Microalbuminuria identifies overall cardiovascular risk in essential hypertension: an artificial neural network-based approach. Giovanna Leoncini a , Giorgio Sacchi b , Francesca Viazzi a , Maura Ravera a , Denise Parodi a , Elena Ratto a , Simone Vettoretti a , Cinzia Tomolillo a , Giacomo Deferrari a and Roberto Pontremoli a Background Ultrasound (US) examination of heart and carotid arteries provides an accurate assessment of target organ damage (TOD) and may influence the stratification of the absolute cardiovascular risk profile. Microalbuminuria has recently proved to be a useful cost- effective marker of increased cardiovascular risk but is still too often neglected in clinical practice. Objective To evaluate how well artificial neural networks (ANNs) predict cardiovascular risk stratification by means of routine data and urinary albumin excretion, as compared to prediction by the clinical work-up suggested by the International Society of Hypertension (ISH), with and without ultrasound-determined TOD. Methods A group of 346 never previously treated essential hypertensives (212 men, 134 women, mean age 47 6 9 years) was studied. Risk was stratified according to the criteria suggested by the 1999 WHO/ISH guidelines; first, by routine procedures alone, and subsequently by reassessment, using data on cardiac and vascular structures obtained by US evaluation. The ANN was trained and tested to predict the overall cardiovascular risk on the basis of routine clinical data and urinary albumin excretion (UAE). The impact of these three approaches on the determination of cardiovascular risk profile was evaluated. Results According to the first classification, 5.5% (n 19) of patients were considered at low risk, 47.3% (n 164) at medium, 26.7% (n 92) at high and 20.6% (n 71) at very high risk. A marked change in risk stratification, namely an increase in the prevalence of high- and very-high-risk patients (2.3% low, 29.8% medium, 42.8% high and 25.2% very high risk; ÷ 2 15.201, P < 0.0001), was obtained when US examination of TOD was taken into consideration. On the basis of routine clinical data and UAE, the artificial neural network successfully predicted overall cardiovascular risk and allocated patients in different classes as accurately as the US-based evaluation. Conclusions The use of US techniques allows a more precise stratification of absolute cardiovascular risk in hypertensive patients as compared to routine clinical data. An ANN can accurately identify the patients’ risk status by using low-cost routine data and UAE. These results further emphasize the value of UAE in the stratification of cardiovascular risk. J Hypertens 20:1315–1321 & 2002 Lippincott Williams & Wilkins. Journal of Hypertension 2002, 20:1315–1321 Keywords: microalbuminuria, artificial neural network, target organ damage, essential hypertension, cardiovascular risk a Department of Internal Medicine and b Department of Experimental Medicine, University of Genoa, Italy. Sponsorship: This work was supported in part by grant number RF99.52 from the Ministero della Sanita ` , Ricerca Finalizzata 1999. Correspondence and requests for reprints to Roberto Pontremoli MD, Department of Internal Medicine, University of Genoa, Viale Benedetto XV, 6 –16132 Genoa, Italy. Fax: +39 02 700 509 259; e-mail: rpontrem@medicina.unige.it Received 6 July 2001 Revised 28 January 2002 Accepted 6 February 2002 See editorial commentary page 1255 Introduction Therapeutic strategies in patients with essential hyper- tension must not be based solely on the severity of blood pressure levels. As a matter of fact, recent guide- lines issued by several national and international agen- cies agree that absolute cardiovascular risk, rather than the severity of hypertension, should guide both the decision to begin treatment as well as the identification of target pressure levels in each patient [1,2]. Non-invasive assessment of target organ damage (TOD) by ultrasound (US) techniques is a reliable way to assess the cardiac and peripheral arterial structures, and has proved to be useful in the evalua- tion of the overall cardiovascular risk profile [3]. While this procedure may have a significant impact on the evaluation of absolute risk, it is not routinely recom- mended by international guidelines, and is only per- formed on a small number of hypertensive patients. An overly restrictive diagnostic approach to risk strati- fication could lead to significant misclassification of patients and to underestimation of the actual absolute Original article 1315 0263-6352 & 2002 Lippincott Williams & Wilkins