Unraveling the Role of Molecular Profiling in Predicting Treatment Response in Stage III Colorectal Cancer Patients: Insights from the IDEA International Study

Simple Summary The treatment and prognosis of colorectal cancer (CRC) patients vary depending on their disease stage at diagnosis. Understanding the processes of tumorigenesis and disease development can reveal new therapeutic targets and guide patient management. On the basis of this, we investigated the molecular profiles of 237 patients with stage III CRC enrolled in the international IDEA study. We also correlated the molecular profile with Toll-like and vitamin D receptor polymorphisms, clinicopathological and epidemiological characteristics, and patient outcomes. This study suggests that the molecular characterization of tumor cells may contribute to the understanding of the biological disease course. Mutations can serve as promising prognostic biomarkers leading to better treatment options. If the results are confirmed in larger patient cohorts, then this is expected to guide clinical decision-making and personalized and improved care, and reduce treatment toxicity and patient and health system costs. Abstract Background: This study aimed to investigate the molecular profiles of 237 stage III CRC patients from the international IDEA study. It also sought to correlate these profiles with Toll-like and vitamin D receptor polymorphisms, clinicopathological and epidemiological characteristics, and patient outcomes. Methods: Whole Exome Sequencing and PCR-RFLP on surgical specimens and blood samples, respectively, were performed to identify molecular profiling and the presence of Toll-like and vitamin D polymorphisms. Bioinformatic analysis revealed mutational status. Results: Among the enrolled patients, 63.7% were male, 66.7% had left-sided tumors, and 55.7% received CAPOX as adjuvant chemotherapy. Whole exome sequencing identified 59 mutated genes in 11 different signaling pathways from the Kyoto Encyclopedia of Genes and Genomes (KEGG) CRC panel. On average, patients had 8 mutated genes (range, 2–21 genes). Mutations in ARAF and MAPK10 emerged as independent prognostic factors for reduced DFS (p = 0.027 and p < 0.001, respectively), while RAC3 and RHOA genes emerged as independent prognostic factors for reduced OS (p = 0.029 and p = 0.006, respectively). Right-sided tumors were also identified as independent prognostic factors for reduced DFS (p = 0.019) and OS (p = 0.043). Additionally, patients with tumors in the transverse colon had mutations in genes related to apoptosis, PIK3-Akt, Wnt, and MAPK signaling pathways. Conclusions: Molecular characterization of tumor cells can enhance our understanding of the disease course. Mutations may serve as promising prognostic biomarkers, offering improved treatment options. Confirming these findings will require larger patient cohorts and international collaborations to establish correlations between molecular profiling, clinicopathological and epidemiological characteristics and clinical outcomes.


Introduction
Colorectal cancer (CRC) is one of the most common malignancies and the second most common cause of death from cancer [1].In 2020, 1.9 million new cases of CRC and approximately 935,000 deaths were reported [2].By 2030, the global burden of CRC is predicted to be 60%, with more than 2.2 million new cases and 1.1 million deaths.By 2035, the total number of deaths from rectal and colon cancer is estimated to increase by 60% and 71.5%, respectively [2].Depending on the stage of the disease at the time of diagnosis, both the treatment and prognosis differ.Patients with stage III CRC have an overall 5-year survival rate of 60%.Adjuvant chemotherapy aims to increase this rate and extend both the overall and disease-free survival [3].Since 2004, folinic acid, fluorouracil, and oxaliplatin (FOLFOX) or capecitabine and oxaliplatin) for six months has been the standard treatment regimen [3,4].However, oxaliplatin can lead to adverse effects, particularly peripheral sensory neuropathy [5].
Considering the increased incidence of the disease, toxicity of the treatment, cost, and efforts to reduce the duration of treatment for all the aforementioned reasons [5], the international IDEA study was designed to evaluate the hypothesis of non-inferiority of the 3-month vs. 6-month adjuvant chemotherapy with FOLFOX or CAPOX [6].
Although strong, AJCC/UICC-TNM staging (American Joint Committee on Cancer/Union Internationale Contre le Cancer-extent of primary tumor, regional lymph node involvement, presence of distant metastases) often fails to provide complete prognostic information because the outcome varies even among patients at the same stage [7].Therefore, there is an urgent need to identify new tools that can contribute to CRC prognosis.The aim of the current study was to investigate the molecular profile of surgical specimens from stage III CRC patients enrolled in the international IDEA study, for whom paraffin-embedded cancer tissue was available.Following genetic profiling, correlations were performed with clinicopathological characteristics, as well as with patient outcomes.Additionally, the patients were tested for vitamin D receptor (VDR) and Toll-like receptor (TLR) gene polymorphisms in peripheral blood, as our previous studies have demonstrated the role of such polymorphisms in tumor development and progression [8][9][10].

