Residence in a rural setting is associated with lower five-year survival rates in patients that undergo spine surgery for breast cancer metastatic to the spine: A propensity matched analysis

Abstract

Background

Social determinants of health and access to care are associated with outcomes of spine surgery. The objective of the present study was to evaluate the association of residence in rural or urban counties with one-, two-, and five-year mortality in patients undergoing spine surgery for metastatic breast cancer to the spine.

Methods

A retrospective propensity matching study of patients over a 10-year period, using data from the U.S. Surveillance, Epidemiology and End Results (SEER) database. Included patients were >18 years of age, underwent spine surgery for breast cancer metastatic to the spine between 2006 and 2015, and were enrolled in the SEER Medicare database. Propensity score through the inverse probability of treatment weighting (IPTW) was conducted to achieve covariate-adjusted results.

Results

Of the 865 female patients, 800 patients resided in urban settings and 65 patients resided in rural settings. Residence in a rural setting was associated with a higher probability of residing in a county with a distress score in the mid-tier, at-risk, or distressed range compared to patients residing in an urban setting (p<.001). After propensity score matching, patients who were living in rural settings had higher probability of mortality at 5 years (HR, 3.39 [95% CI, 1.26–9.12], p=.016).

Conclusions

Rural residence is an independent predictor of 5-year mortality for patients undergoing surgery for metastatic breast cancer to the spine and a possible surrogate marker for access to care. It is imperative for healthcare providers to understand the social determinants of health to develop strategies to address barriers to accessing care.

Level of Evidence

III.

Introduction

Breast cancer is the most common form of malignancy among women [1]. Due to advancements in both surveillance and treatment, the prevalence of metastasis to bone has increased in recent years, with two-thirds of these cases presenting with spinal metastasis [[2], [3], [4]]. These patients often present with increased back pain, physiologic decline, neurological compromise, and increased mortality [3,5,6]. Initial treatment options often consist of nonoperative management including pain control, physical therapy, radiation, and stereotactic radiosurgery (SRS) [4,[7], [8], [9]]. However, surgery is commonly indicated to preserve function, improve mechanical stability, and decrease neurologic compromise [[10], [11], [12]].

While improvements in surgical technique and patient selection has improved survival outcomes in patients with metastatic spine tumors [13], the decision for surgery is still challenging due to increased postoperative complication rate [[14], [15], [16]]. To reduce these complications and improve postoperative outcomes, physicians must understand the medical and social risk factors that predispose to complications and mortality [17,18].

Previous studies that have evaluated presentation, management, and outcomes across the urban-rural continuum for many types of primary cancer [[19], [20], [21]]. These studies mostly report lower cancer survival rates in rural settings [22]. There is also evidence that the disparity in life expectancy overall is widening between urban and rural areas due to poorer general health measures, increased poverty, and reduced access to care [23]. In addition to these challenges, rural Americans with metastatic cancer are more likely to be uninsured or underinsured, face reduced access to specialized care, and a greater distance to specialty services [[24], [25], [26]].

To our knowledge, no prior studies have investigated differences in survival rates based on urban versus rural setting in patients with metastatic breast cancer to the spine. The objective of this study is to compare survival rates at one, two, and 5 years in patients undergoing spine surgery for metastatic breast cancer to the spine based on urban versus rural residence.

Methods

Study design

A retrospective study, with propensity-matching based on the classification of commuting area codes as rural or urban, was conducted using the Surveillance, Epidemiology, and End Results (SEER) Program database [27].

Ethics

The study protocol was approved through the Institutional Review Board. In accordance with CMS policy and data use agreements, no value between 1 and 10 will be directly reported in order to protect confidentiality of Medicare and Medicaid beneficiaries.

Setting and participants

The SEER registry contains data on patient demographics, tumor characteristics, stage of disease, treatment, and outcomes [28]. Female patients 18 years or older with breast cancer metastatic to the spine, undergoing spine surgery between 2006 and 2015, were included. The patients included were SEER-Medicare users who were likely to have complete claims, with both Part A and B coverage, without managed care coverage. Included patients met two conditions: (1) Patients had a known diagnosis of breast cancer (SEER code: 2600) and one of the following ICD-9 diagnostic codes: 198.3 (secondary malignant neoplasm of brain and spinal cord), 198.4 (secondary malignant neoplasm of other parts of nervous system), or 198.5 (secondary malignant neoplasm of bone and bone marrow). (2) The patient had at least one ICD-9-PCS code (Appendix 1) or Current Procedural Terminology (CPT) code (Appendix 2) for a procedure of the spine or central nervous system. SEER is linked with insurance claims, which allows for longitudinal tracking of healthcare outcomes for patients diagnosed with cancer [29]. The following data was sourced from insurance claims: type of procedure, time at which procedure was performed, prior medical encounters including healthcare screening visits and treatments, and survival rates.

