1. Establish a complete and reconciled expenditure framework

A health system emissions assessment must cover the full footprint within its stated boundary, including domestic and international upstream supply chain emissions. A comprehensive expenditure framework provides the financial foundation for identifying all goods, services and capital assets purchased by the health system (Fig. 1).

Assessments should begin with available financial and activity records and improve progressively as coverage and detail increase.

National Health Accounts (NHA) (4) can be used to establish the overall expenditure boundary, including current and capital health spending across public and private providers. Coverage must be checked against the agreed assessment boundary and reporting period, including whether capital expenditure is reported separately.

Additional financial records, including government accounts, organizational ledgers, procurement and capital expenditure records, may be needed to identify purchased goods, services and capital assets within broad NHA categories.

These sources must be reconciled within one coherent framework. One source, usually the NHA, must provide a reference total against which contributing expenditure datasets are reconciled, with any adjustments to its coverage explained.

Transfers between financing bodies and providers, central procurement, and subsequent distribution to facilities must be reconciled so each underlying purchase is only counted once.

Where detailed records cover only part of the system, they may be used as samples to allocate broader totals. Analysts must identify the coverage of each source and document assumptions used to fill gaps, so that gaps in financial records are not carried through as gaps in the assessment.

2. Account for every line of expenditure

A single-sector estimate may be used to provide an initial indication of scale but must not be presented as a complete hybrid assessment (Box 1).

Collecting data, checking assumptions and results with stakeholders, and refining the assessment before publication often take longer than building the model itself. With access to sufficiently detailed financial records, a top-down model based on expenditure can be developed quickly, often within a few months. This provides the foundation for a hybrid model that can be refined using more detailed activity- and expenditure-based data.

Start with an expenditure dataset covering the assessment boundary, using the sources and reconciliation approach described in section 1. Every expenditure line must either be excluded from calculations, mapped directly to an MRIO sector, or allocated to multiple sectors (Fig. 2). Identify exclusions and complete direct mappings before allocating the remaining expenditure.

Assign stable, unique identifiers to organizations, source records and model calculations. Record each dataset’s coverage, reporting period and relationship to other datasets, and retain links between expenditure, activity and emissions estimates. This should make it possible to identify what has been excluded, allocated, replaced or added, and to trace every estimate of emissions back to its source data.

Note: This expenditure-based estimate provides a starting point for a hybrid model. Additional emissions sources must be incorporated to complete an assessment.

Exclude employee remuneration and employer pension contributions, while retaining agency staffing and contracted professional services purchased from third parties. Exclude depreciation and amortization, record capital purchases in the year of expenditure, and treat taxes on the same basis as the MRIO model used.

Clearly defined purchases should be mapped directly to relevant MRIO sectors. For example, an “Essential Medicines” expenditure line may map to “Manufacture of Pharmaceuticals”. Mappings should follow sector definitions, using account and sector descriptions to resolve ambiguous items.

Remaining aggregated expenditure may be allocated using purchasing profiles from representative financial or procurement records or, where these are unavailable, the underlying MRIO model. Profiles must exclude items already accounted for, with the remaining shares rescaled to sum to 100%. Operating expenditure profiles should not be used to allocate capital purchases, which are usually concentrated in a few economic sectors. The purchasing profiles in Annex 3 of the NHS England carbon footprint methodology (3) provide practical examples of expenditure allocation. Their applicability to another health system must be assessed before use.

Detailed records from representative providers can help allocate broader totals, for example using hospital bills to divide utilities expenditure between electricity, gas and water. Where records from one provider group are used as a proxy for another, assess differences in service mix, staffing and outsourcing. Alternatively, the relevant country’s “health services” column in the MRIO intermediate consumption (T) matrix can provide an operating expenditure profile, subject to checks on sector coverage and purchasing patterns. Replace proxy profiles with more representative financial information as it becomes available, and align expenditure with the emission factors’ currency, price year and valuation basis, following the Toolkit’s MRIO guidance.

