Development and validation of a mutation-annotated prognostic score for intrahepatic cholangiocarcinoma after resection: a retrospective cohort study

Abstract

Background:

The value of existing prognostic models for intrahepatic cholangiocarcinoma is limited. The inclusion of prognostic gene mutations would enhance the predictive efficacy.

Methods:

In the screening cohorts, univariable Cox regression analysis was applied to investigate the effect of individual mutant genes on overall survival (OS). In the training set, multivariable analysis was performed to evaluate the independent prognostic roles of the clinicopathological and mutational parameters, and a prognostic model was constructed. Internal and external validations were conducted to evaluate the performance of this model.

Results:

Among the recurrent mutations, only TP53 and KRAS ^G12 were significantly associated with OS across all three screening cohorts. In the training cohort, TP53 and KRAS ^G12 mutations in combination with seven other clinical parameters (tumor size, tumor number, vascular invasion, lymph node metastasis, adjacent invasion, CA19-9, and CEA), were independent prognostic factors for OS. A mutation-annotated prognostic score (MAPS) was established based on the nine prognosticators. The C-indices of MAPS (0.782 and 0.731 in the internal and external validation cohorts, respectively) were statistically higher than those of other existing models (P<0.05). Furthermore, the MAPS model also demonstrated significant value in predicting the possible benefits of upfront surgery and adjuvant therapy.

Conclusions:

The MAPS model demonstrated good performance in predicting the OS of intrahepatic cholangiocarcinoma patients. It may also help predict the possible benefits of upfront surgery and adjuvant therapy.

Introduction

Highlights

• Among the recurrent mutations in intrahepatic cholangiocarcinomas, only TP53 and KRAS ^G12 were significantly associated with overall survival in both Eastern and Western populations.

• A mutation-annotated prognostic score performed better than existing models in predicting postoperative survival in intrahepatic cholangiocarcinomas.

• The mutation-annotated prognostic score model also demonstrated significant value in predicting the possible benefits of upfront surgery and adjuvant therapy.

Intrahepatic cholangiocarcinoma (ICC) is the second most common primary liver cancer with increasing incidence and mortality worldwide^1,2. In recent years, ICC has been recognized as a unique entity, which is distinct from the other hepatobiliary malignancies with regard to clinical features and biological behaviors. However, the complicated etiological factors and various cellular origins contribute to its great heterogeneity among Eastern and Western populations^1–3. Hepatectomy remains the mainstay of curative intent treatment⁴. However, the frequent occurrence of intrahepatic recurrence and distant metastasis still constrains their long-term survival, with the 5-year survival rate ranging from 25 to 40% after liver resection^4,5. Precise prognostication for personalized patient management is therefore urgently needed.

The TNM staging system was the earliest and most widely applied system in the clinical practice of solid malignancies, including ICC. The first specific staging system for ICC (Okabayashi staging) was proposed in 2001⁶. Since then, many staging systems for ICC, such as AJCC^7,8, LCSGJ^9,10, JSHBPS¹¹, and others^12–17 have been proposed. In addition, several scoring systems have been developed to predict survival and assist in clinical decision-making. Traditional scoring systems, including the FUDAN and MEGNA score and others^18–21, assign one point to each independent prognostic factor and calculate the total score for risk evaluation. The nomogram, a model formulated according to the results of multivariable regression, has demonstrated higher prediction efficacy than conventional prognostic models^22,23. In general, these models are mainly based on clinical and pathological variables, including age, cirrhosis, vascular invasion, resection margin, lymph node status, number and maximum size of tumor, CA19-9, CEA, and neutrophil-to-lymphocyte ratio^6–23. Although these prognostic models are somewhat useful, their capability of prognostic discrimination remains limited in external validations, with the C-indices varying from 0.599 to 0.668^24,25. A possible reason is the great heterogeneity of ICC²⁶. Thus, including additional biomarkers, especially those reflecting the biological features of ICC, would further improve the prediction efficacy.

In recent years, multigene signatures based on high-throughput technologies, such as global gene expression profiling and DNA methylation analyses, have been developed to provide additional prognostic information in various cancer types^27–30. These genetic or epigenetic-based multigene signatures, reflecting intrinsic tumor biology, have also demonstrated powerful capability to further optimize conventional prognostic models. Several gene expression signatures have been constructed for the prediction of postoperative survival or recurrence in ICC^31–34. However, the requirement of complicated technique platforms and fresh-frozen tissues for high-throughput gene expression or methylation detection hinders their routine application in clinical practice. By contrast, gene mutation detection by either next-generation sequencing (NGS) or traditional Sanger sequencing has been widely applied in clinical diagnosis and treatment. In recent years, mutational information has been successfully used in the clinical prognostication of various malignancies^35–45.

