In using EEIO for their carbon footprint assessment of the National Health Service (NHS) in England, Tennison et al. note that a major limitation is that the “NHS’s internal spending information does not capture breakdown by economic sector and so was not used; instead, total spending on health care tracked by HM Treasury was proportioned using the transaction matrix in the UK MRIO model.”(3 p. 86–7). However, health care organisations in many jurisdictions globally now use activity based costing to help address economic efficiency and these data can be aligned to economic sectors (21). Australia uses activity-based funding in the healthcare sector whereby public hospitals get paid for the number and mix of patients they treat. To calculate the amount of the Australian Commonwealth Government’s payments to local hospital networks on an activity basis, the Independent Health and Aged Care Pricing Authority is responsible for determining the annual national efficient price and national efficient cost. To inform the calculation of these, Australian public hospitals routinely collect and report cost data as part of annual National Hospital Cost Data Collection (NHCDC) rounds (22). During each round, hospitals identify and classify all relevant costs into predetermined categories according to the NHCDC data request specifications and the Australian Hospital Costing Standards (23). These costs are then aggregated to cost buckets as specified in the relevant pricing framework (23). This process accounts for the direct and indirect costs of a care pathway, including specific procedures, from admission to discharge. This care pathway is referred to as a patient encounter. It includes all expenditure by the health care provider spent on this episode of care, including an apportionment for the general operation of the hospital and wider health system, and the specialised equipment required, such as a heart-lung machine or angiogram device. These cost bucket data can be aligned to economic sectors, enabling EEIO analysis.

Having restricted our sample to patient encounters associated with a pair of clinically equivalent procedures, we further minimised the variables in this novel application of the EEIO methodology by sampling a single hospital for 1 year to maximise the consistency of accounting practices that allocate expenditure to cost buckets. Cost data for NHCDC for the financial year 2021–22 (Round 26) were obtained from the Sydney Local Health District (SLHD) for costs incurred by Royal Prince Alfred Hospital (RPAH), a tertiary and quaternary referral hospital, for the elective surgical treatment of two-vessel coronary stenoses, by either:

The SLHD had information by cost bucket by each procedure and by each patient, supplied for this analysis in the form of an average by cost bucket associated with each procedure. The cost data provided by SLHD were allocated to 16 different cost buckets for each procedure, according to the NHCDC Pricing Framework (23). Addressing the issue raised by Tennison et al. (3), the data were then disaggregated and allocated to sectors within Australia’s economy, informed by the Australian Hospital Patient Costing Standards (23), which provide detail on what expenditure types are accounted for in each cost bucket. Where costs from the same cost bucket were allocated to more than one economic sector, the allocation was weighted based on the ratio of expenditure on those sectors by the Australian healthcare sector in the 2021–22 financial year, which provided the industry average. The types and definitions of cost buckets are summarised in Table 1.

These cost data required disaggregation to the 1,284 economic sectors provided in the single-region input–output table built using the Australian Industrial Ecology Laboratory (23). This disaggregation was executed by building a concordance matrix, allocating the total expenditure through the matrix, and excluding costs that are not included in final demand within the input–output table, which would otherwise result in double counting. These steps are detailed below.

A binary concordance matrix was built based on the details included in NSW Health’s accounting framework. A concordance is a binary matrix which connects two different sets of information, in this case the sector structure used to report economy-wide transactions in the input–output table, and the cost bucket structure used to report the SLHD cost data. An extract of the concordance used for this study is included in Figure 1, where a value of 1 in a cell represents a connection between the data reported within the input–output table (in the rows) and the cost bucket data received from SLHD (in the columns).

Assumptions made in the creation of this concordance include:

Where expenditure in one cost bucket was connected to more than one input–output sector, for example the non-clinical cost bucket, the percentage used to allocate the total expenditure in that cost bucket to the relevant input–output sectors was determined using the Australian healthcare industry average expenditure across each of the component sectors. The data for this was sourced from the intermediate demand matrix within the 2021–2022 input–output table. This meant, for example, that 1.6% of the costs recorded under the non-clinical cost bucket were allocated to expenditure against the electricity generation sectors. At this point, all costs included in each cost bucket were allocated to one or more sectors within the input–output table.

