An AI-based approach driven by genotypes and phenotypes to uplift the diagnostic yield of genetic diseases.

Identifying disease-causing variants in Rare Disease patients' genome is a challenging problem. To accomplish this task, we describe a machine learning framework, that we called "Suggested Diagnosis", whose aim is to prioritize genetic variants in an exome/genome based on the probability of being disease-causing. To do so, our method leverages standard guidelines for germline variant interpretation as defined by the American College of Human Genomics (ACMG) and the Association for Molecular Pathology (AMP), inheritance information, phenotypic similarity, and variant quality. Starting from (1) the VCF file containing proband's variants, (2) the list of proband's phenotypes encoded in Human Phenotype Ontology terms, and optionally (3) the information about family members (if available), the "Suggested Diagnosis" ranks all the variants according to their machine learning prediction. This method significantly reduces the number of variants that need to be evaluated by geneticists by pinpointing causative variants in the very first positions of the prioritized list. Most importantly, our approach proved to be among the top performers within the CAGI6 Rare Genome Project Challenge, where it was able to rank the true causative variant among the first positions and, uniquely among all the challenge participants, increased the diagnostic yield of 12.5% by solving 2 undiagnosed cases.

Clusters of individuals recovering from an exacerbation of chronic obstructive pulmonary disease and response to in-hospital pulmonary rehabilitation.

INTRODUCTION AND OBJECTIVES: Due to the present low availability of pulmonary rehabilitation (PR) for individuals recovering from a COPD exacerbation (ECOPD), we need admission priority criteria. We tested the hypothesis that these individuals might be clustered according to baseline characteristics to identify subpopulations with different responses to PR. METHODS: Multicentric retrospective analysis of individuals undergone in-hospital PR. Baseline characteristics and outcome measures (six-minute walking test - 6MWT, Medical Research Council scale for dyspnoea -MRC, COPD assessment test -CAT) were used for clustering analysis. RESULTS: Data analysis of 1159 individuals showed that after program, the proportion of individuals reaching the minimal clinically important difference (MCID) was 85.0%, 86.3%, and 65.6% for CAT, MRC, and 6MWT respectively. Three clusters were found (C1-severe: 10.9%; C2-intermediate: 74.4%; C3-mild: 14.7% of cases respectively). Cluster C1-severe showed the worst conditions with the largest post PR improvements in outcome measures; C3-mild showed the least severe baseline conditions, but the smallest improvements. The proportion of participants reaching the MCID in ALL three outcome measures was significantly different among clusters, with C1-severe having the highest proportion of full success (69.0%) as compared to C2-intermediate (48.3%) and C3-mild (37.4%). Participants in C2-intermediate and C1-severe had 1.7- and 4.6-fold increases in the probability to reach the MCID in all three outcomes as compared to those in C3-mild (OR = 1.72, 95% confidence interval [95% CI] = 1.2 - 2.49, p = 0.0035 and OR = 4.57, 95% CI = 2.68 - 7.91, p < 0.0001 respectively). CONCLUSIONS: Clustering analysis can identify subpopulations of individuals recovering from ECOPD associated with different responses to PR. Our results may help in defining priority criteria based on the probability of success of PR.

Improving Keyword-Based Topic Classification in Cancer Patient Forums with Multilingual Transformers.

Online forums play an important role in connecting people who have crossed paths with cancer. These communities create networks of mutual support that cover different cancer-related topics, containing an extensive amount of heterogeneous information that can be mined to get useful insights. This work presents a case study where users' posts from an Italian cancer patient community have been classified combining both count-based and prediction-based representations to identify discussion topics, with the aim of improving message reviewing and filtering. We demonstrate that pairing simple bag-of-words representations based on keywords matching with pre-trained contextual embeddings significantly improves the overall quality of the predictions and allows the model to handle ambiguities and misspellings. By using non-English real-world data, we also investigated the reusability of pretrained multilingual models like BERT in lower data regimes like many local medical institutions.

A continuous-time Markov model approach for modeling myelodysplastic syndromes progression from cross-sectional data.

