Impact of primary cancer features on behaviour of colorectal liver metastases and survival after hepatectomy

Background: Markers of tumour biology may be valuable prognostic indicators after hepatic resection of colorectal cancer liver metastases (CRLMs). Identification of the aggressiveness of these metastases might inform the appropriateness of hepatic surgery. Methods: Patients undergoing liver resection for CRLMs between January 2001 and July 2013 in four tertiary hospitals were reviewed. A mathematical model to estimate CRLM doubling times was constructed for patients with metachronous metastases. Tumour doubling time was investigated in relation to the features of colorectal cancer, including KRAS status. The hazard rate for recurrence and death following hepatectomy was explored through the Kernel-smoothed estimator. Results: Of 1063 patients undergoing liver resection for CRLMs, 361 with metachronous metastases undergoing single-stage hepatectomy were analysed. The mean doubling time in patients not receiving chemotherapy between surgery for colorectal cancer and CRLM was 71⋅4 days. Tumour doubling time
was shorter in patients with more advanced primary tumour stages, with mutant KRAS and in those who did not receive chemotherapy. For fast-growing CRLMs (doubling time less than 48 days), the risk of recurrence was highest within the first postoperative year, and was about 7 per cent per month.
Conclusion: Primary features of colorectal cancer were linked to aggressiveness of CRLMs as measured by doubling time.

Introduction
Colorectal cancer is one of the most common malignan- cies worldwide, accounting for at least one million new cancer cases each year1. Spread to the liver occurs com- monly, with roughly one-third of patients developing dis- tant metastases2,3. Although surgical therapy for patients with resectable colorectal liver metastases (CRLMs) offers a 20 per cent possibility of cure 6 – 7 years after resec- tion, many will not achieve long-term benefit3,4. As the indications for liver-directed therapies broaden, the abil- ity to identify patients likely to benefit from intervention is important. Several morphological factors, such as number and size of metastases, carcinoembryonic antigen (CEA)level and the interval between primary tumour and the recognition of hepatic metastases, have been shown to be useful in predicting outcome5.The potential aggressiveness of colorectal cancer is read- ily evident when relapses occur early after resection of the primary tumour, when it recurs in the liver with large or bilobar metastases, and when there is little or no mea- surable response to chemotherapy. Indicators of tumour biology and how they might influence outcome are ofincreasing interest. Mutation of the KRAS gene may be anindicator of biological aggressiveness6 – 8. The relationship between the growth rate of CRLMs and biological features of colorectal cancer may provide additional information related to outcome.The aim of the present study was to estimate CRLM growth rate by assessing tumour doubling time in a con- secutive series of patients with metachronous liver disease who had undergone liver resection. Although assessment of tumour growth rate requires at least two observations of volumes, it is difficult to justify this for every patient in a clinical setting. A mathematical approach was adopted to overcome this problem to validate and investigate the relationship of doubling time with clinical features and out- comes after liver resection.The study was approved by the Institutional Review Boards and ethical committees of all participating centres.Between January 2001 and July 2013, consecutive patients undergoing liver resection for CRLMs with no extrahepatic disease were identified from prospectively developed databases at four Italian tertiary referral hos- pitals for inclusion in the present analysis.

