Comparison of the clinicopathological characteristics and their impact on outcomes of resected colorectal liver metastases across time

Background: Resection of colorectal liver metastases (CLM) has been established as the standard of care. This study aims to compare the change in clinicopathological characteristics of patients who underwent curative resection of CLM across two time periods – 2000 to 2010 (T1) and 2011 to 2016 (T2) and evaluate the prognostic impact of these characteristics on survival outcomes. Methods: Patients who undergo liver resection for CLM at Singapore General Hospital from January 2000 to December 2016 were identified from a prospectively maintained database. The primary end point was overall survival. Results There were 183/318 (57.5%) patients and 135 /318 (42.5%) patients in T1 and T2 respectively. There was a lower proportion of patients who had nodal metastases from primary colorectal cancer and clinical risk score (CRS) less than 3 in T2 when compared to T1. There was no difference in survival between both time periods (HR 0.850 95% CI 0.601 – 1.201 p = 0.357). Independent predictors of survival for the cohort were CEA levels ≥200ng/ml, primary tumour grade and lymph nodal status. Independent predictors of poor survival in T1 were poorly differentiated colorectal cancer and nodal metastases while in T2, independent predictors of poor survival were multiple liver metastases and nodal metastases. Conclusion: Nodal metastases from primary colorectal cancer is an independent predictor of poor survival across time for resectable CLM. Although there is no difference in survival between the two time periods, patients with multiple liver metastases should be carefully considered prior to surgery as it is also an independent predictor of overall survival.


Introduction
Colorectal cancer (CRC) imposes a heavy burden on healthcare and is a leading cause of cancer death worldwide. 1 In Asia, the incidence of CRC has increased by as much as two-to four-fold in the past few decades, and in some countries is now the most common cancer diagnosed. 2 About 25% of patients will present with stage IV CRC (synchronous metastases) and up to 50% of patients overall will go on to develop metachronous liver metastases. 3−4 While the 10-year survival rate for patients with Stage I CRC is about 90%, less than 10% of patients with non-resectable stage IV disease will survive 5 years. 3 Twenty to thirty percent of patients with stage IV CRC will have potentially resectable metastases confined to the liver. In these patients, liver resection with curative intent may be performed, with substantially improved 5-year survival rates ranging between 16% and 71% being reported. 5−8 Well-established independent predictors of disease recurrence and survival after hepatic resection include stage of the primary tumour, preoperative carcinoembryonic antigen (CEA) levels, hepatic tumour size, number of hepatic metastases, time from treatment of primary tumour to diagnosis of hepatic metastases and presence of extrahepatic disease. 9 Extensive research has been conducted on the prognostic value of Fong's clinical risk score 5 and this has been validated in several studies.
However, the treatment of colorectal liver metastases has advanced over the last 2 decades with developments in chemotherapy, surgical technology and also careful patient selection with greater emphasis on multi-disciplinary tumour boards.
Therefore, this paper aims to compare the change in the clinicopathological characteristics of patients who underwent curative resection of colorectal liver metastases across two time periods -2000 to 2010 (T1) and 2011 to 2016 (T2) and evaluate the prognostic impact of these clinicopathological characteristics on the survival outcomes between the two time periods.

