Association of insurance status among cancer patients and survival outcomes: a systematic review and meta-analysis.

Introduction
Since the 21st century, cancer has posed a major challenge to social, economic, and healthcare systems worldwide. In 2022, there were about 20 million new cancer cases and 9.7 million cancer-specific deaths (including non-melanoma skin cancers) [1]. Genetic predispositions, tumor biology, treatment regimens, and socioeconomic status have been well-established determinants of disparities in cancer survival outcomes, both within and across countries [2, 3]. From the health policy perspective, insurance status increasingly represents a structural determinant of disparities in cancer survival outcomes [4]. Although the expansion of insurance coverage in cancer care has reduced the financial burden for patients and contributed to improved survival, the variation in insurance enrollment decisions results in patients making different therapeutic choices due to financial concerns about rising cancer care costs, which may exacerbate survival disparities across insurance statuses [5, 6]. For example, the same national health insurance status in Canada may have contributed to Canadians’ improved survival compared to Americans at the same low-income level [7, 8]. Patients with Medicaid or uninsured status may experience a higher cancer mortality rate than patients who are privately insured [9–11], and Chinese cancer patients with favorable insurance status also have higher survival rates [12, 13]. Some studies found no correlation between survival and insurance status [14–16].
However, most existing research focuses on single-country and single-cancer contexts; international studies on the relationship between insurance status and cancer survival are understudied. To address this gap, we performed a systematic review and meta-analysis focusing on five major cancer types: breast, prostate, lung, colorectal, and liver. These cancers were selected based on their substantial contribution to the global cancer burden in terms of incidence, mortality, and economic cost. According to the International Agency for Research on Cancer (IARC), lung cancer was the most frequently diagnosed cancer worldwide (2.5 million new cases, 12.4%), followed by female breast (11.6%), colorectal (9.6%), prostate (7.3%) in 2022, while four leading cause of cancer death is lung (1.8 million deaths, 18.7%), colorectal (9.3%), liver (7.8%), female breast (6.9%) cancers [1]. Lung, colorectal, breast, and liver cancers are expected to incur the highest economic burdens globally, collectively comprising over 40% of the projected $25.2 trillion in cancer-related healthcare costs between 2020 and 2050 [17]. By quantifying survival disparities across insurance statuses, this study aims to provide a clearer understanding of the potential impact of structural disparities in insurance coverage on cancer survival inequalities and supply evidence for developing equity-oriented insurance policy interventions.

Methods

Study protocol
This article followed the PRISMA 2020 [18] guidelines (Supplementary eTable 1). The protocol of this study was submitted for PROSPERO (CRD42023460395). The protocol detailed the research objective, eligibility criteria, data sources, search strategy, risk of bias assessment, and statistical analysis plan, providing a predefined framework that was adhered to throughout the review process.

Search strategy
Cohort studies published between 1 January 2000 and 15 July 2025 were searched from MEDLINE, PubMed, Embase, and Scopus. We used keywords such as “insurance status”, “medical insurance”, “insurance coverage”, “survival”, and “treatment outcomes” to identify related studies across databases (Supplementary eTable 2). The search strategy focused on identifying studies that examined associations between insurance status, health insurance coverage, and survival outcomes in cancer patients. We included peer-reviewed, English-language cohort studies that reported associations between insurance status and survival outcomes among patients diagnosed with one of five major cancer types: lung, breast, colorectal, liver, or prostate. Excluded from consideration were abstracts, conference proceedings, literature reviews, meta-analyses, editorials, comments, gray literature, and other publications that contained little evidence-based data. In addition, we manually searched the reference lists to supplement omitted studies.
Studies assessed for eligibility must meet al.l of the following exclusion criteria: (1) Studies with ambiguous insurance status classification (e.g., combining public and private insurance without differentiation); (2) did not report survival outcomes or lacked sufficient statistical estimates; (3) examined populations diagnosed before 2000 or used The Surveillance, Epidemiology, and End Results (SEER) Program and The National Cancer Data Base (NCDB) data before 2007; (4) focused on other cancers or comparisons before and after insurance coverage expansion without different insurance statuses; (5) reported inappropriate populations (e.g., populations diagnosed with cancer at age ≤ 15 years), results (e.g., absence of HR or reporting of non-survival endpoints), or study designs (e.g., cross-sectional studies); (6) or full-text data not available.
Two independent reviewers (J.T. and L.Z.) first screened all records by title and abstract. Full texts of the selected articles were then reviewed in detail to assess eligibility based on the predefined criteria. Any disagreements between the reviewers were settled by a third reviewer (N.Z.).

