Claims data tells you what was billed. EHR data tells you what was documented. Neither source alone gives you the complete picture your HEOR team, medical affairs function, or payer relations team actually needs.
When you link them, you can answer questions that neither dataset can answer in isolation: Did the product that was billed actually match the wound type documented? What was the care trajectory before and after a CTP application? How does real-world utilization in SNFs compare to hospital-based wound centers for the same indication?
Claims and EHR data linkage is the methodological backbone of credible real-world evidence in wound care. It is increasingly expected by MACs conducting coverage reviews, payers evaluating LCD compliance, and FDA reviewers assessing PMCF submissions. The 2026 skin substitute payment overhaul has only sharpened that expectation. When the MACs withdrew the unified skin substitute LCDs on December 24, 2025, coverage reverted to legacy MAC policies in three jurisdictions and to case-by-case “reasonable and necessary” review in the other four, while the reclassification of non-BLA products as incident-to supplies proceeded on schedule. Payment changed, coverage standards fragmented, and the WISeR prior authorization model went live in four states. Documentation alignment between what was billed and what was clinically justified is no longer a nice-to-have. It is an audit exposure question.
If your organization is still running claims-only analyses or relying on EHR data without billing context, you are working with a structurally incomplete dataset. The analytic gaps are not minor. Claims cannot capture wound size, tissue type, or prior treatment history. EHR data alone cannot confirm payer adjudication, reimbursement amounts, or cross-setting utilization patterns. The linked dataset is where defensible RWE actually lives.
This guide walks through the six steps required to execute a defensible claims-EHR linkage project in wound care: from defining your analytic question and assessing data source eligibility, through probabilistic or deterministic matching, variable harmonization, bias assessment, and final analytic output. Each step addresses the practical decisions and common failure points that determine whether your linked dataset holds up to scrutiny from a MAC, a payer, or a peer reviewer.
Step 1: Define Your Analytic Question Before Touching the Data
This step sounds obvious. It is also the step most frequently skipped or treated as provisional, and it is the single most common reason claims-EHR linkage projects fail to produce usable output.
The analytic question determines everything downstream: which claims sources you need, which EHR fields are required, what your unit of analysis is, and which care settings must be represented. A vague question produces a dataset that can answer many things weakly and nothing definitively.
Specificity is the requirement. “Product utilization by wound type and care setting” is a starting point, not a question. A workable question looks like: “Among Medicare Part B beneficiaries receiving CTP applications for DFUs in outpatient wound centers between Q1 2025 and Q2 2026, how does episode length and healing trajectory differ by product cohort when controlling for baseline wound severity documented in the EHR?”
That formulation immediately tells you what you need: Part B claims for HCPCS-coded CTP applications, ICD-10 codes for DFU, wound care EHR records capturing wound measurements and tissue classification at baseline, and a care setting filter for outpatient wound centers. It also tells you what you do not need: Part A SNF claims, Part D pharmacy data, home health records.
The unit of analysis must be specified before data acquisition. Is it the application event, the wound episode, the patient encounter, or the product SKU? Each choice produces a different denominator and a different answer. Comparative effectiveness analyses typically use the wound episode as the unit. PMCF tracking may use the application event. LCD compliance auditing may use the individual encounter. Mixing units mid-analysis is a methods error that peer reviewers and MAC auditors will identify.
Care setting matters structurally, not just analytically. SNF claims fall under Medicare Part A with different billing codes and documentation requirements than Part B outpatient wound center claims. Home health records exist in a separate claims stream entirely. If your question spans care settings, your data acquisition plan must account for each setting’s claims structure and EHR documentation norms separately.
The deliverable at this step is a one-page analytic protocol: primary endpoint, secondary endpoints, population definition, timeframe, care settings, and data sources required. Lock it before any data access request is submitted. Teams that treat the protocol as a living document during data acquisition typically discover mid-project that their linked dataset cannot answer the question they actually needed to answer.
Step 2: Assess Data Source Eligibility and Document Structural Limitations
Once the analytic question is locked, the next decision is which data sources can actually support it, and what each source structurally cannot provide. This is not a procurement conversation. It is a methods decision with direct consequences for the defensibility of your findings.