The integration of both genomics and clinical data to model disease progression is now possible, thanks to the increasing availability of molecular patients' profiles. This may lead to the definition of novel decision support tools, able to tailor therapeutic interventions on the basis of a "precise" patients' risk stratification, given their health status evolution. However, longitudinal analysis requires long-term data collection and curation, which can be time demanding, expensive and sometimes unfeasible. Here we present a clinical decision support framework that combines the simulation of disease progression from cross-sectional data with a Markov model that exploits continuous-time transition probabilities derived from Cox regression. Trajectories between patients at different disease stages are stochastically built according to a measure of patient similarity, computed with a matrix tri-factorization technique. Such trajectories are seen as realizations drawn from the stochastic process driving the transitions between the disease stages. Eventually, Markov models applied to the resulting longitudinal dataset highlight potentially relevant clinical information. We applied our method to cross-sectional genomic and clinical data from a cohort of Myelodysplastic syndromes (MDS) patients. MDS are heterogeneous clonal hematopoietic disorders whose patients are characterized by different risks of Acute Myeloid Leukemia (AML) development, defined by an international score. We computed patients' trajectories across increasing and subsequent levels of risk of developing AML, and we applied a Cox model to the simulated longitudinal dataset to assess whether genomic characteristics could be associated with a higher or lower probability of disease progression. We then used the learned parameters of such Cox model to calculate the transition probabilities of a continuous-time Markov model that describes the patients' evolution across stages. Our results are in most cases confirmed by previous studies, thus demonstrating that simulated longitudinal data represent a valuable resource to investigate disease progression of MDS patients.

Ontology-Driven Real World Evidence Extraction from Clinical Narratives.

Unstructured clinical notes contain a huge amount of information. We investigated the possibility of harvesting such information through an NLP-based approach. A manually curated ontology is the only resource required to handle all the steps of the process leading from clinical narrative to a structured data warehouse (i2b2). We have tested our approach at the Papa Giovanni XXIII hospital in Bergamo (Italy) on pathology reports collected since 2008.

Patient similarity for precision medicine: A systematic review.

Evidence-based medicine is the most prevalent paradigm adopted by physicians. Clinical practice guidelines typically define a set of recommendations together with eligibility criteria that restrict their applicability to a specific group of patients. The ever-growing size and availability of health-related data is currently challenging the broad definitions of guideline-defined patient groups. Precision medicine leverages on genetic, phenotypic, or psychosocial characteristics to provide precise identification of patient subsets for treatment targeting. Defining a patient similarity measure is thus an essential step to allow stratification of patients into clinically-meaningful subgroups. The present review investigates the use of patient similarity as a tool to enable precision medicine. 279 articles were analyzed along four dimensions: data types considered, clinical domains of application, data analysis methods, and translational stage of findings. Cancer-related research employing molecular profiling and standard data analysis techniques such as clustering constitute the majority of the retrieved studies. Chronic and psychiatric diseases follow as the second most represented clinical domains. Interestingly, almost one quarter of the studies analyzed presented a novel methodology, with the most advanced employing data integration strategies and being portable to different clinical domains. Integration of such techniques into decision support systems constitutes and interesting trend for future research.

Multicentre registry of brain-injured patients with disorder of consciousness: rationale and preliminary data.

Diagnostic accuracy and reliable estimation of clinical evolution are challenging issues in the management of patients with disorders of consciousness (DoC). Longitudinal systematic investigations conducted in large cohorts of patients with DoC could make it possible to identify reliable diagnostic and prognostic markers. On the basis of this consideration, we devised a multicentre prospective registry for patients with DoC admitted to ten intensive rehabilitation units. The registry collects homogeneous and detailed data on patients' demographic and clinical features, neurophysiological and neuroimaging findings, and medical and surgical complications. Here we present the rationale and the design of the registry and the preliminary results obtained in 53 patients with DoC (vegetative state or minimally conscious state) enrolled during the first seven months of the study. Data at 6-month post-injury follow-up were available for 46 of them. This registry could be an important tool for collecting high-quality data through the application of rigorous methods, and it could be used in the routine management of patients with DoC admitted to rehabilitation settings.

Trusting telemedicine: A discussion on risks, safety, legal implications and liability of involved stakeholders.

