The Make DEEP Human Project

Modeling and Quantification with Secondary Data Resources: Medically Underserved Populations’ Obesity Intervention Outcomes  

Fall 2024

Diet Morbidity Versus Education in Chicago

Eric Hilary Smith, University of Illinois at Chicago, Jane Addams School of Social Work, Midwest Latino Health, Research, Training, and Policy Center

Introduction

        First of all, the “make DEEP Human Project” is a research collaboration of the University of Illinois at Chicago, Jane Addams School of Social Work, Midwest Latino Health Research, Training, and Policy Center.  The purpose of the Midwest Latino Health Research, Training, and Policy Center is not only to explore latino health disparities, but also, all minority health disparities that are classified in medically underserved communities in the city of Chicago.  And, according to the Center of Medicaid and Medicare research, this classification for minority health disparities is now labeled the social determinants of poor health.  However, in the city of Chicago, social determinants of poor health are characterized by geolocation of all minorities.   And, also, within any geolocation, we can begin to quantify random variables such as mean age, graduation rate, and median income, because they can possibly be correlated to diet morbidity, obesity, and diabetes mortality.  

According to the Centers of Disease Control and Prevention reported, “By education and age, in 2022, the prevalence of obesity decreased as educational levels increased:  37.6% of adults without a high school diploma had obesity;  35.9% of adults with some college education had obesity;  and 27.2%  of college graduates had obesity.  Also, the above statistics from the Centers of Disease Control and Prevention can not be easily ignored or go unchecked, because our primary source of data is not available yet (e.g. NAN for primary source data).  So, anytime in obesity research we approach NAN with our primary source of data, we must use CDC as a secondary source of data, because it is updated weekly through the CDC Morbidity and Mortality Weekly Report.  However, for this paper, we are focusing on Chicago in terms of adults who have not completed their college education that are living with diet morbidity, obesity, prediabetes, and diabetes mortality.  And, also, the rate of education by population may vary as a subset of the larger U.S. population of obesity prevalence, which may be different in Chicago, because the context of meaning, definitions, or medical ontology about diet, diabetes, and obesity.

Methodology

In an earlier article, Aida Giachello, the principal investigator from the UIC Midwest Latino Health Research, Training, and Policy Center stated, “...these are largest minority groups (African American, Latinos, Native Americans, Asian Americans, and MidEastern Americans);  … groups are characterized by low levels of education and income and high levels of poverty;  and groups are affected by disparities in health and access to care….groups tend have a strong sense of family, community, and religiosity/spirituality, and groups use home remedies and over - the - counter medications to treat symptoms of illnesses (Giachello, et. al. 2003).” So, when a primary source of data(i.e. Interview, questionnaire, Survey) is not available yet, how can we statistically classify these groups in terms of targeting medically underserved groups with diet morbidity?  And, also, while working on this project on campus at UIC, I had questions about the above statement with Prof. Giachello.  First, how does low levels of education characterize diabetes?  Second, how does income characterize diabetes?  And, finally, how does high levels of poverty characterize diabetes?  It seems like all the above questions must be answered by building a baseline dataset from an open source that contains secondary datasets, because it is important to demonstrate the diabetes mortality rate as a parameter or variable.

Giachello et. al. goes on further to say in her article the following, “Based on available diabetes data and further consultation with key partners, it was determined that these communities could be mobilized around diabetes and reach a state - of - readiness, that is, engaged in targeted actions to reduced diabetes mortality and complications, with ancillary efforts pointed toward increasing adult vaccination for influenza, for which people with diabetes are at elevated risk.”  So, Giachello was right in establishing that in order to target diabetes treatments for medically indigent populations, there must be some steps taken to provision a new baseline dataset from secondary sources like the CDC, NIH, and the city of Chicago data portal.  However, we run into some problems in starting this tedious and very difficult process to model how education affects short term and long term behavior related to diet morbidity, obesity, and diabetes mortality rates among medically indigent populations in Chicago.  