3. Use representative activity data and emission factors

Analysts should use suitable activity data already available and prioritize further collection where it addresses emissions gaps, improves estimates of major sources, or supports analysis and monitoring of decarbonization interventions.

The assessment must capture relevant emissions absent from expenditure-based estimates, including fuel combustion, releases of anaesthetic gases, inhaler propellants and refrigerants, and personal travel not captured by purchasing records. Available activity data can also significantly improve estimates of emissions sources already represented by expenditure-based calculations, such as electricity consumption, water supply, wastewater and waste treatment, or business travel.

Activity data are converted to emissions using factors expressed per physical unit, such as kilograms of carbon dioxide equivalent per kilowatt-hour, litre or passenger-kilometre. Factors must match the units of the activity data and represent the relevant technology, geography and period. Their completeness, reliability and life-cycle coverage must also be assessed.

More specific factors are not necessarily more representative or reliable – a well-documented average may be preferable to a supplier-specific estimate with unclear coverage or assumptions. The selected source, version, reference year and any adjustments must be documented (Box 2).

Technical detail · Box 2Selecting emission factors for activity-based analysis

Emission factors should be assessed against the following criteria, adapted from the GHG Protocol’s data-quality criteria (6). The assumptions, units and coverage of each factor must be documented so it can be reviewed and updated.

Criterion	Questions to address
Technological representativeness	Does the factor reflect the relevant product, production method, equipment or treatment process?
Geographical representativeness	Does it reflect the relevant location, including its energy systems and supply chains?
Temporal representativeness	Do the underlying data reflect the assessment period?
Completeness	Does the dataset adequately represent the activity, including any relevant variation between facilities, suppliers and operating conditions?
Reliability	Are the underlying data, estimation methods and allocation assumptions documented? What review or verification has been undertaken?

Relevant factors published by government agencies, national inventory programmes, regulators or reporting schemes should be reviewed first. They may better represent local energy systems, transport fleets and treatment processes, while supporting consistency with national reporting.

For electricity, analysts should review grid emission factors published by the relevant national authority or grid operator. Their coverage must be checked, including whether upstream fuel emissions and transmission and distribution losses require separate calculations. Other sources include the Intergovernmental Panel on Climate Change Emission Factor Database (7), factor datasets published by other national governments, peer-reviewed life cycle assessments, and product or supplier information.

The Lancet MedZero provides open-access emissions estimates for medicines, medical and surgical devices, diagnostic services and care pathways. It combines published evidence and commissioned product audits with process-based assessments, parametric modelling and MRIO methods, providing source information and guidance on interpretation (8,9). Users can adjust assumptions such as country of use, transport, electricity supply, waste disposal and the number of uses for reusable products to reflect their context (Fig. 3). The database is continually updated as new evidence becomes available, with contributions from researchers, health systems and manufacturers supporting wider coverage and more detailed estimates. Combined with expenditure and activity data, the Lancet MedZero database can support a hybrid emissions assessment (see section 4).

Where activity data cover only part of the system, analysts may extrapolate using representative samples and suitable scaling variables. Provider groups should reflect characteristics affecting consumption, such as facility type, size and service mix.

Analysts must distinguish observed activity from estimates and extrapolations, identify the organizations and reporting periods represented, and disclose material limitations in coverage or representativeness.

4. Combine estimates without gaps or double counting

A hybrid model combines expenditure-based and activity- or process-based estimates within a shared assessment boundary. Before integration, analysts must establish the purchases, organizations, reporting periods and life-cycle stages covered by each calculation.

Data reconciliation and validation checks should be built into the model. These should identify duplicate or unassigned records, allocations that do not sum to the original expenditure, gaps or overlaps in coverage, and replacements that remove emissions outside the coverage of the new calculation. The Toolkit’s guidance on software and technology provides further advice on quality assurance, reproducible analytical pipelines and version control.