Patients with incomplete data, who had undergone preoperative portal vein embolization, or in whom a two-stage hepatectomy had been adopted were excluded. Patients presenting with synchronous CRLMs were also excluded, retaining those with metachronous disease. In this group, those with no history of receiving chemotherapy in the interval between primary colorectal cancer surgery and CRLM resection were used for initial doubling time estimation. Patients were deemed to have technically respectable dis-ease when the metastases could be resected completely and the future remnant liver volume was considered ade- quate. Demographics, clinical data and CRLM features were collected for each patient. The resection margin was defined as R0Par when it was greater than 1 mm and as R1Vasc when the CRLM was detached from a major vessel structure. Both were considered curative, as sug- gested recently9,10. Otherwise, the resection margin was defined as R1Par. Primary tumour characteristics, time from resection of the primary tumour to liver resection and the adoption of chemotherapy (either adjuvant after primary colonic surgery or neoadjuvant before liver resec- tion) were also detailed. When available, the mutational status of KRAS, determined on the primary colorectal can- cer, was recorded. The chemotherapy strategy was pre- scribed by the oncologist in charge. Irinotecan, oxaliplatin, capecitabine and monoclonal antibodies were offered to patients after their approval by the Italian national health- care system. Follow-up included clinical examination, esti- mation of serum CEA level, and CT or MRI at 3, 6 or every 12 months, as determined by the oncologist and clin- ical circumstances. Follow-up was carried out by assessingthe clinical records of the respective institutions or by tele- phone contact with the referring clinician or patient.The solution proposed relied in the concept of the gen- esis of liver metastases from the deposition of unde- tected/microscopic colorectal cancer tumour spheroids that initiate CRLM macroscopic growth. On this basis, the following mathematical solution was adopted.Patients with no history of chemotherapy in the interval between primary colorectal cancer surgery and CRLM resection were considered initially. The formula developedby Schwartz, doubling time = t × ln2/(ln(V 2) − ln(V 1)), was applied for estimation of doubling time, where t is the time interval between measurements, and V 2 and V 1 are the tumour volumes at the end and beginning of the time interval11,12. V 2 was defined as the largest CRLM diameter in the resected specimen. V 1 was obtained by assuming that aggregates of colorectal cancer cells had already set- tled in the liver at the time of primary tumour surgery, but were as yet undetectable.

A plausible range of CRLM sizes that would exist before macroscopic diagnosis at the time of colonic resection was assumed. Research suggested that V 1 could be calculated from a starting size of a tumour spheroid of 200 μm in radius, and a maximum diameter of 6 mm was taken on the basis that this represented the lower limit for CRLM identification by common imaging techniques13,14.Between the two V 1 extremes (from 200 μm to 6 mm), a triangular distribution of V 1 was assumed as the best approximation when the real distribution was unknown15. Ten different random V 1 values within the applied rangewere simulated. The variable t was calculated as the time between primary colorectal cancer surgery, where unde- tectable tumour aggregates were assumed to be already present, and the time of liver resection, at which point the tumour size provided the V 2 estimate. Doubling time was calculated for each of the ten simulated values of V 1, main- taining the value of t fixed as defined above. The average doubling time value over the ten simulations was used as the main outcome measure.To provide plausibility and confirmation of the math- ematical approach, obtained doubling time values were compared with available published data. A systematic search of PubMed and Institute for Scientific Informa- tion (ISI) Web of Knowledge databases was performed for articles published to 1 February 2018 relevant to the tumour doubling time assessment of CRLM. The PubMed database was searched using the following key- words: ‘doubling time’[All Fields] OR ‘growth rate’[All Fields] AND (colorectal[All Fields] AND (‘liver’[MeSH Terms] AND/OR (‘neoplasm metastasis’[MeSH Terms]OR (‘neoplasm’[All Fields] AND ‘metastasis’[All Fields]) OR ‘neoplasm metastasis’[All Fields] OR ‘metastases’[All Fields])). ISI Web of Knowledge was searched using the following terms: ((TS=‘doubling time’ OR TS=‘growth rate’) AND (TS=‘colorectal’ AND TS=‘metastases’)) AND document types: (Article). The search was limited to human subjects and language limitations were imposed. The reference list of identified studies was also searched manually until no additional eligible trials could be identify.The Mann– Whitney test was used to compare doubling time as the distribution was not normal. Natural logarithm conversion of doubling time was used in univariable and multivariable regression models. A nomogram to predict doubling time was generated using coefficients from the multivariable model through the least common denomina- tor approach. Overall survival (OS) was calculated with the Kaplan– Meier method using the log rank test from the date of liver resection until the last follow-up visit (cen- sored) or death. Recurrence-free survival (RFS) was calcu- lated considering intrahepatic, extrahepatic or both types of metastasis as the event and censoring the last follow-up visit and deaths. The hazard rate (HR) of these survival measures was explored by applying the Kernel-smoothed estimator with Epanechnikov function16,17. This analytical technique for lifetime data is flexible, model-free and data-driven, so that no shape assumption is imposed other than that of hazard function as a smooth function over time18. For OS and RFS over the entire study population, the Kernel-smoothed hazard estimator was initially unad- justed for possible confounding factors, whereas when analysing the relationship between doubling time andRFS, the estimator was adjusted at the mean of other co-variables found able to modify prognosis17. Tumour doubling time data retrieved from literature were pooled using the random-effects model of DerSimonian and Laird. Statistical analyses were done using STATA® soft-ware (StataCorp, College Station, Texas, USA). P < 0⋅050 (two-tailed) was considered statistically significant.

