A Transcriptomic-Based Tool to Predict Gemcitabine Sensitivity in Advanced Pancreatic Adenocarcinoma.

Pancreatic ductal adenocarcinoma (PDAC) incidence has increased over the last 30 years.1Siegel R.L. et al.CA Cancer J Clin. 2020; 70: 7-30Crossref PubMed Scopus (12792) Google Scholar When diagnosed at advanced stages, representing approximately 85% of cases, systemic therapy is the only treatment able to improve patient outcomes. For patients with a good performance status, FOLFIRINOX is the preferred choice, but has a high level of toxicity.2Conroy T. et al.N Engl J Med. 2011; 369: 1817-1825Crossref Scopus (5177) Google Scholar For unfit patients, gemcitabine administrated alone or combined with nab-paclitaxel remains the standard treatment.3von Hoff D.D. et al.N Engl J Med. 2013; 369: 1691-1703Crossref PubMed Scopus (4178) Google Scholar,4Burris H.A. et al.J Clin Oncol. 1997; 15: 2403-2413Crossref PubMed Scopus (5171) Google Scholar Treatment choice is currently based on physician evaluation; using tumor molecular analysis to select the most effective and least toxic chemotherapy regimen would represent major progress. In recent years, we and others have described RNA signatures associated with gemcitabine sensitivity. Tiriac et al5Tiriac H. et al.Cancer Discov. 2018; 8: 1112-1129Crossref PubMed Scopus (485) Google Scholar found that RNA signatures derived from organoids could determine chemotherapy sensitivity. We reported GemPred, a gemcitabine RNA signature containing thousands of transcripts and validated in a retrospective cohort of 435 patients.6Nicolle R. et al.Ann Oncol. 2021; 32: 250-260Abstract Full Text Full Text PDF PubMed Scopus (34) Google Scholar As GemPred predictions were associated with the basal-like and classical PDAC subtypes that relate to patient prognosis, organoid models were included in the signature identification strategy. This allowed us to overcome the prognostic limitations of GemPred and generate an improved GemPred signature.7Nicolle R. et al.Transl Oncol. 2022; 16101315Crossref PubMed Scopus (3) Google Scholar Finally, using a strategy based on the selection of a reduced number of transcriptomic-concordant in vitro and in vivo PDAC models, we identified GemCore, a gemcitabine sensitivity signature that has the advantage of containing fewer than 100 transcripts and that has been validated in 2 clinical cohorts of 80 and 305 patients.8Fraunhoffer N.A. et al.Cancer Commun. 2022; 1–5Google Scholar As these signatures were all validated in retrospective cohorts of localized tumors on resected formalin-fixed, paraffin-embedded tissues samples, we decided to analyze their ability to predict gemcitabine sensitivity in advanced PDAC on formalin-fixed, paraffin-embedded microbiopsies from primary tumors and metastatic sites. One hundred and seven patients with advanced PDAC were retrospectively included from 3 hospitals. All patients were treated with gemcitabine as monotherapy in the first line. One hundred and one assessable samples were obtained from 93 patients before treatment (57 unpaired from primary tumors, 28 unpaired from metastatic sites and 16 paired samples). First, we analyzed primary tumors from 65 patients. Five patients (7.7%) had locally advanced disease and 60 (92.3%) had metastatic disease. Median overall survival (OS) was 5.7 months (95% CI, 4.62–8.52 months) and median progression-free survival (PFS) was 2.3 months (95% CI, 1.38–3.44 months). Gem-Tiriac et al, GemPred, and improved GemPred performed