Patient Enrollment
The Hellenic Oncology Research Group (HORG) enrolled 708 patients with CRC in the international IDEA study (ClinicalTrials.govIdentifier: NCT01308086).Of these, 237 stage III patients enrolled from September 2009-July 2015, with formalin-fixed paraffinembedded (FFPE) tissues, were included in the current study (Supplementary Table S1).All patients were aged >18 years and received adjuvant chemotherapy with FOLFOX or CAPOX.None of the enrolled subjects had any other documented malignancies.

Formalin-Fixed Paraffin-Embedded (FFPE) Tissues
All surgical materials were evaluated by a specialized pathologist at the Department of Pathology of the University General Hospital of Heraklion, Crete, and the most representative and enriched tumor areas were selected for dissection.Healthy tissue was used as the control tissue.The specifications of the FFPE sections used for DNA extraction were as follows: a. tissue surface area, 25 mm 2 ; b. section thickness, 10 µm; c.At least 10 sections with >50% cancer cells; d.Cancer cells were collected from areas rich in cancerous tissue and avoiding healthy tissue, adipose tissue, necrotic areas, and lymphocytes that decrease the content of cancer DNA.To facilitate the collection of appropriate cells from the 10 sections, an additional section was stained with hematoxylin-eosin to locate the cancerous areas.

TLR and VDR Genotyping in Blood Samples
A total of 5 mL of peripheral blood in EDTA was collected from each patient, and DNA was extracted using the QIAamp DNA Blood Mini kit (QIAGEN, Hilden, Germany) according to the manufacturer's instructions.The DNA concentration was determined using a NanoDrop ND-1000 v3.3 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA).
To determine the genetic variants of TLRs and VDRs, polymerase chain reaction and restriction fragment length polymorphism (PCR-RFLP) were used to determine the genetic variants of TLRs and VDRs.For TLR2 196-to-174 Ins/Del genetic variants, PCR was used, while TLR4 (Asp299Gly-rs4986790 and Thr399Ile-rs4986791) and TLR9 (T1237C-rs5743836 and T1486C-rs187084) genetic variants were determined using PCR-RFLP.The materials and conditions for each gene target have been previously described [8,9].Similarly, for genotyping of VDR genetic variants at the TaqI (rs731236), ApaI (rs7975232), FokI (rs10735810), and BsmI (rs1544410) positions, PCR-RFLP was used.The reagents and PCR conditions have been previously described in detail [8][9][10].The patients were classified as wild-type, heterozygous, or homozygous for each single nucleotide polymorphism, based on the absence or presence of the restriction site in both alleles.

Whole Exome Sequencing (WES)
The Illumina DNA Prep with Enrichment kit (Illumina, San Diego, CA, USA, 92122) was used for library preparation and enrichment, and sequencing of both tumor and normal tissues was conducted using the NovaSeq 6000 system (Illumina) flow cell.For each sample, 250-350 ng of high-quality DNA was quantified using a Qubit Fluorometer (Thermo Fisher Scientific, Paisley, UK), ensuring consistency across all samples.Samples with quality threshold 260/280 ratio of at least 1.8-2.2were selected to ensure high-quality DNA and used for library preparation.Following the WES, two files (.fastq) containing sequencing data were extracted from each tissue (tumor and normal) using forward and reverse reads (2 × 150 base pairs).The coverage within the designated target region and the number of generated reads exceeded the threshold set by the manufacturer (Illumina).Specifically, the coverage reached a level of 150×, while the number of reads surpassed 40 billion.

