Using Artificial Intelligence to Find the Optimal Margin Width in Hepatectomy for Colorectal Cancer Liver Metastases

Importance

In patients with resectable colorectal cancer liver metastases (CRLM), the choice of surgical technique and resection margin are the only variables that are under the surgeon's direct control and may influence oncologic outcomes. There is currently no consensus on the optimal margin width.

Objective

To determine the optimal margin width in CRLM by using artificial intelligence-based techniques developed by the Massachusetts Institute of Technology and to assess whether optimal margin width should be individualized based on patient characteristics.

Design, Setting, and Participants

The internal cohort of the study included patients who underwent curative-intent surgery for KRAS-variant CRLM between January 1, 2000, and December 31, 2017, at Johns Hopkins Hospital, Baltimore, Maryland, Memorial Sloan Kettering Cancer Center, New York, New York, and Charité-University of Berlin, Berlin, Germany. Patients from institutions in France, Norway, the US, Austria, Argentina, and Japan were retrospectively identified from institutional databases and formed the external cohort of the study. Data were analyzed from April 15, 2019, to November 11, 2021.

Exposures

Hepatectomy.

Main Outcomes and Measures

Patients with KRAS-variant CRLM who underwent surgery between 2000 and 2017 at 3 tertiary centers formed the internal cohort (training and testing). In the training cohort, an artificial intelligence-based technique called optimal policy trees (OPTs) was used by building on random forest (RF) predictive models to infer the margin width associated with the maximal decrease in death probability for a given patient (ie, optimal margin width). The RF component was validated by calculating its area under the curve (AUC) in the testing cohort, whereas the OPT component was validated by a game theory-based approach called Shapley additive explanations (SHAP). Patients from international institutions formed an external validation cohort, and a new RF model was trained to externally validate the OPT-based optimal margin values.

Results

This cohort study included a total of 1843 patients (internal cohort, 965; external cohort, 878). The internal cohort included 386 patients (median [IQR] age, 58.3 [49.0-68.7] years; 200 men [51.8%]) with KRAS-variant tumors. The AUC of the RF counterfactual model was 0.76 in both the internal training and testing cohorts, which is the highest ever reported. The recommended optimal margin widths for patient subgroups A, B, C, and D were 6, 7, 12, and 7 mm, respectively. The SHAP analysis largely confirmed this by suggesting 6 to 7 mm for subgroup A, 7 mm for subgroup B, 7 to 8 mm for subgroup C, and 7 mm for subgroup D. The external cohort included 375 patients (median [IQR] age, 61.0 [53.0-70.0] years; 218 men [58.1%]) with KRAS-variant tumors. The new RF model had an AUC of 0.78, which allowed for a reliable external validation of the OPT-based optimal margin. The external validation was successful as it confirmed the association of the optimal margin width of 7 mm with a considerable prolongation of survival in the external cohort.

Conclusions and Relevance

This cohort study used artificial intelligence-based methodologies to provide a possible resolution to the long-standing debate on optimal margin width in CRLM.

Mutant KRAS as a prognostic biomarker after hepatectomy for rectal cancer metastases: Does the primary disease site matter?

BACKGROUND

The prognostic implication of mutant KRAS (mKRAS) among patients with primary disease in the rectum remains unknown.

METHODS

From 2000 to 2018, patients undergoing hepatectomy for colorectal liver metastases at 10 collaborating international institutions with documented KRAS status were surveyed.

RESULTS

A total of 834 (65.8%) patients with primary colon cancer and 434 (34.2%) patients with primary rectal cancer were included. In patients with primary colon cancer, mKRAS served as a reliable prognostic biomarker of poor overall survival (OS) (hazard ratio [HR]: 1.58, 95% CI 1.28-1.95) in the multivariable analysis. Although a trend towards significance was noted, mKRAS was not found to be an independent predictor of OS in patients with primary rectal tumors (HR 1.34, 95% CI 0.98-1.80). For colon cancer, the specific codon impacted in mKRAS appears to reflect underlying disease biology and oncologic outcomes, with codon 13 being associated with particularly poor OS in patients with left-sided tumors (codon 12, HR 1.56, 95% CI 1.22-1.99; codon 13, HR 2.10 95% CI 1.43-3.08;). Stratifying the rectal patient population by codon mutation did not confer prognostic significance following hepatectomy.

