Automated ICD Coding Using Deep Learning

ICD (international classification of diseases) is a system developed by the WHO in which every unique diagnosis and procedure has a unique code. It provides a standardized way to represent medical information and makes it sharable and comparable across different hospitals and countries. Currently, the task of assigning ICD codes to patient discharge summaries is performed manually by medical coders. Manual coding is costly, time consuming, and inefficient for huge data. So, the healthcare industry requires automated solutions to make the medical coding more efficient, accurate, and consistent. In this study, the automated ICD-9 coding is approached as a multi-label text classification problem. A deep learning system is presented to assign ICD-9 codes automatically to the patient discharge summaries. Convolutional neural networks and word2vec model are combined to automatically extract features from the input text. The best model has achieved 83.28% accuracy. The results of this research prove the usability of deep learning for multi-label text classification and medical coding.

Developing A Deep Learning Natural Language Processing Algorithm For Automated Reporting Of Adverse Drug Reactions

The detection of adverse drug reactions (ADRs) is critical to our understanding of the safety and risk-benefit profile of medications. With an incidence that has not changed over the last 30 years, ADRs are a significant source of patient morbidity, responsible for 5-10% of acute care hospital admissions worldwide. Spontaneous reporting of ADRs has long been the standard method of reporting, however this approach is known to have high rates of under-reporting, a problem that limits pharmacovigilance efforts. Automated ADR reporting presents an alternative pathway to increase reporting rates, although this may be limited by over-reporting of other drug-related adverse events. We developed a deep learning natural language processing algorithm to identify ADRs in discharge summaries at a single academic hospital centre. Our model was developed in two stages: first, a pre-trained model (DeBERTa) was further pre-trained on 150,000 unlabelled discharge summaries; secondly, this model was fine-tuned to detect ADR mentions in a corpus of 861 annotated discharge summaries. To ensure that our algorithm could differentiate ADRs from other drug-related adverse events, the annotated corpus was enriched for both validated ADR reports and confounding drug-related adverse events using. The final model demonstrated good performance with a ROC-AUC of 0.934 (95% CI 0.931 - 0.955) for the task of identifying discharge summaries containing ADR mentions.

693. Performance of ICD Code Versus Discharge Summary based Query for Endocarditis Cohort Identification

Background Studies on infective endocarditis (IE) have relied on International Classification of Diseases (ICD) codes to identify cases but few have validated this method which may be prone to misclassification. Examination of clinical narrative data could offer greater accuracy and richness. Methods We evaluated two algorithms for IE identification from 7/1/2015 to 7/31/2019: (1) a standard query of ICD codes for IE (ICD-9: 424.9, 424.91, 424.99, 421.0, 421.1, 421.9, 112.81, 036.42 and ICD-10: I38, I39, I33, I33.9, B37.6 and A39.51) with or without procedure codes for echocardiogram (93303-93356) and (2) a key word, pattern-based text query of discharge summaries (DS) that selected on the term “endocarditis” in fields headed by “Discharge Diagnosis” or “Admission Diagnosis” or similar. Further coding extracted the nature and type of valve and the organism responsible for the IE if present in DS. All identified cases were chart reviewed using pre-specified criteria for true IE. Positive predictive value (PPV) was calculated as the total number of verified cases over the algorithm-selected cases. Sensitivity was the total number of algorithm-matched cases over a final list of 166 independently identified true IE cases from ID and Cardiology services. Specificity was defined using 119 pre-adjudicated non-cases minus the number of algorithm-matched cases over 119. Results The ICD-based query identified 612 individuals from July 2015 to July 2019 who had a hospital billing code for infective endocarditis; of these, 534 also had an echocardiogram. The DS query identified 387 cases. PPV for the DS query was 84.5% (95% confidence interval [CI] 80.6%, 87.8%) compared with 72.4% (95% CI 68.7%, 75.8%) for ICD only and 75.8% (95% CI 72.0%, 79.3%) for ICD + echo queries. Sensitivity was 75.9% for the DS query and 86.8-93.4% for the ICD queries. Specificity was high for all queries >94%. The DS query also yielded valve data (prosthetic, tricuspid, pulmonic, aortic or mitral) in 60% and microbiologic data in 73% of identified cases with an accuracy of 94% and 90% respectively when assessed by chart review. Table 1. Test Characteristics of Three Electronic Health Record Queries for Infective Endocarditis Conclusion Compared to traditional ICD-based queries, text-based queries of discharge summaries have the potential to improve precision of IE case ascertainment and extract key clinical variables. Disclosures All Authors: No reported disclosures