poorly in identifying gemcitabine sensitivity (Figure 1A–F). Improved GemPred revealed a significant association between PFS and gemcitabine sensitivity, with a hazard ratio (HR) of 0.57 (95% CI, 0.34–0.95; P = .032) (Figure 1F). Of all signatures, GemCore achieved the best performance, classifying 29 patients (44.6%) as GemCore+ and 36 (56.4%) as GemCore– (Figure 1G and H). GemCore+ patients displayed a median OS of 13.9 months (95% CI, 9.51–17.18 months) and a median PFS of 4.85 months (95% CI, 4.29–8.07 months). GemCore– patients had a median OS of 3.1 months (95% CI, 2.33–4.79 months) and a median PFS of 1.15 months (95% CI, 0.49–1.87 months). GemCore was also the only signature to show a significant association with objective response in primary tumors (Table 1 and Supplementary Table 1). In the univariate Cox model, GemCore+ patients showed an OS HR of 0.19 (95% CI, 0.10–0.34; P < .001) and a PFS HR of 0.12 (95% CI, 0.06–0.25; P < .001). When we contrasted the GemCore signature prediction with clinicopathological variables and transcriptomic RNA biomarkers, we found that 5 variables were statistically significant predictors of OS and PFS (P < .05) (Supplementary Table 2): World Health Organization performance status score ≥2, presence of hepatic metastasis, carbohydrate antigen 19-9 levels 59 times higher than the upper limit and poor differentiation were significant for both OS and PFS, whereas number of metastases was only significant for OS and weight loss only for PFS. GemCore was significantly associated with hepatic metastases and the degree of tumor differentiation (Table 1). Despite the observed enrichment of the GemCore stratification with the clinicopathological variables mentioned, GemCore+ remained a predictor of OS (HR, 0.18; 95% CI, 0.09–0.35; P < .001) and PFS (HR, 0.11; 95% CI, 0.04–0.26; P < .001) in a Cox multivariate model (Supplementary Table 2).Table 1Association of GemCore With Clinicopathological and Transcriptional Biomarkers: Samples From Primary Tumors and MetastasesVariableGemCoreP valuePositiveNegativePrimary tumors Age, n (%)1≤65 y6 (9.2)7 (10.8)>65 y23 (35.4)29 (44.6) Sex, n (%)1Female15 (23.1)18 (27.7)Male14 (21.5)18 (27.7) WHO PS score, n (%).436Unknown1 (1.5)0 (0.0)05 (7.7)3 (4.6)18 (12.3)14 (21.5)≥215 (23.1)19 (29.2) Clinical stage, n (%).649Locally advanced2 (3.1)3 (4.6)Metastatic34 (52.3)26 (40.0) Weight loss, kg, mean (SD)8.24 (6.65)8.03 (5.62)1 Primary tumor location, n (%).617Head12 (18.5)18 (27.7)Other17 (26.1)18 (27.7) Tumor thickness, mm, mean (SD)41.9 (18.0)39.3 (16.0).595 Pulmonary metastases, n (%)1No21 (32.3)26 (40.0)Yes8 (12.3)10 (15.4) Hepatic metastases, n (%).001No17 (26.1)7 (10.8)Yes12 (18.5)29 (44.6) No. of metastatic sites, n (%).064123 (35.4)20 (30.1)≥26 (9.2)16 (24.6) Level of CA19-9, n (%)1Unknown5 (7.7)10 (15.4)Normal4 (6.1)5 (7.7)<59× ULN18 (27.7)18 (27.7)>59× ULN2 (3.1)3 (4.6) Differentiation, n (%).005Unknown2 (3.1)2 (3.1)1 (well)25 (38.5)23 (35.8)2 (moderately)2 (3.1)2 (3.1)3 (poor)0 (0.0)9 (13.8) Objective responses, n (%).026Unknown3 (4.6)15 (23.1)Partial response6 (9.2)3 (4.6)Stable disease15 (23.1)6 (9.2)Progressive disease5 (7.7)12 (18.5) PurIST, n (%)1Basal-like2 (3.1)3 (4.6)Classical27 (41.5)33 (50.8) CDA, mean (SD)2.29 (2.40)2.70 (2.26).405 DCK, mean (SD)0.71 (1.45)0.61 (1.55).468 SLC29A1 (hENT1), mean (SD)2.58 (2.54)2.47 (2.79).967 hENT1/CDA ratio, mean (SD)2.00 (2.79)1.89 (3.71).708Metastases Age, n (%).434≤65 y3 (8.3)5 (13.8)>65 y16 (44.4)12 (33.3) Sex, n (%).335Female10 (27.8)6 (16.7)Male9 (25.0)11 (30.6) WHO PS score, n (%).714Unknown, n (%)1 (2.8)1 (2.8)01 (2.8)1 (2.8)16 (16.7)3 (8.3)≥211 (30.6)12 (33.3) Primary tumor location, n (%).516Head9 (25.0)6 (16.7)Other10 (27.8)11 (30.6) Tumor thickness, mm, mean (SD)33.9 (11.8)44.7 (12.4).017 No. of metastatic sites, n (%).048111 (30.6)4 (11.1)≥28 (22.2)13 (36.1) Level of CA19-9, n (%).555Unknown2 (5.6)7 (19.4)Normal2 (5.6)2 (5.6)<59× ULN13 (36.1)7 (19.4)>59× ULN1 (2.8)2 (5.6) Differentiation, n (%)1Unknown2 (5.6)3 (8.3)1 (well)9 (25.0)8 (22.2)2 (moderately)4 (11.1)3 (8.3)3 (poor)4 (11.1)3 (8.3) Objective responses, n (%).493Unknown6 (16.7)14 (38.9)Partial response2 (5.6)1 (2.8)Stable disease4 (11.1)0 (0.0)Progressive disease7 (19.4)2 (5.6) PurIST, n (%).167Basal-like1 (2.8)4 (11.1)Classical18 (50.0)13 (36.1) CDA, mean (SD)3.24 (2.62)4.55 (2.62).181 DCK, mean (SD)0.99 (1.10)0.86 (0.81).871 SLC29A1 (hENT1), mean (SD)3.83 (2.19)3.89 (2.37)1 hENT1/CDA ratio, mean (SD)2.18 (3.83)0.79 (0.43).484CA19–9, cancer antigen 19–9; CDA, cytidine deaminase; DCK, deoxycytidine kinase; hENT1, human equilibrative nucleoside transporter 1; PurIST, Purity Independent Subtyping of Tumours; SLC29A1, solute carrier family 29 member 1; ULN, upper limit of normal; WHO PS, World Health Organization performance status score. Open table in a new tab CA19–9, cancer antigen 19–9; CDA, cytidine deaminase; DCK, deoxycytidine kinase; hENT1, human equilibrative nucleoside transporter 1; PurIST, Purity Independent Subtyping of Tumours; SLC29A1, solute carrier family 29 member 1; ULN, upper limit of normal; WHO PS, World Health Organization performance status score. When possible, biopsies from metastatic sites are frequently used for diagnostic purposes. Therefore, we analyzed the 4 signatures in 36 biopsies from PDAC metastases. Median OS was 3.5 months (95% CI, 2.39–6.00 months) and median PFS was 1.15 months (95% CI, 0.66–2.39 months). As in primary tumors, GemCore was better able to stratify gemcitabine sensitivity in metastasis samples. GemCore+ patients (n = 19 [52.78%]) had a median OS of 6.6 months (95% CI, 4.72–16.13 months) and a median PFS of 2.95 months (95% CI, 1.38–4.36 months). GemCore– patients (n = 17 [47.22%]) displayed a median OS of 2.1 months (95% CI, 1.64–3.48 months) and a median PFS of 0.36 months (95% CI, 0.00–1.34 months). The univariate Cox model confirmed the predictive capability of GemCore to discriminate gemcitabine-sensitive patients. GemCore+ showed an HR of 0.14 (95% CI, 0.06–0.35; P < .001) for OS and 0.17 (95% CI, 0.07–0.42; P < .001) for PFS. Among the clinicopathological variables and transcriptomic RNA biomarkers, tumor thickness was the only variable to predict OS in a univariate Cox model (HR, 1.03; 95% CI, 1.00–1.06). In addition, tumor thickness was significantly lower in GemCore+ than in GemCore– patients (33.9 ± 11.8 vs 44 ± 12.4; P = .017) (Table 1). There was a significant association between GemCore– patients and the number of metastases being ≥2 (P = .048) (Table 1). Finally, our analysis of the paired primary tumor and metastasis samples revealed that the GemCore signature gave a matched prediction in 87.5% of cases (57% of samples were GemCore–, 43% were GemCore+). A weakness associated with drug-response RNA signatures is that they frequently capture the basal-like or classical transcriptomic landscape that is related to the patient's prognosis. However, GemCore did not correlate with any PDAC subtype and was the main OS and PFS predictor in the multivariate Cox