3 September 2020
Lies, Damned Lies and Coronavirus
Death Rates
by David Chilvers
This week*, we look at the factors affecting death rates from COVID-19 which, in spite of repeated assurances from politicians, are still not as clear as they should be some five months into the pandemic.
Early analysis of death rates suggested that there were some key factors affecting death rates from COVID-19 including:
- Gender – males had a higher death rate than females
- Age – older individuals had higher death rates than younger, particularly those aged 75+
- Ethnicity – BAME individuals had higher death rates than others
- Co-morbidities – those with existing health issues had higher than average death rates
- Housing density – those in densely populated areas have higher death rates
- Deprivation – those in deprived areas have higher death rates
- Occupation – those working in some customer facing occupations have higher death rates
The problem with simple analysis, one factor at a time, is that correlations between the factors cannot be taken into account. For example, many areas with high BAME populations are in deprived urban areas, with high population density; and many BAME individuals work as taxi drivers, bus drivers, in the NHS and social care, all occupations with high levels of interaction with other people and with higher than average death rates from COVID-19.
Trying to disentangle the combined effect of these factors is important. For example, if being part of a BAME community is a risk factor irrespective of occupation, then it is understanding the impact of ethnicity that is more important than understanding the impact of occupation. In an ideal world, the importance of each factor having isolated this from the impact of other factors becomes important as strategies and mitigations can be aligned with these key factors. If ethnicity was more important than occupation, then strategies could be aimed at reducing risk for the most vulnerable ethnicities with less focus on occupation. Of course, it might be possible to target all vulnerable groups but in practice resources need to be targeted to those groups most at risk (and groups where the risk was due in large part to other factors would need less targeting).
In May, Professor Chris Whitty confirmed that major studies were under way to help solve this problem and allow the ranking of risk factors to be better determined. In June, PHE produced their report entitled “Coronavirus (COVID-19) related deaths by ethnic group, England and Wales: 2 March 2020 to 15 May 2020”, which in particular looked at differences between ethnic groups having standardised for geographic location and population density, living arrangements, socio-economic profile, and working conditions. The results were adjusted for age, but no account was taken of existing health conditions. Given that at least 80% of those dying from COVID-19 had some pre-existing health condition, this omission renders much of the analysis in this report of limited value.
The ONS Statistician’s comments on the report were “ONS analysis continues to show that people from a Black ethnic background are at a greater risk of death involving COVID-19 than all other ethnic groups. The risk for black males has been more than three times higher than white males and nearly two and a half times higher for black females than white. Adjusting for socio-economic factors and geographical location partly explains the increased risk, but there remains twice the risk for Black males and around one and a half times for black females. Significant differences also remain for Bangladeshi, Pakistani and Indian men. The ONS will continue to research this unexplained increased risk of death, examining the impact of other health conditions.”
Since this report was issued on 19 June, there has been no update and ranking of risk factors remains elusive to the ONS. However, methods of analysis are available which ONS could have used and indeed have been undertaken by this author and supplied to ONS, but with no response. The method builds on the work ONS has undertaken to link death data to Census data but does this for local areas (rather than for individuals which is the method ONS has pursued). Both COVID-19 deaths data and Census data is available for relatively small local areas called MSOA’s (Middle level Super Output Areas).
There are just over 7,000 MSOA’s in England and Wales, each one covering 7-8,000 individuals. If COVID-19 death rates are high in MSOA’s with a greater number of older people, it is a reasonable deduction that age affects death rates. But the advantage of this approach is that all Census variables can be analysed simultaneously against COVID-19 death rates (deaths per 100,000 people) and a ranking of factors derived. Crucially, the Census data contains data on self-reporting of health issues in addition to a very wide range of demographic variables. There are some problems with this approach in that Census data is close to 10 years out of date, but it is likely that characteristics of areas will not have changed fundamentally over time. In addition, the method used has some limitations.
This analysis shows that the factors that are most important in determining COVID-19 death rates are Ethnicity followed by Gender, Age, Economic Activity, Health and Deprivation. Other factors such as Household size, Method of Travel to Work, Social Grade, Qualifications, Industry Sector worked in, number of bedrooms and Occupation are important but less so than those in the first list.
All the results for individual factors are consistent with those shown in the PHE Report. Ethnicity, for example, shows much lower death rates for those of White ethnicity, slightly lower for people of mixed race and above average for all other ethnic groups. Males have higher death rates than females and there is a substantial gradient by age. Those unemployed and with pre-existing health conditions have higher deaths rates, having already taken account of ethnicity, gender, age etc. however, the advantage of this type of analysis is that is does provide at least a basic ranking of the importance of each factor in its own right, allowing strategies and mitigations to be targeted at the most important factors.
Next week, we look at local area statistics, which are becoming increasingly important as strategies focus more on local responses to any uplifts in local infection rates.
*This article is one of a series, find last week’s article “Test and Trace” here.
