Bed Loss & Bedlam (part 3)

A look at bed occupancy within NHS England over the last 13 years

In part 2 of this series I took out my big spanner of ARIMA time series modelling and developed two predictive models for excess non-COVID death for NHS England for the period 2020/w6 - 2022/w33 using a couple of key covariates and the all important series I’m calling Bed Loss Fraction (BLF). Please refer back to part 1 to understand what I’m on about!

In doing so we discovered that non-COVID death is likely subject to certification errors with more deaths pronounced as COVID than the models suggest. This may arise from over-reliance on PCR tests that readily give rise to false positive test results, and it may arise from over-zealous physicians. Whatever the reason the models enabled allowances to be made for certification errors.

This slight correction enabled a more robust assessment of the likely impact of bed loss on non-COVID excess death, which was found to peak 3 - 4 weeks behind bed closure. A cumulative curve was furnished revealing bed loss associated non-COVID death to currently be running at just over 26,000 cases. That’s a lot of death and yet is only part of the story: we have still to estimate the impact of bed loss on COVID death.

In part 1 of this series I pointed out the rather curious decline in bed occupancy over the last 13 years and I wasn’t convinced this was due to improving health of the nation. I suspect a squeeze has been put on beds, so get that cake sliced and the kettle on for we’re going to delve into NHS England quarterly data, all of which may be found here.

Whose Bed Is It, Anyway?

First a quick word about bed types. We have day beds and we have night beds; then we have beds by speciality, for which the NHS suits decree headings of: General & Acute (G&A), Maternity, Learning Disabilities and Mental Illness. Then we have all manner of bed model for specialist purposes, with the mechanical ventilation bed rising to fame during the pandemic. Where a bed is located also matters, since we can have beds in high dependency units (HDU/ICU) and beds on general wards (both day wards and night wards). Then there are beds in places like the Coronary Care Unit and Stroke Unit. In the UK day-to-day bed management is the responsibility of the matron, though the Clinical Nurse Manager (big boss of Division nursing) will have a say, as will the General or Divisional Manager (big boss of cheque book). Then we’ve got the dreaded bed managers - a hospital-wide team whose purpose seems to be to make the life of the Clinical Nurse Managers as stressful as possible. Those beds you’d earmarked for elective cardiac surgery next week? Well, they’ve now been allocated to general surgery and will be full of kidney cases thanks to the bed management team. That sort of thing. Beds, beds, beds, beds, beds.

But it’s not just beds. If you can’t discharge folk back home or get them transferred to another provider then your beds block back. If your beds block back in the wrong direction irate nurse and medical consultants in Accident & Emergency soon let you know because their precious beds need fast turnaround and that requires a supply of G&A beds. Imagine a sausage machine on full pelt running out of skins halfway through Friday evening with no way of turning the machine off.

Then there’s the nurse to patient ratio. You can’t fill beds if you can’t meet the mandatory requirements for staffing for low and dependency care. Beds will go empty if a shift crumbles (nurses calling in sick or failing to appear) and/or the agency or nursing bank can’t supply sufficient of nurses at the right grades. I’ve seen senior nurses work straight through two shifts and more to cover for unavailable colleagues, this being commonplace for a busy city hospital. The Clinical Nurse Manager will ‘call it’ if staffing drops to critical levels, so unsupported beds are roped-off and even whole wards can be closed down. Please bear this in mind when looking at the following charts: a bed may be deemed ‘available’ but the nursing power to staff that bed may not!

The General Picture

To make things simpler I am going to lump together all non-G&A beds; that is to say Maternity, Learning Disabilities and Mental Illness beds will be counted in one big bed mass as ‘other’ beds. I’ve got 49 quarterly bed availability records in my pot stretching from 2010/Q2 to 2022/Q2 but please note these are calendrical quarters where Q1 means Jan - Mar. NHS bed counting suits are concerned with the fiscal year, so their Q1 means Apr - Jun. Please bear these definitions in mind if you decide to browse the source data! I shall start with a rather boring but very useful summary table of bed availability over the last 13 years:

By glancing at mean values we can see that night beds out gun day beds by a factor of 11x (131,957 / 12,013). This gives us a handy snapshot of the relative size of the bed pool with at least ten night beds to every day bed. If we do the same for primary speciality we see that G&A beds out gun all other beds by a factor of 4x (114,744 / 29,226). I shall leave subscribers to jab at their trusty hand-held calculators to derive more factors, thence to get a feel for bed use in English hospitals.

