Explorative data science analysis of the Universal Newborn Hearing Screening in Schleswig-Holstein (2004–2024)

Question: What insights can be gained from the long-term data of the Universal Newborn Hearing Screening in Schleswig-Holstein and how can modern data science methods be used to analyse and visualize these data? Methods: This study is based on a retrospective analysis of screening data collected in over 20 clinics and 50 practices in Schleswig-Holstein between 2004 and 2024. The analysis of these data, which included both initial and re-screening results, aimed to evaluate participation rates and diagnostic trends in newborn hearing screening. Comprehensive data cleaning was required due to various factors. Exploratory statistical methods and data science techniques such as vectorization, embedding and t-SNE (t-distributed Stochastic Neighbour Embedding) were used for the analysis. Results: Analysis of the screening data showed that the number of clinics involved in reporting varied over the years. There were differences in the number of screenings performed directly in the clinics. The number of control screenings were higher in 2015 and 2016. The total number of diagnosed cases of hearing impairment corresponded to the expected prevalence. The number of documented hearing aid fittings varied widely and there was a significant under-reporting of cochlear implant fittings. Natural Language Processing (NLP) was used to extract and visualize additional information from the free text field “History”, which provided new insights into the data. Conclusions: The application of data science methods to the collected data of the universal newborn hearing screening in Schleswig-Holstein provided valuable insights. The results highlight the potential for the development of specific intervention strategies. Recording the date of birth as well as training for the screening staff and improving the completeness of the data sets are important factors for future analyses.