A Belgian Implementation Blueprint for WHO ICD-11/WHO ICHI-Based MDCs and DRGs in Relation to SNOMED CT
Abstract
A defensible Belgian pathway to Major Diagnostic Categories (MDCs) and Diagnosis Related Groups (DRGs) based on WHO ICD-11 and WHO ICHI could be built as a three-layer architecture: SNOMED CT for point-of-care clinical semantics, WHO ICD-11 Mortality and Morbidity Statistics (MMS) and WHO International Classification of Health Interventions (ICHI) for classification of diagnoses and interventions, and a transparent national grouper that consumes standardized episode variables and produces reimbursement classes. That architecture should not assume the existence of an official global SNOMED CT→WHO ICD-11 or SNOMED CT→WHO ICHI reimbursement-grade map. WHO states that the outcome of WHO ICD-11/SNOMED CT mapping collaboration remains under discussion and is not guaranteed, while WHO’s ICD-11 licensing materials also state that mappings and crosswalks are not covered by the base WHO ICD-11 licence and may require separate written agreement. SNOMED International documents official maps to WHO ICD-10, but describes WHO ICD-11 alignment as still in progress. At the same time, Belgium’s current national hospital activity and financing environment remains anchored in ICD-10-BE (US ICD-10-CM/ICD-10-PCS) and APR-DRG, and Belgian policy analysis has stressed that the country still lacks compulsory patient-level cost collection for national DRG pricing. The correct conclusion is therefore not to “flip” Belgium directly into WHO ICD-11/WHO ICHI reimbursement, but to build a phased national transition architecture with governed local maps, a dual-coding bridge, a transparent grouper specification, and a shadow-weight phase before statutory payment use.
1. Introduction: the Belgian problem is institutional, not merely terminological
The first premise is conceptual. WHO ICD-11 and WHO ICHI are classifications; SNOMED CT is a clinical terminology. WHO’s ICD-11 Reference Guide explicitly treats WHO ICD and terminologies as complementary, notes that the WHO ICD-11 Foundation is the terminological component of WHO ICD-11, and explains that tabular classifications derived from the Foundation must remain mutually exclusive and jointly exhaustive. That distinction matters for reimbursement: a terminology is optimized for clinical expression, while a payment grouper needs bounded, versioned, mutually exclusive episode classes. WHO also notes that WHO ICD is used for casemix, resource allocation, and quality uses, which makes WHO ICD-11 an appropriate classification substrate for DRGs, but not a complete DRG design in itself.
The second premise is practical. WHO ICHI is a classification of health interventions built from Target, Action, and Means axes, and WHO is explicit that WHO ICHI is not designed as a fee schedule; it also does not encode provider or setting, while reasons for intervention and outcomes belong elsewhere in the WHO family. That means an ICHI-based DRG design cannot simply “price” WHO ICHI codes one by one. It needs a national resource-significance layer that identifies which interventions are grouping-relevant, which are ancillary, and which are better handled as carve-outs or add-on payments.
The third premise is Belgian. Belgium currently uses ICD-10-BE (US ICD-10-CM/ICD-10-PCS) in hospital data collection and a federal grouper that derives APR-DRGs; KCE’s recent reform work describes a payment system still dominated by budgetary hospital financing and fees, while also noting that Belgium lacks compulsory patient-level cost collection. KCE further states that Finhosta is the only cost source covering all Belgian hospitals, but that it was not designed to measure costs of individual medical activities and leaves substantial discretion in accounting practice. Accordingly, a Belgian WHO ICD-11/WHO ICHI DRG system can only be credible if it separates two tasks: first, episode classification; second, later and more cautiously, reimbursement-weight estimation.
2. The correct relation between SNOMED CT, WHO ICD-11, and WHO ICHI
The most robust national design is to keep SNOMED CT at the point of care and to keep reimbursement grouping downstream of a classification layer. In other words, the Belgian grouper should never take raw SNOMED CT concepts as its legal payment input. It should take a frozen national episode abstraction built from WHO ICD-11 MMS diagnosis clusters, WHO ICHI intervention codes, and a small set of additional grouping variables such as age, discharge status, ICU use, length of stay, transfer status, selected devices or therapies, and episode-level complexity flags. This is not only architecturally cleaner; it is also legally safer, because Belgium’s National Release Centre distributes SNOMED CT under managed licensing arrangements, and Belgian public FHIR guidance already warns implementers that terminology assets such as SNOMED CT and LOINC may have licensing conditions more restrictive than the guide itself.
