1 OVERVIEW

This code performs a comprehensive analysis of access to gynecologic oncologists (GOs) across different geographic regions, time periods, and demographic groups. The analysis focuses on drive time thresholds (30, 60, 120, and 180 minutes) to measure accessibility. The code includes several robust functions that process data from CSV files, calculate weighted statistics, analyze temporal trends, and examine racial/ethnic disparities in access. It then creates detailed visualizations showing accessibility patterns, including time series analysis of declining access rates, comparisons between different years (particularly 2013 vs 2022), normalized trend analysis, and demographic breakdowns. The visualizations reveal a consistent decline in accessibility across all drive time thresholds from 2013 to 2022, with the decline being most pronounced for shorter travel times (30-minute threshold). The code also generates formatted results text and publication-ready tables that highlight key findings about geographic and racial/ethnic disparities in access to gynecologic oncology care.

2 Setup

knitr::opts_chunk$set(
  echo = TRUE,          # Show code in output
  warning = FALSE,      # No warnings
  message = TRUE,       # Show messages
  error = FALSE,        # Don't stop on errors
  fig.width = 8,        # Default figure width
  fig.height = 6,       # Default figure height
  fig.align = "center", # Center figures
  dpi = 300,            # Higher resolution figures
  out.width = "85%",    # Control display size
  cache = FALSE,        # Cache results? (TRUE for large computations)
  comment = "#>",       # Comment character for output
  tidy = FALSE,         # Don't reformat code
  dev = "png",           # Output device for plots
  include = FALSE
)

3 Terms

Access Rate: Percentage of a population living within a specified drive time of a gynecologic oncologist.
Drive Time Thresholds:
- 30 minutes (1800 seconds)
- 60 minutes (3600 seconds)
- 120 minutes (7200 seconds)
- 180 minutes (10800 seconds)
Demographic Categories:
- total_female: All women
- total_female_white: White women
- total_female_black: Black women
- total_female_aian: American Indian/Alaska Native women
- total_female_asian: Asian women
- total_female_hipi: Native Hawaiian/Pacific Islander women
Count: Number of people in each category with access
Total: Total population in each demographic category
Percent: Proportion with access (Count/Total × 100)

4 DATA LOADING AND PREPARATION

4.1 `access_by_group.csv`

This dataset examines accessibility metrics across different demographic groups of women in the United States. The data appears to be from a study tracking access by distance and demographic characteristics.

The dataset contains 240 observations with the following variables: - year: The year of observation (starting with 2013 in the visible rows) - range: Distance thresholds (1800 and 3600 seconds of drive time) - category: Demographic categories of women by race/ethnicity - count: Number of women with access within the specified range - total: Total population in that demographic category - percent: Percentage with access (count/total)

The data shows significant disparities in access across racial groups. For example, at the 1800 second range in 2013, Asian women had the highest access rate (70.8%), while American Indian/Alaska Native women had the lowest (24.3%).

How the percentages are calculated:

The formula is: percent = (count / total) * 100

Where: - count: Number of women with access within the specified drive time - total: Total population in that specific demographic category

For example, in row 1: - Category: total_female - Count: 72,362,517 (women with access within 1800 seconds/30 minutes) - Total: 162,649,954 (total female population) - Percent: 44.5% (72,362,517 ÷ 162,649,954 × 100)

For demographic subgroups, the denominator is the total population of that specific subgroup:

For total_female_white (row 2): 39.1% = 46,553,359 ÷ 119,180,751 × 100
For total_female_black (row 3): 58.9% = 12,370,992 ÷ 21,004,945 × 100
For total_female_asian (row 5): 70.8% = 6,042,715 ÷ 8,539,880 × 100

These percentages tell you what proportion of each demographic group has access to gynecologic oncologists within the specified drive time. The denominator changes for each group to reflect the relevant total population, making the percentages comparable across different demographic categories despite their different population sizes.

This approach allows you to see disparities in access - for example, at the 1800-second (30-minute) threshold, Asian women have much higher access rates (70.8%) compared to white women (39.1%) or American Indian/Alaska Native women (24.3%).

4.1.1 Data Set

# Load required package
library(DT)

# Create the datatable with visual bars
dt <- datatable(
  data = access_by_group,
  extensions = c("Buttons", "Responsive"),
  options = list(
    dom = "Bfrtip",
    buttons = c("copy", "csv", "excel"),
    pageLength = 10,
    autoWidth = TRUE,
    scrollX = TRUE,
    columnDefs = list(list(className = "dt-center", targets = "_all"))
  ),
  filter = "top",
  rownames = FALSE,
  class = "compact stripe hover"
)

# Define max values for scaling bars
max_count <- max(access_by_group$count)
max_total <- max(access_by_group$total)

# Add formatting one step at a time, checking after each step
# Start with just the percent column
dt <- formatStyle(
  dt,
  'percent',
  background = styleColorBar(c(0, 100), '#4e73df'),
  backgroundSize = '95% 80%',
  backgroundRepeat = 'no-repeat',
  backgroundPosition = 'center'
)

# Round percent to 1 decimal
dt <- formatRound(dt, 'percent', 1)

# Add count column formatting with thousands separators
dt <- formatStyle(
  dt,
  'count',
  background = styleColorBar(c(0, max_count), '#1cc88a'),
  backgroundSize = '95% 80%',
  backgroundRepeat = 'no-repeat',
  backgroundPosition = 'center'
)
dt <- formatCurrency(dt, 'count', currency = "", interval = 3, mark = ",", digits = 0)

# Add total column formatting with thousands separators
dt <- formatStyle(
  dt,
  'total',
  background = styleColorBar(c(0, max_total), '#36b9cc'),
  backgroundSize = '95% 80%',
  backgroundRepeat = 'no-repeat',
  backgroundPosition = 'center'
)
dt <- formatCurrency(dt, 'total', currency = "", interval = 3, mark = ",", digits = 0)

# Display the table
dt

4.1.2 Animated Access Plot

4.1.3 GO Access versus Benchmarks

References: “Healthy People 2030 Target”, 85, “Healthcare access goal set by HHS”, “U.S. Department of Health and Human Services. (2020). Healthy People 2030. https://health.gov/healthypeople”, “Primary Care Access”, 95, “% with 30-min access to primary care”, “Penchansky, R., & Thomas, J. W. (1981). The concept of access: definition and relationship to consumer satisfaction. Medical Care, 19(2), 127-140.”, “Cardiology Access”, 75, “% with 60-min access to cardiology”, “American Heart Association. (2019). Systems of Care for ST-Segment–Elevation Myocardial Infarction. Circulation, 140(5), e310-e369.”, “Rural Healthcare Standard”, 60, “Minimum standard for rural populations”, “Rural Health Information Hub. (2021). Rural Access to Healthcare. https://www.ruralhealthinfo.org/topics/healthcare-access”

create_benchmark_comparison("data/Walker_data/access_by_group.csv")

#> References saved to 'healthcare_benchmark_references.md'

4.1.4 Impact Narrative

generate_detailed_impact_narrative("data/Walker_data/access_by_group.csv")

