Meta‐analysis of prognostic factors for patients with colorectal peritoneal metastasis undergoing cytoreductive surgery and heated intraperitoneal chemotherapy

Background Up to 15 per cent of colorectal cancers present with peritoneal metastases (CPM). Cytoreductive surgery and heated intraperitoneal chemotherapy (CRS + HIPEC) aims to achieve macroscopic tumour resection combined with HIPEC to destroy microscopic disease. CRS + HIPEC is a major operation with significant morbidity and effects on quality of life (QoL). Improving patient selection is crucial to maximize patient outcomes while minimizing morbidity and mortality. The aim of this study was to identify prognostic factors for patients with CPM undergoing CRS + HIPEC. Methods A systematic search of MEDLINE, Embase and Cochrane Library electronic databases was performed using terms for colorectal cancer, peritoneal metastasis and CRS + HIPEC. Included studies focused on the impact of prognostic factors on overall survival following CRS + HIPEC in patients with CPM. Results Twenty‐four studies described 3128 patients. Obstruction or perforation of the primary tumour (hazard ratio (HR) 2·91, 95 per cent c.i. 1·5 to 5·65), extent of peritoneal metastasis as described by the Peritoneal Carcinomatosis Index (PCI) (per increase of 1 PCI point: HR 1·07, 1·02 to 1·12) and the completeness of cytoreduction (CC score above zero: HR 1·75, 1·18 to 2·59) were associated with reduced overall survival after CRS + HIPEC. Conclusion Primary tumour obstruction or perforation, PCI score and CC score are valuable prognostic factors in the selection of patients with CPM for CRS + HIPEC.


Introduction
Colorectal peritoneal metastasis (CPM) occurs in up to 15 per cent of patients with colorectal cancer 1 -3 . The prognosis of patients with CPM is poor: untreated median overall survival (OS) is just 6 months 4 . With systemic chemotherapy, median OS is improved to up to 20 months 5 -7 . The standard for CPM is cytoreductive surgery and heated intraperitoneal chemotherapy (CRS + HIPEC), which may improve median OS by 20-63 months 8 -16 . CRS + HIPEC is a long, high-cost operation associated with a protracted inpatient and high-dependency or intensive care unit stay, and an associated mortality rate of 1-12 per cent and morbidity rate of 7-63 per cent 11,14,16 -23 . Improving patient selection is therefore crucial to maximize patient outcomes whilst minimizing morbidity and mortality.
Variation in outcomes for CRS + HIPEC can be explained in part by patient selection, for example necessitating the ability to achieve complete cytoreduction at CRS, the exclusion of patients with extensive CPM as assessed by Sugarbaker's Peritoneal Carcinomatosis Index (PCI), and selection of patients with minimal co-morbidity and good performance status.
Several clinicopathological variables that impact on survival have been identified in the literature, includ-centres performing CRS + HIPEC worldwide, and in turn in their outcomes.
Studies reporting prognostic factors for CRS + HIPEC are cohort in design with small samples, and each examines numerous and varying prognostic factors on differing scales of measurement. No consensus exists as to which prognostic factors contraindicate CRS + HIPEC, or predict a good outcome. A comprehensive evidence synthesis is therefore called for to determine relevant prognostic factors.
The aim of this systematic review and meta-analysis was to analyse all prognostic factors affecting OS in patients with CPM undergoing CRS + HIPEC.

Methods
A comprehensive literature search was conducted in accordance with the PRISMA guidelines 24 . MEDLINE, Embase and Cochrane Library electronic databases, registers of clinical trials (ClinicalTrials.gov and the WHO International Clinical Trials Registry) and the Conference Proceedings Citation Index, Zetoc, were searched from inception to the present. The search strategy captured terms for colorectal cancer, peritoneal metastasis and CRS + HIPEC techniques, separated by the Boolean operator 'AND'. For an example search strategy, see Appendix S1 (supporting information). Searches were supplemented by a hand search of selected journals and the reference lists of all included studies.

