Hospital‐ and county‐level characteristics explain geographic variability in prices of cancer‐related procedures: Implications for policy and interventions

Abstract Background Healthcare costs in the U.S. are high and variable, which can hinder access and impact health outcomes across communities. This study examined hospital‐ and county‐level characteristics to identify factors that explain geographic variation in prices for four cancer‐related procedures. Methods Data sources included Turquoise Health, which compiles publicly‐available price data from U.S. hospitals. We examined list prices for four procedures: abdominal ultrasound, diagnostic colonoscopy, brain MRI, and pelvis CT scan, which we linked to characteristics of hospitals (e.g., number of beds) and counties (e.g., metropolitan status). We used multilevel linear regression models to assess multivariable relationships between prices and hospital‐ and county‐level characteristics. Supplementary analyses repeated these models using procedures prices for commercial insurance plans. Results For each procedure, list prices varied across counties (intraclass correlation: abdominal ultrasound = 23.2%; colonoscopy = 17.1%; brain MRI = 37.2%; pelvis CT = 50.9%). List prices for each procedure were associated with hospital ownership (all p < 0.001) and percent of population without health insurance (all p < 0.05). For example, list prices for abdominal ultrasound were higher for proprietary versus Government‐owned hospitals (β = 539.10, 95% confidence interval [CI]: 256.12, 822.08, p < 0.001) and for hospitals in counties with more uninsured residents (β = 23.44, 95% CI: 2.55, 44.33, p = 0.03). Commercial insurance prices were negatively associated with metropolitan status. Conclusions Prices for cancer‐related healthcare procedures varied substantially, with considerable heterogeneity associated with county location as well as county‐level social determinants of health (e.g., health insurance coverage). Interventions and policy changes are needed to alleviate the financial burden of cancer care among patients, including geographic variation in prices for cancer‐related procedures.

counties (e.g., metropolitan status).We used multilevel linear regression models to assess multivariable relationships between prices and hospital-and countylevel characteristics.Supplementary analyses repeated these models using procedures prices for commercial insurance plans.
Conclusions: Prices for cancer-related healthcare procedures varied substantially, with considerable heterogeneity associated with county location as well as county-level social determinants of health (e.g., health insurance coverage).
Interventions and policy changes are needed to alleviate the financial burden of cancer care among patients, including geographic variation in prices for cancerrelated procedures.

| INTRODUCTION
In the United States, geographic variation in healthcare pricing and spending is evident. 1 For a given procedure, pricing can differ by a factor of 10 or more. 2,3hese variations are related to costs needed to provide services in a specific area (e.g., local rent 1 and wages 4 ), facility-specific factors (e.g., academic status), 3 negotiations between insurers and providers, and patient health status. 1 Prices account for 70% of geographic differences in healthcare spending. 4Across counties, market competition is a key determinant of pricing because providers in low-access counties (e.g., counties with fewer providers or health insurance plans 5 ; rural counties 6 ) can negotiate higher reimbursement rates, resulting in higher prices.
These variations have real implications for healthcare access, utilization, and debt among patients, particularly cancer patients.Half of U.S. adults report foregoing healthcare due to costs, 7 and high healthcare prices disproportionately impact patients with low socioeconomic status and patients without health insurance. 7Healthcare spending associated with cancer diagnosis and treatment has rapidly increased in recent years, resulting in "financial toxicity" which can cause adverse effects such as stress and financial hardship. 8,9In response to these concerns, the Centers for Medicare and Medicaid Services (CMS) implemented a new rule on January 1, 2021, mandating that hospitals disclose their pricing to empower patients, enhance market competition, and curtail healthcare costs. 10,11espite the high burden of healthcare costs, gaps remain in our understanding of geographic variation in prices of cancer-related healthcare procedures.In this study, we leveraged data from 2004 U.S. hospitals to characterize geographic variability in pricing for four procedures that may be used during cancer diagnosis and treatment (abdominal ultrasound, diagnostic colonoscopy, magnetic resonance imaging [MRI] scan of the brain, and pelvis computerized tomography [CT] scan with contrast).These findings provide insight into geographic differences in cancer-related healthcare pricing, with implications for access and accessibility of care among cancer patients.Future research can leverage these findings to inform price transparency interventions 3 or policies 10 to reduce healthcare pricing, spending, and debt.

