Is data quality enough for a clinical decision?: Apply machine learning and avoid bias

This paper provides a practical guideline for the assurance and (re-)usage of clinical data. It proposes a process which aims to provide a systematic data quality assurance even without involving a medical domain expert. Especially when (re-)using clinical data, data quality is an important topic because clinical data are not purposely collected. Therefore, data driven conclusions might be false, because a given dataset is not representative. These false data driven conclusions could even harm the life of patients. Thus, all researchers should adhere to some basic principles that can prevent false conclusions. Twelve empirical experiments were conducted in order to prove that my process is able to assure data quality with respect to the descriptive and predictive analysis. Descriptive results obtained by applying stratified sampling are conflicting in four out of nine population inputs. Sampling is carried based on the top ranked feature drawn by the Contextual Data Quality Assurance (CDQA). Between datasets these features are confirmed by the Mutual Data Quality Assurance (MDQA). Stratified sampled inputs improve predictive results compared to raw data. Both Area Under the Curve (AUC) scores and accuracy scores increase by three percent.