Rotavac Impact: India’s Universal Immunization Program (2016-2020)

Okay,let’s break down the statistical methods used in this study,focusing on how they addressed ‍potential biases and validated their⁢ findings.

1. Main Analysis: Logistic Regression

*⁣ ⁣ ⁣ Goal: To estimate the adjusted vaccine effectiveness (aVE) against ‌hospital admission for acute rotavirus ⁢diarrhea.
* ‌ Method: Logistic regression was used. This is appropriate because the ​outcome (hospital admission) is a binary variable ‍(yes/no).
* ⁣ Covariate Selection: ⁢They carefully selected ​covariates to include in the model. Covariates were ⁤only included if they changed the odds ratio (OR) associated ⁣with vaccination by more than 5%.This helps to avoid overfitting the model and focuses on variables that truly influence the relationship between vaccination‌ and outcome.
* excluded Variables: Sex, birth month, birth⁣ year, year of⁣ admission, hospital, and household characteristics were excluded after consideration. This suggests these variables didn’t substantially alter the OR for vaccination.
* Stratified Analyses: ​ They performed analyses stratified by age, nutritional status, state⁣ of residence, severity of diarrhea, vaccine dose, and circulating genotypes. This allows them to see if the vaccine effectiveness varies ⁤across⁣ these subgroups.
* Interaction Term (Age &⁣ Vaccination): They initially tested for an ‍interaction between vaccination status ​and age. ​An interaction would meen the effect of vaccination differs depending on the child’s age. However, the interaction term was not ​ statistically significant and was removed from the final model.
* ‍ model Fit: A likelihood ratio test ​was used⁢ to compare the model with the interaction⁢ term to ‌a model without it. The non-significant p-value from this ‌test justified removing the interaction term.

2. Sensitivity Analysis: Addressing Confounding

The researchers recognized that observational studies (like this one, where ‍they didn’t randomly assign vaccines) are susceptible to​ confounding. ⁣ Confounding occurs when a third variable is associated with both ​vaccination ‍status and the outcome (hospital admission), potentially distorting the observed relationship. To address this, they used a suite of sensitivity‌ analyses:

* Unmatched Logistic Regression: This is ‌the ⁤main analysis described ‌above, but it’s included here as a baseline for comparison.
* Matched​ Regression Analysis: Participants were matched based on​ relevant covariates before the ‌regression ​analysis. This attempts to create groups of vaccinated and unvaccinated children who are very similar on those⁢ covariates, ⁣reducing confounding.
* Propensity Score Matched Regression: this is a more sophisticated matching technique. Instead of directly matching on all covariates, it estimates⁣ a “propensity⁣ score” for each participant ‌- the probability of being vaccinated given their observed characteristics. Participants‌ are then ​matched‍ based on these propensity scores.
* Propensity Score Weighted Regression: ​ Instead of matching, this method weights each observation based on its⁢ propensity score. This creates a pseudo-population where the distribution of covariates is balanced between the vaccinated and unvaccinated groups.
* ⁣ E-Value Estimation: This is a key part of their sensitivity analysis. The E-value estimates the minimum strength of an unmeasured confounder that would be needed to entirely explain away the observed association between vaccination and hospital admission. A larger E-value suggests that the observed association is less likely to be‌ due to unmeasured confounding.

​ * Formula: ⁢ E = 1/OR + √(1/OR * (1/OR – 1))
* ⁤ Interpretation: The E-value represents how much stronger an unmeasured confounder would need to be (in terms of ⁢its relative risk) to change‍ the observed OR to 1 (no effect).

3. Comparing Results Across Methods

The researchers⁤ compared the vaccine effectiveness estimates obtained from all these different ⁢methods. ​If ⁣the estimates are consistent across methods, ⁤it strengthens their confidence that the ‌observed association is not due to confounding.

In essence, this study ⁣used a robust statistical ​approach to estimate vaccine effectiveness, carefully considering and addressing potential biases through a variety ⁤of sensitivity analyses. The⁢ use of the E-value is ‌notably noteworthy, as it provides a quantitative assessment of the potential impact of unmeasured confounding.