On their website, the Centers for Medicare & Medicaid Services (CMS) have released datasets for 2011 through 2014 which show the average value of hospital-billed charges and Medicare reimbursement rates on a per-hospital basis (for the 100 most common diagnosis related groups (DRGs)). Hospital-billed charges are the amount of money charged by a hospital for performing a certain procedure, whereas the Medicare reimbursement rates are the prices that the government has agreed to pay for those same procedures.

If Medicare payments are below the amount needed by hospitals to recoup their operating costs, as the American Hospital Association suggests is the case, the burden of paying higher premiums may be shifted over to those on private insurance plans. According to the AHA, about 64.8% of hospitals are losing money because of costs of procedures for Medicare patients.

Hospital charges are usually not immediately consequential for determining the amount of money that will be paid by a public or private health insurance plan for a healthcare procedure. However, uninsured patients sometimes still have to pay the full hospital charges and out-of-network patients often have to pay some part of the full charges.

Overall, the high markups in hospital charge prices make it very difficult for patients to gauge relationships between price and quality in healthcare providers, and for anyone outside of a hospital system to gauge the financial health of that system (in particular, whether or not the system is able to obtain positive net earnings).

Below, I walk through my steps for exploratory analysis of some aspects of the 2014 hospital charge dataset, paying particular attention to the light this dataset sheds on differences in healthcare market transparency across the country.

Table of Contents

• Preparing for Analysis
• Loading and Cleaning the Dataset
• Most Common DRGs
• DRG 470 Charges by State
  • Hospital Charges and Medicare Payments in California
  • Hospital Charges and Medicare Payments in Maryland
• Conclusion

First, we'll load a few Python libraries to aid with analyzing and visualizing this dataset. In particular, the Pandas library will allow us to load CSV data, change how it is grouped, and quickly sort and splice the dataset. The Seaborn library will allow us to create some more complex visualizations of this multivariate data and gauge how different charges relate to each other.

#Render out plots inline
%matplotlib inline

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np

plt.style.use('default') # Make the graphs a bit prettier
plt.rcParams['figure.figsize'] = (15, 5)

We will need to use Pandas to load the CSV dataset. The dataset includes the following fields:

• DRG Definition: A description of one of the 100 most common diagnostic resource groups
• Provider Id: The numerical identifier for a specific hospital
• Provider Name: The name of the same hospital
• Provider Street Address: The street address of the same hospital
• Provider City: The city where this specific hospital is located
• Provider State: The state where this specific hospital is located
• Provider Zip Code: The zip code where this specific hospital is located
• Hospital Referral Region (HRR) Description: The Hospital Referral Region (regional health care market for tertiary medical care) where this specific hospital is located.
• Total Discharges: The total number of patients who were released from this specific inpatient hospital after receiving care related to this specific DRG.
• Average Covered Charges: The amount of money charged by the hospital for this medical procedure (how much the provider charges to Medicare).
• Average Total Payments: The amount of money received by the hospital for this medical procedure (This includes the reimbursement from Medicare, as well as third party payments for coordinating benefits, and any co-payment and deductible amounts that the patient is responsible for).
• Average Medicare Payments: The amount of money that Medicare has reimbursed to the hospital for this medical procedure.

In order to be able to chart and compare the various charge and payment variables in this dataset, we will need to remove the dollar symbol ('$') from the Average Covered Charges, Average Total Payments, and Average Medicare Payments columns. Then, we will need to convert the variable type for these columns to a 'float' type.

health_charges = pd.read_csv('Medicare_Provider_Charge_Inpatient_DRGALL_FY2014.csv')
health_charges['Average Covered Charges'] = health_charges['Average Covered Charges'].replace('[\$,]','',regex=True).astype(float)
health_charges['Average Total Payments'] = health_charges['Average Total Payments'].replace('[\$,]','',regex=True).astype(float)
health_charges['Average Medicare Payments'] = health_charges['Average Medicare Payments'].replace('[\$,]','',regex=True).astype(float)

Let's take a look at the first few rows of our dataset to ensure it has loaded properly.

health_charges[:5]

Initially, let's identify which DRGs were most frequently invoked to treat patients under Medicare. We can do this by grouping our dataset by DRG code, aggregating by the sum of total discharges at each hospital under those DRGs, and then sorting in an descending order.

