STEM I

Abstract

Health insurance is costly; Americans spent over $4.3 trillion on healthcare expenses in 2020 alone, with over 85% through private insurance providers (Center for Medicare and Medicaid Expenses, 2020). In 2010, The Patient Protection and Affordable Care Act restricted health insurers' ability to deny and raise rates due to pre-existing medical conditions. However, it was challenging to enforce this section of the act because very little information is known about how insurance providers determined their rates. This study aimed to help resolve this knowledge gap by conducting a correlative study between rates of genetic predisposition and health insurance costs. Health insurance data was collected using public marketplace databases, and information on rates of health conditions was collected. Then, a Pearson Correlation was run on the two datasets. The result was a Pearson Coefficient of r = 0.1198 for all conditions and a statistic of r = 0.1464 among genetic conditions. This result offered significant evidence that health insurance companies used information on genetic predisposition to determine their rates. This conclusion has significant implications for both the healthcare and genetic testing industry. The information provided by this study could be considered by clients interested in proceeding with genetic testing, as being informed of the risk factors associated with health insurance rates resulted in doing so. Additionally, the study provided evidence for conducting a future study using genetic mutation rates rather than incidence rates, which would be used to rephrase legislation and terminology in the Affordable Care Act.

Background

The Patient Protection and Affordable Care Act of 2010 (United States, 2010), also commonly known as Obamacare, has been known to be a rather controversial policy by many Americans. Nevertheless, it has had many undeniable impacts on the medical industry and health insurance providers. The subsection of the law regarding policy relating to pre-existing health conditions is often considered to be one of the most, if not the most impactful part of the act (Kirzinger, 2022). The section states that no individual with a pre-existing medical condition may be refused health insurance because of their pre-existing medical condition. Although aspects of these policies vary from state to state (Kominski, 2017), the particular section mentioned above remains applicable on a national level. However, insurance costs remain significantly expensive for Americans across the nation, and these costs are only growing. In 2021, Americans spent about $4.3 trillion on health care expenses – nearly 85% of these expenses being into Private Insurance Providers- meaning the average American spends approximately $10,976 a year on health insurance. Although the Affordable Care Act (ACA) aimed to combat these costs, in response to the policies put into place by the ACA, health insurance agencies have several methods to combat these restrictive policies by taking advantage of loopholes in the legislation. One such loophole exists due to the poor definition of the phrase “refusing health insurance based on pre-existing medical conditions” (United States, 2010). Because of the wording of this section, although it is illegal for insurance agencies to directly reject someone due to a pre-existing medical condition, they can instead determine the probability that someone has an existing medical condition or will be likely to develop one in the future based on information on their demographics. This process will provide the insurance agencies with a reasonable estimate of whether a potential client has a pre-existing medical condition without needing to directly evaluate if the potential client has a pre-existing condition, thus making the process technically legal while having practically identical results as if they violated the restriction under the ACA.

While this loophole has existed in the past, it is incredibly more prominent today than ever due to the extensive development done with genetic testing technology in recent decades (Rodriguez-Rincon 2022). Almost anyone can get their hands on a genetic test and for much cheaper than it used to cost. Because of this outstanding influx in the percentage of the population taking genetic testing, there is now an incredibly large amount of data available to other genetic testing providers. By taking a genetic test, an individual not only gives their provider access to all the information relating to their genome, but also information on notable statistics such as medical conditions, medical history, age, sex, ethnicity, and geographical location. All of this data is collected and inputted into databases. While these databases have existed for a substantial amount of time, they have not had nearly as much data available as they do now. Because of this, the data available can be compiled into numerous Genome-Wide Association Studies (GWAS) which are ever more accurate and more developed than before.

Importantly, with the increase in data from GWAS available, that information is now also accessible to health insurance agencies. One important thing to note is that how insurers determine their rates is generally very unknown. Because of this, health insurance providers could be using the information provided by Genome-Wide Association Studies to calculate their rates, and it would be unknown to the public. Additionally, along with data from GWAS, insurers likely compile other medical and genetic information such as through screening panels, rates of conditions by demographic, and other similar data, all of which are publicly available, to get a very informative and statistically accurate understanding of how predisposition to developing various conditions varies by demographic. Nevertheless, this information is all related to the rates of the genetic diseases themselves, which is what this study used as data to research whether insurance agencies are performing the previously described procedure. By taking the data from the rates of several conditions and analyzing their correlation with health insurance rates by geographic location, the level of correlation will reveal the likelihood and extent to which insurance providers are using the listed information to determine their rates.

