An Empirical Analysis of Avoidable Medicare Payments and Medicare Payment Variations

Abstract

This study investigates the determinants of average Medicare payments, focusing on the impact of secondary diagnoses, demographic factors, and regional disparities. Our model reveals that major complications and comorbidities (MCCs) significantly increase average Medicare payments compared to less severe complications (CCs), with coefficients of 7674.13 and 1542.67, respectively. Higher percentages of the elderly living in poverty correlate with increased payments, highlighting the influence of social determinants on healthcare costs. Demographic analysis shows that counties with higher populations of Black, Hispanic, Asian, or other races receive higher Medicare payments, indicating potential racial disparities in healthcare access and quality. Political affiliation also plays a role, with Democrat-voting counties incurring higher Medicare costs. Surprisingly, larger populations served correlate with decreased payments, possibly due to cost-containment programs like Medicare’s Competitive Bidding. State-level analysis reveals significant regional variations, with states like Oregon and Wyoming showing higher payments, while Mississippi and Georgia have the lowest. Our findings underscore the complexity of Medicare funding and the need for targeted policy interventions to address health disparities and ensure sustainable healthcare delivery.

1. Introduction

U.S. healthcare spending is high, and it is ultimately unsustainable. According to the Centers for Medicare and Medicaid Services (CMS), the total U.S. health expenditure in 2019 was USD 3.8 trillion. The share of Gross Domestic Product (GDP) spent on healthcare was 17.7%. In 2019, Medicare spending was more than USD 750 billion or 21% of the total national health expenditure which makes Medicare the biggest payer. In this project, we firstly estimate the amount of Medicare payment due to hospital Complications and Comorbidities (CCs) and Major Complications and Comorbidities (MCCs). Secondly, we estimate potential Medicare savings by reducing preventable CC/MCCs. There is little literature studying the potential savings on U.S. healthcare expenditure by reducing avoidable hospital readmissions, but no study focused on the potential savings that can be generated from reducing preventable CC/MCCs. This study will fill this gap. Thirdly, we want to examine if there are any variations in Medicare payments among races, states, and hospital sizes.

Medicare is an extremely important part of the United States healthcare system. The Medicare system, under the Centers for Medicare and Medicaid Services (CMS), was founded in 1965. It allows elderly persons and some disabled beneficiaries to enroll in Part A of Medicare, which pays for hospital care [1]. This is important, as the elderly are more likely to have chronic illnesses, with approximately 80% of seniors having at least one chronic illness and 25% of seniors unable to perform daily activities due to their chronic conditions [2]. The Medicare system provides health and financial security for tens of millions of people in the United States over the age of 65 and takes up a decent amount of the national health expenditure at about 21% of the USD 3.8 trillion total health expenditure [3]. Improving the efficiency and efficacy of Medicare will not only improve the lives of millions of people in the United States, but it could also help to reduce the cost to the American people.

To improve the Medicare system, it is important to understand what causes variations in the average costs of Medicare. An important part of the Medicare system is the idea of a diagnosis-related group (DRG). A DRG allows for the grouping of patients based on the principal diagnosis, secondary diagnosis, and procedures performed [4]. The DRG system was meant to increase the efficiency of care and improve transparency in hospital activities [5]. Payment rates are assigned to each hospital at the beginning of the year, and to calculate a specific payment, each patient is given a DRG with a corresponding weight, which is then multiplied by the hospital payment rate [6].

In this study, we look at the average Medicare payments for a large variety of DRGs and study the difference in average Medicare payments based on a variety of factors. One major factor we look at is secondary diagnoses attached to each DRG, such as complications or comorbidities (CCs) or major complications or comorbidities (MCCs). Complications occur during a patient’s stay in a hospital setting that results in a longer stay, while comorbidities are conditions in the patient that existed before admission [7]. We also attempted to find any effects between average Medicare payments and variables such as the county population, economic status, political affiliation, race, and percentage of adults over the age of 65 who are living in poverty. Furthermore, when analyzing the payments system under Medicare, it is important to distinguish between outpatient and inpatient treatment as those will heavily influence payment. The difference between outpatient and inpatient is that outpatients do not have an order from a doctor to be admitted as an inpatient, while inpatient status begins when the patient is admitted to a hospital with a doctor’s order (Medicare, inpatient, or outpatient hospital status affects your costs).

We mainly focus on inpatient care throughout our study, but we feel it is necessary to mention the role of the site of the service differential, which means that there are different rates for physician’s services depending on the setting [8]. In non-facility settings, physicians must pay for their equipment and supplies, while in facility settings, physicians do not need to supply their own equipment or supplies, due to the facility supplying them. Because of this, reimbursement given to physicians is expected to be less in facility settings than in non-facility settings [9]. Regarding the site of service, studies have shown that the highest cost estimates for hospital outpatients are USD 9.9 to 35.8 billion, followed by hospital inpatients at USD 5.0 to 24.3 billion [10]. For our main study, we excluded outpatient care and instead focused on inpatient discharges for Medicare fee-for-service (FFS) beneficiaries, which reimburse healthcare providers for each service they perform to the patient (Healthcare.gov, Glossary). It is important to note that there is a belief that because fees are paid for each service, healthcare providers will focus mainly on the number of treatments and not the quality since they will be paid regardless of whether the patient’s health improves [11].

