Recurrence and Postoperative Death in Patients with Colorectal Cancer: A New Perspective via Semi-competing Risk Framework

Background/Aims: Cancer studies suffer from an overestimation of prediction of survival when both recurrence and death are of interest. This longitudinal study aimed to mitigate this problem utilizing a semi-competing risk approach evaluating the factors affecting recurrence and postoperative death in patients with colorectal cancer. Materials and Methods: This longitudinal prospective study was conducted in 284 patients with resected colorectal cancer who were referred to the Imam Khomeini Clinic in Hamadan, Iran, during 2001-2017. Primary outcomes were postoperative outcomes and patient survival, including time to recurrence (of colorectal cancer), time to death, and time to death after recurrence. All patients who were alive at the end of the study were censored for death and who did not experience recurrence of colorectal cancer were also censored for recurrent colorectal cancer. The relationship between underlying demographics and clinical factors and the outcomes was assessed using a semi-competing risk approach. Results: The results of the multivariable analysis showed that having metastasis to other sites (hazard ratio = 36.03; 95% CI = 19.48-66.64) and higher pathological node (pN) stage (hazard ratio = 2.46; 95% CI = 1.32-4.56) were associated with a raised hazard of recurrence. The fewer chemotherapies (hazard ratio = 0.39; 95% CI = 0.17-0.88) and higher pN stages (hazard ratio = 4.32; 95% CI = 1.27-14.75) showed significantly higher hazards of death without recurrence. Having metastasis to other sites (hazard ratio = 2.67; 95% CI = 1.24-5.74) and higher pN stages (hazard ratio = 1.91; 95% CI = 1.02-3.61) were linked with the higher hazard of death after recurrence. Conclusion: Considering findings on death/recurrence-specific predictors obtained in this study to manage the outcomes in patients with colorectal cancer, tailored strategies for preventive and interventional plans should be deliberated.


INTRODUCTION
Colorectal cancer (CRC) is the most common gastrointestinal malignancy 1 and is expected to increase by 60% to 2.2 million new cases and 1.1 million deaths globally by 2030. 2 Colorectal cancer is the second leading cause of cancer-related death globally, being the second and third leading cause of cancer-related death in men and women, respectively. 3 The incidence rate of CRC is increasing worldwide, especially in developing countries. 4,5 Thus, there is an urgent need to identify the factors that underlie poor outcomes and drive the recurrence in order to guide appropriate treatment and survival strategies.
Although surgery is the primary treatment, the recurrence rate in the first 5 years after surgery is 12.8% for local recurrence and 25.6% for distant metastasis, 6 with about 60%-80% of recurrences within 2 years after resection. 7 Poor survival is associated with a short recurrence interval 8 and may be improved by curative surgical resection if diagnosed early. 9 The main goal of follow-up programs after CRC treatment is to increase survival, and early diagnosis increases the likelihood of patients' survival. 10 To assess the predictors of recurrence and survival in patients with CRC, Lin et al 11 investigated the clinical features and risk factors of patients with CRC oligo-metastases of the liver by curative resection. Node-positive primary tumor (pN stage) and metastasis were found to significantly affect recurrence. Farhat et al 7 investigated the predictors of recurrence in patients with rectal cancer after curative resection, identifying distal resection margin, extracapsular invasion, tumor stenosis, and degree of parietal invasion as prognostic factors. Gunawardene et al 12 found that the stage of the disease was related to recurrence after surgery. In the study by Duineveld et al. 13 the type of visit and the tumor location significantly affected the type of recurrence. Saso et al 14 showed that tumorrelated characteristics, including preoperative serum carcinoembryonic antigen level, preoperative obstruction, tumor invasion, lymphatic invasion, and venous invasion, were significantly associated with disease-free survival. In a 5-year cohort study by Zare-Bandamiri et al 15 the effects of age, tumor location, lymphovascular invasion, and tumor stage on patient recurrence were significant. Another study found that in patients with CRC, high carcinoembryonic antigen level and lymphovascular invasion factors and in patients with rectal cancer, factors including liver metastasis and venous invasion were identified as risk factors for recurrence. 16 Yazilitas et al 17 investigated the relationship of pathological and clinical features with time to recurrence in patients with early-stage CRC, identifying that grade I and superficial tumors (T1-T2) are predictors of late recurrence. Tsikitis et al 18 examined predictors of relapse-free survival in patients with CRC in stages II and III and reported that only T stage was related to CRC recurrence. Therefore, various factors may affect the recurrence and the time interval between recurrence and death, and there is no appropriate agreement for the predictors.
Competing risks occur frequently in the analysis of survival data. Semi-competing risk framework refers to the general setting where the main scientific concern lies in estimation and inference regarding a nonterminal event (e.g., recurrence), the occurrence of which is dependent on a terminal event (e.g., death). 19 However, due to a strong relationship between the 2 event times, the typical survival modeling for the nonterminal event will cause an overestimation of outcome probabilities. 20 A solution has been suggested in the competing risk framework, 21,22 which does not consider the dependence of the 2 events. However, the semi-competing risk analysis framework appropriately treats the dependence between nonterminal and terminal events as a part of the model specification. 23 Cancer studies suffer from an overestimation of the effects of the predictors on survival when both recurrence and death are of interest. This longitudinal study aimed to mitigate this problem utilizing a semi-competing risk approach evaluating the factors affecting recurrence and postoperative death in patients with CRC. Based on an extensive search in the literature, we did not find any study that uses this framework to assess the outcome of patients with recurrent CRC.

