Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis

Alexander Stojadinovic, Aviram Nissan, John Eberhardt, Terence C. Chua, Joerg O W Pelz, Jesus Esquivel

Research output: Contribution to journalArticle

Abstract

Multimodality therapy in selected patients with peritoneal carcinomatosis is gaining acceptance. Treatment-directing decision support tools are needed to individualize care and select patients best suited for cytoreductive surgery ± hyperthermic intraperitoneal chemotherapy (CRS ± HIPEC). The purpose of this study is to develop a predictive model that could support surgical decisions in patients with colon carcinomatosis. Fifty-three patients were enrolled in a prospective study collecting 31 clinical-pathological, treatment-related, and outcome data. The population was characterized by disease presentation, performance status, extent of peritoneal cancer (Peritoneal Cancer Index, PCI), primary tumor histology, and nodal staging. These preoperative parameters were analyzed using step-wise machine-learned Bayesian Belief Networks (BBN) to develop a predictive model for overall survival (OS) in patients considered for CRS ± HIPEC. Area-under-the-curve from receiver-operating-characteristics curves of OS predictions was calculated to determine the model's positive and negative predictive value. Model structure defined three predictors of OS: severity of symptoms (performance status), PCI, and ability to undergo CRS ± HIPEC. Patients with PCI <10, resectable disease, and excellent performance status who underwent CRS ± HIPEC had 89 per cent probability of survival compared with 4 per cent for those with poor performance status, PCI > 20, who were not considered surgical candidates. Cross validation of the BBN model robustly classified OS (area-under-the-curve = 0.71). The model's positive predictive value and negative predictive value are 63.3 per cent and 68.3 per cent, respectively. This exploratory study supports the utility of Bayesian classification for developing decision support tools, which assess case-specific relative risk for a given patient for oncological outcomes based on clinically relevant classifiers of survival. Further prospective studies to validate the BBN model-derived prognostic assessment tool are warranted.

Original languageEnglish (US)
Pages (from-to)221-230
Number of pages10
JournalAmerican Surgeon
Volume77
Issue number2
StatePublished - Feb 2011

Fingerprint

Colon
Carcinoma
Survival
Neoplasms
Drug Therapy
Area Under Curve
Prospective Studies
ROC Curve
Histology
Patient Care
Therapeutics
Population

ASJC Scopus subject areas

  • Surgery

Cite this

Stojadinovic, A., Nissan, A., Eberhardt, J., Chua, T. C., Pelz, J. O. W., & Esquivel, J. (2011). Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis. American Surgeon, 77(2), 221-230.

Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis. / Stojadinovic, Alexander; Nissan, Aviram; Eberhardt, John; Chua, Terence C.; Pelz, Joerg O W; Esquivel, Jesus.

In: American Surgeon, Vol. 77, No. 2, 02.2011, p. 221-230.

Research output: Contribution to journalArticle

Stojadinovic, A, Nissan, A, Eberhardt, J, Chua, TC, Pelz, JOW & Esquivel, J 2011, 'Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis', American Surgeon, vol. 77, no. 2, pp. 221-230.
Stojadinovic A, Nissan A, Eberhardt J, Chua TC, Pelz JOW, Esquivel J. Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis. American Surgeon. 2011 Feb;77(2):221-230.
Stojadinovic, Alexander ; Nissan, Aviram ; Eberhardt, John ; Chua, Terence C. ; Pelz, Joerg O W ; Esquivel, Jesus. / Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis. In: American Surgeon. 2011 ; Vol. 77, No. 2. pp. 221-230.
@article{d94ae6c2f8a6432da736076f72257843,
title = "Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis",
abstract = "Multimodality therapy in selected patients with peritoneal carcinomatosis is gaining acceptance. Treatment-directing decision support tools are needed to individualize care and select patients best suited for cytoreductive surgery ± hyperthermic intraperitoneal chemotherapy (CRS ± HIPEC). The purpose of this study is to develop a predictive model that could support surgical decisions in patients with colon carcinomatosis. Fifty-three patients were enrolled in a prospective study collecting 31 clinical-pathological, treatment-related, and outcome data. The population was characterized by disease presentation, performance status, extent of peritoneal cancer (Peritoneal Cancer Index, PCI), primary tumor histology, and nodal staging. These preoperative parameters were analyzed using step-wise machine-learned Bayesian Belief Networks (BBN) to develop a predictive model for overall survival (OS) in patients considered for CRS ± HIPEC. Area-under-the-curve from receiver-operating-characteristics curves of OS predictions was calculated to determine the model's positive and negative predictive value. Model structure defined three predictors of OS: severity of symptoms (performance status), PCI, and ability to undergo CRS ± HIPEC. Patients with PCI <10, resectable disease, and excellent performance status who underwent CRS ± HIPEC had 89 per cent probability of survival compared with 4 per cent for those with poor performance status, PCI > 20, who were not considered surgical candidates. Cross validation of the BBN model robustly classified OS (area-under-the-curve = 0.71). The model's positive predictive value and negative predictive value are 63.3 per cent and 68.3 per cent, respectively. This exploratory study supports the utility of Bayesian classification for developing decision support tools, which assess case-specific relative risk for a given patient for oncological outcomes based on clinically relevant classifiers of survival. Further prospective studies to validate the BBN model-derived prognostic assessment tool are warranted.",
author = "Alexander Stojadinovic and Aviram Nissan and John Eberhardt and Chua, {Terence C.} and Pelz, {Joerg O W} and Jesus Esquivel",
year = "2011",
month = "2",
language = "English (US)",
volume = "77",
pages = "221--230",
journal = "American Surgeon",
issn = "0003-1348",
publisher = "Southeastern Surgical Congress",
number = "2",

