Emmanuel Chazard

Toward a Literature-Driven Definition of Big Data in Healthcare

Objective. The aim of this study was to provide a definition of big data in healthcare. Methods. A systematic search of PubMed literature published until May 9, 2014, was conducted. We noted the number of statistical individuals (n) and the number of variables (p) for all papers describing a dataset. These papers were classified into fields of study. Characteristics attributed to big data by authors were also considered. Based on this analysis, a definition of big data was proposed. Results. A total of 196 papers were included. Big data can be defined as datasets with Log⁡(n∗p) ≥ 7. Properties of big data are its great variety and high velocity. Big data raises challenges on veracity, on all aspects of the workflow, on extracting meaningful information, and on sharing information. Big data requires new computational methods that optimize data management. Related concepts are data reuse, false knowledge discovery, and privacy issues. Conclusion. Big data is defined by volume. Big data should not be confused with data reuse: data can be big without being reused for another purpose, for example, in omics. Inversely, data can be reused without being necessarily big, for example, secondary use of Electronic Medical Records (EMR) data.

Data Mining to Generate Adverse Drug Events Detection Rules

Adverse drug events (ADEs) are a public health is sue. Their detection usually relies on voluntary reporting or medical chart reviews. The objective of this paper is to automatically detect cases of ADEs by data mining. 115 447 complete past hospital stays are extracted from six French, Danish, and Bulgarian hospitals using a common data model including diagnoses, drug administrations, laboratory results, and free-text records. Different kinds of outcomes are traced, and supervised rule induction methods (decision trees and association rules) are used to discover ADE detection rules, with respect to time constraints. The rules are then filtered, validated, and reorganized by a committee of experts. The rules are described in a rule repository, and several statistics are automatically computed in every medical department, such as the confidence, relative risk, and median delay of outcome appearance. 236 validated ADE-detection rules are discovered; they enable to detect 27 different kinds of outcomes. The rules use a various number of conditions related to laboratory results, diseases, drug administration, and demographics. Some rules involve innovative conditions, such as drug discontinuations.

The Value of Body Weight Measurement to Assess Dehydration in Children

Dehydration secondary to gastroenteritis is one of the most common reasons for office visits and hospital admissions. The indicator most commonly used to estimate dehydration status is acute weight loss. Post-illness weight gain is considered as the gold-standard to determine the true level of dehydration and is widely used to estimate weight loss in research. To determine the value of post-illness weight gain as a gold standard for acute dehydration, we conducted a prospective cohort study in which 293 children, aged 1 month to 2 years, with acute diarrhea were followed for 7 days during a 3-year period. The main outcome measures were an accurate pre-illness weight (if available within 8 days before the diarrhea), post-illness weight, and theoretical weight (predicted from the child’s individual growth chart). Post-illness weight was measured for 231 (79%) and both theoretical and post-illness weights were obtained for 111 (39%). Only 62 (21%) had an accurate pre-illness weight. The correlation between post-illness and theoretical weight was excellent (0.978), but bootstrapped linear regression analysis showed that post-illness weight underestimated theoretical weight by 0.48 kg (95% CI: 0.06–0.79, p<0.02). The mean difference in the fluid deficit calculated was 4.0% of body weight (95% CI: 3.2–4.7, p<0.0001). Theoretical weight overestimated accurate pre-illness weight by 0.21 kg (95% CI: 0.08–0.34, p = 0.002). Post-illness weight underestimated pre-illness weight by 0.19 kg (95% CI: 0.03–0.36, p = 0.02). The prevalence of 5% dehydration according to post-illness weight (21%) was significantly lower than the prevalence estimated by either theoretical weight (60%) or clinical assessment (66%, p<0.0001).These data suggest that post-illness weight is of little value as a gold standard to determine the true level of dehydration. The performance of dehydration signs or scales determined by using post-illness weight as a gold standard has to be reconsidered.

Lossless watermarking of categorical attributes for verifying medical data base integrity

In this article, we propose a new lossless or reversible watermarking approach that allows the embedding of a message within categorical data of relational database. The reversibility property of our scheme is achieved by adapting the well known histogram shifting modulation. Based on this algorithm we derive a system for verifying the integrity of the database content, it means detecting addition, removal or modification of any t-uples or attributes. Such a content integrity check is independent of the manner the database is stored or structured. We illustrate the overall capability of our method and its constraints of deployment considering one medical database of inpatient hospital stay records. Especially, we reversibly watermark ICD-10 diagnostic codes.

From adverse drug event detection to prevention. A novel clinical decision support framework for medication safety.

BACKGROUND Errors related to medication seriously affect patient safety and the quality of healthcare. It has been widely argued that various types of such errors may be prevented by introducing Clinical Decision Support Systems (CDSSs) at the point of care. OBJECTIVES Although significant research has been conducted in the field, still medication safety is a crucial issue, while few research outcomes are mature enough to be considered for use in actual clinical settings. In this paper, we present a clinical decision support framework targeting medication safety with major focus on adverse drug event (ADE) prevention. METHODS The novelty of the framework lies in its design that approaches the problem holistically, i.e., starting from knowledge discovery to provide reliable numbers about ADEs per hospital or medical unit to describe their consequences and probable causes, and next employing the acquired knowledge for decision support services development and deployment. Major design features of the frameworks services are: a) their adaptation to the context of care (i.e. patient characteristics, place of care, and significance of ADEs), and b) their straightforward integration in the healthcare information technologies (IT) infrastructure thanks to the adoption of a service-oriented architecture (SOA) and relevant standards. RESULTS Our results illustrate the successful interoperability of the framework with two commercially available IT products, i.e., a Computerized Physician Order Entry (CPOE) and an Electronic Health Record (EHR) system, respectively, along with a Web prototype that is independent of existing healthcare IT products. The conducted clinical validation with domain experts and test cases illustrates that the impact of the framework is expected to be major, with respect to patient safety, and towards introducing the CDSS functionality in practical use. CONCLUSIONS This study illustrates an important potential for the applicability of the presented framework in delivering contextualized decision support services at the point of care and for making a substantial contribution towards ADE prevention. Nonetheless, further research is required in order to quantitatively and thoroughly assess its impact in medication safety.

