Deep Learning Methods for Detecting Side Effects of Cancer Chemotherapies Reported in a Remote Monitoring Web Application.

The objective of our work was to develop deep learning methods for extracting and normalizing patient-reported free-text side effects in a cancer chemotherapy side effect remote monitoring web application. The F-measure was 0.79 for the medical concept extraction model and 0.85 for the negation extraction model (Bi-LSTM-CRF). The next step was the normalization. Of the 1040 unique concepts in the dataset, 62, 3% scored 1 (corresponding to a perfect match with an UMLS CUI). These methods need to be improved to allow their integration into home telemonitoring devices for automatic notification of the hospital oncologists.

Designing and describing an electronic referral system to facilitate direct hospital admissions.

BACKGROUND: In France, the progressive use of emergency departments (EDs) by primary care providers (PCPs) as a point of access to hospitalization for nonurgent patients is one of the many causes of their overcrowding. To increase the proportion of direct hospital admissions, it is necessary to improve coordination between PCPs and hospital specialists. The objective of our work was to describe the design and implementation of an electronic referral system aimed at facilitating direct hospital admissions. METHODS: This initiative was conducted in a French area (Hauts-de-Seine Sud) through a partnership between the Antoine-Béclère University Hospital, the Paris-Saclay University Department of General Medicine and the local health care network. The implementation was carried out in 3 stages, namely, conducting a survey of PCPs in the territory about their communication methods with the hospital, designing and implementing a web-based application called "SIPILINK" (Système d'Information de la Plateforme d'Intermédiation Link) and an innovative organization for hospital management of the requests, and analysing through descriptive statistics the platform use 9 months after launch. RESULTS: The e-referral platform was launched in November 2019. First, a PCP filled out an electronic form describing the reason for his or her request. Then, a hospital specialist worked to respond within 72 h. Nine months after the launch, 132 PCPs had registered for the SIPILINK platform, which represented 36.6% of PCPs in this area. Of the 124 requests made, 46.8% corresponded to a hospitalization request (conventional or day hospitalization). The most requested specialty was internal medicine (48.4% of requests). The median time to first response was 43 min, and 43.5% of these requests resulted in direct admission (conventional or day hospitalization). CONCLUSIONS: This type of system responds to a need for coordination in the primary-secondary care direction, which is less often addressed than in the secondary-primary care direction. The first results show the potential of the system to facilitate direct admissions within a short time frame. To make the system sustainable, the next step is to extend its use to other hospitals in the territory.

Deep Learning versus Conventional Machine Learning for Detection of Healthcare-Associated Infections in French Clinical Narratives.

OBJECTIVE: The objective of this article was to compare the performances of health care-associated infection (HAI) detection between deep learning and conventional machine learning (ML) methods in French medical reports. METHODS: The corpus consisted in different types of medical reports (discharge summaries, surgery reports, consultation reports, etc.). A total of 1,531 medical text documents were extracted and deidentified in three French university hospitals. Each of them was labeled as presence (1) or absence (0) of HAI. We started by normalizing the records using a list of preprocessing techniques. We calculated an overall performance metric, the F1 Score, to compare a deep learning method (convolutional neural network [CNN]) with the most popular conventional ML models (Bernoulli and multi-naïve Bayes, k-nearest neighbors, logistic regression, random forests, extra-trees, gradient boosting, support vector machines). We applied the hyperparameter Bayesian optimization for each model based on its HAI identification performances. We included the set of text representation as an additional hyperparameter for each model, using four different text representations (bag of words, term frequency-inverse document frequency, word2vec, and Glove). RESULTS: CNN outperforms all other conventional ML algorithms for HAI classification. The best F1 Score of 97.7% ± 3.6% and best area under the curve score of 99.8% ± 0.41% were achieved when CNN was directly applied to the processed clinical notes without a pretrained word2vec embedding. Through receiver operating characteristic curve analysis, we could achieve a good balance between false notifications (with a specificity equal to 0.937) and system detection capability (with a sensitivity equal to 0.962) using the Youden's index reference. CONCLUSIONS: The main drawback of CNNs is their opacity. To address this issue, we investigated CNN inner layers' activation values to visualize the most meaningful phrases in a document. This method could be used to build a phrase-based medical assistant algorithm to help the infection control practitioner to select relevant medical records. Our study demonstrated that deep learning approach outperforms other classification learning algorithms for automatically identifying HAIs in medical reports.

