Jan 25 19 tweets 15 min read
Does #COVID19 crowd out care for non COVID patients in the #NHS? Has this led to a loss of lives? Are the numbers negligible? The short answers are: yes, yes & no!
Paper ➡️ bit.ly/33XMyHB & a long🧵on how we capture non COVID19 excess deaths & much more ⬇️ 1/n
Lets start with a headline result: we estimate that for every 30 #COVID deaths there is at least one avoidable non COVID excess death in 🏴󠁧󠁢󠁥󠁮󠁧󠁿 hospitals. To arrive at this we use cool #NHS data which makes for a great #EconTwitter #econometrics #DataScience teaching example. 2/n
The #NHS has population individual level hospital episode data (HES) linked to death certificates. For each admitted patient, they predict P(Death|X). This is an out-of-sample prediction coming from Lasso logistic regression model trained on data from the last 3 years. 3/n
The model considers individual variables such as age, mode of admission (ambulance, walk in etc.), pre-existing conditions, diagnosis, etc among others. The model does a good job at predicting out of sample accurately the # of deaths across NHS providers. At least until … 4/n
…Mar 20 from then onwards there is structure in the implicit residuals with E[Observed - Expected(X)|X] >0 suggesting an omitted variable bias in the trained model. Cumulatively this is at least 4000 excess deaths from Mar 20 to Feb 21 alone. Now, you could be worried… 5/n
…that this just captures COVID deaths. Thankfully the @NHSDigital removes ALL hospital episodes that have a #COVID19 diagnosis (which you get testing positive & note, all patients are routinely tested on admission & in hospital). Further, all deaths mentioning COVID19 on... 6/n
the death certificate are removed. This is the most comprehensive measure of COVID19 deaths. So we genuinely capture conservatively the # of excess deaths among non COVID19 hospital episodes. The 4k relative to 120k gives you the 1 in 30 but could be as high a… in 25. Now,.. 7/n
What is driving the variation in non #COVID19 excess deaths post Mar20? Surprise, surprise, pressures as e.g. measured by changes in provider-specific #COVID admissions appears an important omitted variable driving the variation. We can explore a bit what type of patient… 8/n
… ends up being more likely to die under pandemic conditions? This data is not as nice (loads lot of suppression/low case counts… but it is basically a bunch of acute conditions. Most people can relate to "acute myocardial infarcrtion = heart attacks”.. so 9/n
There is a non trivial and large increase in non COVID19 excess deaths among patients seeking help for non COVID19 reasons. You basically dont want to be having a heart attack in COVID times in an area with loads of COVID pressures. We could stop the paper here but,… 10/n
…we do MANY more exercises exploring in essence what is going on under the NHS' hood. Specifically, we look at the pandemic's effect on A&E care & waiting times; specialist referrals; diagnostic waits and… ; access & quality of cancer care…11/n
We study patterns before & after the arrival of #COVID19 & across the recurring waves. For cancer care, the long run mortality impacts are most obvious: e.g. across #NHS patients with lower gastrointestinal cancer are now 20p.p. less likely to begin treatment within 62 days…12/n
The ongoing #pandemic pressures keep cancer treatment delays up. We estimate there to be 32k missing cancer patients and more than 50k patients that had their treatment delayed by > 4 weeks which has mortality implications in the months and years to come. Now you could…13/n
Think that this may all be due to lockdowns causing pent-up demand. There is a some of that but its important that on intensive margin the recurrent demand pressures due to COVID with COVID19 induced staff absence rates among patient facing staff is making things worse…but 14/n
…the good news is that among providers with higher #Vaccination take-up, the non COVID19 excess mortality is weaker suggesting that #MandatoryVaccination may reduce this contributing factor to #NHS pressure. But best would be broad vaccination take up in community…15/n
Lowering the #COVID19 pressures to begin with. The paper iI quite dense. The important message is that COVID19 care does crowd out non COVID care which prevents the NHS to provide its usual quality of care, producing notably worse outcomes for non COVID patients that... 16/n
could be avoided if there was broad vaccination take-up. I worry that recent NHS deals with private providers make the situation not notably better (see bit.ly/3IDSa9b) as private providers suck out more FTE/resources from the NHS while cream-skimming good risks...17/n
Meaning the govt in essence just transfers cash to private providers that further cannibalise the NHS without more output being produced as human capital is the constraining factor. Have a read and happy to get feedback on the paper ➡️ bit.ly/33XMyHB ENDS.

