Postcode versus genetic code: ethnicity and racism in the time of COVID-19

Miqdad Asaria (LSE)
14th May 2020

Disclaimer

  • Data on COVID-19 is constantly being updated as the pandemic progresses
  • Data that is presented here represents the best available information at the time of compiling the slides
  • Data for different measures are available for different time points so care must be taken to read across results
  • Historical data are constantly being revised and corrected by the NHS, PHE and the ONS so precise reproducibility of any analysis presented here may be tricky
  • Despite this conclusions drawn are general and likely robust to these challenges

What do we know?

  • A lot of anecdotal evidence that large numbers of patients in hospital with COVID-19 are from Black, Asian and Minority Ethnic (BAME) communities
  • BAME communities make up 14% of the population of England (ONS 2011)
  • BAME doctors make up 44% of all NHS doctors but 95% of deaths (HSJ, 22nd April 2020)
  • BAME nurses make up 20% of all NHS nurses but 71% of deaths (HSJ, 22nd April 2020)
  • Areas that have the most COVID-19 deaths (population adjusted) are the more ethnically diverse areas such as Newham, Harrow and Brent (ONS)

COVID-19 by local authority

Deaths per 100,000 population

map

Cases & deaths by BAME population (%)

cases

asmr

Is this a London specific thing?

london

COVID-19 in hospital

Data from the NHS

  • Intensive care admissions data by ethnicity
  • COVID-19 death data published daily but only includes Hospital deaths
  • Deaths broken down by age group and ethnicity weekly

Intensive care admissions by ethnicity

icnarc

Excess ICU admissions by ethnicity (%)

icnarc_diff

Deaths by age

age_at_death

Population data from ONS census

pop_structure

  • 26% of the White population is over 60 years
  • 8% of the non-White population is over 60 years

Age adjusted deaths

obs_exp

Excess deaths (%)

excess

Excess deaths (%) detailed

excess_details

Expected versus observed

table

Possible explanations

How do we get more COVID-19 deaths

  • Number of deaths is a function of:
    1. number of cases
    2. severity of cases
    3. treatment of cases
  • How and why might each of these differ in BAME communities?
  • What can we do to mitigate this?

Number of cases

  • Exposure through employment e.g. NHS, precarious or zero hours contracts, front line jobs
  • Exposure due to living conditions e.g. multi-generation household, higher housing density, more urban
  • Exposure due to lack of effective design and communication of public health advice e.g. culture appropriate messaging
  • Exposure due to inadequate testing and tracing
  • All these factors would mean it is harder for BAME people to successfully lockdown so a higher BAME “R”
  • More cases even if severity and treatment are the same will result in more deaths

Severity of cases

  • Increased severity due to co-morbidities e.g. CVD, diabetes
  • Increased severity due overcrowded housing resulting in higher viral load
  • Increased severity associated with poor air quality and other environmental factors
  • More severe cases even if equal numbers of cases and equal treatment will result in more deaths

Treatment of cases

  • Concious or unconcious bias in primary care interactions (111 or GP) e.g. leading to sub-optimal advice
  • Worse treatment by ambulance service e.g. leading to later hospitalisation
  • Worse treatment by hospitals leading to poorer treatment choices (DNR)
  • Poorer outcomes from interaction with health care system even if numbers and severity of cases are the same would result in more deaths

Evidence of socio-economic vulnerability in BAME communities

Index of multiple deprivation & BAME

imd

Index of multiple deprivation (cont.)

imd

  • All dimensions of deprivation worse in more BAME areas
  • Particularly important are income, employment, health, housing, environment

Index of multiple deprivation

imd

Index of multiple deprivation (indicators)

imd

  • Key indicators that may be directly relevant to COVID-19
  • Over-crowded households have fewer bedrooms than needed to avoid undesirable sharing

Over-crowded households

over

Over-crowded households by age

over_age

Workforce in precarious jobs (%)

precarious

Transport, drivers and operatives (%)

drivers

Deaths by occupation

occupation

Studies controlling for health and socio-economic vulnerability

ONS individual level analysis

  • Linked census (2011) records with mortality (2nd March 2020 to 10th April 2020)
  • Logistic model of COVID-19 deaths vs random weighted 1% population sample
  • Separate model run for males and females
  • Control for:
    1. Five-year age bands
    2. IMD decile
    3. Household composition (living alone, family with no children, family with children, other) and country of birth (UK born, non-UK born)
    4. Socio-economic status: highest qualification held, NS-SEC of household person of reference, household tenure
    5. Health: self-reported health and having a limiting health problem or disability.

