Derivation of Decision Algorithms for the Diagnosis of Hepatic Encephalopathy in Patients With Cirrhosis

Abstract

Introduction: The presence of hepatic encephalopathy (HE), whether minimal or overt has a significant detrimental effect on patients’ quality of life, safety and survival. Current diagnostic tests are poorly validated or else are not generally applicable or practical for use in the clinical setting. This aim of this study was to identify simple, accurate algorithms for the diagnosis of HE in patients with cirrhosis that can be used in both clinical and research settings. Methods: Fifty-one patients with cirrhosis (30 men, 21 women; mean [range] age of 57.1 [24–81] years) were classified using clinical, psychometric and electroencephalographic variables, as either neuropsychiatrically unimpaired (n = 24), or as having minimal (n = 18) or overt (n = 9) HE. Patients’ neuropsychiatric status was further assessed using a series of alternative tests including: animal naming, Critical Flicker Fusion Frequency (CFF); Inhibitory Control Test (ICT); the SCAN package, and the Stroop test. Reference data were obtained from 61 healthy individuals (21 men, 40 women; mean age 46.4 [20–70] years). Thresholds for the diagnosis of any degree of HE were identified using receiver operating curve characteristics analysis. Diagnostic algorithms for use in clinical and research settings were then constructed using a Classification and Regression Tree growing method. Results: The Stroop test performed best overall; 80% of its generated variables separated healthy controls from patients with both minimal and overt HE. The single best variable was a combined score of the multiple variables from the SCAN package: sensitivity of 95.8% and specificity of 70.1%. An algorithmic construct, consisting of the ICT, SCAN package, and Stroop test, classified patients with a sensitivity of 92.6% and a specificity of 89.4% (Figure 1A). A modified algorithm, utilizing the Stroop test and the EEG, which would be more applicable in the clinical setting, had a sensitivity of 88.9% and a specificity of 94.7% (Figure 1B). Discussion: The derived algorithms have significantly better diagnostic performance than the individual tests currently used. They can be used in the clinical setting for: (i) the initial assessment of patients with cirrhosis to ensure they receive a diagnosis that may otherwise be overlooked thereby allowing them to receive treatment, if required; (ii) confirming the diagnosis of HE where this is in dispute; and, (iii) evaluating suitability for a transjugular intrahepatic portosystemic shunt procedure. In a research setting the derived algorithms will provide a diagnostic standard that is universally applicable. Further validation in independent cohorts of patients is clearly required. The authors have none to declare.