A new heuristic framework for estimating indirect (Scope 3) emissions of large organizations

Large businesses, both for-profit corporations and not-for-profit organizations, are pledging voluntary carbon mitigation. Alongside such pledges, regulations such as the European Union’s Corporate Sustainability Reporting (CSR) Directive (2022), California’s ‘Climate Corporate Data Accountability Act (2023), U.S. SEC climate-disclosure rules (2024) are advancing mandatory climate-related reporting for large companies1,2,3. For both voluntary action and regulatory compliance, organizations need detailed information about their lifecycle environmental footprint. The lifecycle footprint of a product or an organization can be classified into direct and indirect emissions. Direct emissions are those arising from an organization’s own properties (facilities and vehicles), while indirect emissions are those associated with upstream and downstream activities that enable and support the organization’s own activities. For many industries such as education, finance, health care, hospitality and retail, indirect emissions can be several-times their r direct emissions. An alternative classification popular in corporate sustainability reporting CSR-context is Scope 1, 2, and 3 emissions as described by the GHG Protocol4. Scope 1 emissions are simply the direct emissions. Scope 2 emissions refer to a subset of indirect emissions that are associated only with electricity, steam, heating, and cooling services purchased by an organization for use in its own facilities. Scope 3 emissions are the rest of the indirect emissions, which include lifecycle emissions from all other upstream and downstream activities. The GHG protocol divides these Scope 3 activities in 15 distinct categories that includes purchased goods and services, transportation, employee commuting, business travel, use and end-of-life management of sold products, and waste management, among others5. Here, we present a case study of indirect emissions in healthcare systems, specifically focusing on one category, ‘Purchased Goods and Services’, of the Scope 3 emissions as defined by the GHG Protocol.

The healthcare industry is estimated to be responsible for more than 4% of the global greenhouse gas (GHG) emissions and 8.5% of the domestic GHG emissions in the United States6,7,8. With increasing median income and an aging population, demand for health services is growing rapidly globally9,10,11. This has driven a growing interest in reducing Healthcare industry’s increasing contribution to climate change6,12,13,14,15,16,17 In 2022, the White House and the Human and Health Services (HHS) launched the Health Sector Climate Pledge to encourage the healthcare sector’s commitment and actions in decarbonization. This pledge includes conducting an inventory of Scope 3 supply chain emissions, a task many health systems have never undertaken before. Most voluntary disclosures and pledges thus far have focused on Scope 1 and 2 emissions18,19,20.

Although scope 3 emissions are relatively more difficult to estimate, which motivates this work, the state-of-the-art technique (described below) suggests they might be several-fold greater than Scope 1 and 2 emissions combined21. For instance, Eckelman et al. (2020) estimate that Scope 3 indirect emissions account for 82% of the total US healthcare emissions, while a study undertaken for the University of California Office of the President concluded Scope 3 indirect emissions contributed 72.5% of the total emissions among the five academic health systems in 2022. A widely-used approach, which in this study we will refer to as ‘top-down approach’, proceeds as follows for purchased goods emissions: (i) Collect expenditure data on different products and activities relevant to scope 3 emissions; (ii) Match each product or activity with one of the 400 or more broad industry groups listed in the environmentally-extended input-output (EEIO) tables; (iii) Multiply the expenditure on each individual product and the average GHG emission intensity per dollar of output for the matched g EEIO sector to estimate lifecycle emissions due to that product; (iv) Sum the emissions of each product to determine total scope 3 emissions for the organization22. This approach is also called the ‘expenditure-based’ method in the literature20. The main limitations of using a purely expenditure-based approach to estimate emissions are twofold. One is that it lacks the granularity to adequately accommodate the diversity of the products and services and their lifecycle emissions23. For instance, for the healthcare industry, which we use as a case study here, a single large hospital purchases over 25000 distinct products (grouped under 105 product categories), which are matched to only a handful of the EEIO categories. The second is that both the cost of production and the retail price of a product are poorly correlated with environmental footprint24 (See Fig S1). Therefore, there is a need for better approaches to identify the most polluting products and prioritize mitigation strategies.