Prophylactic use of proton pump inhibitors post-cardiac surgery

Aims/Background Gastrointestinal bleeding significantly increases morbidity and mortality rates postoperatively in patients undergoing cardiac surgery. The prophylactic prescribing of proton pump inhibitors post-cardiac surgery is currently a class IIa recommendation of the European Association of Cardio-Thoracic Surgery. Method A retrospective review of patients who underwent cardiac surgery between July and December 2019 in the authors' hospital was carried out, using discharge summaries. New treatment charts were introduced with a pre-printed proton pump inhibitor included in the ‘regular medication’ section of the treatment chart and two reaudits were performed using the same methodology. Results Before the intervention, 47% were prescribed omeprazole postoperatively, compared to 74% (P<0.001) and 66% (P=0.008) in the first and second reaudits respectively. Gastrointestinal bleeding was more common pre-intervention (4% vs 1% respectively; P=0.10). Conclusions This intervention resulted in a statistically significant improvement in the prescription of postoperative omeprazole and a decrease in gastrointestinal bleeds. However, other risk factors such as diabetes mellitus, arteriosclerosis and procedure urgency may have contributed to the absence of statistical significance in the latter.

EP.TU.281Mind the Gap: Improving the Quality of Neurosurgical Discharge Summaries

Aims A discharge summary is a permanent record of a patient’s hospital visit and the primary means of handover between care providers. Studies show they often lack precision and omit important information. This may compromise quality and continuity of care yet they are frequently written by the most junior clinicians on a ward with little guidance or formal education on how to write one. The aim of this study was to develop some specific guidelines to improve the quality of discharge summaries in a busy neurosurgical unit. Methods A survey was designed to identify the challenges faced by junior medical staff in writing discharge summaries. The essential components of a good neurosurgical discharge summary were identified by group of senior neurosurgeons. Summaries were retrospectively audited against these components. We then designed a simple visual aid and placed it above computer stations in the junior doctors’ offices. Formal departmental teaching session followed. After three months we re-audited the discharge summaries retrospectively to measure any effect of our intervention. Results Half of the neurosurgical team rated summaries as below expectations. Challenges included poor ward round documentation and a lack of clear expectations regarding structure and essential components. After the intervention, ward round documentation and discharge summary quality improved dramatically. Conclusions Although various recommendations about writing good discharge summaries exist, they are generally vague and not specific to neurosurgical practice. The development of a simple specialty specific discharge summary guide can improve discharge summary quality and should be encouraged in all specialties.

EP.TU.740Implementing a departmental handbook to improve induction for FY1s during COVID-19 pandemic

Aims The 2020 cohort of FY1s qualified during the COVID-19 pandemic; most prequalification placements were cancelled. Induction provides crucial information for new FY1s. We assessed the impact of redesigning the General Surgery induction handbook at Hereford County Hospital. Methods A 40-question survey, designed against standards in ’Recommendations for safe trainee changeover’ published by the Academy of Medical Royal Colleges, was sent to all FY1s commencing General Surgery in August 2020 two weeks following induction. The survey contained 5 sections: introduction; testing your knowledge parts A and B; team atmosphere; and demographics. Answer modalities included: binary answer; multiple choice; Net Promoter Scores (NPS); and free text. Submissions were made electronically via Microsoft Forms. Feedback guided redesign of the induction handbook which was sent to FY1s rotating December 2020. Information included: how to contact seniors; requesting investigations; referring to specialties and expected duties. A repeat survey was sent to FY1s 4 weeks following induction. Results 10/11 FY1s responded during the first rotation; 9/11 during the second. FY1s felt more confident requesting bloods (NPS +50 to + 90), requesting imaging (NPS –20 to + 70), completing discharge summaries (NPS +30 to + 80) and referring to specialties (NPS –40 to + 60). There was a better understanding of different shift types (NPS –40 to + 30). More FY1s correctly recalled on call bleeps for the medical registrar (from 56% to 75%) and the anaesthetic registrar (from 50% to 78%). Most (78%, NPS +78) felt they had enough support from other FY1s which remained the same (NPS +78) through the second rotation. As a result of the COVID-19 pandemic, prequalification experience varied greatly between individual FY1s. Conclusion Overall, FY1s were more confident requesting investigations, referring to other specialties and completing discharge summaries after redesign of the induction handbook. Considering disruptions in pre-qualification training as a result of the COVID-19 pandemic, a thorough departmental induction handbook can be an invaluable resource tool to aid rotation into a new specialty.

