Augmented intelligence with natural language processing applied to electronic health records for identifying patients with non-alcoholic fatty liver disease at risk for disease progression

Highlights

• NAFLD is documented poorly in the EMR. We assess how well we can identify it using an NLP approach versus ICD or text search.

• NAFLD can progress to NASH and cirrhosis. We examine our ability to measure disease progression within the EMR with NLP.

• We look at breakdowns in the knowledge chain between doctors, when NAFLD was identified but not mentioned in future notes.

• We identify cases of these breakdowns where the patient developed NASH/cirrhosis without referencing prior NAFLD diagnosis.

Abstract

Objective

Electronic health record (EHR) systems contain structured data (such as diagnostic codes) and unstructured data (clinical documentation). Clinical insights can be derived from analyzing both. The use of natural language processing (NLP) algorithms to effectively analyze unstructured data has been well demonstrated. Here we examine the utility of NLP for the identification of patients with non-alcoholic fatty liver disease, assess patterns of disease progression, and identify gaps in care related to breakdown in communication among providers.

Materials and Methods

All clinical notes available on the 38,575 patients enrolled in the Mount Sinai BioMe cohort were loaded into the NLP system. We compared analysis of structured and unstructured EHR data using NLP, free-text search, and diagnostic codes with validation against expert adjudication. We then used the NLP findings to measure physician impression of progression from early-stage NAFLD to NASH or cirrhosis. Similarly, we used the same NLP findings to identify mentions of NAFLD in radiology reports that did not persist into clinical notes.

Results

Out of 38,575 patients, we identified 2,281 patients with NAFLD. From the remainder, 10,653 patients with similar data density were selected as a control group. NLP outperformed ICD and text search in both sensitivity (NLP: 0.93, ICD: 0.28, text search: 0.81) and F2 score (NLP: 0.92, ICD: 0.34, text search: 0.81). Of 2281 NAFLD patients, 673 (29.5%) were believed to have progressed to NASH or cirrhosis. Among 176 where NAFLD was noted prior to NASH, the average progression time was 410 days. 619 (27.1%) NAFLD patients had it documented only in radiology notes and not acknowledged in other forms of clinical documentation. Of these, 170 (28.4%) were later identified as having likely developed NASH or cirrhosis after a median 1057.3 days.

Discussion

NLP-based approaches were more accurate at identifying NAFLD within the EHR than ICD/text search-based approaches. Suspected NAFLD on imaging is often not acknowledged in subsequent clinical documentation. Many such patients are later found to have more advanced liver disease. Analysis of information flows demonstrated loss of key information that could have been used to help prevent the progression of early NAFLD (NAFL) to NASH or cirrhosis.

Conclusion

For identification of NAFLD, NLP performed better than alternative selection modalities. It then facilitated analysis of knowledge flow between physician and enabled the identification of breakdowns where key information was lost that could have slowed or prevented later disease progression.

Introduction

Artificial intelligence has shown increasing promise when applied identification and prediction of countless medical outcomes. When applied to clinician workflow, it provides a form of augmented intelligence, aiding clinicians with decision support and error avoidance. These applications have predominantly been built leveraging only structured data because of its availability and ease of interpretation, however unstructured data (such as dictated notes) contain critical clinical information and thereby offer the potential to greatly enhance clinical insights than can be derived from use of structured data alone. Several public and proprietary approaches have been taken to developing natural language processing (NLP) systems to make sense of unstructured clinical data [1,2]. NLP approaches have been used successfully for biomedical research such as accurate phenotyping of complex diseases and for clinical tasks including identification of patients with NAFLD [9,10]. Here we examine the use of one SNOMED-based NLP tool for extracting patient features related non-alcoholic fatty liver disease (NAFLD). NAFLD represents a spectrum of liver diseases characterized histologically by macrovesicular fat and ranging in severity from nonalcoholic fatty liver (NAFL) to non-alcoholic steatohepatitis (NASH) [3,4]. A subset of patients with NAFLD progresses to cirrhosis and has an increased risk of hepatocellular carcinoma and liver-related mortality [3]. NAFLD is emerging as one of the most common causes of liver failure in the United States [5]. Multiple professional societies have published guidelines for the diagnosis and management of patients with NAFLD [6]. NAFLD is suspected in patients with metabolic syndrome, hepatomegaly, or mild elevations in aspartate aminotransferase (AST) and alanine aminotransferase (ALT) levels. However, normal levels of AST and ALT do not exclude the presence of NAFLD and hepatomegaly is found only in approximately 20% of patients [7,8]. A key component in the diagnosis of NAFLD is evidence of hepatic steatosis on imaging or biopsy. Many patients have abdominal or chest imaging performed for unrelated disorders which may incidentally find hepatic steatosis and allow for additionally workup for NAFLD including exclusion of other chronic liver diseases and alcohol consumption. The rapid and accurate identification of NAFLD by NLP from unstructured text such as radiology reports is one potential method to address the gap between incidental findings and patient care.

We first assess the accuracy of NLP against other simpler approaches to derive insights into the treating clinician’s understanding of the patient. We then determine the ability of NLP to track the diagnostic process and identify potential breakdowns. We determine the proportion of patients with fatty liver documented in radiology reports in which the presumptive diagnosis was also documented in a progress note from a healthcare provider. Second, to identify communication breakdowns at the point of care, we examine patients where NAFLD was identified in radiology notes but never referenced in progress note. While we focused our approach on NAFLD, the methodology is broadly applicable to other disease processes.