Covid Under Reporting
As with all things data related Covid-19 reporting suffers from a lack of availability, human conservatism and lack of understanding. All of this is completely understandable in the current circumstances but in time I believe the true numbers will slowly emerge and be even more shocking.
The ONS data found here widely acknowledges many of the issues with the data including five day delays in death registration and poor reporting over the Easter weekend.
Utilising this data is it possible to analyse the change in deaths per week from the past five years (2015-2019) and 2020. Using a smoothing methodology to reduce volatility and correction for population numbers gives a 'normal' number of deaths per week as illustrated by the black line.
Adding the reported deaths from Covid-19 to the average gives the grey area showing the steep incline from mid March. However when comparing the average to the 2020 total deaths this leaves the blue gap as unexplained additional deaths. I believe it is fair to assume that the majority of these are also caused by Covid-19. While this may appear relatively small on the chart the data suggests an under-reporting of Covid-19 deaths in England of 38%.
To put that in real terms the latest UK wide deaths on 26th April reported as 20,732 on Wikipedia could in reality be nearly 30,000. The smoothing methodology implemented also results in this being a very conservative estimate. The FT have conducted similar analysis suggesting "a conservative estimate of UK excess deaths by April 21 was 41,102". Further analysis in the US also suggests that real Covid-19 deaths could be 6.7 times higher than reported.
So ultimately what does this mean for us in the UK and around the world?
There is definitely a risk that lockdown measures could be relaxed too soon due to delays in reporting and under-reporting of deaths. This could cause second and further waves down the line. It also indicates how the scale of the pandemic wasn't truly realised until it had taken hold.
I believe that things will improve but tough decisions on maintaining restrictions will need to happen to avoid even greater human cost. I would love to hear your thoughts on the current situation and how data can provide great insight to aid the control of Covid-19 but also how if caveats with the data are not understood the wrong decisions could be made.
In the meantime keep safe, stay at home and we can only hope the total human cost of the pandemic is minimised.
*Update 28th April 2020*
The latest ONS data for week 16 (17th April) shows 21,157 deaths of which 8,758 have been classed as Covid-19. The 'normal' for week 16 is 9,789 deaths so there are an extra 11,368 deaths compared to the expected for that week. Once the Covid-19 deaths are taken off this leaves an extra 2,610 unexplained additional deaths. This suggests an under-reporting of around 30% and reduces the overall figure to 33%. Although this reduces the percentage this week has the largest number of unexplained additional deaths so far.
It is possible to hypothesise that over time reporting is improving as more is understood and issues are ironed out. It may also be possible that as the volume of deaths increase the areas of reporting that are working are larger than the areas that are not thus reducing the percentage. The sample size is also tiny so this could be a one off. I will check back in next week to see what impact another week has on the analysis.
Good article Tristan. Your simple sum is actually better than Chris Giles in the FT, who yet again has a) made a (smallish) error by not bothering to read the detail the ONS data and b) chosen to sensationalise the story by comparing ONS excess morbidity to the NHS daily numbers which everyone knows excludes most deaths outside hospital, rather than the ONS Covid-coded deaths as you have done (it's not in the quote above, but certainly on his tweet). The Economist had a good article looking at excess morbidity across a number of the key countries and regions. On the initial data most seemed to be similar to the UK in unexplained uplift roughly speaking - though in their first few weeks NY excess morbidity was very close to reported Covid deaths. Some of the uplifts will be attributable to other causes - cardiac victims not calling an ambulance, increased suicide etc. - but I suspect that much of your blue area is uncoded/under-reported Covid. It will be interesting to see how it progresses in the coming weeks.
Tom Bent