ABSTRACT
Background
Patients with metastatic cancer at initial diagnosis have the worst prognosis. Despite their involvement in initial assessment, the relationship between Primary Care Physician availability and metastatic cancer at initial diagnosis remains unexplored. Our objective was to determine whether Primary Care Physician availability is associated with patients having metastatic cancer at initial diagnosis.
Methodology
This was a cross‐sectional study of patients aged ≥ 18 and diagnosed with breast, colorectal, liver, lung, ovarian, pancreatic, or prostate cancer from 2015 to 2019. Population‐based Texas and California cancer registries were used for patient data. The main exposure was county‐level Primary Care Physician density per 3500 individuals. Additional exposures included demographic factors, insurance, rural–urban continuum codes, county poverty rates, and US State at Diagnosis. The primary outcome was adjusted odds of metastatic cancer at initial diagnosis. Subgroup analyses based on insurance type, age group, and metropolitan residence were performed. Models were stratified by cancer type.
Results
Overall, 708,451 patients were included. On combined multivariable analysis, with each additional Primary Care Physician per 3500 individuals, the odds for primary outcome decreased by 2% to 6%. In a stratified analysis using Primary Care Physician density per 3500 individuals as a categorical variable, compared with counties with < 1 Primary Care Physician, counties with 3 or more Primary Care Physicians had reduced odds of primary outcome for multiple cancer types with no statistically significant difference for counties with 1–1.9 Primary Care Physicians.
Conclusion
Increased Primary Care Physician availability was associated with a nonlinear decrease in odds for metastatic cancer at diagnosis. Thus, increasing physician density can possibly lower the likelihood of metastatic cancer at initial diagnosis.
Keywords: early cancer diagnosis, metastatic cancer at diagnosis, PCP, primary care, social determinants of health
1. Introduction
Cancer is the second leading cause of death in the United States (US), resulting in over 600,000 deaths annually [1]. A significant proportion of these deaths can be attributed to late diagnosis of cancer, as over half of the cancers that are responsible for most cancer‐related deaths have already metastasized at diagnosis [1]. Despite the enormous efforts invested in improving cancer treatment, the outcomes of patients with metastatic cancer at initial diagnosis remain dismal [1]. At present, efforts to improve early detection of cancer have largely focused on new and improved screening strategies for asymptomatic patients, precluding those cancers for which screening is not available [2].
Early cancer detection requires either diagnosis in the preclinical stage through screening or timely evaluation of symptoms [3]. The World Health Organization (WHO) guidelines emphasize the importance of early diagnosis through timely recognition and investigation of symptoms, distinguishing it from cancer screening of asymptomatic patients [3]. Furthermore, early symptom recognition and investigation are necessary for early diagnosis of all cancers and not just screening‐amenable cancers [3]. Patient‐related factors associated with late‐stage cancer diagnosis include symptom awareness [4, 5], health behaviors [6], and socioeconomic factors such as income [7] and health insurance [8], which also depend on employment status. Beyond individual determinants, system‐level barriers such as limited availability of primary care physicians (PCP) [9], diagnostic facilities [10], and urban–rural disparities [11] —including prolonged travel times—have been implicated as key influences on cancer stage at presentation. In the United States, approximately 14% of all reportable cancers are diagnosed through screening; however, this estimate includes many cancers for which no established screening strategy exists. While screening remains a critical tool for the early detection of cancers with effective screening modalities, timely recognition and evaluation of symptoms remain essential for the diagnosis of cancers that are not amenable to screening as well as cancers that present symptomatically despite available screening programs. Indeed, a study by Philipson et al. estimated the aggregate value of achieving 100% adherence to recommended screening for breast, cervical, colorectal, lung, and prostate cancers would save an additional 3.2–5.1 million life‐years beyond what current adherence achieves, further highlighting the number of patients who are missed even when screening is available (1) [12]. Primary care providers may facilitate both pathways by promoting participation in recommended cancer screening programs and by ensuring timely evaluation of symptoms that may indicate an underlying malignancy.
Although numerous studies have examined the role of cancer screening programs in facilitating timely care [13, 14], there exists a dearth of literature focusing on early diagnosis in symptomatic patients. A recent study from the Veteran Health Administration (VHA) found that an annual PCP visit was associated with a 39% decrease in odds of metastasis at diagnosis [9]. However, findings from other population‐based studies have been equivocal, with associations found only for certain subpopulations [15, 16]. Importantly, these studies have critical limitations, such as misclassification of clinically early‐stage cancers as late‐stage, exclusion of cancers that have a high propensity for metastasis at diagnosis, and controlling for only some factors that may impact cancer diagnosis. In this study, we investigate the relationship of PCP availability and metastatic cancer at diagnosis by using population‐based state‐wide cancer registries from two different states, including multiple cancers and addressing important confounding factors listed above. We hypothesized that greater PCP density would be associated with lower odds of metastatic disease at diagnosis, even when stratifying subpopulations based on insurance type, age group, and metropolitan residence.