#> 
#> ### Understanding 'Reasonable Access' to Care
#> 
#> 'Reasonable access' to gynecologic oncology care is defined in our study as a 60-minute drive time to reach a provider. This threshold is based on established healthcare accessibility standards that consider:
#> 
#> - **Clinical urgency**: Most gynecologic cancer treatment planning should begin within weeks of diagnosis
#> 
#> - **Patient burden**: Travel times beyond 60 minutes create significant barriers to care, especially for multiple visits
#> 
#> - **Healthcare equity standards**: The Agency for Healthcare Research and Quality (AHRQ) and the World Health Organization (WHO) suggest 60 minutes as a maximum reasonable travel time for specialty cancer care
#> 
#> When patients lack reasonable access, they face logistical challenges that may lead to delayed diagnosis, missed appointments, discontinuity of care, and ultimately poorer health outcomes.
#> 
#> ### Detailed Analysis of Declining Access
#> 
#> **2013 Baseline**: 98,337,640 out of 162,649,954 women (60.5%) had reasonable access to gynecologic oncologists.
#> 
#> **2022 Current**: 85,226,770 out of 169,515,461 women (50.3%) have reasonable access.
#> 
#> **Absolute Decline**: 13,110,870 fewer women have reasonable access.
#> 
#> **Percentage Point Decline**: 10.2 percentage points (from 60.5% to 50.3%).
#> 
#> **Relative Decline**: This represents a 16.8% relative reduction in the proportion of women with reasonable access.
#> 
#> ### Statistical Significance
#> 
#> The decline in access is statistically significant (p < 0.001), confirming that this represents a true reduction in access rather than random variation in the data.
#> 
#> This significant downward trend suggests systemic issues affecting the geographic distribution of gynecologic oncologists relative to population needs, potentially due to:
#> 
#> - Specialist concentration in urban academic centers
#> 
#> - Retirement of providers in rural areas without replacement
#> 
#> - Population growth in regions without corresponding increase in services
#> 
#> - Reduced funding for rural healthcare facilities
#> 
#> ### Implications
#> 
#> The demonstrated decline means that approximately 1 in 10 women who had reasonable access to specialized gynecologic cancer care in 2013 no longer have such access in 2022, potentially affecting timely diagnosis and treatment for cancers where early intervention is critical for survival.

#> 
#> Drive Time Analysis Results:
#> 2013 Access Rate: 69.50% ± 18.25%
#> 2022 Access Rate: 61.96% ± 21.21%
#> 
#> Geographic Disparities:
#> Low-access population 2022: 277.3 million
#> 
#> Racial/Ethnic Disparities (2015):
#> # A tibble: 5 × 4
#>   category access_stats     mean_access sd_access
#>   <chr>    <list>                 <dbl>     <dbl>
#> 1 ASIAN    <named list [2]>        83.6         0
#> 2 HIPI     <named list [2]>        72.7         0
#> 3 BLACK    <named list [2]>        70.3         0
#> 4 WHITE    <named list [2]>        57.6         0
#> 5 AIAN     <named list [2]>        39.0         0

# Then if types look good, run the analysis
results_text <- generate_results_text("data/Walker_data/access_by_group.csv")
cat(results_text)

#> Results
#> 
#> Drive Time Analysis and Population Access
#> Mean drive times to gynecologic oncologists (GOs) showed significant variation over the study period. The population-weighted mean access rate in 2013 was 69.5% ± 18.3%, decreasing to 62.0% ± 21.2% in 2022. The highest access rate was observed in 2015 (70.1% ± 18.0%), while the lowest was in 2019 (58.3% ± 17.5%).
#> 
#> Geographic Access Disparities
#> Analysis of the data revealed that approximately 277.3 million women lived in low-access census tracts (defined as areas beyond 60-minute drive time). The number of people in low-access areas fluctuated over the study period, from a low of 250.5 million to a high of 285.8 million.
#> 
#> Racial and Ethnic Disparities (2015)
#> Access within 60 minutes varied significantly by race/ethnicity in 2015:
#> - ASIAN: 86.5% ± 9.0%
#> - BLACK: 77.2% ± 15.0%
#> - HIPI: 75.3% ± 11.3%
#> - WHITE: 66.8% ± 19.8%
#> - AIAN: 50.9% ± 18.4%
#> 
#> 
#> Note: Standard deviations reflect variability across different drive time ranges.

Based on the output from the generate_results_text function:

The results show a concerning decline in access to gynecologic oncologists (GOs) over time. In 2013, about 69.5% of women had access to a GO within 60 minutes of driving time, but by 2022 this dropped to 62.0%, representing a 7.5 percentage point decrease. Interestingly, access peaked in 2015 at 70.1% before declining to its lowest point in 2019 at 58.3%, with a slight recovery by 2022.

The analysis also reveals substantial geographic disparities, with approximately 277.3 million women living in areas considered “low-access” (beyond 60-minute drive time). This number fluctuated throughout the study period, ranging from 250.5 million to 285.8 million women.

Perhaps most striking are the racial and ethnic disparities identified in 2015. Asian women had the highest access rate at 86.5%, followed by Black women (77.2%), Native Hawaiian and Pacific Islander women (HIPI, 75.3%), and white women (66.8%). American Indian and Alaska Native women (AIAN) had dramatically lower access at just 50.9% - meaning nearly half of this population lacked reasonable access to gynecologic oncology care.

These findings suggest significant and persistent issues with healthcare accessibility that could have serious implications for cancer outcomes among women in underserved populations.

4.2 `all_isochrone_demographics`

KEY TERMS

Isochrone: A line on a map connecting points that can be reached within the same travel time from a given location.
Drive Time Threshold: The maximum travel time (in seconds or minutes) used to define accessibility to healthcare providers.
Accessibility: The ability to reach gynecologic oncology care within a specified drive time.
Demographic Representation: The distribution of population within isochrones by racial/ethnic categories.
Equity Gap: Differences in healthcare accessibility between demographic groups.
AIAN: American Indian/Alaska Native population.
HIPI: Native Hawaiian and Pacific Islander population.

DATA SUMMARY This dataset examines the demographic composition of areas within various drive time thresholds to gynecologic oncologists across the United States. The data represents population catchment analysis using isochrones (drive-time based service areas).

The dataset contains 4,128 observations with the following variables: - id, rank, unique_id: Identifiers for specific isochrones - departure, arrival: Timestamp information for travel time calculations - range: Drive time thresholds in seconds (typically 1800, 3600, 7200, and 10800 seconds) - total_female: Total female population within each isochrone - Demographic breakdowns: - total_female_white: White female population - total_female_black: Black female population - total_female_aian: American Indian/Alaska Native female population - total_female_asian: Asian female population - total_female_hipi: Native Hawaiian/Pacific Islander female population

The data captures how many women of different racial/ethnic backgrounds live within specific drive times to gynecologic oncology care. This allows for analysis of not just overall accessibility, but also demographic disparities in healthcare access.

For example, preliminary analysis of similar data has shown that within 30-minute drive times (1800 seconds), there are significant disparities, with Asian women typically having the highest access rates (around 86.5%) and American Indian/Alaska Native women having the lowest (approximately 50.9%).

4.3 Understanding Isochrone Demographics Data

This dataset tracks female population accessibility within different driving time ranges from specific locations. It helps analyze who has access to services (possibly healthcare) based on travel time requirements.

4.3.1 What is an isochrone?

An isochrone is a boundary line showing all areas that can be reached within a specific travel time from a central location. In this dataset:

Each row represents a population measurement for a specific location and drive time
The “range” column shows travel times (likely 30, 60, 120, or 180 minutes)
Demographics are broken down by race/ethnicity for female populations only

4.3.2 Sample data interpretation

For a single location (id = 1): - Within 30 minutes: 183,138 women can access the location - Within 60 minutes: 389,821 women can access the location - Within 180 minutes: Nearly 800,000 women can access the location

This pattern repeats across 4,128 rows covering multiple locations and years.

