Automated Classification of Multi-Labeled Patient Safety Reports: A Shift from Quantity to Quality Measure

Over the past two decades, there have seen an ever-increasing amount of patient safety reports yet the capacity of extracting useful information from the reports remains limited. Classification of patient safety reports is the first step of performing a downstream analysis. In practice, the manual review processes for classification are labor-intense. Studies have shown that the reports are often mislabeled or unclassifiable based on the pre-defined categories, which presents a notable data quality problem. In this study, we investigated the multi-labeled nature of patient safety reports. We argue that understanding multi-labeled nature of reports is a key to disclose the complex relations between many components during the courses and development of medical errors. Accordingly, we developed automated multi-label text classifiers to process patient safety reports. The experiments demonstrated feasibility and efficiency of a combination of multi-label algorithms in the benchmark comparison. Grounded on our experiments and results, we provided suggestions on how to implement automated classification of patient safety reports in the clinical settings.