Outcomes

Primary outcomes were all-cause mortality within 1 year, 2 years, and 5 years after undergoing spine surgery.

Covariates

The following covariates were used for propensity score matching: year and age at diagnosis, race, marital status, dual eligibility/insurance status, year of diagnosis, tumor grade, tumor nodal involvement, primary tumor American Joint Committee on Cancer (AJCC) classification [30], time from diagnosis to procedure, Her2/HR subtypes, estrogen and progesterone receptor status, tumor grade, lymph node metastasis, surgical procedures beyond the primary site, baseline function score, and community-level variables (levels of community distress, income, and education of the county of reference) (Appendix 3). Baseline functional score was defined based on function-related indicators, a combination of social and medical patient characteristics [31]. Community distress score ranges from 0 to 100 and is categorized as follows: Group 1 (0–40; prosperous or comfortable) and Group 2 (41–100; midtier, at-risk, or distressed). The Distressed Communities Index (DCI) database [32] was used to obtain county-level data on education, income, and distress score.

Statistical methods

An exploratory analysis of all variables to evaluate the frequency, percentage, and near-zero variance for categorical variables as well as the distribution for numeric variables and their corresponding missing value patterns, were determined. Sample characteristics were evaluated between the two groups using standardized mean differences (SMD) and p-values from t-tests, ANOVA, and Chi-square tests. p-values below .05 were statistically significant. Kaplan–Meier plots were used to draw survival curves for 1, 2, and 5 years. Propensity score through the inverse probability of treatment weighting (IPTW) was conducted to achieve covariate-adjusted results. Propensity matching balance plot and covariate balancing before and after propensity scoring is shown in Appendix Fig. 1 and Appendix Table 3 and 4.

Results

Demographics

A total of 865 female patients were included in the study. County urban/rural continuum census data was available for all 865 patients. Of these, 800 patients lived in counties described as urban and 65 patients lived in rural counties. There were more white patients than black patients living in urban versus rural residences (p=.026). Rural patients were more likely to reside in a county with a distress score in the midtier, at-risk, or distressed range compared to patients with urban residence (p<.001). Rural residence was also associated with living in a county in which at least 20% live below the poverty line (p<.001) while urban residence was associated with living in a county with income level below the 50th percentile nationally (p<.001; Table 1).

Table 1.

Study sample characteristics and unadjusted comparison between patients treated in urban vs. rural hospitals.

