BACKGROUND: Clinical diagnoses are typically made by following a series of steps recommended by guidelines that are authored by colleges of experts. Accordingly, guidelines play a crucial role in rationalizing clinical decisions. However, they suffer from limitations, as they are designed to cover the majority of the population and often fail to account for patients with uncommon conditions. Moreover, their updates are long and expensive, making them unsuitable for emerging diseases and new medical practices. METHODS: Inspired by guidelines, we formulate the task of diagnosis as a sequential decision-making problem and study the use of Deep Reinforcement Learning (DRL) algorithms to learn the optimal sequence of actions to perform in order to obtain a correct diagnosis from Electronic Health Records (EHRs), which we name a diagnostic decision pathway. We apply DRL to synthetic yet realistic EHRs and develop two clinical use cases: Anemia diagnosis, where the decision pathways follow a decision tree schema, and Systemic Lupus Erythematosus (SLE) diagnosis, which follows a weighted criteria score. We particularly evaluate the robustness of our approaches to noise and missing data, as these frequently occur in EHRs. RESULTS: In both use cases, even with imperfect data, our best DRL algorithms exhibit competitive performance compared to traditional classifiers, with the added advantage of progressively generating a pathway to the suggested diagnosis, which can both guide and explain the decision-making process. CONCLUSION: DRL offers the opportunity to learn personalized decision pathways for diagnosis. Our two use cases illustrate the advantages of this approach: they generate step-by-step pathways that are explainable, and their performance is competitive when compared to state-of-the-art methods.

SUMMARY: Phenotyping consists in applying algorithms to identify individuals associated with a specific, potentially complex, trait or condition, typically out of a collection of Electronic Health Records (EHRs). Because a lot of the clinical information of EHRs are lying in texts, phenotyping from text takes an important role in studies that rely on the secondary use of EHRs. However, the heterogeneity and highly specialized aspect of both the content and form of clinical texts makes this task particularly tedious, and is the source of time and cost constraints in observational studies. To facilitate the development, evaluation and reproducibility of phenotyping pipelines, we developed an open-source Python library named medkit. It enables composing data processing pipelines made of easy-to-reuse software bricks, named medkit operations. In addition to the core of the library, we share the operations and pipelines we already developed and invite the phenotyping community for their reuse and enrichment. AVAILABILITY AND IMPLEMENTATION: medkit is available at https://github.com/medkit-lib/medkit.

BACKGROUND: Clinical pathways are sequences of events followed during the clinical care of a group of patients who meet pre-defined criteria. They have many applications ranging from healthcare evaluation and optimization to clinical decision support. These pathways can be discovered from existing healthcare data, in particular with machine learning which is a family of methods used to learn patterns from data. This review provides a comprehensive overview of the literature concerning the use of machine learning methods for clinical pathway discovery from patient data. METHODS: Guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) method , we conducted a systematic review of the existing literature. We searched 6 databases, i.e., ACM Digital Library, ScienceDirect, Web of Science, PubMed, IEEE Xplore, and Scopus spanning from January 2004 to December 2023 using search terms pertinent to clinical pathways and their development. Subsequently, the retrieved papers were analyzed to assess their relevance to the scope of this study. RESULTS: In total, 131 papers that met the specified inclusion criteria were identified. These papers expressed diverse motivations behind data-driven clinical pathway discovery ranging from knowledge discovery to conformance checking with established clinical guidelines (derived from existing literature and clinical experts). Notably, the predominant methods employed (67.2%, n=88) involved unsupervised machine learning techniques, such as clustering and process mining. CONCLUSIONS: Relevant clinical pathways can be discovered from patient data using machine learning methods, with the desirable potential to aid clinical decision-making in healthcare. However, to reach this objective, the methods used to discover pathways should be reproducible, and rigorous performance evaluation by clinical experts needs to be conducted for validation.

BACKGROUND: The increasing complexity of data streams and computational processes in modern clinical health information systems makes reproducibility challenging. Clinical natural language processing (NLP) pipelines are routinely leveraged for the secondary use of data. Workflow management systems (WMS) have been widely used in bioinformatics to handle the reproducibility bottleneck. OBJECTIVE: To evaluate if WMS and other bioinformatics practices could impact the reproducibility of clinical NLP frameworks. MATERIALS AND METHODS: Based on the literature across multiple researcho fields (NLP, bioinformatics and clinical informatics) we selected articles which (1) review reproducibility practices and (2) highlight a set of rules or guidelines to ensure tool or pipeline reproducibility. We aggregate insight from the literature to define reproducibility recommendations. Finally, we assess the compliance of 7 NLP frameworks to the recommendations. RESULTS: We identified 40 reproducibility features from 8 selected articles. Frameworks based on WMS match more than 50% of features (26 features for LAPPS Grid, 22 features for OpenMinted) compared to 18 features for current clinical NLP framework (cTakes, CLAMP) and 17 features for GATE, ScispaCy, and Textflows. DISCUSSION: 34 recommendations are endorsed by at least 2 articles from our selection. Overall, 15 features were adopted by every NLP Framework. Nevertheless, frameworks based on WMS had a better compliance with the features. CONCLUSION: NLP frameworks could benefit from lessons learned from the bioinformatics field (eg, public repositories of curated tools and workflows or use of containers for shareability) to enhance the reproducibility in a clinical setting.

