Diagnostic Cost Reduction Using Artificial Neural Networks

Introduction

Medical and surgical patients today face a variety of conditions that are both difficult and costly to diagnose and treat. With ever skyrocketing medical costs (Benko, 2004), the use of information technology is seen as a much-needed means to help control and potentially reduce medical direct costs (Intille, 2004). Deep vein thrombosis (DVT) and pulmonary embolism (PE) are medical conditions that are particularly difficult to diagnose in the acute setting (Mountain, 2003). Frequent usage of costly clinical laboratory tests to screen patients for further treatment is commonplace. All too commonly hospitals provide treatment to patients without PE as a preventative measure (Mountain, 2003). Furthermore, patient mortality, morbidity, and both direct and indirect costs for delayed diagnosis of these conditions may also be substantial. Recent studies show that 40 to 80 percent of patients that die from a PE are undiagnosed as having a potential PE (Mesquita et al., 1999; Morpurgo et al., 1998).

DVT may occur as the result of patient genetic and environmental factors or as a side effect of lower extremity immobility (e.g., following surgery). When a blood clot in the veins of a lower extremity breaks away, it may travel to the lungs and lodge in the pulmonary arterial circulation causing PE. If the clot is large enough it may wedge itself into the large pulmonary arteries leading to an acute medical emergency with a significant mortality rate. Approximately 2 million people annually experience DVT, with approximately 600,000 developing a PE and approximately 10% of these PE episodes result in mortality (Mesquita et al. 1999; Labarere et al., 2004). Documented occurrence of DVT in postoperative surgical populations ranges from 10% (Hardwick & Colwell, 2004) to 28% (Blattler et al., 2004).

Direct costs associated with DVT and PE come from the expensive diagnostic and even more expensive treatment protocols. It may be possible to lower these direct costs, especially when additional testing or treatment may be ruled out due to available knowledge. Artificial neural network (ANN) systems enable the economic examination (Walczak, 2001) and nonlinear combination of various readily available clinical laboratory tests. Laboratory tests typically performed on surgical patients, for example, blood chemistry, form the foundation for analysis and diagnostic model development.