gms | German Medical Science

49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds)
19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI)
Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie
Schweizerische Gesellschaft für Medizinische Informatik (SGMI)

26. bis 30.09.2004, Innsbruck/Tirol

Decision Analysis on Diagnostic Procedures for Detecting Pancreatic Cancer and Assessing Resectability

Meeting Abstract (gmds2004)

  • corresponding author presenting/speaker Tania Schink - Institut für Medizinische Biometrie, Charité, Berlin, Deutschland
  • Michael Böhmig - Medizinische Klinik m. S. Hepatologie und Gastroenterologie, Charité, Berlin, Deutschland
  • Klaus-Dieter Wernecke - Institut für Medizinische Biometrie, Charité, Berlin, Deutschland
  • Uwe Siebert - Institute for Technology Assessment, Massachusetts General Hospital, Boston, USA

Kooperative Versorgung - Vernetzte Forschung - Ubiquitäre Information. 49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI) und Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI) der Österreichischen Computer Gesellschaft (OCG) und der Österreichischen Gesellschaft für Biomedizinische Technik (ÖGBMT). Innsbruck, 26.-30.09.2004. Düsseldorf, Köln: German Medical Science; 2004. Doc04gmds341

The electronic version of this article is the complete one and can be found online at:

Published: September 14, 2004

© 2004 Schink et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( You are free: to Share – to copy, distribute and transmit the work, provided the original author and source are credited.




In Germany appoximately 10,000 cases of pancreatic cancer (PC) are diagnosed each year [1]. Therapy and survival depends on the tumor resectability. Patiens with irresectable carcinoma receive palliative therapy and median survival is less than a year. Patients with resectable tumor undergo surgery which increases survival modestly. Therefore, it is not only neccessary to accurately diagnose PC but also to correctly assess resectability. Imaging technologies such as CT have a high sensitivity but a low specitity both in diagnosing PC and assessing resectability. This leads to the question whether the combination of imaging procedures can improve the assessment of disease status and resectability, and if so, which combination of diagnostic procedures optimizes diagnostic accuracy and assessment of resectability in patients with suspected PC.


We used data of a prospective diagnostic study with 195 patients with suspected PC enrolled from 08/1999-11/2001, who underwent six different diagnostic procedures: combined positron emission tomography (PET), computer tomography (CT), magnetic resonance imaging (MRI), ultrasound (US), endosonography (EUS), and endoscopic-retrograde-cholangiopancreaticography (ERCP).

We used the following criteria to reduce the number of possible strategies:

• We considered only single procedures that are able to assess resectability (CT, US, EUS and MR) and combinations of two procedures in which at least one procedure is able to assess resectability.

• Because tests had good sensitivity but only moderate specifity, we chose the believe-the-negative rule to combine discordant test results for diagnois of PC. For resectability, strategies for both believe-the-negative rule and believe-the-positive rule were evaluated in the decision tree.

• To reduce the absolute number of invasive procedures, we set the less invasive procedure as first one in any combination of two procedures. Invasivness increases in following order: US < MR < CT < PET < EUS < ERCP. Combinations of the last three procedures were not considered.

The remaining 22 strategies were evaluated in a clinical decision analysis. We developed a decision tree to predict diagnostic accuracy and resectability with respect to PC. Figure 1 [Fig. 1] displays the core structure of branches used in the decision tree.


The MR alone classified 79% of patients correctly in benign, malign but resectable, and malign and irresectable. The second best strategy is MR followed by ERCP with 78% correctly classified patients. The combinations of US and EUS (BTN and BTP rule) and US followed bei PET had the worst performances with an only 63% correctly classified patients. Figure 2 [Fig. 2] displays the accuracy of the 22 strategies entered into the decision analysis.


We were able to develop a decision tree comparing 22 different diagnostic strategies for detecting and assessing PC. We found only one similar study in the literature which was a decision-analytic cost-effectiveness analysis on assessing resectability in patients with PC [2]. Our study differs from this former study in two points. First, we evaluated the combined performance regarding diagnosis and resectability. Second, we did not assume conditional independence between diagnostic procedures, because we had empirical data of multiple procedures performed on the same patients. Nevertheless strategies including MR were among the most effective in both studies.

Our study has several limitations. First, our results are based on the assumption of unrestricted access to all included technologies, and therefore, may not be generalised to other settings. Second, we did not evaluate combinations of three or more diagnostic procedures and excluded some strategies, and thereby may have missed an accurate and efficient strategy. Third, a small fraction of images have been equivocal and were excluded from analysis. Forth, the goldstandard has not been assessed in 14 patients which may lead to verification bias. The fifth and most important limitation was that we did not distinguish between different types of incorrect assessments. For example, clinical consequences may differ between a patient who has resectable carcinoma and is classiefied as irresectable and a patient with benign tumor who is wrongly classiefied as resectable and undergoes surgery.

Therefore, we intend to extend this model in a future research step to allow for the evaluation of different consequences (e.g., mortality and quality of life) of inadequate treatment following imperfect classifications.

In conclusion, we suggest to perform only MR for patients with suspected pancreas cancer.


Robert Koch-Institut, Berlin,
McMahon et al. Pancreatic Cancer: Cost-Effectivebss of Imaging Technologies for Assessing Resectability. Radiology 2001, 221: 93-106