HCRIS Data

Ian McCarthy | Emory University

Outline for Today

Let’s work with HCRIS using our HCRIS GitHub repository.

1. Data Overview
2. Measuring Prices
3. Measuring Penalties
4. Other key variables

Understanding HCRIS Data

What is HCRIS?

Healthcare Cost Report Information System (‘cost reports’)

• Nursing Homes (SNFs)
• Hospice
• Home Health Agencies
• Hospitals

These reports contain a TON of information about hospital investments, cost structure, etc.

The Raw Data

• Two versions (1996 and 2010)
• Structure (alphanumeric, numeric, report info)
• Extracting variables requires first locating in the report templates

The Raw Data

• Just opening the raw data won’t tell you much
• Need to identify location of cell of interest and merge with report “_RPT” on report record number
• Let’s spend some time reviewing things in the HCRIS GitHub repository.

Some Key Considerations

• Reports submitted by fiscal year, which may not align with other datasets
• Some hospitals submit multiple reports for the same fiscal year
  • Repo code tries to deal with this in a few ways
  • Goal is to end with a single report per hospital (i.e., Medicare provider number) per fiscal year

HCRIS for Emory Hospital

• R
• Python
• Output

hcris.emory <- read_csv("../data/output/hcris-snippets/hcris-emory.csv")

import pandas as pd

hcris_emory = pd.read_csv("../data/output/hcris-snippets/hcris-emory.csv")

# A tibble: 6 × 17
  provider_number fy_end     name     beds tot_charges net_pat_rev tot_discounts
            <dbl> <date>     <chr>   <dbl>       <dbl>       <dbl>         <dbl>
1          110010 1999-08-31 EMORY …   388   482698158   320561077     162137081
2          110010 2000-08-31 EMORY …   397   576571778   344619715     231952063
3          110010 2001-08-31 EMORY …   423   602064553   360298092     241766461
4          110010 2002-08-31 EMORY …   429   669164752   427398291     241766461
5          110010 2003-08-31 EMORY …   429   760913657   410665393     350248264
6          110010 2004-08-31 EMORY …   424   898891053   470083388     428807665
# ℹ 10 more variables: tot_operating_exp <dbl>, ip_charges <dbl>,
#   icu_charges <dbl>, ancillary_charges <dbl>, tot_discharges <dbl>,
#   mcare_discharges <dbl>, mcaid_discharges <dbl>, tot_mcare_payment <dbl>,
#   hrrp_payment <dbl>, year <dbl>

Emory Net Patient Revenue and Medicare Discharges

• R
• Python
• Output

plot.dat <- hcris.emory %>% 
  group_by(year) %>% 
  summarize(net_rev=mean(net_pat_rev, na.rm=TRUE)/1000000, 
            mcare=mean(mcare_discharges,  na.rm=TRUE)) %>%
  ungroup()

rev.plot <- plot.dat %>%
  ggplot(aes(x=as.factor(year), y=net_rev)) + 
  geom_line(linewidth = 1) +
  labs(
    x = "Year",
    y = "$ in millions",
    title = "Mean Patient Revenue Over Time"
  ) +
  theme_bw() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))  
  
mcare.plot <- plot.dat %>%
  ggplot(aes(x=as.factor(year), y=mcare)) + 
  geom_line(linewidth = 1) +
  labs(
    x = "Year",
    y = "Discharges",
    title = "Mean Medicare Discharges Over Time"
  ) +
  theme_bw() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))  

import pandas as pd
import matplotlib.pyplot as plt

# Aggregate to year means
plot_dat = (
    hcris_emory
    .groupby("year", as_index=False)
    .agg(
        net_rev=("net_pat_rev", lambda x: x.mean(skipna=True) / 1_000_000),
        mcare=("mcare_discharges", "mean")
    )
)

plt.figure()
plt.plot(plot_dat["year"], plot_dat["net_rev"], linewidth=1)
plt.xlabel("Year")
plt.ylabel("$ in millions")
plt.title("Mean Patient Revenue Over Time")
plt.xticks(rotation=90)
plt.tight_layout()
plt.show()

plt.figure()
plt.plot(plot_dat["year"], plot_dat["mcare"], linewidth=1)
plt.xlabel("Year")
plt.ylabel("Discharges")
plt.title("Mean Medicare Discharges Over Time")
plt.xticks(rotation=90)
plt.tight_layout()
plt.show()

HCRIS for All Hospitals

• R
• Python
• Output

hcris.data <- read_csv("../data/output/hcris-snippets/hcris-data.csv")
hosp.count.plot <- hcris.data %>% 
  ggplot(aes(x=as.factor(year))) + 
  geom_bar() +
  labs(
    x="Year",
    y="Number of Hospitals",
    title="Number of Hospitals per Year"
  ) + theme_bw() +
  theme(axis.text.x = element_text(angle = 90, hjust=1))

import pandas as pd
import matplotlib.pyplot as plt

hcris_data = pd.read_csv("../data/output/hcris-snippets/hcris-data.csv")

fig, ax = plt.subplots()

ax.hist(hcris_data["year"], bins=len(hcris_data["year"].unique()))
ax.set_xlabel("Year")
ax.set_ylabel("Number of Hospitals")
ax.set_title("Number of Hospitals per Year")
ax.tick_params(axis="x", rotation=90)

plt.tight_layout()
plt.show()

Measuring Prices

Estimating hospital prices

• Negotiated payment rates not easily accessible in public data (if at all)
• Must estimate prices from revenue and costs in HCRIS

\[ \text{Price} = \frac{\text{Charges}*(\text{Discount Factor}) - (\text{Medicare Payments})}{(\text{Total Discharges}) - (\text{Medicare Discharges})}\]

• Approximates average payment from commercial insurers (except for Medicaid)

Emory Prices in HCRIS

• R
• Python
• Output

emory.price <- hcris.emory %>%
  mutate( discount_factor = 1-tot_discounts/tot_charges,
          price_num = (ip_charges + icu_charges + ancillary_charges)*discount_factor - tot_mcare_payment,
          price_denom = tot_discharges - mcare_discharges,
          price = price_num/price_denom) %>%
  select(provider_number, year, price)

emory.price.plot <- emory.price %>%
  group_by(year) %>%
  summarize(mean_price=mean(price, na.rm=TRUE)) %>%
  ungroup() %>%
  ggplot(aes(x=as.factor(year), y=mean_price)) + 
  geom_line(linewidth = 1) +
  labs(
    x = "Year",
    y = "Dollars",
    title = "Mean Price for Emory Hospitals (EUH and Midtown)"
  ) +
  theme_bw() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))  

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

emory_price = (
    hcris_emory
    .assign(
        discount_factor = 1 - hcris_emory["tot_discounts"] / hcris_emory["tot_charges"],
        price_num = (
            hcris_emory["ip_charges"]
            + hcris_emory["icu_charges"]
            + hcris_emory["ancillary_charges"]
        ) * (1 - hcris_emory["tot_discounts"] / hcris_emory["tot_charges"])
        - hcris_emory["tot_mcare_payment"],
        price_denom = hcris_emory["tot_discharges"] - hcris_emory["mcare_discharges"]
    )
    .assign(
        price = lambda d: d["price_num"] / d["price_denom"]
    )
    .loc[:, ["provider_number", "year", "price"]]
)

price_year = (
    emory_price
    .replace([np.inf, -np.inf], np.nan)
    .groupby("year", as_index=False)
    .agg(mean_price=("price", "mean"))
)

fig, ax = plt.subplots()

ax.plot(price_year["year"], price_year["mean_price"], linewidth=1)
ax.set_xlabel("Year")
ax.set_ylabel("Dollars")
ax.set_title("Mean Price for Emory Hospitals (EUH and Midtown)")
ax.tick_params(axis="x", rotation=90)

plt.tight_layout()
plt.show()

