Identification of Long Bone Fractures in Radiology Reports Using Natural Language Processing to support Healthcare Quality Improvement.

2016

1051-1068

2016 Nov 09

1869-0327

<p><strong>BACKGROUND: </strong>Important information to support healthcare quality improvement is often recorded in free text documents such as radiology reports. Natural language processing (NLP) methods may help extract this information, but these methods have rarely been applied outside the research laboratories where they were developed.</p>

<p><strong>OBJECTIVE: </strong>To implement and validate NLP tools to identify long bone fractures for pediatric emergency medicine quality improvement.</p>

<p><strong>METHODS: </strong>Using freely available statistical software packages, we implemented NLP methods to identify long bone fractures from radiology reports. A sample of 1,000 radiology reports was used to construct three candidate classification models. A test set of 500 reports was used to validate the model performance. Blinded manual review of radiology reports by two independent physicians provided the reference standard. Each radiology report was segmented and word stem and bigram features were constructed. Common English "stop words" and rare features were excluded. We used 10-fold cross-validation to select optimal configuration parameters for each model. Accuracy, recall, precision and the F1 score were calculated. The final model was compared to the use of diagnosis codes for the identification of patients with long bone fractures.</p>

<p><strong>RESULTS: </strong>There were 329 unique word stems and 344 bigrams in the training documents. A support vector machine classifier with Gaussian kernel performed best on the test set with accuracy=0.958, recall=0.969, precision=0.940, and F1 score=0.954. Optimal parameters for this model were cost=4 and gamma=0.005. The three classification models that we tested all performed better than diagnosis codes in terms of accuracy, precision, and F1 score (diagnosis code accuracy=0.932, recall=0.960, precision=0.896, and F1 score=0.927).</p>

<p><strong>CONCLUSIONS: </strong>NLP methods using a corpus of 1,000 training documents accurately identified acute long bone fractures from radiology reports. Strategic use of straightforward NLP methods, implemented with freely available software, offers quality improvement teams new opportunities to extract information from narrative documents.</p>

10.4338/ACI-2016-08-RA-0129

Appl Clin Inform

27826610

Clinical Decision Support for a Multicenter Trial of Pediatric Head Trauma: Development, Implementation, and Lessons Learned.

2016

534-42

2016

1869-0327

<p><strong>INTRODUCTION: </strong>For children who present to emergency departments (EDs) due to blunt head trauma, ED clinicians must decide who requires computed tomography (CT) scanning to evaluate for traumatic brain injury (TBI). The Pediatric Emergency Care Applied Research Network (PECARN) derived and validated two age-based prediction rules to identify children at very low risk of clinically-important traumatic brain injuries (ciTBIs) who do not typically require CT scans. In this case report, we describe the strategy used to implement the PECARN TBI prediction rules via electronic health record (EHR) clinical decision support (CDS) as the intervention in a multicenter clinical trial.</p>

<p><strong>METHODS: </strong>Thirteen EDs participated in this trial. The 10 sites receiving the CDS intervention used the Epic(®) EHR. All sites implementing EHR-based CDS built the rules by using the vendor's CDS engine. Based on a sociotechnical analysis, we designed the CDS so that recommendations could be displayed immediately after any provider entered prediction rule data. One central site developed and tested the intervention package to be exported to other sites. The intervention package included a clinical trial alert, an electronic data collection form, the CDS rules and the format for recommendations.</p>

<p><strong>RESULTS: </strong>The original PECARN head trauma prediction rules were derived from physician documentation while this pragmatic trial led each site to customize their workflows and allow multiple different providers to complete the head trauma assessments. These differences in workflows led to varying completion rates across sites as well as differences in the types of providers completing the electronic data form. Site variation in internal change management processes made it challenging to maintain the same rigor across all sites. This led to downstream effects when data reports were developed.</p>

<p><strong>CONCLUSIONS: </strong>The process of a centralized build and export of a CDS system in one commercial EHR system successfully supported a multicenter clinical trial.</p>

10.4338/ACI-2015-10-CR-0144

Appl Clin Inform

27437059