2. Methodology
2.1. Study Setting and Patient Selection
We completed a cross‐sectional retrospective database analysis. Our analytic sample included patients who were aged ≥ 18 years and diagnosed with breast, colorectal, liver, lung, ovarian, pancreatic, or prostate cancer from 2015 to 2019 in the US States of California and Texas, the two largest states by population in the country. The selected cancers are the top five most common causes of cancer‐related death for males and females and include cancers for which screening is and is not available. Data were obtained from the California and Texas State cancer registries. Mandated by law, these registries attempt to include all patients diagnosed with cancer in their respective states. Both registries are Gold Certified by the North American Association of Central Cancer Registries awarded for achieving completeness of case ascertainment of ≥ 95%. We used International Classification of Diseases for Oncology, 3rd edition, codes to identify patients for inclusion. Patients whose cancer was first reported from death certificates or autopsy reports, had unknown stage at diagnosis, or unknown county of residence were excluded.
We used the Health Resources and Services Administration's Area Health Resources Files (AHRF) to extract county‐level data regarding the total population, as well as the number of Doctors of Medicine (MDs) and Doctors of Osteopathic Medicine (DOs) practicing as a PCP in each county. The AHRF physician data that we used was sourced from the 2019 American Medical Association Masterfile, and the population data was sourced from 2020 census county estimates. The study was reviewed and approved by the relevant Institutional Review Board and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
2.2. Variables
2.2.1. Primary Outcome
Surveillance, Epidemiology, and End Results (SEER) Summary staging categorizes the extension of cancer from its site of origin as in situ, localized, regional, or metastatic. The primary outcome was distant‐stage disease at diagnosis as defined by the Summary Stage 2018 classification. Throughout the manuscript, we use the term “metastatic” to refer to Summary Stage distant disease, consistent with the Summary Stage 2018 manual, which describes distant stage as metastatic disease, as this represents the final stage of cancer which has the worst prognosis. In contrast to prior studies, we did not use “late‐stage” disease (combined regional and distant stage), as the regional category encompasses a heterogeneous group of cancers and, for some cancer types, includes tumors that remain potentially curable or correspond to earlier stages in site‐specific staging systems (e.g., limited nodal involvement in breast cancer or involvement of the ampulla of Vater in pancreatic cancer).
2.2.2. Exposures of Interest and Covariates
Our main exposure of interest was county‐level PCP density per 3500 individuals. The density was estimated per 3500 as a level of < 1 PCP for 3500 people is a criterion for Heath Professional Shortage Area (HPSA) designation, making it more actionable for policy makers. Using this threshold, we created a categorical variable as follows: Less than 1, 1–1.9, 2–2.9, 3–3.9, and ≥ 4 PCPs per 3500 individuals.
Additional covariates included were county rural–urban continuum codes (RUCC) categorized as metropolitan area of ≥ 1 million individuals (RUCC 1), metropolitan areas of 250,000–1 million individuals (RUCC 2), metropolitan areas of < 250,000 individuals (RUCC 3), non‐metropolitan counties of ≥ 20,000 individuals (RUCC 4, 5), and non‐metropolitan counties of < 20,000 individuals (RUCC 6 to 9); insurance status (private insurance, Medicaid, Medicare, uninsured, insured but not specified, combined category of VHA, Tricare, and Indian Health Services, and combined category of unknown and missing); patient age in years; gender (male, female, non‐binary); US State of residence (California or Texas) and race/ethnicity (Non‐Hispanic White, Non‐Hispanic Black, Non‐Hispanic Asian, Non‐Hispanic of other races, Hispanic or unknown). Additionally, to account for county socioeconomic demographics, county‐level poverty rates were obtained from the American Community Survey (ACS), and included as a continuous variable in adjusted models [17].
2.3. Statistical Analysis
We used descriptive statistics (proportions, means) to characterize our sample, and completed bivariate analyses (Pearson's Chi Square test for categorical and Student's t‐test for continuous variables) to compare patients with and without metastatic cancer at diagnosis. We estimated multiple binary logistic regression models to examine the association of PCP density with metastatic cancer at initial diagnosis for all cancers combined and stratified by cancer type. Using the continuous PCP density variable, we first estimated a model for all patients (Model A). To isolate the relationship for patients least likely to face barriers, we performed subgroup analyses that used age and insurance types that have the highest acceptance rate as follows: working‐age adults (18–64 years) with private insurance (Model B) and non‐working age adults (≥ 65 years) with Medicare (Model C). Model B controlled for factors such as higher income, full‐time employment, and greater educational attainment in a working‐age population [18, 19]. Model C examined the relationship in a non‐working‐age adult population with high insurance acceptance rates. To control for additional factors related to population density and rurality, we estimated a fourth model restricted to patients meeting the criteria for either Model B or Model C and residing in metropolitan areas of ≥ 1 million individuals (RUCC 1) (Model D). Next, to assess whether PCP density and risk of metastatic cancer might have a non‐linear relationship, we used the categorical PCP density variable. Statistical analyses were completed using STATA version 18 (College Station, TX: Stata Corp LLC).
