Real-World Effectiveness of COVID-19 Vaccines in the Netherlands: Causal Inference From A Retrospective Cohort Study Using the Target Trial Emulation framework Within the PharmLines Initiative

Study type

• Study topic:
  Human medicinal product
• Study type:
  Non-interventional study
• Scope of the study:
  Effectiveness study (incl. comparative)

  Method development or testing
• Data collection methods:
  Secondary use of data

Non-interventional study

• Non-interventional study design:
  Cohort

Study drug and medical condition

• Medicinal product name:
  COMIRNATY

  VAXZEVRIA

  COVID-19 VACCINE (INACTIVATED, ADJUVANTED)
• Medical condition to be studied:
  COVID-19

  SARS-CoV-2 test positive

Population studied

• Short description of the study population:
  Dutch citizens of Northern regions of the Netherlands
• Age groups:
  • In utero
  • Paediatric Population (< 18 years)

    • Neonate

      • Preterm newborn infants (0 – 27 days)
      • Term newborn infants (0 – 27 days)
    • Infants and toddlers (28 days – 23 months)
    • Children (2 to < 12 years)
    • Adolescents (12 to < 18 years)
  • Adult and elderly population (≥18 years)

    • Adults (18 to < 65 years)

      • Adults (18 to < 46 years)
      • Adults (46 to < 65 years)
    • Elderly (≥ 65 years)

      • Adults (65 to < 75 years)
      • Adults (75 to < 85 years)
      • Adults (85 years and over)

Study design details

• Study design:
  Retrospective cohort study
• Main study objective:
  Comparing the estimates of this retrospective cohort study using a sequential Target Trial Emulation approach with the estimates of a Test-Negative Case-Control study (yet to be published)
• Setting:
  Population-based, respondents to the Lifelines questionnaires, registered in IADB.nl (Pharmlines initiative).
• Comparators:
  Unvaccinated participants
• Outcomes:
  SARS-CoV-2 infection; severe COVID-19 disease; hospitalization; ICU-admission; all-cause mortality
• Data analysis plan:
  We apply Inverse Probability of Censor (IPC) weights to adjust for the introduced informative censoring of getting vaccinated after time0 aka those who deviate from the assigned treatment strategy – by weighting the data. IPC upweights those that remain uncensored to get a population that better represent what would have happened without informative censoring.; We apply Inverse Probability of Treatment (IPT) weights to balance the four exposure groups on their observed covariates at baseline.; For studying the time-variation of vaccine protection generalized additive models (GAMs) with smoothing splines will be fitted to the data using the mgcv package in R. The parametric part of the model included linear effects from potential confounders while time elapsed since the index date will be modelled non-parametrically using penalized thin plate regression splines. ; As we would like to compare the results of the retrospective cohort study to the results of the test-negative case control study, which used a conditional logistic regression model yielding odds ratios across post-vaccination time-intervals (14-28 days since vaccination, 29-82 days, 83-181 days, 181-364 days and 365+ days), our primary analysis will make use of a pooled logistic regression model using the same discrete time intervals, since these yield odds ratios. These can then be converted to vaccine effectiveness percentages.; To account for missing covariate data, the models described above will be fitted separately in each of the multiply imputed datasets. ; For the GAM, pooling spline coefficients for penalized smooths can be affected by the smoothing parameter estimation. To avoid this, we will generate model-based predictions and corresponding standard errors for each imputed dataset. At each time elapsed since the index date, the predicted linear-predictor contrasts and their within-imputation variances will be combined across imputations using Rubin’s rules. The pooled effects are then exponentiated.