Administrative data for medical service usage among colorectal cancer patients were obtained from the Korea National Health Information Database (NHID) from January 1, 2002 to December 31, 2018. The NHID is a public database on healthcare services maintained by the National Health Insurance System (NHIS) of Korea, which is a universal health insurance system that covers the medical expenditures of approximately 98% of all Korean citizens [13]. The database includes representative and comprehensive information on medical use among Korean patients, including insurance eligibility, diagnostic codes, prescribed medications and procedures, and billing records [14].

Patients with 2 or more International Classification of Diseases, 10th revision (ICD-10) diagnostic codes for colorectal cancer (C18–C20) and 1 or more admission records between 2004 and 2018 were defined as colorectal cancer patients. In order to exclude prevalent cases who had been diagnosed before NHID follow-up and to include incident cases only, we set a washout period of 2 years and excluded patients diagnosed in 2002–2003 [15]. From the original database, 40% of colorectal cancer patients were randomly selected (n=148,929). We excluded patients aged 18 years or younger (n=81), those diagnosed with cognitive impairment before colorectal cancer diagnosis (n=8,225), and those without administrative records for tumor resection (n=45,320) (Figure 1). In total, 95,303 patients were included in the final analysis (66,733 colon cancer cases and 28,570 rectal cancer cases).

Cognitive impairment was defined as the presence of at least 1 ICD-10 diagnostic code for dementia or minor cognitive impairment [16,17]. The ICD-10 codes for dementia and minor cognitive impairments are listed in Supplementary Material 1. The claim codes for the cancer treatment modality, including surgical resection, chemotherapy, and radiotherapy, were reviewed and confirmed by a colorectal surgeon (CWK), a medical oncologist (HK), and 2 epidemiologists (SJJ, KK). For chemotherapy, regimens that are recommended for first-line chemotherapy in the National Comprehensive Cancer Network (NCCN) Guidelines 2019 were included for analysis (oxaliplatin, capecitabine, 5-fluorouracil [5-FU], irinotecan; the administrative codes are listed in Supplementary Material 2). Patients with at least 1 label for a chemotherapy regimen were considered to be chemotherapy recipients. The date of the first insurance claim for colorectal cancer was considered as the date of colorectal cancer onset.

Monthly insurance premiums were used as a proxy variable for socioeconomic status. The participants’ monthly insurance premium payment records at baseline were collected, and the participants were divided into subgroups according to quintile values of monthly insurance premiums (cut-offs: 18,700, 31,110, 44,169, and 63,749 Korean won/mo in 2004). Medical Aid recipients who did not pay a premium due to their poor economic situation were classified into a separate subgroup. The Charlson comorbidity index values at baseline were calculated to assess medical comorbidities [18]. Participants with ICD-10 diagnostic codes corresponding to individual comorbidities were considered to have those comorbidities (Supplementary Material 3).

We classified participants according to the cancer treatment modality in accordance with claim records for cancer treatment as follows: in colon cancer, (1) primary resection only and (2) primary resection with chemotherapy; in rectal cancer, (1) primary resection only, (2) primary resection with chemotherapy, (3) primary resection and radiotherapy, and (4) primary resection with concurrent chemoradiotherapy (CCRT). Using these categories, we described the baseline characteristics of the participants by presenting mean and standard deviation values for continuous variables and numbers and percentages of participants for discrete variables.

We hypothesized that treatments were provided according to the 2019 NCCN guidelines for colorectal cancer treatment and that participants with the same treatment modality were likely to have a similar tumor burden [19,20]. To control for possible confounding by tumor burden, we excluded patients without claim codes for surgical resection, thereby excluding patients with inoperable tumors (reflecting a higher tumor burden) and chronic patients receiving palliative treatment only.

Hazard ratios (HRs) and their 95% confidence intervals (CIs) for chemotherapy and radiotherapy were estimated using a time-dependent competing risk survival analysis model, with chemotherapy and radiotherapy being considered as time-dependent variables and all-cause mortality considered as a competing risk [21–24]. To avoid immortal time bias, chemotherapy and radiotherapy were constructed as time-dependent variables, and follow-up periods were classified as primary resection with chemotherapy/radiotherapy/CCRT only after patients received the corresponding treatments [25–27]. For instance, if a patient was diagnosed with cognitive impairment after surgical resection and before chemotherapy, the follow-up period of the patient was classified as surgical resection only. Cox regression was conducted by applying the ‘proc phreg’ procedure in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). Time-dependent variables were created by ‘if else’ statements in the ‘proc phreg’ procedure. Censoring, the event (cognitive impairment), and the competing risk (all-cause mortality) were coded as 0, 1, and 2 respectively (Supplementary Material 4). We also estimated HRs for chemotherapy regimen combinations for colorectal cancer treatment, including FOLFOX (folate, 5-FU, and oxaliplatin), FOLFIRI (folate, 5-FU, and irinotecan), FOLFOXIRI (folate, 5-FU, oxaliplatin, and irinotecan), CapeOx (capecitabine and oxaliplatin), capecitabine only, 5-FU only, and irinotecan only. Considering left truncation, the date of colorectal cancer diagnosis was used as the date of follow-up initiation. Age interaction terms were added to the model to assess the moderating effects of age, and conditional HRs of chemotherapy and radiotherapy by age points were estimated [28]. All models were adjusted for age, sex, comorbidities, and the monthly insurance premium.