Variable [Missing]	Total (865)	Urban (800)	Rural (65)	p
Year of diagnosis categorical [0]				p=.996
2006–2010	552 (63.8%)	510 (63.7%)	42 (64.6%)
2011–2015	313 (36.2%)	290 (36.2%)	23 (35.4%)
Age at diagnosis (Mean/SD) [0]	64.1 (+−11.2)	64.1 (+−11.2)	65 (+−10.9)	p=.52
Survival in months all causes (Mean/SD) [1]	22.3 (+−28.3)	22.4 (+−28.3)	21.8 (+−28.8)
Number of patients dying within [0]				p=.714
1 year	>105	>90	<15
2 years	>110	>100	<15
5 years	418	386	>30
more than 5 years	228	211	>15
Baseline function score (Mean/SD) [0]	1.89 (+−1.85)	1.9 (+−1.87)	1.75 (+−1.58)	p=.482
Baseline function score categorical [233]				p=.37
Fit/Well	>475	438	>35
Mild/Severe	>150	144	<15
Function score (Mean/SD) [0]	2.12 (+−2.08)	2.15 (+−2.12)	1.82 (+−1.57)	p=.117
Function score categorical [248]				p=.364
Fit/Well	426 (49.2%)	391 (68.5%)	35 (76.1%)
Mild/Severe	191 (22.1%)	180 (31.5%)	11 (23.9%)
First malignant tumor [0]				p=.174
No	>70	70	<15
Yes	>790	730	>40
Tumor grade [171]				p=.497
Grade I and II	286 (33.1%)	267 (41.7%)	19 (35.8%)
Grade III and IV	408 (47.2%)	374 (58.3%)	34 (64.2%)
Breast subtype [491]				p=.71
Her2+/HR+	>60	57	<15
Her2+/HR−	>40	39	<15
Her2−/HR+	>190	180	<15
Triple Negative	>75	73	<15
Estrogen receptor status [99]				p=.71
Positive	491 (56.8%)	455 (64.4%)	36 (61%)
Negative	275 (31.8%)	252 (35.6%)	23 (39%)
Progesterone receptor status [109]				p=.382
Positive	368 (42.5%)	343 (49.2%)	25 (42.4%)
Negative	388 (44.9%)	354 (50.8%)	34 (57.6%)
Historic status [1]				p=.235
Localized	>205	190	<20
Regional	>270	253	<30
Distant	>320	307	<30
Other	>45	49	<15
Tumor size (millimeters) (Mean/SD) [123]	40 (+−41.8)	40.1 (+−43)	39 (+−24.5)	p=.76
Breast Adjusted AJCC 6th Primary tumor [123]				p=.059
T1	>170	156	<15
T2	>190	181	<15
T3	>65	56	<15
T4	>40	38	<15
Any T Mets	>265	251	<15
Breast Adjusted AJCC 6th metastasis in regional lymph nodes [109]				p=.956
N0	>295	272	<30
N1	>225	210	<30
N2	>100	97	<30
N3	>120	116	<30
Time from diagnosis to procedure [178]				p=.309
Below median	343 (39.7%)	322 (50.5%)	21 (42%)
Above median	344 (39.8%)	315 (49.5%)	29 (58%)
Time from diagnosis to procedure [178]				p=.472
Below 25th centile	>315	299	<30
Between the 25th and 75th centiles	>135	128	<15
Above 75th centile	>225	210	<20
Time from diagnosis to procedure [179]				p=.25
0–30 days	>395	375	>20
31–90 days	>80	>70	<15
More than 90 days	204 (23.6%)	<190 (29.7%)	15 (30.6%)
Diagnosis confirmation [0]				p=.804
Positive histology	>800	>750	>30
Positive cytology	<15	<15	<15
Radiology confirmation	<15	<15	<15
Unknown	<15	<15	<15
Surgery recommendation at DX [5]				p=.128
Surgery performed	610 (70.5%)	556 (69.9%)	>30
Surgery not recommended	>210 (24.9%)	< 210 (26%)	<15
Surgery not performed for other reason	<40 (4.6%)	>25 (4.1%)	<15
Surgery of primary site status at DX [0]				p=.089
None	>250	>240	<15
Tumor destruction, Resection, or Other Surgery	>600	>550	<65
Race [50]				p=.026
White	>690	>630	>45
Black	>120	>120	<15
Ethnicity [0]				p=.116
Non-Spanish-Hispanic-Latino	767	705	62
Spanish-Hispanic-Latino	<100	<90	<15
Marital status at diagnosis [47]				p=.276
Single	>140	>140	<15
Married	379 (43.8%)	350 (46.3%)	29 (46.8%)
Separated/Divorced	134 (15.5%)	122 (16.1%)	12 (19.4%)
Widowed	157 (18.2%)	142 (18.8%)	15 (24.2%)
Poverty Census [11]				p<.001
0% –<20%	651 (75.3%)	617 (77.9%)	34 (54.8%)
20%–100%	203 (23.5%)	175 (22.1%)	28 (45.2%)
Insurance status [37]				p=1
Any Medicaid	130 (15%)	120 (15.6%)	10 (16.4%)
No Medicaid	698 (80.7%)	647 (84.4%)	51 (83.6%)
Dual eligibility [696]				p=1
Not dual eligible	>115	>100	< 15
Dual eligible	>40	>30	< 15
Percentage of adults without a high school degree [11]				p=.336
Below 25th centile	225 (26%)	212 (26.6%)	13 (22.4%)
Between 25th and 50th percentile	214 (24.7%)	202 (25.4%)	12 (20.7%)
Between 50th and 75th percentile	205 (23.7%)	192 (24.1%)	13 (22.4%)
Above 75th centile	210 (24.3%)	190 (23.9%)	20 (34.5%)
Income level [11]				p<.001
Below 50th centile	>435	386	>40
Above 50th centile	>410	410	< 15
Distress score numeric (Mean/SD) [11]	2.8 (+−0.88)	2.72 (+−0.808)	3.98 (+−0.982)	p<.001
Distress score [11]				p<.001
Group 1	>350	>350	<15
Group 2	>490	444	>50

Bolded values represent statistically significant differences between groups, with significance defined as P < 0.05.

