The use of ridge regression for estimating the severity of acute pancreatitis

Abstract

Purpose. Increasing of treatment efficiency for patients with acute pancreatitis by improving objective means of determining the severity of acute pancreatitis.Materials and method. The study was based on a retrospective analysis of 130 cases of acute pancreatitis: 47 cases from «Krasnoyarsk Regional Clinical Hospital» and 83 cases from «Regional Interdistrict Clinical Hospital No 20 named after I.S. Berzon» in the period from 2015 to 2017. The raw data was pre-processed. In particular, different methods (median, linear regression) were used to fill the missing values in the observation matrix. The initial dataset contained features measured in various quantitative and categorical scales. For some features with a pronounced asymmetric distribution, a quantile transformation was applied to initial values. The quantile transformation allows features to be brought to a uniform distribution in order to reduce the risk of excluding significant features. Ridge regression was used in combination with an algorithm for sequential reduction of attribute space.Results. The classifier of three degrees of acute pancreatitis severity was developed. This classifier can help to determine better treatment tactics. During validation, the method of determining the severity of acute pancreatitis classification has proven to be effective. The average accuracy was 92% compared to the experts’ decisions. This procedure for constructing a classifier can be used as part of the basis to the medical decision support system.Conclusion. The results of this study will help to make the choice of a necessary starting therapy, assess the need for surgical intervention and in severe cases, prescribe enhanced antibacterial and detoxification therapy. This will predictably reduce the percentage of septic complications of acute pancreatitis, and consequently will reduce the frequency of fatal outcomes.

Keywords

• острый панкреатит
• степень тяжести
• классификатор
• категориальные признаки
• значимые показатели
• восстановление пропусков
• гребневая регрессия
• auc (area under curve)