Development of a metabolites risk score for one-year mortality risk prediction in pancreatic adenocarcinoma patients
Metrics: PDF 911 views | HTML 868 views | ?
Andrea Fontana1, Massimiliano Copetti1,*, Iole Maria Di Gangi2,*, Tommaso Mazza3, Francesca Tavano4, Domenica Gioffreda4, Fulvio Mattivi2, Angelo Andriulli4, Urska Vrhovsek2, Valerio Pazienza4
1Unit of Biostatistics I.R.C.C.S. “Casa Sollievo della Sofferenza” Hospital, San Giovanni Rotondo (FG), Italy
2Department of Food Quality and Nutrition, Research and Innovation Centre, Fondazione Edmund Mach (FEM), San Michele all’Adige, Italy
3Unit of Bioinformatics, I.R.C.C.S. “Casa Sollievo della Sofferenza” Hospital, San Giovanni Rotondo (FG), Italy
4Gastroenterology Unit, I.R.C.C.S. “Casa Sollievo della Sofferenza” Hospital, San Giovanni Rotondo (FG), Italy
*These authors contributed equally to this work
Valerio Pazienza, e-mail: firstname.lastname@example.org
Keywords: pancreatic cancer, multiple risk score
Received: July 27, 2015 Accepted: November 29, 2015 Published: February 01, 2016
Purpose: Survival among patients with adenocarcinoma pancreatic cancer (PDCA) is highly variable, which ranges from 0% to 20% at 5 years. Such a wide range is due to tumor size and stage, as well other patients’ characteristics. We analyzed alterations in the metabolomic profile, of PDCA patients, which are potentially predictive of patient’s one-year mortality.
Experimental design: A targeted metabolomic assay was conducted on serum samples of patients diagnosed with pancreatic cancer. Statistical analyses were performed only for those 27 patients with information on vital status at follow-up and baseline clinical features. Random Forest analysis was performed to identify all metabolites and clinical variables with the best capability to predict patient’s mortality risk at one year. Regression coefficients were estimated from multivariable Weibull survival model, which included the most associated metabolites. Such coefficients were used as weights to build a metabolite risk score (MRS) which ranged from 0 (lowest mortality risk) to 1 (highest mortality risk). The stability of these weights were evaluated performing 10,000 bootstrap resamplings.
Results: MRS was built as a weighted linear combination of the following five metabolites: Valine (HR = 0.62, 95%CI: 0.11–1.71 for each standard deviation (SD) of 98.57), Sphingomyeline C24:1 (HR = 2.66, 95%CI: 1.30–21.09, for each SD of 20.67), Lysine (HR = 0.36, 95%CI: 0.03–0.77, for each SD of 51.73), Tripentadecanoate TG15 (HR = 0.25, 95%CI: 0.01–0.82, for each SD of 2.88) and Symmetric dimethylarginine (HR = 2.24, 95%CI: 1.28–103.08, for each SD of 0.62), achieving a very high discrimination ability (survival c-statistic of 0.855, 95%CI: 0.816–0.894). Such association was still present even after adjusting for the most associated clinical variables (confounders).
Conclusions: The mass spectrometry-based metabolomic profiling of serum represents a valid tool for discovering novel candidate biomarkers with prognostic ability to predict one-year mortality risk in patients with pancreatic adenocarcinoma.
All site content, except where otherwise noted, is licensed under a Creative Commons Attribution 3.0 License.