Quality assessment
Two independent reviewers (J.T. and L.Z.) used the Newcastle-Ottawa Scale to assess the methodological quality of cohort studies [19]. The Newcastle-Ottawa Scale (NOS) was used to assess selection, comparability, and outcome domains in cohort studies. Studies with NOS ≥ 7 were considered to have high quality.
In addition, the overall risk of bias at the survival outcome level was evaluated using the ROBINS-I (Risk Of Bias In Non-randomized Studies of Interventions), which is specifically designed for non-randomized studies of interventions [20]. It covers seven domains: bias due to confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported result. Specifically, confounding bias is a major concern given that insurance status may be associated with unmeasured variables, such as socioeconomic factors, disease characteristics and treatment processes. We assessed whether the included studies adjusted for these confounders using multivariable Cox models or propensity score methods, particularly for socioeconomic status, tumor-related characteristics, and treatment-related variables. Selection bias was also considered, especially about potential immortal time bias. We carefully evaluated whether follow-up began at the time of cancer diagnosis for the studies utilizing cancer registries. As all included studies were retrospective cohort designs, baseline and outcome data were collected for all eligible participants, minimizing concerns of post-exposure selection. The risk of bias due to misclassification of the exposure (insurance status) was generally low, as insurance information was typically recorded at diagnosis. However, some insurance misclassification could not be ruled out due to unobserved insurance changes in coverage after diagnosis. Since insurance status was not an assigned intervention, the domain of bias due to deviations from intended interventions was considered less applicable. We also evaluated whether these studies reported the reasons for missing data and the proportion of missingness. Measurement bias for survival outcomes was assessed by examining whether the included studies reported that survival data were obtained through active follow-up, death certificates, or linkage to vital statistics and cancer registries. Selective reporting bias was evaluated by comparing study protocols, methods, and reported outcomes to detect any potential selective outcome or subgroup reporting. Studies that received multiple “serious” or “critical” ratings across domains were classified as having a high risk of bias.

Data extraction
J.T. and N.Z. extracted the following information: the first author’s last name, year of publication, country where the study was conducted, sources of data, basic patient characteristics, sample size, hazard ratios (HRs), 95% CI, covariates in the adjusted model, time of diagnosis, insurance status, mortality endpoints, outcome assessment, and duration of follow-up. HR by subgroups of the study, such as gender, race, staging, and different Cox models, were also included. Studies involving the same population during overlapping periods were considered for inclusion if they reported different cancer types, stages, or diverse population characteristics.

Statistical analysis

Definition and classification
Insurance status was defined as the first recorded at the time of cancer diagnosis, in line with standard practice in cancer registry-based epidemiological studies. Reference groups were selected based on policy-relevant contrasts and the dominant insurance structures within each country. For instance, in the United States (US), private insurance was used as the referent category because it generally represents the most comprehensive coverage. The primary outcome of interest was overall survival (OS) and the subsequent outcome was cancer-specific survival (CSS).

