A large dataset of annotated incident reports on medication errors.

Incident reports of medication errors are valuable learning resources for improving patient safety. However, pertinent information is often contained within unstructured free text, which prevents automated analysis and limits the usefulness of these data. Natural language processing can structure this free text automatically and retrieve relevant past incidents and learning materials, but to be able to do so requires a large, fully annotated and validated corpus of incident reports. We present a corpus of 58,658 machine-annotated incident reports of medication errors that can be used to advance the development of information extraction models and subsequent incident learning. We report the best F1-scores for the annotated dataset: 0.97 and 0.76 for named entity recognition and intention/factuality analysis, respectively, for the cross-validation exercise. Our dataset contains 478,175 named entities and differentiates between incident types by recognising discrepancies between what was intended and what actually occurred. We explain our annotation workflow and technical validation and provide access to the validation datasets and machine annotator for labelling future incident reports of medication errors.

Rule-Based Natural Language Processing Pipeline to Detect Medication-Related Named Entities: Insights for Transfer Learning.

We document the procedure and performance of a rule-based NLP system that, using transfer learning, automatically extracts essential named entities related to drug errors from Japanese free-text incident reports. Subsequently, we used the rule-based annotated data to fine-tune a pre-trained BERT model and examined the performance of medication-related incident report prediction. The rule-based pipeline achieved a macro-F1-score of 0.81 in an internal dataset and the BERT model fine-tuned with rule-annotated data achieved a macro-F1-score of 0.97 and 0.75 for named entity recognition and relation extraction tasks, respectively. The model can be deployed to other, similar problems in medication-related clinical texts.

Protocol for the automatic extraction of epidemiological information via a pre-trained language model.

The lack of systems to automatically extract epidemiological fields from open-access COVID-19 cases restricts the timeliness of formulating prevention measures. Here we present a protocol for using CCIE, a COVID-19 Cases Information Extraction system based on the pre-trained language model.1 We describe steps for preparing supervised training data and executing python scripts for named entity recognition and text category classification. We then detail the use of machine evaluation and manual validation to illustrate the effectiveness of CCIE. For complete details on the use and execution of this protocol, please refer to Wang et al.2.

Customised Peripheral Corneal Cross-linking (P-CXL) for Ultra-thin Corneas with Stage III and IV Keratoconus.

PURPOSE: This study aims to determine whether customised peripheral corneal cross-linking (P-CXL) can halt keratoconus progression in ultrathin corneas with stage 3 and 4 keratoconus, with thinnest pachymetry well below 400 µm and therefore excluded from most treatment protocols. METHODS: This retrospective study included 21 eyes with progressive keratoconus and thinnest pachymetry ranging from 97 to 399 µm (mean 315 µm), who underwent P-CXL between 2007 and 2020. The procedure involved preoperative NSAIDs therapy, tomography-guided customized epithelial debridement, the use of both hypo-osmolar and iso-osmolar riboflavin solutions, and 9.0 mW/cm2 UV-A irradiation for 10 minutes. The outcome measures were best spectacle-corrected visual acuity (BSCVA), mean keratometry, maximum keratometry, and thinnest pachymetry. RESULTS: After a minimum follow-up period of 12 months, P-CXL stabilized or improved mean keratometry and maximum keratometry in 85.7% of eyes (Kavg from 57.48 ± 9.38 to 56.43 ± 8.96 D, p < 0.001; Kmax from 72.77 ± 12.74 to 70.00 ± 11.50 D, p < 0.001), BSCVA in 90.5% of eyes (from 4.48 ± 2.85 to 5.72 ± 3.34 decimals, p < 0.001), and thinnest pachymetry in 81% of eyes (from 315.81 ± 90.05 to 342.33 ± 74.22 µm, p = 0.08). No adverse events and no loss of endothelial cell density occurred. CONCLUSIONS: Customised peripheral corneal cross-linking (P-CXL) treated very severe keratoconus with a success rate of 85.7% and improved visual acuity and tomographic indicators in most cases. While a longer follow-up and a larger sample would help to support such conclusions to a greater extent, these results allow to broaden the treatment spectrum for patients with stage 3 and 4 keratoconus and contact lens tolerance.

Epidemiologic information discovery from open-access COVID-19 case reports via pretrained language model.