A final adjustment was made to the expenditure values allocated to each of the economic sectors within the input–output table to exclude any expenditure that did not constitute final demand within them. This includes expenditure on salaries and wages, which is captured in the value-added matrix in the input–output table. The exclusion of any expenditure that did not constitute final demand extends also to capital expenditure, present in the cost bucket data as depreciation, an accounting mechanism for allocating capital expenditure that has already been incurred. This expenditure was assigned by SLHD to the ‘Excluded’ bucket. This exclusion resulted in an underestimate of the systemic footprint, since it did not include emissions associated with expenditure on capital items during the 2021–22 financial year. This is recognised as a limitation of our method, which might be best described as an operating systemic carbon footprint. Both these adjustments are routinely made in studies of this sort (14).

Environmentally-extended input–output (EEIO) analysis, a well-established methodology for calculating carbon footprints (4), was used to calculate the carbon footprint for these two cardiac procedures. This methodology, introduced by Wassily Leontief (24), relies on matrix calculations to connect the flow of money within a given economy, represented in an input–output table, to the greenhouse gases emitted by activities in that economy. This approach has been widely used to understand carbon footprints across the global economy in recent decades but is still novel as a tool for measuring patient care pathways (1, 4). Key to these calculations is Equation 1, which connects the economic transactions captured in the input–output table to the environmental impact generated by those economic transactions.

Here, q represents the direct intensities vector, providing information on the greenhouse gases emitted for every dollar of total output for each sector in the input–output table. The L matrix, also known as the Leontief inverse, provides information on the economic interdependencies within and between all sectors within the economy by quantifying the value of total input required from each sector to produce one dollar worth of output for a given sector. It is derived using Equation 2, where I represents the identity matrix (the matrix equivalent of the number 1) and A represents the direct requirements matrix. The direct requirements matrix A is calculated from the data within the input–output table by dividing each transaction in the intermediate demand matrix, which provides information on the value of transactions between sectors within the economy, by the total output for each sector.

Finally, the y vector in Equation 1 provides information on the expenditure data associated with the entity for which the footprint is being calculated. In this case, two y vectors were created, one each for the CABG and stenting procedures. These expenditure vectors were built by adjusting the disaggregated cost bucket data to exclude depreciation and direct salaries and wages. These transactions are accounted for elsewhere in the input–output tables from which q and L are derived, and their inclusion in the expenditure vector used to calculate the footprint for each procedure would inflate the results.

A single-region input–output table for the 2021–2022 financial year was built using the Australian Industrial Ecology Laboratory (25). This input–output table contained information on the economic transactions within and between 1,284 sectors across the Australian economy, and the associated greenhouse gas emissions generated by each sector during the year, drawn from Australia’s National Greenhouse Inventory Report in line with emissions estimation rules adopted under the Paris Agreement (26). The full list of sectors in the Industrial Ecology Laboratory is given in the Supplementary material.

All greenhouse gases were included in this analysis, but adjustments associated with land use change were excluded.

EEIO analysis can be understood as connecting sectoral GHG emissions to the expenditure that supports each procedure, which we refer to as a systemic carbon footprint. This can be contrasted with the now widespread use of unit process LCA to quantify emissions associated with health sector processes to assess the associated environmental impacts (6, 12, 13). To ensure a manageable analysis, the boundaries in LCA studies tend to be relatively confined, barring an analysis of the complex systems that support patient care pathways. By contrast, the use of EEIO to quantify the carbon footprint associated with these two cardiac procedures supports a boundaryless, systemic assessment of the total carbon emissions generated to support the delivery of these procedures at RPAH. These footprint results incorporate the whole of system emissions attributable to a procedure, including the emissions associated with operating the healthcare system and its facilities, infrastructure, and personnel, along with the apportioning of utilities, transport, and material inputs such as clinical supplies and medical devices and furniture.