2. Literature Review

In this paper, we examined three unique aspects of Medicare on which we strongly believed analytical research was absent. As we’ve mentioned, these are the impact of CC/MCCs on Medicare payments, the estimation of potential Medicare savings by reducing CC/MCCs, and finally, examining variations in Medicare payments among different races, states, and hospital sizes. Research of varying levels has already been completed on these unique aspects that differ in scope and focus. We aimed to establish the depth at which research has been completed and identify how our analysis examines an untouched field of study.

2.1. Impact of CC/MCCs on Medicare Payments

Through our analysis, we have determined a significant impact on Medicare payments due to complications and comorbidities (CCs) and major complications and comorbidities (MCCs). A meta-analysis conducted by Christopher Scally et al. discovered that hospitals that successfully reduced their complications also had a reduction in Medicare payments, indicating a connection between the complications and Medicare payments [12]. Their study examined nearly two million patients undergoing general and vascular procedures and discovered that top hospitals that focused on quality improvement had a significant reduction in payments per patient. This study aims to assess the impact of quality improvement, defined by a change in serious complication rates, on Medicare payments. We believe there is a gap in research where the overarching impact of complications and comorbidities, both major and standard, on Medicare payments is not quantified.

Another study, conducted by Jennifer L. Wolff, discovered that the risk of avoidable inpatient admission, or preventable complications in an inpatient setting, increased dramatically with the number of chronic conditions present within the elderly population hospitalized [13]. The study was conducted on nearly 1.3 million patients over the age of 65 who were enrolled in Medicare Part A and Part B. Multiple logistic regression models were used to analyze how age, sex, and the quantity of chronic conditions impact preventable complications. This study discovered that better primary care, such as coordination of care, can reduce avoidable hospitalization rates, especially when multiple chronic conditions are present. The analysis conducted highlights the net positives of reducing preventable complications and comorbidities in an inpatient setting for the Medicare populace but does not express these benefits in a monetary valuation which we believe to be much more efficient in terms of stressing the importance of reducing preventable complications and comorbidities.

2.2. Estimation of Potential Medicare Savings by Reducing CC/MCCs

Furthermore, in another study conducted by Merath and colleagues, Medicare payments were found to be significantly lower for patients who did not develop complications when compared to patients who did [14]. The retrospective cohort study was conducted between 2013 and 2015 on 13,873 patients using the Medicare Provider Analysis and Review Database, also known as MEDPAR. This study had a differing scope in analysis, focusing on complications following hepatopancreatic surgery. These articles illustrate a genuine impact on Medicare payments due to complications and comorbidities, while simultaneously highlighting the need for an analysis of impact on a national scale.

To further accentuate the need for an accurate estimation of potential Medicare savings by reducing CC/MCCs on a national level, a paper written by Brian Fry et al. analyzed the impact of reducing postoperative complications on Medicare payments [15]. This study was a retrospective review of 37,329 Medicare beneficiaries undergoing bariatric surgery for the years 2005–2006 and 2013–2014. Through their study, they concluded that there is a strong association between reductions in complications and decreased Medicare payments. In fact, the top 20% of hospitals had a decrease in the average serious complication rate of 7.3% and an average per-patient savings of USD 4861, while in comparison, the lowest 20% of hospitals had a smaller decrease in the complication rate of 0.8% and conversely, a smaller average savings of USD 2814. This study indicates a genuine relationship between reducing complications and average Medicare payments, as well as providing quantified estimates of this relationship. It is important to stress that this study was focused on Medicare patients undergoing bariatric surgery, whereas our study aims to apply this observation and analyze it on a national scale.

2.3. Variation in Medicare Payments among Races, States, and Hospital Sizes

Our study also aims to identify variations in average Medicare payments among races, states, and hospitals to identify possible relationships. In an article published by Nora Super, a study was conducted to examine the sources of variation in Medicare payments and costs across different geographic areas and different sites of care. This study analyzes multiple factors that can impact expected Medicare payments and how benefits and premiums can play a role in varying payments among differing regions [16]. An area we aim to touch upon which we believe was left out is an accurate representation of the variation in average Medicare payments when observed through the lens of differing races and hospital size. The tables provided delve into state rankings by quality indicators to provide some justification for variation in Medicare payments. However, we aim to analyze the variation in Medicare payments due to race, state, and hospital sizes, which remains uncharted.