Procedure prices
We extracted pricing data from Turquoise Health. 12urquoise Health compiles publicly-available, machinereadable pricing data from U.S. hospitals (in compliance with the CMS Hospital Price Transparency Regulation), including separate prices by health insurance types/plans.The current analysis focused on hospitals' "list price" for each procedure, as well as supplementary price measures for commercial insurance plans.

| Hospital-level independent variables
Turquoise Health compiles a number of hospital-specific variables, including ownership (government, physician, proprietary, or volunteer [i.e., charity]), type (acute care or critical access), compliance score (range: 1-5, with higher scores indicating greater compliance), and number of beds (range: 1-2891). 12Hospital ownership types have become more complex in recent decades, with many blended models emerging; however, in general, these systems can be considered not-for-profit (i.e., government and volunteer/charity) or for-profit (i.e., physician-owned or proprietary [investor-owned]). 13Compliance scores reflect the degree to which hospitals comply with the CMS price disclosure rules. 12

| County-level independent variables
We collected county-level variables from 2020 Area Health Resource File 14 : Census region (Midwest, Northeast, South, or West), 15 metropolitan status (metropolitan/ urban or non-metropolitan/rural), 16 county population (in 1000s), and density of primary care physicians (PCP; number of PCPs per 100,000 population).We also gathered indicators of social determinants of health: percent without health insurance, percent living below the federal poverty level, percent unemployed (among population ages 16+ in the civilian workforce), and percent with a bachelor's degree (among population ages 25+ years).In addition, we collected life expectancy from 2020 County Health Rankings (CHR) 17 to capture general healthfulness of the population.
We geocoded each hospital's address and linked hospital-and county-level data using a federal information processing system (FIPS) code. 18Multiple locations within a health system were treated as separate hospitals.Our dataset included 2004 hospitals that reported prices for at least one procedure.These hospitals represented 1319 out of 3143 (42.0%)U.S. counties (average hospitals per county = 1.8; range: 1-29).

| Statistical analysis
First, we generated descriptive statistics for the independent variables, i.e., frequency and proportion for categorical variables and median, mean, and 95% confidence interval (CI) for continuous variables.
Then, we generated the median, mean, 95% CI, and coefficient of variation (CV) for the list price of each procedure.We generated choropleth maps to depict variability in prices across counties; for counties with more than one price for a procedure, we calculated the arithmetic mean of the prices within that county.
Next, we used multilevel linear regression models [19][20][21] to assess the relationships between independent variables and the list price for each procedure, using hospital as the unit of analysis.Separately for each procedure, we ran an "empty model" to calculate the intraclass correlation (ICC), 21 which summarizes the degree of clustering in prices for hospitals nested within counties (Model 0).Then, we regressed the list price for each procedure on hospital variables (level-1 variables; Model 1) and county variables (level-2 variables; Model 2).Finally, we regressed the list price on all of the hospital-and county-level variables ("full model"; Model 3).Supplementary analyses repeated these procedures for commercial insurance prices; for parsimony, we report the findings for Models 0 and 3 for the supplementary analyses.
Statistical analyses used a two-sided p-value of 0.05.Analyses were conducted using SAS version 9.4 (Cary, NC).Per U.S. federal regulations, this analysis was exempt from ethics review because it did not involve human subjects.Informed consent was not obtained because all data were aggregated at the hospital or county level (i.e., no participant interaction was involved in the study).