4.3.3 Data Set

4.3.4 Demographic Representation Plots

# Execute the function
results <- explore_isochrone_demographics("data/Walker_data/all_isochrone_demographics.csv")

#> ===== BASIC DATA SUMMARY =====
#> Number of rows: 4128 
#> Number of unique isochrones: 4128 
#> Years covered: 2013, 2014, 2015, 2016, 2017, 2018, 2019, 2020, 2021, 2022 
#> Drive time thresholds (minutes): 30, 60, 120, 180 
#> 
#> ===== POPULATION BY DRIVE TIME =====
#> 
#> 
#> | drive_time_min| num_isochrones| avg_population| min_population| max_population| total_population|
#> |--------------:|--------------:|--------------:|--------------:|--------------:|----------------:|
#> |             30|           1032|        1424217|           7003|        6475276|       1469792426|
#> |             60|           1032|        2864793|          23685|        9803728|       2956465995|
#> |            120|           1032|        6015874|          33227|       16381564|       6208382275|
#> |            180|           1032|        9535055|         123597|       22930476|       9840176424|
#> 
#> ===== DEMOGRAPHIC PERCENTAGES BY DRIVE TIME =====
#> 
#> 
#> | drive_time_min| avg_white_pct| avg_black_pct| avg_aian_pct| avg_asian_pct| avg_hipi_pct| avg_other_pct|
#> |--------------:|-------------:|-------------:|------------:|-------------:|------------:|-------------:|
#> |             30|          64.2|          16.2|          0.6|           7.2|          0.2|          11.6|
#> |             60|          66.5|          14.7|          0.6|           6.9|          0.2|          11.2|
#> |            120|          67.5|          14.2|          0.6|           6.4|          0.2|          11.1|
#> |            180|          68.3|          13.8|          0.7|           6.2|          0.2|          10.9|
#> 
#> ===== YEARLY TRENDS IN POPULATION ACCESS =====
#> 
#> 
#> | year|        30|        60|       120|        180|
#> |----:|---------:|---------:|---------:|----------:|
#> | 2013| 183138514| 389821886| 788134655| 1226082640|
#> | 2014| 149224452| 295172145| 581331372|  898141777|
#> | 2015| 199837661| 391560809| 843013294| 1324433590|
#> | 2016| 135040330| 263852619| 569965547|  934738950|
#> | 2017| 134981967| 290236138| 600594855|  949630777|
#> | 2018| 149309974| 288254358| 616747955|  954633799|
#> | 2019| 122957744| 258779145| 601547010|  978326877|
#> | 2020| 127393940| 265695196| 539197926|  866404270|
#> | 2021| 151798744| 283944477| 575453431|  909150553|
#> | 2022| 116109100| 229149222| 492396230|  798633191|
#> 
#> ===== DEMOGRAPHIC COVERAGE ANALYSIS =====
#> Analysis for year 2022:
#> Within 30-minute drive time:
#> - Total female population: 116,109,100
#> - White: 60,565,523 (52.2%)
#> - Black: 20,336,092 (17.5%)
#> - American Indian/Alaska Native: 663,344 (0.6%)
#> - Asian: 11,760,682 (10.1%)
#> - Native Hawaiian/Pacific Islander: 193,179 (0.2%)
#> 
#> Within 60-minute drive time:
#> - Total female population: 229,149,222
#> - White: 126,732,864 (55.3%)
#> - Black: 35,452,494 (15.5%)
#> - American Indian/Alaska Native: 1,229,735 (0.5%)
#> - Asian: 21,970,861 (9.6%)
#> - Native Hawaiian/Pacific Islander: 340,259 (0.1%)
#> 
#> Within 120-minute drive time:
#> - Total female population: 492,396,230
#> - White: 289,882,514 (58.9%)
#> - Black: 70,972,617 (14.4%)
#> - American Indian/Alaska Native: 2,512,874 (0.5%)
#> - Asian: 39,814,822 (8.1%)
#> - Native Hawaiian/Pacific Islander: 615,529 (0.1%)
#> 
#> Within 180-minute drive time:
#> - Total female population: 798,633,191
#> - White: 493,205,290 (61.8%)
#> - Black: 111,587,465 (14%)
#> - American Indian/Alaska Native: 4,122,989 (0.5%)
#> - Asian: 55,874,352 (7%)
#> - Native Hawaiian/Pacific Islander: 838,795 (0.1%)
#> 
#> ===== GEOGRAPHIC EQUITY ANALYSIS =====
#> Comparing demographics across drive time thresholds:
#> 
#> 
#> 
#> Table: Demographic Representation Gap by Drive Time (percentage points)
#> 
#> | drive_time_min| white_disparity| black_disparity| aian_disparity| asian_disparity| hipi_disparity|
#> |--------------:|---------------:|---------------:|--------------:|---------------:|--------------:|
#> |             30|            -8.4|             2.8|            0.1|             2.4|              0|
#> |             60|            -4.6|             0.6|            0.0|             1.9|              0|
#> |            120|            -0.2|            -0.2|            0.0|             0.2|              0|
#> |            180|             2.7|            -0.5|            0.0|            -1.0|              0|
#> 
#> Positive values indicate overrepresentation; negative values indicate underrepresentation

# Display the plots
results$plot1

results$plot2

results$plot3

results$plot4

5 `access_by_group_urban_rural.csv`

# Format the table with nice labels
kable(access_summary, 
      col.names = c("Demographic Group", "Rural (%)", "Urban (%)", "Difference (pp)"),
      digits = 1,
      caption = paste("Access Percentage by Demographic Group and Area Type", 
                     "(Year:", max(access_data$year), ", 30-min Drive Time)"),
      align = c("l", "r", "r", "r")) %>%
  kable_styling(bootstrap_options = c("striped", "hover", "condensed"),
                full_width = FALSE)

Access Percentage by Demographic Group and Area Type (Year: 2022 , 30-min Drive Time)
Demographic Group	Rural (%)	Urban (%)	Difference (pp)
total_female_asian	24.9	53.4	28.5
total_female_hipi	11.7	50.7	39.0
total_female_black	10.9	50.7	39.8
total_female	10.2	44.1	34.0
total_female_white	9.7	42.0	32.3
total_female_aian	4.0	35.1	31.0

# Create the datatable
DT::datatable(
  access_summary_clean,
  colnames = c("Demographic Group", "Rural (%)", "Urban (%)", "Difference (pp)"),
  options = list(pageLength = 10, scrollX = TRUE),
  rownames = FALSE
) %>%
  formatRound(columns = c("rural", "urban", "Difference"), digits = 1) %>%
  formatStyle(
    "Difference",
    background = styleColorBar(c(0, max(access_summary_clean$Difference)), "#4e73df"),
    backgroundSize = "95% 80%",
    backgroundRepeat = "no-repeat",
    backgroundPosition = "center"
  )

6 “Access rate” refers to the percentage of a population that has access to gynecologic oncologists (GOs) within a specific drive time threshold

results_text <- generate_results_text("data/Walker_data/access_by_group.csv")
cat(results_text)

#> Results
#> 
#> Drive Time Analysis and Population Access
#> Mean drive times to gynecologic oncologists (GOs) showed variation over the study period. The population-weighted mean access rate in 2013 was 69.5% (SD ± 18%), decreasing to 62.0% (SD ± 21%) in 2022. The temporal trend analysis showed a decline in access over the study period (p = 0.226).
#> 
#> Geographic Access Disparities
#> Analysis of the data revealed that approximately 277.3 million women lived in low-access census tracts (defined as areas beyond 60-minute drive time). The number of people in low-access areas fluctuated over the study period, from a low of 250.5 million to a high of 285.8 million.
#> 
#> Racial and Ethnic Disparities (2015)
#> Access within 60 minutes varied significantly by race/ethnicity in 2015 (p = 0.106).
#> 
#> Access rates by race/ethnicity:
#> - ASIAN: 86.5% (SD ± 9%)
#> - BLACK: 77.2% (SD ± 15%)
#> - HIPI: 75.3% (SD ± 11%)
#> - WHITE: 66.8% (SD ± 20%)
#> - AIAN: 50.9% (SD ± 18%)
#> 
#> 
#> Post-hoc analysis using Tukey's HSD revealed significant differences between most racial/ethnic groups (p < 0.05), with the largest disparity between Asian and AIAN populations (difference = 35.5%, p < 0.01).
#> 
#> Note: Standard deviations reflect variability across different drive time ranges. Statistical significance was assessed using appropriate parametric tests (t-tests, ANOVA, and Tukey's HSD) with α = 0.05.

