I conducted a matched test-negative (case-control) approach, where cases consist of individuals who have tested positive for SARS-CoV-2 (for assessing protection against infection) or those who have experienced severe/fatal outcomes following the infection (for assessing protection against disease severity). The control group consists of individuals who have tested negative for SARS-CoV-2.
The matching was done in 1:5 using calipmatch.
We are examining various exposures, specifically different immune histories, such as individuals who have received three vaccine doses in addition to a natural infection, compared to those who are unvaccinated.
We have encountered situations where there are no cases with a specific immune history (e.g., zero vaccinated individuals who developed severe outcomes among cases). As a result, the point estimate from the conditional logistic regression using clogit is 0, and no confidence interval is generated. I tried applying exact logistic regression but it was not feasible as I have a sample of almost 30,000 matched pairs.
I also tried doing bootstrapping and applying clogit but this also did not get me a 95% CI.
I also tried applying the Firth method using the "firthlogit" command in Stata, which does not account for the matching. However, the point estimates we obtained are substantially below 100% effectiveness when there are zero severe outcomes among vaccinated individuals and the CIs between -100% and +100%. I tried applying Firthlogit with Flac correction in R which is supposedly an improvement on the Firthlogit estimates but the method is computationally intensive and R crashes before doing the estimation.
I would greatly appreciate ahving an insight into this matter.
Many many thanks
The matching was done in 1:5 using calipmatch.
We are examining various exposures, specifically different immune histories, such as individuals who have received three vaccine doses in addition to a natural infection, compared to those who are unvaccinated.
We have encountered situations where there are no cases with a specific immune history (e.g., zero vaccinated individuals who developed severe outcomes among cases). As a result, the point estimate from the conditional logistic regression using clogit is 0, and no confidence interval is generated. I tried applying exact logistic regression but it was not feasible as I have a sample of almost 30,000 matched pairs.
I also tried doing bootstrapping and applying clogit but this also did not get me a 95% CI.
I also tried applying the Firth method using the "firthlogit" command in Stata, which does not account for the matching. However, the point estimates we obtained are substantially below 100% effectiveness when there are zero severe outcomes among vaccinated individuals and the CIs between -100% and +100%. I tried applying Firthlogit with Flac correction in R which is supposedly an improvement on the Firthlogit estimates but the method is computationally intensive and R crashes before doing the estimation.
I would greatly appreciate ahving an insight into this matter.
Many many thanks
