Advancing National Health Through Investigations of Nutrition, Medical Expenditure, and Public Insurance Programs
- Author(s): Tavasoli Hozouri, Negar
- Advisor(s): Harding, Matthew
- et al.
Increased legislation and media coverage regarding health care expenditure has brought issues of health and medicine to the forefront of American dialogue. In my thesis, I explore three different topics which both indirectly and directly address issues of long-term health and health expenditure.
In the first chapter of my dissertation, The Impact of Nutritional Content and Claims on Household Dietary Expenditure, I examine the relationship between nutritional content for select food categories, related low-content labels on their packaging, and food expenditure for households with children to assess the salience of nutrition claims. Consumers balance many factors--from price to taste to health benefits--when shopping for food at the supermarket. Household heads must weigh short term gratification of satisfactory tastes or financial savings with the long-term health benefits of purchasing goods with claims that imply positive well-being. I utilize an aggregate hedonic regression model in which households must choose between types of goods within various nutritional categories based on their alimentary contents and the presence of related claims. The model is applied to a proprietary national panel scanner data set on nutritional purchases of households with children to assess the extent to which nutritional claims affect purchasing behavior. The results demonstrate that while the presence of nutritional claims are generally associated with significant increases in purchases for most product categories in question, these effects are minimal in comparison to those from the presence of actual nutritional contents, which also correlate to factors like taste and convenience.
In the second chapter, Forecasting Medical Expenditure in the Era of Health Care Reform through Applications & Innovations of Machine Learning Algorithms, I directly compare several machine learning predictive models to assess which selected tools can best predict the future of medical expenditure, using a nationally representative data set provided by the the Agency of Healthcare Research and Quality (AHRQ). The Medical Expenditure Panel Survey (MEPS) is a rich data source on American health care consumption over the last two decades. Its Household Component consists medical expenses, sources of payment, types of medical expenses and diagnoses, and demographic information on panel members. In 2002, the MEPS data set was supplemented with a largely linear projection for health care expenditure levels over the next fifteen years. Since 2002, advances in statistical computation and the theoretical understanding of panel data machine-learning techniques have greatly advanced, meaning that better projection techniques could be fruitful in examining the future of health care expenditure and proposing policy for such a future. Among the methods reviewed are feedforward and recurrent neural networks, Cluster Ridge and Lasso regression, and a recently developed innovative technique for panel data using penalized forecasting by Harding et al. (2015). I find that this new technique and Cluster Ridge regression outperform the original MEPS projection model as well as the other reviewed methods, which can be attributed to the high level of correlation among groups of variables, particularly in the presence of comorbidities. I use the best projection technique to forecast health care expenditure in the face of the Affordable Care Act for the next nine years after 2016 in order to make policy and savings recommendations for American households.
In the final chapter of my dissertation, The Doctor Will Finally See You Now: Evaluating the Effect of Medi-Cal Expansion on Healthcare Access in Non-Elderly California Adults, I quantify the effect of Medicaid expansion for previously uninsured and non-Medicaid eligible low-income adults on healthcare access. In California, health outcomes and disparities are highly linked to factors of income and health care affordability. The Patient Protection and Affordable Care Act (ACA) of 2010 aimed to reduce these disparities and to allow low-income individuals or those with pre-existing medical conditions to receive medical insurance and care. Existing research has estimated that 2.13 million Californians became eligible for full or low-cost insurance coverage through Medi-Cal, California's Medicaid health care program, after the implementation of the ACA. A reverse difference in differences approach is used on the California Health Interview Survey (CHIS) from 2013 and 2014 to evaluate the extent to which delays in health care due to financial concerns changed immediately after the expansion of Medi-Cal to individuals who were previously ineligible for the benefits, compared to Californians who were already covered by Medi-Cal prior to ACA expansion. The analysis suggests that the ACA was associated with an 9.27 to 9.87 percentage point reduction in the rate of delayed or foregone care for individuals newly eligible for Medi-Cal under the ACA. This is the first study to confirm the improvement of health-seeking behavior in terms of delays for this group of California residents