# We want to find out which DRG codes are most common across the country, so let's create a new column to store the total discharges for a DRG code
DRGgrouped = health_charges.groupby('DRG Definition')
DRGgrouped.agg({'Total Discharges': np.sum}).sort_values(by=['Total Discharges'], ascending=0)[:10]

Since we know that DRG 470 (Joint Replacement/Lower Extremity Reattachment) is the most frequently reimbursed procedure for Medicare, let's take a look at how hospital charges for this DRG differ across states. We can do this by filtering our table by DRG code, and then creating a Seaborn stripplot which compares Average Covered Charges by Provider State.

# Let's take a closer look at how charges for the most common DRG differ across states
drg470 = health_charges[health_charges["DRG Definition"].str.contains("470")]
sns.stripplot(x="Provider State", y="Average Covered Charges", data=drg470)

<matplotlib.axes._subplots.AxesSubplot at 0x1f0f62550>

This chart shows that hospital charges in this DRG group vary greatly across the United States, and often within certain states. In particular, California, Florida, Texas, New Jersey and Pennsylvania seem to have a greater variety in hospital charge amounts for this particular procedure within their states, including charges far beyond those in many other states.

Let's dive deeper into the charge data for one of these states with more variety in hospital charge amounts for DRG 470. We can use 'groupby' on our dataset to allow for easy extraction of California data. Then, we can use a 'pairplot' to plot the pairwise relationship between Average Medicare Payments and Average Covered Charges in California. The scatterplots show the joint relationships between Average Medicare Payments and Average Covered Charges, whereas the histograms show the univariate distributions of each.

drg470_state_grouped = drg470.groupby(["Provider State"])
sns.pairplot(drg470_state_grouped.get_group("CA"), size = 10, vars=["Average Covered Charges", "Average Medicare Payments"])

<seaborn.axisgrid.PairGrid at 0x1f0eea9e8>

What we see is a situation where Medicare payment amounts for these procedures are most often between $10,000 and $20,000, whereas hospital charges tend to be most often between $50,000 and $150,000. The result of these marked up hospital charges is shown in the heavily skewed clusters on the scatter plots above. DRG 470 procedures appear to be an example of the lack of price transparency that healthcare consumers or industry analysts would find in the California marketplace, and many other healthcare markets across the country.

The stripplot we created a bit above showing hospital charges by state for DRG 470 also included some locations that had much less difference in statewide hospital charges than California. One of the lowest and most evenly distributed collection of hospital charges appeared to be in Maryland. We can use the same 'groupby' dataset we created above to take a look at the 'pairplot' for the relationship between Average Medicare Payments and Average Covered Charges for DRG 470 in Maryland.

sns.pairplot(drg470_state_grouped.get_group("MD"), size = 10, vars=["Average Covered Charges", "Average Medicare Payments"])

<seaborn.axisgrid.PairGrid at 0x1ee905128>

These charts appear to depict a very different relationship from the one visible for California's data. We see an almost one-to-one relationship between hospital charge amounts and average Medicare payments in Maryland.

It turns out that Maryland's data for hospital charges is quite unique in comparison to the rest of the country because Maryland is the only state which has a central government body that sets standardized payment rates for hospital procedures, across all public and private insurance plans.

This system was set up in 1971 in reaction to rising hospital costs and large losses incurred by treating uninsured patients. The centralized rate-setting system has created a transparent marketplace for consumers and kept growth of hospital costs below the national trend.

In the above analysis, we took a look at how the relationship between hospital charges and Medicare payments differed across the country in 2014, particularly for procedures related to the most common DRG (470 - Joint Replacement/Lower Extremity Reattachment). The result is a first-hand view of the distortions and lack of transparency for consumers in most of the U.S. healthcare market.

A study describes how, if Maryland had not adopted their unique system, "hospital spending would have been cumulatively $40 billion higher than what resulted under rate setting. On the other hand, had the nation’s costs grown at Maryland’s rate of growth, cumulative savings would have exceeded $1.8 trillion." Despite Maryland's innovation in rate-setting, it is still dealing with the issue of increasing patient volume in hospitals (particularly admissions and out-patient visits) - yet another challenge facing the entire country's entire healthcare marketplace.