Risk assessment of alcohol withdrawal seizures with a Kohonen feature map

Recently, it has been suggested that alcohol-induced hyperhomocysteinaemia in patients suffering from chronic alcoholism might be a risk factor for alcohol withdrawal seizures. In the present follow-up study 12 patients with chronic alcoholism who suffered from withdrawal seizures had significantly higher levels of homocysteine (Hcy) on admission (71.43 ± 25.84 mol/l) than patients (n = 37) who did not develop seizures (32.60 ± 24.87 mol/l; U = 37.50, p = 0.0003). Using a logistic regression analysis, withdrawal seizures were best predicted by a high Hcy level on admission (p < 0.01; odds ratio 2.07). Based on these findings we developed an artificial neural network system (Kohonen feature map, KFM) for an improved prediction of the risk of alcohol withdrawal seizures. Forty-nine patients with chronic alcoholism (12 with alcohol withdrawal seizures and 37 without seizures) were randomized into a training set and a test set. Best results for sensitivity of the KFM was 83.3% (five of six seizure patients were predicted correctly) with a specificity of 94.4% (one false positive prediction of 19 patients). We conclude that in patients with alcohol-induced hyperhomocysteinaemia the KFM is a useful tool to predict alcohol withdrawal seizures.