We conducted this real-world study at two centers, the Southern Center (Cancer Hospital of Shantou University Medical College, Shantou City, Guangdong Province, China) and the Northern Center (Anyang Cancer Hospital, Anyang City, Henan Province, China), as published elsewhere (18). Located in the southeast coastal area, Shantou city has an EC incidence rate (age-standardized: 11.43 per 100,000 person-years) comparable to the national average (19). Anyang City is located in the Taihang Mountain region, which is a high-risk area for EC with an incidence rate up to five times the national average level (19, 20). Both centers are the only Grade-A tertiary cancer specialized hospitals in their respective localities, offering comprehensive cancer treatment services within their catchment regions.

We consecutively recruited patients diagnosed with ESCC who were first admitted to hospital between August 1, 2009 and December 31, 2018 at the Southern Center, and between January 1, 2012 and December 31, 2017 at the Northern Center. The follow-up at the two centers was up to June 7, 2022 and August 23, 2022, respectively. We excluded patients who 1) didn’t receive anticancer treatment; 2) had missing basic information or cost data; 3) had a follow-up period of less than 6 months. (Supplementary Figure S1).

The electronic clinical records of ESCC patients (N = 11037) were extracted from the Hospital Information System. Details of information collection and data quality control are provided in the Supplementary Materials. Missing values for continuous variables were imputed with medians and an “Unknown” group was created to account for missing categorical variables.

Hospital spending including the costs for drugs, surgery, non-surgical treatment, inspection, and others, was exported by the hospital electronic medical record system. Non-surgical treatment included radiation and oxygen therapy. Others included nursing, bed, laboratory test and diagnosis. These costs represent the total hospital spending over the entire treatment period for ESCC, from the first to the last treatment within the same hospital (Figure 1). The mean hospital spending was calculated as the total hospital spending per patients. All costs were reported in Chinese Yuan (CNY) based on the 2023 value, which were inflated using the year-specific personal health care consumer price index (CPI) of Southern Center (Shantou city) and Northern Center (Anyang city), respectively (Supplementary Table S1). The costs were then converted into US Dollars using the 2023 purchasing power parity exchange rate ($1 = ¥3.64) (21). In addition, two sensitivity analysis were performed. The first accounted for total costs from an indirect perspective, including hospital spending and time costs, as detailed in the Supplementary Materials. The second addressed a potential concern: survival time could influence spending group assignment, making it difficult to determine if higher spending improves survival or if longer survival simply allows more spending. To address this, we analyzed spending accumulated only in the first 6 months after treatment.

The characteristics of ESCC patients were stratified by terciles, centers, and TNM stage. The analysis involved using the ANOVA test for continuous variables and the rank sum test for categorical variables.

We explored the temporal trend of spending by estimating the average spending per patient in each month following the onset first hospitalization, as shown in the following formula:

The given month was rounded to the next upward integer (for example, 7.8 months would be rounded up to 8 months).

OS was the primary outcome in our analysis, defined as the time interval from initial admission to either last contact or death. Median follow-up time and median survival time were estimated using the reverse Kaplan-Meier method and the Kaplan-Meier method, respectively. Kaplan-Meier survival curves were used to illustrate prognosis, and differences between curves were assessed with the log-rank test. Multivariable Cox proportional hazards models were used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for comparing OS between patients in tercile 1 (lowest spending) and those in tercile 2, as well as between tercile 1 and tercile 3. Stratified analyses were performed by TNM stage and study center. To evaluate the independent association between spending and OS, we incorporated seven factors in the survival analysis: age, sex, occupation, stage, therapeutic approaches, number of hospital admissions, and length of hospital days. These factors were chosen from an initial set of 13 potential variables, which also considered comorbidities, postoperative complications, type of operation, health insurance status, family history of EC, and study center. All variables were initially evaluated using univariate Cox proportional hazards regression analysis. Those with a P-value<0.05 were further analyzed with multivariable Cox regression, employing backward selection (P < 0.05) to identify potential risk factors for OS. The therapeutic approaches included surgery alone, surgery combined with chemotherapy, surgery with radiotherapy, surgery with chemoradiotherapy, chemotherapy alone, radiotherapy alone, and chemoradiotherapy, with no fixed sequence within treatment plans.

All statistical analyses were performed using STATA version 15.1 (STATA, College Station, Texas, USA). All tests were two-sided and had a significance level of 0.05.

As shown in Table 1, most of the 11,037 ESCC patients were older than 60 years (70.6%), male (66.9%), and predominantly engaged in farming (60.1%). The majority were classified as stage III-IV (36.9%), underwent surgery alone (36.6%), and had a single hospital admission (46.7%). From tercile 1 to tercile 3, higher spending was associated with more advanced TNM stages (stage III: 25.0% vs. 37.6% vs. 48.1%; stage IV: 6.9% vs. 6.9% vs. 9.6%), more comprehensive treatment, and variations in age, sex, occupation, therapeutic approaches, and number of hospital admissions.