Methodology
Patients who had undergone liver resection with curative intent for colorectal liver metastases in Singapore General Hospital from Jan 2000 to Dec 2016 were identified from a prospectivelymaintained database. Patients with extrahepatic metastases, multiple recurrences or second primary tumours were excluded. Patient demographics, primary tumour and CLM clinico-pathological characteristics were collected. Pre-operative imaging included cross-sectional computer tomographic (CT) scans of the abdomen and pelvis both at the time of the diagnosis of CLM and the initial diagnosis of the colorectal primary. The study was approved by the institutional review board.
The two time periods 2000 to 2010 (T1) and 2011 to 2016 (T2) were decided with the development of our institution's dedicated multi-disciplinary tumour boards from 2011 onwards. Synchronous liver metastases were defined as the presence of liver lesions detected at the time of diagnosis of colorectal cancer. The number and size of liver lesions were determined on preoperative crosssectional imaging, predominantly contrast-enhanced CT or MRI scans. The extent of liver resection was defined based on Brisbane 2000 nomenclature 10 . Major liver resection was defined as resection of 3 or more segments. Fong's Clinical Risk Score (CRS) for Colorectal Liver Metastases was calculated. The standard follow-up protocol at outpatient visits every 3 to 6 monthly included clinical examination, serum carcinoembryonic antigen (CEA) level and contrast-enhanced CT scan (chest, abdomen, and pelvis).
Separate survival analyses were conducted for the two study periods. Differences in patient baseline characteristics between the two time periods were compared using chi-square tests for binary variables, and the Student's t-test or Mann Whitney U test for normally and non-normally distributed continuous variables respectively. Overall survival was calculated from the time of hepatic resection until death, and patients who were alive at their last follow-up visit were censored.
Prior to analyses, missing covariate data points were multiply-imputed (M = 50 imputations) using multivariate chained equations, with augmented logistic regression for dichotomous variables and predictive mean matching (five k-nearest neighbours) for continuous variables. Hazard ratios (HRs) were computed by fitting Cox proportional hazard models to each imputed dataset and subsequently combined according to Rubin's rules.
To assess if there was a change in prognostic impact of a given covariate, interaction between the study period and the magnitude of the hazard ratio was evaluated by testing the equality of Cox model regression coefficients obtained at the two time periods using the mi testtransform command in Stata version 16 (StataCorp, TX, USA). For this analysis, nominal interaction P values of less than 0.05 (two-sided) were considered to indicate that the prognostic impact of a given covariate had changed significantly between the two time periods.
In a separate analysis, we sought to identify factors that were independently associated with overall survival within each of the two time periods. This was accomplished by fitting two Cox proportional hazards model (one for each time period) using a stepwise backward elimination procedure at a P threshold of < 0.10 in SPSS version 26.0 (SPSS Inc., Chicago, IL).