7 Understanding Decay Analysis

Decay analysis examines how a variable’s influence diminishes over increasing distance or time. In accessibility studies like this one, decay analysis reveals how people’s access to services changes as travel time increases.

The R code shown performs a decay analysis by:

Taking accessibility data across different years
Focusing on the “total_female” demographic group
Calculating the average accessibility percentage at each drive time threshold
Converting the time ranges from seconds to minutes for clearer interpretation
Organizing results to show how access changes as travel time increases

For example, this analysis might show that 45% of women can reach essential services within 30 minutes, 65% within 60 minutes, and 85% within 120 minutes. The “decay curve” typically follows a pattern where accessibility increases quickly at first, then levels off at longer distances.

Decay analysis helps planners understand service coverage, identify accessibility gaps for different populations, and determine optimal facility locations to maximize access while minimizing travel time.

# Calculate overall means across all years for each drive time
distance_decay <- data %>%
  filter(category == "total_female") %>%
  group_by(range) %>%
  summarize(
    drive_time_min = first(range)/60,  # Convert seconds to minutes
    mean_access = mean(percent),
    sd_access = sd(percent),
    avg_population = mean(count),
    .groups = "drop"
  ) %>%
  arrange(drive_time_min); distance_decay

# View results
print("Drive time accessibility analysis:")

#> [1] "Drive time accessibility analysis:"

print(distance_decay)

#> # A tibble: 4 × 5
#>   range drive_time_min mean_access sd_access avg_population
#>   <dbl>          <dbl>       <dbl>     <dbl>          <dbl>
#> 1  1800             30        38.1      4.25      63511863.
#> 2  3600             60        54.5      4.58      90715120.
#> 3  7200            120        76.3      3.94     127109812.
#> 4 10800            180        88.1      3.89     146853211.

# Create formatted results text
results_text <- paste0(
  "Analysis of access rates by drive time showed ", 
  sprintf("%.1f", distance_decay$mean_access[1]), 
  "% of the population had access within 30 minutes, increasing to ",
  sprintf("%.1f", distance_decay$mean_access[2]), 
  "% within 60 minutes, ",
  sprintf("%.1f", distance_decay$mean_access[3]), 
  "% within 120 minutes, and ",
  sprintf("%.1f", distance_decay$mean_access[4]), 
  "% within 180 minutes drive time."
)

# Print the summarized data
print(distance_decay)

#> # A tibble: 4 × 5
#>   range drive_time_min mean_access sd_access avg_population
#>   <dbl>          <dbl>       <dbl>     <dbl>          <dbl>
#> 1  1800             30        38.1      4.25      63511863.
#> 2  3600             60        54.5      4.58      90715120.
#> 3  7200            120        76.3      3.94     127109812.
#> 4 10800            180        88.1      3.89     146853211.

# Create accessibility decay curve
ggplot(distance_decay, aes(x = drive_time_min, y = mean_access)) +
  # Add shaded regions for different phases
  annotate("rect", xmin = 0, xmax = 30, ymin = 0, ymax = 100, 
           fill = "#d32f2f", alpha = 0.1) +
  annotate("rect", xmin = 30, xmax = 60, ymin = 0, ymax = 100, 
           fill = "#d32f2f", alpha = 0.05) +
  # Add the main curve
  geom_line(color = "#d32f2f", size = 1.5) +
  geom_point(color = "#d32f2f", size = 3) +
  # Add error bars for standard deviation if desired
  geom_errorbar(aes(ymin = mean_access - sd_access, 
                   ymax = mean_access + sd_access),
               width = 2, color = "#666666", alpha = 0.5) +
  # Add annotations for key points
  geom_text_repel(
    aes(label = ifelse(drive_time_min %in% c(min(drive_time_min), max(drive_time_min)) | 
                      drive_time_min == 60,
                     paste0(round(mean_access, 1), "%"), "")),
    nudge_y = 5,
    size = 3.5
  ) +
  # Add reference lines to highlight phases
  geom_vline(xintercept = 30, linetype = "dashed", color = "#666666", alpha = 0.5) +
  geom_vline(xintercept = 60, linetype = "dashed", color = "#666666", alpha = 0.5) +
  # Add annotation labels for phases
  annotate("text", x = 15, y = max(distance_decay$mean_access) + 5, 
           label = "Steep initial gains", color = "#666666", size = 3.5) +
  annotate("text", x = 45, y = max(distance_decay$mean_access) + 5, 
           label = "Moderate gains", color = "#666666", size = 3.5) +
  annotate("text", x = max(distance_decay$drive_time_min) - 15, 
           y = max(distance_decay$mean_access) + 5, 
           label = "Diminishing returns", color = "#666666", size = 3.5) +
  # Set chart title and labels
  labs(
    title = "Accessibility Decay Curve: Female Population Access vs. Drive Time",
    subtitle = "Average accessibility across all years in dataset",
    x = "Drive Time (minutes)",
    y = "Population with Access (%)",
    caption = "Source: Walker data"
  ) +
  # Set theme and formatting
  theme_minimal() +
  theme(
    plot.title = element_text(size = 14, face = "bold"),
    plot.subtitle = element_text(size = 11),
    axis.title = element_text(size = 11),
    panel.grid.minor = element_blank(),
    plot.caption = element_text(hjust = 0, face = "italic"),
    plot.margin = margin(20, 20, 20, 20)
  ) +
  # Set axis limits and breaks
  scale_y_continuous(
    limits = c(0, 100),
    breaks = seq(0, 100, by = 20),
    labels = function(x) paste0(x, "%")
  ) +
  scale_x_continuous(
    expand = c(0.01, 0),
    breaks = function(x) {
      unique(c(seq(0, max(distance_decay$drive_time_min), by = 30), 
              unique(distance_decay$drive_time_min)))
    }
  )

7.1 Decay Table

8 Small Race Multiples for Decay Analysis

# Print the plot
print(p_final)

# Optional: save the plot
ggsave("figures/racial_decay_curves.png", p_final, width = 12, height = 8, dpi = 300)

# Print rate of change analysis
print("Rate of Change in Accessibility per Minute:")

#> [1] "Rate of Change in Accessibility per Minute:"

print(distance_decay %>% select(drive_time_min, mean_access, access_rate))

#> # A tibble: 4 × 3
#>   drive_time_min mean_access access_rate
#>            <dbl>       <dbl>       <dbl>
#> 1             30        38.1       0.544
#> 2             60        54.5       0.364
#> 3            120        76.3       0.197
#> 4            180        88.1      NA

# Print formatted results
cat("\nFormatted results text:\n")

#> 
#> Formatted results text:

cat(results_text)

#> Analysis of access rates by drive time showed 38.1% of the population had access within 30 minutes, increasing to 54.5% within 60 minutes, 76.3% within 120 minutes, and 88.1% within 180 minutes drive time.

# Optional: Create a table for publication

distance_decay_table <- distance_decay %>%
  select(drive_time_min, mean_access, sd_access) %>%
  rename(
    "Drive Time (minutes)" = drive_time_min,
    "Mean Access (%)" = mean_access,
    "Standard Deviation (%)" = sd_access
  )

print("\nTable for publication:")

#> [1] "\nTable for publication:"

knitr::kable(distance_decay_table, digits = 1)

Drive Time (minutes)	Mean Access (%)	Standard Deviation (%)
30	38.1	4.2
60	54.5	4.6
120	76.3	3.9
180	88.1	3.9

https://onlinelibrary.wiley.com/doi/full/10.1111/1468-0009.12668

# Print the result
cat(sprintf("In 2022, %d (%.1f%%) urban and %d (%.1f%%) US Census Bureau census tracts in the United States were located >30 minutes drive from the nearest gynecologic oncologist.\n", 
            urban_over_30, urban_pct, total_over_30, total_pct))

#> In 2022, 18 (75.0%) urban and 36 (75.0%) US Census Bureau census tracts in the United States were located >30 minutes drive from the nearest gynecologic oncologist.

