Overall risk and risk factors for metachronous peritoneal metastasis after colorectal cancer surgery: a nationwide cohort study

Background This study aimed to identify the cumulative incidence and risk factors of metachronous peritoneal metastasis (M‐PM) from colorectal cancer in patients who had intended curative treatment. Methods Patients with colorectal cancer were identified using the Danish Colorectal Cancer Group database for 2006–2015. The Danish Pathology Registry and the Danish National Patient Registry were used to identify M‐PM to 2017. Risk factors were estimated by multivariable absolute risk regression, treating death and other cancers as competing risks. Overall risk and risk differences (RDs) were estimated at 1, 3 and 5 years. Results In 22 586 patients with colorectal cancer, the overall risk of M‐PM was reported to be 0·9 (95 per cent c.i. 0·8 to 1·0) per cent at 1 year, 1·9 (1·8 to 2·1) per cent at 3 years and 2·2 (2·0 to 2·4) per cent at 5 years. Advanced tumour category ((y)pT4 versus (y)pT1) increased the RD of both M‐PM (2·9 (95 per cent c.i. 2·1 to 3·7) at 1 year and 6·0 (4·9 to 7·2) at 3 years) and lymph node involvement ((y)pN2 versus (y)pN0) (2·5 (1·8 to 3·2) at year and 4·3 (3·2 to 5·3) at 3 years). No further increase in risk was observed at 5 years. In a subanalysis, tumour‐involved resection margin (R1 versus R0) was associated with M‐PM with a RD of 3·9 (1·6 to 6·2) at 1 year and 5·9 (2·6 to 9·3) at 3 years. Conclusion The overall risk of M‐PM in patients with colorectal cancer is low, but is increased in advanced T and N status. Follow‐up of at least 3 years after colorectal cancer surgery may be necessary, given the potential curative treatment of early diagnosed M‐PM.


Introduction
The long-term survival of patients with colorectal cancer has improved significantly over the past few years; in Denmark, the relative 5-year survival rate increased from 58-59 per cent in 2001-2004 to 63-65 per cent in 2009-2012 1 . The improvements made so far may be related to several factors, including multidisciplinary team management, the introduction of minimally invasive surgery, implementation of total mesorectal and complete mesocolic excision, specialization and centralization of treatments, pathological/molecular evaluations, and general improvements in radiological assessments, radiotherapy and medical oncology 1,2 . However, recurrence is still an issue.
Registry-based studies have reported the incidence of metachronous peritoneal metastasis (M-PM) to be 3⋅5 per cent at a median of 18 months after diagnosis 3 , rising to 6 per cent within 5 years 4 . Risk factors identified for M-PM include advanced T and N categories 3,5 -8 , bowel perforation, emergency surgery 5,7,9 and non-radical resection 3,5 .
Two different strategies for the prevention and early detection of M-PM have been proposed, including prophylactic adjuvant hyperthermic intraperitoneal chemotherapy (HIPEC) 10 and early detection with second-look surgery plus HIPEC 11 . However, selection of appropriate patients for these treatment options was based on previously identified risk factors for M-PM 5,12 , and needs further investigation.
This study aimed to describe the overall 5-year risk of developing M-PM in patients with colorectal cancer, and to identify risk factors for M-PM following intended curative surgery.

Methods
A nationwide registry-based cohort study was conducted in Denmark according to the STROBE criteria 13 . All 5⋅8 million Danish citizens have access to a public tax-supported healthcare system and are assigned a unique ten-digit personal registration number, enabling unambiguous individual-level record linkage between registers.
Patients diagnosed with colorectal cancer in the Danish Colorectal Cancer Group database between 2006 and 2015 were identified. In March 2014, the implementation of a national screening programme with faecal immunochemical testing 14 led to diagnoses of both symptomatic and asymptomatic patients with colorectal cancer. During the study period, national guidelines 15 recommended that follow-up of patients with colorectal cancer should include, as a minimum, CT of the thorax and abdomen at 12 and 36 months after surgery.
The Danish Data Protection Agency approved the study (number 1-16-02-441-16). Ethical approval is not required for registry-based studies.
Inclusion and exclusion criteria for the study population are summarized in Fig. 1. Patients were included if they had undergone a pathologically confirmed R0 or R1 bowel resection for colorectal cancer. Patients with metastasis to liver and/or lungs were included if the surgery was performed with curative intent. The date of colorectal cancer diagnosis plus 180 days was considered as the index date. To reduce immortal time bias, the index date was considered as the beginning of follow-up 16,17 .
Patients were excluded if they had emigrated or had been diagnosed with other (non-colorectal cancer) malignancies within a period of 5 years plus 180 days before the index date. Finally, patients were excluded if diagnosed with synchronous peritoneal metastasis (S-PM) (identified before the index date) or if they had died between the diagnosis and index dates. Synchronous PMs were defined as PMs identified within 180 days of the diagnosis of colorectal cancer 18,19 .

