AssistMED project: Transforming cardiology cohort characterisation from electronic health records through natural language processing - Algorithm design, preliminary results, and field prospects

INTRODUCTION

Electronic health records (EHR) are of great value for clinical research. However, EHR consists primarily of unstructured text which must be analysed by a human and coded into a database before data analysis- a time-consuming and costly process limiting research efficiency. Natural language processing (NLP) can facilitate data retrieval from unstructured text. During AssistMED project, we developed a practical, NLP tool that automatically provides comprehensive clinical characteristics of patients from EHR, that is tailored to clinical researchers needs.

MATERIAL AND METHODS

AssistMED retrieves patient characteristics regarding clinical conditions, medications with dosage, and echocardiographic parameters with clinically oriented data structure and provides researcher-friendly database output. We validate the algorithm performance against manual data retrieval and provide critical quantitative and qualitative analysis.

RESULTS

AssistMED analysed the presence of 56 clinical conditions, medications from 16 drug groups with dosage and 15 numeric echocardiographic parameters in a sample of 400 patients hospitalized in the cardiology unit. No statistically significant differences between algorithm and human retrieval were noted. Qualitative analysis revealed that disagreements with manual annotation were primarily accounted to random algorithm errors, erroneous human annotation and lack of advanced context awareness of our tool.

CONCLUSIONS

Current NLP approaches are feasible to acquire accurate and detailed patient characteristics tailored to clinical researchers' needs from EHR. We present an in-depth description of an algorithm development and validation process, discuss obstacles and pinpoint potential solutions, including opportunities arising with recent advancements in the field of NLP, such as large language models.

Practical use case of natural language processing for observational clinical research data retrieval from electronic health records: AssistMED project

INTRODUCTION

Electronic health records (EHR) contain data valuable for clinical research but in textual format, requiring encoding to databases by a human- a lengthy and costly process. Natural language processing (NLP) is a computational technique that allows text analysis.

OBJECTIVES

To demonstrate a practical use case of NLP for a large retrospective study cohort characterization and compare it to a human retrieval.

PATIENTS AND METHODS

Anonymized discharge documentation of 10314 patients from the cardiology tertiary care department was analyzed for inclusion in the CRAFT registry (NCT02987062) of patients with atrial fibrillation (AF). Extensive clinical characteristics regarding concomitant diseases, medications, daily dosage and echocardiography were collected manually and through NLP.

RESULTS

There were 3030 and 3029 patients identified by human and NLP-based approaches, respectively, reflecting 99.93% accuracy of NLP in detecting AF. Comprehensive baseline patient characteristics by NLP was faster than human analysis (3 hours and 15 minutes vs 71 hours and 12 minutes). The calculated CHA2DS2VASc and HAS-BLED scores based on both methods did not differ (human vs NLP; median, IQR, P value): 3 (2-5) vs 3 (2-5) P = 0.74 and 1 (1-2) vs 1 (1-2) P = 0.63. For most data, an almost perfect agreement between NLP and human retrieved characteristics was found; daily dosage identification was the least accurate NLP feature. Similar conclusions on cohort characteristics would be made; however, daily dosage detection for some drug groups would require additional human validation in the NLP-based cohort.

CONCLUSIONS

NLP utilization on EHR may accelerate acquisition and provide accurate data for a retrospective study.

Photoplethysmography wave morphology in patients with atrial fibrillation

OBJECTIVE

Most current algorithms for detecting Atrial Fibrillation (AF) rely on Heart Rate Variability (HRV), and only a few studies analyse the variability of Photopletysmography (PPG) waveform. This study aimed to compare morphological features of the PPG curve in patients with AF to those presenting a normal sinus rhythm (NSR) and evaluate their usefulness in AF detection.&#xD;Approach: 10-minute PPG signals were obtained from patients with persistent/paroxysmal AF and NSR. Nine morphological parameters (1/ΔT, Pulse Width [PW], Augmentation Index [AI], b/a, e/a, [b-e]/a, Crest Time [CT], Inflection Point Area [IPA], Area and five HRV parameters (Heart rate [HR], Shannon entropy [ShE], root mean square of the successive differences [RMSSD], number of pairs of consecutive systolic peaks [R-R] that differ by more than 50 ms [NN50], standard deviation of the R-R intervals [SDNN]) were calculated.&#xD;Main Results: Eighty subjects, including 33 with AF and 47 with NSR were recruited. In univariate analysis five morphological features (1/ΔT, p<0.001; b/a, p<0.001; [b-e]/a, p<0.001; CT, p=0.011 and Area, p<0.001) and all HRV parameters (p=0.01 for HR and p<0.001 for others) were significantly different between the study groups. In the stepwise multivariate model (Area under the curve [AUC] = 0.988 [0.974-1.000]), three morphological parameters (PW, p<0.001; e/a, p=0.011; (b-e)/a, p<0.001) and three of HRV parameters (ShE, p=0.01; NN50, p<0.001, HR, p = 0.01) were significant. &#xD;Significance: There are significant differences between AF and NSR, PPG waveform, which are useful in AF detection algorithm. Moreover adding those features to HRV-based algorithms may improve their specificity and sensitivity.&#xD.