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!

# More from @fetzert

Nov 15, 2021
What is the epidemiological impact of a #falsenegative #COVID test? An important question in a high vaxx/low NPI context, but one that cannot be studied in a experiment for obvious reasons. Enter the UK, a reliable supplier of #naturalexperiments. 🧵⬇️
➡️bit.ly/3DhqQv5
On Oct 15, @UKHSA suspended an #Immensa lab, because of community reports of neg PCR tests following a pos lateral flow. There was loads of excellent reporting e.g. by @rowenamason @tomjs @JamieGrierson. NHS TT estimates that 43,000 individuals may have been given a .. 2/N
false negative result most concentrated in South West of England. Even across all of England, a notable increase in both absolute # and relative % of PCR tests matched to a positive LFD tests producing a negative result from early Sept to early Oct 3/N
Jun 23, 2021
Today is the 5th year anniversary of the 2016 EU referendum vote in which the UK had narrowly voted to #Leave the European Union. Unlike Trump, the impact is permanent and already caused notable damage. Here is a 🧵 of 🧵 with some past work and deliberations on #Brexit... 1/...
In one of the first papers we asked "Who voted for #Brexit?". The paper is a systematic correlational analysis of what is common to #Leave support across districts and within cities & we also show that a #Brexit model can predict LePen voting. Link: goo.gl/VzBo57 2/
Similarly, we augment the analysis using individual level data. This helps tackle whether correlational district level evidence is due to ecological fallacy. Open access at @ejprjournal goo.gl/sPzzwf 3/
Mar 9, 2021
So I am going to report on some lack of progress about a #FOIA request we launched to @PHE_uk last Nov to make data available on the #Excel error that resulted in 15k #COVID19 cases to not be contact traced in a timely fashion (whatdotheyknow.com/request/region…). The response so far is ...
quite underwhelming. In the paper we reverse engineer the geographic distribution of the missing cases which is far from perfect. We find that places with higher exposure to the contact tracing error saw a notable differential increase in infections and subsequent deaths.
Naturally we would much rather prefer to work with the actual data as the measurement will be more precise. And further, it would allow for a direct measurement of infections among contacts of individuals that were traced with a delay. But @PHE_uk do not consider this is of
Feb 28, 2021
So @UKHofficial did have a look at my paper on #EOHO and #COVID19 - they have gone to some lengths to try to cast doubt about my research, the methods & results (see ukhospitality.org.uk/page/SafeReope…). So here are their point-by-point lines of attack on my work and my response. Thread 🧵⬇️
Point 1: Misunderstanding the research design and aggregate data fallacy 1/
Point 2: Actually, EOHO did not increase restaurant visits that much. 2/
Nov 24, 2020
Timely #ContactTracing does matter fighting #COVID19. In a new paper (➡️ bit.ly/394Ebuo) we study a bizarre #Excel error in England that caused 16k cases to NOT be contact traced. We econometrically can link this blunder to ~ 120k new cases & 1.5k deaths...🧵⬇️1/N
Studying non-pharmaceutical interventions to fight #COVID19 is HARD because we hardly ever isolate specific individual policies as often many measures are taken together (lockdowns, school closures, masks,...). For #ContractTracing we also have mostly correlational evidence...2/N
Enter the English Test&Trace system hastily built on, what appears to be a set of XLS spreadsheets giving us a consequential natural experiment. On Oct 4, PHE announced that ~ 16k #COVID19 cases were not correctly reported resulting in a large jump in reported cases.. 3/N
Oct 30, 2020
Today I m sharing another paper on unintended consequences of a UK policy which makes me cringe at how my tax money is spent all the while debating #FreeSchoolMeals "Subsidizing the spread of COVID-19: Evidence from the UK’s #EOHO scheme". ➡️ bit.ly/3ed5Slo a thread🧵⬇️
The EOHO scheme was conceived to help the hard-hit restaurant businesses in the UK in the wake of 1st #COVID19 wave. The scheme cut the cost of meals & non-alcoholic drinks by up to 50% across tens of thousands of participating restaurants in the UK from 3 to 31 August 2020. 1...
The research leverages data from #HMRC’s own #EOHO restaurant finder app which was the go-to platform for people searching for EOHO restaurants in their neighbourhood, together with weekly data on new #COVID-19 infections measured at the granular MSOA level (5-10k residents) 2...