Bioinformatic Analysis
After obtaining the raw data from the WES analysis, a bioinformatics pipeline was used to process the data and generate interpretations.This included alignment to the human genome, variant calling, variant filtering, and annotation of the variants.The processed data were analyzed to identify somatic variants and evaluate their functional significance, particularly those associated with CRC (Figure 1) [11][12][13].Initially, the raw sequences were aligned to the human genome (version hg19/GRCh37) [14] utilizing the Burrows-Wheeler Transform [15], and variant calling was conducted using the genome analysis toolkit (GATK, version 4.2.3) for SNPs and insertion/deletions (INDELs).This generates two variant call format files (VCF) for each patient: one for the tumor and one for the normal tissue, respectively [16].The somatic variants were isolated by subtracting the germline variants from the tumor, and a custom gene panel was utilized based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database on CRC-correlated genes [17] (Supplementary Tables S2 and S3).Subsequently, ANNOVAR software (version 2.17) was employed to annotate the SNPs and INDELs, providing functional information that can determine the biological significance of each variant and identify CRC-associated variants [18].
To identify clinically significant variants and the signaling pathways involved, a filtering strategy was employed based on the functional position of the variants.Variants located in exons and splice sites were isolated as they are more likely to cause diseases [19].In addition, the variants located in exons were further filtered based on their functional consequences, excluding variants causing synonymous mutations.This step considered synonymous mutations that do not affect the amino acid sequence of proteins and are therefore less likely to be clinically relevant.All analyses were run in the Anaconda Powershell Prompt (Anaconda3, Inc., Wang and Oliphant, Berlin, Germany) on Ubuntu 20.04.3 LTS.

Consensus Molecular Subgroups (CMS) Assignation
Regarding the CMS subtypes, patients were classified into four subtypes according to the transcriptomics of CRC  To identify clinically significant variants and the signaling pathways involved, a filtering strategy was employed based on the functional position of the variants.Variants located in exons and splice sites were isolated as they are more likely to cause diseases [19].In addition, the variants located in exons were further filtered based on their functional consequences, excluding variants causing synonymous mutations.This step considered synonymous mutations that do not affect the amino acid sequence of proteins and are therefore less likely to be clinically relevant.All analyses were run in the Anaconda Powershell Prompt (Anaconda3, Inc., Wang and Oliphant, Berlin, Germany) on Ubuntu 20.04.3 LTS.

Consensus Molecular Subgroups (CMS) Assignation
Regarding the CMS subtypes, patients were classified into four subtypes according to the transcriptomics of CRC [20].CMS1 (immune) is characterized by microsatellite instability (MSI), high levels of mutations in CpG island methylator phenotype (CIMP) and BRAF genes, and a low prevalence of SNCA gene mutations.It is associated with lymphocyte infiltration and immune activation, along with prominent hypermethylation and reduced signaling through the WNT pathway [21,22].CMS2 (canonical) demonstrates epithelial characteristics and is marked by high chromosomal instability, an elevated count of somatic copy number alterations, as well as mutations in the WNT and MYC genes, leading to heightened activity in these intracellular signaling pathways [21,22].CMS3 (metabolic) exhibits a distinctive global genomic and epigenomic profile with mixed features, including metabolic reprogramming and dysregulated pathways.It displays increased activity in glutaminolysis and lipidogenesis, enriched with KRAS-activating mutations.CMS3 presents a moderate or low mixed state of MSI, intermediate CIMP, and moderate activation of WNT and MYC signaling.Additionally, it is characterized by PIK3CA mutations and IGBP3 overexpression but lacks BRAF mutations [21,22].CMS4 (mesenchymal) is typified by positive gene regulation and the overexpression of proteins involved in stromal infiltration, mesenchymal activation, extracellular matrix remodeling, neoangiogenesis, prominent TGF-β activation, and complement pathways.The presence of CTNNB1 mutations is a distinguishing feature of CMS4 [22].