In the beginning, this seems quite unrealistic to implement and quantify an intervention for medically underserved populations, because each population has a different health belief about the connections between diet, obesity, and diabetes.  Also, patients’ health beliefs can be categorized from variables like education and income.   For example, to further explain the health belief model, Giachello mentioned earlier about different groups or client or patient attributes related to coping mechanisms or coping styles (eg. Strong family, community, or spirituality).  However, does the patient's health belief characterize diet morbidity?   For example, the patient may exclaim, “If I am spiritual, then my blood sugar will decrease!”  Or another example, “If I am religious and go to church every Sunday morning and Tuesday evening, then my blood pressure will decrease!”  These two examples of patient coping mechanisms characterizes the client’s health belief about diabetes self - efficacy.  And, a medically indigent group or population’s  health belief about diabetes self - efficacy is very complex and tedious to model in Chicago.  These different health beliefs may occur at any time or day and can be quantified by gathering what a client may eat.  

So, to make a better connection between the health belief model and diet morbidity,  Boston University, School of Public Health states that “The health belief model derives from psychological and behavioral theory with the foundation that the two components of health related behavior are 1) the desire to avoid illness, or conversely get well if already ill;  and 2) the belief that a specific health action will prevent, or cure illness.  Ultimately, an individual’s course of action often depends on the person’s perceptions of benefits and barriers related to health behavior.  There are six constructs of the Health Belief Model.”  However, in this paper we will discuss the first two constructs to make connections to developing an obesity intervention model for medically indigent groups or populations in Chicago.  And, by using the (HBM) we can begin developing a null hypothesis(H0) and general hypothesis(H1) with our secondary data from the CDC, NIH, and the city of Chicago.  

Also, by quantifying an intervention model for obesity, then we may be able to decrease diabetes mortality rates and increase medically indigent populations quality of life scores.  So, reflecting back to the Health Belief Model, Boston University further states the first construct “Perceived susceptibility - This refers to a person’s subjective perception of the risk of acquiring an illness or disease.  There is a wide variation in a person’s vulnerability to an illness or disease.”  For example, recall the coping mechanism statement:  “If I am a spirtual individual or being, then my blood sugar will decrease day by day.”  Although a patient living with a chronic disease coping statement may be a person’s subjective perception, we must research parameters and variables to classify when their blood sugar decreases only on spiritual days like Sunday morning prayer or Tuesday night prayer meeting.   Also, conversely, parameters and variables may classify on non-spiritual days that a client’s blood sugar may increase on Monday, Wednesdays, Thursdays, Fridays, or Saturdays.

Also, to further support the need to quantify an intervention model for obesity, Boston University, School of Public stated the second construct of the Health Belief Model is “Perceived severity - This refers to a person’s feelings on the seriousness of contracting an illness or disease (or leaving the illness or disease untreated).  This is wide variation in person’s feelings of severity, and often a person considers the medical consequences (eg. disability or medical “death”) and social consequences (eg. family life social relationships) when evaluating the severity.”  Again, just like with the first construct of the Health Belief Model, the second construct is very hard to model without random variables and random events that show medically indigent groups or populations living with obesity.   For example, the second construct yields wide variation in terms of patient attributes or social determinants of disease, because of a change in variable from one geolocation to another geolocation.  And, since there is a change in variable in terms of geolocation, we can begin to quantify an obesity intervention model with independent variables like median household size, mean graduation rate from high school, and median household income, because the diet morbidity rate is dependent on any of these independent variables or random variables in terms of medically indigent groups or populations in Chicago.  And, fortunately, before we can begin to quantify an obesity intervention model, a baseline dataset must be trained and tested from a secondary datasource to show a correlation between income versus household size, or graduation rates versus household size in medically underserved communities, because those attributes, features, and parameters may indicate economic hardship is one of the causes of diet morbidity, obesity, and type 2 diabetes, according to the city of Chicago Health Atlas dataset.  

