STRENGTHS AND LIMITATIONS OF THIS STUDY

A large public database was used.

A Cox proportional hazards model was applied.

To control for inflation of type I error by multiple testing, p values were adjusted by default false discovery rate procedure.

As a retrospective study, selection bias is inevitable.

It is difficult to use the cohort of other regions for external validation.

Introduction

Cervical cancer is the most common malignant tumour in the female reproductive tract and the fourth leading cause of cancer-related death among women worldwide.1 2 According to the database of the International Agency for Research on Cancer, there were more than 500 000 new cases of cervical cancer and 311 000 deaths in 2018.3 4 Due to the development of the human papillomavirus (HPV) vaccine and pelvic examination, the incidence and mortality of cervical cancer are gradually decreasing.5 However, in developing countries or low-income areas, cervical cancer is still one of the most common cancers and the main cause of cancer-related death in women.6 More than 30% of patients develop locally advanced cervical cancer in underdeveloped countries, with an extraordinarily low 5-year survival rate.7 Therefore, it is essential to understand the prognostic factors to better identify high-risk populations and guide clinical management.

Numerous factors are associated with the prognosis of patients with cervical cancer including tumour size,8 lymph node metastasis9 and race.10 11 Recently, accumulating evidence has demonstrated that socioeconomic factors were associated with the survival of patients with cervical cancer.12 13 Higher cervical cancer incidence and mortality rates were found to be associated with lower socioeconomic status.14 Low socioeconomic status has been associated with lower rates of screening15 and/or receipt of standard-of-care therapy,16 17 which have been independently associated with increased late-stage presentations and cervical cancer-specific mortality, respectively. A study by Lee et al hypothesised that increased Medicaid coverage would be associated with improved access to care and thus increase in earlier-stage diagnoses, timely treatment and better survival in cervical cancer.18 The association between multiple socioeconomic factors and survival in patients with cervical cancer requires further assessment. The American Joint Council on Cancer (AJCC) staging system is currently used for routine prognostic assessment of patients with cervical cancer.19 Nevertheless, the AJCC staging system is not comprehensive enough without taking into account other important factors such as age, race, tumour location, cancer grade and clinical treatments. Prediction models for cervical cancer prognosis are therefore still urgently needed. Furthermore, to date, no study constructed a predictive model that includes socioeconomic factors to predict the survival of patients with cervical cancer. A more practicable and effective predictive model compared with adopted staging systems should be introduced to assist clinicians to identify patients at higher risk of poor prognosis.

This study aimed to evaluate the associations of socioeconomic factors with overall survival (OS) in women diagnosed with cervical cancer. Predictive models to predict OS among women diagnosed with cervical cancer based on socioeconomic factors and the AJCC staging system were also constructed. The ability to identify vulnerable patients who have disadvantaged socioeconomic status is important to develop individualised risk stratification and guide targeted interventions.

MethodsStudy design and participants

In this retrospective cohort study, data between 2007 and 2011 were extracted from Surveillance, Epidemiology, and End Results (SEER) Programme (www.seer.cancer.gov; SEER*Stat Database: Incidence-SEER 18 Regs Custom Data (with additional treatment fields), 2018 November Sub (1975–2016 varying); linked to county attributes—total US, 1969–2017 Counties, National Cancer Institute, DCCPS, Surveillance Research Program, released 2019 April, based on the 2018 November submission).20 The year 2007 was selected as the first year, as several covariates were introduced to the database in 2007. Sponsored by the National Cancer Institute, the SEER Programme collects demographic, clinicopathological and survival data from 18 population-based cancer registries (SEER-18) in the USA. Since the SEER-18 covers 27.8% of the population in the USA with a typical distribution, it is thought to be representative of the US population as a whole.21 Women diagnosed with cervical cancer were identified using the International Classification of Diseases (ICD-10) codes to include cancer of the endocervix (ICD-10: 53.0), exocervix (ICD-10: 53.1), overlapping lesion of cervix uteri (ICD-10: 53.8) and unspecified cervix uteri cases (ICD-10: 53.9). Patients with cervical cancer aged ≥18 years, with only one primary tumour and with active follow-up, were included. Patients with distant metastasis at diagnosis, confirmed only by autopsy or death certificates, and patients with incomplete information on the variables included in the analysis were excluded.

Socioeconomic factors

Socioeconomic factors, including marital status, insurance status, local education level (percentage with <high school education), local unemployment rate, residence and local median household income were assessed in this study. Two variables, marital status and insurance status, were determined at the patient level. Marital status was classified as married, single (never married), separated/divorced and widowed, while insurance status was characterised as insured, uninsured and unknown. County-level estimates of median household income, percentage with <high school education and percentage of unemployment rate were assigned to patients according to their state-county recode. The area of residence was classified as a metro area or a non-metro area based on the 2003 Rural-Urban Continuum Codes. Local median annual income, local unemployment rate and local education level were converted into categorical variables according to the median.