Registries
Data from the Danish Colorectal Cancer Group database were merged to identify M-PM and cross-check follow-up. In particular, the Danish National Patient Registry provided information about diagnostic coding of PM, the Danish National Pathology Registry was reviewed for histologically proven PM, and the Danish Civil Registration System for follow-up and vital status. The Danish Colorectal Cancer Group database contains information about all patients with first-time colorectal cancer since 2001, with data completeness of more than 95 per cent. The database also contains information on patient characteristics, radiological evaluation, surgical and oncological treatment, pathology reporting, and the postoperative course within 30 days of surgery 20 .
The Danish National Patient Registry provides longitudinal data from 1977 regarding administrative and clinical data, and contains information about hospital admissions and outpatient contacts with the healthcare system. Diagnoses were recorded using ICD-10 codes from 1994, whereas treatment and procedures are registered by using a Danish version of the Nordic Medico-Statistical Committee (NOMESCO) Classification of Surgical Procedures.
The Danish National Pathology Registry was established in 1997, and all pathological examinations performed in Denmark are registered following a uniform guideline. Each specimen is linked to the personal registration number, the hospital department responsible for treatment, the date of request, the specific Danish Systematized Nomenclature of Medicine codes 21 , and other sources of data.
The Danish Civil Registration System is an administrative register established in 1968 to record information about residency and vital status of all Danish citizens. The register is updated daily and has a high accuracy, allowing for complete long-term follow-up 22 .

Identification of metachronous peritoneal metastasis
In the Danish National Patient Registry, M-PM was identified by two ICD-10 codes: 'metastasis in the retroperitoneal space or in the peritoneum' (C786) and 'metastasis to the ovaries' (C796). In the Danish National Pathology Registry, M-PM was identified as a specimen/biopsy with a topography code as peritoneum, combined with a specific morphology code representing metastatic spread from the colon or rectum (Appendix S1, supporting information).

Potential risk factors
Variables included age (less than 60, 60-75 or more than 75 years), sex, tumour localization (right colon (caecum The index date is 180 days after the date of colorectal cancer diagnosis. and ascending colon), right colonic flexure, transverse colon, left colonic flexure, left colon (descending colon and sigmoid), rectum), surgery (elective or emergency), perforation of the tumour as assessed at operation by the surgeon (no; yes, encapsulated (perforation not free in the peritoneal cavity); or yes, free to the peritoneum), pathologically assessed T category ((y)pT0-1, (y)pT2, (y)pT3 or (y)pT4), pathologically assessed N category ((y)pN0, (y)pN1 or (y)pN2), tumour histology (adenocarcinoma or other), extramural venous invasion (EMVI) (available from 2009), radicality of the resection (R0, no macroscopic or microscopic tumour residual left in resection margins; R1, microscopic tumour residual left 1 mm or less from resection margins (included in 2014 owing to the implementation of new strict national guidelines); R2, macroscopic tumour tissue left during resection of the tumour), and systemic chemotherapy (yes or no). Information on tumour histology was obtained from the Danish National Pathology Registry (Appendix S1, supporting information). Data on systemic chemotherapy before the index date were obtained from the Danish National Patient Registry using the specific NOMESCO codes for systemic chemotherapy (Appendix S1, supporting information); however, no information about cycles, doses or frequency was available 23 .

Statistical analysis
Patient characteristics and demographics are presented as categorical variables by counts and percentages.
Patients were followed up from the index date to the date of diagnosis of M-PM or non-colorectal cancer, death or to 25 January 2017. Cumulative incidence (risk) curves for M-PM were estimated; all-cause mortality (death) and diagnosis of non-colorectal cancer were considered to be competing risks 24,25 .
Analysis of potential risk factors was done as a complete-case analysis: only patients with no missing values for potential risk factors were included. To assess 1-, 3-and 5-year risk differences (RDs) with 95 per cent c.i. for M-PM associated with each risk factor, a multivariable absolute risk regression model including all risk factors (except radicality and EMVI) was conducted, adjusting for year of colorectal cancer diagnosis and co-morbidity, as assessed by the Charlson Co-morbidity Index (CCI) (categorized as low (score 0), medium (score 1-2) or high (score greater than 2)). Death and non-colorectal cancer were considered as competing risks 26,27 .
Details of radicality (R1) and EMVI were not available for the whole study period, and were therefore investigated in subgroups of the cohort restricted to relevant calendar periods, using models adjusted only for age, sex and co-morbidity (CCI) owing to the small number of patients with M-PM.
Statistical analysis was performed using STATA ® software release IC15 (StataCorp, College Station, Texas, USA).