Next, by using the city of Chicago Health Atlas as a secondary datasource, we would like to show how economic hardship may indicate short term behavior of diet morbidity, along with long term behavior indicating high diabetes mortality among medically underserved groups or populations living in disinvested communities.  And, by observing short term behavior in economically disadvantaged communities of Chicago, we can use the (HBM) and (TTM) to target Cognitive Behavior Therapy(CBT), which may be used as a harm reduction strategy for low income and increased weight.   So, economic hardship can be indicated by using logistic regression model, but it is derived after a linear regression model or multilinear regression model, or when polynomial regression models have failed by using secondary training datasets with features like mean household size, median household income, and parameters such as graduating rates at the elementary and secondary levels.  Also, by using these features and parameters in modeling the prediction may underfit or overfit with its target state.  And, it is important to observe overfitting of secondary data, because it is very difficult to visualize how medically underserved populations’ trajectories are affected by a lack of education in terms of diet morbidity.  Then, that is why a logistic regression model must be used to characterize a medically indigent group or population according to the Health Belief Model first and second constructs.  

If you recall, the first construct of the Health Belief Model is that medically underserved clients may have perceived susceptibility about the medical consequences related to diet morbidity.  So, suppose that some groups and populations lived on the westside in communities like Austin, East Garfield Park, North Lawndale, and West Garfield Park, but the problem is that food deserts are causing fluctuations among medically indigent clients and patients in terms of characterizing perceived susceptibility about diet morbidity.  And, it is important to characterize these fluctuations as independent cycles, which may be occurring at the same time, because each neighborhood or group must be represented as independent data clusters.  For example, the Austin community is an independent data cluster;  East Garfield Park can be represented as an independent data cluster;   North Lawndale can be represented as an independent data cluster;  and West Garfield Park can be represented as an independent data cluster.  

Also, knowing diet morbidity is a risk factor of obesity and prediabetes, populations and groups must begin to be targeted for obesity intervention.  Also, the above independent data clusters may be unstable as random events in medically underserved communities on the westside of Chicago.  For example, independent events, such as poor physical activity, decreased number of hours of sleep, or poor glycemic load may quantify a realistic obesity intervention model.  If medically indigent groups and populations, who live on the westside and have at least graduated from high school may manifest short term behaviors like poor physical activity, decreased hours of sleep, or poor glycemic load, then they are susceptible to diet morbidity.  

Groups and populations that are medically underserved are susceptible to diet morbidity, because they live in communities on the westside covered by food deserts, according to the city of Chicago Health Atlas.  And, now, by using the city of Chicago Health Atlas as a new data realm, we can be able to chard, parse, or partition variables and parameters by loading to Microsoft Excel, then migrating it to Google Sheets, and finally cleaning the secondary data source from the city of Chicago with Python libraries such as:  Matplotlib, Pandas, PostgreSQL, Numpy, Scikit-Learn, Seaborn, and SQLite3.  Also, Python and its libraries can chard, parse, partition variables and parameters into a dataframe.  So,  by quantifying an obesity intervention model with python libraries, we may prevent long term behavior effects related to diet morbidity, by accurately targeting groups within a sample population that are living with hyperaldosteronism, hypercholesterolemia, hyperglycemia, or hyperlipidemia.  

However, diabetes complications like hyperaldosteronism, hypercholesterolemia, hyperglycemia, or hyperlipidemia are parameters that have critical values that help characterize the possibility of perceived severity, according to construct 2 of the Health Belief Model (HBM).  For example, there is a 25% chance that a medically indigent client may perceive the severity of hyperaldosteronism;  or a 25% chance that a medically underserved patient may perceive the severity of hypercholesterolemia;  or a 25% chance that a medically indigent group may perceive the severity of hyperglycemia;  or a 25% chance that a medically underserved population may perceive the severity of hyperlipidemia and decide to report to Cook County Hospital for a medical examination and treatment.  Also, all the aforementioned likelihoods of diabetes mortality are equivalent, but which parameter needs to decrease in order for groups or populations to stabilize that live in food deserts on the westside of Chicago.  