As multiple targetable genetic alterations (IDH1 mutation, FGFR2 fusion) have been identified in ICC, clinical guidelines have recommended the routine application of NGS in advanced cholangiocarcinoma⁴⁶. In addition to mutations with therapeutic significance, several mutations have also been reported to influence survival, such as TP53, KRAS, IDH1/2, ATM, and ARID2 ^47–53. However, this evidence was mostly derived from studies with relatively small sample sizes or single center cohort, and has not been comprehensively validated.

In the present study, by systematically investigating large cohorts of ICC patients, we intended to identify the survival-related mutations in ICCs, and further establish a mutation-annotated prognostic score (MAPS) system by integrating genetic and clinicopathological characteristics.

Methods

Patient cohorts and study design

This study followed the REporting recommendations for tumor MARKer prognostic studies (REMARK) guidelines (Supplemental Digital Content 3, http://links.lww.com/JS9/A865)⁵⁴. In this study, we adopted a stepwise strategy with a screening, training, and validation design. This study was conducted in accordance with the Declaration of Helsinki.

In the screening phase, 714 patients with localized ICCs who underwent surgical resection from three independent cohorts (Eastern, Western, and Mixed cohort) were enrolled to identify those survival-related mutations. The Eastern cohort included 407 cases from five countries of Eastern or Southeastern Asian with high ICC incidences (China, n=183; Japan, n=183; Singapore, n=17; Thailand, n=14; South Korea, n=10). The Western cohort included 209 surgically resected ICCs from two Western countries (USA, n=187; Netherlands, n=22). The Mixed cohort included 98 ICCs from the ICGC project including both Western and Eastern populations (Thailand, n=58; France, n=15; Romania, n=12; Singapore, n=11; South Korea, n=2). All patients were pathologically diagnosed as ICCs without extrahepatic metastasis. The individual information of clinical outcome [overall survival (OS)] and clinicopathological parameters were acquired from the original publications^{47–53,55–58}. Those patients whose follow-up data was missed, or postoperative OS was less than 1 month were excluded. The specific information of the enrolled patients were summarized in Supplementary Tables 1 and 2 (Supplemental Digital Content 1, http://links.lww.com/JS9/A863).

A training cohort was recruited to evaluate the independent prognostic roles of clinicopathological, laboratory, and genetic parameters and to construct a prognostic model. This cohort included 227 patients with localized ICCs who underwent hepatectomy from February 2014 to February 2019 at the authors’ institute. The inclusion criteria included: no history of other malignancies; pathologically diagnosed as ICC; curative resection of liver tumors. Patients who underwent transplantation, biopsy or palliative resection, patients with metastatic disease, or those with postoperative survival less than 1 month were excluded. The detailed clinicopathological characteristics of the training cohort was shown in Table 1.

Table 1.

Demographics and clinicopathologic characteristics of patients with ICC from the training, internal validation, and external validation cohorts.