3. Method

3.1. Hypothesis

There are a variety of factors that could potentially play a role in determining the price of healthcare in the United States. For example, there is a large variation in Medicare spending across different regions, which is unrelated to the health outcomes of the patient [17]. In addition to location, we hypothesize that other notable variables play a part in the average Medicare payments that hospitals receive. One such variable we believe plays an important part in average Medicare payments is the DRG assigned to each patient, along with any secondary diagnosis. Two important secondary diagnoses that we believe to have the largest impact on average Medicare payments are the CC and MCC codes, which have been described previously. CC and MCC secondary diagnosis can be assigned to patients with the same main procedure; for example, extracranial procedures can have both diagnoses attached depending on severity. The procedures with their respective secondary diagnosis will be classified as different DRGs [18]. Since MCC secondary diagnoses are considered major, it is a more severe condition, which means it is possible that a physician needs to use more resources on that patient [19]. Therefore, we believe that MCC secondary diagnoses will have a larger impact on average Medicare Payments than CC secondary diagnoses.

We also hypothesize that specific county-level data could play a role in determining average Medicare payments. In this study, we look at county-level data such as the estimated population, political affiliation, and the percentage of adults over 65 who are living in poverty. We are using the estimated population per county as a variable to estimate the population that an individual hospital serves, and we believe that by serving more people, average Medicare payments will increase. We are basing this off the fact that it is normally more expensive to live in more urban areas and we believe that healthcare costs will also increase because of this. When comparing total spending between rural and urban households in 2011, there was an 18% increase in total spending for households located in urban areas [20]. We also linked this idea to political parties, in that Democrat-voting counties are usually more urban areas [21]. Therefore, we believe counties that vote Democrat will have higher average Medicare payments. We also strongly believe that the health of the adult population over the age of 65 plays a critical role in average Medicare payments, as a healthier population should, in theory, lead to less severe conditions resulting in lower payments. Therefore, we also hypothesize that larger percentages of adults over the age of 65 who live in poverty will lead to larger average Medicare payments.

3.2. Data

We used data from four different sources. The main data source we studied is the Inpatient Charge Data from the CMS for the year 2017. The data include information on inpatient discharges for Medicare fee-for-service beneficiaries, and they include over 3000 providers and over 190,000 observations. This data includes information about providers such as the state and zip code, as well as information about DRGs for patients and average Medicare payments [22]. We also included cost reports from the Centers for Medicare and Medicaid Services from the year 2018, which include hospital information such as the total amount of Medicare discharges for each hospital [23]. Another data source we used contains data about county-level information for the location each provider is in, which were obtained through the United States Census Bureau, and we used these data to include the estimated county population in 2017 [24]. The fourth dataset we looked at for our analysis was taken from OpenDataSoft and includes county-level information such as political affiliation in the 2016 presidential election and county-level population health statistics [25]. After joining all our data sources together, and cleaning our data, we are left with a dataset that has 174,323 observations. We also used a model generated that included Outpatient Charge Data from CMS for the year 2017 [26]. We used this model as a robustness check to see if our conclusions from our primary model stayed constant with new but similar data.

We looked at many variables for our analysis. The variables that we used to explore the causal relationships with average Medicare payments include county political affiliation, the total number of Medicare discharges per hospital, secondary diagnosis (CC and MCC), and the percentage of the elderly population over 65 in poverty. Furthermore, we also have the percentage of the population that is White, Black, Hispanic, Asian, Native American, or another race in a county and the estimated population served by the provider, as well as all DRGs which are present within our data. The two measures we selected pertaining to each patient’s DRG are the secondary diagnoses of CC and MCC. These two secondary diagnoses appear the most within our dataset and give a good indication of how severe each diagnosis is. Both variables are coded as 1 if the patient has been classified with one of the diagnoses, and 0 otherwise. Our other variables of importance deal with county- or hospital-level data. For our analysis, we focused mainly on the two variables from the DRGs, CC and MCC. The main reason is that information related to a patient’s DRG, specifically the severity of their diagnosis, is related to the number of resources that are used on that patient, which is explained in the Hypothesis Section. With more resources spent on a single patient, it is logical that the price of care for that patient should go up [27]. Descriptions of all the variables we included in our analysis are shown in

Table 1. Description of variables.

Medicare Payments	The average amount that Medicare pays to the provider.
Political Affiliation	A dummy variable that is 1 if the county voted Democrat in the 2016 presidential election, and 0 otherwise.
Total Hospital Medicare Discharges	The number of discharges billed by the provider for inpatient hospital services.
CC	A dummy variable that is 1 if there is a complication or comorbidity, and 0 otherwise.
MCC	A dummy variable that is 1 if there is a major complication or comorbidity, and 0 otherwise.
Percent 65 or Older in Poverty	The percentage of the county population that is over 65 and living under the poverty line.
Estimated Population Served (thousands)	The estimated population that each hospital serves.
State	The state the specific hospital is in.
Diagnosis-Related Groups (DRGs)	The diagnosis-related group for the patient. They are grouped into their respective major diagnostic categories defined by ICD-10 v37.
Median Income (thousands)	The median income measured in thousands.
Percent White	The percentage of the county population that is White.
Percent Hispanic	The percentage of the county population that is Hispanic.
Percent Black	The percentage of the county population that is Black.
Percent Asian	The percentage of the county population that is Asian.
Percent Native American	The percentage of the county population that is Native American.
Percent Other Race	The percentage of other races not already specified in county.

and

Table 2. Summary statistics on variables of interest.

