Analysing Patient Flow

In this King's Together project, we are investigating opportunities for applying performance modelling and agent-based modelling techniques to the problem of optimising patient flow in hospital settings.

A key performance indicator for NHS A&E departments is that 95% of patients should be admitted, discharged, or transferred within four hours. This KPI is consistently missed across England. A recent briefing report [1] identified reduced patient flow as a key contributing factor to this problem. But what influences patient flow in hospitals? How would changes to bed-capacity planning or the processes involved in preparing beds between patients increase or decrease patient flow? How do “winter pressures” [1] affect patient flow and why? While a number of statistics are available, there is a lack of understanding of the underlying forces and effects. Unless these can be understood in general and in their instantiation to specific hospitals and departments, it will be impossible to improve hospital management and meet the four-hour KPI.

How patients “flow” through hospital departments and their specific interactions also has a strong influence on other aspects of their overall experience – for example patient safety (e.g., determined by the accuracy of initial diagnosis and triage) and patient privacy and dignity – especially at times of high workload.

In computer science, the use of models and simulations for the predictive analysis of system performance have long been studied and are well understood. These models describe system processes focusing on resource requirements. Given a stochastic model of system workload, process models are simulated to derive a prediction of overall performance properties. Changing aspects of the system model and studying their effects on overall performance allows for what-if analyses. This is very similar to the problem of patient flow. Modelling the processes and resources involved in managing beds and patients enables simulation of patient flow and improved understanding of how process changes may affect it. This has generally been recognised: over the last decade there has been substantial research applying different performance-modelling and –analysis approaches to specific aspects of hospital patient flow [2, 3]. However, an initial analysis of the literature suggests that these approaches are either very narrow, so that they cannot provide a full understanding of complex effects and influences of such factors as nurse staffing, or are specific to one particular hospital environment.

Performance models do not account for the effect of people in the system. This is a major difference to hospital systems where people, their situational awareness and their resulting decisions and choices are key for overall effectiveness. Agent-based models (ABMs), where individual actors and their behaviours in a social scenario are modelled and simulated, have been used in the past to model the beliefs, decisions, and interactions of people in complex social systems – for example to study financial markets or the spread of infectious diseases. Agent-based simulation can also improve our understanding of patient flow as demonstrated by some initial research in this area [4]. In particular, ABMs are excellent tools for analysing issues of patient safety, disease spread, or issues created by conflicting or problematic beliefs of people with different roles within the system (e.g., issues of trust between experienced A&E staff and air ambulance staff leading to duplication of activities undertaken). However, to support analysis of system efficiency (as required for KPIs), ABMs need to be integrated with concepts from the performance-modelling world.

Developing the initial models, or adapting them to a new hospital context, is itself a challenge. Presently, this still requires substantial modelling and data acquisition effort from the ground up, making it very challenging to apply any modelling support across hospitals. In software engineering, domain-specific modelling languages (DSMLs) aim to capture recurring characteristics of a problem, allowing modellers to focus exclusively on what is new or different in each problem instance. Thus, the development of a DSML for patient flow may simplify the adaptation of existing models to new contexts, something that has so far not been attempted.

There clearly is the potential to substantially improve understanding and management of hospital patient flow. By combining expertise from computational modelling and analysis, software engineering, clinicians, nurses and hospital management, we will aim to establish King’s leadership in the field of modelling and analysing hospital patient flow. Building on a background of pre-existing work on performance modelling and agent-based modelling of hospital settings, this King’s Together project aims to produce initial proof-of-concept solutions and a substantial grant proposal to develop novel techniques and tools to support decisions by hospital managers in the UK and abroad.

This overarching aim is underpinned by the following objectives:

  1. Undertake a detailed survey of the current state-of-the art in modelling and analysing patient flow, its influencing factors, and its impact on NHS KPIs and patient experience;
  2. Undertake qualitative interviews in a purposive sample of hospital staff whose working lives are most impacted by patient flow in order to identify perceived key factors improving or hindering patient flow;
  3. Develop initial proof-of-concept DSMLs, models and analyses to assess feasibility and identify key scientific challenges;
  4. Publish any initial findings from the proof-of-concept models and analyses as well as from the qualitative study; and
  5. Develop a joint proposal to support substantial research in modelling and analysis of hospital patient flow.


Journal Articles


  1. Interprofessional barriers in patient flow management: an interview study of the views of emergency department staff involved in patient admissions
    Olga Boiko, Matthew Edwards, Steffen Zschaler, and 2 more authors
    Journal of Interprofessional Care 35(3), 2021