This architecture also reflects the present state of documented maps. WHO’s ICD-11 FAQ states that the outcomes of WHO–SNOMED collaboration on mapping WHO ICD-11 and SNOMED CT are still under discussion and not guaranteed. SNOMED International, for its part, documents official maps to WHO ICD-10 and says WHO ICD-11 alignment is in progress. WHO’s ICD-11 licence materials additionally state that mappings and crosswalks between WHO ICD-11 and other classifications or terminologies are not covered by the base WHO ICD-11 licence and require separate agreement. Therefore Belgium should assume four mapping statuses rather than one: an official externally maintained map where one exists; a nationally curated Belgian map; a local implementation map; and “no safe automatic map,” in which case human coding or coding-assist workflows remain necessary.
That caution is especially important for procedures. WHO Family of International Classifications (WHO-FIC) mapping guidance shows that mapping interventions classifications to WHO ICHI routinely produces one-to-one, one-to-many, many-to-one, and many-to-many relationships; it is a normal national task, not an anomaly. Belgium should therefore not pretend that a single global SNOMED CT→WHO ICHI crosswalk will solve procedural grouping. Instead, it should build domain-limited Belgian ConceptMaps and subsets for validated clinical areas, while leaving direct WHO ICHI coding available wherever map coverage is weak or ambiguous.
3. How Belgium should construct WHO ICD-11/WHO ICHI-based MDCs
A Belgian WHO ICD-11/WHO ICHI grouper should begin with a conventional DRG superstructure: a Pre-MDC layer for exceptional high-resource and cross-cutting episodes, followed by organ-system and special-population MDCs. That recommendation is not a claim about any existing official WHO ICD-11 Belgian MDC list; it is a design proposal consistent with the logic of DRG systems and with Belgium’s current familiarity with APR-DRG-style severity stratification. In practice, the Pre-MDC should capture episodes whose resource logic overrides ordinary organ-system assignment, such as transplants, ECMO or prolonged ventilation, major burns, multiple trauma, highly specialized neonatal cases, and other nationally specified exceptional episodes. Belgium’s present APR-DRG environment already uses severity subclasses, so retaining a recognizably hierarchical case-mix structure would lower transition costs.
The ordinary MDC assignment rule should then be diagnosis-led. The grouper should store the full WHO ICD-11 cluster as coded, including sanctioned postcoordination, but it should derive from that cluster a reduced grouping representation that is stable and auditable. This is crucial because the WHO ICD-11 Foundation is rich and linkable, whereas payment grouping requires mutually exclusive categories. Concretely, Belgium should maintain a national `DiagnosisTo MDC` dictionary that maps permitted WHO ICD-11 grouping abstractions to one MDC only, with explicit precedence rules where the same clinical story could otherwise fall into multiple body systems.
A plausible Belgian MDC set would include, at minimum, nervous system, eye, ear/nose/throat, respiratory, circulatory, digestive/hepatobiliary/pancreatic, musculoskeletal, skin and breast, endocrine and metabolic, kidney and urinary tract, male reproductive, female reproductive, pregnancy/childbirth/puerperium, neonates and perinatal conditions, blood and immunological disorders, mental health, and injuries/poisoning/trauma, with a clear policy choice needed on whether certain infectious-disease clusters remain organ-system based or form special cross-cutting groups. The essential rule is that MDCs must remain clinically interpretable, exhaustive for in-scope episodes, and stable enough for year-on-year reimbursement use. That is a normative design judgement built on the role of WHO ICD in casemix and on the need for mutually exclusive classification outputs.