Country-specific meta-analyses
Due to substantial institutional and population-level differences in national health insurance systems, we conducted separate meta-analyses by country. These differences not only influence the categorization of insurance types, but also access to timely diagnosis and treatment, thereby affecting cancer survival outcomes. For example, the US insurance system is characterized by mixed one, where publicly financed government Medicare and Medicaid coexist with privately financed market coverage; while China has tiered multilevel public insurance schemes with structural disparities: Basic Medical Insurance for Urban Employees (UEBMI) and Basic Medical Insurance for Urban and Rural Residents (URRBMI). Pooled analysis across countries may obscure true within-country effects [21]. Country-specific analyses allow for greater interpretability within consistent health system contexts.
Several studies from the US reported multiple survival outcomes for different insurance statuses within the same sample or provided multiple effect estimates, which did not meet the assumption of independent estimation of traditional meta-analysis [22]. To address this, we used multilevel meta-analysis to assess heterogeneity by comparing real effects within (τ2level 2) and between (τ2level 3) studies. The I2 statistic is indicated by I2level 2 and I2level 3. The z-values obtained from the multilevel model were converted to yield correlation coefficients, which were subsequently transformed using the natural logarithm to HR. For studies from other countries where the number of included studies was lower than 10 and estimates were more independent, we applied a traditional meta-analysis. A restricted maximum likelihood random-effects model was used for all studies [23]. Moderator analyses were exploratory and included cancer type, diagnosis period, adjustment covariates, and stage at diagnosis, reflecting theoretically relevant and consistently reported variables. Egger’s test was used to assess funnel plot asymmetry [24]. We analyzed using the ‘dmetar’ [25] and ‘metafor’ [26] packages in R 4.3.2. All tests were two-tailed and P < 0.05 was considered statistically significant.

Results
Preliminary analysis identified 4863 records, of which 2847 were duplicates. 1835 articles were excluded based on titles and abstracts. The remaining 181 articles were carefully examined in full text, with 126 studies excluded (Supplementary eFigure 1). In all, 37 studies were selected for the systematic review and meta-analysis, comprising 31 independent studies that reported OS [9–11, 13, 14, 16, 27–53], yielding 219 effect sizes, and 8 independent studies that reported CSS [10, 12–14, 49, 54–56], contributing 22 effect sizes. Among cancer patients in the US, 27 independent studies provided 204 effect sizes. Four independent studies among Chinese cancer patients with 10 effect sizes. The lack of adequate research from other nations required qualitative synthesis [14, 16, 28, 36, 37, 53]. 32 studies were retrospective cohorts. One was a post-hoc [47]. The number of included patients varied, ranging from 462 to 885,075 individuals. 16 studies obtained patient data from the NCDB (2003–2018) or SEER (2007–2015) [9, 11, 30, 33–35, 38–40, 42–44, 48, 50, 52, 54], and 21 studies originated from hospital cohorts or projects [10, 12–14, 16, 27–29, 31, 32, 36, 37, 41, 45–47, 49, 51, 53, 55, 56]. 4 studies reported data for more than one of the five cancers examined in this study [10, 11, 43, 48].
Most studies accounted for various conventional covariates, including demographic characteristics (such as age, race, sex, and marital status), disease features (such as stage, CCI, and tumor size), and treatment features (such as surgery, chemoradiotherapy, and immunotherapy, hormone therapy, transplantation, adjuvant therapy). A small proportion of studies did not incorporate age as a covariate but instead chose to use the year at diagnosis as a covariate for adjustment [10, 11, 48]. In addition, specific research accounted for variables such as etiology [27], lifestyle score [14], distance [30, 34, 39, 44], smoking status [12, 31, 46], and distance to facility (Supplementary eTable 3–5) [30, 34, 39, 44].
The average NOS score was 8.2, indicating generally good methodological quality due to long follow-up periods, clearly defined outcomes, and accurate reporting of insurance status (Supplementary eTable 6). Detailed assessment using the ROBINS-I tool revealed that most studies (28 out of 37, 75.7%) had a moderate or higher risk of bias. The primary sources of bias included insufficient control for confounding (D1), particularly the lack of adjustment for socioeconomic status, tumor-related characteristics, and treatment-related variables; selection bias (D2), such as inclusion limited to stage IV cancer patients or exclusion of those with uninsured status; and bias due to deviations from intended interventions (D4), where systematic differences in treatment or care received after diagnosis were not directly caused by insurance status but may still influence outcomes, thus compromising the causal interpretation of insurance effects (Supplementary eFigure 2).