Results
Of 1063 patients undergoing liver resection for CRLMs without extrahepatic disease who were identified originally, 97 were excluded because of incomplete data, preoperative portal vein embolization or a two-stage hepatectomy. A further 605 had synchronous CRLMs, leaving 361 patients who had undergone resection for metachronous disease including 149 with no history of chemotherapy in theinterval between primary colorectal cancer surgery and CRLM resection. These patients were used for initial doubling time estimation, the remaining 212 patients being considered afterwards.Baseline characteristics of the study population are shown in Table 1. The mean doubling time value in the 149 patients was 71⋅4 (95 per cent c.i. 62⋅8 to 79⋅9) days. Based on the literature search, 69 abstracts were screened. Ofthese, 61 abstracts were considered not pertinent and eight full texts were reviewed. Five studies19 – 23 were finally con- sidered eligible for inclusion, reporting doubling time for 181 patients with CRLMs. The pooled mean doubling time was 73⋅7 (95 per cent c.i. 43⋅9 to 103⋅5) days. The distribu- tions of doubling time values for patients without and with chemotherapy are depicted in Fig. 1.The relationships between doubling time estimation and features of the primary tumour and CRLMs are reported in Table 2. In univariable analyses, doubling time was sig- nificantly shorter in patients with right-sided colorectal cancer, more advanced tumour stage, mutant KRAS, and in those who did not receive chemotherapy between pri- mary colonic surgery and CRLM resection. Multivariable analysis confirmed stage, KRAS status and chemotherapy as independent predictors of doubling time (Fig. 2).For the subsequent survival analyses, doubling times were divided into tertiles as follows: fast-growing CRLMs, doubling time less than 48 days (120 patients); intermediate-growing CRLMs, doubling time 48 – 82 days (120 patients); slow-growing CRLMs, doubling time more than 82 days (121 patients).Survival analyses after resection of colorectal liver metastasesWithin a median follow-up of 70⋅8 months after liver resec- tion, 243 (67⋅3 per cent) of the 361 patients developed recurrence and 184 died (51⋅0 per cent). Of patients with recurrence, 151 had intrahepatic recurrences (97 confined to the liver, 54 both intrahepatic and extrahepatic).

Median overall survival for the entire cohort was 49⋅1 months, and 1-, 3- and 5-year OS rates were 91⋅2, 58⋅8 and 45⋅0 per cent respectively. Median RFS was 17⋅2 months, and 1-, 3- and 5-year RFS rates were 62⋅8, 32⋅2 and 28⋅0 per cent respec- tively. RFS was significantly reduced by shorter doubling time (P < 0⋅001), higher number of CRLMs (P = 0⋅002), lack of posthepatectomy chemotherapy (P = 0⋅002) and positive resection margins (P = 0⋅006).HRs over time for recurrence and death are plotted in Fig. 3. The HR for recurrence was greatest (approximately 3⋅8 per cent per month) between months 9 and 16 after surgery. The rate then started to decrease, becoming lower than the risk of death from the 36th month onwards. By contrast, the HR for death was greatest (about 1⋅8 per cent per month) between the 20th and 28th month after surgery, and subsequently decreased.Adjusted results on HRs for recurrence by doubling time are depicted in Fig. 4a. Fast-growing tumours had the maximum risk of recurrence (approximately 7 per cent per month) between the ninth and 12th month after surgery. The risk of recurrence then slowly decreased until the 18th month, and then rapidly diminished. Intermediate-growing tumours showed a similar temporalcourse, but with a monthly risk somewhat lower (about 4 per cent per month). In contrast, slow-growing tumours showed their maximum risk of recurrence (about 3⋅3 per cent per month) later in the postoperative period. After the second year from surgery onwards, the HR for recurrence was similar in these three groups.The HR for death of fast-growing CRLMs was greatest (about 2⋅7 per cent per month) between the 16th and 22nd month after surgery, with another increase (around 2⋅0 per cent per month) between the 29th and 39th month. TheHR for death then fell to the rates found for intermediate- and slow-growing tumours (Fig. 4b).All adjusted HRs for recurrence in relation to features found during and after hepatectomy are reported in Table 3. The presence of more than three CRLMs resulted in a monthly HR greater than that that of less advanced CRLMs in the first 2 years after surgery. Positive resection margins led to a consistently higher HR over time com- pared with the HR for negative resection margins. The positive impact of postoperative chemotherapy was con- fined to the first 6 months after operation.

