Erroneous Classification and Coding as a Limitation for Big Data Analyses: Causes and Impacts Illustrated by the...

Background/Objectives: Clavicle injuries are common and seem to be frequently subject to diagnostic misclassification. The accurate identification of clavicle fractures is essential, particularly for registry and Big Data analyses. This study aims to assess the frequency of diagnostic errors in clavicle injury classifications. Methods: This retrospective study analyzed patient data from two Level 1 trauma centers, covering the period from 2008 to 2019. Included were cases with ICD-coded diagnoses of medial, midshaft, and lateral clavicle fractures, as well as sternoclavicular and acromioclavicular joint dislocations. Radiological images were re-evaluated, and discharge summaries, radiological reports, and billing codes were examined for diagnostic accuracy. Results: A total of 1503 patients were included, accounting for 1855 initial injury diagnoses. In contrast, 1846 were detected upon review. Initially, 14.4% of cases were coded as medial clavicle fractures, whereas only 5.2% were confirmed. The misclassification rate was 82.8% for initial medial fractures (p < 0.001), 42.5% for midshaft fractures (p < 0.001), and 34.2% for lateral fractures (p < 0.001). Billing codes and discharge summaries were the most error-prone categories, with error rates of 64% and 36% of all misclassified cases, respectively. Over three-quarters of the cases with discharge summary errors also exhibited errors in other categories, while billing errors co-occurred with other category errors in just over half of the cases (p < 0.001). The likelihood of radiological diagnostic error increased with the number of imaging modalities used, from 19.7% with a single modality to 30.5% with two and 40.7% with three. Conclusions: Our findings indicate that diagnostic misclassification of clavicle fractures is common, particularly between medial and midshaft fractures, often resulting from errors in multiple categories. Further prospective studies are needed, as accurate classification is foundational for the reliable application of Big Data and AI-based analyses in clinical research.