Insurance status in the US
Across 204 effect sizes from 27 independent studies within the US health insurance system, patients insured by Medicare, Medicaid, or uninsured had worse OS than private insurance status (HRMedicare: 1.29, 95% CI: 1.17–1.42; HRMedicaid: 1.39, 95% CI: 1.28–1.51; HRUninsured: 1.42, 95% CI: 1.31–1.53). The pooled effect of OS across the three insurance statuses did not exhibit significant differences (χ2 = 0.79, P = 0.673). Funnel plots did not exhibit asymmetry (Supplementary eFigs. 3 and 4). Although Egger’s test showed significant results (P < 0.05), indicating possible publication bias, after cancer categorization, Egger’s test p-value was insignificant, indicating heterogeneity is mainly due to cancer type (Supplementary eTable 15). The heterogeneity between studies (Ι2level 3) within the Medicare status was 31.89% (τ2level 3 = 0.022), within the Medicaid status was 24.60% (τ2level 3 = 0.016), and within the uninsured status was 30.77% (τ2level 3 = 0.015). Substantial heterogeneity was observed within studies. For Medicare, the Ι2level 2 was 67.28% (τ2level 2 = 0.046); for Medicaid, it was 74.07% (τ2level 2 = 0.049), and in the uninsured status, it was 65.99% (τ2level 2 = 0.032). Significant moderators associated with the design features of studies were identified following exploratory subgroup analyses on OS (Table 1, Fig. 1, Supplementary eTables 7, 8 and 9).

Medicare vs. private insurance
Moderator analysis showed that the OS in Medicare patients was affected by cancer stage (F = 10.78, P < 0.0001), cancer type (F = 4.04, P = 0.006), and multifactorial Cox regression model design (F = 7.44, P = 0.008). Early-stage (I-II) patients (HR: 1.55, 95% CI: 1.34–1.76) with Medicare status were associated with a greater hazard of OS than those with private insurance status. However, no statistical evidence was found in OS between Medicare and private insurance for late-stage (III-IV) patients (HR: 1.16, 95% CI: 0.86–1.54) and studies with unspecified staging (HR: 1.30, 95% CI: 0.88–1.79). Breast (HR: 1.38, 95% CI: 1.20–1.57) and prostate (HR: 1.45, 95% CI: 1.05–1.86) cancer patients exhibited increased hazards of OS with Medicare insurance status than patients with private insurance. In contrast, no evidence of worse OS was observed with Medicare insurance status in the liver (HR: 1.08, 95% CI: 0.74–1.55), lung (HR: 1.15, 95% CI: 0.85–1.52), and colorectal (HR: 1.35, 95% CI: 0.98–1.75) cancer patients. Designs of the multifactorial Cox regression model showed worse OS in Medicare patients after either comprehensive (HR: 1.18, 95% CI: 1.06–1.31) or relatively single (HR: 1.46, 95% CI: 1.12–1.80) adjustments, and the former had a lower OS. Medicare patients diagnosed in 2000–2010 (HR: 1.24, 95% CI: 1.04–1.46) also had a worse OS, but the moderator of year at diagnosis did not show any observed differences in OS (F = 0.57, P = 0.566).

Medicaid vs. private insurance
Moderator analysis results indicated that the OS of Medicaid patients was significantly affected by cancer stage (F = 9.78, P = 0.0002) and type (F = 9.12, P < 0.0001). Early-stage (I-II) cancer (HR:1.69, 95% CI:1.51–1.87) patients insured in Medicaid were associated with poorer OS than private insurance. In contrast, no statistical evidence of OS differences was found between Medicaid and privately insured patients for late-stage (III-IV) (HR: 1.31, 95% CI: 0.99–1.67) and unspecified tumor staging (HR: 1.36, 95% CI: 0.96–1.80). Medicaid patients diagnosed with prostate cancer (HR: 1.75, 95% CI: 1.37–2.09) had the worst OS. Similarly, Medicaid patients with breast (HR: 1.47, 95% CI: 1.30–1.64) or colorectal (HR: 1.46, 95% CI: 1.11–1.82) cancer had worse OS. Nevertheless, no statistically significant difference in OS was found for Medicaid patients with lung (HR: 1.22, 95% CI: 0.93–1.57) or liver (HR: 1.25, 95% CI: 0.90–1.67) cancer than private insurance patients. In the multifactorial Cox regression models, Medicaid patients had worse OS compared to privately insured patients after adjusting for comprehensive (HR: 1.32, 95% CI: 1.18–1.46) or relatively single (HR: 1.47, 95% CI: 1.14–1.82) covariates, moderator of the models had insignificant differences (F = 2.29, P = 0.135). Medicaid patients diagnosed in 2000–2010 had reduced OS (HR: 1.33, 95% CI: 1.12–1.56). However, no significant changes were identified about the diagnosis year moderator (F = 0.34, P = 0.710).

