Emergency Department Admissions: Analysis of CDS Dataset (part 4)
I analyse an anonymised data dump of 1.9 million admissions records to the emergency departments of an undisclosed NHS Trust for the period June 2017 – September 2021
I deliberately short-changed readers with part 3 of this series because I want to introduce folk to the basic demographics of busy A&E departments within a sizeable NHS Trust. Let’s start out with how folk get to A&E in the first instance…
Arrival Mode
‘Ambulance’ includes arrival by air ambulance but I’m guessing readers will be wondering what ‘other’ means. From experience (and my own surveys way back when) I can report that this mostly means private transport, though it will include public transport and police transport in the case of those held in custody.
Bizarre as it may seem, mature males insisting on driving themselves to hospital whilst having a classic heart attack (ST elevation myocardial infarction a.k.a. STEMI) was not that uncommon back when we took a close look at these things.
I think what is going to surprise readers is just how many A&E admissions (75.3% of known modes) start with somebody making their own way to the department, or by having somebody drive them. The expectation of blue flashing lights arises from a weekly diet of TV drama.
Diagnosis
I’ve touched on this in terms of respiratory illness, infectious diseases and probable ILI/COVID but let’s have a look at the bigger picture for the period January 2017 – September 2021.
There were three diagnostic fields available within the CDS type 010 data dump and within each of these clerks had coded a total of 69 different diagnoses. If we take the primary diagnostic field (field 1) here’s what the top ten for this Trust for this period looks like:
I find this fascinating.
Respiratory conditions – other non-asthma catches my eye, these representing 6.1% of all known primary diagnoses. We’ve already seen that respiratory diagnoses peaked prior to the pandemic, with admissions dramatically falling away during 2020, so we need to get it into our heads that respiratory illness is always a thing in these cold, damp islands and always has been.
Cardiac – other non-ischaemia also catches my eye. What we’re essentially looking at here are arrhythmias, atrial fibrillation, the sensation of palpitations, tachycardia and that sort of thing.
Three lists of 69 possible diagnoses per list, whilst making for fascinating bed time reading, are not exactly conducive to statistical analysis so I undertook my usual trick of boiling these down into a few broad-brush categories. After plenty of cogitation I settled on 10 primary diagnostic categories these being:
Cardiac conditions (5.6%)
Respiratory diagnosis (7.3%)
Infectious disease (2.0%)
CNS1 conditions (3.5%)
Diabetic & endocrine conditions (0.8%)
Injury & intoxication (38.0%)
Gastrointestinal conditions (6.5%)
Haemotological & CVA2 (3.0%)
Allergies & Inflammation (6.1%)
No diagnosis made (10.7%)
The incidence of each over the data period is given in brackets and it will not come as a surprise to learn that injury and intoxication leads the way with 38.0% of all diagnoses made.
Treatment
I’ve already had a bit of a dig into this variable but we ought to step back and take in an overview. Some 102 different procedures are coded within the database and it would be mighty cumbersome to list these out, but what I can do is grab a screen shot for the top ten procedures in descending frequency:
There you go! These top ten treatments represent 84.2% of all registered procedures in the database and therefore should be viewed as the bread and butter doings of this particular A&E.
These entries should be self-explanatory but I’ll just mention that electrocardiogram isn’t there because of symptomatic concern, but because of protocol. That is, your ticker is always checked as a matter of course, with this change in practice coming into play back in 2000 when the National Service Framework for Coronary Heart Disease was rolled out by HMG. Those interested might like to click on this link, and the extra nosy might like to learn that this framework and its requirements was why I’d been sent down to A&E as a ‘suit’ back when I bumped into the harsh realities of drug dealers, violent patients and prostitution.
As with diagnosis I decided it would be a good idea to boil this data down into something simple yet tasty. Thus I carefully picked my way through the 102 procedures and assigned them to one of the following 7 main treatment categories. The figures in brackets reveal the incidence of these entries in the database:
Cardiac procedures (8.0%)
Respiratory procedures (1.4%)
IV3/Central Lines (8.3%)
Fracture/joint care (5.2%)
Wound care (6.1%)
Medication administered (23.9%)
Guidance only/no treatment (41.6%)
If I were to place bets at this point I’d bet good money that readers will be surprised to discover that over a third of all A&E admissions ended with just guidance (oral and/or written), though the screenshot above should have given the game away.
