Announcing our OMOP virtual ward and Hospital at Home data generator

Virtual hospitals and virtual wards are a simple idea with big implications for how care is delivered. Instead of keeping people in a physical ward for monitoring, clinical teams can oversee care while patients stay in their own homes. In an NHS virtual ward service people are enrolled into a programme, given equipment or short surveys to complete and are supported by staff who watch for early signs of risk and step in when needed.

Virtual wards are likely to play a more prominent role in how health and care is delivered, supporting the goals of the NHS 10 year plan to enhance care outside of traditional hospital settings and using digital innovation to improve patient safety.

Working with NortHFutures, Teesside University, Newcastle University, and Hull and East Riding Frailty Hospital at Home Virtual Ward, Hippo has been exploring what that can look like in practice through a virtual ward prototype focused on the NHS Hospital at Home programme and healthcare inequalities. As part of this work we have created an OMOP based synthetic data generator for virtual hospital and virtual ward scenarios. It already underpins our Hospital at Home prototype and we have made it available as open source so that the approach is clear, shareable and easier to build on in other settings.

“Ultimately our team will be using real patient datasets extracted from the Hospital at Home service to explore how medical signs associate with risk scores and how well a combination of objective and subjective inputs can predict patient trajectory within the Hospital at Home setting.

However, the data generator that the Hippo team has constructed has already been really helpful to the team for shaping our thoughts around the specific questions we will address, the data inputs we want to have access to, and how the outputs of the modelling will be visualised for use by health care professionals. So, in that sense the data generator has accelerated the timeline of the deliverables from the project”.

Grant Trewartha, Professor of Digital Health Interventions, Teesside University

For this model of care to be safe, teams need to know which patients are stable, who might need a review later in the day and who is likely to deteriorate and they need to see how inequality and place shape demand because communities experience illness and access services in different ways. That depends on having the right data in the right hands, used in a way that protects people and can be shared safely between organisations, which is often difficult while information governance work is still under way or several partners are involved, so progress can feel slow even when the ambition is clear.

Where this work fits

The generator sits inside a joint prototype that looks at how virtual wards and Hospital at Home services can safely support people at home while keeping an eye on inequality.

Together with our partners at NortHFutures, Teesside University, Newcastle University, and Hull and East Riding Frailty Hospital we are exploring how early signals relate to end of episode NEWS2 scores, what different intervention thresholds mean for day to day workload and how risk and demand vary between areas with different levels of deprivation. The aim is to help services move beyond intuition so that when they design or adjust a virtual ward they can see how those choices might affect real people and real teams over time.

We wanted a way to explore those questions before any live data flows were in place, and in a form that could move with partners rather than being tied to a single organisation. That is what led us to OMOP and synthetic data.

Why OMOP and synthetic data

The generator produces data in the OMOP Common Data Model. OMOP offers a standard structure for person, clinical and episode data so that analysis and models do not have to be rebuilt every time work moves from one environment to another.

Using OMOP brings three practical benefits for virtual ward and Hospital at Home services:

Teams can work with a familiar schema instead of a one off dataset for each NHS virtual ward
Queries, dashboards and models built on synthetic data can move more easily into live environments later
Informatics and data science teams can draw on the wider OMOP ecosystem of tools and methods rather than starting from a blank page

The data itself is synthetic. It is created from configuration, probability distributions and clinical logic, not from patient records. The intent is not to recreate real cases, but to create believable patterns of symptoms, measurements and interventions that look like the kind of signals a virtual ward service might see in practice, so that services can test ideas and build confidence without touching identifiable data.

Taken together, this helps reduce friction. Design, modelling and virtual ward pathway conversations can move ahead while work on live data access continues in parallel.

What the generator creates

At a high level the virtual ward OMOP generator creates a complete synthetic view of a virtual hospital service.

It builds a population with basic demographics and observation periods, then assigns conditions that are common in virtual hospitals such as chronic obstructive pulmonary disease, heart failure and postoperative care. Each person is linked to an area using open data from the Office for National Statistics so that attributes such as deprivation are part of the picture from the start.

On top of that population the generator creates virtual ward episodes with different trajectories. During each episode it produces daily symptom surveys and device measurements that change over time in realistic ways. An intervention engine applies clinical rules and records actions such as reviews, visits and escalations when particular thresholds are crossed.

All of this is written into an OMOP compatible DuckDB database with views that make it easier to explore time based signals and episode outcomes. In early testing the datasets support models that can learn to predict urgent reviews from the synthetic signals, which suggests the patterns are rich enough for meaningful modelling work.

How it helps in practice

Within the Hospital at Home prototype the generator is already helping multidisciplinary teams move from ideas to something they can see and question together.

Data scientists can experiment with models that use early observations, symptoms and location based attributes to predict outcomes such as end of episode NEWS2. Clinicians and designers can look at how often alerts would fire under different thresholds, how many reviews or visits a team might be asked to deliver and how that feels across a week or a month of episodes.

Because each person is linked to area level attributes, teams can also look directly at inequality. They can see how risk and service use differ across areas and consider what that might mean for outreach, capacity and pathways, rather than assuming the same patterns everywhere.

Perhaps most usefully, the generator gives partners something tangible to work with. Instead of discussing a future data pipeline in the abstract, they can explore a database, sketch simple dashboards and talk through scenarios with synthetic but joined up data in front of them.

Why we are sharing it

Virtual hospitals and Hospital at Home services are likely to play a growing role in how health and care are delivered. They rely on clinical judgement and thoughtful design, but they also need a grounded understanding of how risk behaves over time for different people in different places.

The virtual ward OMOP generator is one contribution to that wider effort. By sharing it as open source we want to make the approach more visible and give teams a practical example of how synthetic OMOP data can support safe experimentation in this space, alongside the service design and clinical expertise that these services depend on.