The systemic scope of inclusion for EEIO analysis is highlighted in Table 2 when we compare our results with two other recent analysis of cardiac procedure carbon footprints. The 124.3 kg CO₂-e emissions calculated by Grinberg et al. for conventional adult cardiac surgery in France, used an eco-audit approach (31). This simplified LCA approach may have impacted on the precision, robustness, and representativeness of their study (35). A subsequent LCA based estimate of the amount of CO₂-e produced by CABG surgery at Tufts Medical Center was four times larger, at 505.1 kg CO₂ − e per case (33). However, some of this difference may also be attributable to differences in the procedure, like Tuft’s requirement to autoclave all waste from cardiac suites, prior to disposal. Nonetheless both these results were well within the 6–814 kg CO₂ − e range identified by Rizan et al. in their systematic review of surgical operation carbon footprints (36). The very significant difference between the footprint calculated by Grinberg et al. (31) and Hubert et al. (33), and our result of 11.5 tonnes CO₂-e for the CABG procedure, suggests that a narrowly bounded focus on specific surgical or other clinical processes runs the risk of very significantly under-estimating the systemic carbon footprint of patient care pathways, by excluding the emissions associated with the operation of the health system and the comprehensive supply chains that support a particular procedure. For example, Grinberg et al. (31) only included disposable medical products, pharmaceutical products and electricity consumption used in anaesthesia, cardiopulmonary bypass and surgery. Similarly, Hubert et al. (33) estimated by collecting data from anaesthesia, electrical consumption, and generation of solid waste. They acknowledged that emissions from the heating, ventilation and air-conditioning (HVAC) system, reprocessing surgical instruments, pre and post-operative care were excluded. Neither study included a range of other sources of emissions in the patient encounter included in our study, including pathology, critical care, patient transport, as well as non-clinical sources.

The difference between the results in Table 2 is superficially striking. However, these studies have quite substantially different scopes. The American and French LCA studies are designed consequentially, aiming to quantify the marginal additional climate change burden of the CABG procedures studied. By contrast our EEIO study can be understood as an attributional life cycle assessment, aiming to quantify the contribution of the patient care pathway associated with a CABG procedure in Australia to the overall burden of green-house gas emissions driving global heating. The difference between the footprint calculated by Grinberg et al. and Hubert et al. for conventional isolated cardiac procedures and our calculation for a CABG patient care pathway may, at least partially, be explained by these different scopes (31, 33). The scope of our calculations include the full range of activities contributing to each patients’ episode of care from admission to discharge. This includes allied health services, medical support, nursing, critical care, imaging, pathology, prosthetics, specialist procedures, ward and emergency department supplies, non-clinical and on-costs, and patient transport. Importantly, the pathway for a patient undergoing a CABG procedure typically involves an extended period of post operative care prior to discharge, consuming up to 9 days of hospital resources (28, 29). The consumption of these resources all contribute to the carbon footprint of a patient care pathway, but are not within scope for the LCA studies here discussed. All these essential aspects of delivering a cardiac procedure contribute to the health sector’s overall emissions, some 7% of Australia’s total emissions (4).

In making recommendations to patients, healthcare providers have, to date, been almost exclusively influenced by the clinical outcomes of procedures. Increasingly, however, they are acknowledging that treatment choices also have economic and environmental consequences and, in the face of the climate crisis in particular, that environmental impacts also form an important component of healthcare decision making, including informing patients (37, 38). In this context, data about carbon footprinting of different procedures are essential to help healthcare providers understand the environmental impact of their clinical choices.

In this paper, we chose two procedures with relatively similar clinical outcomes, recognising that this is the circumstance where healthcare providers will be most interested in other impacts of their decisions, including those on the environment. There are many other examples in medicine where there are two or more choices for clinical care with similar outcomes, for example in the choice of asthma puffers (pressurised with propellants versus activated by inhalation), anaesthetic agents (intravenous versus volatile gases), bariatric surgery versus GLP-1 agonists for weight loss, and alternative forms of peritoneal dialysis. Such equipoise cases represent relatively uncontroversial entry points for guiding medical practitioners to consider differences in emissions profiles in their advice. Cases where emissions profiles are in tension with health outcomes will require far more public debate, but it can be anticipated that as the impacts of climate change become more severe, debates in health care will, as in other areas of public policy, need to more seriously consider climate impacts.

In addressing climate change impacts, healthcare planners will also benefit from a streamlined mapping of emissions hotspots across patient care pathways, made possible by the systemic carbon footprinting approach demonstrated in this paper. These hotspots can inform systems design for emerging primary care models in areas such as telehealth, introducing consideration of carbon emissions into analysis of health care delivery reform (39). Research in other sectors has shown that integrating activity-based costing evaluation and carbon footprint assessment can help managers incorporate environment costs into decision-making processes (40, 41). The method we describe may achieve similar results in informing health systems funding allocation, by providing planners with a carbon weighting to include in their activity-based funding determinations.