Accuracy of using natural language processing methods for identifying healthcare-associated infections.

OBJECTIVE: There is a growing interest in using natural language processing (NLP) for healthcare-associated infections (HAIs) monitoring. A French project consortium, SYNODOS, developed a NLP solution for detecting medical events in electronic medical records for epidemiological purposes. The objective of this study was to evaluate the performance of the SYNODOS data processing chain for detecting HAIs in clinical documents. MATERIALS AND METHODS: The collection of textual records in these hospitals was carried out between October 2009 and December 2010 in three French University hospitals (Lyon, Rouen and Nice). The following medical specialties were included in the study: digestive surgery, neurosurgery, orthopedic surgery, adult intensive-care units. Reference Standard surveillance was compared with the results of automatic detection using NLP. Sensitivity on 56 HAI cases and specificity on 57 non-HAI cases were calculated. RESULTS: The accuracy rate was 84% (n = 95/113). The overall sensitivity of automatic detection of HAIs was 83.9% (CI 95%: 71.7-92.4) and the specificity was 84.2% (CI 95%: 72.1-92.5). The sensitivity varies from one specialty to the other, from 69.2% (CI 95%: 38.6-90.9) for intensive care to 93.3% (CI 95%: 68.1-99.8) for orthopedic surgery. The manual review of classification errors showed that the most frequent cause was an inaccurate temporal labeling of medical events, which is an important factor for HAI detection. CONCLUSION: This study confirmed the feasibility of using NLP for the HAI detection in hospital facilities. Automatic HAI detection algorithms could offer better surveillance standardization for hospital comparisons.

TIER competency-based training course for the first receivers of CBRN casualties: a European perspective.

INTRODUCTION: Education and training are key elements of health system preparedness vis-à-vis chemical, biological, radiological and nuclear (CBRN) emergencies. Medical respondents need sufficient knowledge and skills to manage the human impact of CBRN events. OBJECTIVE: The current study was designed to determine which competencies are needed by hospital staff when responding to CBRN emergencies, define educational needs to develop these competencies, and implement a suitable delivery method. METHODS: This study was carried out from September 2014 to February 2015, using a three-step modified Delphi method. On the basis of international experiences, publications, and experts' consensus, core competencies for hospital staff - as CBRN casualty receivers - were determined, and training curricula and delivery methods were defined. RESULTS: The course consists of 10 domains. These are as follows: threat identification; health effects of CBRN agents; planning; hospital incident command system; information management; safety, personal protective equipment and decontamination; medical management; essential resources; psychological support; and ethical considerations. Expected competencies for each domain were defined. A blended approach was chosen. CONCLUSION: By identifying a set of core competencies, this study aimed to provide the specific knowledge and skills required by medical staff to respond to CRBN emergencies. A blended approach may be a suitable delivery method, allowing medical staff to attend the same training sessions despite different time zones and locations. The study output provides a CBRN training scheme that may be adapted and used at the European Union level.

Use of emergency department electronic medical records for automated epidemiological surveillance of suicide attempts: a French pilot study.

The aim of this study was to determine whether an expert system based on automated processing of electronic health records (EHRs) could provide a more accurate estimate of the annual rate of emergency department (ED) visits for suicide attempts in France, as compared to the current national surveillance system based on manual coding by emergency practitioners. A feasibility study was conducted at Lyon University Hospital, using data for all ED patient visits in 2012. After automatic data extraction and pre-processing, including automatic coding of medical free-text through use of the Unified Medical Language System, seven different machine-learning methods were used to classify the reasons for ED visits into "suicide attempts" versus "other reasons". The performance of these different methods was compared by using the F-measure. In a test sample of 444 patients admitted to the ED in 2012 (98 suicide attempts, 48 cases of suicidal ideation, and 292 controls with no recorded non-fatal suicidal behaviour), the F-measure for automatic detection of suicide attempts ranged from 70.4% to 95.3%. The random forest and naïve Bayes methods performed best. This study demonstrates that machine-learning methods can improve the quality of epidemiological indicators as compared to current national surveillance of suicide attempts.

Annotation methods to develop and evaluate an expert system based on natural language processing in electronic medical records.