4. How Belgium should construct DRGs within MDCs
Within each MDC, DRG assignment should proceed in three stages. First comes partitioning into intervention-based versus medical groups. Here WHO ICHI is indispensable, but insufficient by itself, because WHO states that WHO ICHI is not a fee schedule and does not encode provider or setting. Belgium should therefore author a national `ICHIResourceSignificance` dictionary that classifies each relevant WHO ICHI stem code into at least four operational classes: major operating-room intervention, minor operating-room intervention, non-operating but high-resource intervention, and ancillary or non-grouping intervention. That dictionary should be evidence-based and versioned.
Second comes base-DRG selection. If the episode contains one or more grouping-relevant interventions, the grouper should choose a dominant intervention using a national dominance hierarchy. Dominance should be determined not merely by chronology, but by expected incremental resource consumption and clinical centrality to the admission. If no qualifying intervention exists, base-DRG selection should revert to the reduced principal WHO ICD-11 diagnosis abstraction. In formulaic terms, Belgium should implement a rule sequence of the form: `Pre-MDC override → MDC assignment → dominant intervention test → intervention DRG or medical DRG`. That logic is much more transparent than a black-box grouper and is compatible with Belgium’s current federal practice of centrally calculating case-mix outputs rather than relying on hospital-submitted DRGs.
Third comes severity or complexity splitting. Belgium is already familiar with APR-DRG’s four severity-of-illness (SOI) and four risk-of-mortality (ROM) levels. For an WHO ICD-11/WHO ICHI national grouper, I would recommend retaining four payment complexity levels in the pilot because that continuity will ease interpretation and simulation, but deriving them from open Belgian logic rather than proprietary APR internals. The payment subclassifier should be driven by observed incremental cost, not by mortality risk as such. Accordingly, Belgium should maintain a separate analytical risk-of-mortality output if desired, but reimbursement severity should be based on a national `Complexity` model combining selected secondary WHO ICD-11 diagnoses, ICU use, ventilation, selected high-cost therapies or devices, neonatal markers, transfer patterns, and other cost-predictive variables.
The key discipline is that postcoordination richness and SNOMED CT detail should be preserved in source data and analytics, but only a bounded subset of clinically and economically relevant abstractions should feed payment grouping. In operational terms, Belgium should distinguish three map-driven workflows: automatic assignment for a small set of high-confidence SNOMED-origin concepts; coder-assisted candidate generation for medium-confidence mappings; and manual coding for the remainder. No national reimbursement logic should depend on a universal unvalidated SNOMED CT crosswalk.
5. A concrete national data model
The data model should have five linked layers.
First, a terminology-asset layer should register every versioned code system used by the programme: WHO ICD-11 MMS release, WHO ICHI release, SNOMED CT International edition, Belgian SNOMED CT edition and namespace, legacy ICD-10-BE, and any nationally maintained value sets or subsets. The Belgian terminology server infrastructure already provides an institutional anchor for this layer (CSIRO Ontoserver).
Second, a mapping layer should store national ConceptMaps and grouping abstractions with explicit provenance and licence status. At minimum, each record should contain source code or expression, target code or grouping abstraction, map type (`exact`, `broader`, `narrower`, `partial`, `none`), any rule expression, authoring body, validation status, effective dates, and redistribution status. This layer is essential because WHO explicitly distinguishes mappings from the base WHO ICD-11 licence.
Third, an episode-clinical layer should contain one record per hospital episode plus linked diagnosis and intervention tables. The diagnosis table should store the clinician-designated main condition, coder-validated principal diagnosis abstraction, all secondary diagnoses, timing or present-on-admission indicators where available, hospital-acquired condition flags, and the full WHO ICD-11 cluster representation. The intervention table should store WHO ICHI stem and extension codes, date/time, theatre flags, anaesthesia flags, and dominance ranking variables. Separate tables should hold ICU days, ventilation, implants, high-cost drugs, transfers, discharge status, and neonatal or obstetric markers. Belgium should also retain legacy ICD-10-BE and APR-DRG outputs in parallel during transition. That bridge is mandatory because Belgium’s statutory environment is still ICD-10-BE/APR-DRG.