CONCLUSIONS

While the left-sided colonic disease is frequently grouped with rectal disease, our analysis suggests that there exist fundamental biologic differences that drive disparate outcomes. Although there was a trend toward significance of KRAS mutations for patients with primary rectal cancers, it failed to achieve statistical significance.

The Prognostic Impact of Primary Tumor Site Differs According to the KRAS Mutational Status: A Study By the International Genetic Consortium for Colorectal Liver Metastasis

OBJECTIVE

To examine the prognostic impact of tumor laterality in colon cancer liver metastases (CLM) after stratifying by Kirsten rat sarcoma 2 viral oncogene homolog (KRAS) mutational status.

BACKGROUND

Although some studies have demonstrated that patients with CLM from a right sided (RS) primary cancer fare worse, others have found equivocal outcomes of patients with CLM with RS versus left-sided (LS) primary tumors. Importantly, recent evidence from unresectable metastatic CRC suggests that tumor laterality impacts prognosis only in those with wild-type tumors.

METHODS

Patients with rectal or transverse colon tumors and those with unknown KRAS mutational status were excluded from analysis. The prognostic impact of RS versus LS primary CRC was determined after stratifying by KRAS mutational status.

RESULTS

277 patients had a RS (38.6%) and 441 (61.4%) had a LS tumor. Approximately one-third of tumors (28.1%) harbored KRAS mutations. In the entire cohort, RS was associated with worse 5-year overall survival (OS) compared with LS (39.4% vs 50.8%, P = 0.03) and remained significantly associated with worse OS in the multivariable analysis (hazard ratio 1.45, P = 0.04). In wild-type patients, a worse 5-year OS associated with a RS tumor was evident in univariable analysis (43.7% vs 55.5%, P = 0.02) and persisted in multivariable analysis (hazard ratio 1.49, P = 0.01). In contrast, among patients with KRAS mutated tumors, tumor laterality had no impact on 5-year OS, even in the univariable analysis (32.8% vs 34.0%, P = 0.38).

CONCLUSIONS

This study demonstrated, for the first time, that the prognostic impact of primary tumor side differs according to KRAS mutational status. RS tumors were associated with worse survival only in patients with wild-type tumors.

Performance of two prognostic scores that incorporate genetic information to predict long-term outcomes following resection of colorectal cancer liver metastases: An external validation of the MD Anderson and JHH-MSK scores

INTRODUCTION

Two novel clinical risk scores (CRS) that incorporate KRAS mutation status were developed: modified CRS (mCRS) and GAME score. However, they have not been tested in large national and international cohorts. The aim of this study was to validate the prognostic discrimination utility and determine the clinical usefulness of the two novel CRS.

METHODS

Patients undergoing hepatectomy for CRLM (2000-2018) in 10 centers were included. The discriminatory abilities of mCRS, GAME, and Fong CRS were evaluated using Harrell's C-index and Akaike's Information Criterion.

RESULTS

In the entire cohort, the C-index of the GAME score (0.61) was significantly higher than those of Fong score (0.57) and mCRS (0.54), while the C-Index of mCRS was significantly lower than that of Fong score. When we compared the models in the various geographical regions, the C-index of GAME score was significantly higher than that of mCRS in North America, Europe, and South America. The AIC of Fong score, mCRS, and GAME score were 14 405, 14 447, and 14 319, respectively.

CONCLUSION

In conclusion, using the largest and most heterogenous population of CRLM patients with known KRAS status, this independent, external validation demonstrated that the GAME score outperforms both the traditional Fong score and mCRS.

