Developing Predictive Models of Health Literacy

Low health literacy (LHL) remains a formidable barrier to improving health care quality and outcomes. Given the lack of precision of single demographic characteristics to predict health literacy, and the administrative burden and inability of existing health literacy measures to estimate health literacy at a population level, LHL is largely unaddressed in public health and clinical practice. To help overcome these limitations, we developed two models to estimate health literacy. We analyzed data from the 2003 National Assessment of Adult Literacy (NAAL), using linear regression to predict mean health literacy scores and probit regression to predict the probability of an individual having 'above basic' proficiency. Predictors included gender, age, race/ethnicity, educational attainment, poverty status, marital status, language spoken in the home, metropolitan statistical area (MSA) and length of time in U.S. All variables except MSA were statistically significant, with lower educational attainment being the strongest predictor. Our linear regression model and the probit model accounted for about 30% and 21% of the variance in health literacy scores, respectively, nearly twice as much as the variance accounted for by either education or poverty alone. Multivariable models permit a more accurate estimation of health literacy than single predictors. Further, such models can be applied to readily available administrative or census data to produce estimates of average health literacy and identify communities that would benefit most from appropriate, targeted interventions in the clinical setting to address poor quality care and outcomes related to LHL.