Search

Objective: Recognizing diseases from discharge letters is crucial for cohort selection and epidemiological analyses, as this is the only type of data consistently produced across hospitals. This is a classic document classification problem, typically requiring supervised learning. However, manual annotation of large datasets of discharge letters is uncommon since it is extremely time-consuming. We propose a novel weakly-supervised pipeline to recognize diseases from Italian discharge letters. Methods: Our Natural Language Processing pipeline is based on a fine-tuned version of the Italian Umberto model. The pipeline extracts diagnosis-related sentences from a subset of letters and applies a two-level clustering using the embeddings generated by the fine-tuned Umberto model. These clusters are summarized and those mapped to the diseases of interest are selected as weak labels. Finally, the same BERT-based model is trained using these weak labels to detect the targeted diseases. Results: A case study related to the identification of bronchiolitis with 33'176 Italian discharge letters from 44 hospitals in the Veneto Region shows the potential of our method, with an AUC of 77.7 % and an F1-Score of 75.1 % on manually annotated labels, improving compared to other non-supervised methods and with a limited loss compared to fully supervised methods. Results are robust to the cluster selection and the identified clusters highlight the potential to recognize a variety of diseases. Conclusions: This study demonstrates the feasibility of diagnosis identification from Italian discharge letters in the absence of labelled data. Our pipeline showed strong performance and robustness, and its flexibility allows for easy adaptation to various diseases. This approach offers a scalable solution for clinical text classification, reducing the need for manual annotation while maintaining good accuracy., Comment: 39 pages, 4 figures

Evaluating hospitals' performance and its relation to patients' characteristics is of utmost importance to ensure timely, effective, and optimal treatment. Such a matter is particularly relevant in areas and situations where the healthcare system must contend with an unexpected surge in hospitalizations, such as for heart failure patients in the Lombardy region of Italy during the COVID-19 pandemic. Motivated by this issue, the paper introduces a novel Multilevel Logistic Cluster-Weighted Model (ML-CWMd) for predicting 45-day mortality following hospitalization due to COVID-19. The methodology flexibly accommodates dependence patterns among continuous, categorical, and dichotomous variables; effectively accounting for hospital-specific effects in distinct patient subgroups showing different attributes. A tailored Expectation-Maximization algorithm is developed for parameter estimation, and extensive simulation studies are conducted to evaluate its performance against competing models. The novel approach is applied to administrative data from the Lombardy Region, aiming to profile heart failure patients hospitalized for COVID-19 and investigate the hospital-level impact on their overall mortality. A scenario analysis demonstrates the model's efficacy in managing multiple sources of heterogeneity, thereby yielding promising results in aiding healthcare providers and policy-makers in the identification of patient-specific treatment pathways., Comment: 29 pages, 12 figures. It includes also the supplementary material

Globally, the incidence of heart failure is increasing, and its principal treatment involves drug therapy. However, widespread non-adherence to therapies is prevalent among heart failure patients and often results in worsening health conditions and an increase in hospital admissions. This study aims to develop an innovative approach, the State-Sequence analysis, to profile heart failure patients based on different drug-utilization patterns. These patterns aim to capture both the multidimensional and dynamic effects of therapies. Subsequently, the study explores how combining clustering algorithms with this technique influences overall patient survival. Findings highlight the importance of continued drug therapy after the first hospitalization in improving heart failure prognosis, irrespective of its severity. The proposed approach can assist healthcare specialists in evaluating the pathways provided to patients, allowing for a change in analysis from a transversal and syntactical approach to a holistic one that leverages statistical tools that are slightly more complex than traditional methods. Moreover, because of the many options available for defining states, temporal granularity, and spacing metrics, SSA is a flexible method applicable to various epidemiological scenarios.

Accurately estimating Origin-Destination (OD) matrices is a topic of increasing interest for efficient transportation network management and sustainable urban planning. Traditionally, travel surveys have supported this process; however, their availability and comprehensiveness can be limited. Moreover, the recent COVID-19 pandemic has triggered unprecedented shifts in mobility patterns, underscoring the urgency of accurate and dynamic mobility data supporting policies and decisions with data-driven evidence. In this study, we tackle these challenges by introducing an innovative pipeline for estimating dynamic OD matrices. The real motivating problem behind this is based on the Trenord railway transportation network in Lombardy, Italy. We apply a novel approach that integrates ticket and subscription sales data with passenger counts obtained from Automated Passenger Counting (APC) systems, making use of the Iterative Proportional Fitting (IPF) algorithm. Our work effectively addresses the complexities posed by incomplete and diverse data sources, showcasing the adaptability of our pipeline across various transportation contexts. Ultimately, this research bridges the gap between available data sources and the escalating need for precise OD matrices. The proposed pipeline fosters a comprehensive grasp of transportation network dynamics, providing a valuable tool for transportation operators, policymakers, and researchers. Indeed, to highlight the potentiality of dynamic OD matrices, we showcase some methods to perform anomaly detection of mobility trends in the network through such matrices and interpret them in light of events that happened in the last months of 2022., Comment: Codes available. Synthetic Data available. Application to train network data. 27 pages, 6 Tables, 16 Figures