If we now flip to considering maxima and minima we find an extraordinary level of elasticity such that total bed availability fell from a peak of 156,238 to a trough of 128,271 beds. That amounts to a withdrawal of 27,967 beds or 17.9% of the national bed stock. The most extreme percentage bed loss is found for the modest number of day beds in the service of Maternity, Learning Disability and Mental Illness units where we observe a decline of 66.2% from maximum levels, the general pattern being a ‘hit’ for day beds. With these figures providing a general background let us now take a look at the dynamics of bed loss over time.

Bed Loss Over Time

They say a picture is worth a thousand words. In this case they may well be swear words for this is what they’ve been doing to the NHS:

Before COVID came along total bed availability across NHS England was declining at a rate of 340.7 beds per quarter if we assume a linear trend (p<0.001). That’s 1,362.8 less beds per year or 13,628 beds per decade. That’s a lot of beds to permanently close in a service that was already under pressure back in 2000 when I joined. Neither is the UK population getting any younger, so I view this as a deliberate attempt by planners to force the service into a crisis so extreme that it will be disbanded. In fact, this is already happening with many internal services now being contracted out. What the public perceive is a façade and we need more than a crayoned rainbow logo or clapping; we need a government and NHS executive that still believe in public health.

Life Without COVID

A cracking question to ask at this point is what would have happened without COVID? The big black bed hole in bed occupancy we discovered in part 1 clearly wasn’t just due to the government, senior advisors and NHS executive deciding to shunt the vulnerable and elderly into care homes in order to make way for the computer-predicted mega-wave of seriously sick cases that never happened. This, incidentally, being the simulation work of just one man and one department at a time of pending global crisis. Go figure the logic behind that and I suspect you’ll come up smelling of shares and cash.

No indeed, for what these illustrious leaders in positions of power decided to do is go and close beds at the height of an emerging pandemic. The NHS’ own quarterly figures reveal this extraordinary fact. There is no logic behind this, only corruption at the highest levels of public health. If that mega-wave did ever crash, what do you think would have happened to the millions unlucky enough to require hospitalisation at a time when managers were rapidly closing beds in a frenzied axe session? Correct! They would have died unnecessarily both at home and on the street in large numbers, and society would have collapsed there and then, leading to an instant “great reset”.

Crystal Ball

With ranting done I must now uncover my crystal ball to judge what might have happened to bed availability if the pandemic had not hit town. There are several ways of doing this but I decided to turn once again to my favourite big spanner of ARIMA time series modelling. If I let the expert modelling module within my stats package do its thing I end up with a model that generates this rather pleasing slide:

Here we see an incredibly simplistic ARIMA(0,1,0)(0,1,0) model do a rather fine job of tracking bed availability over time and providing us with sensible looking estimates of bed availability from 2020/Q2 onward. Those with a keen interest in stats stuff may wish to note a stationary r-square¹ of 0.840, which is rather healthy indication of goodness-of-fit.

This model allows us to estimate the numbers of beds lost because of some seriously warped management policies. The fact that I’m writing about bed loss and not bed gain during a global pandemic beggars belief, but here’s a table of raw numbers:

What we see here are observed and predicted values for mean daily bed availability along with an estimate of the decline in mean daily beds. Because a quarter nominally consists of 13 weeks each with 7 days, then we have to factor that decline up by a factor of 91x to arrive at the figure of estimated actual bed days lost to closure. These are big, big losses.

In Q2 of 2020 (Apr - Jun) alone some 1,063,153 beds were estimated to have been removed from use within NHS England, with the cumulative count peaking at 3,360,266 bed days lost because of management policies. Whilst it’s good to see that trend reversing irrevocable damage to people and the NHS has been done and nobody is being held accountable.

But We Didn’t Have The Staff!

A fair point indeed and one that I laboured at the outset. The good news is that I am sitting on staff absence records as well as bed occupancy levels so I’m in a position to assess the impact of staff absence on bed availability to see just how many beds had to be closed to ensure safe staff to patient ratios. Until then…

Kettle On!

1. Not to be confused with plain old r-square which is inappropriate for time series work.