For sensitivity analyses, we redefined cognitive impairment cases as patients with 2 or more corresponding diagnostic codes for cognitive impairment and repeated the survival analyses. Additionally, we performed landmark analyses by conducting time-fixed Cox regression with lag times of 6 months, 12 months, and 18 months [29]. All statistical analyses were conducted using SAS version 9.4.

The study protocols were approved by the Institutional Review Board of Yonsei University Health System, Seoul, Korea (approval No. 4-2019-0425). Informed consent was waived for this study, since personal information that can be used to identify individuals registered to NHID was removed. All procedures contributing to this work complied with the ethical standards of the relevant national and institutional committees on human experimentation and with the 1975 Declaration of Helsinki, which was revised in 2008.

Among the 66,733 patients with colon cancer at baseline (2004), 14,146 (21.2%) received adjuvant or neoadjuvant chemotherapy until 2018. The mean follow-up duration was longer in patients who did not receive chemotherapy than in those who received it (5.57 vs. 3.21 years, p<0.001). Capecitabine, oxaliplatin, 5-FU, and irinotecan were administered to 3,228 (22.8%), 9,928 (70.2%), 7,709 (54.5%), and 2,444 (17.3%) chemotherapy recipients, respectively. The incidence rates of cognitive impairment were 22.17 per 1,000 person-years in chemotherapy non-recipients and 14.48 per 1,000 person-years in chemotherapy recipients. The all-cause mortality rates were 49.05 per 1,000 person-years in chemotherapy non-recipients and 96.52 per 1,000 person-years in patients who received chemotherapy (Table 1).

Among the 28,570 patients with rectal cancer included in the analyses, 2,604 (9.1%) received chemotherapy, 8,098 (28.3%) received radiotherapy, and 3,161 (11.1%) received CCRT before or after surgical resection. The mean follow-up duration (in years) was relatively longer in the primary resection and resection-radiotherapy combination groups and shorter in the resection-chemotherapy combination and resection-CCRT combination groups. Among 5,765 chemotherapy and CCRT recipients, capecitabine, oxaliplatin, 5-FU, and irinotecan were administered to 1,734 (30.1%), 3,289 (57.0%), 3,499 (60.7%), and 1,228 (21.3%), respectively. The incidence rate of cognitive impairment was highest in the resection only group (23.16 per 1,000 person-years), while mortality was highest in the resection-CCRT combination group (123.66 per 1,000 person-years) (Table 1).

In colon cancer patients, chemotherapy did not increase the risk of cognitive impairment (HR, 0.92; 95% CI, 0.83 to 1.03). In rectal cancer patients, neither chemotherapy (HR, 0.88; 95% CI, 0.75 to 1.04) nor radiotherapy (HR, 0.91; 95% CI, 0.84 to 0.99) was positively associated with cognitive impairment. Folate administration during chemotherapy was negatively associated with cognitive impairment in both colon cancer (HR, 0.66; 95% CI, 0.45 to 0.97) and rectal cancer (HR, 0.52; 95% CI, 0.31 to 0.88) (Table 2). The age-specific HRs of chemotherapy and radiotherapy were larger in older patients, but were not significant in any age spectra. The protective effect of folate administration was more prominent in older adult patients (Figure 2).

When analyzed by regimen, the FOLFOX regimen was negatively associated with cognitive impairment in both colon cancer (HR, 0.44; 95% CI, 0.32 to 0.60) and rectal cancer (HR, 0.53; 95% CI, 0.34 to 0.82). The FOLFOXIRI regimen showed lower HRs in rectal cancer (HR, 0.49; 95% CI, 0.26 to 0.91), but not in colon cancer (HR, 0.85; 95% CI, 0.58 to 1.27). In general, patients who received the CapeOx or capecitabine-only regimen showed increased hazards for cognitive impairment, although the magnitude of the association varied by primary cancer site (Table 2). The age-specific HRs for FOLFIRI and FOLFOXIRI were higher in older patients than in younger patients, although the HRs were non-significant in all age spectra. The direction of the interaction between the effects of the CapeOx regimens and age was positive in colon cancer, but negative in rectal cancer. The capecitabine-only regimen and radiotherapy did not show significant interactions with age (Figure 3).

The sensitivity analysis showed similar results to those of the main analysis. When cognitive impairment was redefined as having 2 or more corresponding ICD-10 codes, the estimated HRs of chemotherapy and radiotherapy did not show significant differences from the main analyses (Supplementary Material 5). In landmark analyses, as the time lag increased, the estimated HRs of chemotherapy decreased and protective effects of folate therapy became more prominent (Supplemental Material 6). The interaction trends detected in landmark analyses were similar to those of the main analyses (Supplemental Material 7). Landmark analyses by regimen showed similar trends to those of the main analyses (Supplemental Material 8).