Results
Overall, 42 250 patients with colorectal cancer were identified in the DCCG database, 22 586 of whom met the study criteria (Fig. 1). These patients did not have a cancer diagnosis 5 years before the diagnosis of colorectal cancer, were assessed negative for synchronous PMs, and underwent intended curative resection (R0 or R1) of the tumour, including concomitant procedures if liver and/or lung metastases were present. Patient, tumour and treatment characteristics at the time of colorectal cancer diagnosis are shown in Table 1.

Overall risk of metachronous peritoneal metastases
Some 533 of the 22 586 patients (2⋅4 per cent) developed peritoneal metastases, among whom 84⋅4 per cent were  identified in the Danish National Patient Registry and 6⋅0 per cent in the Danish National Pathology Registry; an additional 9⋅6 per cent were identified in both registries. The overall risk of M-PM after intended curative surgery for colorectal cancer was 0⋅9 (95 per cent c.i. 0⋅8 to 1⋅0) per cent at 1 year, 1⋅9 (1⋅8 to 2⋅1) per cent at 3 years, and 2⋅2 (2⋅0 to 2⋅4) per cent at 5 years (Fig. 2). Death and non-colorectal cancer were assessed as the major competing risks (Fig. 3).  Metachronous peritoneal metastasis Non-colorectal cancer Death (all-cause mortality)

Risk factors for metachronous peritoneal metastases
Results of the absolute risk regression analyses are shown in Table 2

Tumour localization
Left colon 0 (reference) 0 (reference) Death and other cancer were treated as competing risks. Values in parentheses are 95 per cent confidence intervals. *A total of 21 581 complete cases were included in the multivariable analysis, adjusted for all risk factors in the table, including year of diagnosis and co-morbidity (Charlson Co-morbidity Index score). †The baseline risk of metachronous peritoneal metastases for a reference person was 0⋅2 (95 per cent c.i. 0 to 0⋅7) per cent at 1 year and 0⋅6 (0 to 1⋅5) per cent at 3 years. ‡Data available from 2009, adjusted only for age, sex and co-morbidity, for a restricted group of 13 222 patients (complete cases in the multivariable analysis and complete information for extravenous mural invasion). §Data available from 2014, adjusted only for age, sex and co-morbidity, for a restricted group of 5861 patients (complete cases in the multivariable analysis and complete information for R1 resection available from 2014 and 2015). risk at 3 years. Estimates of the 5-year RD showed similar associations to the 3-year estimates (data not shown).
In the multivariable absolute risk regression analysis, tumour perforation did not correlate with an increased risk of M-PM. Therefore, a post hoc analysis was conducted to compare mortality in patients with tumour perforation and in those without. This analysis showed that the risk of death was substantially higher in patients with tumour perforation (data not shown); thus the null result could be due to competing events.
The baseline risk of M-PM for a reference person (the risk in someone who presented with the reference value for all co-variables) was 0⋅6 (95 per cent c.i. 0⋅0 to 1⋅5) per cent at 3 years. Table 2 shows the absolute RD for each factor; this should be added to the baseline risk to obtain the predictive risk of M-PM for a specific patient. According to this analysis, a patient with (y)pT3 N1 rectal cancer undergoing elective surgery would have an estimated total risk of M-PM of 2⋅2 per cent after 3 years: 0⋅6 per cent (overall risk) −0⋅3 per cent (rectal cancer) + 0⋅6 per cent ((y)pT3) + 1⋅3 per cent ((y)pN1) + 0 per cent (elective surgery).
In contrast, a patient with a right-sided (y)pT4 N2 colonic tumour undergoing emergency surgery would have an estimated risk of M-PM of 13⋅4 per cent (0⋅6 per cent (overall risk) + 0⋅6 per cent (right colonic cancer) + 6 per cent ((y)pT4) + 4⋅3 per cent ((y)pN2) + 1⋅9 per cent (emergency surgery)) at 3 years after intended curative surgery. As EMVI and radicality were not multivariably adjusted, the estimated RDs associated with these variables should be interpreted with caution.