	Mean	SD	Median	Minimum	Maximum
Medicare Payments	11,083.31	11,032.91	7892.67	112.24	491,509.68
Political Affiliation	0.56	0.50	1.00	0.00	1.00
Total Hospital Medicare Discharges	5616.27	4160.14	4642.00	10.00	31,440.00
CC	0.30	0.46	0.00	0.00	1.00
MCC	0.37	0.48	0.00	0.00	1.00
Percent 65 or older in Poverty	9.48	3.72	8.60	0.80	37.10
Estimated Population Served (thousands)	1024.81	1709.02	461.56	1.34	10,103.71
Median Income (thousands)	29.63	5.96	28.90	8.70	55.71
Percent White	66.21	19.24	68.90	2.50	98.70
Percent Hispanic	13.65	14.49	8.00	0.20	97.15
Percent Black	13.63	12.72	9.70	0.00	86.10
Percent Asian	3.92	4.41	2.50	0.00	32.10
Percent Native American	0.58	2.32	0.25	0.05	74.55
Percent Other Race	2.00	0.93	1.80	0.05	13.45

.

3.3. Model

To determine the effects of our independent variables on our outcome variable (average Medicare payments), we used a linear regression model. More specifically, the model used is represented by the following equation:

$$\begin{array}{c}{MP}_i={\beta }_0+{\beta }_1\left({eps}_i\right)+{\beta }_2\left(c{c}_i\right)+{\beta }_3\left(mc{c}_i\right)+{\beta }_4\left(m{d}_i\right)+{\beta }_5\left(\rho a_i\right)+\\ {\beta }_6\left({poverty}_i\right)+{\beta }_7\left({income}_i\right)+{\sum }_{l=1}^{27}{\mathsf{\delta }}_l\left({drg}_{\mathrm{i}l}\right)+\\ {\sum }_{\mathrm{m}=1}^6{\mathsf{\phi }}_{\mathrm{m}}\left({race}_{im}\right)+{\sum }_{\mathrm{n}=1}^{46}{\mathsf{\vartheta }}_{\mathrm{n}}\left({state}_{\mathrm{i}\mathrm{n}}\right)+{\epsilon }_i\end{array}$$

where MP is the average Medicare payment, eps is the estimated population served, cc is the complications or comorbidities, mcc is the major complications or comorbidities, md is the total Medicare discharges, pa is the county political affiliation, poverty is the percentage of the population over 65 living in poverty, and income the median income. For drg, δ represents each DRG we looked at, which we grouped by major diagnostic categories (MDCs) to get a better look at how broader categories affect Medicare payments. For race, $\mathsf{\phi }$ represents the percentage of each race in a county, including the Percent White, Percent Hispanic, Percent Black, Percent Asian, Percent Native American, and Percent Other Races variables. For the state, α represents each state we used in our model, which includes all states excluding Maryland, Alaska, and Hawaii. We looked at the coefficients for each variable to determine how average Medicare payments increase when holding all other variables in the model constant. This will be our method for determining how each variable affects average Medicare payments, and we used p-values (p < 0.05) to determine the statistical significance of each variable [28].

4. Results

Table 3. Findings from the model.

	Coefficient Estimator	95% Confidence Interval
CC [Yes]	1542.67 ***	1442.15–1643.19
MCC [Yes]	7674.13 ***	7578.02–7770.24
Total Hospital Medicare Discharges	0.38 ***	0.37–0.39
Estimated Population Served (thousands)	−0.14 ***	−0.17–−0.11
Percent 65 or Older in Poverty	138.45 ***	122.23–154.66
Percent White	−1439.22 ***	−2249.74–−628.70
Percent Hispanic	−1431.62 ***	−2242.04–−621.19
Percent Black	−1409.51 ***	−2219.98–−599.03
Percent Asian	−1220.83 **	−2030.94–−410.72
Percent Native American	−1445.07 ***	−2256.01–−634.13
Percent Other Race	−1190.78 **	−2004.13–−377.43
Political Affiliation	361.35 ***	242.10–480.60
Median Income (thousands)	3.02	−7.06–13.09
State [AZ]	4708.74 ***	4258.51–5158.97
State [CA]	4551.41 ***	4139.41–4963.41
State [OR]	5097.79 ***	4572.76–5622.83
State [VT]	5140.63 ***	4210.80–6070.47
State [WY]	5429.10 ***	4357.46–6500.73
Heart Transplant or Implant of a Heart Assist System	218,387.79 ***	216,501.81–220,273.777
Liver Transplant	70,700.39 ***	68,572.62–72,841.08
Tracheostomy for Face, Mouth, and Neck Diagnoses or Laryngectomy	24,151.53 ***	22,394.81–25,908.26
Autologous Bone Marrow Transplant or T-cell Immunotherapy	40,385.48 ***	38,388.39–42,382.58
Burns	36,089.23 ***	33,332.04–38,846.42
Multiple Significant Trauma	14,021.78 ***	12,679.41–15,364.14
Extensive O.R. Procedure Unrelated to Principal Diagnosis	16,789.56 ***	16,119.25–17,459.88