OBJECTIVES: The main purpose of the article is to raise awareness among all the involved stakeholders about the risks and legal implications connected to the development and use of modern telemedicine systems. Particular focus is given to the class of "active" telemedicine systems, that imply a real-world, non-mediated, interaction with the final user. A secondary objective is to give an overview of the European legal framework that applies to these systems, in the effort to avoid defensive medicine practices and fears, which might be a barrier to their broader adoption. METHODS: We leverage on the experience gained during two international telemedicine projects, namely MobiGuide (pilot studies conducted in Spain and Italy) and AP@home (clinical trials enrolled patients in Italy, France, the Netherlands, United Kingdom, Austria and Germany), whose development our group has significantly contributed to in the last 4 years, to create a map of the potential criticalities of active telemedicine systems and comment upon the legal framework that applies to them. Two workshops have been organized in December 2015 and March 2016 where the topic has been discussed in round tables with system developers, researchers, physicians, nurses, legal experts, healthcare economists and administrators. RESULTS: We identified 8 features that generate relevant risks from our example use cases. These features generalize to a broad set of telemedicine applications, and suggest insights on possible risk mitigation strategies. We also discuss the relevant European legal framework that regulate this class of systems, providing pointers to specific norms and highlighting possible liability profiles for involved stakeholders. CONCLUSIONS: Patients are more and more willing to adopt telemedicine systems to improve home care and day-by-day self-management. An essential step towards a broader adoption of these systems consists in increasing their compliance with existing regulations and better defining responsibilities for all the involved stakeholders.

Patient similarity by joint matrix trifactorization to identify subgroups in acute myeloid leukemia.

OBJECTIVE: Computing patients' similarity is of great interest in precision oncology since it supports clustering and subgroup identification, eventually leading to tailored therapies. The availability of large amounts of biomedical data, characterized by large feature sets and sparse content, motivates the development of new methods to compute patient similarities able to fuse heterogeneous data sources with the available knowledge. MATERIALS AND METHODS: In this work, we developed a data integration approach based on matrix trifactorization to compute patient similarities by integrating several sources of data and knowledge. We assess the accuracy of the proposed method: (1) on several synthetic data sets which similarity structures are affected by increasing levels of noise and data sparsity, and (2) on a real data set coming from an acute myeloid leukemia (AML) study. The results obtained are finally compared with the ones of traditional similarity calculation methods. RESULTS: In the analysis of the synthetic data set, where the ground truth is known, we measured the capability of reconstructing the correct clusters, while in the AML study we evaluated the Kaplan-Meier curves obtained with the different clusters and measured their statistical difference by means of the log-rank test. In presence of noise and sparse data, our data integration method outperform other techniques, both in the synthetic and in the AML data. DISCUSSION: In case of multiple heterogeneous data sources, a matrix trifactorization technique can successfully fuse all the information in a joint model. We demonstrated how this approach can be efficiently applied to discover meaningful patient similarities and therefore may be considered a reliable data driven strategy for the definition of new research hypothesis for precision oncology. CONCLUSION: The better performance of the proposed approach presents an advantage over previous methods to provide accurate patient similarities supporting precision medicine.

Temporal electronic phenotyping by mining careflows of breast cancer patients.

In this work we present a careflow mining approach designed to analyze heterogeneous longitudinal data and to identify phenotypes in a patient cohort. The main idea underlying our approach is to combine methods derived from sequential pattern mining and temporal data mining to derive frequent healthcare histories (careflows) in a population of patients. This approach was applied to an integrated data repository containing clinical and administrative data of more than 4000 breast cancer patients. We used the mined histories to identify sub-cohorts of patients grouped according to healthcare activities pathways, then we characterized these sub-cohorts with clinical data. In this way, we were able to perform temporal electronic phenotyping of electronic health records (EHR) data.

Comparison of data mining techniques applied to fetal heart rate parameters for the early identification of IUGR fetuses.

The onset of fetal pathologies can be screened during pregnancy by means of Fetal Heart Rate (FHR) monitoring and analysis. Noticeable advances in understanding FHR variations were obtained in the last twenty years, thanks to the introduction of quantitative indices extracted from the FHR signal. This study searches for discriminating Normal and Intra Uterine Growth Restricted (IUGR) fetuses by applying data mining techniques to FHR parameters, obtained from recordings in a population of 122 fetuses (61 healthy and 61 IUGRs), through standard CTG non-stress test. We computed N=12 indices (N=4 related to time domain FHR analysis, N=4 to frequency domain and N=4 to non-linear analysis) and normalized them with respect to the gestational week. We compared, through a 10-fold crossvalidation procedure, 15 data mining techniques in order to select the more reliable approach for identifying IUGR fetuses. The results of this comparison highlight that two techniques (Random Forest and Logistic Regression) show the best classification accuracy and that both outperform the best single parameter in terms of mean AUROC on the test sets.

Template for preparation of papers for IEEE sponsored conferences & symposia.