Statistical Analysis
After characterizing the patients' molecular profiles, the clinical, pathological, and epidemiological characteristics were examined to determine their association with patient outcomes.Disease-free survival (DFS) and overall survival (OS) were calculated from the day of tumor excision until the first documented recurrence or death, respectively.Recurrence was defined as the presence of metastatic disease, local recurrence, or a second primary tumor.The possible associations between baseline characteristics, recurrence, and individual or concurrent mutations were compared using the 2-sided Fisher exact test for categorical variables.The association between risk factors and time-to-event endpoints was evaluated using the log-rank test, and the Kaplan-Meier method was used to generate DFS and OS curves.Univariate and multivariate Cox regression analyses were conducted to evaluate the correlation between the potential prognostic factors and DFS or OS.Statistical significance was defined as p ≤ 0.05, and the statistical tool used was SPSS v. 26.
Additionally, Monte Carlo simulation methods were performed.The Monte Carlo approach, named after the Monte Carlo Casino in Monaco due to its reliance on stochastic principles, is a computational procedure that employs random sampling to tackle intricate mathematical and statistical quandaries.This method entails the generation of random samples from known probability distributions, subsequently utilizing these samples to approximate solutions to problems that defy analytical resolution [23][24][25].The results indicate a high level of confidence (99%), at the 95% significance level, with the predicted statistical power deviating by no more than 4% from the values generated by simulation across various model parameters.This equates to sample size discrepancies of fewer than four subjects and discrepancies in the detectable accuracy difference of less than 0.6%.

