at the industrial parts: outreach cadence, on site scheduling, image intake. Recreating that is effort with no return. Similarly, if your gap is the encounter pipeline itself rather than the round trip, Edifecs is strong there and configuring it is the cheaper answer.

Build when two of these are true. Your provider mix is fragmented enough that retrieval strategy materially affects revenue and no vendor's generic prioritisation reflects it. You run more than one line of business and maintain parallel processes for each, since Medicare Advantage, commercial risk adjustment through the EDGE server and state Medicaid arrangements are effectively three calculation engines with three calendars. You cannot trace a single diagnosis from chart to payment. Or you have been through an audit and found that assembling evidence took weeks. That last one is the honest trigger for most plans that come to us, and it is a good one.

How do hidden costs get into the quote?

Risk adjustment quotes go wrong in five predictable places.

• Each clinical connection is priced as one integration. Every electronic health record and health information exchange carries its own authentication, data quality quirks and legal agreement, and the agreements move at the provider's pace rather than the project's.
• Multi line support is quoted as configuration. Each risk model is a separate calculation engine with its own calendar and submission path, not a setting.
• Model version handling is assumed. The transition from CMS-HCC version 24 to version 28 changed which conditions map to a payment category, so the system has to hold two versions at once during a phase in year and treat a mapping update as a data load with a validation run.
• Coder staffing models are ignored. A mix of internal coders and outsourced partners in the same tool needs separate permission structures and separate audit trails.
• Historical migration is treated as a copy. Rekeying a vendor's identifiers to yours, and deciding which years carry an evidence chain, is real work with a compliance dimension.

The honest bands from Digital Heroes delivery experience are $90,000 to $190,000 over 14 to 20 weeks for a first release covering retrieval orchestration, a coder workspace with two way review, model versioned mapping and diagnosis level submission reconciliation, and $250,000 to $600,000 across 8 to 14 months for a full platform adding prospective suspecting, provider facing gap workflows, multi line support and evidence packaging with self audit.

What separates a build that works from one that fails here?

Symmetry between adds and deletes, first. A review programme that structurally favours adding conditions is the pattern federal enforcement has repeatedly focused on, and a plan that cannot demonstrate it removes unsupported diagnoses with equal diligence has an exposure that coding accuracy does not fix. Build both paths the same way, measure them the same way, and make sure deletes flow through submission, because an unsubmitted delete is a documented exposure you failed to act on.

Second, the coder sees the extracted passage and the surrounding page image together. Validating a snippet is not validating a chart, and the difference shows up in an audit rather than in a productivity report.

Third, model version is data. If a mapping change requires engineering work in your system, you will be late in every year CMS moves the model, and lateness compounds into a suspecting list built on last year's economics.

Fourth, start with reconciliation. It is the fastest payback in the category because it usually surfaces submitted and rejected revenue nobody was working, and it funds the rest of the programme.

Fifth, settle ownership before kickoff, in writing: the repository, the cloud environment and every byte of abstraction history. At Digital Heroes the client owns all of it from the first commit. For a data asset that doubles as your audit defence, any other arrangement is a risk you are choosing to accept.