Uninsured vs. private insurance
Moderator analysis revealed that the OS of uninsured patients was moderated by cancer stage (F = 7.92, P = 0.0008), types (F = 4.28, P = 0.004), and Cox regression model with covariate adjustment (F = 23.78, P < 0.0001). Uninsured cancer patients with early-stage (I-II) (HR: 1.63, 95% CI: 1.44–1.80) had the poorest OS than privately insured individuals, whereas the hazards of OS were decreased at advanced (III-IV) (HR: 1.34, 95% CI: 1.05–1.67). No observed OS differences between privately and uninsured insured in studies of unspecified tumor staging (HR: 1.37, 95% CI: 0.97–1.81). Uninsured patients with breast (HR: 1.50, 95% CI: 1.33–1.67), liver (HR: 1.55, 95% CI: 1.15–1.94), colorectal (HR: 1.44, 95% CI: 1.10–1.80), or prostate cancer (HR: 1.59, 95% CI: 1.21–1.95) experienced worse OS, no statistical differences of these four cancer types were observed. In the moderator of the multifactorial Cox regression model, uninsured patients had worse OS after comprehensive (HR: 1.28, 95% CI: 1.20–1.38) or relatively single (HR: 1.58, 95% CI: 1.37–1.79) covariates, and the latter indicated significantly higher hazards. Regarding the diagnosis year of patients, uninsured cancer patients diagnosed after 2011 (HR: 1.54, 95% CI: 1.13–1.95) had worse OS than those diagnosed in 2000–2010 (HR: 1.30, 95% CI: 1.13–1.49), but year at diagnosis was not moderating factor (F = 2.04, P = 0.138).

Insurance status in China
Across 5 effect sizes from 3 independent studies within the Chinese healthcare system, it was found that patients without Urban Employee Basic Medical Insurance (non-UEBMI) had poorer OS than those with UEBMI status (HR: 1.39, 95% CI: 1.22–1.59). The heterogeneity was 60% (τ2 = 0.012) and visual inspection of the funnel plots revealed limited signs of small study effects (Supplementary eFigs. 3 and 4). Formal asymmetry tests were not performed due to insufficient included studies. Subgroup analysis showed that non-UEBMI patients diagnosed with liver (HR: 1.47, 95% CI: 1.10–1.98, I2 = 88%, P = 0.01) or breast cancer (HR: 1.47, 95% CI: 1.16–1.52, I2 = 0%, P < 0.0001) experienced worse OS (Q = 0.41, P = 0.52) (Figs. 2 and 3, Supplementary eTable 13).

Qualitative analysis
Two studies conducted in South Korea, with 6423 patients, were incorporated. One related 5355 cancer patients (colorectal, stages II-III) showed an insignificant difference in OS between National Health Insurance and the Medical Aid Program groups (HR: 1.17, 95% CI: 0.88–1.54) after adopting a relatively single adjustment Cox regression model [16]. Another for breast cancer patients, using four adjusted models, found no significant diversity of OS between the two insurance statuses in the fully adjusted model (standard error = 0.72) [36]. A 2014 study in Brazil with 3124 patients with breast cancer reported no statistical difference in OS for those with private versus public insurance (HR: 1.89, 95% CI: 0.99–3.57), but the large standard error (1.29) suggests considerable uncertainty [28]. Thailand of 1931 colorectal cancer patients, those with UCS status had worse OS than those with CSMBS (HR: 1.37, 95% CI: 1.09–1.72), with no notable difference in OS between CSMBS and SSS (HR: 0.78, 95% CI: 0.48–1.29) [37]. A German study of 3977 colorectal patients discovered no disparity in OS (HR: 1.03, 95% CI: 0.83–1.30) and CSS (HR: 0.97, 95%CI: 0.77–1.21) between social health insurance (SHI) and private health insurance (PHI) after comprehensive adjustment [14]. Confidence is stronger in studies from Germany, South Korea (for colorectal cancer patients), and Thailand because of smaller standard errors. In the Japanese study included in the review [53], public assistance recipients with EGFR mutation-positive lung cancer did not show a survival advantage (HR: 1.08, 95% CI: 0.67–1.74).