Fourth, a patient-level costing layer should record direct and allocated costs by standardized buckets: ward nursing, ICU, operating room time, recovery, imaging, laboratory, pharmacy, blood, implants and devices, allied health, hotel services, and overheads, with a separate marker for whether a cost component is inside or outside the future DRG bundle. Because Finhosta operates at cost-centre level rather than episode level, Belgium should require reconciliation between patient-level costing submissions and audited hospital financial totals rather than pretending current accounting data already solve episode costing.
Fifth, a grouper-output layer should store MDC, base DRG, payment complexity level, analytical mortality-risk output if used, inlier or outlier status, transfer adjustments, carve-outs triggered, and a machine-readable audit trail of every grouping rule fired. That audit trail is critical for coding governance and for legal defensibility in reimbursement disputes.
A note on DRGs and patient-level costing in Belgium
Patient-level costing means assigning the cost of care to the individual patient episode - for example, one inpatient stay, one day-case, or one outpatient encounter - rather than only knowing average costs at the level of a ward, department, or hospital. In other words, it asks: “What did this specific patient’s care actually cost the hospital?”
For DRGs, patient-level costing is important because it gives the empirical basis for estimating relative weights. Once many patient episodes are costed, you can group them into DRGs and calculate the average or trimmed average cost of each group relative to the overall mean. That is how a country moves from a clinical grouping system to reimbursement weights that reflect actual resource use. In the Belgian context, this matters because KCE has explicitly noted that Belgium does not currently have compulsory collection of patient-level cost data nationwide, which is one reason a move to robust cost-based DRG pricing has to be staged carefully.
Belgium does not have one missing technical tool for patient-based costing; it has a misfit between its accounting architecture, physician-fee visibility, and the data burden required for a compulsory national episode-costing system. Belgium’s core hospital accounting infrastructure, especially Finhosta, was built primarily for hospital budget financing and cost-centre reporting, not for assigning costs to the individual patient episode or even to a single medical act.
A second reason is lack of uniformity across hospitals. Even though hospitals are legally required to report costs within the Finhosta structure, KCE found that the framework still leaves “a lot of freedom” in how hospitals record costs.
A third reason is insufficient granularity. KCE found that many medical activities are recorded in broad buckets such as “other medical-technical services,” and that there is often no one-to-one relationship between clinical units used in direct cost studies and the standardized cost centres in Finhosta.
A fourth reason is poor visibility of physician-related costs, especially in Belgium’s mixed hospital/physician payment arrangements.
A fifth reason is governance and implementation burden. Belgian stakeholders told KCE that registration requirements were already heavy, that multiple authorities asked for overlapping data, and that hospitals were not adequately compensated for coding and data work.
A sixth reason is that Belgium’s payment system could keep moving without true patient-level costing, which reduced the pressure to solve the harder problem.
Belgium did not miss patient-level costing because nobody understood the concept. It stalled because the country’s hospital accounting system was designed for another purpose, the recorded data were not uniform or granular enough, physician-cost visibility was incomplete, nationwide collection remained voluntary rather than compulsory, and the administrative burden of doing it properly was high. The result is a series of useful costing studies and reform steps, but not a mature, mandatory national patient-level costing platform.
6. The grouping dictionaries Belgium must author
The minimum national dictionary set is straightforward. Belgium needs a diagnosis-to-MDC dictionary for WHO ICD-11 grouping abstractions; an WHO ICHI resource-significance dictionary; an intervention dominance hierarchy; a national WHO ICD-11 complication/comorbidity dictionary for payment complexity; a special-episode dictionary for transplants, burns, trauma, neonates, obstetrics, and ventilation-intensive cases; and a carve-out dictionary for cost items that should remain outside DRG weights during the maturity phase, such as exceptionally volatile drugs, devices, or nationally earmarked missions not yet attributable at patient level. None of these assets can be safely inferred from undocumented global maps.
Belgium should also maintain a legacy bridge dictionary linking the current ICD-10-BE/APR-DRG world to the experimental WHO ICD-11/WHO ICHI world. The purpose of that bridge is comparison, not automatic legal equivalence. During the shadow phase, every in-scope admission should receive both outputs: the current statutory APR-DRG and the proposed WHO ICD-11/WHO ICHI DRG. That is the only serious way to study case-mix drift, coding burden, distributional effects, and clinical face validity before any payment decision.