Study selection
English-language articles were eligible for inclusion if they reported on the impact of prognostic factors on OS in patients with CPM undergoing CRS + HIPEC. Where multiple studies described the same cohort of patients, the largest and most complete data set was included. Review articles, case reports and case series of fewer than ten patients were excluded. Additional exclusion criteria included studies involving patients with a primary tumour other than colorectal cancer and studies in which a proportion of the cohort did not receive combined CRS + HIPEC.
After screening the titles and abstracts, articles fulfilling the eligibility criteria were identified and their full-text publications reviewed. Literature search and study selection were done independently by two researchers, and any disagreements were resolved by discussion with senior reviewers. After qualitative assessment, articles were screened to ensure they presented adequate statistical information to be included in the meta-analysis: hazard ratios (HRs) with confidence intervals or Kaplan-Meier curves with the number of events and patients at risk. When HRs, confidence intervals or P values were not provided directly, the methods of Tierney et al. 25 were used to estimate them indirectly from Kaplan-Meier curves, when presented in adequate detail with the numbers of events and patients at risk 26 .

Assessment of risk of bias
The quality and risk of bias of individual studies was assessed using the Quality in Prognosis Studies (QUIPS) tool 27 . This tool reviews each study according to six criteria: study participation, attrition, prognostic factor measurement, outcome measurement, confounding factors, and statistical analysis and reporting. Two authors scored all articles independently.

Data extraction
Data were extracted independently by two reviewers using a dedicated and piloted data extraction form. The number of patients, study design, patient demographics, tumour characteristics, use of adjuvant and neoadjuvant regimens, CRS + HIPEC techniques, survival and prognostic factors were recorded.
The unadjusted HR and its 95 per cent c.i. and P value were extracted. When adjusted HRs (with confidence intervals and P values) were reported, these were extracted along with the set of adjustment factors used. If HRs, confidence intervals or P values were not provided directly, the methods of Tierney and colleagues 25 were used to estimate them indirectly from Kaplan-Meier curves, when presented in adequate detail with the numbers of events and patients at risk 26 .
Prognostic factors reported on a continuous scale were extracted. If results were categorized into three or more categories, results for each comparison were extracted and, when clinically relevant, were grouped to form a binary comparison.

Prognostic factor selection
All prognostic factors described adequately and reported by two or more independent studies were included.

Statistical analysis
Owing to clinical and methodological heterogeneity, a random-effects meta-analysis was used (on the log(HR) scale) using the method of DerSimonian and Laird 28 . The combined effect size was described by the pooled HR, its confidence interval and P value, with P < 0⋅050 considered significant. For prognostic factors reported by more than two studies, a 95 per cent prediction interval (a measure of the variation in treatment effects) is presented 29 . Heterogeneity was also described by the I 2 statistic 30 . All analyses were performed in STATA ® version 15 (StataCorp, College Station, Texas, USA).

Results
The  (Table S2, supporting information), meta-analysis was stratified according to whether the HR was unadjusted or adjusted.

Quality of research
There was a low risk of bias from study participation. The moderate risk of bias due to study attrition reflects the poor reporting of loss to follow-up. There was a low risk of bias due to prognostic factor measurement. Prognostic factors were objective, clearly defined and clinically relevant. Reporting of all prognostic factors and treatment variations was incomplete in the majority of studies, resulting in a moderate risk of bias. The presentation of results and analytical strategy was sufficient in the majority of studies resulting in a low risk of bias (Table S3, supporting information).

Eastern Cooperative Oncology Group performance status
Five studies 18,31,43,52,53 reported on the influence of Eastern Cooperative Oncology Group (ECOG) status on OS, and three presented adequate data to be included in the meta-analysis. The pooled unadjusted HR for an ECOG score of at least 2 was 1⋅77 (95 per cent c.i. 0⋅85 to 3⋅67) (I 2 = 0 per cent) ( Fig. 2; Fig. S1, supporting information). These data provide no evidence that ECOG is a useful predictor of OS.

Adverse primary tumour features
Three studies 18,33,52 reported on the influence of adverse features of the primary tumour (obstruction or perforation) on OS; two presented adequate data to be included in the meta-analysis. The pooled adjusted HR for adverse features of the primary tumour was 2⋅91 (95 per cent c.i. 1⋅5 to 5⋅64) (I 2 = 0 per cent) ( Fig. 2; Fig. S1, supporting information). These data indicate that adverse primary features are a useful predictor of decreased OS (Table S5, supporting information).