Operationally, hotspots identified through EEIO analysis will also constrain parameters for process-based LCA, allowing targeted collection of granular data to inform very specific LCA studies. These studies can be costly, but when targeted, can provide detailed guidance for refining facilities design and management, and for the refining clinical procedures, in areas such as waste management and alternative clinical technologies.

In their study of the carbon footprints of hospital care pathways Zhang et al. (14) conduct a data quality assessment of emissions factors and financial activity data like those used in this study. They found these to be ‘good’ or ‘very good’ in all aspects, although the framework they used is typically applied to physical data. Pertinently, recognising that Leontief’s basic input–output relationship cannot be differentiated analytically, Lenzen (42) describes a range of uncertainties associated with input/output analysis and compares these to truncation errors associated with standard unit process life cycle techniques. He notes that cumulatively, these uncertainties are smaller than the truncation errors for most commodities, and that this uncertainty decreases with the number of components in the assessment. We note that the number of components in our assessment of patient care pathways defies conventional unit-process assessment, suggesting a lower level of uncertainty. In a later paper Lenzen et al. (43) undertake a comprehensive uncertainty analysis to estimate standard deviations for carbon multipliers used in the calculation of the UK’s carbon footprint, using Monte Carlo techniques. They conclude that multipliers for consumer emissions exhibit relatively low Relative Standard Deviations of between 3.0 and 5.1%.

The contrast between the results of LCA studies of the carbon footprint of cardiac procedures and the EEIO study described here emphasizes the difference between these modelling approaches. Each has some limitations. Aside from the major difference in scope, limitations associated with these different methods may contribute to the contrasting scale of results. Unit process LCA, informed by detailed physical data, and EEIO, informed by financial activity data, are subject to different uncertainties (42). LCA studies typically suffer from some degree of truncation error, as they are unable to comprehensively map their entire supply chain. This has been estimated to lead to a more than 50% underestimate of emissions. Studies of more service based activities, such as those in healthcare, tend to have higher truncation errors (42). The degree to which more sophisticated LCA techniques address this limitation continues to be debated in the literature (44, 45). By contrast, EEIO provides a boundaryless and therefore comprehensive account of supply chain emissions, but is limited in the level of granularity that can be achieved given its comprehensiveness. The activity-based costing data used for estimating the footprint of an activity has been subject to several accounting transformations that may also contribute to uncertainty.

Bearing in mind both the different scopes and uncertainties associated with these distinctive modelling approaches, it is important to recognise that they are useful for different purposes: EEIO provides a comprehensive systemic footprint (which could inform strategic decision making such as health systems design and guidance on choice of patient care pathway), LCA provides a detailed specific footprint (which can inform tactical decision making). While unhindered by the need to select a boundary, limitations are introduced to any EEIO analysis through the sectoral resolution of the underlying input–output tables, providing far less granular data than the primary data collected for LCA studies. An EEIO analysis will not provide guidance on procurement decisions between alternate suppliers of goods and services, if they are in the same sector. It will also not provide detailed guidance for refining facilities design and management (for example optimising HVAC, or power standby) or for refining clinical care procedures (for example sterilisation and reuse rather than single use of clinical supplies). Further, specific to the health sector, some authors are cautious of distortions in costing, for example resulting from different drug pricing regimes, that may make it difficult to accurately connect different expenditure values to the resultant GHG emissions (13). Although the subsidies that potentially cause these distortions are incorporated into the input–output tables upon which EEIO analysis is based, these concerns do bear further investigation.

Other limitations of the methodology presented here are introduced with the high level of aggregation used in allocating the costs associated with each procedure to cost buckets, since these data need to be disaggregated to match the economic sectors present in the input–output table. This disaggregation process could be improved if more detailed insights were available on the allocation of detailed costs to each cost bucket, or if cost data were available at a finer resolution, such as at the cost centre level. Further studies using more highly disaggregated activity-based costing data are underway to understand the impact of this limitation.

Finally, the exclusion of depreciation costs is recognised as a limitation of the described method. This exclusion resulted in an underestimate of the systemic footprint, which did not include emissions associated with expenditure on capital items. This does warrant research into practical approaches to estimating capital systemic footprints. Further studies are underway, to explore the implications of these methodological differences and to establish guidelines for the contexts in which the different methodologies will be most appropriate.