Read on Twitter
"Case fatality risk" has various definitions, and it is important to know what it's referring to. This has resulted in substantial confusion when trying to compare influenza and COVID-19. It's complicated. A thread.
There are different ways to estimate the numerator (number of deaths) and different denominators (infection). This has long been debated for influenza, but equally applies to COVID-19.
The obvious way to estimate the number of deaths is to look at reported deaths after a diagnosis of influenza. But this would be a minimum estimate - people may die without having been tested, or the test may be falsely negative.
There are a few sources of mortality data in Australia - notifiable diseases, hospitalisation data (for in-hospital deaths) and the death register.
Notifiable diseases data aren't always a good source of data on mortality - when the disease is reported, the death usually hasn't occurred and this may not be updated if the patient dies. This is less of an issue with COVID-19 as there is active follow up of cases.
Before PCR was widely available, hospitalisation data used to include many patients with "clinically diagnosed influenza", but this is now less common https://www1.health.gov.au/internet/main/publishing.nsf/Content/cda-cdi4004f.htm
The death register is probably as good a data source as is possible, but still may not capture all deaths from influenza because of incomplete testing, or difficulties in attribution (eg secondary bacterial pneumonia or heart attack triggered by influenza).
For COVID, another issue is the delay between infection and death - this delay is usually about 2-3 weeks but can be considerably longer.
While reporting is occurring relatively quickly in notifiable diseases surveillance, the hospitalisations database and death register do several data checks and coding that can take a while before they are available.
Delays in death occurring and being reported are one of many reasons why early estimates of case fatality can't be easily made by simply dividing reported deaths over reported cases https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4504518/
Another way to estimate deaths from influenza is to look at "excess mortality" - how many more deaths occur in the influenza season than you would expect?
This is complicated by the fact that baseline mortality from other causes seems to vary over the year with a peak in winter. This is well shown in fig 8 in the NSW flu reports: https://www.health.nsw.gov.au/Infectious/Influenza/Publications/2020/january-influenza-report.pdf
The original Serfling method simply calculated this baseline using a cyclical function, then looked at how many more deaths occurred in the influenza months. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1915276/
More modern approaches use a regression analysis with some indicator for influenza and other viral activity. https://www.ncbi.nlm.nih.gov/pubmed/19374272 .
But this could be an over-estimate - other viruses that circulate in winter could be responsible for some of these deaths, or perhaps there might be some other confounding factor.
These different methods (reported deaths vs excess mortality) usually produce very different results - often a 5-10 fold difference.
This is becoming important for COVID - for example, UK data suggest that the number of excess deaths is about twice the reported deaths. https://twitter.com/nicktolhurst/status/1252813643927842817?s=20
As compared to influenza, it is unlikely that this sudden spike in deaths in the UK is due to another circulating virus or something else that happens at this time of year.
Deaths that may not show up in official COVID statistics could reflect out of hospital deaths; not testing (sudden deaths or palliative care); slow reporting; or complications not attributed to COVID (eg myocarditis or pulmonary emboli).
It could also include deaths caused by disruptions to the health system - eg patients with stroke or heart attack that don't present promptly, or who might be avoiding hospital altogether. https://thenewdaily.com.au/news/2020/04/17/coronavirus-emergency-departments/
There are some indications that presentations with trauma may be decreased, but other reports suggest more alcohol-related trauma
https://www.surgeons.org/news/media-releases/trauma-surgeons-fear-increased-caseload-as-a-result-of-covid-19
https://www.surgeons.org/news/media-releases/trauma-surgeons-fear-increased-caseload-as-a-result-of-covid-19
The denominator issue for has been well covered - testing criteria and access to testing, ascertainment of mild cases and asymptomatic infections are all important issues. Age-specific mortality is another issue - most cases in Australia to date have been relatively young.
For influenza, the denominator is particularly difficult, as probably less than 1 in 10 cases get a diagnosis - most people with influenza just stay home and don't go the doctor, and even if they do they don't always get tested. https://wwwnc.cdc.gov/eid/article/19/11/12-1878_article
Serological studies suggest that about 20% of the population get influenza, and even PCR testing doesn't pick these all up - in a UK study, only 25% of infections detected by serology had PCR confirmed infection. https://www.thelancet.com/journals/lanres/article/PIIS2213-2600(14)70034-7/fulltext
For COVID-19, access to testing is probably less important in Australia as in other countries. We think that PCRs on swabs are about 80% sensitive. However, we do need serological studies to confirm this and to explore the possibility of asymptomatic infection