*** p-value < 0.001, ** p-value < 0.01.

shows the results that we were able to obtain from our model. As we hypothesized, secondary diagnoses, such as CC and MCC, have a very significant effect on average Medicare payments. The coefficients for CC and MCC diagnosis are 1542.67 and 7674.13, respectively. Therefore, MCC secondary diagnoses lead to much larger average Medicare payments to hospitals than CC secondary diagnoses. This makes sense, as an MCC would take up more of a hospital’s resources than a complication or comorbidity that was not designated as major. The amount of Medicare discharges in a hospital was also a significant variable. With a one-unit increase in total Medicare discharges, we saw a slight increase in the average Medicare payment.

As shown in Table 3, the percent of the population in a county over the age of 65 living in poverty was associated with higher average Medicare payments, which makes sense as poverty is associated with worse health, which would lead to more severe health issues [29]. A higher percentage of a White or Native American population living in a county led to lower average Medicare payments than higher percentages of Blacks, Hispanics, Asians, or Other Races. The Percent Asian and Other Races variables had a larger, negative beta coefficient than any other race variables, indicating that higher percentages of the population in a county that are Asian, or a race not described in our model, lead to much less of a decrease in average Medicare payments. This could signify racial disparities in the quality of treatment different ethnicities receive. The political affiliation of a county, which we initially thought would influence average Medicare payments, was positive and shown to be significant in our model with a p-value < 0.001, indicating that Democrat-voting counties do have higher average Medicare payments, matching our hypothesis. However, a larger estimated population served was also shown to decrease average Medicare payments, which disproves our hypothesis that a larger population served leads to an increase in average Medicare payments. The results for political affiliation could be due to differences in local government policy perhaps, rather than the population as we originally thought. The decrease in average Medicare payments with an increase in the estimated population served could possibly be explained by programs such as Medicare’s Competitive Bidding program, which aims to reduce the price for beneficiaries as well as save Medicare money, in that more populated areas have more competition [30]. Studies have shown that it is possible that programs such as Competitive Bidding could potentially be used to reduce the cost of Medicare [31]. The median earnings for a county were shown not to be significant based on the results from our model, with a p-value of 0.475. Since Medicare focuses on senior people who are most likely at the end of their careers or already retired, it makes sense that income does not affect prices. There was also a difference in payments among states, as shown in Table 3. Alabama was left out and used as a baseline for our other states to reference. In Table 3, it is shown that there are several states that lead to significantly more payments, and our appendix shows values for all other states used in our model. This shows that there is a large disparity in the average Medicare payments received by hospitals for differing states.

There were also several DRGs that had a large and significant effect on average Medicare payments. We used Alcohol/Drug Use and Alcohol/Drug Induced Organic Mental Disorders as a reference level for all our groupings. We were able to identify seven different groups that had a regression coefficient greater than 10,000 and a p-value < 0.001. These groups were heart transplants or the implant of a heart assist system, liver transplants, tracheostomy for face, mouth, and neck diagnoses or laryngectomy, autologous bone marrow transplant or T-cell immunotherapy, burns, multiple significant traumas, and an extensive O.R. procedure unrelated to the principal diagnosis. These findings are shown in Table 3. All these DRGs listed seem to be more severe than the others, such as diseases and disorders of the eye or diseases and disorders of the digestive system. Since these DRGs are more severe, it makes sense that the Medicare payments would be higher.

5. Robustness Check

Our second experiment includes the data for outpatients from the CMS (

Table 4. Summary statistics on variables of interest, second experiment.

