Haukeland University Hospital in Bergen (Norway) and Dynaplan have a long history of working together on topics as varied as healthcare demand forecasting, patient flow management, strategic workforce planning, and clinic resource optimisation.
Right from the start it was clear that COVID-19 would have an impact both on supply and demand for healthcare services; but nobody could know to what extent and for how long. When the situation today – and the outlook for many tomorrows to come – are significantly different than what we are used to, simulation-based scenario planning is what we want.
Thanks to Pål Ove Vadset and his colleagues for the opportunity to work on this project and for their contribution to shaping the result. Thanks also to my great colleagues, Finn Olav Sveen and John Huertas, who created the data transformations and cockpits, respectively.
This project involved a team of people from Haukeland university Hospital and Dynaplan. Initially, several on-line meetings were devoted to answer the following questions:
- What will be the benefit of the Smia models, and to whom?
- When and how will the Smia models be used, and by whom?
Variation in supply and demand is relevant not only during the pandemic; fluctuations are normal for healthcare service provides. The team wanted a solution that can assist planning in stable as well as turbulent times.
During the pandemic, but also in general, patients are prioritized according to criteria such as severity of their condition. It is of critical importance that acute cases and intensive care are treated in a timely manner. At the same time, care for elective patients (patients who are referred to the hospital by their doctor) should be optimised.
The overall objective was formulated like this:
The project should develop Smia models that can help hospitals prioritize acute and urgent cases when capacity is insufficient, while ensuring full utilization of capacity for handling remaining patients.
The team identified a need both for monitoring the development of the situation, and for planning ahead. Users of the Smia models would be planners at different levels of secondary healthcare (hospitals).
For the pilot phase, Section of Health Service Development would use the models to service other entities of the organisation.
The Smia models coming out of the project should provide these features:
- A base view of the “normal” situation (e.g., before the pandemic).
- A snapshot view of the current “as-is” situation.
- A simulator for studying future scenarios subject to user-specified timeframes, assumptions about inflow and capacity, and patient flow management policies.
Two separate, but interconnected, Smia models were developed by Dynaplan: “Patient flow metrics” and “Patient flow balancer”.
Patient flow metrics
The metrics model is a static model (no simulation, only data transformation and statistics) importing patient flow data. This model is used both for the base and the snapshot view of the patient flow.
Inputs to this model are obtained from the recorded registrations, examinations, treatments, controls, and discharges of patients at all or a selection of the entities. It turned out to be a formidable job to validate the raw data and make a transformation into a format suitable for analysing and visualizing the stocks (waiting lists) and flows (production) of secondary healthcare.
For data security and GDPR reasons, the processing of information involving patients and treatment of patients is performed by the hospital and stays within the hospital.
Patient flow balancer
The Patient flow balancer receives data containing no individual patient information from the metrics model described above. Filled with data, the Patient flow balancer is shared via Dynaplan’s cloud service for model sharing.
Any user within the health region who has been granted access to the shared Patient flow balancer can develop scenarios for managing the patient flow at any level of the organisation.
Under the hood, the Patient flow balancer implements an elaborate patient flow network, illustrated below:
Twelve states and near 150 transitions are prioritized by urgency (coloured boxes) and constrained by capacity.
The time horizon for the simulation can be freely set from a few weeks up to a full year. The simulation is performed with a time step of one day.
The simulation cockpit provides overviews as well as detailed views of the state and development of the patient flow. Any imbalances between supply and demand are visualized, alerting the user to bottlenecks and declining service level for the different patient groups.
Assumptions about the future are initialized from the base data. The user can make modifications to the assumptions by increasing or decreasing supply (capacity) and demand (registrations).
Policies for managing the patient flow involves measures like the following:
- Increase time between controls.
- Reduce number of controls per patient case.
- Forward registration to private healthcare providers.
At the time of writing (October 2020), the rollout of the Smia models is still in its pilot phase. Already the first runs provided insights – some quite counterintuitive. As an example, users expected that increasing the time between controls would reduce demand. Counter to intuition, there will be only a temporary drop in demand (not shown by the model) before demand returns to its original level. To get a lasting reduction in capacity needed for controls, the average number of controls per patient case must be lowered.
As another example, the team discovered that the hospital might be missing out on an opportunity to reduce waiting times for elective patients. Under COVID-19, referrals to the hospital has in general declined. If the hospital manages to run at full capacity, the overall net patient flow will become negative, i.e., the capacity will exceed demand and make it possible to reduce waiting times for elective patients. This opportunity seems not to be used at the moment, as the potential capacity is reduced by management as part of a policy to limit spread of COVID-19 among employees.
As we have discussed earlier in the waiting list project, bringing capacity above demand might not lead to a reduction in waiting times for healthcare. A change in waiting list management policies is called for.
Use of the solution for other hospitals
The models described here cover any number of different types of policlinics and wards. The models support arbitrarily complex organisational structures.
When adapting the solution to other hospitals, queries for obtaining patient data from the local system must be created. Apart from that, customisation is mainly done via parameters and metadata (such as organisational hierarchy).