Subsequent outcome
In four independent studies with 16 effect sizes conducted within the US health insurance system, not privately insured (Medicare: HR, 1.19; 95%CI, 1.04–1.36. Medicaid: HR, 1.35; 95%CI, 1.11–1.60. uninsured: HR, 1.30; 95%CI, 1.21–1.40) patients exhibited worse CSS than privately insured patients. All three insurance statuses have worse CSS for breast cancer than for other cancers. Subgroup analysis showed that uninsured breast, colorectal, lung, or prostate cancer patients had worse CSS than privately insured patients (Fig. 4, Supplementary eTable 10–12). Three studies conducted in China found a substantial correlation between non-UEBMI and worse CSS. Subgroup analysis revealed that breast cancer patients with non-UEBMI had worse CSS (Figs. 2 and 3, Supplementary eTable 14).

Sensitivity analysis
We included a patient sample from a study conducted in the US between 1997 and 2003 for a multilevel meta-analysis on the OS [10]. The summary effect estimates for cancer-specific survival (CSS) in the trials conducted in the US and China were obtained using the ‘Paule-Mandel’ rather than the ‘REML’. In summary, sensitivity analysis did not yield any alterations in the significance or direction of the main results (Supplementary eTables 16–18, eFigure 5–9).

Discussion
Patients with early-stage (I-II) cancer exhibited larger disparities in OS compared with patients with advanced-stage (Ⅲ-Ⅳ) cancer in multilevel meta-analyses of three insurance statuses compared with private health insurance. This pattern suggests that the impact of insurance policy may be more elastic in early-stage cancer care. The concept of insurance elasticity highlights that financial protection may contribute to more apparent health benefits at earlier stages of the cancer, through insurance, when timely access to screening and curative treatment is both more feasible and more effective. Insurance influences out-of-pocket costs through variations in coverage scope and reimbursement rates, thereby affecting the price elasticity of demand for cancer care [57]. As such, patients with less generous or no coverage may be discouraged from initiating early treatment, leading to poorer survival outcomes, while the marginal benefit of insurance in advanced-stage cancers may diminish, as treatment decisions are less discretionary and prognosis less modifiable [58–60]. Although contemporary medicine has made progress in improving the complexity of cancer treatment methods, patients in advanced stages of the disease still have difficulties in obtaining remission [61, 62]. Moreover, improper treatment protocols might result in financial toxicity [63], which could potentially contribute to the narrowing gap in survival rates among individuals with advanced cancer. On the other hand, by providing equitable financial support, health insurance facilitates early cancer diagnosis, increases opportunities for patients to access primary healthcare, reduces expenses, and alleviates economic burdens, enhancing the price elasticity and adaptability of insurance payments in cancer treatment [64]. Under the Affordable Care Act (ACA), most insurance plans must cover physical exams and preventive services [65]. Nevertheless, the current body of research suggests that insufficient insurance coverage [66] or its absence may reduce the chances for cancer screening and basic healthcare [67]. Research has indicated that around 30% of people without health insurance undergo the most up-to-date screenings for breast or colorectal cancer. In comparison, the percentage is over 65% among those who have private health insurance [68]. Uninsured persons rarely benefit from early screening and evaluation of symptoms, leading to a high probability of being diagnosed with advanced cancer, which prompts some to seek enrollment in public insurance to gain access to improved treatment outcomes [69, 70], but cancer registries usually record only the insurance status at initial diagnosis, which partially accounts for the found similarities in cancer stage and OS among patients with these three different insurance statuses. Staging was not reported in the included Chinese studies, but China has had a rise in cancer screening usage in recent decades. As of 2024, China has initiated four national cancer screening programs (NCSP) encompassing eight target cancer types, funded through the Basic Public Health Service special funds. However, screening has not been integrated into basic health insurance schemes [71]. In the future, it is essential to focus on studying the specific benefits of insurance status in preventing and treating cancer while also considering how changes in status over time can affect the progression of cancer and the chances of survival to gain a more comprehensive understanding of how insurance status impacts cancer development.
The same insurance status on OS or CSS varies across countries for different types of cancer. The US multilevel meta-analysis found that cancer type moderates the heterogeneity of results. Compared to other cancers, the OS disparity between breast and prostate cancer patients with Medicaid or uninsured status and those with private insurance is larger. However, subgroup analysis found no changes in OS or CSS among Chinese with liver, lung, and breast cancer who had non-UEBMI insurance. These variations in relative survival rates among different cancers are influenced by tumor malignancy, early detection of the disease, accessibility of diagnostic techniques, and advancements in treatment regimens tailored to specific tumor types [72–74]. Liver and colorectal cancer are more malignant and have fewer early symptoms with shorter survival, but breast and prostate cancer are less malignant, diagnostic techniques are developing more rapidly, and many effective treatments can lead to clinical cures and improved survival [75]. Furthermore, disparities in cancer survival are influenced by variances in healthcare systems’ performance [76]. The 5-year survival rates for prostate and breast cancer in the US were 93.4% and 86.1%; however, the survival rates for colorectal (60.8%), hepatic (20.8%), and lung (5.4%) cancer were lower [77]. In China, breast cancer (73.1%) patients had higher 5-year survival rates compared to lung (16.1%) and liver (10.1%) cancer [78]. Comparisons of survival rates between health systems in different countries must, therefore, be interpreted and analyzed carefully to avoid misinterpretation [79].