The objective of the SYNODOS collaborative project was to develop a generic IT solution, combining a medical terminology server, a semantic analyser and a knowledge base. The goal of the project was to generate meaningful epidemiological data for various medical domains from the textual content of French medical records. In the context of this project, we built a care pathway oriented conceptual model and corresponding annotation method to develop and evaluate an expert system's knowledge base. The annotation method is based on a semi-automatic process, using a software application (MedIndex). This application exchanges with a cross-lingual multi-termino-ontology portal. The annotator selects the most appropriate medical code proposed for the medical concept in question by the multi-termino-ontology portal and temporally labels the medical concept according to the course of the medical event. This choice of conceptual model and annotation method aims to create a generic database of facts for the secondary use of electronic health records data.

Can epidemic detection systems at the hospital level complement regional surveillance networks: case study with the influenza epidemic?

BACKGROUND: Early knowledge of influenza outbreaks in the community allows local hospital healthcare workers to recognise the clinical signs of influenza in hospitalised patients and to apply effective precautions. The objective was to assess intra-hospital surveillance systems to detect earlier than regional surveillance systems influenza outbreaks in the community. METHODS: Time series obtained from computerized medical data from patients who visited a French hospital emergency department (ED) between June 1st, 2007 and March 31st, 2011 for influenza, or were hospitalised for influenza or a respiratory syndrome after an ED visit, were compared to different regional series. Algorithms using CUSUM method were constructed to determine the epidemic detection threshold with the local data series. Sensitivity, specificity and mean timeliness were calculated to assess their performance to detect community outbreaks of influenza. A sensitivity analysis was conducted, excluding the year 2009, due to the particular epidemiological situation related to pandemic influenza this year. RESULTS: The local series closely followed the seasonal trends reported by regional surveillance. The algorithms achieved a sensitivity of detection equal to 100% with series of patients hospitalised with respiratory syndrome (specificity ranging from 31.9 and 92.9% and mean timeliness from -58.3 to 20.3 days) and series of patients who consulted the ED for flu (specificity ranging from 84.3 to 93.2% and mean timeliness from -32.3 to 9.8 days). The algorithm with the best balance between specificity (87.7%) and mean timeliness (0.5 day) was obtained with series built by analysis of the ICD-10 codes assigned by physicians after ED consultation. Excluding the year 2009, the same series keeps the best performance with specificity equal to 95.7% and mean timeliness equal to -1.7 day. CONCLUSIONS: The implementation of an automatic surveillance system to detect patients with influenza or respiratory syndrome from computerized ED records could allow outbreak alerts at the intra-hospital level before the publication of regional data and could accelerate the implementation of preventive transmission-based precautions in hospital settings.

Evaluation of syndromic algorithms for detecting patients with potentially transmissible infectious diseases based on computerised emergency-department data.

BACKGROUND: The objective of this study was to ascertain the performance of syndromic algorithms for the early detection of patients in healthcare facilities who have potentially transmissible infectious diseases, using computerised emergency department (ED) data. METHODS: A retrospective cohort in an 810-bed University of Lyon hospital in France was analysed. Adults who were admitted to the ED and hospitalised between June 1, 2007, and March 31, 2010 were included (N=10895). Different algorithms were built to detect patients with infectious respiratory, cutaneous or gastrointestinal syndromes. The performance parameters of these algorithms were assessed with regard to the capacity of our infection-control team to investigate the detected cases. RESULTS: For respiratory syndromes, the sensitivity of the detection algorithms was 82.70%, and the specificity was 82.37%. For cutaneous syndromes, the sensitivity of the detection algorithms was 78.08%, and the specificity was 95.93%. For gastrointestinal syndromes, the sensitivity of the detection algorithms was 79.41%, and the specificity was 81.97%. CONCLUSIONS: This assessment permitted us to detect patients with potentially transmissible infectious diseases, while striking a reasonable balance between true positives and false positives, for both respiratory and cutaneous syndromes. The algorithms for gastrointestinal syndromes were not specific enough for routine use, because they generated a large number of false positives relative to the number of infected patients. Detection of patients with potentially transmissible infectious diseases will enable us to take precautions to prevent transmission as soon as these patients come in contact with healthcare facilities.

The use of regional platforms for managing electronic health records for the production of regional public health indicators in France.