Potential covariates

Demographic and clinical variables included age (>65 years, ≤65 years), race (black, other (American Indian/Alaska Native, Asian/Pacific Islander) and white), treatments (no radiation and/or cancer-directed surgery, both radiation and surgery were given), tumour/node (TN) category, clinical stage (I, II, III/IV), tumour size and chemotherapy (no/unknown or yes). The clinical stage and TN categories were measured using the seventh edition of the AJCC staging system.22

Outcomes

The outcomes of this study were 1-year OS and 5-year OS. OS was defined as the time (in months) from the date of diagnosis until death due to any cause within the follow-up period. If the patient died, the follow-up was terminated.

Development and evaluations of the predictive models

We first conducted an AJCC staging system for predicting 1-year and 5-year OS of patients with cervical cancer. Then, the prognostic socioeconomic factors of cervical cancer were initially determined by univariate Cox proportional hazards models. Variables with p<0.05 were entered into the multivariable Cox proportional hazards models. The AJCC staging system and independent socioeconomic prognostic factors determined by the multivariate analysis were used to construct predictive models for 1-year and 5-year OS. The concordance indexes (C-indexes)23 were used to assess the performance of the predictive models by applying AJCC models as a reference. The value of the C-index ranges from 0.5 to 1.0, with a higher C-index indicating a better ability to separate patients with different survival outcomes. The net reclassification improvement (NRI) was proposed to detect differences in model discrimination.24 An external validation set was generated to validate the predictive models using data from 18 states of the SEER Database between 2012 and 2016.

Statistical analysis

Measurement data are presented as median with quartile (M (Q1, Q3)), and the Mann-Whitney U test was used for intergroup comparison. Counting data were presented as the number of cases/constituent ratio (n (%)), and the χ2 test or Fisher’s exact test was applied for intergroup comparison.

Univariate and multivariate Cox proportional hazards models were employed to estimate and analyse the associations of socioeconomic factors with OS of patients with cervical cancer. In the case of significant results for multicategorical variables, additional overall tests were performed. The R language ‘anova()’ was used to compare models with covariates (age, race, radiotherapy, tumour size, chemotherapy and AJCC) and models with covariates plus insurance/marital status. Data distributions of baseline socioeconomic factors from 2007 to 2011 and from 2012 to 2016 were visualised using stacked histograms. To control for inflation of type I error by multiple testing, p values were adjusted by default false discovery rate procedure. P<0.05 or adjusted p<0.05 was considered the threshold of statistical significance. Kaplan-Meier curves were used to assess the prediction of OS based on the number of risk factors of socioeconomic variables. Kaplan-Meier survival curves were analysed using pairwise log-rank tests. HRs and 95% CIs were used as the evaluation indexes. All descriptive statistics were performed using SAS V.9.4 (SAS Institute). We used the ‘survival’ package (2023; A Package for Survival Analysis in R. R package V.3.5-5 (https://CRAN.R-project.org/package=survival)).25

Patient and public involvement

None.

ResultsClinical features and characteristics of included participants

Initially, 13 080 patients between 2007 and 2011 were identified from the SEER Database. Based on the inclusion and exclusion criteria, 5954 patients who were diagnosed with cervical cancer were included in this study. At the end of the follow-up period, 5820 patients had 1-year mortality and 5460 patients had 5-year mortality. A flow diagram of patients’ selection is shown in figure 1. Overall, 88.02% of patients were aged >65 years, and 11.98% of patients were aged ≤65 years. A higher proportion of women (4594; 77.16%) were white. No radiation and/or cancer-directed surgery was performed on 72.51% of patients, and 27.49% received both radiation and surgery. Regarding the TN stages, more than half of the patients (66.26%) were at the T1 stage and more than half of the patients (81.31%) were at the N0 stage. Most of the local patients (91.00%) were uninsured. More than half of the local population had a median annual income of ≥US$56 270. The marital status data indicated that there were 2882 (48.40%) married patients, 895 (15.03%) separated/divorced patients, 1743 (29.27%) single patients and 434 (7.29%) widowed patients. Age, race, radiation/surgery treatment, tumour size, AJCC T stage and N stage, chemotherapy, AJCC stage, local education level, local unemployment rate, marital status and local median annual income were significantly different between patients who survived and those who died (all p<0.05). The baseline characteristics of the study population are depicted in table 1. The stacked histograms showed that the socioeconomic factors were relatively balanced between 2007–2011 and 2012–2016 in the SEER Database. The distribution of baseline socioeconomic factors between 2007–2011 and 2012–2016 is depicted in online supplemental figure 1.

Table 1

Characteristics of included participants

Figure 1

Flow diagram for patient selection. AJCC, American Joint Council on Cancer.

