This survey was conducted at a Class A tertiary specialized cancer hospital in Nanning, the capital city of Guangxi. The inclusion criteria were as follows: (1) patients whose first hospital visit for CRC diagnosis and/or treatment occurred between 1 January 2005 and 31 December 2014; (2) patients who received CRC therapy at the study hospital; and (3) patients whose complete clinical information could be captured, particularly details on cancer staging and pathological diagnosis. Patients who had only visited the hospital for diagnosis, post-treatment follow-up, or outpatient radiotherapy within the study period were excluded. Patients were not contacted directly, and information was obtained from their medical records in the study hospital (24, 25). The clinical stage of cancer was defined utilizing the TNM classification by the American Joint Committee on Cancer (AJCC) and the Union for International Cancer Control (UICC).

The survey was approved by the Institutional Review Board of the cancer hospital at the Chinese Academy of Medical Sciences (which waived the need to obtain informed consent from participants because of the retrospective and noninterventional design of this survey) as a component of a broader screening program (program 15–071/998).

The patient selection process is illustrated in Figure 1. Initially, 1,065 patients diagnosed with primary colorectal cancer between 2005 and 2014 were identified from the hospital’s electronic medical records. Subsequently, a multi-step exclusion criteria was applied to ensure data quality and homogeneity of the study cohort. First, 20 patients with incomplete key data (e.g., missing cost records or staging information) were excluded. Second, one patient whose surgical site was the appendix was removed to maintain consistency in tumor localization. Finally, 15 patients who were concurrently diagnosed with other primary malignancies were excluded to avoid potential confounding effects on medical expenditure analysis. After these exclusions, the final analytical cohort comprised 1,029 eligible patients with primary colorectal cancer, who were then stratified into three calendar periods for analysis: 2005–2008 (n = 390), 2009–2011 (n = 327), and 2012–2014 (n = 312).

We included at least 100 patients each year from 2005 to 2014. We retrieved the date of the first clinical visit of each patient to the study hospital. We defined the first patients diagnosed in May 2005, June 2006, July 2007, August 2008, September 2009, October 2010, November 2011, December 2012, March 2013, and April 2014 as the first participants in their respective years (January and February were not considered since the fewest number of patients were hospitalized for diagnosis and treatment around the Spring Festival in China). Subsequently, we collected each eligible case whose date of clinical visit was after that of the first participant, until the figure of 100 or more patients was reached. A simplified form was used to classify patients with primary and nonprimary CRC. Patients with nonprimary CRC were excluded; those with primary CRC were administered a survey via a structured case report form (CRF), which was provided by the Chinese National Cancer Center (CNCC). The following data were collected: (1) basic demographic information, including hospital ID, name, gender, age, insurance status, occupation, etc.; (2) clinical information, such as clinical stage, pathological diagnosis, disease metastasis, main diagnosis, etc.; (3) details related to the expenditure, including the date of the first visit, hospitalization days, overall expenditure, component of expense, etc.

Our team was managed by an individual who had attended a training course organized by the CNCC in Beijing. Six postgraduate students from the school of public health in the Guangxi medical university were recruited as investigators and then trained and examined; only those who passed the examination became qualified investigators. There were specific staff who worked in the study hospital to assist with data collection. The investigators worked in pairs; one investigator retrieved data from the electronic information system of the hospital and recorded this information on the paper-based CRF, while the other investigator rechecked the data. Data of the CRFs were entered onto the CNCC’s purpose-built database (EpiData 3.1, Odense, Denmark); an investigator double-entered the CRF data, then the data were subsequently submitted to the CNCC system. The CNCC checked the data which we submitted and raised suggestions about the unqualified data; we supplemented or modified the information based on these.

All statistical analyses were performed using SPSS version 20 (IBM, D0EJNLL, New York, United States). The records of 36 patients were excluded from analysis due to missing data. Based on the timing of the announcement of the official government reform plan, health reform in China can be broadly classified into two phases: Phase One (2009 to 2011) and Phase Two (2012 onward) (26). Accordingly, we stratified the study period as follows: 2005–2008; 2009–2011; and 2012–2014.

The expenditure data from this study and other cited studies were converted into the 2011 Chinese yuan (CNY; 1 CNY = 0.155 US dollar), which was inflated using the year-specific healthcare consumer price index of China obtained from the official website of the National Bureau of Statistics of China (27). Specially, expenditure data converted to 2011 values were consistent with a previous national study (24), Shi Jufang aimed to conduct direct comparisons with the national average, the average levels of different regions, the average levels of different hospital types, the average levels of different hospital, and so on.

