Does getting more doses of COVID-19 vaccinations correlate to more hospitalisations in NSW, Australia?

Question

Twitter user @LCHF_Matt has created a NSW Surveillance Report which is a dashboard that compare the rate of "events" (which can be configured to include COVID-19 deaths, COVID-19 ICU admissions, COVID-19 non-ICU hospital admissions or reported COVID-19 infections) per capita from New South Wales, Australia, against their vaccination status. He says it is based on data published by NSW Health.

The graph below (which includes hospital admissions + deaths) appears to show that people with four or more doses are at most risk, and unvaccinated people at the least risk.

The Notes tab includes this summary:

It is well known that younger people are less likely to need hospitalisation as a result of Covid-19 than older people and that vaccination rates are lower in the very young. The hospitalisation and death observations in the unvaccinated cohort is no doubt skewed by this fact. These data still however indicate that the cohort of people currently unvaccinated are, collectively at least, not being as adversely affected by Covid-19 resulting in the need for hospitalisation as are the vaccinated cohorts. This could be the result of lower infection rates, lower hospitalisation rates when infected, both, and/or other mechanisms unknown […]

Note that apparently NSW Health doesn't publish the data needed to normalize for age. The author acknowledges that for this reason, the chart cannot demonstrate causation.

The graph has been referenced by Joel Smalley's blog, on Del Bigtree's show The Highwire, and endorsed by the Director General of the Israel Institute for Biological Research (IIBR) Shmuel Shapira.

Is this chart accurate? Does it misinterpret the NSW Health data? Is there a correlation between COVID-19 vaccine doses and hospitalizations in NSW?

Answer 1

Yes on correlation, but an emphatic no on causation.

Firstly, the claim being made is in fact a highly common data misinterpretation from anti-vaccination groups, and has previously been widely debunked.

To explain further, assuming the data being presented is accurate, it is impossible to draw a causative conclusion from this. This paper explains that positive test rates strongly correlates with age, whereas the data presented in the graph has not been through this basic correction. This is problematic, as the NY Department of Health explains:

Age confounding occurs when the two populations being compared have different age distributions and the risk of the disease or outcome varies across the age groups.

In fact, this is precisely what is happening. The NSW vaccination eligibility website states that:

From July 2022, you are recommended to have an additional winter COVID-19 vaccine (second booster / fourth dose) if you are: aged 50 years or over, a resident of an aged care or disability care facility, aged 16 years and over and severely immunocompromised (this will be a fifth dose), aged 16 years and over years and have complex, chronic or severe conditions

Additionally, the same page also states that:

People aged 12 to 15 years with complex health needs are also recommended to get a booster vaccination. This includes those who:are severely immunocompromised, have a disability with significant or complex health needs, have complex and/or multiple health conditions that increase the risk of severe COVID-19

Due to the government's vaccination efforts, only people who are highly susceptible to the virus are encouraged to obtain the fourth vaccination, as well as people aged 12-15 who are encouraged to obtain the third vaccination, where they would not be recommended otherwise.

Due to the large difference in populations between these two groups, unless all reasonable confounding factors (including but not limited to: age, travel habits, location, etc.) are accounted for, this data is not useful in determining the effectiveness of vaccination.

In fact, age based data is clearly available in the first link provided in the source, making it reasonable to conclude that the authors of the page are either incompetent in basic statistical analysis, or maliciously interpreting the data to support their beliefs.

Answer 2

Without getting into age stratification (which I happen to think any such analysis should do) etc:

To me the bar charts are all over the place, so I don't think the dashboard shows a correlation between doses and hospitalisations or deaths.

The impression given by the chart changes week to week. For "Hospital (not ICU)", in some weeks two doses appear better than one dose; for "Deaths", in some weeks one and two doses seem far better than no doses. What is the reasonable conclusion to draw if, solely going by the chart, in eight out of nine weeks there is a higher rate of death for "no doses" than "two doses?

Indeed the Doses Summary table bears this out: "two doses" are half as likely to die as "no doses".

Is the risk U-shaped?

Also the vaccination status "Unknown" comprises 25% of the "Hospital (not ICU)" and "Hospital in ICU" events respectively, which seem large proportions to disregard.

---

You can drill down into age groups in these charts purporting to represent Covid-19 mortality by vaccination status

https://ourworldindata.org/covid-deaths-by-vaccination#data-on-covid-19-mortality-by-vaccination-status