BACKGROUND: In France, recent developments in healthcare system organization have aimed at strengthening decision-making and action in public health at the regional level. Firstly, the 2004 Public Health Act, by setting 100 national and regional public health targets, introduced an evaluative approach to public health programs at the national and regional levels. Meanwhile, the implementation of regional platforms for managing electronic health records (EHRs) has also been under assessment to coordinate the deployment of this important instrument of care within each geographic area. In this context, the development and implementation of a regional approach to epidemiological data extracted from EHRs are an opportunity that must be seized as soon as possible. Our article addresses certain design and organizational aspects so that the technical requirements for such use are integrated into regional platforms in France. The article will base itself on organization of the Rhône-Alpes regional health platform. DISCUSSION: Different tools being deployed in France allow us to consider the potential of these regional platforms for epidemiology and public health (implementation of a national health identification number and a national information system interoperability framework). The deployment of the Rhône-Alpes regional health platform began in the 2000s in France. By August 2011, 2.6 million patients were identified in this platform. A new development step is emerging because regional decision-makers need to measure healthcare efficiency. To pool heterogeneous information contained in various independent databases, the format, norm and content of the metadata have been defined. Two types of databases will be created according to the nature of the data processed, one for extracting structured data, and the second for extracting non-structured and de-identified free-text documents. SUMMARY: Regional platforms for managing EHRs could constitute an important data source for epidemiological surveillance in the context of epidemic alerts, but also in monitoring a number of indicators of infectious and chronic diseases for which no data are yet available in France.

Evaluation study of different strategies for detecting surgical site infections using the hospital information system at Lyon University Hospital, France.

OBJECTIVE: To evaluate different strategies for detecting surgical site infections (SSIs) using different sources (notification by the surgeon, bacteriological results, antibiotic prescription, and discharge diagnosis codes). BACKGROUND: Surveillance plays a role in reducing the risks of SSIs but the performance of case reports by surgeons is insufficient. Indirect methods of SSI detection are an alternative to increase the quality of surveillance. METHODS: A retrospective cohort study of 446 patients operated consecutively during the first half of 2007 was set up in a 56-bed general surgery unit in Lyon University Hospital, France. Patients were followed up 30 days after intervention. Different methods of detection were established by combining different data sources. The sensitivity and specificity of these methods were calculated by using, as reference method, the manual review of the medical records. RESULTS: The sensitivity and specificity of SSI detection were, respectively, 18.4% (95% confidence interval [CI]: 7.9-31.6) and 100% for surgeon notification; 63.2% (95% CI: 47.3-78.9) and 95.1% (95% CI: 92.9-97.1) for detection based on positive cultures; 68.4% (95% CI: 52.6-81.6) and 87.5% (95% CI: 84.3-90.7) using antibiotic prescription; 26.3% (95% CI: 13.2-42.1) and 99.5% (95% CI: 98.8-100) using discharge diagnosis codes. By combining the latter 3 sources, the sensitivity increased at 86.8% (95% CI: 76.3-97.4) and the specificity was lowered at 85.5% (95% CI: 82.1-89.0). CONCLUSIONS: SSI detection based on the combination of data extracted automatically from the hospital information system performed well. This strategy has been implemented gradually in Lyon University Hospital.

Early-onset ventilator-associated pneumonia incidence in intensive care units: a surveillance-based study.

BACKGROUND: The incidence of ventilator-associated pneumonia (VAP) within the first 48 hours of intensive care unit (ICU) stay has been poorly investigated. The objective was to estimate early-onset VAP occurrence in ICUs within 48 hours after admission. METHODS: We analyzed data from prospective surveillance between 01/01/2001 and 31/12/2009 in 11 ICUs of Lyon hospitals (France). The inclusion criteria were: first ICU admission, not hospitalized before admission, invasive mechanical ventilation during first ICU day, free of antibiotics at admission, and ICU stay ≥ 48 hours. VAP was defined according to a national protocol. Its incidence was the number of events per 1,000 invasive mechanical ventilation-days. The Poisson regression model was fitted from day 2 (D2) to D8 to incident VAP to estimate the expected VAP incidence from D0 to D1 of ICU stay. RESULTS: Totally, 367 (10.8%) of 3,387 patients in 45,760 patient-days developed VAP within the first 9 days. The predicted cumulative VAP incidence at D0 and D1 was 5.3 (2.6-9.8) and 8.3 (6.1-11.1), respectively. The predicted cumulative VAP incidence was 23.0 (20.8-25.3) at D8. The proportion of missed VAP within 48 hours from admission was 11% (9%-17%). CONCLUSIONS: Our study indicates underestimation of early-onset VAP incidence in ICUs, if only VAP occurring ≥ 48 hours are considered to be hospital-acquired. Clinicians should be encouraged to develop a strategy for early detection after ICU admission.

