Working with the national collaboration to coordinate covid-related analysis, and the NHSE/I Midlands region, the Strategy Unit has produced a ‘systems dynamics’ model of waiting lists for planned care. The model is freely available for non-commercial use across the NHS. Here, Steven Wyatt and Mike Woodall explain what we did and how we did it.

Health services have changed radically and rapidly in response to the covid pandemic.  Services were redesigned to manage anticipated surges in covid cases and associated risks.  As the initial system shock subsided, attention has turned to the unintended effects of these changes.  Access to planned acute care, diagnostics, outpatient care and planned surgery are now significant concerns.

The pre-covid position was far from perfect. For several years prior, waiting lists for planned care had grown; waiting times had increased; breaches of constitutional standards had become commonplace. 

Yet the analytics and management of waiting list had been largely perfected.  Processes can always be improved, but the story was one of demand outstripping supply - rather than of poor analysis or management.  We understood the dynamics of waiting lists and times. We knew what to do: even if resource constraints meant we couldn’t always do it.

This certainty has gone.  Covid changed many of the rules of the game simultaneously and radically. 

Patients’ willingness to present at primary care; their ability to get an appointment; GPs’ willingness to refer to secondary care; the willingness of patients to take up hospital appointments; mortality whilst waiting; hospital staffing levels; diagnostic, bed and theatre capacity.   All of these parameters have changed. Some will return to pre-covid levels; others will find a new natural level. 

Circumstances have changed – and so have potential management responses.  New national resources may become available; Nightingale capacity could be repurposed; constitutional standards might be dropped or reset.  Responses that would have been unfeasible, uneconomic, or inappropriate only a few months ago might become real possibilities.

How will clinicians and managers respond to such fluidity, dynamism and uncertainty? 

Pre-covid models will be of limited use. Their founding assumptions have melted.  Some will therefore abandon any attempt to model - relying instead on instinct. But crisis management can’t be sustained. It will prove inefficient and error prone. 

A better option is to embrace the complexity and uncertainty: to evolve a new set of methods and models better suited to new and fluid circumstances. 

The key here is not to crave certainty. It does not stop everyone, but nobody can say with confidence how the next few months will pan out. We need approaches that understand which, of the many new unknowns, have the greatest potential to influence outcomes. We need approaches that can identify, prioritise - and then remedy – limiting factors in our knowledge.

This thinking underpins our waiting list simulation model.  We wanted to offer analysts and managers a facility to safely explore uncertainties and scenarios. We wanted to them to see how different strategies might fair under changing circumstances.

So we chose to build these models within a system dynamics (SD) framework.  SD is a well-tested, agile modelling approach that tracks changes in stocks and flows.  Stocks are quantities of interest (e.g. patients on a waiting lists, staff, beds). Flows are mechanisms by which stocks increase or deplete.  SD has its limitations. But where delayed effects, accumulations and feedback loops are present, then SD is a good fit. (Alternative modelling paradigms, such as discrete event simulation would be more suitable if the key concern is understanding how the distribution of waiting times are likely to change in the future).

In producing the model, we found that:

Waiting lists did not increase in the few months after the lockdown. This is not cause for celebration

The two main factors that determine the change in waiting list size are the rates of planned care activity (removing patients from the waiting list) and rate of new referrals (adding new patients onto the list).  In the months immediately following lockdown, rates of outpatient and inpatient activity reduced dramatically.  But referrals also reduced. Patients stayed away from - or were unable to book an appointment with - their GP.  The number of patients waiting for treatment fell from 4.4m at the end of February to 3.8m at the end of May.  A considerable referral backlog is building up in the community and in GP practices. 

Services are planning the dark. Modelled results are sensitive to factors we know little about

To produce estimates of the future waiting list size, our model requires a set of assumptions.  Supplying these assumptions is a sobering process.  There are many factors where our estimates are speculative.  How quickly will GPs clear the referral backlog? How many symptoms will clear before the referral takes place?  How will theatre productivity be affected by infection control measures?  How many patients will refuse appointment slots?  When will these patients feel more comfortable entering a hospital? 

Analysis and research needs to be directed towards these uncertainties. In the meantime any plans should incorporate sensitivity analyses allowing these values to vary within plausible ranges.

Waiting lists may yet reach unprecedented levels

Despite these uncertainties, we can use a set of realistic assumptions and observe the model outputs.  This process indicates that waiting lists could double or even triple by the end of March 2021. 

This accumulation takes place over three phases.  In phase 1, referrals are supressed at something like the same level as hospital activity resulting in small changes to the waiting list.  The second phase sees the waiting list grow rapidly as the referral backlog is cleared and hospital activity begins to increase. The third phase sees sustained waiting list growth.  The referral backlog is cleared, and referral rates return to pre-covid levels.  Meanwhile hospital activity plateaus at a level below pre-covid level as a result of infection control measures.

Hope of a rapid return to pre-covid waiting lists is misplaced

One of the values of simulation models is that they allow us to ask questions that could never be entertained in real life.  So we asked: by how much would hospital resources (staff, beds, diagnostics and theatres) need to be increased to return the waiting list to pre-covid levels by the end of the financial year?  The answer is that a step change of around 80% in resource levels would be required.

The answer’s lack of realism is instructive in itself. So too was the fact that different resource types - staffing levels, theatres, beds, diagnostic capacity - needed to be increased simultaneously. There is simply no rapid way back to pre-covid waiting lists.

Feel free to use or adapt our model

We are part of the NHS analytical community. So our model is freely available for non-commercial use. It is part of a broader collaborative effort to coordinate covid-related analytical outputs. We want it to be widely shared and we hope it is of value for local planning.

There are many different questions about the impact of Covid on waiting lists for planned care.  No single model can address all questions well. The chances that our model will coincide with the exact questions local teams might have are slim.  So here too are suggestions for using our work to support your own:

  • As a point of reference.  It’s often useful to see how someone else has approached a similar problem.  You can compare approaches and consider the strengths and limitations of the two methods, the assumptions made, the simplifications applied, the model functionality etc.
  • As a source of assumptions. One of the key challenges when modelling the outcome of some future scenario is to parameterise your model with assumptions about the timing and scale of certain impacts.  No-one can know these with any certainty. So our assumptions can be a useful reference point.
  • As a source of data (or data wrangling code). We have assembled large quantities of data.  This data wrangling process is time consuming.  You might therefore want to use these datasets as inputs into your own model(s), or adapt the data wrangling scripts (r and t-sql) to produce your own bespoke data tables.
  • To run specific scenarios supported by the model Our model allows parameters to be adjusted to create new scenarios.  You may wish to run locally relevant scenarios within our models.
  • As a starting point to develop a more complex model.  It may be that with some additional variables or functionality, our model might be able to address some new questions that are relevant in your area.  Feel free to use our model as a starting point and add complexity and functionality as required. 

Making our work freely and easily available means that we don’t always hear how useful it has been. So we would also ask that you let us have your reflections and suggestions:

Email: Steven Wyatt ( and Mike Woodall (  

Twitter: @Strategy_Unit


Instructions on accessing the model and working with it are available here; 


For those requiring an introduction to the model and the System Dynamics method, an explanatory video is available here or here.

Supporting / related analysis is available on: