Introduction

For my final project, the question I would like to answer if a small number of high-cost drugs account for a disproportionately large share of total Medicare drug spending. I will use several different methods to try to reach the answer to this question.

Load packages

library(dplyr)
## Warning: package 'dplyr' was built under R version 4.4.2
## 
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
## 
##     filter, lag
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, setequal, union
library(ggplot2)
## Warning: package 'ggplot2' was built under R version 4.4.2
library(stringr)
library(tidyr)

Import dataframe

df <- "https://raw.githubusercontent.com/j-song-npc/Data607-Final-Project-/refs/heads/main/Medicare_Part_D_Spending_by_Drug_2022.csv" 
medicare <- read.csv(df)

Clean and filter data

The observations are broken out by manufacturer but I think it will be easier to look at the overall drug instead of broken out for this analysis.

# Filter and reorder dataset
medicare_overall <- medicare %>%
  filter(Mftr_Name == "Overall") %>%
  arrange(desc(Tot_Spndng_2022))

# Long table for years 
medicare_long <- medicare_overall %>%
  pivot_longer(cols = starts_with("Tot_Spndng_"), 
               names_to = "year", 
               values_to = "total_spending") %>%
  mutate(year = gsub("Tot_Spndng_", "", year),
         year = as.integer(year))

Summary stats and ratios

medicare_overall$cum_spnd_2022 <- cumsum(medicare_overall$Tot_Spndng_2022)
Total_Spend <- sum(medicare_overall$Tot_Spndng_2022, na.rm =TRUE)
Total_Benes <- sum(medicare_overall$Tot_Benes_2022, na.rm = TRUE)

medicare_overall$ratio_total_spending <- (medicare_overall$cum_spnd_2022/Total_Spend) * 100 
medicare_overall$ratio_drugs <- (1:nrow(medicare_overall) / nrow(medicare_overall)) *100
medicare_overall$ratio_benes <- (medicare_overall$Tot_Benes_2022/Total_Benes)* 100

Spending concentration

drugs80 <- min(which(medicare_overall$ratio_total_spending >= 80))
percent_drugs80 <- (drugs80 / nrow(medicare_overall)) * 100

head(percent_drugs80)
## [1] 7.272727

This tells us that 7.3% of drugs account for 80% of total spending. However, some drugs may have higher utilization though lower cost, while other drugs are higher cost with lower utilization.

Visualize spending

# Arrange data by descending spending
medicare_overall_sorted <- medicare_overall %>%
  arrange(desc(Tot_Spndng_2022)) %>%
  mutate(
    drug_rank = row_number(),
    pct_drugs = drug_rank / n(),
    pct_spending = cumsum(Tot_Spndng_2022) / sum(Tot_Spndng_2022))

# Plot cumulative spending curve
ggplot(medicare_overall_sorted, aes(x = pct_drugs, y = pct_spending)) +
  geom_line() +
  geom_vline(xintercept = 0.073, linetype = "dashed") +
  geom_hline(yintercept = 0.8, linetype = "dashed") +
  labs(
    title = "Cumulative Drug Spending Distribution",
    subtitle = "7.3% of drugs account for 80% of Medicare drug spending",
    x = "Cumulative % of Drugs",
    y = "Cumulative % of Total Spending")

Top 10% high-spend drugs

0.17% of medicare beneficiaries are using the top 10% costliest drugs.

# Top 10% by total spending 
top10_count <- ceiling(0.1 * nrow(medicare_overall))
top10_spending <- sum(medicare_overall$Tot_Spndng_2022[1:top10_count])
top10_benes <- sum(medicare_overall$Tot_Benes_2022[1:top10_count])
percent_spending_top10 <- ((top10_spending / Total_Spend) * 100)
percent_benes_top10 <- (top10_benes / Total_Benes) * 100

head(top10_count)
## [1] 358
head(top10_spending)
## [1] 205013579285
head(top10_benes)
## [1] 239783940
head(percent_spending_top10)
## [1] 85.26892
head(percent_benes_top10)
## [1] 53.69217
# Top 10% by average spending per beneficiary 
top10_threshold <- quantile(medicare_overall$Avg_Spnd_Per_Bene_2022, 0.9, na.rm = TRUE)
high_cost_drugs <- medicare_overall %>%
  filter(Avg_Spnd_Per_Bene_2022 >= top10_threshold)

top10_total_benes <- sum(high_cost_drugs$Tot_Benes_2022, na.rm = TRUE)
percent_benes_top10_avgcost <- (top10_total_benes / Total_Benes) * 100

head(percent_benes_top10_avgcost)
## [1] 0.1736131

Utilization categories

# Set thresholds to categorize each drug
high_cost <- quantile(medicare_overall$Avg_Spnd_Per_Bene_2022, 0.9, na.rm = TRUE)
low_cost <- quantile(medicare_overall$Avg_Spnd_Per_Bene_2022, 0.1, na.rm = TRUE)
high_util <- quantile(medicare_overall$Tot_Benes_2022, 0.9, na.rm = TRUE)
low_util <- quantile(medicare_overall$Tot_Benes_2022, 0.1, na.rm = TRUE)

# Categorize drugs
medicare_overall <- medicare_overall %>%
  mutate(
    cost_category = case_when(
      Avg_Spnd_Per_Bene_2022 >= high_cost ~ "High Cost",
      Avg_Spnd_Per_Bene_2022 <= low_cost ~ "Low Cost",
      TRUE ~ "Medium Cost"),
    util_category = case_when(
      Tot_Benes_2022 >= high_util ~ "High Util",
      Tot_Benes_2022 <= low_util ~ "Low Util",
      TRUE ~ "Medium Util"),
    cost_util_group = paste(cost_category, util_category, sep = " / "))

# Count drugs in each category
group_counts <- medicare_overall %>%
  count(cost_util_group) %>%
  arrange(desc(n))

print(group_counts)
##             cost_util_group    n
## 1 Medium Cost / Medium Util 2408
## 2   High Cost / Medium Util  294
## 3    Medium Cost / Low Util  261
## 4   Medium Cost / High Util  228
## 5    Low Cost / Medium Util  189
## 6      Low Cost / High Util  111
## 7      High Cost / Low Util   45
## 8       Low Cost / Low Util   39

Plot of utilization categories

ggplot(medicare_overall, aes(x = Tot_Benes_2022, y = Avg_Spnd_Per_Bene_2022, color = cost_util_group)) +
  geom_point(alpha = 0.7) +
  scale_y_log10() + 
  scale_x_log10() +
  labs(
    title = "Drug Cost vs. Utilization",
    x = "Number of Beneficiaries (log scale)",
    y = "Average Spending per Beneficiary (log scale)",
    color = "Category"
  ) +
  theme_minimal()
## Warning: Removed 189 rows containing missing values or values outside the scale range
## (`geom_point()`).

Conclusion

This analysis revealed that a small number of drugs account for the majority of Medicare drug spending, a finding I explored using several statistical techniques in R that were new to me. The project not only reinforced my assumptions about spending concentration but also expanded my ability to manipulate and visualize data.

One limitation is that the dataset does not account for patients taking multiple medications, which could affect interpretations of utilization. For future analysis, it may be valuable to incorporate information on drug classifications or disease categories to better understand the types of drugs driving spending. I would also be interested in analyzing trends over time using the 2018–2022 data and further exploring the components of prescription claim costs.