To Register      SMDM Homepage

Tuesday, 19 October 2004

This presentation is part of: Poster Session - Clinical Strategies; Judgment and Decison Making

DIAGNOSTIC PROCEDURES FOR DETECTING PANCREATIC CANCER AND ASSESSING RESECTABILITY: A CLINICAL DECISION ANALYSIS

Tania Schink, MSc1, Michael Böhmig, MD2, Chin Hur, MD, MPH3, Inga Koch, MD2, Stefan Rosewicz, Prof.2, Klaus-Dieter Wernecke, Prof.1, and Uwe Siebert, MD, MPH, MSc3. (1) Charité, Humboldt University Berlin, Department of Medical Biometry, Berlin, Germany, (2) Charité, Humboldt University Berlin, Medical Department, Division of Hepatology and Gastroenterology, Berlin, Germany, (3) Massachusetts General Hospital, Harvard Medical School, Institute for Technology Assessment and Department of Radiology, Boston, MA

Purpose: To assess different strategies for diagnosis and determining resectability in patients with suspected pancreatic cancer (PC).

Methods: We used the data from a prospective study of 193 patients with suspected PC who were enrolled at the Charité University Hospital from 08/1999-11/2000. These patients underwent each of the following six different diagnostic procedures to determine diagnosis of pancreatic cancer and assessment of resectability: ultrasound (US), magnetic resonance imaging (MR), computed tomography (CT), endoscopic ultrasound (EUS), fluorodeoxyglucose positron emissiontomography (PET), and endoscopic retrograde cholangio-pancreaticography (ERCP). We developed a decision tree to predict diagnostic accuracy and resectability using the conditional probabilities derived from our study. We evaluated 41 strategies comprising strategies based on single tests and combinations of two tests. As both correct diagnosis and resectability assessment are crucial for an adequate treatment plan, we sought to maximize the number of patients having both the diagnostic and resectability state correctly classified. In the base case analysis, we used the point estimates of the conditional probabilities to perform a cohort simulation. We evaluated the impact of a reduced initial prevalence in a one-way sensitivity analysis. As numbers in some sub-branches were small, we also explored the degree of uncertainity of our result by means of probabilistic sensitivity analysis using Monte-Carlo-simulation.

Results: In the base-case analysis, the best test performance was achieved with MR alone, classifying 79% of patients correctly in benign, malignant/resectable, and malignant/unresectable. The combination of PET and EUS had the worst performance with only 61% of patients correctly classified. The choice of the best strategy depends on prevalence. If the prevalence is less than 23%, US followed by CT has the highest expected value of correctly classified patients. For a prevalence between 23% and 53%, MR followed by ERCP is optimal. Probabilistic sensitivity analysis showed robust results. In a MC-simulation with 10,000 samples, the probability of beeing the optimal strategy was 57% for MR alone (man accuracy 78%, standard deviation 3.1%) and 24% for MR followed y ERCP (mean accuracy 76%, standard deviation 3.3%)

Conclusion: Among the evaluated strategies, we suggest performing MR alone as first choice for patients with suspected pancreatic cancer in the setting of a specialized medical center with high prevalence of PC. Our model will be expanded to perform a cost-effectiveness analysis.


See more of Poster Session - Clinical Strategies; Judgment and Decison Making
See more of The 26th Annual Meeting of the Society for Medical Decision Making (October 17-20, 2004)