| Multilevel characteristics and prices for abdominal ultrasound
In Model 0 (the empty model estimating clustering in prices), the ICC for abdominal ultrasound prices indicated that 23.2% of the variation in prices across hospitals was attributable to county location.In Model 1 (assessing the relationships between prices and hospital-level variables), prices differed by hospital ownership and type (both p < 0.01).In Model 2 (assessing the relationships between prices and county-level variables), prices differed by Census region, county population, and percent uninsured (all p < 0.05).In Model 3 (the full model assessing the relationships between prices and hospitaland county-level variables), prices were higher for hospitals that were proprietary versus government ownership (β = 539.10,95% CI: 256.12, 822.08, p < 0.001) and lower for critical access versus acute care hospitals (β = −355.11,95% CI: −573.81,−136.41,p < 0.01) (Table 1).In addition, prices were higher for hospitals in counties in the South versus Midwest region (β = 266.47,95% CI: 24.68, 508.26, p = 0.03) and with higher percent uninsured (β = 23.44,95% CI: 2.55, 44.33, p = 0.03).The ICC for the Model 3 was 13.6%, indicating that these variables accounted for 41.2% of the variation in prices.

| Multilevel characteristics and prices for brain MRI
In Model 0, the ICC for brain MRI was 37.2%.Prices differed by hospital ownership and number of beds (Model 1) as well as county-level Census region, population, and percent uninsured (Model 2; all p < 0.05).In Model 3, prices were higher for hospitals that were proprietary or volunteer

| Multilevel characteristics and prices for pelvis CT scan
In Model 0, the ICC for pelvis CT was 50.9%.Prices differed by hospital ownership, type, and number of beds (Model 1), as well as county-level Census region, percent   1).The ICC for Model 3 was 41.5%, indicating that these variables accounted for 18.4% of the variation in prices.

| Multilevel characteristics and commercial insurance prices
Commercial insurance prices were generally lower for hospitals in urban counties (Table S3).Commercial insurance prices also varied by hospital compliance score, with a negative association observed for abdominal ultrasound but positive associations observed for the remaining procedures.The ICCs for each procedure were moderate (abdominal ultrasound: ICC = 29.3%;diagnostic colonoscopy: ICC = 18.8%; brain MRI: ICC = 38.6%;pelvis CT: ICC = 30.0%).