7. How Belgium should determine reliable and robust DRG weights
The correct principle is to separate technical cost weights from policy reimbursement. Technical weights should estimate relative resource use. Policy reimbursement can then apply budget neutrality, explicit add-ons, outlier rules, and carve-outs. This separation is especially important in Belgium because hospital financing mixes budgetary and fee-based elements, because patient-level cost data are not yet compulsory nationally, and because existing national cost sources do not cleanly attribute all financing layers to episodes.
The technical weight formula should be simple and transparent::
Weight g= Trimmed mean standardized inlier cost for DRG g / Trimmed mean standardized inlier cost for all in-scope DRGs
The numerator should use standardized patient-level episode costs after exclusion of policy carve-outs and after application of national cost-accounting rules. The denominator should use the same standardized cost basis across all in-scope inlier episodes. Payment should then be determined as `base rate × weight`, plus transfer rules, outlier payments, and explicitly defined add-ons. That design keeps relative weights about resource intensity, not hidden policy subsidies. This recommendation follows the logic of national DRG pricing models and is consistent with Germany’s annual recalculation of cost weights from hospital cost-accounting data and Australia’s stability-oriented national pricing framework.
For Belgium, the decisive issue is data maturity. Because KCE has documented that patient-level costing is not yet compulsory and that Finhosta was not built for activity-level costing, the first national WHO ICD-11/WHO ICHI weights should be shadow weights, not statutory reimbursement weights. The shadow phase should rely on a representative hospital costing sample using a common costing manual and reconciliation to audited accounts. Those shadow weights can be compared with Belgium’s existing annual APR-DRG/severity cost averages in MORSE-type analyses, but they should not be mistaken for final legal tariffs.
Robustness requires explicit stabilization rules. Australia’s national pricing model provides a useful benchmark: it uses defined inlier bounds, removes extreme cost outliers, pools current and prior years for low-volume groups, and constrains year-to-year movement of some low-volume price weights. Belgium should adopt the same philosophy. In practice, that means: define inlier episodes using a pre-specified stay-based and cost-based rule set; analyze outliers separately; pool multiple years for low-volume DRGs; merge clinically similar very-low-volume groups instead of publishing spurious weights; and impose controlled weight-movement limits unless a clinically justified regrouping explains the change. Belgian evidence on APR-DRG outliers also supports treating high-cost outliers as a distinct analytical problem rather than forcing them into ordinary weights.
Note: Data maturity in healthcare refers to an organization's readiness and capability to effectively collect, manage, analyze, and utilize data to improve patient outcomes, operational efficiency, and strategic decision-making. It represents a journey from reactive, siloed data usage toward proactive, predictive, and even AI-driven insights that foster continuous improvement.
8. Governance, licensing, and publication rules
The governance structure should be split into four permanent bodies. A Classification and Grouper Board should own MDC and DRG logic. A Terminology and Mapping Board should own SNOMED CT subsets, ConceptMaps, WHO ICD-11/WHO ICHI grouping abstractions, and map validation. A Costing and Pricing Board should own the costing manual, weight estimation, outlier policy, and annual recalibration. An Audit and Data Quality Board should own coding audit, costing audit, inter-rater reliability, upcoding surveillance, and release assurance. Belgium already has the institutional pieces to anchor such a model in FPS Public Health, NIHDI/RIZIV-INAMI, the Belgian SNOMED CT National Release Centre, eHealth terminology infrastructure, hospital federations, and academic centres.
Licensing should shape publication policy. Belgium should publish its grouper logic, variable definitions, release notes, quality indicators, and annual methodological reports openly. But it should distribute licensed terminology artefacts, and especially any WHO ICD-11 mapping assets whose redistribution is contractually limited, through controlled channels or a secured terminology service rather than assuming that every table can be posted publicly. This is a direct consequence of WHO’s ICD-11 mapping licence language and Belgium’s managed SNOMED CT distribution arrangements.
