Home
Welcome to the Wiki of this Linear Regression Analysis!
Regression analysis is an important statistical method for the analysis of medical data. It enables the identification and characterization of relationships among multiple factors. It also enables the identification of prognostically relevant risk factors and the calculation of risk scores for individual prognostication.
Using data from the 1979 - 2014 NLSY Investigator, the effects of government-funded health insurance on the lives of Americans is analyzed. In this analysis, health insurance is controlled, leading to the observation that there is an increasing trend in BMI.
Looking at the effects Medicaid has had on United States health is especially important when it provides medical care to four out of ten American children, pays for the care of two thirds of the people in nursing homes, and provides for ten million children and adults with physical and mental disabilities, according to the New York Times. It accounts also for sixty percent of U.S. federal funding. While many assume that Medicaid is expensive in every state, Illinois spending per person was well below the U.S. average for all major eligibility groups and spending per enrollee of $4,682 according to Voices for Illinois Children. If block grants were passed to all the states, Illinois and many other states would have to reduce Medicaid spending, number of programs being covered, and enrollment. With low spending on Medicaid being so prevalent already, it is important to study the impact on low-income, disabled, and children in Illinois. Medicaid is a topic that is of recent interest to this new presidential administration. With the current president threatening to repeal President Obama’s Affordable Care Act (ACA), looking at the results that Medicaid has had on the United States health is important. As one party is discrediting its return on (taxpayer dollars) investment, it is critical to review how government funded health insurance have impacted the well-being of Americans.
In 2008, the Oregon Health Insurance Experiment revolutionized the scientific community's view of the effects of Medicaid because it was the first time Medicaid had been assessed through a random-lottery selection. Out of approximately 90,000 people, 25,169 individuals that included children, pregnant women, the disabled, and families, received one year of insurance. It not only gave pertinent data about Medicaid, but it also demonstrated what kind of results could be gathered from such a large scale experiment. From the experiment, Baicker and Finkelstein (2016) discovered that Medicaid improved the participant's financial security and decreased depression. However, there was no statistically significant evidence for "better physical health, employment rates, or earnings." Baicker and Finkelstein's (2016) results of health care utilization, labor market outcomes, income, and depression can be used for comparison in this study. While this experiment does not directly address BMI, the factors being studied all affect a person's BMI.
Despite using a random-lottery selection, there will never be a perfect group that represents the United States population as a whole. While the experiment is the closest representation of an ideal sampling of data, one could assume that the action of deciding who receives and does not receive Medicaid is unethical. However, most of the people in the experiment were low-income uninsured adults. It is arguable that they would have never had the opportunity of obtaining any insurance or the benefits associated without the experiment. The benefits of this experiment outweighed the cost-- through this experiment, results were shared with policymakers, impacting healthcare policy direction. As noted, excess medical expenditures result from treating those with obesity-related diseases. For example, in an earlier study by Finkelstein in 2003, it was noted that the overweight- and obesity-attributable medical spending accounted for 9.1% of total annual U.S. medical expenditures in 1998. The results of Finkelstein’s study indicate that Medicaid has the highest prevalence of obesity-- ten percentage points higher than other insurance categories. In addition, out of the total aggregate medical spending attributable to overweight and obesity, Medicaid accounts for the highest spending, out of private, out-of-pocket and Medicare. While Finkelstein’s regression framework does convey that obesity does contribute to the nation’s growing health care bill, it does not touch upon Medicaid’s effectiveness on decreasing BMI. However, the correlation between Medicaid and obesity is the essence of this research.
In a 2008 study between BMI and medical expenditures for North Carolina adolescents enrolled in Medicaid, the results were similar to Finkelstein (2003). Overweight adolescents and at-risk for overweight adolescents had higher Medicaid expenditures than did normal-weight adolescents. Some of these differences were statistically significant. Overweight adolescents were significantly more likely to have a paid claim for services related to variables such as diabetes or asthma. While some variables in this study are statistically significant, the survey of North Carolina adolescents is not representative of the whole country, nor of adults.
Chester’s “Medicaid at 50: A Look at the Long-Term Benefits of Childhood Medicaid” (2015) is perhaps the best indicator of Medicaid health impact amongst adults. Chester concluded that amongst teenagers, there is a correlation between Medicaid and BMI, in that a 10-percentage point increase in average childhood Medicaid reduced their BMI by 3.9 percent as teenagers. Using longitudinal data, Chester’s study also presented that children with access to Medicaid showed a 26 percentage point decline of high blood pressure in adulthood. While Chester found statistically significant correlations that will be helpful for this study, Chester’s presentation of variables seems one-sided and biased. Miller’s "The Long-Term Effects of Early Life Medicaid Coverage," (2014) also look at the long-term outcomes of Medicaid. Miller studies the outcome of Medicaid on people born between 1979 and 1993 who are still in their mother’s womb or as children. For mothers who gained Medicaid during prenatal care, they had lower obesity, better self-reported health, and a better chance of completing high school. Miller’s research relates to this study because the NLSY also follows people from a young ago into their adulthood. While this study focuses on birth outcomes in the United States, the "long-run perspective when evaluating the benefits of early public intervention" will be applicable. While the data in this study does not go back to before a child was born, both demonstrate the product of Medicaid coverage over a long period of time. The long run perspective will be easier to compare the data to since this study uses a twenty year comparison. Since Miller (2014) focuses on fetuses or young children, they could be "underestimating the effect of Medicaid on hospitalization later in life." This could be a problem if looking at older adults since the data would not be comparable.
The aforementioned studies all will all be helpful to this research project. While the factors presented in some of the studies do not directly correlate to the dependent variable, the correlation between some factors, such as depression and obesity, can be a good indicator of a person’s overall health.
The null hypothesis is that there is not any effect of Medicaid on BMI. The alternative hypothesis is that there is an effect of Medicaid on BMI, which is consistent with the modern research. In this research, the findings typically demonstrate a positive correlation between individuals with Medicaid and low BMI.
In the second iteration of this model, we would be interested to explore the effect on prediction accuracy of utilizing different machine learning algorithms (eg. Random Forest, KNN) in addition to decision tree. In addition, while we completed a linear regression analysis and basic principal component analysis (PCA) to better understand which variables maximize the variance of the data, we would like to further this analysis to explore the effects of additional variables available in the NLSY dataset such as preexisting conditions, use of tobacco, and pregnancy. We would employ fuzzy matching so that the end user does not need to enter the symptoms exactly to produce an accurate output, and would like to design a user-friendly interface to interact with the predictive feature of the tool.