Discussion
The development of CRLMs, recurrence after liver resec- tion and life expectancy depend on the complex interac- tion between the primary colorectal cancer, treatments and patient response. In this study a significant relationship between doubling time in the primary tumour and CRLM growth rate was observed. The fact that T status, N category and KRAS mutational status accounted for different CRLM doubling times relies on the ‘seed and soil’ concept in stochastic and deterministic ways24,25. From a stochastic point of view, circulating tumour cells from the primary colorectal cancer are already present in the liver at the time of primary surgery, or might extravasate from the bloodstream at a second or different time point. In both cases they can either form macrometastases or remain dormant, often for a long period of time24,26 – 28. The initial colorectal cancer stages may involve the dissemination of solitary and rare cell aggregates that cannot proceed to macroscopic tumour development, resulting in long doubling times (for example, T1 of 125⋅9 days). Conversely, more advanced colorectal cancer stages could determine the release of more tumour cell aggregates to seed the liver, with higher probabilities that one (or more) of them will grow into clinically detectable macrometastases, resulting in shorter doubling times (for example, T4 of 52⋅4 days). The deterministic point of view is based on the fact that neoplasms contain genetically diverse tumour cell sub- populations (genetic heterogeneity), each with a different metastatic potential.

In the initial stages of colorectal can- cer, genetic subpopulations with low metastatic potential may be less capable of progressing to macrometastases. On the other hand, advanced colorectal cancers may be more likely to give rise to macrometastases because fur- ther genetic changes have appeared. The present findings support this process. In fact, CRLMs from mutant KRAS tumours grew faster than those derived from wild-type colorectal cancers. The KRAS mutation has been associated with an increased rate of primary colorectal cancer vascu- lar invasion and decreased expression of regulatory mecha- nisms, which result in faster proliferation, and an increased rate of haematogenous spread. The complex interaction between tumour biology, treatments and patient responses after liver resection was evident in the variability of the monthly hazard of recurrence and death in different subgroups. Fast-growing tumours showed a peak in the HR for recurrence within the first year after surgery (up to 7⋅0 per cent). This suggests that aggressive colorectal cancers may have already seeded in the liver and other organs, and that their aggressiveness is maintained after hepatectomy32. This might account for high rates of early recurrence25. This is supported by the observation that slow-growing CRLMs showed a peak HR late after surgery, usually at 1 – 2 years. Considering OS, the peak HR for death was advanced by about 1 year for fast-growing tumours with respect to the recurrence peak, and by about 2 years for more indolent tumours. After this time point, the recurrence HR for faster-growing tumours returned to a level similar (2⋅2 per cent) to that of less aggressive tumours (1⋅9 per cent), although the risk of death remained higher.

The ability to achieve an R0 resection clearly affected RFS. Almost all authors agree that R1 resection is asso- ciated with a higher local recurrence risk9,10,33,34, but this cannot be the sole cause of intrahepatic recurrence. R1 resection may also indicate more diffuse intrahepatic disease, which in turn leads to a more demanding surgical procedure and a lower likelihood of obtaining an R0 resection35. This latter contention is supported by the present results. If an R1 margin was simply due to surgical failure, the HR for recurrence should be high early after surgery and then decrease over time. Instead, it remained consistently above that for R0 margins, suggesting that an R1 margin reflects a more advanced tumour burden with higher metastatic potential over the entire postoperative time period. The present study has limitations. First, only patients with metachronous disease were considered. Estimation of tumour doubling times in patients with synchronous metastases would provide a more comprehensive picture. The second limitation relates to the reliability of the doubling time estimate. The literature measuring CRLM doubling time is more than 10 years old and, with mod- ern treatments, comparability over such a long time interval may not be BI-2852 valid19 – 23. Despite this, the present mathematical model has been built on a sound scientific rationale11 – 13, and findings in the cohort not receiving chemotherapy after primary colorectal cancer resection were in keeping with observed doubling time values in earlier cohorts19 – 23.The present study has provided a comprehensive estima- tion of CRLM growth rates in patients with metachronous liver involvement. It would seem reasonable to explore this further in patients with synchronous metastases and to see how doubling time is affected by different chemotherapy regimens.