The optimal cut-off values for tumor size, number of lesions, and CEA levels in patients with surgically treated colorectal cancer liver metastases: An international, multi-institutional study

BACKGROUND AND OBJECTIVES

Despite the long-standing consensus on the importance of tumor size, tumor number and carcinoembryonic antigen (CEA) levels as predictors of long-term outcomes among patients with colorectal liver metastases (CRLM), optimal prognostic cut-offs for these variables have not been established.

METHODS

Patients who underwent curative-intent resection of CRLM and had available data on at least one of the three variables of interest above were selected from a multi-institutional dataset of patients with known KRAS mutational status. The resulting cohort was randomly split into training and testing datasets and recursive partitioning analysis was employed to determine optimal cut-offs. The concordance probability estimates (CPEs) for these optimal cut offs were calculated and compared to CPEs for the most widely used cut-offs in the surgical literature.

RESULTS

A total of 1643 patients who met eligibility criteria were identified. Following recursive partitioning analysis in the training dataset, the following cut-offs were identified: 2.95 cm for tumor size, 1.5 for tumor number and 6.15 ng/ml for CEA levels. In the entire dataset, the calculated CPEs for the new tumor size (0.52), tumor number (0.56) and CEA (0.53) cut offs exceeded CPEs for other commonly employed cut-offs.

CONCLUSION

The current study was able to identify optimal cut-offs for the three most commonly employed prognostic factors in CRLM. While the per variable gains in discriminatory power are modest, these novel cut-offs may help produce appreciable increases in prognostic performance when combined in the context of future risk scores.

Evaluation of the stage classification of anal cancer by the TNM 8th version versus the TNM 7th version

Background: The UICC TNM 7th edition introduced stage groups for anal cancer which in 2019 has not yet come into general use. The new TNM 8th edition from 2016 defines 7 sub-stages. Background data for these changes are lacking. We aimed to investigate whether the new classification for anal cancer reliably predict the prognosis in the different stages.Patients and methods: The Nordic Anal Cancer Group (NOAC) conducted a large retrospective study of all anal cancers in Norway, Sweden and most of Denmark in 2000-2007. From the Nordic cohort 1151 anal cancer patients with follow-up data were classified by the TNM 4th edition which has identical T, N and M definitions as the TNM 7th edition, and therefore also can be classified by the TNM 7th stage groups. We used the Nordic cohort to translate the T, N and M stages into the TNM 8th stages and sub-stages. Overall survival for each stage was assessed.Results: Although the summary stage groups for TNM 8th edition discriminates patients with different prognosis reasonably well, the analyses of the seven sub-stages show overlapping overall survival: HR for stage IIA 1.30 (95%CI 0.80-2.12) is not significantly different from stage I (p = .30) and HR for stage IIB 2.35 (95%CI 1.40-3.95) and IIIA 2.48 (95%CI 1.43-4.31) are also similar as were HRs for stage IIIB 3.41 (95%CI 1.99-5.85) and IIIC 3.22 (95%CI 1.99-5.20). Similar overlapping was shown for local recurrence and distant spread.Conclusion: The results for the sub-stages calls for a revision of the staging system. We propose a modification of the TNM 8th edition for staging of anal cancer into four stages based on the T, N and M definitions of the TNM 8th classification.

Association of BRAF Mutations With Survival and Recurrence in Surgically Treated Patients With Metastatic Colorectal Liver Cancer

Importance

BRAF mutations are reportedly associated with aggressive tumor biology. However, in contrast with primary colorectal cancer, the association of V600E and non-V600E BRAF mutations with survival and recurrence after resection of colorectal liver metastases (CRLM) has not been well studied.

Objective

To investigate the prognostic association of BRAF mutations with survival and recurrence independently and compared with other prognostic determinants, such as KRAS mutations.

Design, Setting, and Participants

In this cohort study, all patients who underwent resection for CRLM with curative intent from January 1, 2000, through December 31, 2016, at the institutions participating in the International Genetic Consortium for Colorectal Liver Metastasis and had data on BRAF and KRAS mutational status were retrospectively identified. Multivariate Cox proportional hazards regression models were used to assess long-term outcomes.