Rectal or colonic primary
Eight studies 20,34,36,40,43,47,53,54 reported on the influence of a rectal or colonic primary on OS and six presented adequate data to be included in the meta-analysis. The pooled unadjusted HR for rectal primary (compared with colonic primary) was 1⋅48 (95 per cent c.i. 0⋅71 to 3⋅09) (I 2 = 73⋅8 per cent) ( Fig. 2; Fig. S1, supporting information). These data provide no evidence that a rectal primary is a useful predictor of OS.

Tumour differentiation
Two studies 34,40 reported on the effect of primary tumour differentiation on OS. The pooled unadjusted HR for well differentiated tumours was 0⋅90 (95 per cent c.i. 0⋅18 to 4⋅40) (I 2 = 72⋅4 per cent) ( Fig. 2; Fig. S1, supporting information). These data provide no evidence that primary tumour differentiation is a useful predictor of OS.

Timing of colorectal peritoneal metastasis (synchronous or metachronous)
Three studies 11,40,54 reported on the effect of the timing of CPM (synchronous or metachronous) on OS. Two presented adequate data to be included in the meta-analysis. The pooled unadjusted HR for synchronous CPM was 1⋅13 (95 per cent c.i. 0⋅56 to 2⋅28) (I 2 = 57⋅1 per cent) ( Fig. 2; Fig. S1, supporting information), indicating that the timing of CPM is not a useful predictor of OS.

Signet ring histology
Three studies 11,42,46 reported on the effect of signet ring histology on OS. Two included adequate data to be included in the meta-analysis. The pooled adjusted HR for signet ring histology was 1⋅65 (95 per cent c.i. 0⋅68 to 4⋅03) (I 2 = 0 per cent) ( Fig. 2; Fig. S1, supporting information), suggesting that signet ring histology is not a useful predictor of OS.

Mucinous histology
Four studies 11,43,46,49 reported on the influence of mucinous histology on OS. Three studies presented adequate data to be included in the meta-analysis. The pooled unadjusted HR for mucinous histology was 1⋅10 (95 per-cent c.i. 0⋅70 to 1⋅72) (I 2 = 0 per cent) ( Fig. 2; Fig.  S1, supporting information), suggesting that mucinous histology is not a useful predictor of OS.

Previous surgical score
Three studies 11,34,42 reported on the influence of the previous surgical score (PSS) on OS. The pooled unadjusted HR for a PSS of at least 2 was 1⋅23 (95 per cent c.i. 0⋅78 to 1⋅95) (I 2 = 0 per cent) ( Fig. 2; Fig. S1, supporting information). These data provide no evidence that PSS is a useful predictor of OS.

Postoperative morbidity
Three studies 40,43,50 reported on the association between postoperative morbidity and OS. The pooled unadjusted HR for a Clavien-Dindo complication of grade III or above was 1⋅56 (95 per cent c.i. 0⋅74 to 3⋅31) (I 2 = 48⋅7 per cent) (Fig. 2; Fig. S1, supporting information). These data provide no evidence that postoperative morbidity is a useful predictor of OS.

Neoadjuvant and adjuvant chemotherapy
Neoadjuvant and adjuvant treatment of the primary tumour was reported poorly (Table S6, supporting information). No study reported response to treatment, or whether it was completed as planned.
Two studies 11,43 reported on the effect of neoadjuvant chemotherapy before CRS + HIPEC on OS. The pooled unadjusted HR for the use of neoadjuvant chemotherapy was 1⋅00 (95 per cent c.i. 0⋅63 to 1⋅58) (I 2 = 0 per cent) ( Fig. 2; Fig. S1, supporting information). These data provide no evidence that neoadjuvant chemotherapy is a useful predictor of OS.

