SM5002 Chap.6 Parametric Modelling and Lancet MedZero

Parametric Modelling and Lancet MedZero

The scale problem behind this week

Process-based life cycle assessment, as practised in the openLCA workshops, produces careful results one product at a time. That precision has a cost.

The Lancet comment that introduced the Lancet MedZero notes that a full assessment of one product takes between 2 and 12 months, that carbon data exist for less than 1% of the roughly 2 million medical products used worldwide, and that filling the gap product by product would need an indicative 333,000 person-years of analytical capacity.

The same comment points to further barriers: companies rarely release the underlying data, independent assessments without commercial bias are rare, and published footprints are often tied to one place and time and hard for clinicians to use.

Lecture 5 frames the result as four problems for healthcare users: scale, accessibility, trust, and quality and standardisation.

What parametric modelling is

The recorded lecture on chemicals and pharmaceuticals starts from a simple idea. Every model runs on input variables, and in life cycle assessment the direct inputs are called parameters.

Parameters may be numbers or categories, and numeric ones may be whole counts or continuous values, and they can be direct flows or indirect drivers of impact. Parametric modelling uses a selection of parameters, trained on a set of products whose results are already known, to predict or estimate impacts for a product that has not been assessed.

The approach is especially useful for highly variable products, where there are endless possible mixes and conditions.

Chemicals show why it is needed. The lecture gives a ladder of numbers: about 2 x 10^11 theoretically possible chemicals, around 10^8 registered identifiers, roughly 10^5 chemicals in commerce, and only about 10^3 with life cycle inventory data.

No process-based programme can close a gap of that size directly.

The four steps

The lecture sets out four steps. First, select parameters by asking which processes are common across products and which drive emissions. Second, determine the potential values each parameter can take. Third, find data or physical relationships to train the model.

Fourth, use statistical, machine learning or AI techniques to find the best fitting model.

Estimation approaches sit on a trade-off between data requirements and uncertainty, from simple process calculations and stoichiometry through to detailed process simulation, which the lecture describes as the most accurate approach because it also gives a process breakdown.

Parameterising a medicine

For pharmaceuticals the product is split into parts and each part is modelled in the way that suits it.

Synthesis of the active pharmaceutical ingredient uses a machine learning model; excipients are a fixed percentage; manufacturing and packaging each use a regression model.

The synthesis model draws on four kinds of input: whether the ingredient is a small molecule or a biologic, the complexity of synthesis in steps and purification, the generic market price as a proxy for resource use, and molecular complexity such as molecular weight and functional groups.

The Lancet MedZero platform

The Lancet MedZero is described in Lecture 5 as an open, dynamic and collaborative global database of carbon figures covering medicines, medical and surgical devices, diagnostics and care pathways, intended for health professionals and policymakers.

Its principles include independence, best available methods and data, transparency with quality assurance, decision readiness, and operation at scale. Footprints can be customised for where a product is used and made, its shipping mode and distance, its disposal method, and the number of uses of a reusable.

The platform combines individual product assessments with parametric modelling for products that share similar processes and multi-region input-output modelling that covers the whole economy.

The Lancet comment adds that each footprint will carry a quality rating, that users will be able to amend assumptions and parameters for their own clinical context, and that the aim is coverage of 80% of the health system footprint within five years.

Using the platform for interventions

The workshop gives each group a hospital case with two candidate initiatives, such as reusable surgical gowns, fewer urinary catheters, low-flow anaesthesia, IV-to-oral paracetamol, or fewer blood tests.

Groups fill a semi-populated spreadsheet with local inputs and platform carbon factors, compare scenarios, test sensitivity to key assumptions, and present projected financial and carbon savings with a recommendation.

Several cases are designed to surface tension between objectives, for example a cheaper medicine supplier whose product carries a higher footprint, so a good presentation names where cost and carbon disagree rather than assuming they always move together.

How the platform plans to grow

Lecture 5 lists five routes to scale: expanding data sources, prioritising the most commonly procured items, letting users amend assumptions, using novel data collection methods including artificial intelligence and parametric models, and letting industry add its own data to the platform.

The Lancet comment describes the platform as a collaboration between The Lancet and academic partners in Thailand, Japan, the United States, Singapore and Australia, independent of industry, with dedicated audits of prioritised products supplementing published and grey literature.

Industry submissions are accepted only if strict transparency standards are met, and an annual publication in The Lancet is planned to set out all available footprints.

What the numbers look like in practice

The answer workbook makes the scale of decisions concrete. In the surgical gown case, 27,000 single-use gowns a year cost $69,592.50 including incineration and emit about 78.14 t CO2e.

Diverting the same gowns to recycling saves about 18.69 t, while moving 70% of procedures to reusable gowns used for their full 75 lives saves about 46.84 t and $22,943.72 a year. Even with lifetime use cut by 30%, the reusable option still saves about 46.49 t, which shows how a sensitivity check can confirm a recommendation rather than overturn it.

What this chapter covers

• 01

  Why product-by-product assessment cannot cover health care

• 02

  Parameters and what parametric modelling predicts

• 03

  The four steps of building a parametric model

• 04

  How a pharmaceutical product is parameterised

• 05

  Customising footprints on the Lancet MedZero platform

• 06

  Costing and carbon-counting a hospital intervention

Worked example · free

Switching eligible paracetamol doses from IV to oral

Key terms

Parameter

A direct input variable to an LCA model, which may be categorical or numerical and a direct flow or an indirect driver.

Parametric Model

A model trained on products with known results that estimates impacts for an unassessed product from its parameters.

Active Pharmaceutical Ingredient

The therapeutic substance in a medicine, whose synthesis is estimated with a machine learning model in the lecture approach.

Multi-Region Input-Output

Economy-wide modelling that by design covers any product or service used in health care.

Quality Rating

A score attached to each platform footprint showing how robust its underlying data and methods are.

Parametric Modelling and Lancet MedZero FAQ

Assessment move

Work one intervention end to end with a calculator before opening the spreadsheet, writing the baseline, the changed quantities and the savings as separate lines. Then change one assumption at a time to see which input moves the answer most, and say so in your recommendation.