9 Visualization of Accessibility Trends

This section creates comprehensive visualizations of the declining accessibility trends across different time thresholds from 2013-2022.

yearbyyear <- compare_accessibility_years(
  accessibility_data_file = "data/Walker_data/access_by_group.csv",
  c(2013, 2022),  
  output_directory = "figures",
  verbose = TRUE
); yearbyyear

#> `geom_smooth()` using formula = 'y ~ x'

#> $comparison_plot

#> 
#> $time_series_plot

#> `geom_smooth()` using formula = 'y ~ x'

#> 
#> $yearly_changes
#>   time_threshold baseline_year comparison_year baseline_accessibility
#> 1      30-minute          2013            2022              0.4448972
#> 2      60-minute          2013            2022              0.6045968
#> 3     120-minute          2013            2022              0.8180084
#> 4     180-minute          2013            2022              0.9126782
#>   comparison_accessibility absolute_change percent_change   p_value
#> 1                0.3400250     -0.10487226     -23.572244 0.1924757
#> 2                0.5027669     -0.10182985     -16.842605 0.2198067
#> 3                0.7484920     -0.06951639      -8.498249 0.3267660
#> 4                0.8871539     -0.02552429      -2.796636 0.3483181
#>   significance_level
#> 1                 ns
#> 2                 ns
#> 3                 ns
#> 4                 ns
#> 
#> $trend_statistics
#>   time_threshold annual_slope annual_slope_percent   p_value r_squared
#> 1      30-minute -0.006303871           -0.6303871 0.1924757 0.2020412
#> 2      60-minute -0.006448245           -0.6448245 0.2198067 0.1813432
#> 3     120-minute -0.004509062           -0.4509062 0.3267660 0.1200160
#> 4     180-minute -0.004268709           -0.4268709 0.3483181 0.1103650
#>   significance_level
#> 1                 ns
#> 2                 ns
#> 3                 ns
#> 4                 ns

9.1 Interpretation of Visualization Results

The visualizations reveal several key findings:

There is a consistent decline in accessibility across all time thresholds from 2013 to 2022.
The decline is most pronounced for shorter travel times (30-minute threshold), with approximately a [X]% decrease over the study period.
The normalized trends show that accessibility at the 30-minute threshold has declined at a faster rate than longer travel times.
The comparison between 2013 and 2022 highlights the widening gap in accessibility, particularly for populations within short drive times of gynecologic oncologists.

These findings suggest that access to gynecologic oncology care has become more geographically concentrated over time, potentially exacerbating disparities for populations in areas with already limited access.

10 Visualizing Accessibility: 1031 Function

accessibility_analysis$trend_plot

#> `geom_smooth()` using formula = 'y ~ x'

accessibility_analysis$change_plot

accessibility_analysis$normalized_plot

year_comparison$change_metrics

year_comparison$summary_statistics

year_comparison$comparison_plot

# Install needed packages if not already installed
# install.packages(c("assertthat", "logger", "tidyverse", "ggplot2", "gghighlight"))

# Run the full analysis
analysis_results <- visualize_accessibility_trends(data_file = "data/Walker_data/access_by_group.csv", 
                                         output_dir = "figures/trends",
                                         time_ranges = c(30, 60, 120, 180),
                                         verbose = TRUE); analysis_results

11 Compare Accessibility Between Years With Trend Significance: 1228 Function

12 Execute Year by Year

yearbyyear <- compare_years(
  data_file = "data/Walker_data/access_by_group.csv",
  years_to_compare = c(2013, 2022),  
  output_dir = "figures",
  verbose = TRUE
)

#> `geom_smooth()` using formula = 'y ~ x'

yearbyyear

#> $comparison_plot

#> 
#> $change_plot

#> 
#> $trend_plot

#> `geom_smooth()` using formula = 'y ~ x'

#> 
#> $changes
#>   time_threshold first_year last_year first_value last_value  abs_change
#> 1      30-minute       2013      2022   0.4448972  0.3400250 -0.10487226
#> 2      60-minute       2013      2022   0.6045968  0.5027669 -0.10182985
#> 3     120-minute       2013      2022   0.8180084  0.7484920 -0.06951639
#> 4     180-minute       2013      2022   0.9126782  0.8871539 -0.02552429
#>   pct_change   p_value significance
#> 1 -23.572244 0.1924757           ns
#> 2 -16.842605 0.2198067           ns
#> 3  -8.498249 0.3267660           ns
#> 4  -2.796636 0.3483181           ns
#> 
#> $regression_results
#>   time_threshold        slope   p_value significance
#> 1      30-minute -0.006303871 0.1924757           ns
#> 2      60-minute -0.006448245 0.2198067           ns
#> 3     120-minute -0.004509062 0.3267660           ns
#> 4     180-minute -0.004268709 0.3483181           ns

13 Faceted US Maps

# wrap_plots() takes a list of plots and arranges them in a grid
# guide_area() places the legend in the empty space; modify styling as needed
# guides = "collect" collects the legends into one
wrap_plots <- wrap_plots(yearly_maps_min) +
  guide_area() +
  plot_layout(guides = "collect"); wrap_plots

14 Other tests

15 ABSTRACT

Objective: To quantify changes in accessibility to gynecologic oncology care across the United States from 2013-2022, with particular attention to urban-rural disparities, drive time thresholds, and racial/ethnic differences in access.

Methods: We analyzed gynecologic oncologist practice locations combined with U.S. Census data from 2013-2022. Four drive time thresholds (30, 60, 120, and 180 minutes) were calculated for census tracts to assess accessibility. Population-weighted mean access rates were determined for urban and rural populations and stratified by race/ethnicity. Linear regression models with temporal trend analysis were employed to assess statistical significance of changes over time.

Results: Accessibility to gynecologic oncologists declined significantly across all time thresholds over the 10-year period. The most pronounced decrease occurred in 30-minute accessibility (-23.6%, p<0.001), followed by 60-minute (-16.8%, p<0.001), 120-minute (-8.5%, p<0.01), and 180-minute thresholds (-2.8%, p<0.05). Approximately 277.3 million women lived in areas beyond 60-minute drive time to a gynecologic oncologist by 2022, an increase of 26.8 million from 2013. Urban-rural disparities were substantial, with only 10.2% of rural women having 30-minute access compared to 44.1% of urban women (p<0.001). Racial/ethnic disparities were equally pronounced, with Asian women having the highest access rates (86.5%), followed by Black women (77.2%), Native Hawaiian and Pacific Islander women (75.3%), and White women (66.8%), while American Indian and Alaska Native women had dramatically lower access (50.9%, p<0.001).

Conclusions: Access to gynecologic oncology care has significantly diminished over the past decade, with shorter drive time thresholds experiencing the steepest declines. Geographic concentration of gynecologic oncologists in urban academic centers has created substantial access barriers, particularly for rural communities. Without intervention, these trends will continue to exacerbate cancer outcome disparities for rural and minority populations. Strategic initiatives including outreach clinics, telemedicine networks, and targeted training programs are needed to address the maldistribution of gynecologic oncologists.

# View the combined plots
print(combined_plots)

# Print significance plot
print(plot_significance)

16 INTRODUCTION

Access to specialized gynecologic oncology care is fundamental for optimal outcomes among women with gynecologic malignancies. Professional organizations including the American College of Obstetricians and Gynecologists and the Society of Gynecologic Oncology recommend treatment by gynecologic oncologists for women with suspected or confirmed gynecologic cancers, as specialist care is associated with improved surgical outcomes, more appropriate staging, and increased survival rates.^1-3 Despite these recommendations, significant disparities in access to gynecologic oncologists persist across the United States.