	Training cohort	Internal validation cohort	External validation cohort
Variables	N=227	N=95	N=230
Age
Median (range)	58 (28–84)	62 (39–86)	62 (27–86)
≥60 years	108 (47.6%)	58 (61.1%)	134 (58.3%)
<60 years	119 (52.4%)	37 (38.9%)	96 (41.7%)
Sex
Male	137 (60.4%)	63 (66.3%)	131 (57.0%)
Female	90 (39.6%)	32 (33.7%)	99 (43.0%)
HBV/HCV
HBV	82 (36.1%)	21 (22.1%)	64 (27.8%)
HCV	1 (0.4%)	2 (2.1%)	NR
HBV+HCV	2 (0.9%)	0 (0.0%)	NR
NBNC	142 (62.6%)	72 (75.8%)	166 (72.2%)
Hepatolithiasis
Yes	20 (8.8%)	13 (13.7%)	8 (3.5%)
No	207 (91.2%)	82 (86.3%)	222 (96.5%)
Liver fluke
Yes	1 (0.4%)	0 (0.0%)	15 (6.5%)
No	226 (99.6%)	95 (100.0%)	215 (93.5%)
CA19-9 (U/ml)
Median (range)	47.2 (0.6–10000)	76.8 (0.6–6810)	46.5 (0.6–10000)
>100	86 (37.9%)	44 (46.3%)	84 (36.5%)
≤100	141 (62.1%)	51 (53.7%)	146 (63.5%)
CEA (ng/ml)
Median (range)	2.5 (0.2–1000)	2.84 (0.5–909)	2.7 (0.4–301.8)
>10	38 (16.7%)	17 (17.9%)	23 (10.0%)
≤10	189 (83.3%)	78 (82.1%)	207 (90.0%)
Morphologic type
Mass-forming	213 (93.8%)	86 (90.5%)	NR
Papillary	3 (1.3%)	3 (3.2%)	NR
Periductal infiltrating	11 (4.8%)	6 (6.3%)	NR
Tumor size (cm)
Median (range)	5.5 (1–13)	5 (2–17.7)	5.5 (1.3–15)
>5	120 (52.9%)	46 (48.4%)	128 (55.7%)
≤5	107 (47.1%)	49 (51.6%)	102 (44.3%)
Tumor number
Solitary	180 (79.3%)	69 (72.6%)	158 (68.7%)
Multiple	47 (20.7%)	26 (27.4%)	72 (31.3%)
Cirrhosis
Positive	81 (35.7%)	23 (24.2%)	20 (8.7%)
Negative	146 (64.3%)	72 (75.8%)	210 (91.3%)
Major vascular invasion
Positive	40 (17.6%)	32 (33.7%)	96 (41.7%)
Negative	187 (82.4%)	63 (66.3%)	134 (58.3%)
Adjacent invasion
Positive	29 (12.8%)	15 (15.8%)	29 (12.6%)
Negative	198 (87.2%)	80 (84.2%)	201 (87.4%)
Necrosis
Positive	29 (12.8%)	13 (13.7%)	NR
Negative	198 (87.2%)	82 (86.3%)	NR
Lymphadenectomy
Yes	81 (35.7%)	62 (65.3%)	NR
No	146 (64.3%)	33 (34.7%)	NR
Surgical margin
Negative	204 (89.9%)	83 (87.4%)	NR
Positive	23 (10.1%)	12 (12.6%)	NR
Perineural invasion
Positive	12 (5.3%)	17 (17.9%)	45 (19.6%)
Negative	215 (94.7%)	78 (82.1%)	185 (80.4%)
Microvascular invasion
Positive	37 (16.3%)	27 (28.4%)	NR
Negative	190 (83.7%)	68 (71.6%)	NR
Nodal status
N0	34 (15.0%)	36 (37.9%)	188 (81.7%)
N1	47 (20.7%)	26 (27.4%)	42 (18.3%)
Nx	146 (64.3%)	33 (34.7%)	NR
Grade
Well	2 (0.9%)	4 (4.2%)	NR
Poor	91 (40.1%)	27 (28.4%)	NR
Moderate	134 (59.0%)	64 (67.4%)	NR
AJCC 8th Stage
IA	54 (23.8%)	22 (23.2%)	47 (20.4%)
IB	52 (22.9%)	12 (12.6%)	31 (13.5%)
II	49 (21.6)	20 (21.1%)	76 (33.0%)
IIIA	7 (3.1%)	5 (5.3%)	5 (2.2%)
IIIB	65 (28.6%)	36 (37.9%)	71 (30.9%)
TP53
Mutant	46 (20.3%)	24 (25.3%)	46 (20.0%)
Wildtype	181 (79.7%)	71 (74.7%)	184 (80.0%)
KRAS G12
Mutant	28 (12.3%)	25 (26.3%)	31 (13.5%)
Wildtype	199 (87.7%)	70 (73.7%)	199 (86.5%)
Adjuvant therapy
Yes	84 (37.0%)	52 (54.7%)	79 (34.3%)
No	137 (60.4%)	43 (45.3%)	153 (66.5%)
Unknown	6 (2.6%)	0 (0.0%)	0 (0.0%)
Follow-up time
Median (95% CI)	51 (40.5-61.5)	63.6 (50.9-76.3)	32.9 (29.4-36.4)

AJCC, American Joint Committee on Cancer; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; HBV, hepatitis B virus; HCV, hepatitis C virus; NBNC, nonhepatitis B, nonhepatitis C virus.

In the validation phase, two independent cohorts (internal and external validation cohort), were enrolled to evaluate the predictive accuracy of the model and compare it with the other existing prognostic models. The internal validation cohort of 95 consecutive ICC patients who underwent hepatectomy at the same center from February 2014 to May 2022 was recruited, using the same inclusion and exclusion criteria. Another cohort of 230 ICC patients from the FU-iCCA study was used for external validation⁵⁹. The detailed clinical and pathological characteristics of the two cohorts are shown in Table 1.

In addition, 20 locally advanced and unresectable ICCs who diagnosed from April 2020 to May 2022 at the authors’ institute were recruited to evaluate the surgical benefit of patients in the high risk group. They were pathologically diagnosed as ICC by biopsy, MAPS greater than or equal to 6; received palliative standard combination chemotherapy or hepatic arterial infusion chemotherapy, or in combination with targeted therapy and immunotherapy.

Diagnosis and treatment procedures

A preoperative diagnosis of ICC was obtained based on a special medical history (such as viral hepatitis, hepatolithiasis, primary sclerosing cholangitis, and liver fluke infection), raised serum tumor markers and particularly imaging manifestation. Once the clinical diagnosis was made, positron emission tomography CT (PET-CT) was performed to exclude extrahepatic metastases.

Major or minor hepatectomy was carried out according to tumor size, number, location, liver cirrhosis, vascular or bile duct invasion, hepatic function reserve, and the estimated volume of the future liver remnant. Routine dissection of hepatoduodenal ligamental and retropancreatic lymph nodes was performed as guideline suggested since 2017, except for those patients with severe cirrhosis or other contraindications. Hepaticojejunostomy was performed when tumor involved the primary or secondary bile ducts.