Interventions

Hepatectomy in patients with CRLM.

Main Outcomes and Measures

The association of V600E and non-V600E BRAF mutations with disease-free survival (DFS) and overall survival (OS).

Results

Of 853 patients who met inclusion criteria (510 men [59.8%] and 343 women [40.2%]; mean [SD] age, 60.2 [12.4] years), 849 were included in the study analyses. Forty-three (5.1%) had a mutated (mut) BRAF/wild-type (wt) KRAS (V600E and non-V600E) genotype; 480 (56.5%), a wtBRAF/wtKRAS genotype; and 326 (38.4%), a wtBRAF/mutKRAS genotype. Compared with the wtBRAF/wtKRAS genotype group, patients with a mutBRAF/wtKRAS genotype more frequently were female (27 [62.8%] vs 169 [35.2%]) and 65 years or older (22 [51.2%] vs 176 [36.9%]), had right-sided primary tumors (27 [62.8%] vs 83 [17.4%]), and presented with a metachronous liver metastasis (28 [64.3%] vs 229 [46.8%]). On multivariable analysis, V600E but not non-V600E BRAF mutation was associated with worse OS (hazard ratio [HR], 2.76; 95% CI, 1.74-4.37; P < .001) and DFS (HR, 2.04; 95% CI, 1.30-3.20; P = .002). The V600E BRAF mutation had a stronger association with OS and DFS than the KRAS mutations (β for OS, 10.15 vs 2.94; β for DFS, 7.14 vs 2.27).

Conclusions and Relevance

The presence of the V600E BRAF mutation was associated with worse prognosis and increased risk of recurrence. The V600E mutation was not only a stronger prognostic factor than KRAS but also was the strongest prognostic determinant in the overall cohort.

Double KRAS and BRAF Mutations in Surgically Treated Colorectal Cancer Liver Metastases: An International, Multi-institutional Case Series

BACKGROUND

While previously believed to be mutually exclusive, concomitant mutation of Kirsten rat sarcoma viral oncogene homolog (KRAS)- and V-raf murine sarcoma b-viral oncogene homolog B1 (BRAF)-mutated colorectal carcinoma (CRC), has been described in rare instances and been associated with advanced-stage disease. The present case series is the first to report on the implications of concurrent KRAS/BRAF mutations among surgically treated patients, and the largest set of patients with surgically treated colorectal liver metastasis (CRLM) and data on KRAS/BRAF mutational status thus far described.

CASE SERIES

We present cases from an international, multi-institutional cohort of patients that underwent hepatic resection for CRLM between 2000-2015 at seven tertiary centers. The incidence of KRAS/BRAF mutation in patients with CRLM was 0.5% (4/820). Of these cases, patient 1 (T2N1 primary, G13D/V600E), patient 2 (T3N1 primary, G12V/V600E) and patient 3 (T4N2 primary, G13D/D594N) succumbed to their disease within 485, 236 and 79 days respectively, post-hepatic resection. Patient 4 (T4 primary, G12S/G469S) was alive 416 days after hepatic resection.

CONCLUSION

The present case series suggests that the incidence of concomitant KRAS/BRAF mutations in surgical cohorts may be higher than previously hypothesized, and associated with more variable survival outcomes than expected.

Population-based study on resection rates and survival in patients with colorectal liver metastasis in Norway

Background

Detailed knowledge about the proportion of patients with colorectal liver metastases (CLM) undergoing resection is sparse. The aim of this study was to analyse cumulative resection rates and survival in patients with CLM.

Methods

For this population-based study of patients developing CLM during 2011–2013, data were extracted from the Norwegian Patient Registry and the Cancer Registry of Norway.