On the claims side, the relevant Medicare sources for wound care are well-defined. Part B captures HCPCS-coded CTP and CAMP applications, HBOT, and NPWT in outpatient settings. Part A captures SNF stays and inpatient encounters. Part D captures systemic antibiotic fills, which function as a proxy for wound infection events when EHR documentation is incomplete. Commercial claims add coverage for non-Medicare populations but introduce payer heterogeneity that complicates pooled analyses.
The HCPCS coding environment for CTPs has added complexity in recent cycles. Effective January 1, 2026, non-BLA skin substitutes are paid as incident-to supplies rather than as biologicals, which made the JW and JZ modifiers inapplicable to those products and made discarded product non-payable under any circumstance. Products marketed under a BLA continue to be paid as biologicals, and the modifiers still apply to them. Only the administered portion is now billable for everything else. That changes how application quantities appear in claims, and it changes them for some products and not others. If your analytic window spans January 1, 2026, your claims-derived utilization series has a structural break that must be accounted for explicitly, and so does any cost variable you derive from it.
On the EHR side, the distinction between wound care-specific EHRs and general EHRs is analytically significant. General EHRs capture encounter data but typically lack structured fields for wound measurements, tissue classification, exudate characteristics, and prior treatment history. Wound care-specific EHRs capture this clinical granularity in structured, queryable form. For wound care RWE, the difference between these source types is not marginal: it determines whether your linked dataset can answer clinical questions or only billing questions.
Temporal coverage alignment is a common failure point. Your claims window and EHR window must overlap sufficiently to capture the full wound episode, not just the billing event. A CTP application claim with no corresponding pre-application wound documentation in the EHR cannot support a trajectory analysis. Confirm the date range of both sources before committing to a study window.
Geographic and population concentration in the EHR source also requires assessment. If the EHR dataset is concentrated in specific regions or practice types, your findings may not generalize to the broader population implied by your analytic question. Document this explicitly.
Data use agreements, IRB requirements, and HIPAA-compliant de-identification protocols must be confirmed and documented before any linkage attempt begins. This is not administrative overhead. For PMCF submissions and peer-reviewed publication, the data governance documentation is part of the methods record. State data limitations explicitly in your analytic protocol: claims do not capture wound size, depth, or tissue type; EHRs may not capture payer adjudication data or post-discharge outcomes. An audience evaluating your work for regulatory or coverage purposes will look for this disclosure. Its absence signals methodological naivety.
Step 3: Execute the Linkage Without Introducing Systematic Bias
The linkage step is where the methodological rigor of the project becomes visible. Two approaches are available: deterministic and probabilistic. The choice is not a preference; it is determined by what identifiers exist in both datasets.
Deterministic linkage matches records on shared identifiers: patient MRN, date of birth, date of service, NPI, and procedure code. It requires both datasets to contain at least one common key with sufficient specificity to avoid false matches. Most wound care-specific EHRs can export encounter-level data with NPI and date-of-service fields, which makes deterministic linkage feasible for outpatient wound center data where both the claim and the EHR record reference the same provider and encounter date.
Probabilistic linkage is used when direct identifiers are absent or partially missing. Algorithms assign match weights based on combinations of demographic and encounter variables, and records above a defined threshold are treated as matches. This approach requires validation against a gold-standard subset where true match status is known. The validation step is not optional: an unvalidated probabilistic linkage is not a methods choice, it is an assumption.
Define match thresholds explicitly before running the algorithm. Post-hoc threshold adjustment to maximize match rate is a methodological red flag that peer reviewers and MAC auditors will identify. If your threshold was set after observing the match rate distribution, document why, and expect scrutiny.
Match rates will vary by care setting, and this variation is itself analytically meaningful. SNF claims and wound care EHR records typically have lower match rates than outpatient wound center data, reflecting documentation fragmentation across care settings. If your SNF match rate is substantially lower than your outpatient match rate, your SNF cohort may be systematically less representative of the true SNF population. That is a finding, not just a limitation.
Quantify and report unmatched records. What proportion of claims had no corresponding EHR record, and what proportion of EHR records had no corresponding claim? Systematic missingness concentrated in one care setting or patient subgroup signals a bias source that must be addressed in the analysis, not silently absorbed into the exclusion count.
Document the linkage algorithm, match rate by care setting, and all exclusion criteria in a methods appendix. For PMCF submissions and peer-reviewed publication, this documentation is non-negotiable. Build it as you go, not after the analysis is complete.