TP9.2.20Improving surgical discharge summaries

Introduction Discharge summaries are a means of communication to the patient, the GP and for medical records. An initial audit showed surgical discharge summaries contained misleading information and sometimes omitted relevant information. Changes were implemented to improve the accuracy of surgical discharge summaries. Method The initial audit assessed the accuracy of discharge summaries over a two-week period and the re-audit was conducted after implementation of change over a similar time period. Data was extracted from electronic patient records (EPR). Change implementation included educating the surgical team on the need for accurate discharge summaries. The EPR team was notified of the intrinsic error in the PowerChart system which is widely used in various NHS Trust. Results Incidence of misdiagnosis or misleading diagnosis in discharge summaries reduced from 42% to zero, lack of relevant investigations decreased from 7% to 1%, No follow up status reduced from 23% to 10% (usually post appendicectomy patients which are not routinely followed up but this needs to be stated in the discharge summary for clarity), at both initial audit and re-audit all patients had relevant surgery or procedures done included in their discharge summaries while the rate at which relevant medications were not stated in the discharge summary decreased from 4% to zero. Conclusions Discharge summaries are vital for record keeping and are usually the only written information a patient receives regarding their hospital stay. It is important that errors in EPR systems be flagged up for review.

EP.TU.79Assessed and Discharged - An Audit of Discharge Summaries Completion During On-Calls

Aim Effective health care provision is heavily dependent on timely, reliable transfer of patient information. Failure of this communication between professionals could result in redundancy of tests, delay in treatment, which may in turn endanger patient safety. The NHS Standard Contract requirements state discharge summaries should be completed within 24 hours of hospital assessment and discharge. Discharge summaries for patients who were reviewed but not admitted have been observed to be poorly completed during on-calls and this audit aims to clarify this. Method On-Call Patient Lists between 1 December to 14 December 2020 were studied retrospectively. Patients who were assessed by the on-call surgical team but not admitted were included in the audit. Patients referred to other specialties were excluded. Hospital electronic system was reviewed for electronic records from the encounter including clinical note or discharge summary. Results In total, 47 patients were identified during the 2 week- period. 40/47 patients were referred from AE and 9 of these patients were discharged from AE directly. 3 of the patients had a clinical note or discharge summary completed on the hospital electronic system. Overall, 18 of the 47 (38.3%) patients had a clinical note or discharge summary on the electronic system, with 6 (12.8%) of them being recorded as discharge summaries. Conclusion The overall completion of discharge summaries for this group of patients was poor. Awareness of this failing and the importance of professional communication should be highlighted with the juniors during surgical meeting to improve compliance.

Mining Clinical Process from Hospital Information System: A Granular Computing Approach

Data mining methods in medicine is a very important tool for developing automated decision support systems. However, since information granularity of disease codes used in hospital information system is coarser than that of real clinical definitions of diseases and their treatment, automated data curation is needed to extract knowledge useful for clinical decision making. This paper proposes automated construction of clinical process plan from nursing order histories and discharge summaries stored in hospital information system with curation of disease codes as follows. First, the system applies EM clustering to estimate subgrouping of a given disease code from clinical cases. Second, it decomposes the original datasets into datasets of subgroups by using granular homogenization. Thirdly, clinical pathway generation method is applied to the datasets. Fourthly, classification models of subgroups are constructed by using the analysis of discharge summaries to capture the meaning of each subgroup. Finally, the clinical pathway of a given disease code is output as the combination of the classifiers of subgroups and the the pathways of the corresponding subgroups. The proposed method was evaluated on the datasets extracted hospital information system in Shimane University Hosptial. The obtained results show that more plausible clinical pathways were obtained, compared with previously introduced methods.