Results

A total of 2960 patients had CLM; their median overall survival was 10·9 months. Liver resection was performed in 538 patients. The cumulative resection rate was 20·0 per cent. The cumulative resection rate was 23·3 per cent in patients aged less than 40 years, 31·1 per cent in patients aged 40–59 years, 24·7 per cent in those aged 60–74 years, 17·9 per cent in those aged 75–79 years and 4·7 per cent in patients aged 80 years or more (P &lt; 0·001). In multivariable analysis, resection rate was associated with age, extrahepatic metastases, disease-free interval and geographical region. Overall survival after diagnosis of CLM was affected by liver resection (hazard ratio (HR) 0·54, 95 per cent c.i. 0·34 to 0·86), rectal cancer (HR 0·82, 0·74 to 0·90), metachronous disease (HR 0·66, 0·60 to 0·74), increasing age (HR 1·32, 1·28 to 1·37), region, and extrahepatic metastases (HR 1·90, 1·74 to 2·07). Three- and 4-year overall survival rates after hepatectomy were 73·2 and 54·8 per cent respectively.

Conclusion

The cumulative resection rate in patients with CLM in Norway between 2011 and 2013 was 20 per cent. Resection rates varied across geographical regions, and with patient and disease characteristics.

Intra-patient Inter-metastatic Genetic Heterogeneity in Colorectal Cancer as a Key Determinant of Survival after Curative Liver Resection

Chromosomal instability is a well-defined hallmark of tumor aggressiveness and metastatic progression in colorectal cancer. The magnitude of genetic heterogeneity among distinct liver metastases from the same patient at the copy number level, as well as its relationship with chemotherapy exposure and patient outcome, remains unknown. We performed high-resolution DNA copy number analyses of 134 liver metastatic deposits from 45 colorectal cancer patients to assess: (i) intra-patient inter-metastatic genetic heterogeneity using a heterogeneity score based on pair-wise genetic distances among tumor deposits; and (ii) genomic complexity, defined as the proportion of the genome harboring aberrant DNA copy numbers. Results were analyzed in relation to the patients' clinical course; previous chemotherapy exposure and outcome after surgical resection of liver metastases. We observed substantial variation in the level of intra-patient inter-metastatic heterogeneity. Heterogeneity was not associated with the number of metastatic lesions or their genomic complexity. In metachronous disease, heterogeneity was higher in patients previously exposed to chemotherapy. Importantly, intra-patient inter-metastatic heterogeneity was a strong prognostic determinant, stronger than known clinicopathological prognostic parameters. Patients with a low level of heterogeneity (below the median level) had a three-year progression-free and overall survival rate of 23% and 66% respectively, versus 5% and 18% for patients with a high level (hazard ratio0.4, 95% confidence interval 0.2-0.8, P = 0.01; and hazard ratio0.3,95% confidence interval 0.1-0.7, P = 0.007). A low patient-wise level of genomic complexity (below 25%) was also a favorable prognostic factor; however, the prognostic association of intra-patient heterogeneity was independent of genomic complexity in multivariable analyses. In conclusion, intra-patient inter-metastatic genetic heterogeneity is a pronounced feature of metastatic colorectal cancer, and the strong prognostic association reinforces its clinical relevance and places it as a key feature to be explored in future patient cohorts.

Impact of KRAS, BRAF, PIK3CA, TP53 status and intraindividual mutation heterogeneity on outcome after liver resection for colorectal cancer metastases