Although open-access data are increasingly common and useful to epidemiological research, the curation of such datasets is resource-intensive and time-consuming. Despite the existence of a major source of COVID-19 data, the regularly disclosed case reports were often written in natural language with an unstructured format. Here, we propose a computational framework that can automatically extract epidemiological information from open-access COVID-19 case reports. We develop this framework by coupling a language model developed using deep neural networks with training samples compiled using an optimized data annotation strategy. When applied to the COVID-19 case reports collected from mainland China, our framework outperforms all other state-of-the-art deep learning models. The information extracted from our approach is highly consistent with that obtained from the gold-standard manual coding, with a matching rate of 80%. To disseminate our algorithm, we provide an open-access online platform that is able to estimate key epidemiological statistics in real time, with much less effort for data curation.

Annotation Guidelines for Medication Errors in Incident Reports: Validation Through a Mixed Methods Approach.

At present no adequate annotation guidelines exists for incident report learning. This study aims at utilizing multiple quantitative and qualitative evidence to validate annotation guidelines for incident reporting of medication errors. Through multiple approaches via annotator training, annotation performance evaluation, exit surveys, and user and expert interviews, a mixed methods explanatory sequential design was utilized to collect 2-stage evidence for validation. We recruited two patient safety experts to participate in piloting, three annotators to receive annotation training and provide user feedback, and two incident report system designers to offer expert comments. Regarding the annotation performance evaluation, the overall accuracy reached 97% and 90% for named entity identification and attribute identification respectively. Participants provided invaluable comments and opinions towards improving the annotation methods. The mixed methods approach created a significant evidential basis for the use of annotation guidelines for incident report of medication errors. Further expansion of the guidelines and external validity present options for future research.

Automatic Patient Fall Outcome Extraction Using Narrative Incident Reports.

Patient outcome is one of the key information categories in incident reporting. Being able to extract meaningful patient fall outcomes would allow better analysis of the consequences and possible mitigating actions for in-hospital fall incidents. This study aims to automate the extraction of patient outcomes from narrative fall incident reports by decomposing this into three classification subtasks: injured or not, injury types, and the number of injuries. Implementing an existing incident report classification (IRC) framework, the experimental results demonstrated that oversampling and structured features were effective to achieve better overall performances across all three subtasks. The study further validated the application of an IRC framework to deal with imbalanced classification problems found in fall patient outcome classification and advanced the science of automatic patient outcomes extraction.

Evaluating resampling methods and structured features to improve fall incident report identification by the severity level.

OBJECTIVE: This study aims to improve the classification of the fall incident severity level by considering data imbalance issues and structured features through machine learning. MATERIALS AND METHODS: We present an incident report classification (IRC) framework to classify the in-hospital fall incident severity level by addressing the imbalanced class problem and incorporating structured attributes. After text preprocessing, bag-of-words features, structured text features, and structured clinical features were extracted from the reports. Next, resampling techniques were incorporated into the training process. Machine learning algorithms were used to build classification models. IRC systems were trained, validated, and tested using a repeated and randomly stratified shuffle-split cross-validation method. Finally, we evaluated the system performance using the F1-measure, precision, and recall over 15 stratified test sets. RESULTS: The experimental results demonstrated that the classification system setting considering both data imbalance issues and structured features outperformed the other system settings (with a mean macro-averaged F1-measure of 0.733). Considering the structured features and resampling techniques, this classification system setting significantly improved the mean F1-measure for the rare class by 30.88% (P value < .001) and the mean macro-averaged F1-measure by 8.26% from the baseline system setting (P value < .001). In general, the classification system employing the random forest algorithm and random oversampling method outperformed the others. CONCLUSIONS: Structured features provide essential information for categorizing the fall incident severity level. Resampling methods help rebalance the class distribution of the original incident report data, which improves the performance of machine learning models. The IRC framework presented in this study effectively automates the identification of fall incident reports by the severity level.

The impact of performance status on tuberculosis-related death among elderly patients with lung tuberculosis: A competing risk regression analysis.

While advanced age is a main prognostic factor in patients with tuberculosis, the factors that specifically affect tuberculosis-related death are unclear because elderly people are at a risk for other age-related lethal diseases. We aimed to assess the impact of performance status on tuberculosis-related death among elderly patients with lung tuberculosis. Elderly patients (≥65 years of age) admitted to our hospital for bacteriologically-diagnosed lung tuberculosis were included, and analyzed the influence of performance status on tuberculosis-related in-hospital death, with non-tuberculosis-related death as a competing risk. Forty and 19 of the 275 patients died from tuberculosis-related causes and non-tuberculosis-related causes, respectively. The tuberculosis-related death group had a greater number of patients with a poor performance status (defined as category 3 and 4 [HR 21.022; 95%CI 2.881-153.414; p = 0.003]), a lower serum albumin level (HR 0.179; 95%CI 0.090-0.359; p < 0.001) and a higher C-reactive protein level (HR1.076; 95%CI 1.026-1.127; p = 0.002). A multivariate competing risk regression analysis showed that a poor performance status (HR 7.311; 95%CI 1.005-53.181; p = 0.049) and low albumin level (HR 0.228; 95%CI 0.099-0.524); p = 0.001) significantly predicted tuberculosis-related death. Performance status can be a useful scale for predicting tuberculosis-related death among elderly patients with pulmonary tuberculosis.