3. Results
3.1. Descriptive and Bivariate Analyses
Overall, a total of 708,451 patients were included (Figure 1). Among them, the most common cancer type was breast (35.8%), followed by prostate (19.2%), whereas the least common cancer type was ovarian (2.5%). Patients were predominantly female (57.6%), from California (58.9%), Non‐Hispanic White (60.8%), on Medicare (42.7%), and lived in RUCC 1 areas (71.1%) (Table 1).
FIGURE 1.
Flow diagram of analytic cohort derivation with inclusion and exclusion criteria and proportion of patients with metastatic cancer.
TABLE 1.
Comparisons of patients with vs. without metastatic cancer at initial diagnosis.
| Variables | Overall n = 708,451 | Patients with metastasis at initial diagnosis n = 150,453 | Patients without metastasis at initial diagnosis n = 557,998 | p | |
|---|---|---|---|---|---|
|
Age in Years (Mean ± SD) |
65.2 ± 12.5 | 68.0 ± 12.5 | 64.5 ± 12.4 | < 0.001 | |
| Gender | Male | 300,312 | 76,449 (25.5) | 223,863 (74.5) | < 0.001 |
| Female | 408,006 | 73,978 (18.1) | 334,028 (81.9) | ||
| Non‐binary | 133 | 26 (19.6) | 107 (80.4) | ||
| State of Diagnosis | California | 416,973 | 87,185 (20.9) | 329,788 (79.1) | < 0.001 |
| Texas | 291,478 | 63,268 (21.7) | 228,210 (78.3) | ||
| Race/Ethnicity | Non‐Hispanic White | 430,481 | 91,933 (21.4) | 338,548 (78.6) | < 0.001 |
| Non‐Hispanic Black | 68,385 | 16,057 (23.5) | 52,328 (76.5) | ||
| Non‐Hispanic Asian | 66,248 | 13,931 (21.0) | 52,317 (78.9) | ||
| Non‐Hispanic of Other Race | 6976 | 1427 (20.5) | 5549 (70.5) | ||
| Hispanic | 130,768 | 26,861 (20.5) | 103,907 (79.5) | ||
| Unknown | 5593 | 244 (4.4) | 5349 (95.6) | ||
| Insurance Status | Private | 272,637 | 43,686 (16.0) | 228,951 (84.0) | < 0.001 |
| Uninsured | 20,419 | 6945 (34.0) | 13,474 (66.0) | ||
| Medicaid | 52,479 | 14,441 (27.5) | 38,038 (72.5) | ||
| Medicare | 302,302 | 74,519 (24.7) | 227,783 (75.3) | ||
| VHA/Tricare/IHS/Military | 10,818 | 2239 (20.7) | 8579 (79.3) | ||
| Insurance NOS | 24,852 | 4617 (18.6) | 20,235 (81.4) | ||
| Unknown/Missing | 24,944 | 4006 (16.1) | 20,938 (83.9) | ||
|
Residential County Poverty Rate (Mean ± SD) |
12.5 ± 7.5 | 13.3 ± 7.7 | 12.3 ± 7.4 | < 0.001 | |
| Rural–Urban Continuum Codes (RUCC)* | 1 | 503,741 | 103,599 (20.6) | 400,142 (79.4) | < 0.001 |
| 2 | 117,531 | 25,601 (21.8) | 91,930 (78.2) | ||
| 3 | 35,448 | 8370 (23.6) | 27,078 (76.4) | ||
| 4–5 | 19,490 | 4800 (24.6) | 14,690 (75.4) | ||
| 6–9 | 32,241 | 8083 (25.1) | 24,158 (74.9) | ||
| Primary Site | Breast | 253,649 | 12,199 (4.8) | 241,450 (95.2) | < 0.001 |
| Colorectal | 108,919 | 24,665 (22.7) | 84,254 (77.3) | ||
| Liver | 26,239 | 4462 (17.0) | 21,777 (83.0) | ||
| Lung | 129,469 | 66,733 (51.5) | 62,736 (48.5) | ||
| Ovarian | 17,525 | 9410 (53.7) | 8115 (46.3) | ||
| Pancreas | 36,992 | 19,361 (52.3) | 17,631 (47.7) | ||
| Prostate | 135,658 | 13,623 (10.0) | 122,035 (90.0) | ||
| PCP Density per 3500 people | 0 to less than 1 | 25,059 | 6229 (24.9) | 18,830 (75.1) | < 0.001 |
| 1 to less than 2 | 91,279 | 21,183 (23.2) | 70,096 (76.8) | ||
| 2 to less than 3 | 169,400 | 36,969 (21.8) | 132,431 (78.2) | ||
| 3 to less than 4 | 284,467 | 58,853 (20.7) | 225,614 (79.3) | ||
| 4 or more | 138,246 | 27,219 (19.7) | 111,027 (80.3) | ||
1 = Counties in metro areas of 1 million or more individuals, 2 = Counties in metro areas of 250,000 to 1 million individuals, 3 = Counties in metro areas of fewer than 250,000 individuals, 4–5 = Non‐Metro: Urban population of 20,000 or more individuals, 6–9 = Non‐Metro: Urban population of fewer than 20,000 individuals.