Moreover, the payment elasticity of insurance policies may also determine cancer survival outcomes. Previous studies revealed that patients with lower insurance status and financial constraints faced restrictions in receiving and adhering to treatment [80, 81]. For instance, in China’s non-UEBMI population, the likelihood of liver cancer patients undergoing adjunctive transarterial chemoembolization was reduced; in the US, non-privately insured patients received fewer standard treatments and adjuvant radiotherapy [27, 82]. Colorectal cancer patients with SHI status also less used targeted therapies in Germany, however, no association was observed between insurance status and OS, possibly attributed to insurance policies promptly paying expansion for valuable treatment modalities tailored to specific cancers by recommended treatment guidelines [14]. Thus, researchers should gain further insights into a comprehensive comparative analysis of policy payment elasticity across countries to provide valuable insights for policy transfer and adjustment strategies in the future.
Adjustment variables moderated the Medicare and uninsured status of US patients’ OS. Specifically, after comprehensive adjustments for sociodemographic characteristics, disease physiology, and treatment features, we still observed an impact of insurance on survival, albeit diminished. However, an insignificant OS difference was shown between the single and comprehensive Cox models. The potential cause could be the differences in measures among various studies focusing on different types of malignancies, clinical environments, geographic regions, and socioeconomic conditions [83]. Examination of specific adjustment variables is complex, but insurance policy consistency and justice can reduce these complicated variables and improve survival rates [84]. The basic health insurance system in China, which includes UEBMI and URRBMI, provides coverage for more than 95% of the population [85], with the URRBMI system featuring highly unified internal policies and comprehensive protection measures, such as payment ratios, enrollment criteria, and reimbursement proportions [86–88]. Chinese medication price negotiation and reimbursement policies reduce the economic burden on cancer patients, facilitate access to innovative drugs, and achieve relatively equitable policy objectives [89, 90], establishing a healthcare safety net for economically disadvantaged patients [91].In contrast, the US insurance statuses and plans are fragmented [92]. The fragmented nature of insurance coverage may further amplify the influence of certain adjustment variables, such as insurance-based racial discrimination rates and low coverage rates among socioeconomically disadvantaged groups [93–95]. However, the OS of Medicaid patients was unmoderated by adjustment variables design, indicating achievement by the Medicaid scheme in improving coverage and equity. Recent research indicates that implementing Medicaid expansion under the ACA diminishes discrepancies in survival rates by ensuring equitable access to healthcare [96]. In addition to the ACA, some states are expanding additional healthcare subsidy programs, like the Medicaid Cancer Treatment Program (MCTP) and the National Breast and Cervical Cancer Early Detection Program (NBCCEDP), to reduce disparities [68, 97]. The American Rescue Plan Act (ARPA), a legislation worth $1.9 trillion, was enacted in 2021 to improve healthcare accessibility and affordability for cancer patients without insurance status. Additionally, it provides incentives to low-income adults in unexpanded Medicaid states [98]. After conducting comprehensive multivariable Cox regression, the persistent influence of insurance status indicates its potential unidentified factors, impacting patient survival. Despite having a large amount of patient data, most studies still rely on database information that does not include measurements and records of important variables such as patients’ nutritional dietary structures, health-seeking behaviors (including insurance enrollment behavior), insurance awareness, and cultural beliefs [14, 48, 82].These unidentified variables are likely to interact with the variables currently being examined in research, exerting additional influence on measurement outcomes [83]. Therefore, conducting a thorough analysis to verify these findings and gain a deeper understanding of these mechanisms is imperative.
Although our study did not find that year at diagnosis moderated the results, this does not diminish its importance. The prompt implementation of expanding insurance coverage can lead to better survival outcomes for patients diagnosed at that time [99, 100]. Our study validates that after the implementation of the ACA in 2010 [101], patients with Medicare and Medicaid insurance status diagnosed with cancer after 2011 showed no significant differences in OS than those with private insurance [102]. However, uninsured patients diagnosed after 2011 experienced worse OS and CSS. Future research should focus on understanding the long-term survival trajectories of cancer patients under different insurance policies.
There is a lack of extensive worldwide research on the correlation between these aspects, specifically regarding the impact of inventive provisions inside healthcare insurance policies on the outlook of individuals with cancer. Many aspects of cancer policy remain to be explored, such as Payment and Delivery System Reform, Interruption and change of insurance status, new issues due to policy reforms, and the similarities and differences in healthcare insurance policies among different countries on cancer survivorship [84, 103].