Discussion
In this large population-based registry study, the risk of M-PM was nearly 1 per cent after 1 year, increasing to 2⋅2 per cent within 5 years. Overall, (y)pT4 and (y)pN2 categories were assessed as independent risk factors for M-PM, driving the increased risk between 1 and 3 years. All estimates of the 5-year RD showed similar associations to the 3-year estimates. Additionally, right-sided colonic cancers and tumours that required emergency surgery independently increased the risk of M-PM. EMVI and microscopic tumour-involved margins (R1 resections) were also associated with an increased risk, although the estimated RDs for these may require further analysis.
In addition, the present study excluded patients who had non-colorectal cancer within 5 years before the colorectal cancer diagnosis, and non-colorectal cancer diagnosed during follow-up (Fig. 3) was considered as a competing risk to minimize the chances of including PMs that originated from other locations.
Previous studies 26,27 have reported different ranges for M-PM, a variation that may be explained by methodological issues and different time periods. In a prospective clinical study 6 , 5⋅3 per cent (135 of 2542) of the patients were diagnosed with M-PM by CT. All patients included were diagnosed with colorectal cancer between 1989 and 1999, and the incidence of M-PM was not reported at specific time points 6 . In other clinical studies, rates of up to 19 per cent were reported, although these studies analysed M-PM before the further optimization of colorectal surgery 26 . In comparison, registry-based studies 3,5 have found the risk of M-PM to be in accordance with the results reported here.
In the present study, strict inclusion criteria were used, which could explain the lower incidence compared with that reported in other studies. Other reports included patients receiving a R2 resection, distinguished between synchronous and metachronous PMs as early as 30 days after colorectal cancer resection, included patients alive at 30 days after surgery, and did not report any information regarding the presence of other cancers. However, the low incidence observed in the present study may be related to the multidisciplinary improvement in surgical, radiological, oncological and pathological management of colorectal cancer.
The potential risk factors for M-PM were in accordance with those of previous studies 5, 28 , including T and N categories, surgical radicality and emergency surgery as independent risk factors for M-PM. Several other studies 3,8,12,29 have reported similar associations. Still, the identification of patients at high risk of developing M-PM with the aim of including them in preventive and prophylactic clinical trials is challenging 30 . The effects of early detection with second-look surgery including HIPEC were investigated in 41 patients with colorectal cancer 1 year after curative resection with no signs of clinical, biochemical or radiological signs of recurrence 11 . The study documented PMs in 23 of the 41 patients after the second-look procedure; these metastases were treated with cytoreductive surgery and HIPEC, whereas other patients were treated using HIPEC alone. The results suggested a beneficial overall survival and low recurrence rate of PM at a median follow-up of 30 (9-109) months 11 . However, the patients selected for that study included those with S-PM, synchronous ovarian metastasis and tumour perforation. In this respect, the results of the present study suggest that patients with tumour perforation represent a very fragile subgroup with high short-term mortality. This should be taken into consideration when including these patients in future trials.
Of note, the impact of cytoreductive surgery and HIPEC in the present cohort was not investigated as it was considered beyond the scope of this analysis, which aimed to identify risk factors for M-PM. The recently published RCT 31 investigating adjuvant HIPEC in patients with T4 tumours or perforated colorectal cancer (COLOPEC trial), documented no benefit of adjuvant HIPEC in terms of peritoneal metastases-free survival at 18 months. However, during follow-up, PMs were reported in 21 per cent of the overall study population, indicating the magnitude of the risk in patients with high-risk colorectal cancer 31 .
Given the potential for curative treatment of M-PM, the present results indicate that follow-up of at least 3 years after colorectal cancer surgery may be warranted to detect the majority of incident cases.
Although the registries provide complete information regarding follow-up, allowing assessment of the risk of M-PM at specific time points after colorectal cancer surgery, a general limitation of using these registry-based data is that the assessment of PM may not be uniform; the registration originates from diverse centres throughout Denmark and, according to the longitudinal design, treatments changed over the years 32 . This might introduce information bias, although data were adjusted for the year of colorectal cancer diagnosis in the multivariable absolute risk regression model.
In addition, M-PM was identified by the use of two nationwide registries: the Danish National Pathology Registry, where the diagnosis of M-PM is based on pathological examination of the tissue specimen, and/or the Danish National Patient Registry, where the diagnosis is based on the clinician's reporting of an ICD-10 diagnosis. Thus, the registration of PMs may be reported insufficiently.
Finally, the statistical model applied in the present study does not restrict probabilities to the interval of 0-1. The c.i. of some baseline risk estimates included negative numbers, in which case the lower limit was set to zero. Furthermore, the prediction model presented here has not been validated. The aim of the study was to determine individual risk factors rather than to predict M-PM, and thus the model should not be used for prediction in future patients without external validation.