ORIGINAL RESEARCH Influence of Disadvantaged Socioeconomic and Demographic Status on Overall Survival in Patients with Kaposi Sarcoma

Kaposi

nodes may occur. 2 The classic form of KS typically arises in older men of Mediterranean and Eastern European descent on the lower extremities, as well as an endemic form predominantly in sub-Saharan Africa primarily causing lymph node involvement in pediatric patients. 3In iatrogenic and HIV patients, it may present with skin, oral (palatal and gingival), and less commonly, visceral involvement. 4Typically, iatrogenic and AIDSrelated KS respond well to immune reconstitution and anti-retroviral therapy respectively, thus many patients present with breakthrough localized skin involvement. 5solated skin lesions may be treated with excision, liquid nitrogen, or vincristine injection, while cytotoxic chemotherapy is the standard of treatment for systemic disease. 5,6 is an AIDS-defining illness and the second-most common neoplasm associated with AIDS. 7Among people with HIV (PWH) living in the United States from 2000-2015, KS rates were elevated 521-fold compared to the general population and have declined at an annual percentage change of −6%. 8KS most often occurs in PWH with uncontrolled viral replication and decreased CD4 T cell levels. 9Since the advent of highly active antiretroviral therapy (HAART), incidence of KS and other AIDS-related malignancies such as non-Hodgkin's Lymphoma have declined substantially while non-AIDS related malignancies have increased in patients with HIV. 10 However, new and recurrent forms of KS despite adequate viral suppression have been documented as an novel issue in PWH. 9 Previous studies by Royse et al. examining the Surveillance, Epidemiology, and End-Results (SEER) database demonstrated that in men younger than 55 years, the annual percent change (APC) for KS incidence significantly decreased for white men between 2001 and 2013 (APC -4.52, p=0.02), whereas the APC for African American (AA) men demonstrated a non-significant decrease from 2000-2013 (APC -1.84, p=0.09). 11Among AA men in the South, however, APC has significantly increased between 2000 and 2013 (+3.0, p=0.03). 11dditionally, compared with white men diagnosed with KS during the same time period, AA men were also more likely to die from all causes and KS cancer-specific causes (aHR 1.52, 95% CI 1.34-1.72,aHR 1.49, 95% CI 1.30-1.72respectively). 11herefore, despite recent advances in the treatment of PWH using HAART, KS continues to be one of the most common malignancies that occurs in PWH even in the presence of adequate viral suppression.Moreover, there appear to be geographic and racial disparities in both incidence and survival of KS in the US that require further analysis.
In an effort to better understand the current outcomes of KS patients on a national scale, we examined the National Cancer Database (NCDB) to study the epidemiology and demographics of KS patients, and how these factors impact overall survival (OS) on a national scale.As a large oncology database that compiles data from over 1,400 cancer programs and including about 75% of newly diagnosed cancers in the United States, it is the largest database of these cancers currently available. 12

Patient Selection
Patients diagnosed with KS (identified by ICD-O-3 histology code 9140) from 2004-2017 were collected from the NCDB.This included demographic, socioeconomic, and survival data for patients aged 18 and up.

METHODS
After selecting for variables of interest (Figure 1), a total of 4,034 patients met study inclusion criteria.Data from the NCDB is deidentified, and as such this project was determined to be institutional review board (IRB) exempt give that it is not human subjects research.Analyses were performed using IBM SPSS Statistics for Mac, Version 27 (Armonk, NY: IBM Corp), with propensity score matching utilizing the PSM 2.0.1 and FUZZY 2.0.1 extensions with Python 3.8 module in SPSS.

Patient Characteristics
Baseline demographic, socioeconomic, and survival data was determined using NCDB codes for the entire cohort.These features were also compared between racial identities and ethnicity.Race is patient-identified and was reported as White, Black, Asian and Pacific Islander (AAPI), and other.The other group included smaller racial groups and those who did not identify according to the available categories.Spanish/Hispanic ethnicity was also patient-identified.Patients who were identified to be Spanish/Hispanic based on surname alone were excluded.The Charlson-Deyo comorbidity score was used to define the presence of comorbidities based on International Classification of Disease (ICD) codes for common medical conditions, and included values of 0, 1, 2, and ³3.Higher scores are typically associated with greater mortality and morbidity. 13The 2016 American Community Survey Data was used to determine income and education level quartiles of patient zip code of residence.Patient insurance status was defined as uninsured, privately insured, or government insured.Facility types were defined as nonacademic and academic facilities.Categorical variables were first compared by Chi-square and Fischer's exact test for significance.Age at diagnosis was compared between groups using one-way ANOVA with Dunnett's T3 post-hoc comparison.Statistical significance was defined as p value <0.05.

