Reading the hospital data stack: EHR clinical data versus claims

• Nursing flowsheets with hourly vitals
• The operative note describing the surgical approach
• Medication administration records (the exact times each dose was given)
• Free-text progress notes from three providers
• The physical therapy evaluation and daily sessions

On the UB-04, all of that collapses into structured codes:

• A DRG (Diagnosis Related Group), the payment category Medicare and many insurers use to reimburse the whole stay as a bundle. A major joint replacement maps to a specific DRG family.
• ICD-10-CM diagnosis codes (why she needed care: osteoarthritis of the knee)
• ICD-10-PCS procedure codes (what was done, in inpatient coding language)
• Revenue codes grouping charges (operating room, pharmacy, physical therapy)
• Total charges and dates of service

Notice the compression. Days of clinical narrative become perhaps a dozen codes. That compression is where the two datasets diverge.

Why the same admission produces two different truths

1. Purpose shapes content

The EHR records clinical reality. The claim records billable, defensible reality. If a nurse documents a symptom that never becomes a formal diagnosis, it lives in Epic but never reaches the claim.

2. Coding is an interpretation, not a copy

A trained medical coder (or increasingly, computer-assisted coding software) reads the chart and translates it into codes. That translation involves judgment. Did the fever count as a documented complication? Only if a physician wrote a diagnosis that supports it. This is why hospitals invest in CDI (Clinical Documentation Integrity) teams whose job is to ensure the chart supports accurate coding.

3. Timing differs

Epic data exists in real time. The claim is assembled after discharge, sometimes weeks later, and may be corrected and resubmitted. So on any given day, the two systems disagree simply because the claim has not caught up.

4. Granularity differs

Epic knows the implant serial number for a device recall. The claim does not care. Conversely, the claim carries the final DRG that drives revenue, a field that does not naturally "live" in the clinical chart.

A concrete data example

Imagine you pull both records for Mrs. Alvarez. Simplified, they look like this:

Same encounter_id links them. But the EHR has 214 vitals rows and an implant serial; the claim has a DRG and revenue codes. Neither is "complete." A hospital analyst who wants the full picture must join both.

Note the DRG detail: 470 versus 469 differs by whether the stay had an MCC (Major Complication or Comorbidity). If Mrs. Alvarez's post-op fever had been documented as a qualifying diagnosis, the stay might have grouped differently and reimbursed at a higher rate. That single documentation decision moves money. This is exactly why CDI exists, and why regulators scrutinize it.

Why leaders constantly reconcile the two

Hospital executives make decisions from both datasets, and the gaps cause real problems.

Quality reporting. Public quality programs often use claims because claims are standardized and available across all hospitals. But claims miss clinical nuance. A patient who was genuinely sicker may look like a "complication" in claims data if documentation was weak. The Agency for Healthcare Research and Quality maintains widely used claims-based quality indicators that illustrate both the power and the limits of billing data.

Revenue integrity. If the EHR shows care that never made it onto the claim, the hospital may be under-billing. If the claim shows more than the chart supports, that is a compliance risk. Reconciliation protects both.

Population health and value-based care. In arrangements where hospitals are paid for outcomes, leaders need clinical detail (from the EHR) matched to cost detail (from claims). Neither alone is enough.

The interoperability layer

A modern development matters here: FHIR (Fast Healthcare Interoperability Resources), the industry standard for exchanging healthcare data via APIs. Federal rules now require certified EHRs to expose data through FHIR APIs, which makes it easier to pull structured clinical data programmatically rather than scraping notes. You can explore the standard at HL7 FHIR.

FHIR helps on the clinical side. It does not replace claims. The two data models remain distinct because their purposes remain distinct. FHIR just makes the clinical half more accessible for analytics.

How analysts actually use each dataset

Think of it as choosing the right tool.

Use claims when you need:

• Comparability across hospitals or over time (everyone codes to the same standards)
• Total cost and reimbursement per stay
• Broad denominators for population trends

Use the EHR when you need:

• Clinical detail: lab values, medication timing, vitals trends
• Real-time or near-real-time signals (sepsis alerts, deteriorating patients)
• Device tracking, allergy checks, or anything the claim never carries

Use both, joined, when you need:

• Risk-adjusted outcomes (clinical severity from the EHR, cost from claims)
• Revenue integrity checks
• Value-based care performance

A practical trap for new analysts: assuming a claims count equals a clinical count. "How many patients got a blood thinner?" is an EHR question. The claim only shows a pharmacy revenue code, not each dose. Ask the claim a clinical question and you will get a confidently wrong answer.

A note on the human systems behind the data

The divergence is not a flaw to eliminate. It reflects two legitimate workflows: caring for patients and getting paid for that care. The reconciliation work (coders, CDI specialists, revenue integrity analysts, and increasingly automated tools) is a permanent feature of hospital operations, not a temporary bug.

For a non-technical leader, the key insight is this: when someone shows you a "hospital metric," your first question should be "is this from the clinical record or the claim?" The answer changes what the number can and cannot tell you.

Key Takeaways

• One admission, two datasets. The EHR (Epic, Oracle Health) records clinical reality; the UB-04 claim records billable, coded reality. They diverge by design.
• Coding is interpretation. Human coders and software translate charts into ICD-10 and DRG codes. Documentation quality (the job of CDI teams) directly affects both accuracy and reimbursement.
• Match the dataset to the question. Use claims for comparability and cost, the EHR for clinical detail and real-time signals, and both joined for risk-adjusted outcomes.
• Always ask the source. Before trusting any hospital metric, determine whether it came from the clinical record or the claim. That single question reframes what the data can prove.
• FHIR helps, but does not merge the two. Interoperability standards make clinical data more accessible, yet claims remain a separate model built for payment.