A release cycle should be annual and at least bilingual (Dutch/French). Each cycle should include data freeze, draft dictionaries, clinical review, simulation against legacy APR-DRG outputs, public methodological notice, final release, and post-implementation audit. Every rule change should be versioned, backward-comparable where possible, and accompanied by a quantified impact note on case-mix shifts and hospital revenue simulation. That level of change control is indispensable for reimbursement legitimacy.
9. A pilot recalibration protocol tailored to Belgium
The pilot should begin with acute inpatient activity, not the entire hospital sector at once. Day cases can be added after the inpatient grouper and cost model stabilize; psychiatry, rehabilitation, and other sectors with distinct cost structures should be treated separately at first. Each pilot hospital should continue ordinary statutory coding while also producing shadow WHO ICD-11/WHO ICHI-coded episodes for selected service lines, then for full in-scope inpatient activity. This dual run is necessary because Belgium’s legal reimbursement environment remains ICD-10-BE/APR-DRG-based.
The pilot sample should be stratified rather than merely convenient: university and non-university hospitals, both language communities, large and small networks, and a mix of case-mix complexity profiles. Each pilot site should implement the same national costing manual, the same grouper input specification, and the same terminology governance process. A hospital should be excluded from weight estimation for a given cycle if its coding quality, map validation, or cost reconciliation fails predefined thresholds. That is a methodological necessity in a setting where KCE has already documented heterogeneity in Belgian analytic accounting practice.
The recalibration steps should be fixed in advance. First, assemble episodes and apply scope exclusions. Second, assign experimental WHO ICD-11/WHO ICHI DRGs while retaining legacy APR-DRGs. Third, determine inliers and outliers using a prepublished rule set. Fourth, calculate standardized episode costs. Fifth, estimate provisional weights with trimming and multi-year pooling where needed. Sixth, examine statistical and clinical coherence using dispersion metrics, outlier shares, and clinician review. Seventh, simulate revenue neutrality and redistribution against current Belgian payment streams. Eighth, publish shadow weights and methodological diagnostics, but move only low-volatility, high-confidence groups into any limited reimbursement pilot. This is the appropriate order for a country whose case-costing maturity is slowly improving but not yet uniform and of sufficient quality.
The go/no-go rule should be strict, not laying the bar low and then still crawling underneath it. Belgium should not convert shadow weights into statutory national reimbursement weights until four conditions are met: first, sufficiently reliable patient-level costing coverage; second, stable year-on-year DRG weights after outlier handling and pooling; third, acceptable (competent) coding and mapping audit performance; and fourth, transparent legal authority to use and distribute the required terminology and mapping assets. Without those conditions, Belgium would risk building a modern classification façade on an insufficient pricing base.
Conclusion
The correct Belgian strategy is neither to reject WHO ICD-11/WHO ICHI nor to overclaim readiness for immediate payment use. It is to build a national WHO ICD-11/WHO ICHI DRG architecture in which SNOMED CT remains the clinical source terminology, WHO ICD-11 MMS and WHO ICHI form the classification layer, and a transparent Belgian grouper converts standardized episode abstractions into MDCs, base DRGs, and open payment-complexity classes. Because no documented official global SNOMED CT→WHO ICD-11 or SNOMED CT→WHO ICHI reimbursement-grade map can safely be assumed, Belgium must govern its own maps, subsets, and manual-coding fallback routes. Because Belgium’s national costing maturity remains incomplete, it should estimate shadow weights first, using a standardized patient-level costing pilot reconciled to hospital accounts, and only later move to statutory reimbursement weights. That is the academically defensible and institutionally realistic route to reliable national DRG pricing under WHO ICD-11 and WHO ICHI.
However, as a concluding remark one could wonder why the European Union (EU) is not capable to design, develop and maintain a European DRG-system based on WHO ICD-11 and WHO ICHI?
Bibliography
EU CORDIS. EuroDRG – Diagnosis-Related Groups in Europe: towards Efficiency and Quality.
KCE. Analytic accounting in Belgian hospitals: state of the art and options for reform [Indirecte ziekenhuiskosten als mark-up percentage op de directe kosten] (KCE Report 380). 2024.