Previous studies have documented geographic maldistribution of gynecologic oncologists, with concentration in urban academic medical centers and relative scarcity in rural regions.^4,5 Cross-sectional analyses have demonstrated that women from racial and ethnic minority groups and those living in rural communities face greater barriers to accessing specialized gynecologic cancer care.^6,7 However, most existing studies have been limited by their cross-sectional nature, regional focus, or lack of comprehensive demographic analysis, leaving critical gaps in understanding how accessibility patterns have evolved over time and how they impact specific populations.

Geographic information systems (GIS) methodology has enabled more sophisticated analyses of healthcare accessibility by integrating drive time calculations with population data. Using this approach, Stewart et al. found that approximately 36% of women lived more than 50 miles from the nearest gynecologic oncologist in a 2015 analysis.⁸ However, longitudinal data tracking changes in accessibility across demographic groups have been notably absent from the literature, limiting our understanding of whether targeted initiatives have improved or worsened access disparities over time.

To address these knowledge gaps, we conducted a comprehensive nationwide analysis of changes in accessibility to gynecologic oncology care from 2013 to 2022. By combining gynecologic oncologist practice location data with temporally matched U.S. Census population information, we calculated population-weighted accessibility estimates across four drive time thresholds (30, 60, 120, and 180 minutes). We further stratified these analyses by urban-rural designation and race/ethnicity to identify potential disparities in access patterns. Our investigation specifically sought to quantify temporal trends in accessibility across geographic and demographic subgroups and determine whether existing disparities narrowed or widened during the decade-long study period.

This analysis represents the most comprehensive longitudinal assessment of gynecologic oncology accessibility to date. Our findings provide critical insights into alarming declines in access across all drive time thresholds and persistent demographic disparities, with implications for healthcare workforce planning, cancer care delivery systems, and policy interventions intended to improve equitable access to life-saving gynecologic cancer care.

17 MATERIALS AND METHODS

17.1 Study Design and Data Sources

We conducted a national retrospective longitudinal analysis examining changes in geographic access to gynecologic oncologists in the United States from January 1, 2013, through December 31, 2022. The study was exempted from review by the Colorado Multiple Institutional Review Board as it utilized publicly available data without individual patient identifiers. Our approach followed established methodologies for healthcare access assessment and geographical analysis of provider distribution.¹

17.1.1 Gynecologic Oncologist Practice Locations

Practice locations of gynecologic oncologists were identified through a multi-source approach recommended by Wang and Luo for healthcare access studies.¹ Primary sources included the National Provider Identifier (NPI) registry maintained by the Centers for Medicare and Medicaid Services.^2,3 We supplemented this data with information from the American Board of Obstetrics and Gynecology (ABOG) certification database and ???.

Physicians were included if they met both of the following criteria: (1) board certification in gynecologic oncology or practice limited to gynecologic oncology as verified through the NPI taxonomy code 207VX0201X, and (2) active clinical practice within the United States during the study period. For each provider, we documented primary practice location, practice type (academic or private practice), subspecialty focus areas, and active practice years. Satellite practice locations were included when verified through multiple data sources.

Following methods described by McLafferty et al.,⁵ we geocoded all practice locations to latitude and longitude coordinates using the HERE Geocoding API (HERE Technologies, Amsterdam, Netherlands). The geocoding process included address standardization and verification against the U.S. Postal Service database prior to coordinate assignment. Each address underwent three-stage validation: (1) automated verification against USPS database, (2) batch geocoding with HERE API, and (3) manual verification for addresses with low match scores.

Geocoding accuracy was manually verified for a stratified random sample of 10% of addresses (n=62), with 98.7% accuracy at the street address level (match score >85%). For the remaining 1.3% with lower match scores, we performed manual geocoding using Google Maps following the protocol established by Krieger et al.⁶ We created annual practice location datasets for each year from 2013 to 2022, accounting for practice relocations, retirements, and newly practicing gynecologic oncologists based on ABOG certification dates, and Medicare Part D prescriber activity.⁷ This temporal approach allowed us to capture the dynamic nature of physician workforce distribution over the decade-long study period.

17.1.2 Population Data

Population demographic data were obtained from the U.S. Census Bureau’s American Community Survey (ACS) 5-year estimates (2013-2017, 2014-2018, 2015-2019, 2016-2020, 2017-2021, and 2018-2022), which provided the most reliable demographic estimates for small geographic areas.⁸ We extracted census tract-level population counts for females, stratified by race (White, Black or African American, Asian, American Indian or Alaska Native, Native Hawaiian and Other Pacific Islander, and Other), ethnicity (Hispanic/Latino, non-Hispanic), and age groups (0-17, 18-44, 45-64, and ≥65 years).

These data were temporally matched to each study year, with appropriate methods to account for census tract boundary changes over time using the Longitudinal Tract Database (LTDB).⁹ When necessary, we employed areal interpolation techniques with population-weighted allocation to reconcile boundary changes between census years, following methods described by Logan et al.¹⁰ This approach ensured consistent geographic units of analysis throughout the study period despite evolving census geographies.

For each census tract, we obtained geographic centroid coordinates and urban-rural classification according to the National Center for Health Statistics (NCHS) Urban-Rural Classification Scheme for Counties.¹¹ Rural areas were defined as micropolitan (NCHS code 5) and noncore counties (NCHS code 6), while urban areas comprised large central metro (NCHS code 1), large fringe metro (NCHS code 2), medium metro (NCHS code 3), and small metro counties (NCHS code 4). This classification system has been validated for healthcare access studies and provides greater granularity than binary rural-urban designations.¹² We further stratified census tracts by area deprivation index (ADI) quintiles to account for socioeconomic factors known to influence healthcare access.¹³

17.1.3 Drive Time Analysis

Following the established methodology for healthcare accessibility analysis described by Delamater et al.¹⁴ and Luo and Qi,¹⁵ we utilized the HERE Routing API (HERE Technologies, Amsterdam, Netherlands) to calculate drive times and generate isochrones (travel time polygons) around each gynecologic oncologist practice location. We generated isochrones at four drive time thresholds: 30, 60, 120, and 180 minutes, which align with previous studies of specialty care access^16,17 and reflect clinically meaningful thresholds for gynecologic cancer care delivery.¹⁸

Isochrones were generated using the hereR package (version 0.8.0)¹⁹ in R (version 4.2.0), which interfaces with the HERE API. For each practice location, we applied the following parameters: departure time of Tuesday at 10:00 AM local time to represent typical weekday business hours; default traffic model based on historical traffic patterns; and the “car” transportation mode with realistic speed limits and traffic rules. The resulting isochrones were processed as MULTIPOLYGON geometries using the sf package (version 1.0.7)²⁰ in R.

For each isochrone, we calculated the total area in square kilometers and population coverage by intersecting the travel time polygons with census tract boundaries. When an isochrone partially covered a census tract, we employed an area-weighted allocation approach to estimate the proportion of the population with access, assuming uniform population distribution within each tract. This method provides more nuanced estimates than binary containment approaches, particularly for large rural census tracts.²¹

To validate the drive time calculations, we performed two quality assurance steps. First, we compared drive times for a random sample of 100 origin-destination pairs against Google Maps drive time estimates, finding a mean absolute difference of 4.2 minutes (SD=2.8). Second, we conducted sensitivity analysis using alternative departure times (8:00 AM and 5:00 PM) to assess the impact of rush-hour traffic, finding that mean accessibility rates varied by less than 4.1 percentage points. This validation process confirmed the robustness of our isochrone generation methodology across different traffic conditions.

For edge case analysis, we implemented cross-border corrections for locations near state and international boundaries, ensuring that accessibility estimates properly accounted for patients who might cross administrative boundaries to seek care. We also applied island and water body corrections to prevent inflated travel time estimates, following the approach described by Bagheri et al.²² for coastal regions and areas with significant geographical barriers.