Socioeconomic factors and OS in patients with cervical cancer

Regarding OS, lower education level (HR: 1.15, 95% CI: 1.04 to 1.27, p=0.005) and being widowed (HR: 1.28, 95% CI: 1.06 to 1.55, p=0.009) were associated with worse OS for patients with cervical cancer. Having insurance (HR: 0.75, 95% CI: 0.62 to 0.90, p=0.002), earning a local median annual income of ≥56 270 (HR: 0.83, 95% CI: 0.75 to 0.92, p<0.001) and being married (HR: 0.79, 95% CI: 0.69 to 0.89, p<0.001) were related to better OS in patients with cervical cancer. The associations of socioeconomic factors with OS of patients with cervical cancer are available in online supplemental table 1. Multivariable Cox proportional hazards analysis showed that a lower local education level (HR: 1.47, 95% CI: 1.19 to 1.80, p<0.001) and being widowed (HR: 1.83, 95% CI: 1.55 to 2.16, adjusted p=0.003) were factors associated with 1-year OS of patients with cervical cancer. Better 1-year OS was observed in patients with a higher median annual income (HR: 0.72, 95% CI: 0.59 to 0.89, p=0.002) and in married patients (HR: 0.64, 95% CI: 0.57 to 0.73, adjusted p=0.002). The associations of socioeconomic factors with 1-year OS of patients with cervical cancer are shown in table 2. Lower local education level (HR: 1.16, 95% CI: 1.04 to 1.29, p=0.006) and being widowed (HR: 1.32, 95% CI: 1.08 to 1.61, adjusted p=0.018) correlated with worse 5-year OS in patients with cervical cancer. Having insurance (HR: 0.80, 95% CI: 0.66 to 0.96, adjusted p=0.04) and a higher local median annual income (HR: 0.84, 95% CI: 0.76 to 0.94, p=0.002), and being married (HR: 0.79, 95% CI: 0.69 to 0.90, adjusted p=0.003) were related to preferable 5-year OS in patients with cervical cancer. Table 3 presents the associations of socioeconomic factors with 5-year OS of patients with cervical cancer. The comparison of models with covariates and models with covariates plus insurance/marital status showed that there were no significant differences between models with and without insurance (online supplemental table 2). Kaplan-Meier curve results indicated that the more risk factors were present in socioeconomic variables, the worse the 1-year and 5-year OS were. Kaplan-Meier survival analyses of 1-year and 5-year OS with different numbers of risk factors of socioeconomic variables among patients with cervical cancer are shown in online supplemental figures 2 and 3.

Table 2

Association of socioeconomic factors with 1-year OS for patients with cervical cancer

Table 3

Association of socioeconomic factors with 5-year OS for patients with cervical cancer

Predictive model constructions and evaluations for predicting OS of patients with cervical cancer

We first constructed the AJCC staging system. Based on the AJCC staging system and socioeconomic factors of the local education level, local median annual income and marital status, the predictive models for predicting 1-year and 5-year OS were established. The C-index of the predictive model for predicting 1-year OS (C-index: 0.781, 95% CI: 0.760 to 0.802) was higher than that of the AJCC staging system (C-index: 0.741, 95% CI: 0.719 to 0.763). Similarly, a higher C-index of the predictive model for predicting 5-year OS (C-index: 0.744, 95% CI: 0.732 to 0.757) was found in patients with cervical cancer compared with the AJCC staging system (C-index: 0.715, 95% CI: 0.703 to 0.727). The predictive models were improvements on the AJCC staging system with NRI being 0.21 (95% CI: 0.13 to 0.25) in 1-year OS and 0.03 (95% CI: −0.06 to 0.05) in 5-year OS, respectively. Performance evaluations of predictive models for predicting OS in patients with cervical cancer are available in table 4. The validation also indicated that the predictive models had better predictive performance compared with the AJCC staging system (table 4).

Table 4

Performance evaluations of predictive models for predicting OS in patients with cervical cancer

Discussion

Previous studies have been conducted in several areas to explore the relationship between cervical cancer survival and socioeconomic factors.15 To the best of our knowledge, this is the first attempt to develop predictive models to predict survival in patients with cervical cancer based on socioeconomic factors. A lower local education level was associated with a higher risk of 1-year mortality and 5-year mortality in patients with cervical cancer. Being widowed compared with being single correlated with worse 1-year OS and 5-year OS in patients with cervical cancer. Better 1-year and 5-year OS were observed in patients with a higher local median annual income. Those who were married had better 1-year and 5-year OS compared with those who were single. Having insurance was also related to preferable 5-year OS in patients with cervical cancer. Our predictive models were conducted based on AJCC staging system, local education level, local median annual income and marital status. Our predictive models showed better predictive performances compared with the AJCC staging system.