Over 50% of lung, ovarian, and pancreatic cancer patients and only 4.8% of breast cancer patients were diagnosed with metastatic cancer at initial diagnosis. Patients with (vs. without) metastasis at initial diagnosis were also significantly older (mean age: 68.0 years vs. 64.5 years, p < 0.001) and lived in counties with a higher poverty rate (mean residential county poverty rate: 13.3% vs. 12.3%, p < 0.001). A greater proportion of patients who did (vs. did not) have metastatic cancer at initial diagnosis were male, Non‐Hispanic Black, and uninsured. The proportion of patients with metastasis at initial diagnosis also increased as the residential county population decreased, being highest for RUCC 6–9 counties (25.1%). Finally, the proportion of patients with metastatic cancer at initial diagnosis was highest in counties with < 1 PCP per 3500 people (24.9%) and lowest in counties with ≥ 4 PCPs per 3500 people (19.7%) (Table 1).
3.2. Multivariable Analysis
Our combined models (Table S1, Models A–D) showed that the risk of metastatic cancer at initial diagnosis increased significantly with each additional year of age at diagnosis. Non‐Hispanic Black and Non‐Hispanic Asian (vs. Non‐Hispanic White) patients had significantly higher odds of having metastasis at initial diagnosis. With each percentage point increase in county poverty rate, the odds of metastatic cancer at diagnosis increased by 1%. The odds of an individual having metastasis at initial diagnosis increased incrementally as county population decreased, with patients living in RUCC 6–9 areas having 13% higher odds than those living in RUCC 1 areas (OR 1.13, 95% CI 1.09–1.16) (Table S1, Model A). Compared with patients with private insurance (Table S1, Model A), patients with all other insurance types had higher odds of having metastatic cancer at initial diagnosis: Uninsured 181% (OR 2.81, 95% CI 2.72–2.90), Medicaid 110% (OR 2.10, 95% CI 2.06–2.15), Medicare 21% (OR 1.21, 95% CI 1.19–1.23), VHA/Tricare/IHS/Military 8% (OR 1.08, 95% CI 1.03–1.13).
For all cancer types, patients who were uninsured or Medicaid‐insured consistently had significantly higher odds of having metastatic cancer at initial diagnosis when compared with privately insured patients (Table S2, Model A). For uninsured patients, the risk was 327% higher for prostate cancer (OR 4.27, 95% CI 3.84–4.74), followed by 260% higher for breast cancer (OR 3.60 95% CI 3.30–3.92). Similarly, for patients insured through Medicaid, the risk was 196% higher for prostate cancer (OR 2.96, 95% CI 2.73–3.22), followed by 151% higher for breast cancer (OR 2.51 95% CI 2.36–2.66).
3.3. PCP Density
The model that combined all cancer types and included PCP density as a continuous variable (Table 2, Model A combined; Figure 2, Panel A) found that the odds of being diagnosed with metastatic cancer at initial diagnosis decreased by 3% with each additional PCP per 3500 individuals (OR 0.97, 95% CI 0.96–0.98). In combined models for the various subsample analyses (Table 2, Model B–D; Figure 2, Panel A) the odds of being diagnosed with metastatic cancer at initial diagnosis decreased by 2% to 6% for each additional PCP per 3500 individuals. In models stratified by cancer type, PCP density was significantly associated with decreased odds of being diagnosed with metastatic cancer at initial diagnosis for liver and lung (Table 2, Models A–D; Figure 2, Panel A), breast (Table 2, Models A–B; Figure 2, Panel A), ovarian (Table 2, Model B; Figure 2, Panel A), and colorectal cancers. The odds of metastatic cancer at initial diagnosis increased with an increase in PCP density for prostate cancer (Table 2, Models A, C, D; Figure 2, Panel A). Although not statistically significant, the overall trends for most of the remaining models and cancer types were similar (Figure 2, Panel A).
TABLE 2.
Association of PCP density per 3500 individuals (continuous variable) with metastatic cancer at the time of diagnosis from adjusted multivariable models.