Procedure

The health exchange data website, HIX Compare, has publicly accessible reports on health plans by health insurance rate area, as well as crosswalk datasets between health insurance rates and county. This resource is updated annually, so this source was used to obtain the average silver plan premium by county from 2014 to 2018. Data obtained from this resource was compiled and stored in one large spreadsheet.

Numerous reports, studies, and organizations have publicly accessible data on rates of health conditions by county in the United States. This study obtained data from numerous reports, such as the National Cancer Institute’s State Cancer Profiles, which models the incidence rates of cancers by county from 2016 to 2020. Similarly, the Institute for Health Metrics and Evaluation provides a report on the Mortality rates of diseases such as sexually transmitted infections, diabetes and kidney disorders, and non-health related injuries from 2000-2019. This report also used national atlases, such as the Center for Disease Control and Prevention’s Interactive Atlas of Heart Disease and Stroke in 2019 and the Centers for Medicare & Medicaid Services annual reports from 2007-2018 on Chronic Conditions. For a preliminary model, this system utilized annual reports on Chronic Conditions as health condition datasets.

As mentioned above, the health insurance data collected was displayed by the health insurance area, whereas the health condition data was collected by county. To be able to map the data, a dataset of rates by county had to be obtained. Conversely, the most accurate Pearson Correlation depended upon having datasets of rates by health insurance area. For these reasons, using the previously provided crosswalk dataset, datasets were converted such that there was a dataset of health insurance rates and health condition incidence rates by county as well as by health insurance rate area. This calculation was performed in Google Sheets using the “IF”, “REGEXMATCH”, “XLOOKUP”, and other functions provided by the software.

While there are other factors considered in health insurance cost calculations, this study identified smoking and state-by-state differences as two primary external factors from health rates as contributors in health insurance price calculations. However, whether the client is smoking is asked and considered in the premium calculation process. Therefore, differences in rates of smoking by county are not considered in premium calculation. Conversely, state-by-state differences in policy, as well as the environment, are not avoided by this; a great portion of how much one’s health insurance costs depends upon their state rather than their county. Therefore, intending to avoid this, this study correlated the amount of change over the years in percent difference. By eliminating smoking and state-by-state policies, this study created a more accurate and predictive model. Finally, rather than investigating the lowest policy by the state as many popular models do, this study decided to utilize one health insurance plan to project how one plan in particular changes as health condition incidence rates are adjusted, choosing to only utilize Silver-Tiered Blue Cross Blue Shield Plans, as it is one of the more popular providers, spanning over 20 states. There are many advantages of this approach in comparison to using the lowest rate. For one, it will eliminate the variable that different health insurance providers are using different data to determine their rates. This benefit in turn enables the study to determine exactly what factors are used for each plan. Conversely, a negative of this approach is in increasing the demand for datasets included if the goal is to gain an understanding of all health insurers. However, this model functions solely as a more preliminary study, so it does not necessarily require this. See below for maps of the data obtained from this process created using Datawrapper.

Discussion

While it is known that a Pearson value less than or equal to 0.5 is not statistically significant evidence for a correlation, this study is rather a comparison of Pearson coefficients. Given that there is a larger correlation coefficient for one health condition with health insurance premiums than another, there will be a greater overall correlation between the first two than the second.

As this model analyzed is a preliminary version, there are several limitations to the design that should be taken into account. For one, the amount of data that this study was able to compile for statistical testing became exponentially limited as mapping health insurance plans was undergone. To avoid this, this model may be expanded into considering additional plans rather than one, along with adding more health condition incidence rates to create a more informed final result.

This study expands upon several preexisting studies to go about answering its central question of “How do health insurance companies determine their rates?” One study which a great portion of the idea for this project stems from is “Assessing the Impact of Developments in Genetic Testing on Insurers’ Risk Exposure” by Rodriguez-Rincón et. al., 2022. As a study investigating a very closely related problem from the perspective of insurers, it had a significant impact on the development of this study.

Following expanding this model for more insurance plans and/or health conditions, if there is sufficient preliminary evidence gathered from this initial model, the results of this study could be used as significant evidence for running a similar study using genetic predisposition rates.

STEM I

Course Description

Assessing the Correlation Between Genetic Predisposition and Health Insurance Rates

Overview

Abstract

Research Proposal

Phrase 1

Phrase 2

Background

Procedure

Analysis

Discussion

Conclusion