Limitations
This study has several limitations. First, selection bias remains a major concern. The included studies primarily relied on retrospective cohort data from cancer registries, which may not adequately record pre-diagnosis characteristics. Moreover, studies typically use an individual’s insurance status at the time of cancer diagnosis as an exposure in Cox regression models. Changes in insurance status during cancer therapy are often unmeasured or not reported in administrative databases. Uninsured patients may enroll in public insurance programs at cancer diagnosis, and then lose eligibility for insurance coverage after completion of initial treatment, potentially narrowing the survival gap between insurance groups and making adverse effects on access to high-quality survivorship care [10, 104]. These unobserved, time-varying insurance statuses may lead to misclassification and limited true effect estimates. Second, it is unclear whether the interaction between insurance status and adjustment variables, such as sociodemographic, clinical and cancer characteristics, partly explains these survival inequalities. Some studies lacked detailed data on patient-level socioeconomic factors, lifestyle behaviors, or comorbid conditions, all of which can influence both insurance status and survival. Identifying specific, modifiable insurance-related factors that contribute to survival disparities remains a challenge. Finally, the geographic distribution of included studies was imbalanced, with the majority originating from the US, which may limit the generalizability of findings to countries with more universal healthcare coverage. Therefore, the results should be interpreted and discussed with caution. Future studies should employ prospective longitudinal designs with clear documentation of follow-up duration and changes in insurance status over time. These methodological improvements would contribute to refining the evidence base and enhancing the robustness of conclusions regarding the relationship between insurance status and cancer survival.

Conclusions
Insurance status is a structural determinant of cancer survival equity worldwide, though its association with survival outcomes may vary across national health insurance systems. In countries with multilevel health insurance systems, such as the US and China, significant survival outcome disparities were observed in different insurance statuses. On the other hand, studies from countries with universal social insurance coverage, such as Germany and South Korea, reported no such associations. This observation suggests the importance of considering institutional context when evaluating the impact of insurance on cancer survival. Cancer stage, cancer type, and adjustment variables moderated the heterogeneity of the effect sizes in the included studies. These results highlight the importance of insurance elasticity that support early-stage management and screen-detectable cancers with typically favorable prognoses, such as breast and prostate cancer, while also extending coverage protections to uninsured groups in both national and global settings. Future research should establish international prospective cohorts to explore how insurance status and covariate interactions affect survival disparities and to inform equitable cancer policy reform globally.

Supplementary Information