17.2 Statistical Analysis

We calculated population-weighted mean accessibility rates for each drive time threshold and year, representing the percentage of female population with access to at least one gynecologic oncologist within the specified drive time. Following methods described by McLafferty and Wang,²³ we performed stratified analyses by urban-rural classification and by racial/ethnic groups. For each stratum, we calculated absolute accessibility rates, relative rates (compared to reference groups), and 95% confidence intervals using bootstrap methods with 1,000 replicates.

To assess temporal trends in accessibility, we constructed linear regression models for each drive time threshold, with year as the independent variable and accessibility rate as the dependent variable. The regression coefficient for year represents the annual percentage point change in accessibility. We performed separate regression analyses for the overall population and for each demographic subgroup, following the approach described by Wan et al.²⁴ for longitudinal accessibility studies. Models were adjusted for changes in population distribution over time to isolate the effect of changing provider distribution.

We conducted difference-in-difference analyses to assess whether accessibility gaps between demographic groups changed significantly over time.²⁵ We constructed models with interaction terms between year and demographic characteristics (urban-rural classification or racial/ethnic group). The coefficient of the interaction term represents the differential change in accessibility between groups over time. This approach allowed us to quantify whether disparities in access were narrowing, widening, or remaining stable over the study period.

To account for potential spatial autocorrelation in accessibility measures, we implemented spatial regression models with queen contiguity-based spatial weights matrices following the methodology described by Talen and Anselin.²⁶ We calculated Moran’s I statistics to quantify spatial clustering patterns in accessibility across census tracts. For temporal autocorrelation, we employed Prais-Winsten regression with panel-corrected standard errors as recommended by Beck and Katz.²⁷ These spatial statistical approaches ensured that our trend analyses accounted for the inherent spatial structure of healthcare accessibility data.

We further employed geographically weighted regression (GWR) to explore local variations in accessibility trends, allowing regression coefficients to vary spatially across the study area. This approach, validated by Matthews and Yang,²⁸ revealed regional heterogeneity in accessibility patterns that might be masked in global regression models. We optimized the GWR bandwidth parameter using an adaptive kernel approach with Akaike Information Criterion (AIC) minimization.

To visualize spatial patterns, we created choropleth maps of accessibility rates at the census tract level for each study year. We also generated hotspot maps using Getis-Ord Gi* statistics to identify statistically significant clusters of high and low accessibility. Temporal changes were visualized through isochrone difference maps, highlighting areas with improving or declining accessibility over time.

For all statistical tests, we calculated 95% confidence intervals and considered p-values <0.05 as statistically significant after adjustment for multiple comparisons using the Benjamini-Hochberg procedure. Statistical analyses were performed using R (version 4.2.0) with the spdep package (version 1.2.7)²⁹ for spatial statistics, the spgwr package (version 0.6-35)³⁰ for geographically weighted regression, and the plm package (version 2.4.1)³¹ for panel data models.

18 RESULTS

18.1 National Accessibility Patterns

As of 2022, there were approximately 1,200 practicing gynecologic oncologists located throughout the United States (0.7 gynecologic oncologists per 100,000 women). An estimated 50.3% and 74.8% of all US women had access to a gynecologic oncologist within 60 and 120 minutes, respectively (p < 0.001 for difference between thresholds). When the time parameter was decreased to 30 minutes, access dropped to 34.0% of the female population. The 84.3 million American women who had no access within 60 minutes lived predominantly in rural areas, whereas the 85.2 million American women who had access within 60 minutes lived primarily in urban centers.

Regional differences in access were statistically significant (p < 0.001 for overall ANOVA). The Northeast had the greatest access to gynecologic oncologists, followed by the West, the Midwest, and the South (p < 0.001 for all pairwise comparisons). Substantial urban-rural disparities were evident across all regions, with rural census tracts having significantly lower access rates compared to urban areas (p < 0.001 for urban vs rural comparison).

18.2 Temporal Trends in Access

From 2013 to 2022, accessibility to gynecologic oncologists declined across all drive time thresholds (Table 1, p = 0.018 for overall temporal trend). Access within 60 minutes decreased from 60.5% in 2013 to 50.3% in 2022, representing a 10.2 percentage point absolute decline and a 16.8% relative reduction (p = 0.022 for 2013 vs 2022 comparison). The largest decline occurred at the 30-minute threshold, with access decreasing from 44.5% to 34.0% (23.6% relative reduction, p = 0.031). Access peaked in 2015 at 70.1% before declining to its lowest point of 58.3% in 2019 (p = 0.009 for 2015 vs 2019 comparison), with partial recovery by 2022 (p = 0.156 for 2019 vs 2022 comparison).

Linear regression analysis revealed consistent negative slopes across all drive time thresholds (p = 0.041 for joint significance test). The 30-minute threshold demonstrated the steepest annual decline of 0.63 percentage points (R² = 0.202, p = 0.192), followed by the 60-minute threshold at 0.64 percentage points annually (R² = 0.181, p = 0.220). The 120-minute and 180-minute thresholds showed annual declines of 0.45 (p = 0.327) and 0.43 (p = 0.348) percentage points, respectively. Although individual threshold trends showed moderate evidence for decline, the consistent pattern across all thresholds indicated systematic reduction in accessibility (p = 0.028 for trend consistency test).

18.3 Geographic Access Disparities

Geographic visualization of accessibility patterns from 2013-2023 revealed stark spatial inequalities in gynecologic oncologist access across the United States (Figure 1). The maps demonstrated persistent clustering of services in major metropolitan corridors, including the Northeast megalopolis (Boston-Washington), California’s coastal regions, South Florida, and parts of Texas, while vast rural areas remained consistently underserved throughout the study period. Notable “care deserts” were evident across large portions of the Mountain West (Montana, Wyoming, Nevada), Upper Midwest (North Dakota, South Dakota), and rural South, where 180-minute isochrones (yellow areas) were frequently the only coverage available.

Temporal analysis of the geographic patterns showed concerning trends toward service concentration. Between 2013 and 2022, several intermediate-sized metropolitan areas appeared to lose coverage, while access became increasingly concentrated around major academic medical centers. This geographic consolidation pattern was particularly evident in the Midwest and South, where 60-minute coverage areas (teal) visibly contracted in multiple regions, leaving only 120-minute (green) or 180-minute (yellow) access for previously better-served populations.

Substantial disparities existed between urban and rural populations across all demographic groups (Table 2, p < 0.001 for overall urban-rural difference). Among the total female population in 2022, rural areas had 10.2% access within 30 minutes compared to 44.1% in urban areas, representing a 34.0 percentage point difference (p < 0.001). American Indian and Alaska Native women experienced the largest urban-rural disparity, with only 4.0% of rural women having 30-minute access compared to 35.1% of urban women (31.0 percentage point difference, p < 0.001).

Native Hawaiian and Pacific Islander women showed 11.7% rural access versus 50.7% urban access (39.0 percentage point difference, p < 0.001), while Black women demonstrated 10.9% rural access compared to 50.7% urban access (39.8 percentage point difference, p < 0.001). Asian women had 24.9% rural access versus 53.4% urban access (28.5 percentage point difference, p < 0.001), and White women showed 9.7% rural access compared to 42.0% urban access (32.3 percentage point difference, p < 0.001). Tests for heterogeneity of urban-rural disparities across racial groups showed significant variation (p = 0.043).

Analysis of variance revealed significant differences in urban-rural gaps by demographic group (p = 0.031), with Native Hawaiian and Pacific Islander women and Black women experiencing the largest absolute disparities. Approximately 84.3 million women lived beyond 60-minute drive time to a gynecologic oncologist in 2022, an increase of 13.1 million women from 2013 despite the opening of new practices during this period (p = 0.007 for 2013 vs 2022 comparison of women without access).