In the context of clinical and biomedical studies, joint frailty models have been developed to study the joint temporal evolution of recurrent and terminal events, capturing both the heterogeneous susceptibility to experiencing a new episode and the dependence between the two processes. While discretely-distributed frailty is usually more exploitable by clinicians and healthcare providers, existing literature on joint frailty models predominantly assumes continuous distributions for the random effects. In this article, we present a novel joint frailty model that assumes bivariate discretely-distributed non-parametric frailties, with an unknown finite number of mass points. This approach facilitates the identification of latent structures among subjects, grouping them into sub-populations defined by a shared frailty value. We propose an estimation routine via Expectation-Maximization algorithm, which not only estimates the number of subgroups but also serves as an unsupervised classification tool. This work is motivated by a study of patients with Heart Failure (HF) receiving ACE inhibitors treatment in the Lombardia region of Italy. Recurrent events of interest are hospitalizations due to HF and terminal event is death for any cause.

In recent years, research interest in personalised treatments has been growing. However, treatment effect heterogeneity and possibly time-varying treatment effects are still often overlooked in clinical studies. Statistical tools are needed for the identification of treatment response patterns, taking into account that treatment response is not constant over time. We aim to provide an innovative method to obtain dynamic treatment effect phenotypes on a time-to-event outcome, conditioned on a set of relevant effect modifiers. The proposed method does not require the assumption of proportional hazards for the treatment effect, which is rarely realistic. We propose a spline-based survival neural network, inspired by the Royston-Parmar survival model, to estimate time-varying conditional treatment effects. We then exploit the functional nature of the resulting estimates to apply a functional clustering of the treatment effect curves in order to identify different patterns of treatment effects. The application that motivated this work is the discontinuation of treatment with Mineralocorticoid receptor Antagonists (MRAs) in patients with heart failure, where there is no clear evidence as to which patients it is the safest choice to discontinue treatment and, conversely, when it leads to a higher risk of adverse events. The data come from an electronic health record database. A simulation study was performed to assess the performance of the spline-based neural network and the stability of the treatment response phenotyping procedure. In light of the results, the suggested approach has the potential to support personalized medical choices by assessing unique treatment responses in various medical contexts over a period of time.

Survival analysis is a fundamental tool in medicine, modeling the time until an event of interest occurs in a population. However, in real-world applications, survival data are often incomplete, censored, distributed, and confidential, especially in healthcare settings where privacy is critical. The scarcity of data can severely limit the scalability of survival models to distributed applications that rely on large data pools. Federated learning is a promising technique that enables machine learning models to be trained on multiple datasets without compromising user privacy, making it particularly well-suited for addressing the challenges of survival data and large-scale survival applications. Despite significant developments in federated learning for classification and regression, many directions remain unexplored in the context of survival analysis. In this work, we propose an extension of the Federated Survival Forest algorithm, called FedSurF++. This federated ensemble method constructs random survival forests in heterogeneous federations. Specifically, we investigate several new tree sampling methods from client forests and compare the results with state-of-the-art survival models based on neural networks. The key advantage of FedSurF++ is its ability to achieve comparable performance to existing methods while requiring only a single communication round to complete. The extensive empirical investigation results in a significant improvement from the algorithmic and privacy preservation perspectives, making the original FedSurF algorithm more efficient, robust, and private. We also present results on two real-world datasets demonstrating the success of FedSurF++ in real-world healthcare studies. Our results underscore the potential of FedSurF++ to improve the scalability and effectiveness of survival analysis in distributed settings while preserving user privacy.

