Background: Cardiovascular sequelae after COVID-19 are frequent. However, the predictors for their occurrence are still unknown. In this study, we aimed to assess whether myocardial injury during COVID-19 hospitalization is associated to CV sequelae and death after hospital discharge. Methods: In this prospective observational study, consecutive patients who were admitted for COVID-19 in a metropolitan COVID-19 hub in Italy, between March 2021 and January 2022, with a ≥ 1 assessment of high sensitivity cardiac troponin I (hs-cTnI) were included in the study, if they were alive at hospital discharge. Myocardial injury was defined as elevation hs-cTnI > 99th percentile of the upper reference limit. The incidence of all-cause mortality and major adverse cardiovascular and cerebrovascular events (MACCE, including cardiovascular death, admission for acute or chronic coronary syndrome, hospitalization for heart failure, and stroke/transient ischemic attack) at follow-up were the primary outcomes. Arrhythmias, inflammatory heart diseases, and/or thrombotic disorders were analyzed as well. Results: Among the 701 COVID-19 survivors (mean age 66.4 ± 14.4 years, 40.2% female), myocardial injury occurred in 75 (10.7%) patients. At a median follow-up of 270 days (IQR 165, 380), all-cause mortality (21.3% vs. 6.1%, p < 0.001), MACCE (25.3% vs. 4.5%, p < 0.001), arrhythmias (9.3% vs. 5.0%, p = 0.034), and inflammatory heart disease (8.0% vs. 1.1%, p < 0.001) were more frequent in patients with myocardial injury compared to those without. At multivariate analysis, myocardial injury (HR 1.95 [95% CI:1.05–3.61]), age (HR 1.09 [95% CI:1.06–1.12]), and chronic kidney disease (HR 2.63 [95% CI:1.33–5.21]) were independent predictors of death. Myocardial injury (HR 3.92 [95% CI:2.07–7.42]), age (HR 1.05 [95% CI:1.02–1.08]), and diabetes (HR 2.35 [95% CI:1.25–4.43]) were independent predictors of MACCE. Conclusion: In COVID-19 survivors, myocardial injury during the hospital stay portends a higher risk of mortality and cardiovascular sequelae and could be considered for the risk stratification of COVID-19 sequelae in patients who are successfully discharged.

Background: Systemic lupus erythematosus (SLE) is a chronic disease characterized by a broad spectrum of involved organs, including neurological, renal, and vascular domains, with disease activity manifesting through unpredictable patterns that vary across individuals and over time, making the prediction of activity events particularly challenging. Objective: This paper proposes a hierarchical machine learning model to predict a 12-month SLE activity, defined as the occurrence of at least one event among SLE hospitalization, new organ-involved domain, and neurological, renal, or vascular manifestation within the following year. At each patient’s visit, the model considers all the features at the current time point, the information about the patient’s clinical history, and about its last 12 months, to predict the outcome for the next 12 months. Methods: The study cohort consists of 262 patients with at least an outpatient visit and an SLE admission from 2012 to 2020, at the Italian Gemelli Hospital, comprising a retrospective longitudinal dataset of 5962 contacts. The data include demographics, laboratory, clinical features (eg, domain involvements and manifestations), treatments, and pathways (eg, contact types as outpatients, hospitalizations, day hospitals, and visit frequency). The variables consider 3 time ranges: features about the current contact and the last 12 months, and the previous patient’s clinical history. The main model was developed by testing different machine learning approaches within a cross-validation setup. The predicted probability outputs were used in a risk stratification analysis, identifying 3 groups of predictions: strong, moderate, and mild. Mild samples were then passed through a second cascade model. The integration of the main model (applied to strong and moderate samples) with the cascade model (applied to mild contacts) forms our final hierarchical model. Results: The hierarchical model, resulting from the ensemble of the main random forest and cascade decision tree, demonstrated enhanced performance, increasing the area under the receiver operating characteristic curve from 0.696 (95% CI 0.672‐0.719) in the original main model to 0.743 (95% CI 0.717-0.769), particularly for specific patient characteristics. Through the application of explainable artificial intelligence methods, we also identified the key features that significantly influence the model’s predictions. Among the 185 collected features, 15 emerged as the most impactful, including age at contact, response to therapy modifications, abnormal laboratory tests, and clinical manifestations. This analysis plays a crucial role in enhancing model transparency, which is essential for fostering the adoption of artificial intelligence in health care settings. Conclusions: Our study introduces an explainable and reliable tool for predicting 1-year SLE activity, supporting physicians with an advanced decision-support system to improve patient management. The model identifies key features that may help characterize patient phenotypes, enabling personalized treatment plans and better outcomes. In addition, the methodology can be generalized for predictive analytics in other chronic autoimmune diseases.