	Mean	SD	Median	Minimum	Maximum
Medicare Payments	4223.36	5273.55	2207.00	2.00	43,018.00
Political Affiliation	0.49	0.50	0.00	0.00	1.00
Total Hospital Medicare Discharges	4415.25	3605.30	3568.00	1.00	31,440.00
Percent 65 or Older in Poverty	9.34	3.54	8.55	0.80	37.10
Estimated Population Served (thousand)	904.15	1610.52	353.34	1.34	10,103.71
Median Income (thousands)	29.13	5.84	28.31	8.70	55.71
Percent White	68.54	19.18	71.85	2.50	98.70
Percent Hispanic	12.82	14.09	7.35	0.25	97.15
Percent Black	12.47	12.55	8.45	0.00	77.70
Percent Asian	3.55	4.22	2.20	0.05	32.10
Percent Native American	0.64	2.07	0.25	0.05	72.25
Percent Other Race	1.98	0.96	1.80	0.05	13.45

). Our finished dataset for this experiment included 55,880 observations after joining datasets together and cleaning the data, which is significantly less than our inpatient dataset. A major difference in outpatient classification systems is the use of an Ambulatory Payment Classification (APC), which is used to pay hospitals for outpatient services. We separated the APC level and procedure from the specific APC description and examined the APC levels like our CC and MCC codes, since both were attached to their respective classifications or groupings, and the procedures like the DRGs from the inpatient experiment. Each procedure can have different levels attached to it. For example, an Airway Endoscopy can be Level 3, 4, or 5. For this experiment, we ran a similar regression model to our inpatient data:

$$\begin{array}{c}{MP}_i={\beta }_0+{\beta }_1\left({eps}_i\right)+{\beta }_2\left({md}_i\right)+{\beta }_3\left({pa}_i\right)+{\beta }_4\left(poverty\right)+{\beta }_5\left({income}_i\right)+{\sum }_{l=1}^6{\mathsf{\delta }}_l\left({apclevel}_{\mathrm{i}l}\right)+\\ {\sum }_{k=1}^{18}{\mathsf{\Phi }}_k\left({procedure}_{ik}\right)+{\sum }_{\mathrm{m}=1}^6{\mathsf{\phi }}_{\mathrm{m}}\left({race}_{im}\right)+{\sum }_{\mathrm{n}=1}^{46}{\mathsf{\vartheta }}_{\mathrm{n}}\left({state}_{\mathrm{i}\mathrm{n}}\right)+{\epsilon }_i\end{array}$$

where MP is the average Medicare payment, md represents the total hospital Medicare discharges, pa represents the county political affiliation, poverty represents the percentage of the population over 65 living in poverty, and income represents the median income. For the APC level, δ represents Levels 2 through 7 regarding the APC description. For the procedure, Φ represents each procedure associated with a specific APC description; there are 18 of these procedures used, not including the level held out as a baseline. For race, φ represents the percentage of each race in a county, including the Percent White, Percent Hispanic, Percent Black, Percent Asian, Percent Native American, and Percent Other Races variables. For the state, ϑ represents each state used in our model, which includes all states excluding Maryland, Alaska, and Hawaii. Alabama was used as a baseline, like the inpatient model.

From the results of our model, we discovered that a higher level associated with an APC led to an increase in average Medicare payments. For the levels, APC Level 1 was used as a reference for the dummy variable. Thus, like our inpatient model with CCs and MCCs, a higher APC level is associated with larger costs for Medicare. Other similarities include more payments to counties with a larger percentage of adults over 65 in poverty, fewer payments to a larger estimated population served, and some procedures with drastically different effects on payments. For the procedures, the Airway Endoscopy was used as a reference level since it was a dummy variable. Some of the procedures with large effects that were shown to be significant in our model are Electrophysiologic Procedures, ICD, and Similar Procedures, and Neurostimulator and Related Procedures.

Some major differences in our model using outpatient data are that race is no longer significant, and neither is the political affiliation of a county. The beta coefficient for total hospital Medicare discharges is also shown to be zero in our outpatient model. These differences could perhaps be explained by differences in the payment systems between inpatients and outpatients.

For the states in our inpatient model, Oregon (OR), Vermont (VT), and Wyoming (WY) led to the highest increase in average Medicare payments from the reference level. For the outpatient model, Oregon (OR) and Wyoming (WY) also seemed to contribute a large amount towards affecting average Medicare payments the most. Oregon (OR) led to the fourth highest increase in Medicare payments and Wyoming (WY) led to the ninth highest increase in Medicare payments in the outpatient model. Arizona (AZ) led to the fourth highest increase in average Medicare payments in the inpatient model and was third in our outpatient model. California (CA) also led to the highest increase in payments in our outpatient model and was the fifth highest state leading to higher average Medicare payments in the inpatient model.

For the inpatient model, Mississippi (MS), Georgia (GA), and Arkansas (AR) led to the lowest increase in average Medicare payments out of all significant variables, with Mississippi (MS) having the lowest beta coefficient, −488.24. Virginia (VA) had the second lowest beta coefficient and Tennessee (TN) had the third for the inpatient model, but they were not significant. For the outpatient model, the lowest beta coefficients for significant states were Tennessee (TN), West Virginia (WV), and South Carolina (SC). Mississippi had the lowest beta coefficient, but it was not significant. Georgia (GA), which had one of the smallest beta coefficients for the inpatient model, has the fifth smallest beta coefficient for the states that are significant in the outpatient model (

Table 5. Description of variables.