Comparison of video laryngoscopy versus direct laryngoscopy for intubation in emergency department patients with cardiac arrest: A multicentre study.

AIM: To compare the tracheal intubation performance between video laryngoscopy (VL) and direct laryngoscopy (DL) in patients with cardiac arrest in the ED. METHODS: This is an analysis of the data from a prospective, multicentre study of 15 EDs in Japan. We included consecutive adult patients with cardiac arrest who underwent intubation with VL or DL from 2012 through 2016. The primary outcome was first-attempt success. The secondary outcomes were glottic visualisation assessed with Cormack grade (1 vs. 2-4) and occurrence of oesophageal intubation. To examine the between-device difference in outcome risks, we analysed the whole data and 1:1 propensity score matched data. RESULTS: Among 9694 patients who underwent intubation in the EDs, 3360 cardiac arrests (35%) were included in the analysis (90% were non-traumatic cardiac arrests). The first-attempt success rate was higher in the VL group compared to those in the DL (78% vs 70%; unadjusted OR 1.61 [95%CI 1.26-2.06] P < 0.001). This association remained significant after adjusting for six potential confounders and within-ED clustering (adjusted OR 1.33 [95%CI 1.03-1.73] P = 0.03). VL use was also associated with a better glottic visualisation (adjusted OR 3.84 [95%CI 2.81-5.26] P < 0.001) and lower rate of oesophageal intubation (adjusted OR 0.45 [95%CI 0.24-0.85] P = 0.01) compared to DL. These results were consistent in the propensity score matched analysis. CONCLUSIONS: Based on large multicentre prospective data of ED patients with cardiac arrest, the use of VL was associated with a higher first-attempt success rate compared to DL, with a better glottic visualisation and lower oesophageal intubation rate.

Artificial Intelligence for infectious disease Big Data Analytics.

BACKGROUND: Since the beginning of the 21st century, the amount of data obtained from public health surveillance has increased dramatically due to the advancement of information and communications technology and the data collection systems now in place. METHODS: This paper aims to highlight the opportunities gained through the use of Artificial Intelligence (AI) methods to enable reliable disease-oriented monitoring and projection in this information age. RESULTS AND CONCLUSION: It is foreseeable that together with reliable data management platforms AI methods will enable analysis of massive infectious disease and surveillance data effectively to support government agencies, healthcare service providers, and medical professionals to response to disease in the future.

Transmission of methicillin-resistant Staphylococcus aureus in the long term care facilities in Hong Kong.

BACKGROUND: The relative contribution of long term care facilities (LTCFs) and hospitals in the transmission of methicillin-resistant Staphylococcus aureus (MRSA) is unknown. METHODS: Concurrent MRSA screening and spa type analysis was performed in LTCFs and their network hospitals to estimate the rate of MRSA acquisition among residents during their stay in LTCFs and hospitals, by colonization pressure and MRSA transmission calculations. RESULTS: In 40 LTCFs, 436 (21.6%) of 2020 residents were identified as 'MRSA-positive'. The incidence of MRSA transmission per 1000-colonization-days among the residents during their stay in LTCFs and hospitals were 309 and 113 respectively, while the colonization pressure in LTCFs and hospitals were 210 and 185 per 1000-patient-days respectively. MRSA spa type t1081 was the most commonly isolated linage in both LTCF residents (76/121, 62.8%) and hospitalized patients (51/87, 58.6%), while type t4677 was significantly associated with LTCF residents (24/121, 19.8%) compared with hospitalized patients (3/87, 3.4%) (p<0.001). This suggested continuous transmission of MRSA t4677 among LTCF residents. Also, an inverse linear relationship between MRSA prevalence in LTCFs and the average living area per LTCF resident was observed (Pearson correlation -0.443, p=0.004), with the odds of patients acquiring MRSA reduced by a factor of 0.90 for each 10 square feet increase in living area. CONCLUSIONS: Our data suggest that MRSA transmission was more serious in LTCFs than in hospitals. Infection control should be focused on LTCFs in order to reduce the burden of MRSA carriers in healthcare settings.