Abstract

Context: Low health literacy (HL) is a public health issue, with impacts on population health and illness, however there are few tools for collecting health literacy data in large populations.

Objective: To develop a method of deriving indicative functional HL levels from routinely collected socio-demographic data.

Method: We investigated which socio-demographic variables would best depict whether an individual is above or below a constructed HL competency threshold. Weighted logistic regression was used to estimate Odd Ratios for being below the threshold. Weighted Receiver Operating Characteristic (ROC) analysis examined which variables best predicted low HL. Specificity, sensitivity and area under (AU) the ROC were descriptors for ability to predict risk.

Results: Three models were developed; one using all nine variables; a pragmatic model using the four most predictive variables (Qualification (whether the individual had achieved the level expected by age 16 years), Ethnicity, Home ownership, and Area Deprivation); and one using only “Qualification” (the single most predictive variable). All models showed good prediction of low HL (AUROC 0.73 (95% CI 0.71; 0.74) to 0.78 (95% CI 0.76; 0.79)), with predictive power increasing with more complex models.

Conclusion: The most important predictor of low HL is achievement of the qualification level expected by age 16 years, with additional variables adding more predictive power. The developed formulae can be used to estimate functional HL levels in populations from routinely collected socio-demographic data, and hence facilitate effective development and targeting of public health communications. The method to derive the formulae will be applicable in other industrialized countries.