Results
We identified 318 patients for this analysis (mean age, 60. 6  respectively. The clinicopathological characteristics of the patients in both time periods are described in Table 1. There was no statistical difference between the 2 time periods with regards to clinical characteristics such as age, gender, ASA score, gender and CEA levels. There was no difference in primary colorectal cancer pathological characteristics such as T-stage, location of the primary colorectal cancer and also the characteristics of the liver metastases such as size, distribution and number of liver metastases.   Table 3 summarizes the univariate survival analysis performed separately in the 2 time periods.
Raised CEA levels ≥ 200 ng/ml, nodal metastases from primary colorectal cancer, and CRS score ≥ 3 were associated with poor survival in both time periods and there was no significant statistical difference upon comparison of their hazard ratios. Disease free interval (DFI) ≤ 12months was associated with poorer overall survival in both time periods. In T1, DFI ≤ 12 months had a higher hazard ratio when compared to T2 and this difference was statistically significant (p = 0.0178). Poor primary tumour grade, synchronous liver metastases and neoadjuvant chemotherapy were associated with poorer overall survival in T1 but did not affect survival in T2. The differences in hazard ratios for these factors were not significant.  Table 4 summarizes the multivariate survival analysis performed separately in the 2 time periods. In T1, independent predictors of poor survival were poor tumour grade and nodal metastases. In T2, independent predictors of poor survival were multiple liver metastases and nodal metastases. These results demonstrate that nodal metastases from the primary colorectal cancer continue to remain an independent predictor of survival across the 2 time periods. Table 4 Multivariable In this study there were fewer patients with nodal metastases and high CRS scores who underwent curative surgery from T2 compared with T1. This could be attributed to multidisciplinary meetings which account for better patient selection. There was suggestion of improved survival in T2 compared with T1 as demonstrated in the univariate survival analysis. However, after adjusting for CEA levels, primary tumour grade and nodal metastases there was no significant difference in survival between the 2 time periods. It is possible that the magnitude of the prognostic impact of variables used in the CRS differs, thus accounting at least partially for this.
Although liver resection for surgically operable CLM is associated with survival benefit, appropriate patient selection ensures that the appropriate treatment is given to each patient 13 . Technical resectability refers to the feasibility of obtaining a margin-negative resection while preserving adequate functional liver remnant (FLR) with vascular inflow and outflow and biliary drainage 14 .
However, resectability of CLM should be determined with a multidisciplinary team with input from hepatobiliary surgeons, oncologists, radiologists and pathologists. In addition to technical considerations, there is increasing emphasis on oncological and prognostic evaluation prior to surgical resection. The aim is to select patients with better chance of cure or sustained disease remission with consideration of the underlying tumour biology. Traditional prognostic factors include characteristics of primary colorectal cancer such as the T stage and nodal status and characteristics of the CLM such as size, number of lesions, margin status as well as CEA levels and disease-free interval between primary colorectal cancer and development of the liver metastases 15,16 .
Clinically, the prognostic value of these individual factors are limited. The Fong clinical risk score (CRS) together with other scoring systems were developed to provide an overall risk assessment to predict long term survival for patients prior to curative resection. However, these scores fail to demonstrate predictive accuracy for long term survival and their clinical utility remains uncertain 17,18 .
Similarly, in this study CRS score was calculated and a high score of ≥ 3 was associated with poor survival on univariate analysis demonstrated for the overall cohort and even across both time periods.
However, with multivariate survival analysis CRS score was not an independent predictor of survival.
Regardless, our study demonstrated that individual prognostic factors such as nodal metastases from primary colorectal cancer remain a strong independent predictive factor for overall survival in the overall cohort and even across both time periods. Interestingly, poorly differentiated colorectal cancer was an independent predictor for poor overall survival in T1 but it was not predictive for survival in T2. This could be due to smaller proportion of patients with poorly differentiated colorectal cancer in T2 compared to T1, however the difference was not statistically significant.
Although there was similar proportion of patients with multiple liver metastases in both time periods, it was an independent predictor of overall survival in T2. This is supported by studies that demonstrate multiple liver metastases were associated with negative survival outcomes 5,9 . This result also demonstrates that there are better survival outcomes for patients with solitary liver metastases in T2 compared with T1. This difference could be attributed to better surgical techniques, development of energy devices and increasing use of liver parenchyma sparing operations. Given that multiple liver metastases are a poor predictor of survival, this could be a contributing factor to no significant difference in survival outcomes being seen between the two time periods based on the multivariate survival analysis of the overall cohort. The implications to clinical practice would be that patients with multiple liver metastases which appear technically resectable would need to be assessed carefully for the oncological benefit of curative liver resection given that it is a poor predictor of overall survival.
This study has evaluated the trends in management of CLM as well as the changes in prognostic factors over the 2 time periods. This has provided information on clinical practice that may influence patient outcomes. However, the limitations of this study would include the small sample size which upon subgroup survival analysis result in the study being under powered and therefore the predictive value of other prognostic factors may not have been identified. Another limitation would be the presence of missing data which will affect the sample size for survival analysis. Given the retrospective observational nature of this study from a single centre, the results are not entirely representative. However, this study highlights the need for further multicentre trials or populationbased cohort studies to properly evaluate the changes in trends and management of CLM over time.

Conclusion
In conclusion, nodal metastases from primary colorectal cancer continues to remain an independent predictor of poor survival across time for patients who undergo curative liver resection for colorectal liver metastases. Although there are changes in patient selection over time, there is no difference in survival outcomes between the two time periods. Therefore, patients with multiple liver metastases should be carefully evaluated prior to curative liver resection given that it is also an independent predictor of overall survival.