Step 4: Harmonize Variables Across Two Structurally Different Datasets
Claims and EHR data do not speak the same language. Claims encode procedures via HCPCS and CPT, diagnoses via ICD-10, and reimbursement via adjudicated amounts. EHRs encode clinical state via structured measurement fields, tissue classification schemas, and, frequently, free text that requires parsing. These representations do not map one-to-one, and the harmonization decisions you make here directly determine what your linked dataset can and cannot support analytically.
Build a variable crosswalk before running any analysis. For each analytic variable, specify the claims source field, the EHR source field, the harmonization rule, and the expected discordance rate. This document is your methods record for the harmonization step and should be treated as a formal deliverable, not a working note.
The wound care-specific harmonization challenges are well-defined. ICD-10 codes for DFUs identify laterality and complication status but do not capture Wagner grade or wound area. Those variables require EHR documentation. When your analytic question requires wound severity stratification, the ICD-10 code is insufficient, and the EHR field is the authoritative source. HCPCS codes for CTPs identify product categories but coding specificity varies by product and billing staff. For SKU-level utilization analysis, HCPCS codes alone are not reliable, and EHR documentation of the specific product applied must be the primary source.
Temporal alignment requires an explicit decision rule. A claims date of service reflects the billing date, which may differ from the application date documented in the EHR by one to several days depending on billing workflow. Define your temporal matching window before running the linkage, apply it consistently, and document it. A window of plus or minus three days is common in wound care linkage studies, but the right window depends on your care setting and billing lag assumptions.
Handle discordant records explicitly. When the product billed on the claim does not match the product documented in the EHR, do not silently drop these records from the analysis. Flag them as a separate analytic category. Discordance between billed and documented products is itself a finding with direct relevance to audit defense and LCD compliance. As documented in analyses of skin substitute audit patterns, billing-documentation misalignment is a primary driver of claim denials and recoupment actions. Your linked dataset may be the only source that can quantify this discordance at scale.
Validate harmonized variables against a random sample before running the full analysis. Pull a subset of matched records, verify the harmonized variable values against the source fields manually, and confirm that your crosswalk rules are producing the intended output. Errors discovered at this stage are recoverable. Errors discovered after the full analysis has run are not.
Step 5: Identify and Report Bias Sources Before Interpreting Findings
A linked dataset that has not been evaluated for bias is not an analytic asset. It is a liability. The audiences evaluating your work for PMCF submissions, payer dossiers, or peer-reviewed publication will assess your bias disclosure as a signal of methodological credibility. A clean narrative with no limitations section is more suspicious than a thorough one.
Selection bias is the first concern. The patients whose records successfully linked may differ systematically from those who did not link. In wound care, higher-volume practices with more complete documentation and more consistent billing workflows will produce higher match rates. This means your linked cohort may overrepresent well-documented patients in higher-volume settings. Characterize both the linked and unlinked populations on observable variables before proceeding to the primary analysis.
Informative censoring is a structural problem in wound care episode analysis. Patients who switch care settings mid-episode, from SNF to home health, or from outpatient wound center to inpatient, may drop out of one data stream while continuing to appear in another. If your claims source does not capture the receiving setting, or your EHR does not follow the patient across settings, the episode will appear to end when it did not. This distorts episode length, healing trajectory, and cost-per-episode estimates in ways that are difficult to detect without cross-setting data coverage.
Coding variability in HCPCS billing for CTPs and CAMPs is a documented source of measurement error. Utilization counts derived from claims may undercount applications, misclassify product categories, or reflect billing staff decisions rather than clinical decisions. Claims-derived utilization is a billing record, not a clinical record. Your EHR data is the corrective source, and discordance between the two is information, not noise.
EHR documentation completeness varies by clinician and practice workflow, even within wound care-specific EHR systems. Fields that are required by the system may still be populated inconsistently if clinical workflows allow workarounds. Assess completeness rates for your key analytic variables before treating EHR fields as authoritative.
Confounding by indication is the analytic challenge that claims-only studies cannot address and that linked datasets can only partially mitigate. Sicker patients with more severe wounds receive more aggressive products. Without risk adjustment using baseline wound severity variables from the EHR, comparisons across product cohorts will reflect patient selection as much as product effect. The linked dataset gives you the baseline clinical variables needed for risk adjustment. Use them.
Report all identified bias sources in the limitations section. Frame each one specifically: what the bias source is, which direction it likely pushes the estimates, and what steps were taken to assess or mitigate it. This is the standard expected by regulatory reviewers, payers, and peer reviewers alike.