Livia Lilli, Stefano Patarnello, Carlotta Masciocchi, Valeria Masiello, Fabio Marazzi, Tagliaferri Luca, Nikola Capocchiano. Proceedings of the 6th Clinical Natural Language Processing Workshop. 2024.

Electronic Health Records (EHRs) contain a wealth of unstructured patient data, making it challenging for physicians to do informed decisions. In this paper, we introduce a Natural Language Processing (NLP) approach for the extraction of therapies, diagnosis, and symptoms from ambulatory EHRs of patients with chronic Lupus disease. We aim to demonstrate the effort of a comprehensive pipeline where a rule-based system is combined with text segmentation, transformer-based topic analysis and clinical ontology, in order to enhance text preprocessing and automate rules’ identification. Our approach is applied on a sub-cohort of 56 patients, with a total of 750 EHRs written in Italian language, achieving an Accuracy and an F-score over 97% and 90% respectively, in the three extracted domains. This work has the potential to be integrated with EHR systems to automate information extraction, minimizing the human intervention, and providing personalized digital solutions in the chronic Lupus disease domain.

Natural language processing is emerging as a key application of artificial intelligence in oncology. This systematic review aims to evaluate the performance and methodological frameworks of natural language processing systems in gynecologic oncology. We conducted a systematic review following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis 2020 guidelines. MEDLINE, EMBASE, and Web of Science were searched for studies published between January 2015 and February 2025. Outcomes were synthesized across 3 research questions: the accuracy of natural language processing systems used as consultation service platforms; the accuracy of natural language processing systems used as clinical decision support systems; and the benchmarking methodologies applied, including their associated methodological outcomes. Consultation service platforms deliver general medical information, whereas clinical decision support systems provide recommendations that are integrated into the patient's clinical workflow. This review analyzed 12 retrospective studies. Consultation service platforms were less accurate than clinicians (60% vs 86.7%) and rated lower in response quality (2.96/5 vs 4.2/5) but outperformed guideline-based answers (1.54/2 vs 1.38/2). In cervical cancer, ChatGPT surpassed experts (7.0 vs 6.1). ChatGPT-4 showed a concordance of 70% with the National Comprehensive Cancer Network and 60% with the European Society of Gynaecological Oncology guidelines in clinical decision support tasks, with an overall recommendation accuracy of 75%. IBM Watson achieved a 72.8% concordance with guidelines. Prompting was applied from 100% to 37.5% across studies. Qualitative benchmarking varied across studies: 83.3% used clinical guidelines and 37.5% of consultation service platforms studies used expert answers. Four- or 5-point scales and binary scoring were used to assess consultation service platforms and clinical decision support systems, respectively. Clinicians remain superior in complex reasoning, but natural language processing systems demonstrate robust performance in guideline-driven tasks, with advantages in speed, readability, and reproducibility. However, performance declined in nuanced scenarios and among under-represented patient sub-groups. Large language models currently play a supportive rather than substitutive role in gynecologic oncology.

Objectives: Electronic health records (EHRs) contain a wealth of unstructured patient data that can be leveraged using artificial intelligence (AI). This study aimed to develop a natural language processing (NLP) pipeline to identify clinical phenotypes and disease trajectories in patients with systemic lupus erythematosus (SLE) from EHRs. Methods: EHR data from SLE patients were included. A standardized stepwise framework combining AI and human intelligence (HI) was designed. Ontology-based definitions were developed for clinical domains, flares and disease complexity phenotypes (low, medium, high) at the first contact, and corresponding data were extracted using an NLP-based pipeline. Results: Out of 1,000 extraxcted patients, inclusion criteria were met by 262 who had ≥1 hospitalization, ≥1outpatient visit, and a follow-up ≥1.5 years. Among these, 88% were female, median age was 43 years, median follow-up 6 years. At first contact, the most frequently involved clinical domains were hematological (64%), articular (47%), cutaneous (59%) and renal (58%). At first contact, 43% of patients presented with a high-complexity phenotype, 35% medium, 22% low complexity: the first group experienced more flares over time (5 vs 3 and 3, P < 0.001). Patients with a low and medium-complexity phenotype showed a higher increase in new clinical domains and in the use of conventional immunosuppressants, biologics and glucocorticoids during follow-up. Conclusion: This novel framework, based on real-world data, enables longitudinal phenotype characterization of SLE patients. It demonstrates promise as a feasible tool to study the heterogeneity of SLE and its progression over time, offering insights into potential applications in clinical research and patient management.