Viazzi Francesca, Leoncini Giovanna, Sacchi Giorgio, Parodi Denise, Ratto Elena, Falqui Valeria, Parodi Angelica, Vaccaro Valentina, Tomolillo Cinzia, Deferrari Giacomo, Pontremoli Roberto
Department of Internal Medicine and Cardionephrology, Azienda Universitaria Ospedale San Martino, Italy.
J Hypertens. 2006 Jul;24(7):1281-6. doi: 10.1097/01.hjh.0000234107.08368.e5.
A slight reduction in estimated creatinine clearance is a predictor of unfavorable outcome in patients with primary hypertension. We evaluated how well an artificial neural network (ANN) can assess cardiovascular risk profile on the basis of estimated creatinine clearance and routine, low-cost clinical data, as compared with thorough clinical work-up, which includes an accurate assessment of target organ damage.
A group of 404 untreated patients with essential hypertension (250 men, 154 women; mean age, 47 +/- 9 years) were studied. We compared two different approaches that can be used to allocate patients into different risk classes according to the European Society of Hypertension-European Society of Cardiology guidelines: thorough clinical work-up, including cardiac and vascular ultrasound scan and microalbuminuria; and prediction by an ANN on the basis of estimated creatinine clearance and routine clinical data.
Thorough evaluation, as recommended by the guidelines, showed that 6% (n = 24) of our patients were at low risk, 20% (n = 81) were at medium risk, 45% (n = 182) were at high risk, and 29% (n = 117) were at very high risk. The ANN approach yielded almost superimposable results (sensitivity, 94%; positive predictive value, 96%; r = 0.95).
An ANN can accurately identify the patient's risk status using low-cost, clinical data and estimated creatinine clearance. These results emphasize the value of even a mild reduction in creatinine clearance for the stratification of cardiovascular risk in primary hypertension.
估算的肌酐清除率略有降低是原发性高血压患者不良预后的一个预测指标。我们评估了人工神经网络（ANN）基于估算的肌酐清除率和常规低成本临床数据评估心血管风险概况的能力，并与全面的临床检查（包括对靶器官损害的准确评估）进行比较。
研究了一组404例未经治疗的原发性高血压患者（250例男性，154例女性；平均年龄47±9岁）。我们比较了两种不同的方法，可根据欧洲高血压学会-欧洲心脏病学会指南将患者分为不同风险类别：全面的临床检查，包括心脏和血管超声扫描以及微量白蛋白尿；以及基于估算的肌酐清除率和常规临床数据的人工神经网络预测。
按照指南建议进行的全面评估显示，我们的患者中6%（n = 24）处于低风险，20%（n = 81）处于中风险，45%（n = 182）处于高风险，29%（n = 117）处于非常高风险。人工神经网络方法得出的结果几乎重叠（敏感性94%；阳性预测值96%；r = 0.95）。
人工神经网络可以使用低成本的临床数据和估算的肌酐清除率准确识别患者的风险状态。这些结果强调了即使肌酐清除率轻度降低对于原发性高血压心血管风险分层的价值。