| Cancer type | Variables | Odds ratios | p | |
|---|---|---|---|---|
| All cancers combined | PCP density per 3500 | Model A | 0.97 (0.96–0.98) | < 0.001 |
| Model B | 0.94 (0.93–0.96) | < 0.001 | ||
| Model C | 0.98 (0.97–0.99) | < 0.001 | ||
| Model D | 0.98 (0.97–0.99) | < 0.001 | ||
| Breast | PCP density per 3500 | Model A | 0.95 (0.93–0.97) | < 0.001 |
| Model B | 0.95 (0.91–0.98) | 0.006 | ||
| Model C | 0.98 (0.95–1.01) | 0.30 | ||
| Model D | 0.98 (0.95–1.02) | 0.40 | ||
| Colorectal | PCP density per 3500 | Model A | 0.99 (0.98–1.01) | 0.83 |
| Model B | 0.94 (0.92–0.97) | < 0.001 | ||
| Model C | 1.03 (1.00–1.05) | 0.02 | ||
| Model D | 1.02 (1.00–1.05) | 0.02 | ||
| Liver | PCP density per 3500 | Model A | 0.91 (0.88–0.94) | < 0.001 |
| Model B | 0.85 (0.78–0.93) | < 0.001 | ||
| Model C | 0.91 (0.86–0.97) | 0.004 | ||
| Model D | 0.91 (0.86–0.96) | 0.001 | ||
| Lung | PCP density per 3500 | Model A | 0.96 (0.95–0.97) | < 0.001 |
| Model B | 0.95 (0.92–0.98) | 0.002 | ||
| Model C | 0.97 (0.95–0.98) | < 0.001 | ||
| Model D | 0.97 (0.95–0.98) | < 0.001 | ||
| Ovary | PCP density per 3500 | Model A | 0.97 (0.93–1.00) | 0.08 |
| Model B | 0.93 (0.88–0.98) | 0.01 | ||
| Model C | 0.97 (0.91–1.04) | 0.43 | ||
| Model D | 0.97 (0.91–1.03) | 0.34 | ||
| Pancreas | PCP density per 3500 | Model A | 0.98 (0.96–1.00) | 0.001 |
| Model B | 0.97 (0.92–1.02) | 0.26 | ||
| Model C | 0.99 (0.96–1.03) | 0.89 | ||
| Model D | 0.99 (0.96–1.03) | 0.79 | ||
| Prostate | PCP density per 3500 | Model A | 1.03 (1.00–1.05) | 0.005 |
| Model B | 0.97 (0.92–1.03) | 0.37 | ||
| Model C | 1.03 (1.00–1.06) | 0.01 | ||
| Model D | 1.03 (1.00–1.06) | 0.01 | ||
Note: Model A: All patients. Adjusted for age, gender, US State at diagnosis, Race/Hispanic ethnicity, Insurance status, poverty index, RUCC. Model B: Subgroup – Private Insurance Aged 18 to 64. Adjusted for gender, US State at diagnosis, Race/Hispanic ethnicity, poverty index, RUCC. Model C: Subgroup‐Medicare aged 65 and above. Adjusted for gender, US State at diagnosis, Race/Hispanic ethnicity, poverty index, RUCC. Model D: Subgroup—Model B or Model D patients living in RUCC 1 counties. Adjusted for gender, US State at diagnosis, Race/Hispanic ethnicity, poverty index.
FIGURE 2.
Adjusted odds of metastatic cancer at initial diagnosis by PCP density per 3500 individuals.
In the overall models that used PCP density as a categorical variable (Table 3; Figure 2, Panel B), compared with patients from counties < 1 PCP per 3500 individuals, odds of metastatic cancer at initial diagnosis were significantly lower with incremental increases in PCP density. However, stratified by cancer type, there was no statistically significant difference in the odds of metastatic cancer between counties with < 1 PCP per 3500 individuals (the reference category) and counties with 1–1.9 PCP per 3500 except for in breast cancer. However, the odds of metastatic cancer at initial diagnosis were 6% to 20% lower for patients living in counties with 3 to 3.9 PCP per 3500 individuals for 3 of the 7 cancer types examined.
TABLE 3.
Association of PCP density per 3500 individuals (categorical variable) with metastatic cancer at the time of diagnosis stratified by cancer type from adjusted multivariable models.
| Cancer type | Variables | All cancers combined | p | |
|---|---|---|---|---|
| All cancers combined | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.96 (0.92–0.99) | 0.013 | ||
| 2 to 2.9 | 0.93 (0.90–0.97) | < 0.001 | ||
| 3 to 3.9 | 0.88 (0.85–0.91) | < 0.001 | ||
| 4 or more | 0.87 (0.84–0.90) | < 0.001 | ||
| Breast | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.92 (0.82–1.03) | 0.15 | ||
| 2 to 2.9 | 0.86 (0.76–0.95) | 0.006 | ||
| 3 to 3.9 | 0.82 (0.73–0.91) | < 0.001 | ||
| 4 or more | 0.78 (0.69–0.88) | < 0.001 | ||
| Colorectal | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.96 (0.88–1.04) | 0.35 | ||
| 2 to 2.9 | 1.01 (0.93–1.09) | 0.76 | ||
| 3 to 3.9 | 0.95 (0.87–1.03) | 0.22 | ||
| 4 or more | 0.91 (0.83–1.00) | 0.05 | ||
| Liver | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.96 (0.81–1.05) | 0.69 | ||
| 2 to 2.9 | 0.85 (0.71–1.01) | 0.07 | ||
| 3 to 3.9 | 0.80 (0.67–0.96) | 0.01 | ||
| 4 or more | 0.69 (0.57–0.85) | < 0.001 | ||
| Lung | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 1.00 (0.94–1.06) | 0.92 | ||
| 2 to 2.9 | 1.02 (0.96–1.08) | 0.48 | ||
| 3 to 3.9 | 0.94 (0.89–1.00) | 0.06 | ||
| 4 or more | 0.92 (0.86–0.99) | 0.01 | ||
| Ovary | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.94 (0.76–1.16) | 0.55 | ||
| 2 to 2.9 | 0.85 (0.70–1.05) | 0.14 | ||
| 3 to 3.9 | 0.81 (0.66–0.99) | 0.04 | ||
| 4 or more | 0.84 (0.68–1.05) | 0.14 | ||
| Pancreas | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.95 (0.83–1.08) | 0.42 | ||
| 2 to 2.9 | 0.96 (0.84–1.09) | 0.54 | ||
| 3 to 3.9 | 1.09 (0.88–1.14) | 0.96 | ||
| 4 or more | 0.92 (0.80–1.05) | 0.24 | ||
| Prostate | PCP density per 3500 | 0 to 0.9 | Ref | |
| 1 to 1.9 | 0.96 (0.86–1.07) | 0.49 | ||
| 2 to 2.9 | 0.98 (0.88–1.10) | 0.74 | ||
| 3 to 3.9 | 0.97 (0.87–1.09) | 0.62 | ||
| 4 or more | 1.03 (0.91–1.16) | 0.64 | ||
Note: Multivariable binary logistic regression models. Each model is adjusted for age, gender, US State at diagnosis, race, Hispanic ethnicity, Insurance Status, RUCC, Poverty Index.
