Limen

The hospitalisation risk, patient by patient.

For patients who take many medicines, Limen calculates this risk from their records.

A tilted white and green capsule in a sage and mint gradient; a tablet and a small capsule fade into the background.

One risk per patient
A probability of hospitalisation for each record, not an average.
Judged on each group
Measured separately on hospitalised patients and on the others.
A threshold set with you
The alert threshold is set with the care team, according to what each error costs.
Records stay with you
Hosting is chosen with you: on your servers or in the cloud.

The number that matters is not the success rate.

Hospitalised patients are a minority: a model that answered "no" for everyone would have an excellent success rate and miss every one of them. So Limen is read on both populations: patients who are not hospitalised cluster at low probabilities, hospitalised ones at high probabilities.

Diagram based on the project's chart.

What Limen changes for care teams.

A risk estimated for each patient

From what each record contains, Limen gives the probability that a patient will be hospitalised. Care teams see first the patients above the threshold.

The threshold is set with those who decide

The 0.5 threshold is only a default setting. In a hospital, missing a patient at risk costs more than one alert too many: the threshold is set with the care team and can change from one ward to another.

InsightPolypharmacy: why a good success rate is not enough

What the project taught us about how to judge such a model.

Read the insight

Frequently asked questions

What is Limen?

A model built by GTS that estimates, for each patient, the risk of hospitalisation linked to taking many medicines.

What data was it built on?

On 9,066,106 records. This volume lets the model see rare situations.

Why not judge it on its success rate?

Because hospitalised patients are a minority: a model that answered "no" for everyone would have an excellent success rate and miss every one of them. We measure it on each group.

Who sets the alert threshold?

The care team, with us. The 0.5 threshold is only a default setting; it is set according to what each error costs.