This study aims to analyse the effects of reducing Received Dose Intensity (RDI) in chemotherapy treatment for osteosarcoma patients on their survival by using a novel approach. In this scenario, toxic side effects are risk factors for mortality and predictors of future exposure levels, introducing post-assignment confounding. Chemotherapy administration data from BO03 and BO06 Randomized Clinical Trials (RCTs) in ostosarcoma are employed to emulate a target trial with three RDI-based exposure strategies: 1) standard, 2) reduced, and 3) highly-reduced RDI. Investigations are conducted between subgroups of patients characterised by poor or good Histological Responses (HRe). Inverse Probability of Treatment Weighting (IPTW) is first used to transform the original population into a pseudo-population which mimics the target randomized cohort. Then, a Marginal Structural Cox Model with effect modification is employed. Conditional Average Treatment Effects (CATEs) are ultimately measured as the difference between the Restricted Mean Survival Time of reduced/highly-reduced RDI strategy and the standard one. Confidence Intervals for CATEs are obtained using a novel IPTW-based bootstrap procedure. Significant effect modifications based on HRe were found. Increasing RDI-reductions led to contrasting trends for poor and good responders: the higher the reduction, the better (worsen) was the survival in poor (good) reponders. This study introduces a novel approach to (i) comprehensively address the challenges related to the analysis of chemotherapy data, (ii) mitigate the toxicity-treatment-adjustment bias, and (iii) repurpose existing RCT data for retrospective analyses extending beyond the original trials' intended scopes.

Heart failure (HF) is a severe and costly clinical syndrome associated with increased healthcare costs and a high burden of mortality and morbidity. Although drug therapy is the mainstay of treatment for heart failure, non-adherence to prescribed therapies is common and is associated with worse health outcomes and increased hospitalizations. In this study, we propose a novel approach using Latent Markov models to analyze drug adherence to polytherapy over time using a secondary database. Our methodology enables us to evaluate patients' drug utilization behaviour, identify complex behavioural patterns, and incorporate them into predictive models to improve clinical outcomes. The significance of adhering to prescribed therapies for patients' prognosis has been highlighted in this study. Our findings show that adherent patients gained an additional ten months of life over a seven-year follow-up period compared to non-adherent patients. This underscores the importance of secondary prevention and continuous monitoring of heart failure patients. These procedures are crucial for identifying areas of improvement and promoting better adherence to prescribed therapies., Comment: 15 pages; 5 figures; 4 tables

We introduce a novel statistical significance-based approach for clustering hierarchical data using semi-parametric linear mixed-effects models designed for responses with laws in the exponential family (e.g., Poisson and Bernoulli). Within the family of semi-parametric mixed-effects models, a latent clustering structure of the highest-level units can be identified by assuming the random effects to follow a discrete distribution with an unknown number of support points. We achieve this by computing {\alpha}-level confidence regions of the estimated support point and identifying statistically different clusters. At each iteration of a tailored Expectation Maximization algorithm, the two closest estimated support points for which the confidence regions overlap collapse. Unlike the related state-of-the-art methods that rely on arbitrary thresholds to determine the merging of close discrete masses, the proposed approach relies on conventional statistical confidence levels, thereby avoiding the use of discretionary tuning parameters. To demonstrate the effectiveness of our approach, we apply it to data from the Programme for International Student Assessment (PISA - OECD) to cluster countries based on the rate of innumeracy levels in schools. Additionally, a simulation study and comparison with classical parametric and state-of-the-art models are provided and discussed.

The comprehension of the mechanisms behind the mobility of skilled workers is of paramount importance for policy making. The lacking nature of official measurements motivates the use of digital trace data extracted from ORCID public records. We use such data to investigate European regions, studied at NUTS2 level, over the time horizon of 2009 to 2020. We present a novel perspective where regions roles are dictated by the overall activity of the research community, contradicting the common brain drain interpretation of the phenomenon. We find that a high mobility is usually correlated with strong university prestige, high magnitude of investments and an overall good schooling level in a region.

The concept of care pathways is increasingly being used to enhance the quality of care and to optimize the use of resources for health care. Nevertheless, recommendations regarding the sequence of care are mostly based on consensus-based decisions as there is a lack of evidence on effective treatment sequences. In a real-world setting, classical statistical tools resulted to be insufficient to adequately consider a phenomenon with such high variability and has to be integrated with novel data mining techniques suitable of identifying patterns in complex data structures. Data-driven techniques can potentially support the empirical identification of effective care sequences by extracting them from data collected routinely. The purpose of this study is to perform sequence analysis to identify different patterns of treatment and to assess the most efficient in preventing adverse events. The clinical application that motivated the study of this method concerns the several problems frequently encountered in the quality of care provided in the mental health field. In particular, we analyzed administrative data provided by Regione Lombardia related to all the beneficiaries of the National Health Service with a diagnosis of schizophrenia from 2015 to 2018 resident in Lombardy, a region of northern Italy. This methodology considers the patient's therapeutic path as a conceptual unit, i.e., a sequence, composed of a succession of different states that can describe longitudinal patient's status. This kind of information, such as common patterns of care that allowed us to risk profile patients, can provide health policymakers an opportunity to plan optimum and individualized patient care by allocating appropriate resources, analyzing trends in the health status of a population, and finding the risk factors that can be leveraged to prevent the decline of mental health status at the population level.