Medicare Payments	The average amount that Medicare pays to the provider.
Political Affiliation	A dummy variable that is 1 if the county voted Democrat in the 2016 presidential election, and 0 otherwise.
Total Hospital Medicare Discharges	The number of discharges billed by the provider for inpatient hospital services.
Percent 65 or Older in Poverty	The percentage of county population that is over 65 and living under the poverty line.
Estimated Population Served (thousands)	The estimated population that each hospital serves.
Median Income (thousands)	The median income measured in thousands.
State	The state the specific hospital is in.
APC Level	The level associated with a specific APC description.
Procedure	The procedure is associated with a specific APC description.
Percent White	The percentage of the county population that is White.
Percent Hispanic	The percentage of the county population that is Hispanic.
Percent Black	The percentage of the county population that is Black.
Percent Asian	The percentage of the county population that is Asian.
Percent Native American	The percentage of the county population that is Native American.
Percent Other Race	The percentage of other races not already specified in the county.

and

Table 6. Findings from model, secondary experiment.

	Coefficient Estimator	95% Confidence Interval
Total Hospital Medicare Discharges	−0.00 *	−0.01–−0.00
Estimated Population Served (thousands)	−0.06 ***	−0.07–−0.04
Percent 65 or Older in Poverty	18.97 ***	13.20–24.75
Percent White	−109.00	−391.35–173.35
Percent Hispanic	−109.93	−392.22–172.37
Percent Black	−109.38	−391.71–172.94
Percent Asian	−88.56	−370.78–193.66
Percent Native American	−109.24	−391.79–173.32
Percent Other Race	−125.62	−408.86–157.63
Political Affiliation	0.56	−41.23–42.35
Median Income (thousands)	17.16 ***	13.66–20.65
State [AZ]	1146.28 ***	989.48–1303.08
State [CA]	1766.29 ***	1621.02–1911.56
State [OR]	1140.10 ***	967.13–1313.06
State [VT]	791.11 ***	467.77–1114.46
State [WY]	957.33 ***	658.76–1255.90
APC [Level 2]	2787.55 ***	2734.46–2840.64
APC [Level 3]	5875.61 ***	5812.86–5938.35
APC [Level 4]	8720.85 ***	8651.71–8790.00
APC [Level 5]	9750.64 ***	9671.18–9830.10
APC [Level 6]	12,321.04 ***	12,175.55–12,466.53
APC [Level 7]	18,653.92 ***	18,435.57–18,872.28
Electrophysiologic Procedures	13,345.55 ***	13,229.93–13,461.18
ICD and Similar Procedures	29,409.21 ***	29,299.57–29,518.84
Neurostimulator and Related Procedures	15,416.44 ***	15,311.64–15,521.23

*** p-value < 0.001, * p-value < 0.05.

).

6. Discussion

In this study, we explored how different variables affect average Medicare payments. We were mainly interested in how the secondary diagnoses of CC and MCC affected average Medicare payments, and we were able to determine that both of these secondary diagnoses have a large, significant impact on average Medicare payments. We were also able to explore how a large variety of other variables affected average Medicare payments and make connections as to how and why these variables in our model affect average Medicare payments.

Since our inpatient data were for the year 2017, we were unable to use the most recent ICD classification codes for DRGs and resorted to using ICD-10 v37. ICD stands for “International Classification of Diseases” and is used to track public health conditions, improve data, measure outcomes and care of patients, make decisions, identify fraud, and design payment systems [32]. Some of our DRGs did not fall under a specific major diagnostic category (MDC), so they were kept as what they were originally called. These include heart transplant or the implant of a heart assist system, liver transplants, tracheostomy for face, mouth, and neck diagnoses or laryngectomy, autologous bone marrow transplant or T-cell immunotherapy, an extensive O.R procedure unrelated to the principal diagnoses, and non-extensive O.R procedures unrelated to the principal diagnoses. All other groups were categorized into their defined MDC, such as diseases and disorders of the eye, or diseases and disorders of the circulatory system. Each MDC contains multiple diagnoses, which are related to a single system of the body [33]. These groups allowed us to easily see how DRGs that were closely related to each other affected average Medicare payments.

We realize that there are several limitations to our study. For one, we realize that by joining multiple datasets together we lost some valuable information regarding the number of observations we can use. Although we were still able to evaluate our hypothesis on a sizable dataset of 174,323 observations, we lost about 20,000 data points from joining more datasets and cleaning the data. We also acknowledge that the model we evaluated our hypotheses with only included 47 out of the 50 states. We dropped Maryland, Hawaii, and Alaska from our final model. This is because Maryland uses an all-payer model instead of an IPPS system, and it is the only state to use an all-payer hospital rate regulation system [34]. Hawaii and Alaska were also dropped because states not connected to the mainland United States use a different methodology for their fee schedules [35]. We also used an estimator of the county population to mirror what we believe would be the total population that a specific hospital would serve. We understand that this is not necessarily the case and that people travel all over the country to receive medical treatment or certain procedures, but we believe that it is a good estimate of the everyday people who visit specific hospitals for treatment.