| DISCUSSION
This study examined the relationship between hospital and county characteristics and healthcare prices for four procedures commonly used in cancer care.In terms of hospital characteristics, prices were higher at proprietary hospitals compared to government-owned hospitals for all four procedures; abdominal ultrasound prices were lower at critical access hospitals compared to acute care hospitals; and prices for diagnostic colonoscopy, brain MRI, and pelvis CT were higher at hospitals with more hospital beds.3][24] In terms of county characteristics, prices differed by region and health insurance coverage.The ICCs suggest that county location accounted for a substantial proportion of variation in prices.Geographic location of the hospitals helped explain pricing, with elevated prices for abdominal ultrasound, brain MRI, and pelvis CT in the South.Many hospitals with the highest price markups are located in Southern states, which motivates calls for federal and state policies  to limit the charge-to-cost ratios. 25Further, we found that county-level social determinants of health were associated with procedure prices.Specifically, list prices were positively associated with the percent of the population that was uninsured, which supports prior literature suggesting that hospitals raise prices to subsidize the risk of uncompensated care delivered to patients without health insurance. 22It is important to note that list price is often different from the actual out-of-pocket costs incurred by uninsured patients, which can be substantially lower.One study reported that most of the hospital revenue from care for the uninsured was from a subset of patients who paid full or near-full list price. 26verall, our study did not find consistent differences in prices by metropolitan status or density of PCPs, which differs from previous studies demonstrating that providers in low-access counties have greater negotiation power, resulting in higher prices for primary care visits. 5However, our results suggest that this relationship does not hold true for the four cancer-related procedures included in this study.
List prices differed from commercial insurance prices, even after accounting for county-and hospitalcharacteristics, which is consistent with previous findings. 2,27Procedure prices for commercial insurance were generally related to hospital compliance score and countylevel urbanicity.Price variation for commercial insurance may relate to negotiations between insurers and hospitals, or they may indicate that the markets are not operating efficiently. 28In general, as we observed in this study, greater market power among providers (e.g., in rural counties) can lead to higher prices for commercial insurers. 28hese findings have implications for interventions and policies focused on healthcare costs and accessibility.The CMS policy requiring hospitals to publish procedure prices was implemented in 2021 to empower patients and energize market competition. 10However, in low-access counties (including rural counties), patients have fewer options for accessing healthcare; as a result, competition is limited.Complementary price transparency policies could also impact costs for other aspects of care.For example, a recent study showed that prices for commonly-used chemotherapeutics at cancer centers were 188-634% higher than the estimated costs to manufacturers. 29It is important to remember that healthcare prices are not the same as costs incurred to insurers or out-of-pocket costs for patients, 30 so additional research is needed to understand these interrelationships and how patients can understand and navigate costs.Additional interventions and policies are needed to ensure healthcare accessibility and minimize costs, including efforts to ensure prices are available to patients in a digestible and consumer-friendly format.In addition, efforts are needed to reduce the burden of cancer diagnosis and treatment, particularly for patients in certain geographic areas, e.g., in the South and in counties with low health insurance coverage. 31,32Overall, additional research is needed on the correlation between listed prices and incurred costs.
These findings have implications for cancer inequities and for clinical practice with at-risk populations.Lowresource patients may delay, refuse, or discontinue cancer screening and treatment. 33For example, low-income women may delay cervical cancer screening because of high screening costs. 34Clinicians' role in helping patients manage treatment costs include screening patients for financial toxicity, discussing healthcare costs and concerns, and working with an interdisciplinary team that provides supportive care services. 35Patient navigators, nurse case managers, and social workers can assist patients in understanding healthcare costs and connecting patients with available resources. 35n terms of study strengths, our analysis leveraged a large dataset with coverage across 2004 hospitals and 1319 counties in the U.S. We conducted a comprehensive assessment of several multilevel characteristics theoretically and empirically related to hospital prices, across four illustrative cancer-related healthcare procedures, to demonstrate the breadth and reliability of our findings.In particular, analyzing differences in list prices and commercial insurance prices extended on prior literature (often focused on Medicare/Medicaid prices) and highlighted differences in multilevel characteristics associated with prices across health insurance types.
In terms of study limitations, the price dataset is limited to hospitals that published publicly-available, machinereadable data on prices, and therefore is incomplete.Thus, hospitals included in the dataset may differ from excluded hospitals in systematic ways.Further, data availability limited our analysis to four cancer-related procedures; clearly, additional procedures are administered for cancer diagnosis and treatment, and the included procedures could be used for non-cancer purposes.Future research should examine additional cancer-related procedures to extend our findings.Excluded variables may be important correlates of healthcare prices, which could bias the study findings.However, our models included random countylevel intercepts, which serve to minimize the influence of unmeasured county-level characteristics.

| CONCLUSIONS
Prices for cancer-related hospital procedures varied substantially by hospital, county, and health insurance type.County-level social determinants of health (e.g., levels of health insurance coverage and unemployment, metropolitan status) were consistently associated with procedure prices, helping to explain geographic variation in the financial burden of cancer care.Patients from lower resource communities may be charged the highest prices for cancer-related healthcare procedures.Further research is needed to determine whether price transparency programs drive reductions in healthcare costs and spending and ultimately impact cancer outcomes.Policy and programmatic interventions to alleviate financial burden are needed to ensure that patients can access the healthcare they need to effectively diagnose and treat cancer.

F I G U R E 1
County-level average of hospital list prices for (A) abdominal ultrasound, (B) diagnostic colonoscopy, (C) magnetic resonance imaging (MRI) of the brain, and (D) pelvis computerized tomography (CT) scan with contrast.T A B L E 1 Multilevel relationships between hospital-and county-level characteristics and hospital list prices for cancer-related procedures.