Personalized medicine is the future of medical practice. In oncology, tumor heterogeneity assessment represents a pivotal step for effective treatment planning and prognosis prediction. Despite new procedures for DNA sequencing and analysis, non-invasive methods for tumor characterization are needed to impact on daily routine. On purpose, imaging texture analysis is rapidly scaling, holding the promise to surrogate histopathological assessment of tumor lesions. In this work, we propose a tree-based representation strategy for describing intra-tumor heterogeneity of patients affected by metastatic cancer. We leverage radiomics information extracted from PET/CT imaging and we provide an exhaustive and easily readable summary of the disease spreading. We exploit this novel patient representation to perform cancer subtyping according to hierarchical clustering technique. To this purpose, a new heterogeneity-based distance between trees is defined and applied to a case study of prostate cancer. Clusters interpretation is explored in terms of concordance with severity status, tumor burden and biological characteristics. Results are promising, as the proposed method outperforms current literature approaches. Ultimately, the proposed method draws a general analysis framework that would allow to extract knowledge from daily acquired imaging data of patients and provide insights for effective treatment planning.

Statistical methods to study the association between a longitudinal biomarker and the risk of death are very relevant for the long-term care of subjects affected by chronic illnesses, such as potassium in heart failure patients. Particularly in the presence of comorbidities or pharmacological treatments, sudden crises can cause potassium to undergo very abrupt yet transient changes. In the context of the monitoring of potassium, there is a need for a dynamic model that can be used in clinical practice to assess the risk of death related to an observed patient's potassium trajectory. We considered different dynamic survival approaches, starting from the simple approach considering the most recent measurement, to the joint model. We then propose a novel method based on wavelet filtering and landmarking to retrieve the prognostic role of past short-term potassium shifts. We argue that while taking into account past information is important, not all past information is equally informative. State-of-the-art dynamic survival models are prone to give more importance to the mean long-term value of potassium. However, our findings suggest that it is essential to take into account also recent potassium instability to capture all the relevant prognostic information. The data used comes from over 2000 subjects, with a total of over 80 000 repeated potassium measurements collected through Administrative Health Records and Outpatient and Inpatient Clinic E-charts. A novel dynamic survival approach is proposed in this work for the monitoring of potassium in heart failure. The proposed wavelet landmark method shows promising results revealing the prognostic role of past short-term changes, according to their different duration, and achieving higher performances in predicting the survival probability of individuals.

Recent evidence highlights the usefulness of DNA methylation (DNAm) biomarkers as surrogates for exposure to risk factors for non-communicable diseases in epidemiological studies and randomized trials. DNAm variability has been demonstrated to be tightly related to lifestyle behavior and exposure to environmental risk factors, ultimately providing an unbiased proxy of an individual state of health. At present, the creation of DNAm surrogates relies on univariate penalized regression models, with elastic-net regularizer being the gold standard when accomplishing the task. Nonetheless, more advanced modeling procedures are required in the presence of multivariate outcomes with a structured dependence pattern among the study samples. In this work we propose a general framework for mixed-effects multitask learning in presence of high-dimensional predictors to develop a multivariate DNAm biomarker from a multi-center study. A penalized estimation scheme based on an expectation-maximization algorithm is devised, in which any penalty criteria for fixed-effects models can be conveniently incorporated in the fitting process. We apply the proposed methodology to create novel DNAm surrogate biomarkers for multiple correlated risk factors for cardiovascular diseases and comorbidities. We show that the proposed approach, modeling multiple outcomes together, outperforms state-of-the-art alternatives, both in predictive power and bio-molecular interpretation of the results.