Annotated Variants for Each Position
The KEGG CRC panel detected 85,871 uniquely annotated variants.Of these, were detected in exonic positions, including 25 splice variants.The remaining variants were non-coding, with 81,555 intronic variants being the most common, followed by UTR variants, 788 downstream variants, and 742 upstream variants.
The objective of this study was to analyze genetic changes in surgical samples from patients with stage III CRC and detect VDR and TLR gene polymorphisms in peripheral blood samples.The study included 237 patients, 84 of whom had available blood samples.The WES and KEGG gene panel for CRC revealed 59 mutated genes belonging to 11 distinct signaling pathways.Of these, mutations in APC2, BRAF, MAPK10, MLH1, MSH6, RHOA, TGFB, and TGFB2 have been linked to a significant impact on patient survival.APC, TP53, KRAS, and MSH3 were the most commonly observed mutations in this study.
APC encodes an anti-tumor protein that competes with the Wnt signaling pathway and is involved in cell migration, adhesion, and apoptosis.APC mutations are responsible for familial adenomatous polyposis (FAP), an autosomal dominant precancerous disease that typically leads to malignancy.APC mutations are commonly observed in CRC cases [45].Similarly, APC2 mutations, which are directly associated with APC's tumor-suppressive function [46], have been linked to worse prognosis in CRC patients [47,48].This study confirms that APC2 mutations in patients with stage III CRC are associated with lower overall survival but do not represent an independent prognostic factor.TP53 encodes an anti-tumor protein that regulates the expression of target genes, leading to cell cycle arrest, apoptosis, senescence, DNA repair, or metabolic changes.Similarly to APC, TP53 are frequently observed in CRC cases [45].Furthermore, mutations in the APC2 gene, which are directly linked to APC's tumor-suppressive function [46], are also associated with worse prognosis in CRC patients [47,48].This study confirms that while APC2 mutations in stage III CRC patients are linked to lower overall survival, they do not represent an independent prognostic factor.Various human cancers, including approximately 60% of CRC, are associated with mutations in the TP53 gene [49,50].Prior studies have shown that mutations in TP53 resulting in the loss of its transcriptional activity can lead to uncontrolled cellular proliferation in multiple organs, including the colon [51].Similarly, KRAS mutations are the primary indicators of gastrointestinal cancers and are found in approximately 40% of patients with CRC (stage II-IV) [52].They serve as negative prognostic factors for carcinogenesis and anti-EGFR therapy [53] because intracellular signal interruption leads to uncontrolled cellular proliferation and cancer.MSH3 mutations have been mainly linked to endometrial cancer, but there are reports of its relationship with inflammatory processes, such as ulcerative colitis and Crohn's disease, which considerably increase the likelihood of CRC development [54,55].MSH3-associated CRC seems to follow the classic APC pathway, as patients with adenomas and CRC carrying APC mutations showed MSH3 deficiency [56], as confirmed in this study.In addition to the common mutations detected in the patient group, mutations in AKT1, ARAF, BAD, MAPK10, RAC3, RHOA, TGFB2, and TGFB3 were associated with worse prognosis in this study.Furthermore, mutations in ARAF and MAPK10 were identified as independent prognostic factors for DFS, whereas mutations in RAC3 and RHOA were identified as independent prognostic factors for decreased OS.
Regarding the detection of gene polymorphisms in the blood, it was demonstrated that patients with VDR gene polymorphisms have shorter survival rates, and this is in agreement with our previous demonstrations in various cohorts [8][9][10].
This study sheds light on the association between mutations in genes involved in signaling pathways such as PI3K-Akt, MAPK, apoptosis, and CRC.To the best of our knowledge, this is the first report of its kind in the literature.The reactivation of embryonic self-renewal pathways, such as Hedgehog, Notch, and TGFB/Stat3, is characteristic of most tumors, including CRC.The Wnt pathway is also essential in most CRC.Targeting embryonic pathways directly is likely to be more effective against stem and differentiated cancer cells [57][58][59].Tumors that are addicted to increased regulated activity of the embryonic pathway, in combination with high tumor heterogeneity, may be more vulnerable to such therapies [60][61][62].Patients with VDR polymorphisms had concurrent mutations in genes involved in cell cycle, apoptosis, PI3K-Akt, WNT, MAPK, ErbB, MSI, and RAS.Similarly, TLR9 polymorphisms were associated with mutations in genes involved in apoptotic signaling pathways, PI3K-Akt, and Wnt.Previous studies from our group have demonstrated that higher detection of TLR and VDR polymorphisms in CRC patients, especially advanced-stage patients, highlights the role of these polymorphisms in carcinogenesis, disease progression, and ultimately, patient survival [9,10,63].Regarding DFS, tumors in the sigmoid or right colon and mutations in the ARAF and/or MAPK10 genes were associated with shorter DFS, a fact that has been confirmed in previous studies [64][65][66].To our knowledge, this is the first study to highlight the role of ARAF and MAPK10 mutations as independent prognostic factors for decreased DFS.
The observation of statistically lower OS in patients with right colon tumors and gene mutations has been confirmed in the literature.Borakati et al. conducted a retrospective study and found that tumors in the right colon were independent prognostic factors for reduced OS after hepatic metastasectomy, regardless of the higher rates of liver metastases and larger metastases in the left colon [67].Patients with mutations in genes, such as AKT1, APC2, ARAF, BAD, MAPK10, RAC3, RHOA, TGFB2, and TGFB3 had significantly reduced OS, as reported in other studies that also linked APC2, RHOA, and TGFB mutations to worse prognosis [47,48,68].Conversely, studies have shown that mutations in MSH6 are associated with a lower risk of developing CRC, and patients with such mutations have a milder clinical presentation [69,70].In the present study, patients with MSH6 mutations had a significantly longer OS, confirming that MSH6 mutations are good prognostic factors.Moreover, regarding the CMS subtypes, in the current study, it was demonstrated that the patients were classified as CMS1 subtype based on BRAF and MSI pathway mutations.Such a subtype may lead to worse survival rates after relapse, as compared to CMS2 subtype, which has been classified based on Myc and Wnt pathway mutations.CMS3 and CMS4 have also been associated with bad survival rates [21,22].
Furthermore, whether the mutations related to patients' survival were due to chance was assessed via Monte Carlo simulations.Sidedness and eight mutations were associated with patients' survival, which confirms the univariate analysis results.
To determine whether the results of the present work could be further validated by other studies, we examined the available data on the National Cancer Institute's GDC data portal [71].Case filters specifying the location of the primary tumor and clinical filters specifying gender, age, and location were applied, and the information regarding the genes of the study and the corresponding survival plots were analyzed.Consistent with the result of the current study, patients with tumors in the colon had mutations in the KRAS gene more frequently than patients with tumors in the sigmoid (p < 0.001).Moreover, patients with AKT1, ARAF, and RHOA mutations exhibited a significantly shorter OS compared to wild-type patients (p < 0.001, p < 0.001, p < 0.001, and p < 0.001, respectively) [72].All of the above are in agreement with our results.Moreover, MSH6 mutations are significantly correlated with previous studies, as demonstrated on the National Cancer Institute's GDC data portal [73], and this is also in agreement with our results [71].Concurrent mutations (co-mutations) are a significant factor that have been minimally investigated in CRC.Studies in patients with non-small cell lung cancer have shown distinct biological behavior and prognosis in KRAS/LKB1, KRAS/TP53, or KRAS/p16 mutated tumors [74].Additionally, our group has previously reported the importance of evaluating the loss of LKB1 through immunohistochemistry in early-stage CRC, particularly in BRAF V600E mutated tumors [75].In the present study, several concurrent mutations were detected in patients, but no correlation was found with clinical/pathological characteristics or patient prognosis.