To improve the access to medical information is necessary to design and implement integrated informatics techniques aimed to gather data from different and heterogeneous sources. This paper describes the technologies used to integrate data coming from the electronic medical record of the IRCCS Fondazione Maugeri (FSM) hospital of Pavia, Italy, and combines them with administrative, pharmacy drugs purchase coming from the local healthcare agency (ASL) of the Pavia area and environmental open data of the same region. The integration process is focused on data coming from a cohort of one thousand patients diagnosed with Type 2 Diabetes Mellitus (T2DM). Data analysis and temporal data mining techniques have been integrated to enhance the initial dataset allowing the possibility to stratify patients using further information coming from the mined data like behavioral patterns of prescription-related drug purchases and other frequent clinical temporal patterns, through the use of an intuitive dashboard controlled system.

From data to the decision: A software architecture to integrate predictive modelling in clinical settings.

The application of statistics and mathematics over large amounts of data is providing healthcare systems with new tools for screening and managing multiple diseases. Nonetheless, these tools have many technical and clinical limitations as they are based on datasets with concrete characteristics. This proposition paper describes a novel architecture focused on providing a validation framework for discrimination and prediction models in the screening of Type 2 diabetes. For that, the architecture has been designed to gather different data sources under a common data structure and, furthermore, to be controlled by a centralized component (Orchestrator) in charge of directing the interaction flows among data sources, models and graphical user interfaces. This innovative approach aims to overcome the data-dependency of the models by providing a validation framework for the models as they are used within clinical settings.

Big data and biomedical informatics: a challenging opportunity.

Big data are receiving an increasing attention in biomedicine and healthcare. It is therefore important to understand the reason why big data are assuming a crucial role for the biomedical informatics community. The capability of handling big data is becoming an enabler to carry out unprecedented research studies and to implement new models of healthcare delivery. Therefore, it is first necessary to deeply understand the four elements that constitute big data, namely Volume, Variety, Velocity, and Veracity, and their meaning in practice. Then, it is mandatory to understand where big data are present, and where they can be beneficially collected. There are research fields, such as translational bioinformatics, which need to rely on big data technologies to withstand the shock wave of data that is generated every day. Other areas, ranging from epidemiology to clinical care, can benefit from the exploitation of the large amounts of data that are nowadays available, from personal monitoring to primary care. However, building big data-enabled systems carries on relevant implications in terms of reproducibility of research studies and management of privacy and data access; proper actions should be taken to deal with these issues. An interesting consequence of the big data scenario is the availability of new software, methods, and tools, such as map-reduce, cloud computing, and concept drift machine learning algorithms, which will not only contribute to big data research, but may be beneficial in many biomedical informatics applications. The way forward with the big data opportunity will require properly applied engineering principles to design studies and applications, to avoid preconceptions or over-enthusiasms, to fully exploit the available technologies, and to improve data processing and data management regulations.

Multivariate analysis based on linear and non-linear FHR parameters for the identification of IUGR fetuses.

Fetal Heart Rate (FHR) monitoring represents a powerful tool for checking the arousal of pathological fetal conditions during pregnancy. This paper proposes a multivariate approach for the discrimination of Normal and Intra Uterine Growth Restricted (IUGR) fetuses based on a small set of parameters computed on the FHR signal. We collected FHR recordings in a population of 120 fetuses (60 normals and 60 IUGRs) at approximately the same gestational week through a standard CTG non-stress test. A set of 8 linear and non-linear indices were selected and computed on each recording, on the basis of their "stand-alone" discriminative properties, demonstrated in previous studies. By using the Orange® data mining suite we checked various multivariate discrimination models. The results show that a Logistic Regression performed on a limited set of only 4 parameters can reach 92.5% accuracy in the correct identification of fetuses, with 93% sensitivity and 91.5% specificity.

A probabilistic method for computing quantitative risk indexes from medical injuries compensation claims.

BACKGROUND: The increasing demand of health care services and the complexity of health care delivery require Health Care Organizations (HCOs) to approach clinical risk management through proper methods and tools. An important aspect of risk management is to exploit the analysis of medical injuries compensation claims in order to reduce adverse events and, at the same time, to optimize the costs of health insurance policies. OBJECTIVES: This work provides a probabilistic method to estimate the risk level of a HCO by computing quantitative risk indexes from medical injury compensation claims. METHODS: Our method is based on the estimate of a loss probability distribution from compensation claims data through parametric and non-parametric modeling and Monte Carlo simulations. The loss distribution can be estimated both on the whole dataset and, thanks to the application of a Bayesian hierarchical model, on stratified data. The approach allows to quantitatively assessing the risk structure of the HCO by analyzing the loss distribution and deriving its expected value and percentiles. RESULTS: We applied the proposed method to 206 cases of injuries with compensation requests collected from 1999 to the first semester of 2007 by the HCO of Lodi, in the Northern part of Italy. We computed the risk indexes taking into account the different clinical departments and the different hospitals involved. CONCLUSIONS: The approach proved to be useful to understand the HCO risk structure in terms of frequency, severity, expected and unexpected loss related to adverse events.