Prices in the Full HRIS Data

• R
• Python
• Output

price.plot <- hcris.data %>% group_by(year) %>% 
  summarize(mean_price=mean(price, na.rm=TRUE)) %>%
  ggplot(aes(x=as.factor(year), y=mean_price)) + 
  geom_line(aes(group=1)) +
  labs(
    x="Year",
    y="Average Hospital Price",
    title="Hospital Prices per Year"
  ) + scale_y_continuous(labels=comma) +
  theme_bw() + theme(axis.text.x = element_text(angle = 90, hjust=1))

import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.ticker import FuncFormatter
import numpy as np

price_year = (
    hcris_data
    .groupby("year", as_index=False)
    .agg(mean_price=("price", "mean"))
)

comma_fmt = FuncFormatter(lambda x, pos: f"{int(x):,}")

fig, ax = plt.subplots()

ax.plot(price_year["year"], price_year["mean_price"], linewidth=1)
ax.set_xlabel("Year")
ax.set_ylabel("Average Hospital Price")
ax.set_title("Hospital Prices per Year")

ax.yaxis.set_major_formatter(comma_fmt)
ax.tick_params(axis="x", rotation=90)

plt.tight_layout()
plt.show()

Removing very high prices

• R
• Python
• Output

price.plot2 <- hcris.data %>% 
  filter(price>0) %>%
  group_by(year) %>% 
  mutate(
    p95 = quantile(price, 0.95, na.rm = TRUE),
    p05 = quantile(price, 0.05, na.rm = TRUE),
    price = pmin(pmax(price, p05), p95)
  ) %>% 
  summarize(mean_price=mean(price, na.rm=TRUE)) %>%
  ggplot(aes(x=as.factor(year), y=mean_price)) + 
  geom_line(aes(group=1)) +
  labs(
    x="Year",
    y="Average Hospital Price",
    title="Hospital Prices per Year"
  ) + scale_y_continuous(labels=comma) +
  theme_bw() + theme(axis.text.x = element_text(angle = 90, hjust=1))

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.ticker import FuncFormatter

# comma formatter (ggplot::comma analogue)
comma_fmt = FuncFormatter(lambda x, pos: f"{int(x):,}")

# filter, winsorize within-year, then average within-year
df2 = hcris_data.loc[hcris_data["price"] > 0, ["year", "price"]].copy()

def winsorize_group(g):
    p95 = g["price"].quantile(0.95)
    p05 = g["price"].quantile(0.05)
    g["price"] = g["price"].clip(lower=p05, upper=p95)
    return g

price_year2 = (
    df2.groupby("year", group_keys=False)
       .apply(winsorize_group)
       .groupby("year", as_index=False)
       .agg(mean_price=("price", "mean"))
)

# plot
fig, ax = plt.subplots()
ax.plot(price_year2["year"].astype(str), price_year2["mean_price"], linewidth=1)  # factor(year) analogue
ax.set_xlabel("Year")
ax.set_ylabel("Average Hospital Price")
ax.set_title("Hospital Prices per Year")
ax.yaxis.set_major_formatter(comma_fmt)
ax.tick_params(axis="x", rotation=90)

plt.tight_layout()
plt.show()

HRRP Penalties

HRRP in HCRIS

• Variable recorded in our data as hrrp_payment
• Recall HRRP is entirely punitive, so hrrp_payment should always be the same sign

Share of Penalized Hospitals

• R
• Python
• Output

share.hrrp <- hcris.data %>% 
  mutate(penalized=if_else(hrrp_payment>0 & !is.na(hrrp_payment), 1, 0)) %>%
  group_by(year) %>% 
  summarize(share_hrrp=mean(penalized, na.rm=TRUE)) %>%
  ggplot(aes(x=as.factor(year), y=share_hrrp)) + 
  geom_line(aes(group=1)) +
  labs(
    x="Year",
    y="Share of Hospitals",
    title="Penalized Hospitals by Year"
  ) + scale_y_continuous(labels=comma) +
  theme_bw() + theme(axis.text.x = element_text(angle = 90, hjust=1))

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.ticker import FuncFormatter

share_year = (
    hcris_data
    .assign(
        penalized = np.where(
            (hcris_data["hrrp_payment"] > 0) & (~hcris_data["hrrp_payment"].isna()),
            1, 0
        )
    )
    .groupby("year", as_index=False)
    .agg(share_hrrp=("penalized", "mean"))
)

fig, ax = plt.subplots()

ax.plot(share_year["year"].astype(str), share_year["share_hrrp"], linewidth=1)
ax.set_xlabel("Year")
ax.set_ylabel("Share of Hospitals")
ax.set_title("Penalized Hospitals by Year")