Evaluation of natural language processing from emergency department computerized medical records for intra-hospital syndromic surveillance.

BACKGROUND: The identification of patients who pose an epidemic hazard when they are admitted to a health facility plays a role in preventing the risk of hospital acquired infection. An automated clinical decision support system to detect suspected cases, based on the principle of syndromic surveillance, is being developed at the University of Lyon's Hôpital de la Croix-Rousse. This tool will analyse structured data and narrative reports from computerized emergency department (ED) medical records. The first step consists of developing an application (UrgIndex) which automatically extracts and encodes information found in narrative reports. The purpose of the present article is to describe and evaluate this natural language processing system. METHODS: Narrative reports have to be pre-processed before utilizing the French-language medical multi-terminology indexer (ECMT) for standardized encoding. UrgIndex identifies and excludes syntagmas containing a negation and replaces non-standard terms (abbreviations, acronyms, spelling errors...). Then, the phrases are sent to the ECMT through an Internet connection. The indexer's reply, based on Extensible Markup Language, returns codes and literals corresponding to the concepts found in phrases. UrgIndex filters codes corresponding to suspected infections. Recall is defined as the number of relevant processed medical concepts divided by the number of concepts evaluated (coded manually by the medical epidemiologist). Precision is defined as the number of relevant processed concepts divided by the number of concepts proposed by UrgIndex. Recall and precision were assessed for respiratory and cutaneous syndromes. RESULTS: Evaluation of 1,674 processed medical concepts contained in 100 ED medical records (50 for respiratory syndromes and 50 for cutaneous syndromes) showed an overall recall of 85.8% (95% CI: 84.1-87.3). Recall varied from 84.5% for respiratory syndromes to 87.0% for cutaneous syndromes. The most frequent cause of lack of processing was non-recognition of the term by UrgIndex (9.7%). Overall precision was 79.1% (95% CI: 77.3-80.8). It varied from 81.4% for respiratory syndromes to 77.0% for cutaneous syndromes. CONCLUSIONS: This study demonstrates the feasibility of and interest in developing an automated method for extracting and encoding medical concepts from ED narrative reports, the first step required for the detection of potentially infectious patients at epidemic risk.

The prevalence of celiac disease in Europe: results of a centralized, international mass screening project.

INTRODUCTION: Although the prevalence of celiac disease (CD) has been extensively investigated in recent years, an accurate estimate of CD frequency in the European population is still lacking. The aims of this study were: 1) to establish accurately the prevalence of CD in a large sample of the European population (Finland, Germany, Italy, and UK), including both children and adults; and 2) to investigate whether the prevalence of CD significantly varies between different areas of the European continent. MATERIALS AND METHODS: Samples were drawn from the four populations. All 29,212 participants were tested for CD by tissue transglutaminase (tTG) antibody test. Positive and border-line findings were further tested for serum endomysial antibodies (EMA). All serological determinations were centrally performed. Small-bowel biopsies were recommended to autoantibody-positive individuals. Previously diagnosed cases were identified. RESULTS: The overall CD prevalence (previously diagnosed plus anti-tTG and EMA positives) was 1.0% (95% CI 0.9-1.1). In subjects aged 30-64 years CD prevalence was 2.4% in Finland (2.0-2.8), 0.3% in Germany (0.1-0.4), and 0.7% in Italy (0.4-1.0). Sixty-eight percent of antibody-positive individuals showed small-bowel mucosal changes typical for CD (Marsh II/III lesion). CONCLUSIONS: CD is common in Europe. CD prevalence shows large unexplained differences in adult age across different European countries.

Evaluation of a French medical multi-terminology indexer for the manual annotation of natural language medical reports of healthcare-associated infections.