analysis (Supplementary Table 2 and Table 1). These observations suggest that GemCore has a predictive, not prognostic, capacity. Gemcitabine is the main drug used in unfit patients with metastatic PDAC because it has reduced infusion times and fewer adverse effects than polychemotherapy regimens (ie, FOLFIRINOX and gemcitabine/nab-paclitaxel). To avoid any potential biases derived from a combined treatment, here we focused on patients treated with gemcitabine alone. However, further validation of GemCore is needed in patients treated with gemcitabine plus nab-paclitaxel to enlarge the scope of this signature. We noted that the median OS of GemCore+ patients with biopsied primary tumors was longer than that of those with biopsies from metastatic sites. Although GemCore was able to identify responders to gemcitabine in both, the difference in the median OS is suspected to be because of the small number of patients in the metastatic group and/or because the biopsies of metastatic tissue correspond to those patients with the most advanced disease; further validation on larger metastasis cohorts is needed to elucidate this discrepancy. Development of predictive signatures is challenging and in permanent evolution. These predictors depend on the technology used for RNA sequencing and even more on the site from which the biopsy is taken. In this work, we challenged in a multicentric cohort of advanced PDAC patients the GemCore signature alongside 3 other signatures previously validated for gemcitabine as adjuvant treatment for patients who have undergone surgery. GemCore represents the RNA-based signature best able to predict gemcitabine response not only in resected but also in advanced PDAC patients and in all types of samples (ie, resections or microbiopsies from primary tumors and metastatic sites). PDAC Chemo Sensitivity Prediction Working Group includes Martin Bigonnet,1 Claire Bongrain,2 Emilie Lermite,3 Patrick Pessaux,4 Fabio Giannone,4 Marie-Pierre Chenard,4 Sophie Michalak,5 Rémy Nicolle,6 Marion Rubis,2 Flora Poizat,7 Marc Giovannini,7 Fabrice Caillol,7 and Philippe Rochigneux7; from 1Predicting Med, Luminy Science and Technology Park, Marseille, France; 2Cancer Research Center of Marseille (CRCM), INSERM, CNRS, Institut Paoli-Calmettes, Aix-Marseille University, Marseille, France; 3Endocrine and Visceral Surgery Department, University Hospital Angers, Angers, France; 4Department of General, Digestive, and Endocrine Surgery, Nouvel Hôpital Civil, Strasbourg, France; 5Department of Pathology, University Hospital Angers, Angers, France; 6Université Paris Cité, Centre de Recherche sur l'Inflammation (CRI), INSERM, U1149, CNRS, ERL 8252, Paris, France; and 7Medical Oncology Department, Institut Paoli-Calmettes, Marseille, France. Nicolas Fraunhoffer, PhD (Data curation: Lead; Formal analysis: Lead; Investigation: Lead; Methodology: Lead; Software: Lead; Validation: Lead; Visualization: Lead; Writing – original draft: Lead). Brice Chanez, MD, PhD (Data curation: Lead; Formal analysis: Supporting; Investigation: Supporting; Methodology: Supporting; Resources: Lead; Validation: Lead; Visualization: Supporting; Writing – review & editing: Equal). Carlos Teyssedou, MD (Data curation: Lead; Resources: Lead) Martin Bigonnet, Mr (Methodology: Equal). Claire Bongrain, MD (Methodology: Supporting). Emilie Lermite, MD (Resources: Supporting). Patrick Pessaux, MD (Resources: Supporting). Fabio Giannone, MD (Resources: Supporting). Marie-Pierre Chenard, MD (Resources: Supporting). Sophie Michalak-Provost, MD (Resources: Supporting). Rémy Nicolle, PhD (Formal analysis: Supporting; Validation: Supporting). Marion Rubis, Ms (Methodology: Supporting). Flora Poizat, MD (Methodology: Supporting; Resources: Supporting). Marc Giovannini, MD (Resources: Supporting). Fabrice Caillol, MD (Resources: Supporting). Philippe Rochigneux, MD (Resources: Supporting). Juan L. Iovanna, MD, PhD (Conceptualization: Lead; Data curation: Lead; Formal analysis: Lead; Funding acquisition: Lead; Investigation: Lead; Supervision: Lead; Validation: Lead; Visualization: Lead; Writing – review & editing: Lead). Emmanuel Mitry, MD, PhD (Data curation: Supporting; Formal analysis: Supporting; Investigation: Supporting; Supervision: Supporting; Validation: Supporting; Writing – review & editing: Equal). Dusetti J. Nelson, PhD (Conceptualization: Lead; Data curation: Lead; Formal analysis: Lead; Funding acquisition: Lead; Investigation: Lead; Methodology: Lead; Supervision: Lead; Validation: Lead; Visualization: Lead; Writing – original draft: Lead; Writing – review & editing: Lead). This study retrospectively included patients from 3 hospitals using the following as inclusion criteria: confirmed diagnosis of PDAC at an advanced stage; treated with gemcitabine monotherapy in the first line; and tumor sample availability (formalin-fixed, paraffin embedded [FFPE] tissues). A total of 107 consecutive patients, diagnosed during 2010–2021, were included from 3 hospitals (95 from the Institut Paoli-Calmettes [IPC], 6 from the University Hospital Angers and 6 from the Nouvel Hôpital Civil, Strasbourg). Fourteen patients (13.1%) were excluded, as samples had poor RNA quality, leaving 93 assessable patients (101 samples). All samples were collected before any treatment. Samples were distributed as follows: 57 unpaired from primary tumors, 28 unpaired from metastatic sites, and 16 paired primary tumor and metastasis samples from 8 patients. The IPC patients were recorded using the ConSoRe (Continum Soin Recherche) clinical data mining interface, using the following keywords: “pancreas adenocarcinoma” as primary tumor and “gemcitabine” as chemotherapy regimen to identify consecutive patients. Patients from the University Hospital Angers and the Nouvel Hôpital Civil, Strasbourg were manually selected. The IPC Internal Review Board approved the study as no. IPC2021-070 (Gempred-Retro), and the research was conducted in accordance with the Helsinki Declaration. The consent forms of informed patients were collected according to ethics principles. Total RNA was extracted from FFPE tissue sections using the RNeasy FFPE kit (Qiagen, Hilden, North Rhine-Westphalia, Germany) following the manufacturer’s instructions. Briefly, the presence of neoplastic cells and the percentage of cellularity were evaluated by a pancreatic pathologist using H&E staining. Only nucleated cells were considered for the calculation of cellularity (red blood cells were not considered). From each FFPE block, between 4 and 5 sections of 10 μm were cut and manually macrodissected to enrich for neoplastic cells. Samples with neoplastic cellularity of >10% and that produced >30 ng of total RNA were used for transcriptomic analysis. The quality of FFPE-derived RNA was measured by the proportion of fragments above 200 base pairs (DV200) and ranged from 17% to 77% (mean 51.3%) assessed by use of the Agilent 2100 Bioanalyzer System. RNA libraries were prepared with the QuantSeq 3′ mRNA-Seq kit (Lexogen, Vienna, Austria) and run on the Illumina NovaSeq 6000 for 50 base pair single-end reads. The expression matrix was obtained using the Rsubread R package. Then, the RNA reads were normalized using trimmed mean of M-values and log2 transformed. Different gemcitabine RNA signatures were assessed using the parameters defined by the original description. For the Gem-Tiriac et al, signature, the transcripts positively correlated with sensitivity (r < –0.38) were selected to compute the mean z-score. GemPred and improved GemPred were computed using the web application (https://app.gebican.fr/pdac-gempred/). Finally, GemCore stratification was calculated through a binary classifier defined by logistic regression. OS was defined as the time from diagnosis to death. PFS was measured from the date of first gemcitabine injection to the time of disease progression or death. Objective responses were assessed by using the Response Evaluation Criteria in Solid Tumours, version 1.1 criteria. Survival curves were estimated using the Kaplan–Meier technique and compared with the log-rank test. Qualitative variables were compared with χ2 test or Fisher test, and quantitative variables with the use of Student t test or a nonparametric Wilcoxon test. Normality was tested with the Shapiro–Wilk test. For each test, statistical significance was set at a 2-sided P value of <.05. Univariate and multivariate Cox regression analyses and Kaplan–Meier curves were computed using the survival R package. Variable selection for the Cox multivariate analysis was performed applying Lasso regression with lambda cross-validation. Variables with non-zero coefficients were selected. The Cox proportional hazard regression model was used for univariate and multivariate analyses to estimate the hazard ratio with a 95% CI. Proportional hazards assumption was tested using the Schoenfeld residuals.Supplementary Table 1Association of Gem-Tiriac et al, GemPred, and Improved GemPred With Objective Responses: Samples From Primary Tumors and MetastasesVariableGem-Tiriac et al.GemPredImproved GemPredPositive, n (%)Negative, n (%)P valuePositive, n (%)Negative, n (%)P valuePositive, n (%)Negative, n (%)P valuePrimary tumors.747.084.662 Objective responsesPartial response3 (4.6)6 (9.2)3 (4.6)6 (9.2)6 (9.2)3 (4.6)Stable disease11 (16.9)10 (15.4)11 (16.9)10 (15.4)11 (16.9)10 (15.4)Progressive disease8 (12.3)9 (13.8)3 (4.6)14 (21.5)11 (16.9)6 (9.2)Unknown9 (13.8)9 (13.8)6 (9.2)12 (18.5)6 (9.2)12(18.5)Metastases.485.906.5134 Objective responsesPartial response1 (2.8)2 (5.6)1 (2.8)2 (5.6)3 (8.3)0 (0.0)Stable disease3 (8.3)1 (2.8)2 (5.6)2 (5.6)1 (2.8)3 (8.3)Progressive disease4 (11.1)5 (13.9)4 (11.1)5 (13.9)3 (8.3)6 (16.7)Unknown16 (44.4)4 (11.1)9 (25.0)11 (30.5)13 (36.1)7 (19.4)NOTE. The objective response was determined according to the Response Evaluation Criteria in Solid Tumours, version 1.1. Open table in a new tab Supplementary Table 2Univariate and Multivariate Cox Analysis of Clinicopathological Variables to Determine Associations With Overall Survival and Progression-Free Survival: Samples From Primary TumorsVariableDataUnivariateMultivariateHR (95% CI)P valueHR (95% CI)P valueOverall survival GemCore, n (%)—<.001GemCore–36 (55.4)—RefGemCore+29 (44.6)—0.18 (0.09––0.35) Age, n (%).743—≤65 y13 (20.0)Ref—>65 y52 (80.0)1.11 (0.60–2.06)— Sex, n (%).443—Female33 (50.8)Ref—Male32 (49.2)0.82 (0.50–1.36)— WHO PS score, n (%)Unknown1 (1.5)————08 (12.3)Ref—Ref—122 (33.8)1.74 (0.77–3.93).1841.62 (0.70–3.74).257≥234 (52.3)2.33 (1.05–5.15).03714.00 (1.65–118.67).006 Clinical stage, n (%).335—Locally advanced5 (7.7)Ref—Metastatic60 (92.3)1.57 (0.63–3.96)— Weight loss, kg, n (%)8.1 (6.0)1.04 (0.99–1.08).110—— Primary tumor location.196—Head30 (46.2)Ref—Other35 (53.8)1.40 (0.84––2.34)— Tumour thickness, mm, mean (SD)40.4 (16.8)1.00 (0.99–1.02).541— Pulmonary metastases, n (%).950No47 (71.9)Ref—Yes18 (28.1)1.02 (0.58–1.78)— Hepatic metastases, n (%).012.362No24 (36.9)RefRefYes41 (63.1)1.94 (1.16–3.26)1.31 (0.74–2.32) No. of metastatic sites, n (%).005.0120–143 (66.2)RefRef≥222 (33.8)2.18 (1.27–3.75)2.15 (1.18–3.89) Level of CA19-9, n (%)—Unknown15 (23.1)————Normal9 (13.8)Ref———<59× ULN36 (55.4)1.67 (0.79–3.53).179——>59× ULN5 (7.7)3.70 (1.18–11.63).025—— Differentiation, n (%)Unknown4 (6.1)————1 (well)48 (73.8)Ref———2 (moderately)4 (6.1)1.31 (0.47–3.69).605——3 (poor)9 (13.8)3.33 (1.58–7.03).002—— PurIST, n (%)Basal-like5 (7.7)Ref———Classical60 (92.3)0.85 (0.34–2.16).736—— CDA, mean (SD)2.5 (2.3)1.04 (0.93–1.16).499—— DCK, mean (SD)0.7 (1.5)1.07 (0.90–1.28).443—— SLC29A1 (hENT1), mean (SD)2.5 (2.7)0.99 (0.89–1.09).778—— hENT1/CDA ratio, mean (SD)1.9 (3.3)1.01 (0.91–1.13).846——Progression–free survival GemCore, n (%)—<.001GemCore–36 (55.4)—RefGemCore+29 (44.6)—0.11 (0.04-0.26) Age, n (%).9—≤65 y13 (20.0)Ref—>65 y52 (80.0)1.04 (0.56–1.93)— Sex, n (%).93.273Female33 (50.8)RefRefMale32 (49.2)0.98 (0.59–1.61)1.41 (0.76-2.62) WHO PS score, n (%)Unknown1 (1.5)————08 (12.3)Ref—Ref—122 (33.8)2.08 (0.88–4.92).12.64 (1.01-6.88).047≥234 (52.3)2.50 (1.10–5.70).033.81 (1.56-9.27).003 Clinical stage, n (%)——Locally advanced5 (7.7)Ref——Metastatic60 (92.3)1.25 (0.50–3.12).64— Weight loss, kg, mean (SD)8.1 (6.0)1.05 (1.00–1.11).03— Primary tumor location, n (%)Head30 (46.2)Ref———Other35 (53.8)0.70 (0.42–1.18).18—— Tumor thickness, mm, mean (SD)40.4 (16.8)1.00 (0.98–1.01).54—— Pulmonary metastases, n (%)No46 (71.9)Ref—Ref—Yes18 (28.1)0.84 (0.48–1.45).520.57 (0.28-1.15).116 Hepatic metastases, n (%).01—No24 (36.9)Ref—Yes41 (63.1)2.00 (1.19–3.35)— No. of metastatic sites, n (%).070.099143 (66.2)RefRef≥222 (33.8)1.63 (0.96–2.75)1.68 (0.91-3.10) Level of CA19-9, n (%)Unknown15 (23.1)————Normal9 (13.8)Ref———<59× ULN36 (55.4)1.57 (0.72–3.41).25——>59× ULN5 (7.7)3.22 (1.02–10.20).05—— Differentiation, n (%)Unknown4 (6.1)————1 (well)48 (73.8)Ref—Ref—2 (moderately)4 (6.1)1.90 (0.67–5.37).231.85 (0.60-5.70).2843 (poor)9 (13.8)5.16 (2.32–11.48)<.0012.39 (1.00-5.74).051 PurIST, n (%)Basal-like5 (7.7)Ref———Classical60 (92.3)0.79 (0.31–1.99).616—— CDA, mean (SD)2.5 (2.3)1.03 (0.93–1.14).618—— DCK, mean (SD)0.7 (1.5)1.03 (0.85–1.25).785—— SLC29A1 (hENT1), mean (SD)2.5 (2.7)1.06 (0.96–1.18).245—— hENT1/CDA ratio, mean (SD)1.9 (3.3)1.06 (0.96–1.16).281——CA19–9, cancer antigen 19–9; CDA, cytidine deaminase; DCK, deoxycytidine kinase; hENT1, human equilibrative nucleoside transporter 1; PurIST, Purity Independent Subtyping of Tumours; SLC29A1, solute carrier family 29 member 1; ULN, upper limit of normal; WHO PS, World Health Organization performance status. Open table in a new tab NOTE. The objective response was determined according to the Response Evaluation Criteria in Solid Tumours, version 1.1. CA19–9, cancer antigen 19–9; CDA, cytidine deaminase; DCK, deoxycytidine kinase; hENT1, human equilibrative nucleoside transporter 1; PurIST, Purity Independent Subtyping of Tumours; SLC29A1, solute carrier family 29 member 1; ULN, upper limit of normal; WHO PS, World Health Organization performance status.