MATERIALS AND METHODS Study Design and Setting
This longitudinal prospective study included a cohort of 284 patients after CRC curative resection who were admitted to Imam Khomeini Clinic in Hamadan, Iran, during 2001-2015. The patients were followed until August 2017, and the data were analyzed accordingly. Patients who were diagnosed with colorectal cancer and had undergone surgery for colorectal cancer were included in this study. Patients undergoing transanal surgery were excluded. Data on patient demographics, treatment, mortality and morbidity, and survival were collected.

Predictors
All demographic and clinical/pathological information were extracted from patients' medical records and administrative resources. These included demographic variables such as age at diagnosis (years), gender (female: 1; male: 2), body mass index (kg/m 2 ), and clinical/pathological variables such as metastasis to other sites (no: 0; yes: 1), cancer site (colon: 1; rectum: 2), surgery (no: 0; yes: 1), radiotherapy (no: 0; yes: 1), chemotherapy (no: 0; yes: MAIN OUTCOME VARIABLES Primary outcomes were postoperative outcomes and patient survival, including time to recurrence (of CRC), time to death, and time to death after recurrence. All patients who were alive at the end of the study were censored for death and who did not experience recurrence of CRC were also censored for recurrent CRC. Patients' recurrence status was determined from the patient records for those patients who experienced CRC recurrence, which was computed from the date of surgery to local/distant recurrence in months. The CRC recurrence was defined as coming back of the disease after treatment, which was determined by the patients' doctor. When cancer shows up in the same/different organs (like the liver or lungs) similar to the first time it happened, it is called a local/distant recurrence. Besides, the time to death was computed from the date of surgery to the patients' death. The follow-up was done by the main researcher via a telephone call to confirm the patients' vital status (death, recurrence, or alive) in August 2017. After this time, the data were extracted from the hospital database and considered for the process of analyses.

Ethics Committee Approval
The institutional review board of Tabriz University of Medical Sciences approved the protocol of the study (ethics code: IR.TB ZMED. REC.1 400.4 57). The participants' privacy was preserved. All the processes were in accordance with international agreements (World Medical Association, Declaration of Helsinki, Ethical Principles for Medical Research Involving Human Subjects).

Informed Consent
All participants, or their legal guardian, provided informed written consent on registration in the database. Also, all methods were carried out in accordance with relevant guidelines and regulations.

Statistical Analysis
Data are summarized and reported as mean (SD) and median (minimum-maximum) for the normal and non-normal numeric variables, respectively, and as frequency (percent) for categorical variables. The recurrence/death rates were computed per 1000 persons. Log-rank tests were carried out to compare the survival rates across groups. Primary outcomes were considered as recurrence (of CRC) (called nonterminal event), time to death (called terminal event) without recurrence, and time to death after recurrence. To model the specific risk factors of recurrence and death, semi-competing risk analysis was utilized under the illness-death multistate model. We utilized proportional hazards models characterized by 3 hazard functions, illustrated in Figure 1.
The following specification for hazard functions was considered: 1. a cause-specific hazard for the nonterminal event,   : where h 0g is an unspecified baseline hazard function and β g is a vector of log-hazard ratio (HR) regression parameters associated with the covariates x ig , and γ i is a study subject-specific shared frailty following a Gamma(θ −1 , In our study setting, equations 1-3 are the probability of recurrence of CRC, the probability of death from CRC without any recurrence, and the probability of death from CRC after any recurrence, respectively. To model the risk factors with these outcomes, in univariable and multivariable terms, HRs of the outcomes were estimated using the R 4.2 software utilizing the SemiCompRisks package. 23 The significance level was set at .05.  Table 1 for more information).

Outcome Rates
For both recurrence and death, significantly higher outcome rates were observed among higher age categories, with substantially higher rates in the age group above 70 years. Patients with metastases to other sites also had much higher rates of both outcomes. In addition, patients who had received less than 6 types of chemotherapy were associated with higher outcome rates for both recurrence and death; however, the rates decreased for patients who had received more than 6 chemotherapies. Nonterminal and terminal event rates raised significantly as the disease stage, pT stage, and pN stage levels, increased (all P < .05) (see Table 2 for more information).