For predicting the survival of patients with cervical cancer, several predictive models have been widely conducted.26–28 A study based on the SEER Database to establish nomograms for predicting the survival rate of patients with cervical cancer who undergo radiation therapy, which showed good predictive performance of nomograms, however, lacked external validation.27 Yan et al26 developed a prognostic nomogram for OS of patients aged 50 years or older with cervical cancer. However, this nomogram was based only on women older than 50 years old, and the predictive model was not externally validated. Feng et al29 constructed nomograms to predict the OS and cancer-specific survival of patients with stage IIIC1 cervical cancer. The C-index for the nomogram of OS was only 0.687. A recent study by Fan et al28 developed a prediction tool for the OS of patients with cervical cancer aged 25–69 years. However, the predictive model was validated in the internal validation set. Moreover, none of the above studies developed models based on socioeconomic factors. In this study, socioeconomic factors were independent predictive factors for OS and were included in the predictive models. Additionally, we compared the C-indexes of the predictive models with that of the AJCC stage and found that the C-indexes of the predictive models for predicting OS were higher than that of the AJCC staging system. Moreover, external validation of the predictive models was carried out, with the C-index being above 0.700.

Socioeconomic factors have been reported to be implicated in cancer survival outcomes.30–32 In this study, socioeconomic factors including local education level, local median annual income and marital status were significantly associated with the OS of patients with cervical cancer. Our findings are supported by a study by Ibfelt et al that uncovers that socioeconomic factors such as education and income influence survival after cervical cancer diagnosis and treatment.31 A number of reasons may explain the association between socioeconomic factors and cervical cancer survival. First, lower education at the time of diagnosis was associated with advanced cervical cancer, suggesting that the cancer stage partly explains socioeconomic differences in survival.31 Second, lower education and income were also associated with delayed screening and low HPV vaccination rates.33 In addition, women of lower social status also tend to have comorbidities and risky health behaviours, such as smoking, which may affect cervical cancer incidence, comorbidities, treatment options and survival rates.34 Previous research on the association of marital status and cancer suggested that married patients have an advantage in early diagnosis of cancer, including cervical cancer.35 In addition, married patients are able to receive more comprehensive adjuvant treatment, leading to a better prognosis for cervical cancer.36 Interventions to promote cervical cancer prevention strategies should be targeted at women and girls with lower education levels and lower income. Taken together, these findings underscore the importance of socioeconomic factors in cervical cancer outcomes.

The present study aimed to investigate the association of socioeconomic factors with the OS of patients with cervical cancer and establish predictive models to predict OS, thereby providing potentially relevant data on resource allocation targeting vulnerable groups for increased survival. Our findings suggest the importance of paying attention to socioeconomic factors for cervical cancer prognosis. Regions or countries with low socioeconomic factors should improve HPV vaccination rates and develop strategies to improve cancer screening. Furthermore, assessment of patient prognosis is crucial to formulate individualised treatment and follow-up plans.

Our study also has several limitations. First, as a retrospective study, this research filtered data from datasets and excluded patients with missing data for the collected variables, leading to selection bias. Second, due to the limitations of data in the SEER Database, it is difficult to include all data related to socioeconomic factors in the study. Third, it is difficult to use the cohort of other regions for external validation. Fourth, in this study, possible demographic and clinical characteristics, and other influencing factors were adjusted based on the SEER Database, but the possible confounding variables in the follow-up process, such as living habits and other interventions, were not obtained and adjusted. Fifth, socioeconomic variables can affect outcomes through different causal pathways and at the individual, household and neighbourhood levels. However, as several socioeconomic variables were done at an ecological level in the study, we did not adjust for micro-level socioeconomic variables that could have a potential confounding or modifying effect on the relationship between socioeconomic factors and survival of patients with cervical cancer. Sixth, changes in socioeconomic factors (such as marital status) may occur after database registration or during treatment, thus affecting patient survival. Finally, the Cox proportional hazards model depends on the assumption of a constant hazard over time. Therefore, prospective cohort studies are necessary to further explore the link between socioeconomic factors and survival in cervical cancer.

Conclusion

Compared with the AJCC staging system, our predictive models incorporating socioeconomic factors were more comprehensive with a high degree of C-index, which may highlight the potential clinical application of incorporating socioeconomic factors in predicting OS in cervical cancer.

Data availability statement

Data from this analysis can be available with the consent of the corresponding author upon reasonable request. SEER data can be accessed directly via www.seer.cancer.gov.

Ethics statementsPatient consent for publication

Obtained.

Ethics approval

The requirement of ethical approval for this study was waived by the Institutional Review Board of the First Affiliated Hospital of Xinjiang Medical University, because the data were accessed from SEER (a publicly available database). All methods were carried out in accordance with relevant guidelines and regulations (Declaration of Helsinki). Due to the retrospective nature of the study, informed consent was waived.