Supporting translational research on inherited cardiomyopathies through information technology.

OBJECTIVES: The INHERITANCE project, funded by the European Commission, is aimed at studying genetic or inherited Dilated cardiomyopathies (DCM) and at understanding the impact and management of the disease within families that suffer from heart conditions that are caused by DCMs. The biomedical informatics research activity of the project aims at implementing information technology solutions to support the project team in the different phases of their research, in particular in genes screening prioritization and new gene-disease association discovery. METHODS: In order to manage the huge quantity of scientific, clinical and patient data generated by the project several advanced biomedical informatics tools have been developed. The paper describes a layer of software instruments to support translation of the results of the project in clinical practice as well as to support the scientific discovery process. This layer includes data warehousing, intelligent querying of the phenotype data, integrated search of biological data and knowledge repositories, text mining of the relevant literature, and case based reasoning. RESULTS: At the moment, a set of 1,394 patients and 9,784 observations has been stored into the INHERITANCE data warehouse. The literature database contains more than 1,100,000 articles retrieved from the Pubmed and generically related to cardiac diseases, already analyzed for extracting medical concepts and genes. CONCLUSIONS: After two years of project the data warehouse has been completely set up and the text mining tools for automatic literature analysis have been implemented and tested. A first prototype of the decision support tool for knowledge discovery and gene prioritization is available, but a more complete release is still under development.

Trustworthy reuse of health data: a transnational perspective.

BACKGROUND: The widespread adoption of electronic health records (EHRs) is accelerating the collection of sensitive clinical data. The availability of these data raises privacy concerns, yet sharing the data is essential for public health, longitudinal patient care, and clinical research. METHOD: Following previous work in the United States [1,2], the International Medical Informatics Association convened the 2012 European Summit on Trustworthy Reuse of Health Data. Over 100 delegates representing national governments, academia, patient groups, industry, and the European Commission participated. In all, 21 countries were represented. The agenda was designed to solicit a wide range of perspectives on trustworthy reuse of health data from the participants. RESULTS AND CONCLUSIONS: Delegates agreed that the "government" should provide oversight, that the reuse should be "fully regulated," and that the patient should be "fully informed." One important reflection was that doing nothing will have negative implications across the European Union (EU). First, continued fragmented parallel non-standards-based developments in multiple sectors entail a substantial duplication of costs and human effort. Second, a failure to work jointly across the stakeholders on common policy frameworks will forego a crucial opportunity to boost key EU markets (pharmaceuticals, health technology and devices, and eHealth solutions) and counter global competition. Finally, and crucially, the lack of harmonized policy across EU nations for trustworthy reuse of health data risks patient safety. The productive dialog, initiated with multiple stakeholders from government, academia, and industry, will have to continue, in order to address the many remaining issues outlined in this white paper.

Supporting regenerative medicine by integrative dimensionality reduction.

OBJECTIVE: The assessment of the developmental potential of stem cells is a crucial step towards their clinical application in regenerative medicine. It has been demonstrated that genome-wide expression profiles can predict the cellular differentiation stage by means of dimensionality reduction methods. Here we show that these techniques can be further strengthened to support decision making with i) a novel strategy for gene selection; ii) methods for combining the evidence from multiple data sets. METHODS: We propose to exploit dimensionality reduction methods for the selection of genes specifically activated in different stages of differentiation. To obtain an integrated predictive model, the expression values of the selected genes from multiple data sets are combined. We investigated distinct approaches that either aggregate data sets or use learning ensembles. RESULTS: We analyzed the performance of the proposed methods on six publicly available data sets. The selection procedure identified a reduced subset of genes whose expression values gave rise to an accurate stage prediction. The assessment of predictive accuracy demonstrated a high quality of predictions for most of the data integration methods presented. CONCLUSION: The experimental results highlighted the main potentials of proposed approaches. These include the ability to predict the true staging by combining multiple training data sets when this could not be inferred from a single data source, and to focus the analysis on a reduced list of genes of similar predictive performance.