Simon Lande
From computational breakthrough to strategic decision-making – reflections from the World Congress of the International Microsimulation Association
I recently attended the World Congress of the International Microsimulation Association (IMA), where researchers from around the world presented the latest advances in microsimulation across multiple fields.
One thing immediately stood out.
There was remarkably little representation from healthcare and the pharmaceutical industry, despite microsimulation having been applied for decades in the academic and public sector of these domains, as set out in the seminal review paper by Professor Deborah Schofield et al. Microsimulation is well established in taxation, pensions, labour markets and social policy. Yet life sciences, despite being dominated by individual variation, long-term risk and interconnected outcomes, remains comparatively under-represented.
Pharmaceutical companies increasingly face exactly the types of questions that microsimulation-based modelling is best placed to answer: understanding long-term disease progression, patient heterogeneity, multimorbidity, and the future impact of treatment, prevention, and other interventions.
Historically there has been a practical reason for this.
Patient-level microsimulation is computationally intensive. Running millions of individual life-course simulations across complex diseases has traditionally required hours – or even days – of computation. That has inevitably limited both its adoption and the types of questions it could realistically address.
At the Congress, our CTO, Odhrán McConnell, presented our work on executing 100 million individual life-course simulations in 100 seconds using a cloud-native architecture. While this was just one contribution among many excellent presentations, it reflects a broader trend: computational performance is becoming far less of a constraint than it once was.
The significance is not simply that models run faster. Reducing simulation times from hours or days to seconds removes many of the practical constraints that have shaped modelling approaches. Larger populations can be simulated, more scenarios explored, and more complex patient pathways represented without compromising methodological rigour.
The real performance metric is therefore not “How fast does the model run?” but “How quickly can we move from a strategic question to a scientifically robust answer?”
For chronic diseases, that shift is particularly important. Treatment pathways are becoming increasingly complex, patients accumulate multiple conditions over decades, and decision-makers need timely insights across clinical, economic and operational dimensions. Patient-level simulation is uniquely suited to this complexity, provided it can operate at the necessary scale.
That then leads to the key question. If computational performance is no longer the principal limitation, which pharmaceutical and healthcare decisions require patient-level simulation, and where does it provide insights that simpler approaches cannot?
Microsimulation becomes valuable when averaging patients would materially distort the decision. Here are some examples where that threshold may be met.
Which indication should be developed?
Consider an asset with five possible indications before Phase II.
Portfolio teams already compare market size, competitive intensity, development risk and commercial potential. But the epidemiological evidence underpinning those comparisons is often inconsistent: different definitions, assumptions and patient segments are used for each disease.
A common microsimulation framework would compare:
The purpose is not to replace commercial forecasting. It is to answer a more precise strategic question: Does a large headline prevalence conceal a much smaller addressable population, or does a smaller disease contain a subgroup with high unmet need? Here, patient trajectories could change how indications are ranked.
What is the real addressable population?
The pharmaceutical industry frequently needs more than a prevalence estimate. It needs to know:
This is particularly relevant in rare diseases, and diseases with substantial underdiagnosis. Microsimulation can model how individuals enter, leave and move between eligible groups over time.
To what extent does earlier diagnosis change outcomes?
The value of screening cannot be established simply by counting additional diagnoses. The decision depends on what happens next:
This requires linking the timing of diagnosis to the subsequent individual pathway. It is especially important in “silent” diseases where patients may progress for years before identification, including CKD, hypertension and some rare kidney diseases.
How do interconnected cardiovascular, renal and metabolic pathways shape long-term outcomes?
Obesity, hypertension, diabetes, CKD, heart failure and cardiovascular disease are usually analysed separately, despite continuously influencing one another.
A therapy or prevention strategy may affect weight, blood pressure, glucose, kidney function and cardiovascular events over different time horizons. Its full value may therefore be missed by a model confined to one condition.
Microsimulation can follow individuals across these interacting pathways and answer questions such as:
This is not simply “future burden modelling.” It is disease-system modelling for portfolio, access and policy decisions.
Where AI fits
AI could fundamentally change how microsimulation models are built and maintained.
One of the biggest barriers to microsimulation has not only been computation, but it has been the effort required to construct and continuously maintain highly parameterised disease models.
Building a credible model often requires many epidemiological assumptions drawn from disparate sources, each of which must be identified, interpreted, validated and periodically updated.
This is where AI may have its greatest impact.
Rather than replacing the disease model itself, AI has the potential to become an intelligent evidence layer that continuously:
If computational scalability and AI-enabled evidence maintenance mature together, microsimulation could become a continuously updated decision-support capability.
Where next?
Perhaps the real lesson from this year’s IMA Congress is that the next breakthrough in microsimulation will not be computational.
It will come from applying it to the decisions where patient heterogeneity, long-term disease dynamics and complex intervention pathways genuinely change the outcome, and simpler methods are inadequate;
The question is no longer “Where can microsimulation be applied?” It is “Which decisions cannot be made properly without it?”