ONS individual level analysis results (age)

logistic Source: Coronavirus (COVID-19) related deaths by ethnic group, England and Wales: 2 March 2020 to 10 April 2020 (ONS)

  • Black people 4 times as likely to die of COVID-19
  • Bangladeshi/Pakistani 3.5 times as likely to die of COVID-19

ONS individual level analysis results (full model)

logistic Source: Coronavirus (COVID-19) related deaths by ethnic group, England and Wales: 2 March 2020 to 10 April 2020 (ONS)

  • Black people 1.9 times as likely to die of COVID-19
  • Bangladeshi/Pakistani men 1.8 and women 1.6 times as likely to die of COVID-19
  • Beware of table 2 fallacy

ONS individual level odds ratios

odds Source: Coronavirus-related deaths by ethnic group, England and Wales methodology

  • Shows how odds of dying of COVID-19 change as you control for more factors
  • After controlling for age and geography other factors make little difference

Patient level data

  • OpenSAFELY study using patient level GP data for 17 million patients (7th May 2020)
  • Controls for host of socio-economic factors, health risk factors and comorbidites
  • Ethnicity has independent effect of similar magnitude to ONS data after controls
  • Black people 1.7 and Asian people 1.6 times as likely to die as Whites

Discussion of studies

  • Should we be controlling for indirect pathways e.g. socio-economic factors or is total effect more relevant?
  • The ethnicity effect does not appear to be mediated through comorbidities or obvious socio-economic factors
  • What is causing residual BAME effect?
  • increased exposure on ethnic grounds e.g. more frontline roles, cultural differences
  • increased health risk on ethnic grounds e.g. higher allostatic load, vitamin D
  • poorer interaction with the health service e.g. discrimination in 111/GP/Ambulance/Hospital, reluctance to use healthcare
  • These are all above and beyond health and socio-economic factors controlled for in the study

Challenges

Data challenges

  • Death certificates do not record ethnicity
  • Ethnicity not always recorded on health records
  • Hard to find ethnicity data in other key administrative datasets
  • Bad data science already used in many areas of policy to step in and fill data gaps by extrapolating societal prejudices implicitly encoded into administrative data

Structural vs individual factors

  • more frontline roles and poor public health advice versus cultural differences result in poor isolation
  • higher allostatic load versus vitamin D deficiency
  • discrimination in 111/GP/Ambulance/Hospital verus reluctance to use healthcare
  • Probably a mixture of causes any role found for individual factors should not conveniently obscure structural factors
  • How do we unpick the causal effects of the various factors using the observational data available

Structural issues

belly Source: Guardian, 12th May 2020

  • Belly Mujinga was a railway ticket office worker at London Victoria station
  • Her employers knew she had respiratory problems but still insisted she work on the concourse without PPE and interact with passengers
  • Man claiming to have coronavirus spat at her on the 22nd March 2020
  • She died from COVID-19 on the 5th April 2020

Individual narrative

civitas Source: Telegraph, 5th May 2020

  • structural issues seen as a “victimhood agenda”

“As it happens, prior medical complications are not found in equal proportions in all ethnic groups. 73% of black adults in England are overweight or obese … people classified as black or south Asian have much higher rates of diabetes than the population as a whole. And ‘older Pakistani men’ have ‘particularly high’ levels of cardiovascular disease. These differences have no connection to discrimination.“

“Then there are cultural differences. South Asians are more likely to live in large households comprising three generations, thereby bringing vulnerable older people into more frequent contact with younger people who may carry the disease without realising. … These are lifestyle choices unrelated to discrimination.”

"Political" science

bmj Theodoratou, Evropi, et al. Bmj 348 (2014): g2035.

  • Poorly controlled observational studies and badly designed trials
  • No convincing evidence of vitamin D effect

beta blocker Terbeck, Sylvia, et al. Psychopharmacology 222.3 (2012): 419-424.

  • Double blind placebo controlled RCT the gold standard of clinical study design
  • Propranolol abolished implicit racial bias

Government response

  • NHS England released advice to all hospitals to consider BAME staff vulnerable
  • Public Health England has launched an inquiry advised by Trevor Phillips
    1. Former chairman of Equalities and Human Rights Commission who has a history of leading inquiries that identify 'cultural deficit' and overlook structural issues
    2. Is currently suspended from the Labour party over allegations of Islamophobia
    3. Works for a private sector firm that produces software for the police: “data-driven identification technologies and databases [that] can replicate racialised stereotypes and reinforce institutionalised prejudices” (Guardian, 6 May 2020)
  • It will be unsurprising to see this inquiry reporting back citing the individual factors
  • The enquiry is unlikely to win the confidence of people from BAME communities

PHE Terms of Reference

ToR

ToR “The review will not ascertain root causes of findings that are likely to be driven by complex interactions”

Source: Public Health England, May 2020

Research questions

  • What interventions exists to help mitigate these impacts in the immediate term?
  • Is it possible to unpick the causal effects of the various individual and structural factors using the observational data available?
  • What does the response to the disease both in clinical terms and in economic terms tell us about the implicit valuation of different lives - i.e. what do the policy choices made so far reveal about the societal exchange rate between Black lives and White lives - how does this compare across countries and correlate with other social values?
  • Do people voluntarily take on uncompensated risk or must power dynamics be included in economic studues: implications for VSL, WTP etc.
  • Should we be conducting less simplistic epidemiological and economic analysis in order to understand the impact of policies on health inequalities rather than roll out policies that exacerbate these inequalities?

References & data sources

References

Data Sources