There were no differences in baseline functional scores and the percentage of patients classified as fit/well versus mild/severe functional impairment. There were no differences in any of the following: age or year at diagnosis, marital status, insurance status, dual Medicare-Medicaid eligibility, average county-level educational attainment, number of patients dying within 1 year, 2 years, or 5 years (Fig. 1, Fig. 2, Fig. 3), whether the breast cancer was the first primary tumor to cause metastatic disease, tumor grade, receptor status (HER 2, hormone receptor, estrogen receptor, or progesterone receptor), historic tumor status, tumor size, breast-adjusted AJCC score, time from diagnosis to procedure (based on percentile and duration), tumor confirmation (radiographic, histology, or cytology), surgery recommendation, type of surgery at primary cancer site, and overall survival duration (Table 1).

Fig. 1.

Kaplan–Meier plots for survival probability within one year from diagnosis showing no significant difference urban vs. rural.

Fig. 2.

Kaplan–Meier plots for survival probability within two years from diagnosis showing no significant difference urban vs. rural.

Fig. 3.

Kaplan–Meier plots for survival probability within five years from diagnosis showing no significant difference urban vs. rural.

Propensity matched analysis

After propensity score matching, patients who were living in rural settings were associated with a higher risk of all-cause mortality at 5 years (HR, 3.39 [95% CI, 1.26–9.12], p=.016) (Fig. 4). There were no differences in all-cause mortality at 1 year or 2 years after diagnosis (Table 2).

Fig. 4.

After propensity score matching, patients who were living in rural settings were associated with a higher risk of all-cause mortality at 5 years.

Table 2.

Probability of death, unadjusted and through propensity scores following spine surgery.

	Death within one year	Death within two years	Death within five years
Unmatched
Urban	(HR) 1 [Referent]	(HR) 1 [Referent]	(HR) 1 [Referent]
Rural	1.21 (0.6–2.42) [p=.599]	1.37 (0.677–2.79) [p=.378]	1.58 (0.758–3.3) [p=.221]
Matched
Urban	(HR) 1 [Referent]	(HR) 1 [Referent]	(HR) 1 [Referent]
Rural	2.01 (0.868–4.66) [p=.103]	2.31 (0.992–5.37) [p=.052]	3.39 (1.26–9.12) [p=.016]

Discussion

The present study aimed to investigate the impact of urban versus rural care on long-term outcomes in patients who received spine surgery for metastatic breast cancer to the spine. Our results indicate that there was no statistically significant difference in one-, two-, and five-year all-cause mortality between patients receiving care in urban versus rural settings in the overall population. However, with propensity matching, our findings showed a higher risk of all-cause mortality among patients living in rural areas at 5 years, indicating that patients with metastatic breast cancer to the spine living in rural areas may experience a greater risk of long-term mortality after spine surgery.

Although this is the first study, to our knowledge, to report geographic disparities among patients with metastatic breast cancer to the spine, our findings were similar to previously reported findings in the nonspinal oncology literature [[33], [34], [35]]. A recent study by Semprini et al. that investigated socioeconomic disparities of head and neck cancer treatment in the United States reported lower long-term survival rates among patients with a low socioeconomic status living in rural areas [36]. Similar findings were also reported by Vedire et al. who found a significantly worse overall survival among patients with esophageal cancer living in rural areas [37]. Both studies speculate that the poorer outcomes seen among patients living in rural areas is likely attributed by a combination of socioeconomic and systemic barriers that prevent access to timely, quality care. Prior studies have shown that spinal metastatic disease is a particularly complex diagnosis that requires frequent, surveillance and close management of medical comorbidities and systemic disease burden to increase overall survival [38,39]. Interestingly, in a New York State database study of patients with extradural metastatic spine tumors who underwent surgery, urban hospitals were associated with significantly higher 30-day mortality compared to rural hospitals [40]. Similarly, a large database study using the National Inpatient Sample (NIS) by Im et al. examined patients undergoing lumbar spinal fusion and found higher rates of perioperative complications occurring in urban hospitals compared to rural [41]. Thus, it may be speculated that urban hospitals may take on more medically complex cases which could be a driving factor for higher rates of complication and mortality. The results of the present study may be less dependent on an urban vs. rural hospital but rather patients with metastatic breast cancer to the spine living in rural areas are more prone to geographic barriers that limit timely and frequent access to important long-term surveillance and management options, which likely contributes to the higher risk of long-term mortality.