Proposal and evaluation of FASDIM, a Fast And Simple De-Identification Method for unstructured free-text clinical records.

PURPOSE Medical free-text records enable to get rich information about the patients, but often need to be de-identified by removing the Protected Health Information (PHI), each time the identification of the patient is not mandatory. Pattern matching techniques require pre-defined dictionaries, and machine learning techniques require an extensive training set. Methods exist in French, but either bring weak results or are not freely available. The objective is to define and evaluate FASDIM, a Fast And Simple De-Identification Method for French medical free-text records. METHODS FASDIM consists in removing all the words that are not present in the authorized word list, and in removing all the numbers except those that match a list of protection patterns. The corresponding lists are incremented in the course of the iterations of the method. For the evaluation, the workload is estimated in the course of records de-identification. The efficiency of the de-identification is assessed by independent medical experts on 508 discharge letters that are randomly selected and de-identified by FASDIM. Finally, the letters are encoded after and before de-identification according to 3 terminologies (ATC, ICD10, CCAM) and the codes are compared. RESULTS The construction of the list of authorized words is progressive: 12h for the first 7000 letters, 16 additional hours for 20,000 additional letters. The Recall (proportion of removed Protected Health Information, PHI) is 98.1%, the Precision (proportion of PHI within the removed token) is 79.6% and the F-measure (harmonic mean) is 87.9%. In average 30.6 terminology codes are encoded per letter, and 99.02% of those codes are preserved despite the de-identification. CONCLUSION FASDIM gets good results in French and is freely available. It is easy to implement and does not require any predefined dictionary.

DicomWorks Teleradiology: Secure transmission of medical images over the Internet at low cost

We developed a completely secured teleradiology solution tailored for e-mail teleradiology applications at low- cost. Data processing consists in creating a couple of files with an encrypted and compressed image archive and a 128 bits decoding key file. No proprietary file format or encryption scheme is used. Files are exchanged using the e-mail (SMTP and POP) protocols, but FTP or sFTP can be used for better performances. Software includes original features such as realtime interactive JPEG compression, instant archive preview or secured data cleanup when a user logs off. We believe that this flexible, integrated and easy to use solution is a robust alternative to more complex architectures for simple image transmissions or occasional circumstances.

Personalized and automated remote monitoring of atrial fibrillation

AIMS Remote monitoring of cardiac implantable electronic devices is a growing standard; yet, remote follow-up and management of alerts represents a time-consuming task for physicians or trained staff. This study evaluates an automatic mechanism based on artificial intelligence tools to filter atrial fibrillation (AF) alerts based on their medical significance. METHODS AND RESULTS We evaluated this method on alerts for AF episodes that occurred in 60 pacemaker recipients. AKENATON prototype workflow includes two steps: natural language-processing algorithms abstract the patient health record to a digital version, then a knowledge-based algorithm based on an applied formal ontology allows to calculate the CHA2DS2-VASc score and evaluate the anticoagulation status of the patient. Each alert is then automatically classified by importance from low to critical, by mimicking medical reasoning. Final classification was compared with human expert analysis by two physicians. A total of 1783 alerts about AF episode >5 min in 60 patients were processed. A 1749 of 1783 alerts (98%) were adequately classified and there were no underestimation of alert importance in the remaining 34 misclassified alerts. CONCLUSION This work demonstrates the ability of a pilot system to classify alerts and improves personalized remote monitoring of patients. In particular, our method allows integration of patient medical history with device alert notifications, which is useful both from medical and resource-management perspectives. The system was able to automatically classify the importance of 1783 AF alerts in 60 patients, which resulted in an 84% reduction in notification workload, while preserving patient safety.

Adverse drug events with hyperkalaemia during inpatient stays: evaluation of an automated method for retrospective detection in hospital databases

BackgroundAdverse drug reactions and adverse drug events (ADEs) are major public health issues. Many different prospective tools for the automated detection of ADEs in hospital databases have been developed and evaluated. The objective of the present study was to evaluate an automated method for the retrospective detection of ADEs with hyperkalaemia during inpatient stays.MethodsWe used a set of complex detection rules to take account of the patient’s clinical and biological context and the chronological relationship between the causes and the expected outcome. The dataset consisted of 3,444 inpatient stays in a French general hospital. An automated review was performed for all data and the results were compared with those of an expert chart review. The complex detection rules’ analytical quality was evaluated for ADEs.ResultsIn terms of recall, 89.5% of ADEs with hyperkalaemia “with or without an abnormal symptom” were automatically identified (including all three serious ADEs). In terms of precision, 63.7% of the automatically identified ADEs with hyperkalaemia were true ADEs.ConclusionsThe use of context-sensitive rules appears to improve the automated detection of ADEs with hyperkalaemia. This type of tool may have an important role in pharmacoepidemiology via the routine analysis of large inter-hospital databases.

Clinical evaluation of the ADE scorecards as a decision support tool for adverse drug event analysis and medication safety management.

The prevention of adverse drug events (ADEs) demands co‐ordination of different health care professionals. ADE scorecards are a novel approach to raise the team awareness regarding ADE risks and causes. It makes information on numbers and on possible causes of possible ADE cases available to the clinical team. The aim of the study was to investigate the usage and acceptance of ADE scorecards by healthcare professionals and their impact on rates of possible ADEs.