Our dataset of 174,323 observations, which uses data primarily from the CMS, includes 51,841 (29.74%) observations that had CC secondary diagnoses and 63,865 (36.64%) observations that had MCC secondary diagnoses. In 2017, there were 61,405,844 people enrolled in Medicare (Chronic Conditions Data Warehouse) and 11,252,922 inpatient claims (Chronic Conditions Data Warehouse). A study of 500 consecutive elderly patients admitted into a single medical facility showed that 29% of elderly patients had a complication because of their hospitalization [36]. Out of the 11,252,922 inpatient claims in 2017, this would result in an estimated 3,263,347 complications. The chronic and acute ambulatory care sensitive conditions (ACSCs) rates for Medicare are metrics used to measure hospital stays due to complications of ACSCs that are potentially preventable. The combined rate of the chronic and acute ACSCs rates was 38.2 per 1000 beneficiaries in 2018 [37]. Using this information, we can estimate that there were 124,659 preventable Medicare inpatient claims with a CC secondary diagnosis.

In 2006, a study was completed using California hospitalizations from the California Office of Statewide Health Planning and Development that showed 27.63 out of 1000 patients had a major potentially preventable complication [38]. This number of major potentially preventable complications includes all patients of all ages. We assume that older patients are more at risk of complications from medical care than younger patients. Therefore, the major complication number would be higher if it only included patients enrolled in Medicare [36]. However, we can still use this number as a low estimate for what we believe would be the total number of inpatient claims that included an MCC secondary diagnosis. From this study, we can estimate that there were 90,166 potentially preventable MCC secondary diagnoses.

Using these studies, we settled on using low estimates that 20% of the elderly population that visits a hospital for treatment will have a complication. We also used low estimates of 25 out of 1000 beneficiaries that have potentially preventable complications because of their hospital stay and 15 out of 1000 patients that have a potentially preventable major complication because of their hospital stay. By using lower estimates than what has been previously shown in other studies, we can find the lowest amount of potential savings Medicare could have if there was a shift in focus to preventing complications and comorbidities.

7. Conclusions

Our results provide a nuanced understanding of how various factors influence average Medicare payments, reflecting broader issues within the U.S. healthcare system. The significant impact of secondary diagnoses, such as complications and comorbidities (CCs) and major complications and comorbidities (MCCs), on Medicare payments underscores the complexity and resource intensity of treating patients with severe health conditions. This finding aligns with the current emphasis on value-based care in the U.S. healthcare system, which aims to optimize resource allocation and improve patient outcomes while controlling costs. However, the significant cost associated with MCC diagnoses raises concerns about the financial sustainability of Medicare, especially as the population ages and the prevalence of chronic conditions increases.

The observed racial disparities in Medicare payments, where counties with higher percentages of Black, Hispanic, Asian, or other races receive higher payments, point to systemic issues in healthcare access and quality. These disparities may reflect differences in the prevalence of chronic conditions, access to high-quality care, and social determinants of health among different racial and ethnic groups. The U.S. healthcare system must continue to address these disparities through targeted interventions, such as culturally competent care, increased funding for community health programs, and policies aimed at reducing discrimination and bias in healthcare delivery. Addressing these disparities is crucial for the U.S. healthcare system, as it seeks to provide equitable care and reduce overall healthcare costs. Policies aimed at improving access to preventive care, enhancing social services, and addressing the root causes of poverty could mitigate these disparities and reduce the financial burden on Medicare.

State-level disparities in Medicare payments reveal significant regional differences in healthcare costs and practices. States like Oregon and Wyoming, which showed higher Medicare payments in both inpatient and outpatient models, may have higher costs of living or different healthcare infrastructure and practices compared to states like Mississippi and Georgia, which had the lowest increases in payments. These regional differences highlight the complexity of implementing national healthcare policies in a diverse country and the need for state-specific strategies to address unique healthcare challenges.

In summary, we were able to determine that CC and MCC secondary diagnoses have a significant effect on average Medicare payments. We concluded that Medicare could have potentially saved about USD 86,000,000 from preventable CCs and USD 259,000,000 from preventable MCCs. In total, it is possible that Medicare spent an extra USD 345,000,000 on CC and MCC secondary diagnoses that could have been prevented with higher-quality and more timely care. Based on these findings, increasing the quality of care that Medicare patients receive can not only help to prevent health issues in the elderly but also save Medicare a considerable amount of money. We hope our study can lead to more research on how to improve medical care, especially related to Medicare, to reduce the number of complications and comorbidities to improve health outcomes for elderly patients as well as reduce cost.