Our study demonstrated no difference in age at diagnosis, tumor stage at diagnosis, tumor subtype, time from diagnosis to procedure, or insurance status. The increased risk of death at 5 years after propensity matching may be attributed to several factors, including a higher prevalence of comorbidities and living in areas with higher poverty rates. This study demonstrated an increased likelihood of living in a county with higher poverty levels in the rural population. The rural population was more likely to reside in counties with an income level below the 50% percentile. With regards to the poverty census, the rural population was more to reside in counties in which 20%–100% of the population lives below the poverty line.

Additionally, the rural population was more likely to reside in counties with a higher overall distress score. There are likely several factors that contribute to this finding. Han et al. found in patients who undergo surgery for colorectal cancer, younger age, lower income, advanced tumor stage, poorer social support, and poorer baseline health related quality of life all led to increased distress levels [42]. While the current study did not find any differences in age at diagnosis or tumor stage, there were differences in county-level income level. Moss et al. reviewed rural-urban differences in health-related quality of life across multiple types of cancer and found that rural cancer survivors had poorer vitality, and decreased social, physical, and emotional functioning compared to both urban cancer patients and matched rural controls [43]. Persons living in rural areas may be less likely to use secondary social support services including counseling, support groups, and wellness programs [[44], [45], [46]]. These persons are also less likely to see any value in these programs [47].

The increases in county-level distress scores found in the current study may be theorized to be due to a lack of social support and resources, increased community or family poverty level, and overall worse general health metrics. Prior studies have also shown patients living in rural areas have limited access to specialized care, limited availability of advanced treatment options, and challenges with postoperative follow-up care [[48], [49], [50]]. Historically, lower rates of insured patients in rural areas has played a factor in health care disparities as well [51]. With the expansion of Medicaid through the Affordable Care Act, there is some hope that this will reduce these differences, but more research is needed [34,52].

The findings of this study suggest that further research is needed to understand the unique challenges faced by rural patients in accessing specialized care and to develop strategies to improve outcomes for this patient population. Overall, this study highlights the need for targeted interventions to improve access to specialized care for patients with metastatic breast cancer to the spine, particularly those living in rural areas, to improve long-term survival outcomes. Potential strategies to mitigate these disparities include regionalized referral networks, telemedicine-enabled multidisciplinary tumor boards, and targeted outreach programs aimed at improving access to specialized oncologic care in rural populations.

There are several limitations to this study. Due to the retrospective nature of the study, data accuracy is contingent on the accuracy of the utilized SEER-Medicare linked database. The study controlled for stage at presentation as well as tumor receptor status but could not control for true disease burden at time of initial presentation. While the SEER-Medicare database provides robust demographic and oncologic data, important clinical variables such as spinal instability scores, epidural disease burden, and detailed systemic therapy information are not captured. As such, residual confounding related to disease severity or treatment pathways cannot be excluded. Furthermore, the SEER-Medicare dataset does not include precise geographic identifiers or travel distance to treatment facilities, preventing analysis of the relationship between travel distance and survival outcomes. Referral bias may also exist whereby rural patients presenting to tertiary centers represent a subset with more advanced or complex disease compared to those treated in rural areas. Using ICD-9 codes in the SEER-Medicare dataset to identify patients with later metastatic disease may underpredict the cohort of patients with metastatic disease because of possible limitations in the follow-up care in patients with metastatic disease. Furthermore, due to the nature of this database, patients diagnosed with breast cancer before age 65 may have had prior treatment which would be unknown given they were not Medicare eligible until age 65 or older. This also introduces selection bias as patients diagnosed with breast cancer prior to age 65 must have survived until age 65 (becoming eligible for Medicare on the basis of age) to be included. Therefore, patients with more aggressive disease resulting in mortality prior to age 65 were not evaluated in the present study. The authors acknowledge the disproportionate representation of urban and rural patients within the SEER-Medicare cohort. This imbalance reflects the underlying population distribution captured within the registry and may limit statistical power when evaluating outcomes in rural populations. Although inverse probability weighting was used to mitigate confounding, residual bias may remain due to unmeasured variables not accounted for in the analysis. Future studies will evaluate the role of access to care in outcomes in spine surgery for metastatic breast cancer by assessing hospital quality, hospital and surgeon volume, and the distance from patient residence to the hospital.

Conclusion

This study was the first to demonstrate that rural residence may contribute to increased mortality rates in patients undergoing surgery for metastatic breast cancer to the spine. Given the retrospective nature of this analysis, rural residence should be interpreted as a surrogate marker for potential disparities in healthcare access rather than a causal determinant of mortality. It is important for healthcare providers to understand these differences and develop strategies to address specific barriers in accessing care.

Declarations of competing interests

The authors report there are no relevant conflicts of interest to disclose for this manuscript.