International audience

Word embeddings are shown to be a great asset for several Natural Language and Speech Processing tasks. While they are already evaluated on various NLP tasks, their evaluation on spoken or natural language understanding (SLU) is less studied. The goal of this study is two-fold: firstly, it focuses on semantic evaluation of common word embeddings approaches for SLU task; secondly, it investigates the use of two different data sets to train the embeddings: small and task-dependent corpus or huge and out-of-domain corpus. Experiments are carried out on 5 benchmark corpora (ATIS, SNIPS, SNIPS70, M2M, MEDIA), on which a relevance ranking was proposed in the literature. Interestingly, the performance of the embeddings is independent of the difficulty of the corpora. Moreover, the embeddings trained on huge and out-of-domain corpus yields to better results than the ones trained on small and task-dependent corpus.

INTRODUCTION: Clinical data warehouses are often oriented toward integration and exploration of coded data. However narrative reports are of crucial importance for translational research. This paper describes Dr. Warehouse®, an open source data warehouse oriented toward clinical narrative reports and designed to support clinicians' day-to-day use. METHOD: Dr. Warehouse relies on an original database model to focus on documents in addition to facts. Besides classical querying functionalities, the system provides an advanced search engine and Graphical User Interfaces adapted to the exploration of text. Dr. Warehouse is dedicated to translational research with cohort recruitment capabilities, high throughput phenotyping and patient centric views (including similarity metrics among patients). These features leverage Natural Language Processing based on the extraction of UMLS® concepts, as well as negation and family history detection. RESULTS: A survey conducted after 6 months of use at the Necker Children's Hospital shows a high rate of satisfaction among the users (96.6%). During this period, 122 users performed 2837 queries, accessed 4,267 patients' records and included 36,632 patients in 131 cohorts. The source code is available at this github link https://github.com/imagine-bdd/DRWH. A demonstration based on PubMed abstracts is available at https://imagine-plateforme-bdd.fr/dwh_pubmed/.

OBJECTIVE: The repurposing of electronic health records (EHRs) can improve clinical and genetic research for rare diseases. However, significant information in rare disease EHRs is embedded in the narrative reports, which contain many negated clinical signs and family medical history. This paper presents a method to detect family history and negation in narrative reports and evaluates its impact on selecting populations from a clinical data warehouse (CDW). MATERIALS AND METHODS: We developed a pipeline to process 1.6 million reports from multiple sources. This pipeline is part of the load process of the Necker Hospital CDW. RESULTS: We identified patients with "Lupus and diarrhea," "Crohn's and diabetes," and "NPHP1" from the CDW. The overall precision, recall, specificity, and F-measure were 0.85, 0.98, 0.93, and 0.91, respectively. CONCLUSION: The proposed method generates a highly accurate identification of cases from a CDW of rare disease EHRs.

OBJECTIVE: In the context of rare diseases, it may be helpful to detect patients with similar medical histories, diagnoses and outcomes from a large number of cases with automated methods. To reduce the time to find new cases, we developed a method to find similar patients given an index case leveraging data from the electronic health records. MATERIALS AND METHODS: We used the clinical data warehouse of a children academic hospital in Paris, France (Necker-Enfants Malades), containing about 400,000 patients. Our model was based on a vector space model (VSM) to compute the similarity distance between an index patient and all the patients of the data warehouse. The dimensions of the VSM were built upon Unified Medical Language System concepts extracted from clinical narratives stored in the clinical data warehouse. The VSM was enhanced using three parameters: a pertinence score (TF-IDF of the concepts), the polarity of the concept (negated/not negated) and the minimum number of concepts in common. We evaluated this model by displaying the most similar patients for five different rare diseases: Lowe Syndrome (LOWE), Dystrophic Epidermolysis Bullosa (DEB), Activated PI3K delta Syndrome (APDS), Rett Syndrome (RETT) and Dowling Meara (EBS-DM), from the clinical data warehouse representing 18, 103, 21, 84 and 7 patients respectively. RESULTS: The percentages of index patients returning at least one true positive similar patient in the Top30 similar patients were 94% for LOWE, 97% for DEB, 86% for APDS, 71% for EBS-DM and 99% for RETT. The mean number of patients with the exact same genetic diseases among the 30 returned patients was 51%. CONCLUSION: This tool offers new perspectives in a translational context to identify patients for genetic research. Moreover, when new molecular bases are discovered, our strategy will help to identify additional eligible patients for genetic screening.

Phenome-Wide Association Studies (PheWAS) investigate whether genetic polymorphisms associated with a phenotype are also associated with other diagnoses. In this study, we have developed new methods to perform a PheWAS based on ICD-10 codes and biological test results, and to use a quantitative trait as the selection criterion. We tested our approach on thiopurine S-methyltransferase (TPMT) activity in patients treated by thiopurine drugs. We developed 2 aggregation methods for the ICD-10 codes: an ICD-10 hierarchy and a mapping to existing ICD-9-CM based PheWAS codes. Eleven biological test results were also analyzed using discretization algorithms. We applied these methods in patients having a TPMT activity assessment from the clinical data warehouse of a French academic hospital between January 2000 and July 2013. Data after initiation of thiopurine treatment were analyzed and patient groups were compared according to their TPMT activity level. A total of 442 patient records were analyzed representing 10,252 ICD-10 codes and 72,711 biological test results. The results from the ICD-9-CM based PheWAS codes and ICD-10 hierarchy codes were concordant. Cross-validation with the biological test results allowed us to validate the ICD phenotypes. Iron-deficiency anemia and diabetes mellitus were associated with a very high TPMT activity (p = 0.0004 and p = 0.0015, respectively). We describe here an original method to perform PheWAS on a quantitative trait using both ICD-10 diagnosis codes and biological test results to identify associated phenotypes. In the field of pharmacogenomics, PheWAS allow for the identification of new subgroups of patients who require personalized clinical and therapeutic management.