And, each of those random diabetic events represents independent variables that must be quantifiable for an obesity intervention model, because it possibly may infer poor self - efficacy related to the education rate.  For example, here are some self - efficacy If - then statements about diet morbidity:  If I buy sugary soda pop at the corner store, then I may have kidney problems;  If I buy a mister submarine sandwich from the submarine shop, then I may have problems with my cholesterol;  and, If I buy a meal combo from McDonald’s, then I may have problems with my blood sugar.  So, all these aforementioned poor self - efficacy If-then statements may represent hypothesis1 (H1), which may result in an outcome of data of 1 = a diabetic event occurring or 0 = otherwise.  And, these two outcomes of 1 or otherwise 0, may help us indicate behavior modification is needed.

However, an example of a behavior modification is the Transtheoretical Model of Change or stages of change (TTM), and can be implemented to deduce unhealthy eating habits, changes in sleeping patterns, and increased physical inactivity that causes medically indigent populations to be unstable.  And, the (TTM) can be used to flatten the curve that indicates diabetes mortality.  Also, according to Boston University, “The Transtheoretical Model of Change (also called the Stages of Change Model), developed by Prochaska and DiClemente in the late 1970s,...focuses on the decision-making of the individual and is a model of intentional change.  The (TTM) operates on the assumption people do not change behaviors quickly and decisively.  Rather, change in behavior, especially habitual behavior, occurs continuously through a cyclic process…The (TTM) posits that individuals move through six stages of change:  precontemplation, contemplation, preparation, action, and maintenance, and termination.”  So, for the purpose of writing this paper, we will discuss stage 1:  precontemplation and its connection to the medically underserved groups’ health beliefs about reducing the disparities of diabetes mortality on the westside of Chicago.  To further explain, Prochaska and DiClemente states that “Precontemplation - In this stage, people do not intend to take action in the foreseeable future (defined as within the next 6 months).  People are unaware that the behavior is problematic or produces negative consequences.  People in this stage often underestimate the pros of changing behavior and place too much emphasis on the cons of changing behavior.”  And, if medically indigent groups, who have graduated from high school, begin precipitating health behaviors such as poor sleep, increased caloric intake, and low physical activity, then the Transtheoretical Model of Change can be implemented to decrease high blood sugar, weight gain, and high blood pressure.

Conclusion

Earlier, if we recall, high blood pressure, weight gain, and high blood sugar is related to poor self - efficacy about diet morbidity and its health consequences.  And, poor self - efficacy about diet morbidity may be from a poor educational background to read and understand food labeling, because it may lead to obesity and prediabetes without a quantifiable intervention model.  Also, a quantifiable obesity intervention model can be categorized and scaled with the precontemplation stage of the Transtheoretical Model of Change.  For example, we can create a scale from 0 - 6 to categorize a medically indigent client’s precontemplation about changing his or her food choice from the mister submarine shop to Pete’s Produce, because Prochaska and DiClemente stated earlier it could be “defined as within the next 6 months.”  So, suppose we wanted to evaluate the response of westside communities that are living with diet morbidity by asking:  When do you think you would be ready to buy fresh food from Pete’s Produce instead of buying mister submarine?  And, recall from earlier that each westside community is it on individual data cluster, but however, suppose each individual westside community is now its own state space, because the precontemplation stage of the (TTM) is probable, and regardless if a medically indigent clients makes a good food choice or bad food choice on some given day.