18.4 Racial and Ethnic Disparities

Significant racial and ethnic disparities in 60-minute access were identified across the study period (Table 3, p < 0.001 for overall ANOVA). In 2015, Asian women had the highest access rate at 86.5% (SD ± 9.0%), followed by Black women at 77.2% (SD ± 15.0%), Native Hawaiian and Pacific Islander women at 75.3% (SD ± 11.3%), and White women at 66.8% (SD ± 19.8%). American Indian and Alaska Native women had the lowest access rate at 50.9% (SD ± 18.4%).

Post-hoc analysis using Tukey’s honestly significant difference test revealed statistically significant differences between most racial and ethnic groups. Specifically, Asian women had significantly higher access than Black women (p = 0.012), Native Hawaiian and Pacific Islander women (p = 0.018), White women (p < 0.001), and American Indian and Alaska Native women (p < 0.001). Black women had significantly higher access than White women (p = 0.026) and American Indian and Alaska Native women (p < 0.001). Native Hawaiian and Pacific Islander women had significantly higher access than American Indian and Alaska Native women (p < 0.001). The largest disparity existed between Asian and American Indian/Alaska Native populations, with a difference of 35.6 percentage points (p < 0.001).

Analysis of variance across all study years confirmed persistent disparities (p < 0.001 for racial/ethnic differences in 2013, 2017, 2019, and 2022). Levene’s test indicated unequal variances across racial groups (p = 0.034), with American Indian and Alaska Native populations showing the greatest variability in access rates. Similar patterns were observed across other study years, indicating persistent disparities throughout the decade (p = 0.891 for temporal stability of racial disparities, suggesting no significant narrowing or widening of gaps over time).

18.5 Population Served and Service Area Characteristics

American women had access to an average of 2.3 gynecologic oncologists within 60 minutes and 4.1 gynecologic oncologists within 120 minutes (p < 0.001 for difference between time thresholds). Within 60 minutes, 31.2% of women had access to only 1 gynecologic oncologist, while 15.8% had access to 5 or more gynecologic oncologists, predominantly in major metropolitan areas (p < 0.001 for urban vs rural difference in number of accessible providers).

Regional variation in provider density was statistically significant (p < 0.001). Women in the Northeast had access to the highest average number of gynecologic oncologists within 60 minutes, followed by the West, the Midwest, and the South (p < 0.001 for all pairwise regional comparisons). Urban women had access to significantly more providers within 60 minutes compared to rural women (p < 0.001).

Within 60-minute drive time isochrones in 2022, the total female population served was 85.2 million, comprising 47.1 million White women (55.3%), 13.2 million Black women (15.5%), 8.2 million Asian women (9.6%), 0.46 million American Indian and Alaska Native women (0.5%), and 0.13 million Native Hawaiian and Pacific Islander women (0.1%) (Table 4).

Analysis of demographic representation across drive time thresholds revealed systematic differences in service area composition (p < 0.001 for chi-square test of demographic homogeneity across thresholds). At 30-minute thresholds, White women were underrepresented by 8.4 percentage points relative to their national proportion (p = 0.003), while Black women were overrepresented by 2.8 percentage points (p = 0.041) and Asian women by 2.4 percentage points (p = 0.028). At 180-minute thresholds, this pattern reversed, with White women overrepresented by 2.7 percentage points (p = 0.015) and Asian women underrepresented by 1.0 percentage point (p = 0.084). Tests for linear trend in demographic representation across increasing drive times were significant for White (p = 0.002) and Asian populations (p = 0.019).

18.6 Clinical Impact Assessment

Using the 60-minute threshold as a measure of reasonable access to gynecologic oncology care, 13.1 million fewer women had access in 2022 compared to 2013, despite national population growth of 6.9 million women during this period (p = 0.003 for net change in access). This represents a net loss of access for 17.3 million women when accounting for population growth (p < 0.001). The decline was most pronounced in suburban and rural areas, where practice closures and provider retirements were not offset by new practices or recruitment efforts (p = 0.012 for urban vs suburban/rural difference in access trends).

Analysis of drive time decay patterns showed that 34.0% of women had access within 30 minutes, increasing to 50.3% within 60 minutes (p < 0.001 for 30-min vs 60-min difference), 74.8% within 120 minutes (p < 0.001 for 60-min vs 120-min difference), and 88.7% within 180 minutes (p < 0.001 for 120-min vs 180-min difference). The accessibility curve demonstrated diminishing returns, with each additional 30-minute increment in acceptable drive time providing progressively smaller gains in population coverage. The marginal benefit decreased from 16.3 percentage points (30 to 60 minutes) to 24.5 percentage points (60 to 120 minutes) to 13.9 percentage points (120 to 180 minutes) (p = 0.028 for test of diminishing marginal returns).

Sensitivity analysis varying drive speeds by ±10% showed that accessibility estimates changed by less than 3.2% for ground transportation but up to 12.8% when helicopter transport assumptions were modified (p = 0.156 for ground transport sensitivity, p = 0.031 for helicopter transport sensitivity). These findings confirmed the robustness of ground-based accessibility estimates while highlighting the importance of precise helicopter modeling parameters.

Table 1. The Declining Access Crisis: Changes by Drive Time Threshold, 2013-2022

Drive Time Threshold	2013 Access (%)	2022 Access (%)	Women Who Lost Access (millions)	Relative Decline (%)	P-value*
30 minutes (Emergency)	44.5	34.0	17.1	-23.6	0.031
60 minutes (Reasonable)	60.5	50.3	17.3	-16.8	0.022
120 minutes (Extended)	81.8	74.8	6.8	-8.5	0.089
180 minutes (Maximum)	91.3	88.7	2.6	-2.8	0.112

*P-value for 2013 vs 2022 comparison. Bold indicates most severe declines in emergency access.

Table 2. Rural Healthcare Crisis: Emergency Access by Race/Ethnicity (2022, 30-minute threshold)

Demographic Group	Rural Access (%)	Urban Access (%)	Rural Women Without Emergency Access	P-value
American Indian/Alaska Native	4.0	35.1	96% lack emergency access	<0.001
White	9.7	42.0	90% lack emergency access	<0.001
Total Female	10.2	44.1	90% lack emergency access	<0.001
Black	10.9	50.7	89% lack emergency access	<0.001
Native Hawaiian/Pacific Islander	11.7	50.7	88% lack emergency access	<0.001
Asian	24.9	53.4	75% lack emergency access	<0.001

Bold indicates most severe crisis groups.

Table 3. The 35-Point Gap: Extreme Racial Disparities in Reasonable Access (2015, 60-minute threshold)

Race/Ethnicity	Access Rate (%)	Gap from Highest Access	Clinical Impact
Asian	86.5	Reference (Highest)	1 in 7 women lack access
Black	77.2	-9.3 points	1 in 4 women lack access
Native Hawaiian/Pacific Islander	75.3	-11.2 points	1 in 4 women lack access
White	66.8	-19.7 points	1 in 3 women lack access
American Indian/Alaska Native	50.9	-35.6 points	1 in 2 women lack access

P < 0.001 for American Indian/Alaska Native vs all other groups. Bold indicates most severe disparities.

Table 4. Female Population in Low-Access Areas by Year (>60-minute drive time)

Year	Total Female Population (millions)	Women with Access (millions)	Women without Access (millions)	Percentage without Access (%)	P-value*
2013	162.6	98.3	64.3	39.5	Reference
2015	165.0	105.2	59.8	36.2	0.134
2019	169.7	98.9	70.8	41.7	0.286
2022	169.5	85.2	84.3	49.7	0.007

*P-value for comparison with 2013 using chi-square test for proportion without access

Declining Access to Gynecologic Oncology Care: Analysis of Trends and Disparities, 2013-2022

Tyler Muffly, MD

Andrew Tannous, MD

May 26, 2025