Univariable Semi-competing Risk Model
The study results indicated significant and positive associations between upper age categories with a higher hazard of recurrence (HR = 2.06) and death without recurrence (HR = 3.62). Metastasis to other sites was linked with a raised risk of recurrence (HR = 36.26) and death after recurrence (HR = 2.56). The higher number of chemotherapies significantly decreased both death without recurrence (HR = 0.39) and the hazard of death after recurrence (HR = 0.30). Also, a poor level of differentiation was connected with the outcome of death after recurrence (HR = 3.64). Patients with higher disease stages had significantly higher hazards for all 3 outcomes occurring (all HR >4). Similar results were observed for pT stage and pN stage (Table 3).

Multivariable Semi-competing Risk Model
Metastasis to other sites was associated with a raised hazard of recurrence (HR = 36.03) and the hazard of death after recurrence (HR = 2.67). The higher number of chemotherapies significantly decreased the hazard of death without recurrence (HR = 0.39). Also, patients with a higher pN stage showed significantly higher hazards of all 3 outcomes occurring (all HR > 1.9) ( Table 4).
Higher recurrence and death rates were observed among patients with metastasis, among patients who had less than 6 chemotherapies, and among patients with higher pN stage (Figures 2-4).

DISCUSSION
This study aimed to model the effect of demographics and clinical characteristics on recurrence and postoperative death in patients with CRC, utilizing a semi-competing risk analysis. The study showed that metastasis to other sites, higher pN stage, and undergoing fewer chemotherapies were associated with a raised risk of recurrence and death. Higher pN stage was associated with significantly higher hazards for recurrence, death without recurrence, and death with recurrence, while metastasis to other sites was associated with significantly higher hazards of recurrence and death after recurrence.
Recurrence after surgery is one of the major problems affecting the long-term survival of CRC patients. Often occurring within the first 2 years after surgery, recurrence is associated with poor survival outcomes in the first 5 years. 25 In the present study, about 80% of patients who experienced a recurrence of disease after surgery died by the end of the study, while in patients who did not   experience a recurrence, the mortality rate was 10.5%. The recurrence rate in the first 5 years post-curative resection has been reported to be 25%-37%. 11,12 In line with our results, other studies showed a direct link between age and recurrence. 15 Some studies have also demonstrated a significant connection with age and local and distant recurrence rates and excess mortality, 26 although this was not demonstrated in a study by Kawai et al. 27 Therefore, diagnosis of the disease at an early age can lead to a complete cure and reduce the risk of recurrence.
Metastasis to other sites showed a significant association with both nonterminal and terminal events, as was evident in similar studies, 26 emphasizing that diagnosis of the disease before metastasis occurs is crucial in improving outcomes.
Similar to our findings, complementary treatment has been shown to have a significant preventive effect on recurrence and mortality and chemotherapy to effectively reduce recurrence, 28 suggesting that chemotherapy can decline the hazard of recurrence and death. Although curative surgical resection is an effective treatment in patients with CRC, poorer diagnosis is associated with later stages of the disease, and about 30% of patients have a recurrence after surgery, 4,29 which is compatible with the results of this study. In other studies, the stage of disease was significantly related to death, 26 recurrence, 26 and death after recurrence. 11    In the present study, we did not separate recurrences into early and late events in order to fulfill the minimum required sample size for the semi-competing risk analysis. There was no statistically significant difference in the median survival between early (<2 years) and late recurrence (>2 years) (results not shown). In other studies, no significant difference was observed between early and late recurrence. 30 We utilized a semi-competing risk approach to model the risk factors of recurrence and death. This framework provides us with unbiased estimates of probabilities of the outcomes. Since, in studies like the present study, there is a strong correlation between the terminal and nonterminal events (in the present study: r = 0.92, P < .001), simple utilization of a conventional survival model leads to an overestimation of the terminal event rates. In the classical approaches, the terminal event is considered as an independent censoring mechanism, which is not the case in reality and would lead to an overestimation of the effects of the risk factor on the outcomes. 20,23 Utilizing semicompeting risk analysis, we considered the dependence between the 2 events to be part of the model specifications and therefore avoid the bias. The largest limitation of the study was the width of the CIs in some analyses, especially in the multivariate analysis, and the CIs for the non-computable parameter estimates. The small sample size in the subgroups is the likely major contributor to this limitation. The Bayesian approach in estimating parameters and their credibility intervals may be the solution. In addition, the introduction of nonterminal-and terminal-specific prediction nomograms into the results could be useful for risk assessment. 31

CONCLUSION
Identifying the factors affecting the recurrence of disease after surgery and the duration of recurrence until the death of patients can provide more appropriate treatment and follow-up strategies for high-risk patients.

Availability of Data and Materials:
The data that support the findings of this study are available from MAJ, but restrictions are applied to the availability of these data, which were used under license for the current study, and are not publicly available. Data are, however, available from the authors upon reasonable request by MAJ.
Ethics Committee Approval: The institutional review board of Tabriz University of Medical Sciences approved the protocol of the study (ethics code: IR.TB ZMED. REC.1 400.4 57).
Informed Consent: All participants filled and signed an informed consent and assent, and all methods were carried out in accordance with relevant guidelines and regulations.