The ideal approach, which we refer to as a ‘bottom-up approach’, is to conduct a life cycle assessment (LCA) for each product or activity contributing to scope 3 emissions25. An LCA is a standardized methodology (ISO 14040, ISO 14044) to estimate the different types of environmental and resource burdens associated with the entire life cycle of a product or service. The GHG protocol defines this approach as ‘supplier-specific method’ where a business has to “collect product-level cradle-to-gate GHG inventory data from goods or services suppliers”5. There is growing literature on LCA of healthcare-related products and services. Examples include personal protective apparel (PPE) such as scrubs and head covers26,27,28,29,30, face masks31,32,33, gloves34, and specific medical instruments and devices, e.g., anesthetic equipment35, laryngoscope36, ureteroscope37, catheter38, and knee implants39. Going beyond individual products, literature also analyzes different medical procedures that are multi-product systems, e.g., prostatectomy40, and cataract surgery41,42. A limited but growing literature analyzes emissions for specialized care, such as for bariatric surgery43, hemodialysis treatment44, dental care45,46, and intensive care47. Drew et al. (2021) reviewed the landscape of LCA studies related to surgical and anesthetic care48. However, the number of peer-reviewed, published and publicly accessible LCAs is but a tiny fraction of the tens of thousands of products consumed in hospitals, and more generally, in the healthcare sector. But since conducting an LCA for each product is data intensive and costly in time and resources, a purely bottom-up approach to Scope 3 emissions appears impractical. An LCA of each products is perhaps even unnecessary given the distribution of products in terms of the quantity of and expenditure on each (more details below).

The few hospital-level studies that exist apply either a top-down approach or a hybrid approach, where some products are subjected to a bottom-up method, with the rest subjected to the expenditure-based method. A hybrid approach used in a study of a German hospital includes the bottom-up method for Scope 1 and 2, but the top-down method for Scope 3 emissions25. A study of a Dutch hospital finds the annual carbon footprint to be 209 kilotons, with Scope 3 contributing the majority share at 72%49. One study that employs the bottom-up approach exclusively is an Organizational LCA of a Canadian hospital23. Organizational LCA (O-LCA) is a framework that expands LCA methodology for an organization where the functional unit is a financial reporting year50. O-LCA methodology applied to the relevant products can be used for calculating the Scope 3 emissions of the organization for that reporting year. The challenge with this approach is selecting a representative sample from the large number of products consumed in the organization to best approximate total emissions or identifying the major contributing products. The O-LCA guidance does not mandate any specific sampling approach except recommending a few alternatives and requiring transparent communication of the choices made. Cimprich and Young (2023) adopt a stratified random sampling method to analyze a portfolio of 2927 unique products, where the sample size is determined by the LCAs of pilot products in different categories. They compile life cycle inventory (LCI) data for a sample of around 200 products based on existing literature and primary data collection23.

Here, we present a different heuristic framework for estimating scope 3 emissions and apply it to data from the University of California, San Francisco Medical Centre, a tertiary academic medical center with a total of 796 beds across its various facilities. We specifically analyze one category of scope 3 emissions called purchased products, which is currently estimated to constitute the biggest part of the Scope 3 emissions. Our dataset comprises over 25,000 unique medical products categorized into 105 different categories and amounting to over $290 million in annual expenditure by UCSF. Given the large number of products, the lack of readily available estimates of lifecycle emissions for most of these products and the challenges in performing an LCA for each, we develop a heuristic technique to identify a subset of all products for each of which we develop an estimation of life cycle emissions which we then scale to estimate emissions for all the products in our dataset.

We present a heuristic to obtain a sample of ~ 1000 products spanning 45 categories from a population of over 25,000 products spanning 105 categories. Our heuristic uses information about both the expenditure on and quantity of different products. Next, we estimate lifecycle emissions for each of these products using another heuristic. From this, we calculate emissions for each category. We do this by totaling the emissions for all products in the sample from a given category and then scaling this by the inverse of the ratio of total expenditure on all products in that category to the total expenditure on the products in the category that are in the sample. In this way, we calculate the total emissions across the sampled categories. We then scale this by the inverse of the ratio of total expenditure across all categories (i.e., 105 categories) to the total expenditure on the categories in the sample (See Method for more details and Fig. 1 for a schematic of our approach). The estimate of total emission derived this way is what we refer to as the “bottom-up money heuristic”. Using the same two-step process, we also derive a “bottom-up quantity heuristic” wherein the scaling at the first step is done based on the share of total quantity that products in the sample represent for a category and then again in the second step, scaling is done based on the share of total quantity across all categories represented by the categories in the sample (see Fig. 1 for schematic of our approach).

Fig. 1

Schematic of the proposed bottom-up approach in this paper.

The main objective of this work is to present an alternative to the expenditure-based heuristic and show how this suggests a different set of priorities for emission reduction. Indeed, one could improve our heuristic or develop alternative heuristics, and for this reason, we don’t claim ours is optimal, but as one that does not rely purely on cost and average emissions per unit cost for highly-aggregated economic sectors, as the expenditure-based method does. Indeed, our heuristic does entail arbitrary cut-offs such as the share of expenditure and quantity yet these are based on plausible justifications. To address this, we perform a sensitivity analysis. Additionally, we develop a metric of uncertainty using machine learning techniques (natural language processing using neural networks and unsupervised learning) to analyze text descriptions of the products. This metric is a measure of how aligned the sample for each category is with the product diversity within that category. We believe that the heuristic approach presented here, combined with an understanding of sensitivity and uncertainty, holds promise for identifying cost-effective interventions for reducing scope 3 emissions.