Derivation and validation of a bayesian network to predict pretest probability of venous thromboembolism

Abstract

AMI) has been excluded and provocative cardiac testing obtained. An ACI-TIPI score was assigned on arrival to the chest pain center, the results of which were blinded to the treating physician. Subsequently admitted patients underwent medical record review for death, AMI, unstable angina, cardiac interventions, and final diagnosis. All patients received a telephone call at 30 days, medical records review, and social security death index review for follow-up. Institutional review board approval was obtained with waiver of informed consent.Results: One thousand seven hundred fifty-one consecutive patients with low to intermediate risk were evaluated from June 1, 2000, through March 22, 2001, with our chest pain protocol in the chest pain center. One thousand four hundred seventy-eight patients underwent stress testing and ACI-TIPI score assignment. Of the study group, 400 patients had an ACI-TIPI score below 20; 4 of these patients were found to have an acute coronary syndrome, which leads to a miss rate of 0.01 and a negative predictive value of 0.99. In the study cohort, 265 patients had an ACI-TIPI score below 20 and were men younger than 35 and women younger than 45 years. For men younger than 35 and women younger than 45 years of age and with an ACI-TIPI score below 20, the observed cardiac event rate was zero, resulting in a negative predictive value of 100%.Conclusion: The use of the ACI-TIPI score could be applied to further risk stratify patients safely following a chest pain protocol that rules out AMI. If all male patients younger than 35 years and all female patients younger than 45 years in whom AMI was excluded and the ACI-TIPI score was less than 20 had been sent home, 217 (15%) stress tests would have been safely avoided, with no observed adverse cardiac events.

A probabilistic network for the diagnosis of acute cardiopulmonary diseases

“…A second innovative aspect of our work is represented by the use of both expert knowledge and patient data to estimate the quantitative part of the probabilistic network. This approach is quite uncommon in the literature, where only one of the two sources of information is typically exploited (Andreassen et al, 1991;Charitos et al, 2009;Galán et al, 2002;Kline et al, 2005;Lacave & Díez, 2003;Leibovici et al, 2007;Luciani et al, 2007;Middleton et al, 1991;Nathwani et al, 1997;Suojanen et al, 1999;Van der Gaag et al, 2002). Applications where both expert knowledge and collected data are exploited include the expert systems DIAVAL (Díez et al, 1997) and HEPAR II (Wasyluk et al, 2001).…”

“…On the other hand, the quantitative part of probabilistic networks applied to medical problems is typically estimated by exploiting either expert knowledge (Andreassen, Hovorka, Benn, Olesen, & Carson, 1991;Charitos, Van der Gaag, Visscher, Schurink, & Lucas, 2009;Díez, Mira, Iturralde, & Zubillaga, 1997;Galán, Aguado, Díez, & Mira, 2002;Lacave & Díez, 2003;Leibovici, Paul, Nielsen, Tacconelli, & Andreassen, 2007;Luciani et al, 2007;Nathwani, et al, 1997;Suojanen, Andreassen, & Olesen, 1999;Van der Gaag, Renooij, Witteman, Aleman, & Taal, 2002), or a dataset of patient cases (Kline, Novobilski, Kabrhel, Richman, & Courtney, 2005;Middleton et al, 1991;Wasyluk, Onisko, & Druzdzel, 2001). In this paper, we illustrate the development of a probabilistic network for the diagnosis of acute cardiopulmonary diseases, where both beliefs from medical experts and clinical data were exploited to estimate the quantitative part.…”

Accuracy of Very Low Pretest Probability Estimates for Pulmonary Embolism Using the Method of Attribute Matching Compared with the Wells Score

“…These are different patients than are in the validation set in this report. 14,[25][26][27][28] Eight of the predictor variables were previously selected from over 25 candidate variables using a stepwise logistic regression technique and are used in the pulmonary embolism rule-out criteria (PERC) rule.…”