We determined prognostic impact of KRAS, BRAF, PIK3CA and TP53 mutation status and mutation heterogeneity among 164 colorectal cancer (CRC) patients undergoing liver resections for metastatic disease. Mutation status was determined by Sanger sequencing of a total of 422 metastatic deposits. In univariate analysis, KRAS (33.5%), BRAF (6.1%) and PIK3CA (13.4%) mutations each predicted reduced median time to relapse (TTR) (7 vs. 22, 3 vs. 16 and 4 vs. 17 months; p < 0.001, 0.002 and 0.023, respectively). KRAS and BRAF mutations also predicted a reduced median disease‐specific survival (DSS) (29 vs. 51 and 16 vs. 49 months; p <0.001 and 0.008, respectively). No effect of TP53 (60.4%) mutation status was observed. Postoperative, but not preoperative chemotherapy improved both TTR and DSS (p < 0.001 for both) with no interaction with gene mutation status. Among 94 patients harboring two or more metastatic deposits, 13 revealed mutation heterogeneity across metastatic deposits for at least one gene. Mutation heterogeneity predicted reduced median DSS compared to homogeneous mutations (18 vs. 37 months; p = 0.011 for all genes; 16 vs. 26 months; p < 0.001 analyzing BRAF or KRAS mutations separately). In multivariate analyses, KRAS or BRAF mutations consistently predicted poor TRR and DSS. Mutation heterogeneity robustly predicted DSS but not TTR, while postoperative chemotherapy improved both TTR and DSS. Our findings indicate that BRAF and KRAS mutations as well as mutation heterogeneity predict poor outcome in CRC patients subsequent to liver resections and might help guide treatment decisions.

What's new?

Preliminary evidence suggests that poor outcome after liver resection in metastatic colorectal cancer (CRC) is predicted by mutations in KRAS and BRAF and by intra‐individual heterogeneity involving copy number alterations that vary from one metastatic lesion to the next. Little is known, however, about the clinical implications of intra‐individual mutation heterogeneity in CRC. Here, in a comparison of KRAS and BRAF wild‐type status, mutational homogeneity, and mutational heterogeneity, mutation heterogeneity was found to be the strongest predict or of reduced disease‐specific survival following liver resection in metastatic CRC. Knowledge of intra‐individual mutation heterogeneity in KRAS and BRAF in CRC could facilitate therapeutic decisions.

Predictive factors for time to recurrence, treatment and post-recurrence survival in patients with initially resected colorectal liver metastases

BACKGROUND

Despite progress in resection for colorectal liver metastases (CLM), the majority of patients experience recurrence. We aimed to evaluate factors influencing time to recurrence (TTR), treatment and post-recurrence survival (PRS) related to site of recurrence.

METHODS

This is a retrospective population-based cohort study (1998-2012) of consecutive patients without extrahepatic disease treated with resection for CLM in a referral centre.

RESULTS

A total of 311 patients underwent resection for CLM. After a median follow-up of 4.2 years (range 1.2-15.2), 209 (67.4 %) patients developed recurrence, hepatic 90, extrahepatic 59 and both 60. Median TTR was 14.0 months, and 5-year recurrence-free status was 25.7 %. Five- and 10-year overall survival (OS) was 38.8 and 22.0 %, respectively. Median OS was 45 months. A multivariate analysis displayed synchronous disease (hazard ratio (HR) 1.50), American Society of Anaesthesiologists (ASA) score (HR 1.40), increasing number (HR 1.24) and size of metastases (HR 1.08) to shorten TTR (all p < 0.05). Perioperative chemotherapy (n = 59) increased overall TTR (HR 0.63) and overall survival (OS; HR 0.55). Hepatic TTR was correlated to synchronous disease (HR 2.07), number of lesions (HR 1.20), R1 resection (HR 2.00) and ASA score (HR 1.69), whereas extrahepatic TTR was correlated to N stage of the primary (HR 1.79), number (HR 1.27) and size of metastases (HR 1.16). Single-site recurrence was most common (135 of 209, 64.5 %), while 58 patients had double- and 16 triple-site relapses. Median PRS was 24.3 months. There was a difference in median PRS (months) according to site of relapse: liver 30.5, lung 32.3, abdominal 22.0, liver and lung 14.3, others 14.8 (p = 0.002). Repeated liver resections were performed in n = 57 patients resulting in 40.6 months median OS and 36.8 % 5-year OS.

CONCLUSIONS

An adverse overall TTR was correlated to number and size of metastases, ASA score and synchronous disease. Perioperative chemotherapy increased TTR and OS after surgery for CLM. Patients with solitary post-resection relapse in the liver or lungs had the potential for longevity due to multimodal treatment.