Quantifying the Socioeconomic Burden of Epilepsy: A Partial Identification Analysis of US National Survey Data

Socioeconomic inequalities in health outcomes remain a central concern in public health research because chronic conditions frequently interact with income disparities to produce persistent patterns of social disadvantage. Epilepsy, one of the most prevalent neurological disorders globally, illustrates this challenge clearly, as individuals living with the condition often experience reduced labour market participation, income instability, and long-term economic vulnerability. A key measurement challenge arises when researchers have access only to aggregate prevalence statistics without the joint distribution of health status and income rank, the concentration index is not point-identified, and a wide range of index values including values of opposite sign) may be consistent with the observed data. This paper develops a partial identification framework that makes this uncertainty explicit and derives sharp bounds on the concentration index from a single moment condition on aggregate prevalence. The model represents the health burden as a feasible distribution across the income rank of the population and formulates a linear optimisation problem whose extreme points characterise the tightest bounds consistent with observed prevalence levels. A central analytical finding is that the identification interval always contains zero when prevalence is below one, implying that the direction of socioeconomic inequality cannot be established from marginal prevalence data alone. Analytical derivations are complemented by computational simulations applied to anationally representative survey of adults with epilepsy, which illustrate the width of the identification interval and the conditions under which calibrated scenario estimates suggest progressive concentration of work disability. The framework equips researchers working with aggregated epidemiological data to assess which concentration index values are compatible with their observations, to communicate identification uncertainty transparently, and to specify the individual-level data required to achieve point identification.