The appropriate treatment depends on the relationship between the estimates (Fig. 4):

• Replacement: where a calculation covers emissions already represented, the overlapping expenditure-based emissions must be removed.

• Addition: where a calculation covers emissions outside the existing estimate, these must be added while retaining the existing coverage.

• Replacement and addition: where a calculation covers both overlapping and additional emissions, only the overlapping estimate is removed and the full new calculation is included.

Note: Spending is removed from the expenditure-based calculation only where the associated emissions are fully replaced. Partial replacements retain uncovered emissions; additional sources are added separately.

Where activity data cover only selected hospitals, replacement must be limited to those hospitals’ corresponding estimates unless the data have been appropriately extrapolated to the wider system. Expenditure-based estimates must remain for the coverage not represented by the new calculation.

Life-cycle coverage must also match. A combustion-only fuel calculation adds use-phase emissions while retaining the upstream production estimate. Remove an entire expenditure-based estimate only when the replacement calculations cover all the emissions it represents. Any retained component of an MRIO estimate must be identifiable and its calculation must be supported and documented.

Product, procedure and care-pathway footprints may overlap with existing estimates for energy consumption, purchased products, transport or waste treatment. Their coverage must be checked before integration within a hybrid model.

Expenditure associated with replacements should remain traceable within the model, including how the corresponding amount was identified or estimated. The net effects of each replacement or addition must be documented.

Material differences between the original expenditure-based estimate and the combined calculation must be reviewed and explained. These may indicate errors or reflect differences between the methods. Distinguish changes in underlying activity or emissions intensity from changes caused by revised methods, coverage or data quality. Where material methodological changes affect a trend, recalculate earlier estimates consistently where possible, or clearly explain a break in comparability.

Where an intervention affects emissions that the current model cannot distinguish, improve the relevant calculation using representative activity data and emission factors where feasible. Modelled savings may also be incorporated as an explicit adjustment, provided they are reconciled with the affected estimate, use compatible boundaries, reporting periods and assumptions, and are not already captured through changes in expenditure, activity data, emission factors or other adjustments. Only adjustments that cover emissions within the inventory boundary and reporting period can enter the inventory; counterfactual or avoided emissions should be reported separately.

Apply the replacement and reconciliation rules above, and document the adjustment’s basis, assumptions and uncertainty. Where estimated savings cannot be integrated consistently, they can be reported alongside the inventory. The Toolkit’s guidance on projecting health system emissions provides further advice on modelling intervention effects.

5. Enable disaggregated estimates consistent with the whole-system model

Developing disaggregated estimates that are consistent with the national model is a priority for supporting decarbonization decisions. Disaggregation can distinguish regions, organizations, health care functions, levels of care, or public and private provision.

The longer-term ambition should be to build as much of the model as possible from organizational-level financial and activity records, connecting emissions to the places where changes can be implemented directly (Fig. 5). This development may be incremental. For example, hospital electricity and fuel records could support hospital-level energy calculations, while national purchasing profiles remain for categories with less detail.

Disaggregated estimates should use consistent boundaries, classifications and methods, and should reconcile with the system total. For each emissions category, record which organizations are represented by local data and which rely on national estimates or proxy allocations. Central procurement, shared services and any remaining national estimates must be accounted for consistently.

Note: Spending is removed from the expenditure-based calculation only where the associated emissions are fully replaced. Partial replacements retain uncovered emissions; additional sources are added separately.

Where organizational records are incomplete or unavailable, national estimates may still be disaggregated using proxies. Floor area may provide a basis for allocating building-energy emissions, while expenditure on relevant purchases may be more appropriate for allocating supply chain emissions. Allocation measures must be appropriate to the emissions category; a single proxy should not automatically be applied to the entire footprint.

Proxy allocations reflect assumptions about the distribution of emissions. They must not be used alone to assess relative emissions intensity or environmental performance. Comparisons must also consider differences in scale, services delivered, and access to and quality of care.