Step 6: Structure Outputs That Hold Up to Scrutiny
The analytic output of a claims-EHR linkage project must be structured for its intended audience from the start. A PMCF report for EU MDR requires different structure, evidence standards, and disclosure language than a MAC coverage dossier, an HEOR manuscript, or a payer value dossier. Retrofitting output structure after the analysis is complete is inefficient and often produces documents that do not meet the applicable standard.
Identify the primary output format before the analysis runs and align your output structure to the applicable guidance document. For PMCF submissions, EU MDR Annex XIV Part B defines the required content elements. For FDA RWE submissions, the agency’s Real-World Evidence Framework and subsequent guidance documents establish expectations for data source characterization, linkage methodology disclosure, and bias assessment. For MAC coverage dossiers, the output must map directly to the LCD criteria being documented. Know which standard applies before you start drafting.
The core analytic outputs for wound care linked datasets are well-defined. Product utilization rates by wound type and care setting answer the market intelligence question. Episode length and healing trajectory by product cohort answer the comparative effectiveness question. Cost-per-episode analysis combining claims reimbursement with EHR-documented clinical outcomes answers the payer value question. LCD compliance rates, measured as the alignment between billed indication and documented clinical indication, answer the audit defense question. Intellicure Analytics’ skin sub cost versus healing analysis infrastructure is a practical example of how these output types can be generated from a pre-built linked dataset rather than constructed from scratch.
Stratify all primary findings by care setting. SNF, home health, and hospital-based wound center populations have different wound severity distributions, different product access patterns, and different billing structures. Pooling them without justification produces estimates that may not accurately represent any of the three settings. Your competitive intelligence team and payer relations team both need setting-specific outputs: the product that leads in outpatient wound centers may rank differently in SNFs, and that difference is strategically significant.
Include a data flow diagram in every output document: total claims records, matched EHR records, and analytic cohort after exclusions. Reviewers will ask for this regardless of the output format. Build it from the start of the project, not at the end.
State what the linked dataset cannot answer. If your EHR source does not capture outcomes beyond 90 days, say so. If your claims source does not capture Medicaid or commercially insured patients, quantify the coverage gap. Transparency about dataset boundaries is a credibility signal, not a weakness.
For organizations evaluating whether to build a linkage infrastructure from scratch or access an existing linked wound care dataset, Intellicure Analytics’ wound care-specific dataset combines structured EHR documentation with claims context across multiple care settings. The infrastructure for comparative effectiveness studies, product utilization tracking, and PMCF-grade output generation already exists. The more productive question for most organizations is whether their specific analytic question can be addressed within that existing infrastructure rather than whether to build it independently.
Your Linkage-Ready Checklist
Executed correctly, claims and EHR data linkage produces something neither source can provide alone: a dataset that connects billing events to clinical reality, at the encounter level, across care settings, with sufficient granularity to support regulatory submissions, payer negotiations, and competitive intelligence simultaneously.
The six-step sequence in this guide maps to six checkpoints that determine whether your project will produce defensible output or a dataset that raises more questions than it answers.
Analytic question locked: Primary endpoint, unit of analysis, care settings, and data sources specified before any data access request is submitted.
Data sources assessed: Claims sources and EHR sources evaluated for temporal coverage alignment, population representativeness, and structural limitations. Data governance documentation confirmed.
Linkage executed and documented: Deterministic or probabilistic approach selected based on available identifiers. Match thresholds defined pre-hoc. Match rates quantified by care setting. Unmatched records characterized and reported.
Variables harmonized: Crosswalk built before analysis. Temporal alignment window defined and applied consistently. Discordant records flagged as a separate analytic category, not dropped.
Bias sources assessed: Selection bias, informative censoring, coding variability, documentation completeness, and confounding by indication all evaluated and disclosed explicitly.
Output structured for audience: Format aligned to applicable guidance document before drafting. Data flow diagram included. Dataset limitations stated plainly.
The two failure points that most frequently derail linkage projects are starting data acquisition before locking the analytic question, and failing to document match rate and exclusion criteria prospectively. Both are avoidable with discipline at the protocol stage.
For organizations that need linked wound care data without building the infrastructure from scratch, Learn more about our services at Intellicure Analytics, where pre-built linked datasets and comparative effectiveness study infrastructure are available for wound care-specific analytic questions across CTPs, CAMPs, HBOT, and multiple care settings.