Shingles Vaccine Non-vaccination Remains Common in U.S. Adults

SHINGLES vaccine non-vaccination fell among older U.S. adults, but prediction models lost calibration as uptake changed.

Shingles Vaccine Non-vaccination Remains Common

The proportion of U.S. adults aged 50 years and older who had not received a shingles vaccine declined from 73.7% in 2019 to 56.1% in 2024, according to an analysis of National Health Interview Survey data.

Although this trend indicates improving shingles vaccine uptake, more than half of eligible adults remained unvaccinated in 2024. Non-vaccination was highest among adults aged 50–59 years, at 71.4%, compared with 59.6% among those aged 60–64 years and 44.6% among adults aged 65 years and older.

Researchers analyzed data from 96,663 adults aged 50 years and older. Models were developed using 2019–2023 data from 79,245 participants and tested against a separate 2024 sample of 17,418 adults.

Machine Learning Provides Limited Improvement

The study compared survey-weighted logistic regression with XGBoost, a machine-learning approach capable of identifying nonlinear relationships and interactions. Both models used the same 15 predictors to estimate the probability that an adult would remain unvaccinated.

In the 2024 test sample, logistic regression achieved an area under the receiver operating characteristic curve of 0.764, while XGBoost achieved 0.771. Although the difference was statistically significant, the absolute improvement was only 0.0065. Brier scores were also similar, at 0.210 and 0.207, respectively.

Influenza vaccination, age, education, and race and ethnicity were the four most influential predictors in both models. These shared predictors suggest that a more complex algorithm added relatively little predictive value over conventional regression.

Changing Uptake Affects Model Reliability

Both models systematically overestimated the probability of shingles vaccine non-vaccination in 2024, even though their calibration slopes remained close to 1. This pattern reflected temporal calibration drift as vaccination prevalence changed.

Restricting model development to more recent 2022–2023 data produced similar discrimination but substantially reduced overprediction. The findings indicate that models intended to guide population-level outreach may require regular updating as vaccine uptake, access, and healthcare use evolve.

For clinicians and public health teams, prediction tools could help identify groups more likely to remain unvaccinated. However, the results caution against deploying older models without temporal and external validation. Recent data may be particularly important when models inform vaccine education, preventive care discussions, or targeted outreach.

Reference
Costa NS et al. Temporal validation of machine-learning models for shingles vaccine non-vaccination among U.S. adults aged 50 years and older. Vaccine. 2026;92:129099.

Featured Image: bernardbodo on Adobe Stock.