Neural-network analysis of socio-medical data to identify predictors of undiagnosed hepatitis C virus infections in Germany (DETECT)

Abstract

BACKGROUND: Chronic hepatitis C virus (HCV)-infection is a slowly debilitating and potentially fatal disease with a high estimated number of undiagnosed cases. Given the major advances in the treatment, detection of unreported infections is a consequential step for eliminating hepatitis C on a population basis. The prevalence of chronic hepatitis C is, however, low in most countries making mass screening neither cost effective nor practicable. METHODS: We used a Kohonen artificial neural network (ANN) to analyze socio-medical data of 1.8 million insurants for predictors of undiagnosed HCV infections. The data had to be anonymized due to ethical requirements. The network was trained with variables obtained from a subgroup of 2544 patients with confirmed hepatitis C-virus (HCV) infections excluding variables directly linked to the diagnosis of HCV. All analyses were performed using the data mining solution "RayQ". Training results were visualized three-dimensionally and the distributions and characteristics of the clusters were explored within the map. RESULTS: All 2544 patients with confirmed chronic HCV diagnoses were localized in a clearly defined cluster within the Kohonen self-organizing map. An additional 2217 patients who had not been diagnosed with hepatitis C co-localized to the same cluster, indicating socio-medical similarities and a potentially elevated risk of infection. Several factors including, age, diagnosis codes and drug prescriptions acted only in conjunction as predictors of an elevated HCV risk. CONCLUSIONS: This ANN approach may allow for a more efficient risk adapted HCV-screening. However, further validation of the prediction model is required.