Conclusions
In conclusion, molecular characterization of cancer cells can enhance our understanding of the biological progression of this disease [76,77].The findings of this study suggest that mutations are promising prognostic biomarkers.As personalized medicine has become the primary mode of therapy, knowledge of the precise mutation status of patients with CRC can lead to better therapeutic choices.However, further research is necessary with a larger patient cohort and international collaborations to confirm the correlation between patients' molecular profiles, clinicopathological and epidemiological characteristics, and outcomes.Such research is expected to contribute to more precise clinical decision-making, personalized and improved care, and reduced toxicity of treatment, costs to patients, and burden on health systems.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers15194819/s1,Table S1: Raw patient data; Table S2: Gene panel for colorectal cancer (CRC) based on Kyoto Encyclopedia of Genes and Genomes (KEGG); Table S3: Signaling pathways and the associated genes for colorectal cancer (CRC) based on Kyoto Encyclopedia of Genes and Genomes (KEGG); Table S4: Detection of genes and position in chromosomes for tumor samples.
Funding: This research was partially funded by the Hellenic Society of Medical Oncology (HeSMO) and the Gastrointestinal Cancer Study Group (GIC-SG).

Institutional Review Board Statement:
This study was approved by the Ethics Committee/ Institutional Review Board of the University Hospital of Heraklion (Number 7302/ 19-8-2009).All procedures were performed in accordance with the ethical standards of the institutional and/or national research committee and the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.All patients signed a written informed consent form for participation.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: All relevant data are within the paper and its Supplementary Information files.Supplementary Table S1: Raw patient data.

Acknowledgments:
The authors would like to thank NATERA Inc., San Carlos, CA 94070, USA, for providing whole exome sequencing analysis and sharing the raw data for further bioinformatic analysis.

Figure 1 .
Figure 1.Overview of the next generation sequencing (NGS) analysis pipeline.

[ 20 ]
. CMS1 (immune) is characterized by microsatellite instability (MSI), high levels of mutations in CpG island methylator phenotype (CIMP) and BRAF genes, and a low prevalence of SNCA gene mutations.It is associated with lymphocyte infiltration and immune activation, along with prominent hypermethylation and reduced signaling through the WNT pathway [21,22].CMS2 (canonical) demonstrates epithelial characteristics and is marked by high chromosomal instability, an elevated count of somatic copy number alterations, as well as mutations in the WNT and MYC genes, leading to heightened activity in these intracellular signaling pathways [21,22].CMS3 (metabolic) exhibits a distinctive global genomic and epigenomic profile with mixed features, including metabolic reprogramming and dysregulated pathways.It displays increased activity in glutaminolysis and lipidogenesis, enriched with KRAS-activating mutations.CMS3 presents a moderate or low mixed state of MSI, intermediate CIMP, and moderate activation of WNT and MYC signaling.Additionally, it is characterized by PIK3CA mutations and IGBP3 overexpression but lacks BRAF mutations [21,22].CMS4 (mesenchymal) is typified by positive gene regulation and the overexpression of proteins involved in stromal infiltration, mesenchymal activation, extracellular

Figure 1 .
Figure 1.Overview of the next generation sequencing (NGS) analysis pipeline.

Figure 2 .
Figure 2. Frequency of mutations in each gene located in exons and splicing sites.Figure 2. Frequency of mutations in each gene located in exons and splicing sites.

Figure 2 .
Figure 2. Frequency of mutations in each gene located in exons and splicing sites.Figure 2. Frequency of mutations in each gene located in exons and splicing sites.

Table 2 .
VDR and TLR gene polymorphism detection.

Table 3 .
Frequency of mutated patients in each gene.

Table 4 .
Frequency of mutations according to patients' characteristics.

Table 5 .
Correlation of mutated signaling pathways and patients characteristics (X=where correlation was observed.

Table 6 .
Univariate and multivariate Cox regression analysis.