}

TY - JOUR

T1 - Development of a Bayesian belief network model for personalized prognostic risk assessment in colon carcinomatosis

AU - Stojadinovic, Alexander

AU - Nissan, Aviram

AU - Eberhardt, John

AU - Chua, Terence C.

AU - Pelz, Joerg O W

AU - Esquivel, Jesus

PY - 2011/2

Y1 - 2011/2

N2 - Multimodality therapy in selected patients with peritoneal carcinomatosis is gaining acceptance. Treatment-directing decision support tools are needed to individualize care and select patients best suited for cytoreductive surgery ± hyperthermic intraperitoneal chemotherapy (CRS ± HIPEC). The purpose of this study is to develop a predictive model that could support surgical decisions in patients with colon carcinomatosis. Fifty-three patients were enrolled in a prospective study collecting 31 clinical-pathological, treatment-related, and outcome data. The population was characterized by disease presentation, performance status, extent of peritoneal cancer (Peritoneal Cancer Index, PCI), primary tumor histology, and nodal staging. These preoperative parameters were analyzed using step-wise machine-learned Bayesian Belief Networks (BBN) to develop a predictive model for overall survival (OS) in patients considered for CRS ± HIPEC. Area-under-the-curve from receiver-operating-characteristics curves of OS predictions was calculated to determine the model's positive and negative predictive value. Model structure defined three predictors of OS: severity of symptoms (performance status), PCI, and ability to undergo CRS ± HIPEC. Patients with PCI <10, resectable disease, and excellent performance status who underwent CRS ± HIPEC had 89 per cent probability of survival compared with 4 per cent for those with poor performance status, PCI > 20, who were not considered surgical candidates. Cross validation of the BBN model robustly classified OS (area-under-the-curve = 0.71). The model's positive predictive value and negative predictive value are 63.3 per cent and 68.3 per cent, respectively. This exploratory study supports the utility of Bayesian classification for developing decision support tools, which assess case-specific relative risk for a given patient for oncological outcomes based on clinically relevant classifiers of survival. Further prospective studies to validate the BBN model-derived prognostic assessment tool are warranted.

AB - Multimodality therapy in selected patients with peritoneal carcinomatosis is gaining acceptance. Treatment-directing decision support tools are needed to individualize care and select patients best suited for cytoreductive surgery ± hyperthermic intraperitoneal chemotherapy (CRS ± HIPEC). The purpose of this study is to develop a predictive model that could support surgical decisions in patients with colon carcinomatosis. Fifty-three patients were enrolled in a prospective study collecting 31 clinical-pathological, treatment-related, and outcome data. The population was characterized by disease presentation, performance status, extent of peritoneal cancer (Peritoneal Cancer Index, PCI), primary tumor histology, and nodal staging. These preoperative parameters were analyzed using step-wise machine-learned Bayesian Belief Networks (BBN) to develop a predictive model for overall survival (OS) in patients considered for CRS ± HIPEC. Area-under-the-curve from receiver-operating-characteristics curves of OS predictions was calculated to determine the model's positive and negative predictive value. Model structure defined three predictors of OS: severity of symptoms (performance status), PCI, and ability to undergo CRS ± HIPEC. Patients with PCI <10, resectable disease, and excellent performance status who underwent CRS ± HIPEC had 89 per cent probability of survival compared with 4 per cent for those with poor performance status, PCI > 20, who were not considered surgical candidates. Cross validation of the BBN model robustly classified OS (area-under-the-curve = 0.71). The model's positive predictive value and negative predictive value are 63.3 per cent and 68.3 per cent, respectively. This exploratory study supports the utility of Bayesian classification for developing decision support tools, which assess case-specific relative risk for a given patient for oncological outcomes based on clinically relevant classifiers of survival. Further prospective studies to validate the BBN model-derived prognostic assessment tool are warranted.

UR - http://www.scopus.com/inward/record.url?scp=79952407655&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79952407655&partnerID=8YFLogxK

M3 - Article

C2 - 21337884

AN - SCOPUS:79952407655

VL - 77

SP - 221

EP - 230

JO - American Surgeon

JF - American Surgeon

SN - 0003-1348

IS - 2

ER -