4. Discussion
Access to primary care services is crucial for early cancer detection, yet research on this topic remains limited. To the best of our knowledge, this study is among the first to examine associations between PCP availability and metastatic cancer at initial diagnosis. We found a negative association between metastatic cancer at initial diagnosis and increased PCP density, even for subpopulations of privately insured patients living in densely populated metropolitan areas. Arguably, these patients have greater access to health care and increased health awareness given their higher likelihood of being educated. The relationship also held for non‐working age patients with Medicare, which, like private insurance, has a high acceptance rate. Similar trends were observed across all included cancer types, but statistical significance was limited to certain cancers. Having 3 or more PCPs per 3500 individuals was associated with reduced odds of metastatic cancer at diagnosis for most cancer types, whereas 1 to 1.9 PCPs per 3500 people showed no significant difference. This suggests that the HPSA threshold of 1 PCP may be inadequate, and a higher density could possibly act as a mechanism to meaningfully lower metastatic cancer rates at diagnosis.
Recent evidence indicates that only 14% of cancers in the United States are diagnosed following a screening test, with the vast majority identified through other means, including symptom onset or incidental detection during unrelated medical care [12]. In the present analysis, more than half of patients with certain malignancies presented with metastasis—a finding made particularly notable by the observation that this proportion was lower in counties with greater PCP density. The observed association may reflect several functions of primary care, including increased screening utilization for cancers with established screening recommendations as well as earlier recognition, evaluation, and referral of patients presenting with potentially cancer‐related symptoms. Furthermore, although the propensity and timing of cancer metastasis is determined primarily by tumor biology [20], a patient's response to symptoms and availability and use of resources (e.g., primary care services) can contribute to the cancer stage at diagnosis. In an analysis of over 2.75 million patients in the US, Hostetter et al. demonstrated that increased primary care visits per year were significantly associated with improved longitudinal health outcomes and reduced healthcare expenditures [21]. Another noteworthy finding was that prostate cancer patients in certain subgroups exhibited increased odds of metastatic disease with higher PCP density. A possible explanation for this can stem from the 2012 update to the USPSTF guidelines, which prior work using SEER data has shown declining localized disease incidence and increasing metastatic disease incidence after the 2012 recommendation (1). Furthermore, the association of higher metastatic disease in higher PCP density areas aligns with work by Garg et al., who found that men in urban areas had lower odds of any prostate cancer screening when compared with non‐urban areas (2). Notably, prostate cancer has previously been shown to be the only cancer type where higher PCP rate was associated with increased late‐stage incidence, with one explanation being that PCP‐dense areas adopted the USPSTF recommendation more aggressively, leading to less screening and fewer urology referrals (3). The findings from this study suggest that one mechanism of improved outcomes may be early diagnosis of cancer.
Consistent with studies that identify urban–rural disparities in cancer care [11], we found that as population density decreased, the risk of metastatic cancer at diagnosis increased. The risk was also highest for uninsured and Medicaid‐insured patients and lowest for those privately insured. Although urbanicity and insurance are important, they can be reflective of patients' circumstances. For instance, the studies indicate that factors unique to rural populations, such as transportation challenges, lower health literacy, and higher prevalence of risky health behaviors, lead to worse cancer outcomes [11]. Similarly, poor cancer outcomes among patients who are uninsured or Medicaid‐insured may be related to limited health care access, lower rates of full‐time employment, lower educational attainment, and lower income. Importantly, we find that even for patients who are less likely to be impacted by these factors, higher PCP density is associated with lower risk of delayed cancer diagnosis.
