Background: Estimating the effects of vaccine schedules on important health outcomes remains a challenge due to missing or delayed vaccine administration. Modern causal inference methods that account for informative non-adherence may be useful in such settings, but each method requires different statistical assumptions which may impact performance. For instance, inverse probability weighting (IPW) requires properly specified weighting models. Augmented inverse probability weighting (AIPW) takes both outcome and weighting models as inputs and is consistent if at least one of the input models is correctly specified, a property known as double robustness.
Objectives: Using IPW and AIPW estimators, we compared the 2-year risk of acute gastroenteritis (AGE) - related emergency department (ED) visits between infants who received 0 versus 3 doses of the 3-dose RotaTeq® rotavirus vaccine (RV).
Methods: Using the IBM MarketScan® Commercial database (2007-2015), we identified 1433925 infants who received a diphtheria-tetanus-acellular pertussis (DTaP) vaccine between 38 and 92 days of age. Follow-up began on the date of the first DTaP dose and continued for a maximum of 720 days. The data were setup so that there were two copies of each infant record. In the first copy, representing no vaccine, infants were censored on the date of rotavirus vaccine receipt. In the second copy, representing full adherence to the 3-dose RV schedule, infants were censored on the date of nonadherence to the 3-dose RV schedule. Within each group, Cox proportional hazard models were fit to estimate risk of protocol nonadherence and AGE-related ED visits conditional on patient-, provider-, and insurance plan-level covariates. The fitted models were used to construct weights and outcome predictions for the IPW and AIPW estimators.
Results: Using IPW, the 2-year risks of AGE-related ED visits with no vaccine was 4.46%. With the fully adherent 3-dose RV schedule, the risk was 3.29%, for a risk difference of -1.17% (95% CI: -1.34%, -1.01%). Using AIPW, the risks were 5.01% and 3.51% for no vaccine and the fully adherent schedule, respectively, for a risk difference of -1.50% (95% CI: -1.67%, -1.34%).
Conclusions: Using both IPW and AIPW, we found that adherence to the 3-dose rotavirus vaccine schedule prevented AGE-related ED visits, though the AIPW estimate was larger in magnitude than the IPW estimate. Notably, the AIPW double robustness property came with no loss of precision. Given the improved robustness with no precision loss, AIPW may be preferred for estimating the effects of vaccine schedules, or more generally any treatments or protocols that require adherence over time.