Discussion
This systematic review and meta-analysis examined the effect of prognostic factors on OS following CRS + HIPEC for CPM. Emergency presentation with obstruction or perforation of the primary tumour as well as the extent of CPM and the completeness of resection, as described by the PCI and the CC score respectively, were the only significant prognostic factors.
An emergency presentation of the primary tumour with obstruction or perforation was predictive of reduced OS (for both synchronous and metachronous CPM). A number of factors may contribute to this. In the primary setting, colorectal cancers presenting with obstruction or perforation are associated with decreased cancer-specific survival and increased postoperative mortality 56 . Obstructed or perforated colorectal cancer is, by definition, advanced in stage and increases the risk of metastasis; additionally, emergency presentation limits the possibility of neoadjuvant treatment and may delay adjuvant treatment owing to postoperative morbidity. The extent of peritoneal metastasis as described by the PCI was predictive of reduced OS as a continuous variable and when the PCI score was 12 or above. A complete cytoreduction (CC0) was predictive of improved OS. Improving patient selection is therefore reliant on the ability to predict accurately the extent of peritoneal metastasis and the ability to resect it completely. Specialist radiologists have demonstrated good concordance between radiological and surgical PCI estimations, particularly when combining modalities; however, these tend to be most accurate in patients with high PCI scores 57,58 . In some centres this is used in combination with diagnostic laparoscopy before CRS + HIPEC. This is feasible in the majority of patients and may help to reduce the laparotomy rate in patients for whom CRS + HIPEC may not be possible 59 .
Included patients were relatively young at 54 (range 45-69) years compared with the incident age of colorectal cancer (80-90 years) 60 In addition, performance status was not reported by the majority of studies in the review, and limited to an ECOG score of less than 2 by a further nine. Within these limits, no other patient or tumour factor was predictive of OS.
Details of neoadjuvant and adjuvant treatments were reported poorly in the included studies. Within these limits, the use of neoadjuvant or adjuvant chemotherapy was not predictive of OS after CRS + HIPEC. One meta-analysis, by Kwakman and colleagues 61 from 2016, examined the effect of clinicopathological variables only on OS following CRS + HIPEC. Significant prognostic factors identified in the present review (adverse features of the primary tumour, PCI, CC score) were not comparable with those from Kwakman et al. 61 as they were not examined. In contrast to the present study, Kwakman and co-workers 61 found performance status, the presence of lymph node or hepatic metastasis, tumour differentiation, signet ring histology, a rectal primary and the use of neoadjuvant chemotherapy to be predictive prognostic factors. A number of differences may explain this variation in findings: the exclusion of 73 papers described as unavailable in full text may introduce a potential selection bias 61 . In addition, a number of studies included by Kwakman and colleagues were excluded in the present study for the following reasons: presentation of inadequate data to estimate the HR accurately 13,52,54,62 the inclusion of mixed primary tumours 48,63,64 , and the inclusion of patients having repeat CRS + HIPEC procedures. Finally, the meta-analysis 61 combined all studies regardless of the presentation of unadjusted or adjusted HR.
The strengths of the present study include the comprehensive and systematic literature search including 3128 patients with CPM undergoing CRS + HIPEC, all potential prognostic factors were included, and a consistent association was found between predictive prognostic factors across different studies. The low heterogeneity associated with these factors adds to the strength and generalizability of the findings. The application of strict inclusion criteria limits the potential impact of factors such as mixed primary tumour origin. The present analysis takes into account the adjustment factors used in primary studies to ensure meta-analysis of time to event data was performed only when data were comparable.
A number of limitations must, however, be acknowledged. This review is limited by the quality of primary studies and the heterogeneity of the population. Prognostic factor systematic reviews, by their nature, represent one of the most difficult categories due to their retrospective and observational nature. As CRS + HIPEC is performed at tertiary centres, data concerning the primary tumour and its treatment may not have been captured or reported fully. Additionally, the statistical analysis and presentation of data necessary for accurate meta-analysis varied widely across primary studies, which may introduce a degree of inclusion bias.
There was a lack of molecular and genetic data concerning patients with CPM undergoing CRS + HIPEC in comparison with studies of primary colorectal cancer, and this is an area for future research. Discordance between primary genetic mutations and those in peritoneal metastases may identify novel therapeutic targets and prognostic markers in this metastatic group. Recent research by Schneider and colleagues 65 found that RAS/RAF mutations impair survival after CRS/HIPEC, although this is the only study that has considered this prognostic factor and validation is required. Further large-scale research is needed to account for these factors; this would require collaboration between CRS + HIPEC centres, standardization of the analysis and presentation of prognostic factor time to event data.