The pathological diagnosis of ICC was made by two independent experienced pathologists. Information on critical clinicopathological parameters were documented in a unified template. Major vascular invasion was defined as tumor invading the main branches of the portal vein or hepatic veins. Adjacent invasion was defined as any perforation of the liver capsule and involvement of surrounding organs, such as the colon, omentum, stomach, and adrenal gland, except for the gallbladder. Since standard adjuvant therapy for biliary tract cancer was not established before 2019, various adjuvant therapies were adopted, including gemcitabine-based systemic chemotherapy, oral S-1 or capecitabine monotherapy, transarterial chemoembolization or observation. Since 2019, routine oral capecitabine was taken in most patients according to the guideline suggestions⁶⁰.

Mutational analysis

To incorporate all potential driver mutated genes in ICC, we first comprehensively reviewed the genomic spectrum of ICC from three large cohort studies from both Western (MSK MetTropism cohort, n=407, https://www.cbioportal.org/study/summary?id=msk_met_2021), Eastern (OrigiMed2020 cohort, n=508, https://www.cbioportal.org/study/summary?id=pan_origimed_2020) and mixed (ICGC cohort, n=269, https://www.cbioportal.org/study/summary?id=chol_icgc_2017) populations from the cBioPortal database, and selected genes with greater than 1.5% mutation frequency across all of the three cohorts. In addition, some other critical genomic alterations identified in ICC, including FGFR2 fusion, CDKN2A deletion, as well as the amplifications of ERBB2, MDM2, CCND1, and MYC, were also covered in selected cases. Finally, a total of 32 ICC-related genes were selected for further prognostic evaluation (Supplementary Table 3, Supplemental Digital Content 1, http://links.lww.com/JS9/A863).

In the screening phase, different sequencing methods were adopted for the included three cohorts. Whole exome/genome sequencing was performed in the Eastern cohort. MSK-IMPACT NGS panels were applied in the Western cohort. A 404 gene panel was applied for the mixed cases from the ICGC cohort. Their processed genomic data were obtained from the cBioPortal database (https://www.cbioportal.org/study/summary?id=ihch_smmu_2014; https://www.cbioportal.org/study/summary?id=chol_icgc_2017), ICGC database (https://dcc.icgc.org/projects/LIRI-JP), the cBioPortal database (https://www.cbioportal.org/study/summary?id=ihch_msk_2021; https://www.cbioportal.org/study/summary?id=chol_icgc_2017), or the original repertoires.

For the training cohort, the mutational status of target genes (TP53 and KRAS ^G12 ) was detected by Sanger sequencing as previously described². In the internal validation cohort, the mutational status of target genes (TP53 and KRAS ^G12 ) was detected by Sanger sequencing or the NGS platform. In the external validation cohort (FU-iCCA cohort), the mutation status of TP53 and KRAS ^G12 was determined by whole exome sequencing, and the processed genomic data were retrieved from the original publication⁵⁷. The mutational status of target genes (TP53 and KRAS ^G12 ) was detected by the NGS platform for the locally advanced ICCs using the biopsy tissue.

Follow-up

The patients of the training and internal validation cohorts were followed-up every 2 months in the first two years and every 3 months thereafter at the authors’ institute after operations. At each of the follow-up visit, abdominal ultrasonography and blood tests (liver function, serum levels of CEA and CA19-9) were routinely carried out. A liver contrast-enhanced CT or MRI was performed every 6 months or when tumor recurrence or metastasis was suspected. The recurrence or metastasis was defined as the detection of a newly appeared tumor confirmed on two radiologic images, with or without elevated tumor markers. OS was used as the primary end point, which was defined as the interval from surgical treatment to death or the last follow-up. The follow-up procedure and OS information of other cohorts was available in the supplementary data of the original studies, respectively.

Classification of ICCs using different prognostic models

For the internal validation cohort, ICC cases were classified according to the existing TNM staging systems (AJCC 8th, modified AJCC 8th-1, modified AJCC 8th-2, LCSGJ 6th), nomograms (EHBH and JHU nomograms), and the prognostic scores (MEGNA and PRS score). Those patients of the external cohort were classified according to the TNM staging systems, rather than the nomograms or prognostic scores because they lacked specific information on microvascular invasion or growth patterns.

Statistics

All statistical analyses were performed using R Studio (version 1.1.463) and SPSS (version 23.0). The median follow-up time, median OS time, and 3-year and 5-year OS rates were calculated using the Kaplan–Meier method. The survival curves of the different risk groups were depicted and compared using the Kaplan–Meier method and log-rank test. The effects of individual clinical and mutational variables on OS were evaluated using univariate and multivariate Cox proportional hazards regression models. Harrell’s concordance index (C-index) was calculated using the R package rms to evaluate the performance of the MAPS and other existing models. Calibration plots were drawn using a bootstrapped sample to illustrate the agreement between the predicted and observed probabilities of 3-year and 5-year survivals. Statistical significance was set at two-tailed P<0.05.