# keep comma formatter to mirror ggplot call (though shares typically don't need it)
comma_fmt = FuncFormatter(lambda x, pos: f"{x:,.2f}")
ax.yaxis.set_major_formatter(comma_fmt)

ax.tick_params(axis="x", rotation=90)

plt.tight_layout()
plt.show()

Amount of Penalty

• R
• Python
• Output

hrrp.stats <- hcris.data %>% filter(!is.na(hrrp_payment), hrrp_payment!=0) %>%
  mutate(hrrp_percent=hrrp_payment/tot_mcare_payment) %>%
  group_by(year) %>% 
  summarize(mean=mean(hrrp_payment), 
            max=max(hrrp_payment),
            mean_percent=mean(hrrp_percent, na.rm=TRUE),
            max_percent=max(hrrp_percent, na.rm=TRUE))

import pandas as pd
import numpy as np


hrrp_stats = (
    hcris_data
    .loc[
        (~hcris_data["hrrp_payment"].isna()) &
        (hcris_data["hrrp_payment"] != 0)
    ]
    .assign(
        hrrp_percent=lambda d: d["hrrp_payment"] / d["tot_mcare_payment"]
    )
    .groupby("year", as_index=False)
    .agg(
        mean=("hrrp_payment", "mean"),
        max=("hrrp_payment", "max"),
        mean_percent=("hrrp_percent", "mean"),
        max_percent=("hrrp_percent", "max")
    )
)

# A tibble: 4 × 5
   year    mean     max mean_percent max_percent
  <dbl>   <dbl>   <dbl>        <dbl>       <dbl>
1  2012  33177.  479553     0.000802      0.0209
2  2013  99257. 2259874     0.00262       0.0142
3  2014  94899. 2490722     0.00271       0.0187
4  2015 145433. 3286700     0.00430       0.0278

Other Key Variables

HRRP Details

• Hospitals penalized as a percentage of base Medicare DRG payments
• More Medicare patients implies more “exposure” to the program
• Two hospitals with different amounts of Medicare patients (but otherwise identical) will have different penalties

Medicare Patients

• R
• Python
• Output

mcare.hrrp <- hcris.data %>% 
  filter(year %in% c(2011, 2013)) %>%
  group_by(provider_number) %>%
  summarize(
    mcare_2011    = mcare_discharges[year == 2011][1]/1000,
    hrrp_pay_2013 = hrrp_payment[year == 2013][1]/1000000,
    .groups = "drop"
  ) %>%
  drop_na(mcare_2011, hrrp_pay_2013)  

mcare.hrrp.plot <- ggplot(mcare.hrrp, aes(x = mcare_2011, y = hrrp_pay_2013)) +
  geom_point(alpha = 0.6) +
  labs(
    x = "Medicare Discharges (2011) in thousands",
    y = "HRRP Penalty (2013) in millions $",
    title = "HRRP Penalties vs Pre-Period Medicare Volume"
  ) +
  theme_bw()

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

mcare_hrrp = (
    hcris_data
    .loc[hcris_data["year"].isin([2011, 2013])]
    .groupby("provider_number", as_index=False)
    .agg(
        mcare_2011=("mcare_discharges",
                    lambda x: x[hcris_data.loc[x.index, "year"] == 2011].iloc[0] / 1_000),
        hrrp_pay_2013=("hrrp_payment",
                       lambda x: x[hcris_data.loc[x.index, "year"] == 2013].iloc[0] / 1_000_000)
    )
    .dropna(subset=["mcare_2011", "hrrp_pay_2013"])
)

fig, ax = plt.subplots()

ax.scatter(
    mcare_hrrp["mcare_2011"],
    mcare_hrrp["hrrp_pay_2013"],
    alpha=0.6
)

ax.set_xlabel("Medicare Discharges (2011) in thousands")
ax.set_ylabel("HRRP Penalty (2013) in millions $")
ax.set_title("HRRP Penalties vs Pre-Period Medicare Volume")

plt.tight_layout()
plt.show()