Effectiveness of Covid-19 vaccination vgainst risk of intensive care hospitalization derived from NL NICE and CWZ data.
Source data: NICE and CWZ
Introduction
You may have seen the following article. We can now use this approach to derive the observed VE as it evolves in time from the real world NL hospital data (NICE and CWZ). The result is shown below for two sets of data.
NICE: This data is reporting IC (intensive care patients by vaccination status)
CWZ: This data is reporting Covid-19 patients in the need of oxygen supply Those are less acute cases than the NICE IC patients obviously. Therefore we expect the VE against this level of illness to be lower than for the IC.
As we can see, the VE against ICU is relatively stable. It has to be mentioned, that this may be confounded by the booster campaign.
Method
We solve the following equation for VE as function of:
Share vaccinated patients in ICU: π_ICU
Vaccination level: π_π£ππ₯
Resulting in
VE=1β(π_πΌπΆπβ(1βπ_π£ππ₯))/(π_π£ππ₯β(1βπ_πΌπΆπ))
We now apply this to extract thh VE to NL hospital field data and assuming a vaccination level of 90%-92% for NL population cohort above 60 years. This is the cohort that dominates in the intensive care admissions. We can neglect younger cohorts as those don't end up in ICU (vaccinated or not) in numbers relevant for this calculation. The method is shown below graphically.
Using the latest data from NL includes November and gives the following result.
With this we finally get the VE as function of time as shown in the introduction. We can now compare this with the published Sweden study βEffectiveness of Covid-19 Vaccination Against Risk of Symptomatic Infection, Hospitalization, and Death Up to 9 Months: A Swedish Total-Population Cohort Studyβ in the Lancet, 25 Oct 2021.
The results obtained on the NICE data is in line with the reported VE against ICU for the age group above 70 (VE_ICU 92%-95%).
The VE waning appears slower as compared with the Lancet paper study. This may be related by assuming the NL vaccination day zero too far in the past. In the graph above, we assumed the date for 90% double vaccinated as 30/03/2021. Below is the result if assuming 15/05/2021 instead.
Limitations of the simple model
After some valid remarks (here and here), the following caveats shall be pointed out. While I could make a proper paper, substack gives a more rapid turnaround and allows adding peer remarks quicker.
Over 80 year patients may not be admitted to IC. This will underestimate the share of vaccinated patients (as they donβt make it to ICU).
Died in care home; not transferred to hospital. This will underestimate the share of vaccinated patients (as they donβt make it to ICU).
The hospital admittance viewpoint is limited.
Naturally immune are confounded in the cohort.
Boosters campaign is confounded.
The assumed day zero for the vaccination level may be wrong. In reality, the vaccination campaign (February-May) was spread over several months for the elderly cohort.
Asymmetric C19 case counting (e.g. RKI counts a breakthrough case only if clinically diagnosed as a symptomatic case in addition to a positive PCR requirement), cohort confounding and test behaviour bias.
Undercounting of vaccinated patient cases in general.
Obviously, if the vaccination level is overestimated, the derived VE will be lower.
The vaccination level progress (dose 1) for the above 60 year cohort can be seen here.