Recognizing the impact of limited access to health professionals, the US government established HPSA, a system for identifying healthcare provider shortage areas. Prior work has demonstrated the importance of HPSA designation, with one recent study reporting a mean life expectancy nearly 311 days shorter in counties with primary care physician density below the HPSA threshold. We leveraged the HPSA criterion for designation of primary care shortage areas to define our outcome of interest and found that the risk of delayed cancer diagnosis was negatively associated with PCP availability per 3500 individuals. This indicates that the impact of PCP density extends to non‐HPSA designated areas, as the odds of patients with metastatic cancer at initial diagnosis were lowest for counties that had 4 or more PCPs per 3500 individuals. Efforts to address physician shortage have been made for close to 50 years, yet a shortage of up to 86,000 physicians has been projected for the year 2036 [22]. The Association of American Medical Colleges calls for a multi‐faceted approach that includes increasing residency training opportunities and adoption of innovative and technologically advanced means of delivering patient care [23]. The findings from this study contribute to the expanding literature on the impact of physician availability on health outcomes that transcend specialties and disease complexities.
Taken together, the results of this study provide important and practical insights into the relationship between availability of PCPs and cancer stage at diagnosis. However, certain limitations need to be considered. We estimated the average relationship of PCP density at the county level, which reflects the broader healthcare environment rather than individual access to care, which may vary within counties. Furthermore, patients may obtain care outside their county of residence, limiting the ability of this measure to fully capture physician access. Given the observational and cross‐sectional design of this study, causality cannot be inferred. Although higher PCP density was associated with lower odds of metastatic disease at diagnosis, these findings should be interpreted as hypothesis‐generating and do not establish a definitive workforce threshold for policy implementation. The impact of access to other resources such as diagnostic facilities or other specialists on cancer stage at diagnosis is not accounted for in our analyses, nor are certain other factors that may impact cancer stage at diagnosis but are not easily attainable through administrative data, such as patient income and health behaviors. Furthermore, our analysis did not account for advanced practice providers, whose availability may influence access to primary care and cancer detection, particularly in rural and underserved areas. Although physician density reflects the local supply of primary care services, actual access may be influenced by factors such as insurance status, transportation, appointment availability, and other patient‐level barriers. Furthermore, healthcare system characteristics such as diagnostic imaging availability, specialist density, hospital resources, and screening utilization were not available for inclusion in the analysis. Therefore, the observed association between PCP density and stage at diagnosis may reflect broader healthcare infrastructure and access‐to‐care factors rather than PCP density alone. In addition, causal inferences cannot be made from cross‐sectional data, and there may be errors in the administrative records. Finally, regional idiosyncrasies present among individuals residing in CA and TX may limit generalizability of our findings to other parts of the US.
5. Conclusion
Our results demonstrate an association between greater PCP availability and a lower likelihood of metastatic cancer at initial diagnosis. However, this relationship appeared nonlinear in stratified analyses, with no significant difference observed between counties with 1–1.9 PCPs per 3500 individuals and those with fewer than 1 PCP per 3500 individuals. These findings suggest that primary care availability may be associated with more timely cancer detection, although the observational nature of this study precludes causal inference. Further research is needed to better characterize this relationship and determine whether specific PCP density thresholds are associated with meaningful differences in cancer outcomes. The efforts to increase access to primary care, even among individuals who are privately insured and reside in metropolitan areas, can possibly be an important step toward timely diagnosis and better cancer outcomes.
Author Contributions
Muhammad Sohaib Khan: conceptualization, data curation, formal analysis, visualization, writing – original draft, methodology, investigation. Bella Etingen: conceptualization, methodology, supervision, writing – review and editing. Herbert J. Zeh: writing – review and editing, project administration, supervision, validation, investigation. Patricio M. Polanco: conceptualization, data curation, writing – review and editing, project administration, supervision, investigation, methodology, software, validation, resources. Tammy Leonard: conceptualization, writing – review and editing, project administration, supervision, validation. Natalie Williams: conceptualization, data curation, methodology, investigation. Saad Mallick: data curation, visualization, writing – original draft, writing – review and editing, software, resources, project administration.
Funding
The authors have nothing to report.
Ethics Statement
The study was reviewed and approved by our Institutional Review Board with exempted consent and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
Conflicts of Interest
Dr. Zeh reports that he is on the Scientific Advisory Board of Surgical Safety Technologies. Dr. Polanco reports that he is a Consultant at Iota Bioscience, Proctor for Intuitive Surgical, and Consultant for Medtronics Inc. The rest of the authors have no related competing interest to declare.
Supporting information
Table S1: Factors Associated with Metastatic Cancer at Time of Diagnosis, All Cancers Combined
Table S2: Association of PCP Density per 3500 Individuals (Continuous Variable) with Metastatic Cancer at the Time of Diagnosis from Adjusted Multivariable Models#.
Data Availability Statement
The data that support the findings of this study are available from the Texas Cancer Registry (TCR), Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services.