Results

Identification of survival-related mutations in ICCs

A total of 714 ICC patients from three cohorts (Eastern, n=407; Western, n=209; and mixed, n=98) were enrolled in the screening analysis (Fig. 1). The median follow-up times were 47.3, 80.7, and 42.7 months for Eastern, Western, and mixed cohort, respectively. In the univariable COX proportional hazards analyses, 12 of the 32 candidate genes were identified to be significantly associated with poorer OS of ICC in any one of the three cohorts (Supplementary Table 4, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). Among them, only TP53 and KRAS mutations were significantly associated with OS (P<0.05) across all three cohorts (Fig. 2A, Supplementary Figure 1A and 1B, Supplemental Digital Content 2, http://links.lww.com/JS9/A864).

Figure 1.

Flow diagram showing the study design. The number of patients enrolled in each analysis and the reasons for exclusion are depicted. MAPS, mutation-annotated prognostic score.

Figure 2.

Survival analysis identified prognosis-related gene mutations in the screening cohorts. (A) Venn diagram showing overlaps of prognostic gene mutations from three independent screening cohorts. (B) Kaplan–Meier survival plots showing overall survival after hepatectomy in the combined screening cohort according to the mutational subtypes of TP53. (C) Kaplan–Meier survival plots showing overall survival after hepatectomy in the combined screening cohort according to the mutational subtypes of KRAS.

We further evaluated the association between different mutation subtypes of TP53 (missense, truncation, frameshift, splice site, in-frame indels, and complicated) and KRAS (G12V, G12D, other G12, and non-G12 mutations) with OS in the screening cohorts. We found that all mutation subtypes of TP53 were associated with a poorer OS of ICCs (Fig. 2B), whereas only KRAS ^G12 mutations (G12V/D/S/C/A), but not KRAS ^non-G12 mutations (Q61R/L/K/H; G13D/C; A146T), were predictive of a shorter OS (Fig. 2C). These results suggest that TP53 and KRAS ^G12 mutations may be used to predict OS in ICC patients.

Development of a mutation-annotated prognostic scoring system

We then evaluated the prognostic role of the identified mutations (TP53 and KRAS ^G12 ) and the clinicopathological and laboratory factors in the training cohort (n=227). The median follow-up duration in this cohort was 51 months. The median OS was 21 months, and the 3-year and 5-year OS rates were 34.8 and 26.2%, respectively.

According to the results of the univariate analysis, 14 parameters were significantly associated with poorer OS (P<0.05, Table 2). In the multivariable analyses, nine factors were identified as independent predictors for OS, which included tumor size [hazard ratio (HR), 1.52; 95% CI: 1.07–2.16; P=0.02], tumor number (HR, 1.79; 95% CI: 1.18–2.71; P=0.006), vascular invasion (HR, 1.69; 95% CI: 1.12–2.55; P=0.012), lymph node metastasis (HR, 2.34; 95% CI: 1.54–3.56; P<0.001), adjacent invasion (HR, 2.04; 95% CI: 1.28–3.26; P=0.003), CA19-9 greater than 100 (HR, 1.91; 95% CI: 1.33–2.76; P=0.001), CEA greater than 10 (HR, 1.98; 95% CI: 1.27–3.10; P=0.003), TP53 mutation (HR, 3.43; 95% CI: 2.29–5.15; P<0.001) and KRAS ^G12 mutation (HR, 2.01; 95% CI: 1.23–3.27; P=0.005).

Table 2.

Univariable and multivariable analyses of factors associated with overall survival in the training cohort.

	Univariable model		Multivariable model
Variables	HR (95% CI)	P	HR (95% CI)	P
Sex (male/female)	1.176 (0.851-1.625)	0.326
Age (>60/≤60 years)	1.104 (0.806–1.512)	0.539
HBV/HCV (positive/negative)	1.003 (0.727–1.384)	0.987
CEA (>10 /<10 ng/ml)	2.880 (1.985–4.180)	<0.001	1.984 (1.272–3.097)	0.003
CA19-9 (>100 /<100 U/ml)	2.424 (1.768–3.324)	<0.001	1.913 (1.328–2.756)	0.001
Tumor size (>5/≤ 5 cm)	1.883 (1.365–2.599)	<0.001	1.518 (1.069–2.156)	0.02
Tumor number (multiple/single)	2.322 (1.619–3.331)	<0.001	1.788 (1.179–2.711)	0.006
Major vascular invasion (yes/no)	2.319 (1.595–3.373)	<0.001	1.690 (1.120–2.548)	0.012
Microvascular invasion (yes/no)	1.557 (1.039–2.333)	0.032	1.432 (0.911–2.253)	0.12
Perinrural invasion (yes/no)	1.933 (1.017–3.673)	0.044	1.191 (0.568–2.495)	0.644
Adjacent invasion (yes/no)	2.611 (1.710–3.987)	<0.001	2.044 (1.283–3.256)	0.003
Lymphadenectomy (yes/no)	1.214 (0.877–1.678)	0.242
Nodal status (yes/no)	2.646 (1.852–3.781)	<0.001	2.343 (1.542–3.561)	<0.001
Morphologic type (PI/other)	1.997 (1.015–3.929)	0.045	0.540 (0.235–1.237)	0.145
Grade (poor/moderate-well)	1.386 (1.009–1.906)	0.044	1.346 (0.960–1.888)	0.085
Surgical margin (positive/negative)	1.638 (1.011–2.652)	0.045	0.765 (0.430–1.362)	0.363
Necrosis (yes/no)	1.099 (0.694–1.743)	0.687
Cirrhosis (yes/no)	1.247 (0.903–1.722)	0.179
Adjuvant therapy (yes/no)	1.141 (0.823–1.581)	0.428
TP53 mutation (yes/no)	2.451 (1.706–3.522)	<0.001	3.430 (2.286–5.146)	<0.001
KRAS G12 mutation (yes/no)	2.877 (1.877–4.411)	<0.001	2.008 (1.232–3.272)	0.005

