Among 506 healthcare worker infections from a cohort of 12,248 at PGI Chandigarh, 64% were unvaccinated, 9.5% were fully vaccinated (>2 weeks past II dose)
However, the shorter follow up post 2nd dose (time bias) means the real % could be higher.
1/10
nejm.org/doi/pdf/10.105…
However, the shorter follow up post 2nd dose (time bias) means the real % could be higher.
1/10
nejm.org/doi/pdf/10.105…
Cont’d
The graph does not factor in the duration of follow up. During an ongoing vaccination process, those who are unvaccinated “get longer time” (compared to the vaccinated healthcare workers) to pick up the infection.
2/10
The graph does not factor in the duration of follow up. During an ongoing vaccination process, those who are unvaccinated “get longer time” (compared to the vaccinated healthcare workers) to pick up the infection.
2/10
Cont’d
Since vaccination takes time, those who were 2 weeks past second dose in the pie chart would have had fewer days of observation, than others.
This means that the odds of picking up infection will be smaller by default. This also gets reflected in the pie chart.
3/10
Since vaccination takes time, those who were 2 weeks past second dose in the pie chart would have had fewer days of observation, than others.
This means that the odds of picking up infection will be smaller by default. This also gets reflected in the pie chart.
3/10
Cont’d
In other words, the pie chart will give a true picture only if (theoretically) indefinite follow up occurred in the three subgroups (partially and fully vaccinated, unvaccinated) so they get equal chance to be exposed to infection on a timeline.
4/10
In other words, the pie chart will give a true picture only if (theoretically) indefinite follow up occurred in the three subgroups (partially and fully vaccinated, unvaccinated) so they get equal chance to be exposed to infection on a timeline.
4/10
Cont’d
Hence it is important to factor in this lead-time bias while reading such pie charts that are done during an ongoing vaccination process, regardless of the place it is done at.
This also applies to the early data that came out of Israel.
5/10
Hence it is important to factor in this lead-time bias while reading such pie charts that are done during an ongoing vaccination process, regardless of the place it is done at.
This also applies to the early data that came out of Israel.
5/10
Cont’d
To illustrate this bias with an example, imagine you are calculating the risk of a road accident based on 3 groups of drivers based on their age.
Your question is whether older age groups have a lower risk of accidents.
6/10
To illustrate this bias with an example, imagine you are calculating the risk of a road accident based on 3 groups of drivers based on their age.
Your question is whether older age groups have a lower risk of accidents.
6/10
Cont’d
To do the study properly, you must ask each group to drive for the same DURATION and DISTANCE in SIMILAR conditions.
But if you ask the older group to only drive 10 Km, middle-aged group 100 km and the youngest group 1000 km, you can’t compare the accident rates.
7/10
To do the study properly, you must ask each group to drive for the same DURATION and DISTANCE in SIMILAR conditions.
But if you ask the older group to only drive 10 Km, middle-aged group 100 km and the youngest group 1000 km, you can’t compare the accident rates.
7/10
Cont’d
In this hypothetical study, we will come up with an erroneous conclusion that older people are much less likely to have accidents. This result could occur simply because the older people did not drive long enough, and not because they were necessarily safer drivers.
8/10
In this hypothetical study, we will come up with an erroneous conclusion that older people are much less likely to have accidents. This result could occur simply because the older people did not drive long enough, and not because they were necessarily safer drivers.
8/10
Cont’d
A potential solution would be to calculate “man-days of exposure” and then redo the pie chart based on number of days of exposure, and the % infections acquired between the three subgroups.
9/10
A potential solution would be to calculate “man-days of exposure” and then redo the pie chart based on number of days of exposure, and the % infections acquired between the three subgroups.
9/10
Note: This thread was made only to highlight a common bias that occurs while reporting breakthrough infections, and it does not take anything away from PGI for publishing their data.
The study does indicate that unvaccinated people are more likely to pick up infection.
10/10
The study does indicate that unvaccinated people are more likely to pick up infection.
10/10
• • •
Missing some Tweet in this thread? You can try to
force a refresh
