COIOS
Emergent technology Item confirms preprint

Gradient boosting did no better than logistic regression on HPV vaccine hesitancy

In 1,156 Cameroonian parents of children aged 9 to 18, XGBoost, random forest and logistic regression predicted HPV vaccine hesitancy almost identically (AUC 0.841-0.870, XGBoost 0.870 against 0.869, p=0.965), with logistic regression the best calibrated.

A secondary analysis of a cross-sectional survey of 1,156 parents in the Buea Health District, Cameroon, developed and internally validated three models of HPV vaccine hesitancy, in a preprint. Discrimination was comparable across logistic regression, random forest and XGBoost (AUC 0.841-0.870), and logistic regression had the highest sensitivity (73.1%) and lowest Brier score (0.1435). Most predictive weight fell on insufficient vaccine information, perceived unsafety and distrust in the Ministry of Health rather than on sociodemographic variables.

Why it is interesting: Another case of algorithmic complexity buying nothing over logistic regression on tabular survey data, here in a low-resource vaccination setting.

Source
medRxiv, 20 September 2026
DOI
10.64898/2026.09.18.26363393
Type
Preprint
Design
Secondary analysis of a cross-sectional survey of 1,156 parents in Buea Health District, Cameroon; 80/20 split, internal validation only
Verdict
Confirms prior evidence
Driver
AI in medicine
Driver
Infectious disease and pandemics