Factors that were significantly associated with overall survival in the univariable or multivariable analyses were demonstrated in bold.

Factors showing significance by univariable analysis were adopted when multivariable analysis was performed.

CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; HBV, hepatitis B virus; HCV, hepatitis C virus; HR: hazard ratio; PI: periductal infiltrating.

Based on these results, we established a risk model, termed the MAPS, which incorporates these nine prognostic parameters. In this model, each parameter was assigned one point and the total score was calculated for each case, with increasing scores indicating a higher risk of death. Finally, five prognostic groups were identified: Group 1, extremely low risk (MAPS=0); Group 2, low risk (MAPS=1); Group 3, intermediate risk (MAPS=2); Group 4, high risk (MAPS=3–5); and Group 5, extremely high risk (MAPS≥6). The five risk groups showed well-separated survival curves (P<0.001, Fig. 3), with gradually decreasing survival rates and increasing HRs for death (Supplementary Table 5, Supplemental Digital Content 1, http://links.lww.com/JS9/A863).

Figure 3.

Kaplan–Meier survival plot showing overall survival after hepatectomy in the training cohort (n=227) according to different MAPS groups. MAPS, mutation-annotated prognostic score.

Evaluation of the predictive efficacy of MAPS for ICC

An internal validation cohort (n=95) was recruited to evaluate the predictive accuracy of MAPS and compare it with other existing prognostic models. With a median follow-up time of 60.8 months, the median OS of this cohort was 21.8 months, and the 3-year and 5-year OS rates were 33.9 and 25.5%, respectively.

The performance (predictive accuracy of OS) of the MAPS model was evaluated using the C-index, Kaplan–Meier method, and bootstrapped calibration plot. The C-index of the MAPS was 0.782 (95% CI: 0.728–0.837), which was significantly higher than that of the other prognostic models (AJCC 8th staging:0.682, modified AJCC 8th staging-1 0.667, modified AJCC 8th staging-2 0.671, LCSGJ staging:0.667, EHBH nomogram:0.755, JHU nomogram:0.723, MEGNA score: 0.673, PRS score: 0.715). The five subgroups according to MAPS, but not the others, showed clearly separated survival curves (P<0.001, Fig. 4A–I) and increased HRs for death (Supplementary Table 5, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). These results suggest that MAPS demonstrated significantly better accuracy in predicting OS than the existing prognostic models. Moreover, the calibration curve showed excellent concordance between MAPS predicted and observed probabilities of 3-year and 5-year survival (Supplementary Figs 2A and 2B, Supplemental Digital Content 2, http://links.lww.com/JS9/A864).

Figure 4.

Kaplan–Meier survival plots showing overall survival after hepatectomy in the internal validation cohort (n=95) according to different prognostic models. The analyzed models include (A) MAPS, (B) MEGNA, (C) PRS, (D) EHBH nomogram, (E) JHU nomogram, (F) LCSGJ 6th, (G) AJCC 8th, (H) modified AJCC 8th-1, (I) modified AJCC 8th-2.

External validation of the predictive efficacy of MAPS

The predictive accuracy of the MAPS was further evaluated in an external validation cohort (n=230). With a median follow-up time of 32.9 months, the median OS time was not reached and the 3-year OS rate was 54.9%.

The C-index of the MAPS was 0.731 (95% CI: 0.681–0.781). The C-index of the MAPS was significantly higher than that of the existing staging systems (P<0.05), but no significant difference was observed between them (AJCC 8th staging,0.693; modified AJCC 8th staging-1,0.667; modified AJCC 8th staging-2,0.676; LCSGJ 6th staging,0.705). Only the subgroups by MAPS and mAJCC 8th-1, but not the other models, showed clearly separated survival curves (P<0.001, Fig. 5A–E) and increased HRs for death (Supplementary Table 5, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). The calibration curve also showed excellent concordance between MAPS predicted and observed probabilities of 3-year and 5-year survival (Supplementary Fig. 3A and 3B, Supplemental Digital Content 2, http://links.lww.com/JS9/A864). These results further confirmed that MAPS demonstrated good performance in the prognostication of ICC.