Due to the presence of multiple types of adverse events with different levels of severity, the analysis of longitudinal toxicity data is a difficult task in cancer studies. In this work, a novel approach based on latent Markov models and compositional data techniques is proposed. The latent status of interest is the Latent Overall Toxicity (LOTox) condition of each patient. The main objectives consist in identifying different latent states of overall toxicity burden and investigating the evolution of individual toxicity risk during cancer treatment. This methodology is applied to osteosarcoma treatment data to provide novel techniques that may support medical decisions in childhood cancer therapy.

EEG technology finds applications in several domains. Currently, most EEG systems require subjects to wear several electrodes on the scalp to be effective. However, several channels might include noisy information, redundant signals, induce longer preparation times and increase computational times of any automated system for EEG decoding. One way to reduce the signal-to-noise ratio and improve classification accuracy is to combine channel selection with feature extraction, but EEG signals are known to present high inter-subject variability. In this work we introduce a novel algorithm for subject-independent channel selection of EEG recordings. Considering multi-channel trial recordings as statistical units and the EEG decoding task as the class of reference, the algorithm (i) exploits channel-specific 1D-Convolutional Neural Networks (1D-CNNs) as feature extractors in a supervised fashion to maximize class separability; (ii) it reduces a high dimensional multi-channel trial representation into a unique trial vector by concatenating the channels' embeddings and (iii) recovers the complex inter-channel relationships during channel selection, by exploiting an ensemble of AutoEncoders (AE) to identify from these vectors the most relevant channels to perform classification. After training, the algorithm can be exploited by transferring only the parametrized subgroup of selected channel-specific 1D-CNNs to new signals from new subjects and obtain low-dimensional and highly informative trial vectors to be fed to any classifier.

Class imbalance is a common issue in many domain applications of learning algorithms. Oftentimes, in the same domains it is much more relevant to correctly classify and profile minority class observations. This need can be addressed by Feature Selection (FS), that offers several further advantages, s.a. decreasing computational costs, aiding inference and interpretability. However, traditional FS techniques may become sub-optimal in the presence of strongly imbalanced data. To achieve FS advantages in this setting, we propose a filtering FS algorithm ranking feature importance on the basis of the Reconstruction Error of a Deep Sparse AutoEncoders Ensemble (DSAEE). We use each DSAE trained only on majority class to reconstruct both classes. From the analysis of the aggregated Reconstruction Error, we determine the features where the minority class presents a different distribution of values w.r.t. the overrepresented one, thus identifying the most relevant features to discriminate between the two. We empirically demonstrate the efficacy of our algorithm in several experiments on high-dimensional datasets of varying sample size, showcasing its capability to select relevant and generalizable features to profile and classify minority class, outperforming other benchmark FS methods. We also briefly present a real application in radiogenomics, where the methodology was applied successfully.

Logistic Regression (LR) is a widely used statistical method in empirical binary classification studies. However, real-life scenarios oftentimes share complexities that prevent from the use of the as-is LR model, and instead highlight the need to include high-order interactions to capture data variability. This becomes even more challenging because of: (i) datasets growing wider, with more and more variables; (ii) studies being typically conducted in strongly imbalanced settings; (iii) samples going from very large to extremely small; (iv) the need of providing both predictive models and interpretable results. In this paper we present a novel algorithm, Learning high-order Interactions via targeted Pattern Search (LIPS), to select interaction terms of varying order to include in a LR model for an imbalanced binary classification task when input data are categorical. LIPS's rationale stems from the duality between item sets and categorical interactions. The algorithm relies on an interaction learning step based on a well-known frequent item set mining algorithm, and a novel dissimilarity-based interaction selection step that allows the user to specify the number of interactions to be included in the LR model. In addition, we particularize two variants (Scores LIPS and Clusters LIPS), that can address even more specific needs. Through a set of experiments we validate our algorithm and prove its wide applicability to real-life research scenarios, showing that it outperforms a benchmark state-of-the-art algorithm.

This paper proposes a clustering procedure for samples of multivariate functions in $(L^2(I))^{J}$, with $J\geq1$. This method is based on a k-means algorithm in which the distance between the curves is measured with a metrics that generalizes the Mahalanobis distance in Hilbert spaces, considering the correlation and the variability along all the components of the functional data. The proposed procedure has been studied in simulation and compared with the k-means based on other distances typically adopted for clustering multivariate functional data. In these simulations, it is shown that the k-means algorithm with the generalized Mahalanobis distance provides the best clustering performances, both in terms of mean and standard deviation of the number of misclassified curves. Finally, the proposed method has been applied to two real cases studies, concerning ECG signals and growth curves, where the results obtained in simulation are confirmed and strengthened., Comment: 22 pages