Disposal
Not a pleasant word, I admit, but this is how the CDS type 010 dataset goes about coding what happens to each and every admission when stuff has been done and dusted. In terms of the entire data sample of 1,928,918 admissions between January 2017 and September 2021, this is how things went for folk:
I guess I better trundle out a few words of explanation, starting with three acronyms: NFA = no further action; F/U = follow-up; OP = outpatients.
‘Other’ is the weirdest discharge category since this includes discharge to police custody as well as discharge to prison/custodial centres (among other things). ‘Referred to other’ means other healthcare provider, for example to physiotherapy. ‘Transferred’ means direct transfer to another hospital (i.e. an admission to a ward somewhere else). The way to look at this is that if somebody survives their visit to A&E they can be discharged somewhere, referred somewhere or admitted to a hospital.
As can clearly be seen the most likely outcome of a visit to A&E is discharge with no further action, this representing 55.8% of all known outcomes over the period in question. In second place we have admitted to hospital representing 16.1% of all known outcomes, with the bronze medal going to discharge home with follow-up at 10.6% of all known outcomes. Together these three big hitters represent 82.5% of all known routes to disposal.
When it comes to analysis I shall simplify matters by lumping together all five routes of discharge, then lumping together all four routes of referral. I’ll then lump together admitted to hospital/transferred and call these ‘hospitalised’, leaving died in department as a sole, sorry category. This simplistic 4-level categorical variable will account for 1,896,647 of the 1,928,918 admissions (98.3%) owing to missing data which isn’t bad going for the sharp end. Now for some demogs…
Age & Sex
We observe a pretty equal split between male (49.6%) and females (50.4%) for those 1,927,244 admissions where sex has been registered, with similar mean ages of 39.5 years for males and 41.7 years for females. I think the surprise here for some readers is that A&E really is a youthful place, and I shall furnish a bar chart of decadal age group by sex to bring out a couple of interesting points:
The A&E departments for this NHS Trust are dominated by the younger members of the public, with the 20 – 29 year age group putting in a particularly strong showing. At the older end of the spectrum females tend to out-gun males (possibly because fewer males survive to old age) but in the youngest category of 0 – 9 years it is males who out-gun females in terms of admissions.
Quite how representative these profiles are for the catchment population I cannot say; there may simply be more young chaps and older women, but these two groups may also be more susceptible in some way. We simply cannot tell. What we can glean from this analysis is the somewhat youthful vigour of the case profile that A&E staff in this NHS Trust face on a daily basis.
Vaccination Status
Bringing up the rear is vaccination status and I fancied a quick and dirty cut ‘n’ paste:
Don’t forget that this analysis is for the whole period January 2017 – September 2021 so a whole bunch of unvaccinated is to be expected: I’ll be getting into the nitty-gritty of vaccination outcomes later on.
What this table usefully reveals is that nobody had received a booster (dose 3) prior to admission and that we’re talking a total sample size of 137,491 admission records for vaccinated folk that should be juicy enough to get a decent handle on things. But all in good time!
Kettle On!
Central Nervous System.
Cerebrovascular Accident (stroke).
Intravenous.
Some ideas:
1. Maybe do the same things you did in parts 2 and 3 (i.e. look for disproportionately of occurrence around covid time), but do it brute-force across all diagnoses and all treatments (or at least the most common ones). In other words, don't assume what covid looks like a priori, but see if something can imply what covid looks like. I doubt going by code groupings would work for this, as opposed to isolated codes.
2. In the above, see if all the presumable negative controls (e.g. laceration) don't proportionally spike around covid.
3. Make a group of diagnoses composed of the highest hits from the above (i.e cherry pick) and see if it's proportionality is totally nonsensical or not.
4. Look at rates of death and hospitalization for every diagnosis. See how those rates changed around covid.
I would posit that a simultaneous spike in both proportionately and in fatality rate (or admission rate) of a given diagnosis could indicate covid. Maybe this could form a code grouping as well.
sublime come sempre !!!!♥