Figure 5.

Kaplan–Meier survival plots showing overall survival after hepatectomy in the external validation cohort (n=230) according to different prognostic models. The analyzed models include (A) MAPS, (B) LCSGJ 6th, (C) AJCC 8th, (D) modified AJCC 8th-1, (E) modified AJCC 8th-2.

The implication of MAPS in the treatment selection of ICC patients

Then we investigated the possibility of using the MAPS as a tool to identify patients who might benefit from adjuvant therapy. In the 321 ICC patients from the training and internal validation cohorts, the HRs reflecting the benefit from adjuvant therapy decreased from group 1 to group 5, and only group 5 patients showed a significant benefit from adjuvant therapy (HR=0.269, 95% CI: 0.077–0.937, P=0.039) (Supplementary Table 6, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). Consistently, in the external validation cohort, a marginally significant benefit from adjuvant therapy was observed only in group 5 (HR=0.324, 95% CI: 0.092–1.143, P=0.08) (Supplementary Table 6, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). These results indicate that only ICC patients with higher MAPS scores would benefit from adjuvant therapy.

Considering the extremely poor OS of ICC patients with MAPS greater than or equal to 6 (Group 5) after surgery, we further investigated whether surgical resection (n=16) could bring more benefit to OS in Group 5 ICCs than nonsurgical palliative treatment (n=20) in a preliminary study (Supplementary Table 7, Supplemental Digital Content 1, http://links.lww.com/JS9/A863). No significant difference was found in the OS of ICC patients with MAPS greater than or equal to 6 between resection (median OS, 5.067 months) and nonpalliative treatment (median OS, 5.533 months; P=0.820) (Fig. 6). These results suggest that surgical treatment is not the only or the best choice for patients with higher MAPS (≥6).

Figure 6.

Kaplan–Meier survival plot showing overall survival after hepatectomy (n=16) or palliative treatment (n=20) among patients with MAPS greater than or equal to 6 ICCs.

Discussion

In the present study, we systemically evaluated the prognostic value of recurrent somatic mutations in large cohorts of localized ICCs after surgical treatment from both Western and Eastern populations. By integrating prognosis-related mutations and traditional clinicopathological factors, we developed a novel prognostic model to predict OS and help guide decision-making for ICC.

ICC demonstrates a high discrepancy in incidence between Eastern and Western populations, largely due to its complicated etiologies^1–3. Accordingly, the genomic landscapes of ICC differ greatly between Western and Eastern populations. Among Eastern populations, TP53 and KRAS have been recognized as the most important driver genes with the highest mutation rates. In contrast, BAP1 and IDH1/2 mutations, FGFR2 fusion, and CDKN2A deletion are more prevalent among Western populations⁶¹. In previous study, we have also found that the prognostic values of driver mutations in ICC tended to be etiologically and geographically specific^62,63. So, we at first analyzed the prognostic values of the most recurrent mutations in large cohorts of ICCs from Western, Eastern, or both to elucidate their prognostic values for ICCs from different regions. We found that the prognostic values of TP53 and KRAS mutations were consistently significant in both Eastern and Western populations. However, the other frequently occurred mutations, such as IDH1/2, BAP1, or ARID1A, did not show prognostic significance in either the Eastern or Western cohorts. Furthermore, many previous studies have reported that specific, but not all mutation subtypes of TP53 and KRAS were associated with poor outcomes in various cancer types^64–66. Similarly, the presence of KRAS ^G12 mutations but not KRAS ^non-G12 mutations was reported to be associated with a worse prognosis in ICC⁶⁷. In the present study, we found that all TP53 mutation subtypes were associated with poorer OS, while only the KRAS ^G12 mutation rather than the KRAS ^non-G12 mutation demonstrated a negative prognostic value in ICC. As such, TP53 and KRAS ^G12 mutations were identified and used in further investigations.

Based on the COX regression analyses of both univariate and multivariate, we developed a prognostic score (MAPS model) by incorporating clinicopathological, laboratory variables, as well as the mutational information. The included clinicopathological and laboratory parameters were the same as that of the Wang’s nomogram²². This is reasonable since the patients enrolled in these two studies were from the same region. These results were also consistent with an international multicohort study²⁴. To make the MAPS model as simple as possible, we did not weight factors by HR.

