Answering FAQs: Part 2
What about my personalised risk? How many would die anyway?
To help the Royal Statistical Society, I wrote answers to frequently asked questions about COVID-19.
The main constraint was a limit of 400 words. These are the original version of the articles. The posts were later improved by other authors.
How can I find my personalised risk?
In this pandemic, there are several risks to consider, including:
- The risk of getting a SARS-CoV-2 infection;
- The risk of infecting others;
- The risk of dying due to COVID-19.
There are many factors which could affect your personal risk of infection and fatality. Analysis by the Office for National Statistics suggests age and sex matter for mortality due to COVID-19. Older people are more likely to die with the disease. In each age group, estimated mortality rates are higher for men than women. That analysis covers England and Wales, using death certificates registered up to 4th July 2020.
The University of Manchester produced the ‘Your COVID-19 Risk’ tool. The tool takes inputs of country, age, gender, working arrangements, and self-reported behaviours. The risk model is based on expert judgment and published research. A major limitation is this tool does not account for health conditions.
The output represents categories of estimated probabilities: getting an infection and infecting others. The tool recommends following guidance: keep your distance, wash your hands, quarantine if infected.
The University of Exeter produced a tool for calculating an individual risk score. Their analysis studied associations with demographic and health characteristics. The measures were hospitalisation with COVID-19 and dying due to the disease.
The British Medical Association adopted this tool for healthcare workers. This simplified score uses age, sex, ethnicity, and health conditions.
There are other tools, serving different purposes.
Calculators could miss major factors to your personal risk. The calculations often rely on statistical associations. There may be confounding factors. Such tools provide simplified estimates: your personal risk could be somewhat higher or lower.
The results do not override government guidance or advice from doctors. Scientific understanding of personal risks develops with further research.
As this is a pandemic: this is not only about personal risk, but the risk of spreading infections to others.
How many Covid deaths would have happened this year anyway?
There are still people dying due to COVID-19. This figure is not easy to estimate.
When a heat wave or cold spell hits, there can be a temporary increase in the number of deaths. Given deaths were higher than expectations, researchers term these fatalities excess deaths. Mortality may then fall beneath expectations during later weeks (deficit deaths).
Researchers suggest: as some people are frail, heat waves bring forward their deaths. Other names for this pattern are mortality displacement or harvesting. This reasoning applies to other phenomena, like pandemics.
In the limited published research, evidence for mortality displacement is inconsistent. There are disagreements over how large these displacements are. Some studies report little evidence of harvesting effects after heat waves. There are differences between places: in heat waves and numbers of frail people.
The counterfactual would be: what if the event had not happened? A counterfactual time series represents estimations of how many deaths would have occurred. These deaths can be from all causes, or for mortality in age groups or specific types of deaths.
Suppose researchers are studying a heat wave. A model might express deaths on a given day as a function of temperature on that day and past days. Such models are sensitive to how many past days researchers include. Different models produce different estimates.
We can estimate initial excess deaths and later deficit deaths. The mortality displacement ratio is the deficit estimate divided by the excess estimate. The major problem in this approach is the natural variation in when people die.
There are other methods, such as reviewing a sample of linked patient records.
Uncertainty matters. We should focus less on a single number, and more on the range of plausible values.