Survival Analysis
Mean and median overall survival (OS) in months from diagnosis were calculated using Kaplan Meier analysis, and significance was evaluated using pairwise log-rank tests.A multivariate Cox proportional hazards model was used to determine the risk of death based on demographic and socioeconomic variables of interest.

Propensity Score Matched Analysis
Using a multivariate logistic regression model that included age, sex, ethnicity, median education and income level of zip code, insurance status, facility location, facility type, HIV status, and Charlson-Deyo comorbidity score, a propensity score was determined by the conditional probability of a patient identifying as Black.These factors were chosen by their significant variation between race groups and theoretical impact on survival.White patients were then matched 1:1 to Black patients using this propensity score, with preference for exact matches and a match tolerance of 0.01.Following matching, Kaplan Meier survival analysis for median overall survival was repeated.

RESULTS
Figure 1.Selection of study patient population and exclusion criteria ethnicity are described in Tables 1 and 2, respectively.
Compared to White patients, Black patients were younger at diagnosis (mean age 58 versus 51 years old, p<0.001), with 75.3% of Black patients diagnosed before age 55 compared to 53.5% of White patients (p<0.001).Other racial and ethnic groups presented at a similar age compared to White patients (Tables 1 and 2).Black patients (37.7%) were more likely to have Charlson-Deyo scores of 3 and above compared to White (29.5%) and AAPI patients (34.6%) (p=0.001).82.7% of Black patients were HIVpositive compared to 62.7% of White and 61.5% of AAPI patients (p<0.001).Compared to White patients (9.2%), 12.8% of Black patients and 13.5% of AAPI patients were uninsured (p<0.001).Hispanic patients were more likely to be uninsured (16.7%) compared to non-Hispanic patients (9.1%) (p<0.001).Black and Hispanic patients resided in lower income areas, with 39.5% and 31.0%,respectively, living in zip codes with an average income below the 25 th percentile, compared to 18.7% of White and 11.5% of AAPI patients (p<0.001).A greater percentage of Black (37.9%) and Hispanic (57.8%) patients resided in zip codes where 17.6% or more of residents did not attain a high school degree, compared to 27.4% of White patients (p<0.001).