BACKGROUND: Surveillance of healthcare-associated infections is essential to prevention. A new collaborative project, namely ALADIN, was launched in January 2009 and aims to develop an automated detection tool based on natural language processing of medical documents. OBJECTIVE: The objective of this study was to evaluate the annotation of natural language medical reports of healthcare-associated infections. METHODS: A software MS Access application (NosIndex) has been developed to interface ECMT XML answer and manual annotation work. ECMT performances were evaluated by an infection control practitioner (ICP). Precision was evaluated for the 2 modules and recall only for the default module. Exclusion rate was defined as ratio between medical terms not found by ECMT and total number of terms evaluated. RESULTS: The medical discharge summaries were randomly selected in 4 medical wards. From the 247 medical terms evaluated, ECMT proposed 428 and 3,721 codes, respectively for the default and expansion modules. The precision was higher with the default module (P1=0.62) than with the expansion (P2=0.47). CONCLUSION: Performances of ECMT as support tool for the medical annotation were satisfactory.

Reduction of urinary tract infections acquired in an intensive care unit during a 10-year surveillance program.

PURPOSE: To describe trends of urinary catheter-related infections (UCRIs) acquired by patients hospitalized in intensive care units (ICU) in relation with an infection control program. MATERIALS AND METHODS: Prospective surveillance in one ICU of a university hospital in Lyon (France) between 1995 and 2004. RESULTS: A 66% reduction of urinary catheter-related infections (UCRIs) acquired by patients hospitalized was observed between 1995 and 2004 after adjustment on age, gender, antibiotic use at admission, and duration of exposure to urinary catheter. CONCLUSIONS: These results, obtained by continuous epidemiological monitoring of nosocomial infections, are encouraging with regard to the improvement of infection control measures and the evolution of medical practices. Further studies in ICUs are needed to confirm this trend.

Quality of information: a European challenge.

Since the end of the 1970s, many countries have started to set up programmes to control healthcare-associated infections (HAIs) and to achieve a safe and sustainable development of their healthcare facilities that minimises the risk of infection. Surveillance is a usual component of any organised programme to address the problem either at national, regional or local level. So a considerable effort has been made by the national Public Health Authorities of EU Member States to foster and extend the surveillance of HAI via the production of increasingly standardised indicators. This information is used by Infection Control teams to implement preventive strategies, to evaluate the magnitude of the problem and to understand variations in the risks of HAI. At the same time, Public Health authorities and healthcare financing agencies in several countries have attempted to generalise the production of such indicators at an official level and use them as a global approach for hospital quality assessment, accreditation, continuous quality improvement and communication with patients and the general population.

Mortality excess in individuals with elevated IgA anti-transglutaminase antibodies: the KORA/MONICA Augsburg cohort study 1989-1998.

OBJECTIVES: Immunoglobulin A (IgA) autoantibodies to tissue transglutaminase (tTG) are commonly used for screening and diagnosing of celiac disease. We examined the hypothesis that elevated IgA anti-tTG antibodies were associated with higher all-cause mortality risk. METHODS: The cohort, 2333 men and 2300 women, was based on the follow-up of participants of a representative population-based survey in Southern Germany (KORA/MONICA Augsburg project) conducted in 1989-1990. The endpoint for the vital status with cause of death was the year 1998. The sera drawn at baseline and stored at -80 degrees C, were recently screened with an IgA enzyme-linked immunosorbent assay (ELISA) using human recombinant tTG. Age-standardized mortality rates and age-adjusted hazard ratios were calculated. RESULTS: From the 4633 sera analyzed, 63 had an IgA anti-tTG concentration>or=7 AU/ml. Of these 63 individuals, 15 died between 1989 and 1998. The age-adjusted hazard ratio (HRa) of all-cause mortality was 1.86 (95% CI: 1.01-3.41) and 3.92 (95% CI: 1.44-10.71) for men and women, respectively. The excess of cancer mortality was even higher with an HR(a) of 2.47 (95% CI: 0.89-6.83) in men and of 6.65 (95% CI: 2.04-21.63) in women. CONCLUSIONS: Individuals with elevated IgA anti-tTG antibodies had a highly increased mortality risk, particularly due to cancer. New studies are necessary to clarify if this increased risk is due to undiagnosed celiac disease or/and if this elevated IgA anti-tTG antibodies level is a marker of serious diseases like cancer, chronic liver disease or end-stage heart failure.