26 October 2026

^{Boehringer Ingelheim and Microsoft are collaborating on patient-centered, AI-supported clinical development  (image: Boehringer Ingelheim)}

 

Boehringer Ingelheim and Microsoft are partnering to research digital patient twins: computational models that could help forecast disease progression and treatment outcomes. Beginning with lung health, they will investigate whether these predictions can support better-designed clinical trials and potentially reduce the number of people needed in control groups. The collaboration will examine scientific and technical feasibility through modelling, prototypes and evidence generation. 

The partnership's ambition addresses a practical problem in drug development. Researchers need convincing evidence that a treatment works, but recruiting enough suitable participants and following their progress takes time and resources. Better use of existing patient information could make that process more efficient.

 

What would a digital patient twin do?

 

Imagine a person entering a lung-disease trial. Their starting lung function, medical history and other measurements could help a computer model forecast how their condition might progress under standard care.

That forecast could help researchers account for the person’s expected disease course when analysing treatment results. Someone expected to deteriorate quickly and someone likely to remain stable bring different starting circumstances to the comparison.

In this setting, the term digital twin need not mean a three-dimensional virtual body or a model continuously fed by wearable sensors. It can mean a patient-specific prediction generated from measurements taken when someone enters a study. The model captures selected features relevant to the research question.

 

How predictions could strengthen a trial

 

Clinical trials commonly compare an experimental treatment with a placebo, standard care or another appropriate treatment. The control group helps researchers distinguish an intervention’s effects from changes that would have happened anyway.

Boehringer Ingelheim and Microsoft will explore whether digital twins can support synthetic or hybrid control arms, drawing on existing information or combining it with a conventional control group.

Several approaches sit beneath this terminology. An external control group uses data from patients outside the current trial. Model-generated forecasts use patterns learned from earlier patients to predict outcomes for particular individuals. A hybrid design can supplement an enrolled control group with external information.

The potential benefit is substantial: less unexplained variation can make a treatment effect easier to detect. Depending on the method and its performance, that could allow a smaller study to answer the same question with adequate confidence. 

 

What Unlearn’s experience shows

 

The Boehringer Ingelheim/Microsoft partnership enters a field where other companies are already pursuing defined applications. Unlearn, for example, builds disease-specific machine-learning models that forecast how individual trial participants might progress under a control condition.

Its Prognostic Covariate Adjustment methodology, or PROCOVA, uses predicted outcomes to help account for differences between participants, improving the precision of the treatment comparison.

In September 2022, the European Medicines Agency’s human medicines committee qualified PROCOVA as a statistical approach for randomised controlled trials with continuous outcomes, such as numerical changes in a clinical score. The methodology can support greater precision and, under appropriate assumptions, a smaller sample. For Boehringer Ingelheim and Microsoft, this provides relevant context.

 

The difficulty of comparing patients across time

 

The quality of historical data is central here. Patients in earlier studies may have received different treatment, undergone different tests or entered research at a different stage of illness.

If standard care has improved, a forecast based on older outcomes could make a new medicine appear more effective than it really is. Differences in measurement practices can also distort the comparison.

Patient diversity creates another challenge. A model developed mainly from people with one condition may perform differently for someone who also has kidney disease and takes several medicines. An accurate average forecast can conceal weaker performance in underrepresented groups.

Validation therefore needs to establish where predictions hold, how uncertain they are and which patients fall outside the model’s reliable range. The intended analysis must also address the risk of drawing an incorrect conclusion about the treatment.

 

What would count as progress?

 

For drug developers, the commercial attraction is clear. Reducing recruitment demands or avoiding an inefficient study design could save time and resources. Those gains depend on preserving evidence strong enough to judge a medicine.

The partnership’s lung-health starting point gives that broad ambition a focus. A useful next step would be to define the patient population, the outcome being predicted and the role the prediction would play in a study.

Researchers could then compare forecasts with observed outcomes in relevant patients and assess whether the approach improves trial design or analysis. Any claimed efficiency gain would need to survive those checks and regulatory scrutiny.

Boehringer Ingelheim and Microsoft have announced an investigation into that possibility. Its eventual value will be measured by whether researchers can obtain a dependable answer about a treatment with less time, fewer resources or fewer participants, and without weakening the evidence patients need.

 

 

 

Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.

👉 https://buymeacoffee.com/movetheneedle.news