Abstract: MATH/CHEM/COMP 2002, Dubrovnik, June 24-29, 2002

MACHINE LEARNING BASED ANALYSIS OF BIOCHEMICAL AND MORPHOLOGIC PARAMETERS IN PATIENTS WITH DIALYSIS RELATED AMYLOIDOSIS

Nikola stambuk¹, Igor Barisic², Vladimir Wilhem³, Stipan Jankovic²,

Pasko Konjevoda¹,  and Biserka Pokric¹

¹Rudjer Boskovic Institute,POB 180, HR-10002 Zagreb, Croatia

²Division of Radiology, Clinical Hospital Split, Soltanska 1, HR-21000 Split, Croatia

³Department of Nephrology, Clinical Hospital Split, Soltanska 1, HR-21000 Split, Croatia

Dialysis related amyloidosis is defined as an accumulation and deposition of β₂-microglobulin derived fibrils, especially in bones and joints, due to insufficient elimination during therapy. The syndrome has also been reported in patients with slowly progressive renal failure who had never been dialysed. The aim of this study was to analyse biochemical, morphologic and anamnestic parameters that may be relevant for the onset and developement of dialysis related amyloidosis. In addition to standard statistical procedures we also applied  the machine learning based methods of data mining to quantify the risk factors for asymptomatic patients. The extraction of risk factors for the onset of the dialysis related amyloidosis syndrome could enable us to predict the symptoms and consider medical procedures to prevent the onset of the disease. The C4.5 machine learning algorithm extracted simple and highly accurate tree for the discrimination of asymptomatic and symptomatic patients suffering from dialysis related amyloidosis. It remains an open question if our findings may contribute to the problem of accurately predicting the onset of dialysis related arthropathy in asymptomatic patients group.