In addition, the gross domestic product per capita was converted into the 2011 Chinese yuan by the year-specific gross domestic product per capita index. Demographic and clinical characteristics were described using proportions; age at diagnosis was described using the mean value; the number of clinical visits/admissions, the hospitalization days, and all expenses were described using the median. The expenditure data were compared within subgroups of characteristics using the nonparametric test; the Chi-square test or Fisher exact test were performed to detect any differences in clinical and demographic data among three calendar periods, as appropriate. Temporal trends of all medical expenses, the number of clinical visits/admissions, and hospitalization days from 2005 to 2014 were investigated using Spearman correlation analysis. The correlation analysis between medical expenses and the gross domestic product per capita was also investigated using Spearman correlation analysis. We transformed the medical costs into binary categorical variables based on the median. The medical cost that was lower than the median was classified as the ‘low-cost group,’ and that which was equal to or higher than the median was classified as the ‘high-cost group.’ The influencing factors of medical expenditure were identified utilizing the binary logistic regression analysis; to achieve an optimal goodness-of-fit test of the binary logistic regression model, the number of clinical visits/admissions and inpatient days were excluded from the analysis of the medical expense per clinical visit/admission. The AAPC of medical expenditure per patient, per clinical visit/admission, and per day was calculated; the AAPC of hospitalization days was also computed. The medical expenditure was evaluated in a whole period (2005–2014) and three calendar periods: 2005–2008, the period before the healthcare reform; 2009–2011, the first phase of the healthcare reform; and 2012–2014, the second phase of the healthcare reform. All statistical tests were two-tailed, and p-values less than 0.05 were considered statistically significant.

To assess the potential for selection bias and evaluate the robustness of our primary findings regarding sample representativeness, we conducted sensitivity analyses using subgroup consistency tests. This approach tests whether the key temporal trends in medical expenditures were consistent across diverse patient strata, which would indicate that the overall findings were not overly sensitive to the specific composition of our single-center sample. We stratified the study cohort by key demographic and clinical characteristics, including gender (male, female), age group at diagnosis (<45, 45–54, 55–64, ≥65 years), health insurance status (insured, uninsured), cancer stage at diagnosis (I-II, III-IV), and primary treatment regimen (chemotherapy alone, surgery alone, surgery combined with chemotherapy). For each subgroup, we calculated the average AAPC from 2005 to 2014 for the three primary expenditure outcomes: medical expenditure per patient, per clinical visit/admission, and per day. The consistency of the AAPC estimates and their directions (increasing or decreasing) across the various subgroups was then qualitatively evaluated. Robust and generalizable trends would be evidenced by similar AAPC values and uniform directional changes in expenditure outcomes across the majority of patient strata.

In total, records of 1,065 patients diagnosed with primary CRC from 2005 to 2014 in our study hospital were collected, of which 1,029 were available. Male patients constituted 60.64% of the total (male-to-female ratio: 1.54). The mean age at diagnosis was 57.25 years, with 34.31% of them being 65 years or more. In all, 28.18% of patients had received treatment at other hospitals. The most frequently observed stage at diagnosis was the advanced stage (59.87% of stage III–IV), and the least was stage I (3.89%). The proportions of rectal and colon cancer were similar. Adenocarcinoma was the most common histological type of cancer (92.71%). The majority of patients (72.30%) were diagnosed without metastatic disease. Most patients received surgery combined with chemotherapy. Approximately half of the cases had only one clinical visit/hospitalization. The median of inpatient days was 35 days. Among the three calendar periods, there were differences in distributions of health insurance status, number of clinical visits/admissions, or therapeutic patterns. The details are displayed in Table 1.

The results of the subgroup consistency tests are summarized in Table 7. The upward temporal trend in medical expenditure per patient observed in the overall cohort (AAPC: 5.59%) was consistently positive across all investigated patient subgroups. The AAPC estimates ranged from 4.82 to 6.15%, demonstrating remarkable homogeneity. Notably, the most substantial increases were observed among patients with health insurance (AAPC: 5.89%) and those diagnosed with stage I-II disease (AAPC: 6.15%). Similarly, the medical expenditure per day exhibited a pronounced and consistent increasing trend across all subgroups, with AAPCs ranging from 9.50 to 11.20%. The rapid growth in daily costs was universal, regardless of patient demographics, disease severity, or treatment received. In contrast, the trend for medical expenditure per clinical visit/admission, which showed a modest overall increase (AAPC: 1.41%), displayed greater variability across subgroups. While the direction of change was generally positive or neutral, the magnitude of the AAPC varied, from a slight decrease of −0.45% in the surgery-alone treatment subgroup to an increase of 3.02% in female patients.

The sensitivity analysis confirmed that the core findings of rapidly increasing per-patient and per-day costs were robust and not driven by the specific distribution of characteristics within our sample. The consistent trends across diverse patient strata strengthen the validity of our primary conclusions regarding the escalating economic burden of CRC care in Guangxi during the study period.

This study has several limitations. First, as a single-center study conducted in a tertiary hospital, the findings are representative of specialized, high-level cancer care settings in Guangxi but may not be broadly generalizable to patients in primary care or other regions. The sample had a higher male proportion and younger age at diagnosis compared to national averages, which might introduce selection bias. However, sensitivity analyses (e.g., subgroup consistency and weight adjustment) showed that the key trends were robust across different patient characteristics. Second, cost data were obtained solely from our institution, and some patients had prior diagnoses or treatments elsewhere, potentially leading to underestimation of total expenses. Third, the reliability of results depends on the completeness and accuracy of medical records.

Despite these limitations, as part of the health economics evaluation of the Cancer Screening Program in Urban China, this study offers several important practical implications. It provides a comprehensive overview of actual medical expenditures for CRC diagnosis and treatment, including decadal trends and influencing factors; supplies valuable evidence to inform the design of future CRC screening programs in urban Guangxi; and establishes a useful baseline for subsequent research in Guangxi, the broader southwestern region, and other comparable provinces in China.