Prior Risk Of Death (part 2)

I attempt to derive a more sophisticated measure for how sick in-patients are in the first instance and find something rather unexpected

In part 1 of this mini series I introduced the concept of PROD (prior risk of death) as being an embellishment on the total number of diagnoses made in the EPR. I’m pretty sure that readers will understand why it is vital to assess just how ill folk were prior to admission and death, even for a sample of deceased patients, so today I will pursue the most peculiar finding of a drop-off in diagnosed conditions for the elderly.

The more I think about it the variable I am calling prior risk of death (PROD) is essentially prior risk of diagnosis. Hence it will favour the more common entries in the EPR. The danger here is to have minor conditions like the common cold dominate matters, which is why the score was derived using 11 major morbidity groupings rather than an abundance of specific and somewhat trivial diagnoses.

With all that said and done here’s another break down by age band but for the number of diagnoses made per patient prior to death:

If we compare this to the slides presented in part 1 of this series we observe an increase in case complexity all the way through a peak in the 90 – 94y age band, whereas the PROD peak stops short in the in the 70 – 74y age band. Curious indeed that a complexity score should out-run the commonality score.

Two Horns On The Same Goat

It occurs to me that the prior risk of death score, together with total diagnoses made, both serve to distinguish cases in two slightly different ways. To illustrate this I’ve derived mean prior risk of death and mean diagnoses made for each of the 18 age bands and plonked these down on a scatterplot:

This is most satisfying! We have a wonderful progression that reveals a complex relationship that sets in beyond 74 years of age. At this point the number of diagnoses continues to climb but their frequency amongst the general pool of the deceased does not, this being suggestive of an increasing diagnostic for less common conditions. Beyond 94 years of age even the number of diagnoses dwindles, this likely being a function of complex issues such as decision to treat and protocols employed on care of the elderly wards, especially for those with a poor prognosis. It is somewhat thought-provoking that those living to 104 years and beyond are numerical bedfellows with those who fail to reach their 24th birthday in terms of the diagnostic count – yet another issue that I need to address through model development.

This slide may be viewed as a kind of map, showing analysts which age bands to focus on if they want to refine results. I am inclined to identify four major groupings of: 18 – 64y, 65 – 79y, 80 -94y and >94y and may well use this knowledge in future analyses. By way of example, here’s a crosstabulation of this age grouping by incidence of chronic respiratory disease:

This is surprising because we’d bet good money on this increasing with age, yet it doesn’t according to the EPR! Thus, we either have to accept that chronic respiratory conditions don’t tend to plague those who manage to survive to extreme age or that diagnosis in the elderly isn’t what it should be. This sort of bias is going to make a mess of any analysis that is reliant on accurate reporting of a condition that impacts on the severity of COVID, and thus on any notion of vaccine benefit.

In attempting to model retrospective observational data arising within NHS patient management and data capture of such I appear to be wrestling a greased pig!

Summary

• Prior risk of death (PROD) is a score derived from the frequency of 11 diagnostic groupings within the dataset. As applied to an individual it provides a proxy measure of base likelihood of death at admission.

• Both PROD and the number of diagnoses made provide evidence of diagnostic tail-off in the elderly, which will impinge on assessment of vaccine harm/benefit.

Kettle On!