Survival Analysis
OS for all race groups was analyzed using Kaplan-Meier curves.(Figure 2) The median OS for all KS patients was 133 months (95% confidence interval [CI] 119-148 months).Black patients had a shorter median OS of 99 months (95% CI 73-124 months) compared to White patients with a median OS of 140 months (95% CI 122-158 months) (p<0.001).All other minority groups trended towards improved OS compared to White patients.The AAPI, Other, and Hispanic cohorts did not reach median survival during the followup period (p=0.584,0.473, and 0.001, respectively).
To assess for the independent contribution of socioeconomic and demographic variables, multivariate Cox regression was performed for age, sex, race, ethnicity, Charlson-Deyo score, education, income, insurance, location, facility type, and HIV status (Table 3).Age greater than 70 years old was associated with reduced OS compared to patients aged 40-54 years old (hazard ratio [HR] 2.06, 95% CI 1.67-2.55,p<0.001).Black race was independently associated with reduced OS compared to White race (HR 1.31, 95% CI 1.16-1.47,p<0.001).No statistically significant difference was seen in survival for other racial groups.Hispanic ethnicity was associated with improved survival compared to non-Hispanic ethnicity (HR 0.80, 95% CI 0.69-0.93,p=0.003).Residence in areas with high educational attainment was associated with improved survival, with a survival benefit seen for patients residing in areas with no greater than 10.8% of residents not graduating high school (HR 0.83, 95% CI 0.71-0.98,p=0.026).Having private insurance was associated with improved survival (HR 0.74, 95% CI 0.62-0.89,p=0.002); however, no statistically significant difference was seen in survival with government insurance compared to having no insurance.Improved survival was associated with academic centers (HR 0.76, 95% CI 0.68-0.83,p<0.001).Positive HIV status portended a poorer prognosis (HR 1.93, 95% CI 1.61-2.32,p<0.001), as did any Charlson-Deyo score greater than 0 (HR 1.49, 95% CI 1.26-1.76,p<0.001).Sex, income, and facility location did not independently have any statistically significant effect on survival.To our knowledge, our study is the first to describe the impact of key demographic and socioeconomic factors on survival in KS on a large national scale.Similar to prior studies, we found that Black patients were diagnosed at younger ages and had poorer survival outcomes compared to White patients, and that patients living with HIV have reduced OS. 14,15,16On a national level, it is wellestablished that URM individuals are more likely to have lower socioeconomic status, experience poorer healthcare outcomes regardless of SES, and be disproportionately affected by HIV. 17,18Our study builds on these past findings with the addition of multivariate statistical analysis, comprehensive socioeconomic factors, and propensity score matching to help determine what features have the most significant impact on these outcomes.
We found significant variation in population characteristics between KS patients of White and URM background.Black patients were more likely to be underinsured and reside in area of significantly lower education and income levels.Moreover, they were more likely to be affected by multiple comorbidities and be diagnosed with HIV.On multivariate analysis, Black race was independently associated with reduced OS, which persisted when controlled for covariates including age, sex, ethnicity, median education and income level of zip code, insurance status, facility location, facility type, HIV status, and Charlson-Deyo comorbidity score.This may be due to comorbidities that are not accounted for by the Charlson-Deyo Comorbidity Index, or additional economic and social factors not represented in the available codes, including structural racism and other surrogate markers of healthcare access.Another reason for lagging survival in Black patients living with HIV could be disparities in access to HAART.Furthermore, prior research has shown that racial and ethnic minorities, including Black patients, have decreased ART adherence, which could be attributed to historical discrimination and racism experienced by these patients causing mistrust of the medical system and increased treatment-associated psychological distress. 19,20ultivariate analysis indicated that in addition to Black race, the features of advanced age, HIVpositive status, and Charlson-Deyo score greater than 0 were each independently associated with decreased survival, while Spanish/Hispanic ethnicity, private insurance, residence in areas of high educational attainment, and treatment at academic centers were associated with improved survival.Past work has indicated that comorbidities such as diabetes may promote susceptibility to developing KS, which may explain the detrimental effect of higher Charlson-Deyo score on KS survival. 21everal studies have noted that despite poorer socioeconomic factors, including residence in areas with reduced educational attainment and income, improved OS has been seen in Spanish/Hispanic patients compared to non-Spanish/Hispanic patients  in several malignancies, although the reason for this is not well-understood. 22ivate insurance is known to confer a survival benefit across cancers, which our study replicated; however, our study revealed no survival benefit of having government insurance compared to having no insurance at all. 23Prior studies have shown that patients who are uninsured or have Medicaid insurance experienced worse survival across a variety of malignancies. 24Greater study is needed to understand why survival continues to lag for cancer patients using government insurance.Improved KS survival was noted with increased education, a trend which has been described in other cancers and attributed to a variety of causes, including stronger psychosocial support and earlier stage at diagnosis. 25Treatment at an academic center also was associated with improved survival, which has been demonstrated by similar work across malignancies.It is hypothesized that access to specialized therapies and enhanced care logistics at academic medical centers could be contributors to improved survival. 26,27ere are several limitations to this study.The NCDB is limited to Commission on Cancer accredited facilities, which covers approximately 75% of new cancer cases in the U.S. Thus, a significant fraction of hospitals may be underrepresented in this data due to selection bias, particularly against rural critical access hospitals.Moreover, the NCDB is a hospital-based registry, so generalizability on a populationlevel is more limited.Racial and ethnic identity in the NCDB is self-identified and are provided as broad categories, therefore racial and ethnic identities may not be fully described by the available categories.While powerful due to its size, the NCDB also has limited information on treatment and diseasespecific features and/or outcomes.

DISCUSSION
In summary, our study reveals the impact of key socioeconomic and demographic factors on survival in KS in the NCDB.Notably, we highlight significant racial disparities with poor survival outcomes in Black patients with KS, which persisted even after controlling for socioeconomic features and other surrogate markers of healthcare access in a 1:1 propensity-score matched model.While we propose unaccounted comorbidities and decreased access to HAART in the setting of discrimination and systemic racism as potential reasons for these survival disparities, more study is warranted to investigate why Black patients in KS continue to have poorer outcomes.Other socioeconomic and URM groups were also noted to have disparate insurance status, highlighting the importance of healthcare access in this population.We call for greater efforts in promoting access to healthcare and providing equitable medical care to all KS patients.

Figure 3 .
Figure 3. Kaplan Meier curves comparing median survival between Black and White patients in 1:1 propensity score matching model.

Table 1 .
Demographic and socioeconomic features of KS by race.

Table 2 .
Demographic and socioeconomic features of KS by ethnicity.
Figure 2. Kaplan Meier curves comparing survival between race groups.Median survival for White and Black patients are noted.

Table 3 .
Cox regression for demographic and socioeconomic factors of all KS patients.