Patients at the Southern Center were younger (median age: 61 vs. 65), had a higher percentage of males (75.8% vs. 61.0%), and a lower proportion of farmers (16.9% vs. 89.1%) and cases treated with surgery only (24.4% vs. 44.8%). Additionally, they had a higher proportion of advanced-stage cases (stage III: 52.7% vs. 26.3%; stage IV: 9.7% vs. 6.5%) and longer hospital stays (64.3 days vs. 51.0 days) compared to patients at the Northern Center (Supplementary Table S2).

As shown in Table 1, the mean spending per ESCC patient was $19,461, which increased by 218% from tercile 1 to tercile 3 ($10,012, $16,511, $31,859). Higher costs were associated with more advanced TNM stages (stage 0-I: $14,450; stage II: $18,391; stage III: $21,900; stage IV: $21,252). Spending also varied by age, sex, occupation, therapeutic approaches, and number of hospital admissions. Except for tercile 1, which had lower costs for advanced TNM stages, tercile 2 and tercile 3 exhibited similar patterns. These trends were consistent across both the Southern and Northern Centers (Supplementary Table S2).

A higher degree of treatment complexity was associated with increased spending. As shown in Supplementary Table S3, the proportion of stage 0-II patients undergoing surgery alone sharply decreased from tercile 1 to tercile 3 (83.0% vs. 73.0% vs. 40.6%). Patients in higher spending terciles tended to receive more comprehensive treatments, such as additional adjuvant therapy. Similarly, Supplementary Table S4 shows that a similar trend was observed among stage III-IV patients undergoing surgery alone, with proportions decreasing from tercile 1 to tercile 3 (35.99% vs. 21.70% vs. 7.90%).

As shown in Figure 1, the costs associated with drugs, surgery, non-surgical treatments, inspections, and other expenses were 38%, 7%, 24%, 13%, and 18%, respectively. Drug costs constituted the largest portion for all patients (38%), the Southern Center (36%), and the Northern Center (39%). Figure 2 illustrates that the time trends for monthly average hospital spending since the first hospitalization followed an “L-shaped” pattern, with the majority of lifetime hospital spending incurred within the first six months (overall: 77.73%; stage 0-II: 81.40%; stage III-IV: 76.48%). This trend remained consistent across patients stratified by study center (Supplementary Figure S2).

As shown in Supplementary Tables S3 and S4, mean hospital spending for ESCC patients, stratified by stage, varied significantly based on the type of therapy received. Surgery combined with adjuvant therapy was more costly than surgery alone. For patients with stage 0-II, the spending on surgery combined with chemoradiotherapy, surgery with radiotherapy, surgery with chemotherapy, and surgery alone was $32,225, $23,584, $21,521, and $14,917, respectively. In comparison, spending for patients with stage III-IV was $32,992, $25,504, $25,230, and $16,247, respectively. Additionally, chemoradiotherapy was more expensive than both radiotherapy and chemotherapy (stage 0-II: $20,684 vs. $13,304 and $19,425; stage III-IV: $23,061 vs. $13,751 and $15,066).

After a maximum follow-up of 12.52 years, the median follow-up time, median survival time, and age-standardized (ASR) 5-year survival rate of ESCC patients was 6.79 years (95% CI: 6.68-6.85 years), 4.70 years (95% CI: 4.39-5.10 years), and 49.58% (95% CI: 47.90%-51.23%), respectively. As shown in Figure 3, Kaplan-Meier survival curves for ESCC patients showed tercile 3 (median survival time:3.19 years) was linked to worse OS compared with tercile 1 (median survival time:6.97 years) and tercile 2 (median survival time:7.12 years). This pattern was consistent for stage 0-II patients but reversed for those with stage III-IV. Similar patterns were observed in the Kaplan-Meier survival curves stratified by study center (Supplementary Figure S3).

Figure 4 illustrates that multivariable Cox regression models demonstrated higher spending was associated with better OS (adjusted HR_{tercile 2 vs 1} = 0.85, 95% CI: 0.79-0.91; adjusted HR_{tercile 3 vs 1} = 0.88, 95% CI: 0.80-0.97). These findings were consistent for patients with stage III-IV (adjusted HR_{tercile 2 vs 1} = 0.82, 95% CI: 0.74-0.90; adjusted HR_{tercile 3 vs 1} = 0.73, 95% CI: 0.64-0.83), but reversed for those with stage 0-II, where higher spending was associated with worse OS (adjusted HR_{tercile 3 vs 1} = 1.55, 95% CI: 1.27-1.89). Similar patterns were observed in the multivariable Cox regression models stratified by study center (Supplementary Figure S4).

The findings from the indirect perspective were consistent with those from the direct perspective (Supplementary Materials, Supplementary Table S5, Supplementary Figures S5-S8). Similarly, the analysis of spending accumulated only within the first 6 months also yielded consistent results, supporting the robustness of our primary findings (Supplementary Materials, Supplementary Table S6, Supplementary Figures S9, 10).