References
- 1. Siegel R. L., Kratzer T. B., Giaquinto A. N., Sung H., and Jemal A., “Cancer Statistics, 2025,” CA: a Cancer Journal for Clinicians 75 (2025): 10–45. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. National Cancer Plan‐Detect Cancer Early, 2025.
- 3. Organization WH , Guide to cancer early diagnosis, 2017.
- 4. Koo M. M., Swann R., McPhail S., et al., “Presenting Symptoms of Cancer and Stage at Diagnosis: Evidence From a Cross‐Sectional, Population‐Based Study,” Lancet Oncology 21 (2020): 73–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Green S. M. C., Lloyd K. E., Smith S. G., and investigators E , “Awareness of Symptoms, Anticipated Barriers and Delays to Help‐Seeking Among Women at Higher Risk of Breast Cancer: A UK Multicentre Study,” Preventive Medicine Reports 34 (2023): 102220. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Klein W. M. P., O'Connell M. E., Bloch M. H., et al., “Behavioral Research in Cancer Prevention and Control: Emerging Challenges and Opportunities,” Journal of the National Cancer Institute 114 (2022): 179–186. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Ruan Y., Heer E., Warkentin M. T., et al., “The Association Between Neighborhood‐Level Income and Cancer Stage at Diagnosis and Survival in Alberta,” Cancer 130 (2024): 563–575. [DOI] [PubMed] [Google Scholar]
- 8. Zhao J., Han X., Nogueira L., et al., “Health Insurance Status and Cancer Stage at Diagnosis and Survival in the United States,” CA: a Cancer Journal for Clinicians 72 (2022): 542–560. [DOI] [PubMed] [Google Scholar]
- 9. Qiao E. M., Guram K., Kotha N. V., et al., “Association Between Primary Care Use Prior to Cancer Diagnosis and Subsequent Cancer Mortality in the Veterans Affairs Health System,” JAMA Network Open 5 (2022): e2242048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Moscovice I. S., Parsons H., Bean N., et al., “Availability of Cancer Care Services and the Organization of Care Delivery at Critical Access Hospitals,” Cancer Medicine 12 (2023): 17322–17330. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Bhatia S., Landier W., Paskett E. D., et al., “Rural‐Urban Disparities in Cancer Outcomes: Opportunities for Future Research,” Journal of the National Cancer Institute 114 (2022): 940–952. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Percent of Cancers Detected by Screening in the U.S.: NORC at the University of Chicago, 2022.
- 13. Waks A. G., King T. A., and Winer E. P., “Timeliness in Breast Cancer Treatment‐The Sooner, the Better,” JAMA Oncology 2 (2016): 302–304. [DOI] [PubMed] [Google Scholar]
- 14. Doubeni C. A., Gabler N. B., Wheeler C. M., et al., “Timely Follow‐Up of Positive Cancer Screening Results: A Systematic Review and Recommendations From the PROSPR Consortium,” CA: a Cancer Journal for Clinicians 68 (2018): 199–216. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Plascak J. J., Fisher J. L., and Paskett E. D., “Primary Care Physician Supply, Insurance Type, and Late‐Stage Cancer Diagnosis,” American Journal of Preventive Medicine 48 (2015): 174–178. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Ananthakrishnan A. N., Hoffmann R. G., and Saeian K., “Higher Physician Density Is Associated With Lower Incidence of Late‐Stage Colorectal Cancer,” Journal of General Internal Medicine 25 (2010): 1164–1171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. U.S. Census Bureau , “2015‐2019 American Community Survey 5‐year Public Use Microdata Samples [SAS Data file],” (2024), https://www.census.gov/programs‐surveys/acs/data.html.
- 18. Dewar D. M., “Do Those With More Formal Education Have Better Health Insurance Opportunities?,” Economics of Education Review 17 (1998): 267–277. [Google Scholar]
- 19. Keisler‐Starkey K. and Bunch L. N., “Health Insurance Coverage in the United States: 2023,” U.S. Census Bureau, Current Population Reports (2024): 60–284. [Google Scholar]
- 20. Fares J., Fares M. Y., Khachfe H. H., Salhab H. A., and Fares Y., “Molecular Principles of Metastasis: A Hallmark of Cancer Revisited,” Signal Transduction and Targeted Therapy 5 (2020): 28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Hostetter J., Schwarz N., Klug M., Wynne J., and Basson M. D., “Primary Care Visits Increase Utilization of Evidence‐Based Preventative Health Measures,” BMC Family Practice 21 (2020): 151. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Tim Dall R. R., Chakrabarti R., Ruttinger C., Zarek P., and Parker O., The Complexities of Physician Supply and Demand: Projections From 2021 to 2036, 2024.
- 23. AAMC , Addressing the Physician Workforce Shortage, 2024.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Factors Associated with Metastatic Cancer at Time of Diagnosis, All Cancers Combined
Table S2: Association of PCP Density per 3500 Individuals (Continuous Variable) with Metastatic Cancer at the Time of Diagnosis from Adjusted Multivariable Models#.
Data Availability Statement
The data that support the findings of this study are available from the Texas Cancer Registry (TCR), Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services.