Using Biochemical Indexes to Prognose Paraquat-Poisoned Patients: An Extreme Learning Machine-Based Approach

Abstract

The biochemical indexes are used to assess the hepatic and renal function of paraquat (PQ) poisoning patients. However, these indexes correlated with the prognosis of patients are unidentified. This paper aims to explore useful indexes from biochemical tests and to identify their predictive value. A total of 101 PQ poisoning patients including 51 dead patients and 50 survived patients is involved in this study. The biochemical indexes of PQ poisoning patients in different status are collected and analyzed by the independent-sample test. After that, Fisher scores feature selection is applied to screen prognostic factors from biochemical indexes. Based on the results of Fisher selection, an effective extreme learning machine (ELM) is applied to diagnose the prognosis status of PQ poisoning patients. The created ELM method is rigorously evaluated for accuracy, sensitivity, and specificity. The results show that there is statistical significance between dead and survived people in biochemical indexes (P<;0.01). Feature selection revealed that direct bilirubin, alanine aminotransferase (ALT), total bilirubin, aspartate aminotransferase (AST), the ratio of ALT/AST, and creatinine are the most crucial indexes, which correlated with the prognosis of PQ poisoning. The maximal classification accuracy is 79.6% when these six indexes are selected as the dataset. In conclusion, the biochemical test is related to the prognosis of PQ poisoning patients. It provides a new method for prognosis of PQ poisoning with feature selection and ELM model.

Keywords

• Extreme learning machine
• biochemical index
• paraquat
• medical prognosis
• patients