COVID‐19 vaccination status and testing rates in Finland—A potential cause for bias in observational vaccine effectiveness analysis

Abstract COVID‐19 vaccination effectiveness has been monitored in observational studies (test‐negativity design or traditional cohort design), but these studies have not addressed the potential behavioral bias between vaccinated and unvaccinated individuals. We aimed to address this by comparing COVID‐19 testing rates between vaccination status and whether vaccination changes the testing rates. We found that three times vaccinated had least tests performed during the pandemic and unvaccinated had the highest testing rate. Each vaccination dose increased the testing rate. In conclusion the observational studies addressing vaccine effectiveness should also present testing rates between vaccinated and unvaccinated to address the potential behavioral bias.

1 | BACKGROUND COVID-19 vaccinations had all excellent efficacy against severe disease (i.e., hospitalization, intensive care unit [ICU] admission, or death) in the original randomized controlled trials run by the vaccine developers, and those studies also showed protection against the SARS-CoV-2 virus infection. [1][2][3] Alongside these trials, real life observational data have been used to monitor the vaccine efficacy continuously. [4][5][6] As the observational studies are prone to certain biases, testnegativity design has been used to monitor the vaccine efficacy against infection (especially popular previously in influenza vaccinations), in which the vaccine rates between those who tested positive are compared to those who tested negative. 7,8 Although this design acknowledges some of the biases in the more traditional vaccinated versus unvaccinated cohort/case-control designs, it is still vulnerable to behavioral bias. 7 In order to work properly, test-negativity design would need the vaccinated and unvaccinated samples to have similar healthcare seeking behavior. Thus, if either group does not seek for testing in symptoms, it may lead to neglected selection bias in the sample. 7,9 This bias is not seen when the outcome is objective, such as ICU admission or death, but it may cause bias, if only infection and detection rates are measured. 9 We aimed to assess the potential bias due to the healthcare seeking behavior related to COVID-19 vaccination by comparing testing rates between vaccinated and unvaccinated population.

| Materials
We conducted an observational retrospective analysis in Southern Savonia region of Finland. Our study period was from August 1, 2020, to January 31, 2022. During this period, all people in this area had the possibility to book COVID-19 polymerase chain reaction (PCR) testing free of charge either by phone call or via internet. The testing guidelines were practically unchanged during this period as the recommendation was to test everyone with the slightest symptom of COVID. Testing was recommended if any of the following symptoms were found: Fatigue, headache, fever, cold, cough, difficulty in breathing, vomiting, or diarrhea.
We included all citizens aged 18 or more in our region. We generated the study groups according to vaccination status on January

| DISCUSSION
We found that over half of the three times vaccinated had not had a single COVID-19 PCR test performed. This indicates that either the vaccination reduces the symptomatic SARS-CoV-2 infections, or that individuals rely on the vaccine so much that they do not see the testing necessary despite the recommendations. However, in the assessment of the testing rates in relation to vaccination, the testing rates increased after received vaccinations in each group. Still, it seems evident that those vaccinated three times had the lowest testing rate, which should be taken into account in the vaccine efficacy analyses.
Initially, the vaccination was hoped to reduce the possible viral spreading, but during the omicron period, it became clear fast that the vaccinated individual may spread the disease and get infected as well as those unvaccinated. 10 Three doses though protect still from serious events (ICU admission and death). 11 The largest test-negativity study from England addressed the vaccination status against symptomatic F I G U R E 2 Number of tests per individuals stratified by the vaccination status F I G U R E 3 COVID-19 testing incidence prior and after vaccination doses stratified by the final vaccination status. Fourth dose was left out of this analysis due to the low number of individuals omicron infection and showed low protection with two doses and better protection after booster shots, but it did not analyze the healthcare seeking behavior. 12 Most likely this will lead to bias in the results and overestimate the booster shot effectiveness, as those having booster could have the lowest testing rate based on our results and therefore the vaccine effectiveness results might be due to reduced testing in this group instead of true effect.
We have some limitations to our data. First, we do not have the numbers of home testes performed. These have been recommended in our area since September 2021, but all of the positive home test findings are guided to be confirmed in an official PCR test. A second limitation is the lack of testing data on private sector but based on the reports published in local newspaper the private testing capacity in our region has been less than 5% of the publicly funded. Third limitation is that we do not know the symptoms which lead to the test. It may be that those vaccinated have got tested in order to get their COVID-19 passport eligible for example to traveling (although this was not officially allowed indication for free testing).
The main strengths are that the testing has been free for citizens and the testing time can have been booked directly from internet. Furthermore, our region has the highest vaccination coverage in the country, and we have had the lowest COVID-19 incidence from August 2020 to the mid-January 2022, although our region has had one of the highest testing rates during this time period.

| CONCLUSION
When conducting a test-negativity designed study to analyze vaccine efficacy against infection based on observational data, the testing rates in the selected population should be presented, as there might be substantial behavioral differences in the groups, which may cause selection bias for the study cohorts and therefore bias the efficacy estimated. In our example the results in a classic test-negativity design would have overestimated the vaccine efficacy against COVID-19 infection due to the substantially lower testing rates among vaccinated individuals. The test-negativity design is most suitable when the outcome does not depend on the patient's behavior (e.g., ICU admission or death).

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1111/irv.12993.