For example, the category and scale is as follows:  0 means 0 months to precontemplate;  1 means 1 month to precontemplate;  2 means 2 months to precontemplate;  3 means 3 months to precontemplate;  4 means 4 months to precontemplate;  5 means 5 months to precontemplate;  and 6 means 6 months to precontemplate.  Based on the category and scale for (TTM),  the survey says 0.01% of westside residents that live in food deserts may buy from Pete’s in 1 month;  another 9% would change to Pete’s in 2 months;  another 17% would precontemplate Pete’s from mister submarine in 3 months;  another 34% of medically indigent groups would precontemplate to fresher foods in 4 months from Pete’s;  another 51% of medically underserved on the westside may contemplate leaving mister submarine in 5 months;  and the last 68% of westside residents may precontemplate Pete’s Produce in 6 months.  

However, the shorter times to precontemplate leaving mister submarine for Pete’s Produce seems to be better, because we would only be concerned if some small number of households in each westside community answered 0, and some other households in each of the same westside communities answered 3.  So, what is preventing these small number of households from reducing their time of precontemplation down from at least 6 months to 3 months?  First, we must consider the change in variable to reduce the time of precontemplation from mister submarine to Pete’s Fresh Produce.   It is important to consider the change in variable, because it may indicate that medically indigent households may have access to care issues or their basic needs are not being met.  

And, also, earlier, Giachello pointed out in her article, “...these are the largest minority groups;  both groups are characterized by low levels in education and income and high levels of poverty;  and both groups are affected by disparities in health and access to health care.”  So, as aforementioned by Giachello, and in terms of diet morbidity, we must continue to make connections that a change in variable with median income may vary with mean household size, or medically at risk households that may be experiencing overcrowding on the westside of Chicago.  Also, in Chicago, not just graduation parameters or rates are important, but mean household size is a very important variable for individuals to determine perceived susceptibility about diet morbidity and its connections to obesity and prediabetes.  For example, the larger the household, individuals may not have a perceived susceptibility of poor eating habits, lack of physical activity, or irregular sleeping patterns, which may have long term behavior effects with parameters or rates related to diabetes mortality.  

Implications for Further Study

However, we may run into a problem with simply using the mean household size for designing a quantifiable obesity intervention model for households that maybe or overrepresented or underrepresented with some number of individuals that may be represented with human pronouns such as:   he/she, him/her, them/they/theirs, female/male, or x/y.  And, knowing human pronouns, now data can be clustered together about each individual represented in a medically underserved household size, which may result in categorizing or miscategorizing definitions about obesity:  body fat, body shaming, fat, thick, weight bias, or weight shaming.  So, measuring the mean household size of Chicago westside neighborhoods does not provide enough data to create an accurate obesity intervention model, because each individual represented does not live in the same type of housing, such as an apartment, shelter, single family housing, single room occupancy, or multifamily housing, and with no regards to time.  For example, each type of housing in terms of the median household size has a maximum carrying capacity that may affect a medically underserved individuals’ blood pressure, blood sugar, cholesterol level, height, or weight, because at any time the household size may increase in a fibonacci sequence:  (0, 0, 1, 1, 2, 3, 5, 8…), and which is also based on a change in variable of median household income and education parameters and rates.  

Furthermore, recall the use case of raw data collection methods with the city of Chicago Health Atlas to help solve the problem of overcrowding in medically indigent households, which are in four westside neighborhoods.  For example,

And, within Austin, East Garfield Park, North Lawndale, and West Garfield Park, medically underserved individuals may increase like a fibonacci sequence to the carrying capacity of the mean household size or increase beyond the carrying capacity of the mean household size, which may vary based on median household income as a variable, and parameters or rates of graduation from at least high school.  So, next, we must explore the aforementioned four communities datasets within the city of Chicago Health Atlas to calculate the correlation and linear regression of mean household size, mean population, education rates, and obesity rates.  

Also, it is important to use correlation and linear regression modeling, because calculating the central tendency of all four neighborhoods may not provide enough data realization about creating an accurate obesity intervention model.