As the first clinical risk scoring model incorporating clinically available genetic information (TP53 and KRAS ^G12 mutations) of ICC, MAPS showed several advantages over the others. First and most important, MAPS performs much better in terms of the predictive efficacy of OS compared with the other existing ones. Second, the MAPS is easy to calculate, and convenient to implement in clinical practice. Another advantage of the current model is the identification of an extremely good (Group 1: MAPS=0) and an extremely dismal (Group 5: MAPS≥6) prognostic groups of ICCs, which may help guide clinical treatment stratification. Among the previous models, only the LCSGJ 6th staging system could identify a subgroup of ICCs that have a good prognosis with a 5-year survival rate greater than 80%¹⁰. However, the criteria for this subtype are very strict, which includes solitary tumor size less than 2 cm, and no lymph node invasion. As a result, no such patient could be identified in the internal validation cohort of our study. By MAPS model, 10, 8, and 28 patients were identified as extremely good prognosis (Group 1 with MAPS=0, with 5-year OS rate >75%) in the training, internal validation, and external validation cohorts, respectively. Considering the extremely good prognosis of this subgroup, whether adjuvant chemotherapy is necessary and would bring more benefit to them needs further investigation. On the contrary, we also identified a subgroup with an extremely poor prognosis (Group 5, MAPS≥6), these patients showed much more benefit from adjuvant therapy compared with other groups. These results supported that MAPS could be used to further stratify the subgroup of ICC patients that are more possible to get benefit from adjuvant therapy. Considering the median OS of advanced and metastatic ICC after regional or systemic treatments has exceeded 12 months⁶⁸, and the high complication rate of extended major liver resection, whether these patients could benefit from upfront surgery deserves further elucidation. Recent studies have found that the loco-regional therapies, such as hepatic arterial infusion chemotherapy, could provide similar OS versus hepatectomy in multifocal or node-positive ICCs^69,70. In another recent study, neoadjuvant therapy is recommended for those ICC patients with a high risk of early recurrence after hepatectomy^71,72. Our findings indicate that MAPS may be also helpful in the selection of ICC patients who would be benefit from surgical treatment.

Certainly, there are still some limitations with the MAPS model. First, although the screening cohorts were from both Eastern and Western populations, MAPS was developed and validated mainly in Chinese ICC patients. Further validation in Western populations is needed. Second, the predictive efficacy of a prognostic model is influenced by the therapeutic strategy evolution. Much progress in the surgical treatments of ICC has been obtained during the past decades, such as routine lymphadenectomy^73,74. More and more ICC patients may undergo preoperative neoadjuvant treatments such as chemotherapy, molecular targeted therapy, and immunotherapy, or their combinations^75,76. Third, the evidence supporting the implication of MAPS score in treatment selection for ICC patients is still weak due to the relative small sample size, and the clinicopathological parameters might be unmatched in the two arms (surgery vs. palliative treatment). Further studies are needed to confirm the ability of this score to select patients for upfront surgery or neoadjuvant therapy. Lastly, it would need to be further updated including more prognostic markers from both cancer itself and the tumor microenvironment^77,78.

Conclusion

In conclusion, we established a MAPS model by integrating genetic mutations and clinicopathological characteristics using large cohorts of ICCs. This model demonstrates good performance in predicting postoperative OS of ICC patients, and may be also helpful in predicting the possible benefit of upfront surgery and adjuvant therapy. However, further studies are still needed to validate the ability of this score in clinical prognostication and treatment stratification for ICC patients.

Ethical approval

The Ethical Committee of Huashan Hospital, Fudan University approved this study (KY2022-945).

Consent

Written informed consent was obtained from the patient for publication of this case report and accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal on request.

Sources of funding

This work was supported by the National Natural Science Foundation of China (No. 81930074, No. 91959203 to LX-Q, No. 82272836 to JH-C, No. 82072696 to WW-Z, No. 82203792 to XY-W) and the Program of Shanghai Academic/Technology Research Leader (22XD1400300 to JH-C).

Author contribution

X.Y.W., W.W.Z., and L.L.: contributed equally to this work; L.X.Q. and J.H.C.: had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis; L.X.Q., J.H.C., and X.Y.W.: conceptualization; X.Y.W., W.W.Z., L.L., and Y.T.L.: formal analysis; X.Y.W., Y.T.L., and J.F.: methodology; L.X.Q., J.H.C., X.Y.W., and W.W.Z.: funding acquisition; L.X.Q., J.H.C., B.B.Y., J.B.Z., H.L.J., X.Y., and C.H.: resources; L.X.Q., J.H.C., and X.Y.W.: project administration; L.X.Q. and J.H.C.: supervision; X.Y.W., W.W.Z., and L.L.: writing – original draft; L.X.Q. and J.H.C.: writing – review and editing. All authors contributed in data curation, investigation, and validation.

Conflicts of interest disclosure

The authors declare that they have no conflicts of interest.

Research registration unique identifying number (UIN)

1. Name of the registry: Chinese Clinical Trial Registry (ChCTR).

2. Unique identifying number or registration ID: ChiCTR2300070574.

3. Hyperlink to your specific registration (must be publicly accessible and will be checked): https://www.chictr.org.cn/showprojEN.html?proj=192058

Guarantor

Lun-Xiu Qin and Jin-Hong Chen.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available since this could compromise the privacy of research participants.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Supplementary Material

js9-109-3506-s001.docx ^{(57.7KB, docx)}

js9-109-3506-s002.docx ^{(855.6KB, docx)}

js9-109-3506-s003.docx ^{(17.6KB, docx)}