Rosa Morros

Background: Unstructured free-text narratives in electronic health records (EHRs) contain critical clinical information that is not captured by structured standard codes. In Catalonia, primary care notes follow a semi-structured format known as MEAP (in Catalan). However, extracting structured variables using cloud-hosted commercial large language models (LLMs) raises substantial data privacy concerns and incurs high operational costs. Objective: To develop and validate a privacy-preserving local hybrid pipeline combining a lightweight small language model (SLM) with deterministic regular expressions for extracting six uncoded urinary tract infection (UTI) clinical features (fever, nitrites, leukocytes, lumbar pain, abdominal pain, and haematuria) from primary care EHR narratives. Methods: We deployed the open-weight model (Phi4-mini, 3.8B parameters) within the institutional firewall using Ollama. Prompts were iteratively refined in collaboration with clinicians, and regular expressions were applied post-inference. Validation was conducted across two independent arms: (1) a double-blind clinician gold standard review of 60 real patient records (consensus-resolved cases), and (2) an adversarial synthetic dataset of 720 notes enriched with linguistic noise, generated using GPT-4.1 and Grok-4.1. Point estimates and 95% confidence intervals (CIs) were computed for key performance metrics. Results: The pipeline processed 15,498 MEAP narratives from a matched cohort of 2,962 primary care patients and identified 4,663 clinical feature occurrences. Patients progressing to acute pyelonephritis (cases) presented a higher features burden than non-progressing controls (71.3% vs. 55.8%; standardized mean difference [SMD] = 0.326), particularly for fever (33.0% vs. 9.0%; SMD = 0.616) and lumbar pain (29.0% vs. 9.8%; SMD = 0.502). In real-world clinician validation, overall performance yielded 93.8% accuracy (95% CI 90.6%-96.1%), 83.6% sensitivity (95% CI 73.0%-91.2%), 96.6% specificity (95% CI 93.6%-98.4%), 87.1% positive predictive value (PPV; 95% CI 77.0%-93.9%), and 95.5% negative predictive value (NPV; 95% CI 92.3%-97.7%). In synthetic stress-testing, the pipeline demonstrated 87.2% accuracy (95% CI 84.6%-89.6%), 73.9% sensitivity (95% CI 68.9%-78.5%), near-perfect specificity (99.5%, 95% CI 98.1%-99.9%), and 99.2% PPV (95% CI 97.2%-99.9%). Conclusions: A privacy-preserving local hybrid framework combining a compact open-weight SLM with regular expression rules achieves high specificity and precision for extracting six uncoded clinical features from multilingual primary care narratives. Running entirely within institutional servers, this approach keeps patient data secure, meets privacy standards, and avoids Application Programming Interface (API) costs, making it a practical tool for EHR research, though with lower sensitivity for narratively complex descriptions.

BACKGROUND Unstructured free‑text narratives in electronic health records (EHRs) contain critical clinical information that is not captured by structured standard codes. In Catalonia, primary care notes follow a semi‑structured format known as MEAP (in Catalan). However, extracting structured variables using cloud‑hosted commercial large language models (LLMs) raises substantial data privacy concerns and incurs high operational costs. OBJECTIVE To develop and validate a privacy-preserving local hybrid pipeline combining a lightweight small language model (SLM) with deterministic regular expressions for extracting six uncoded urinary tract infection (UTI) clinical features (fever, nitrites, leukocytes, lumbar pain, abdominal pain, and haematuria) from primary care EHR narratives METHODS We deployed the open‑weight model (Phi4-mini, 3.8B parameters) within the institutional firewall using Ollama. Prompts were iteratively refined in collaboration with clinicians, and regular expressions were applied post‑inference. Validation was conducted across two independent arms: (1) a double‑blind clinician gold standard review of 60 real patient records (consensus‑resolved cases), and (2) an adversarial synthetic dataset of 720 notes enriched with linguistic noise, generated using GPT‑4.1 and Grok‑4.1. Point estimates and 95% confidence intervals (CIs) were computed for key performance metrics. RESULTS The pipeline processed 15,498 MEAP narratives from a matched cohort of 2,962 primary care patients and identified 4,663 clinical feature occurrences. Patients progressing to acute pyelonephritis (cases) presented a higher features burden than non-progressing controls (71.3% vs. 55.8%; standardized mean difference [SMD] = 0.326), particularly for fever (33.0% vs. 9.0%; SMD = 0.616) and lumbar pain (29.0% vs. 9.8%; SMD = 0.502). In real-world clinician validation, overall performance yielded 93.8% accuracy (95% CI 90.6%-96.1%), 83.6% sensitivity (95% CI 73.0%-91.2%), 96.6% specificity (95% CI 93.6%-98.4%), 87.1% positive predictive value (PPV; 95% CI 77.0%-93.9%), and 95.5% negative predictive value (NPV; 95% CI 92.3%-97.7%). In synthetic stress-testing, the pipeline demonstrated 87.2% accuracy (95% CI 84.6%-89.6%), 73.9% sensitivity (95% CI 68.9%-78.5%), near-perfect specificity (99.5%, 95% CI 98.1%-99.9%), and 99.2% PPV (95% CI 97.2%-99.9%). CONCLUSIONS A privacy-preserving local hybrid framework combining a compact open-weight SLM with regular expression rules achieves high specificity and precision for extracting six uncoded clinical features from multilingual primary care narratives. Running entirely within institutional servers, this approach keeps patient data secure, meets privacy standards, and avoids Application Programming Interface (API) costs, making it a practical tool for EHR research, though with lower sensitivity for narratively complex descriptions.

Long COVID-19 is a post-infectious syndrome with persistent symptoms that can involve multiple organ systems. Evidence suggests that SARS-CoV-2 infection may disrupt gut microbiome composition, potentially contributing to long-term effects. As treatment remains symptom-based, interest has grown in repurposing drugs like montelukast. However, non-antibiotic medications may also alter gut microbial communities, raising questions about their impact. Here, we compare gut microbiota between long COVID patients and healthy controls and examine how montelukast treatment affects microbial composition. We analysed stool samples from long COVID patients and healthy controls using 16S rRNA gene sequencing (Illumina MiSeq). We evaluate alpha (Shannon) and beta (Bray-Curtis) diversity, followed by relative abundance and linear discriminant effect size analysis, to identify differentially abundant taxa. This proof-of-concept study included a cross-sectional comparison and a longitudinal analysis of montelukast-treated patients vs. placebo. Cross-sectional analysis revealed a significant structural reorganisation of the gut microbial community in long COVID patients, although overall species richness was largely maintained. Linear discriminant effect size analysis revealed that this architectural shift was driven by an enrichment of Firmicutes (Agathobacter and Faecalibacterium genera) in the long COVID group, while healthy controls were characterised by higher abundances of the phyla Verrucomicrobiota and Actinobacteriota, as well as genera Alistipes and Akkermansia. Longitudinal analysis demonstrated that the broader community structure remained stable in both groups; however, montelukast treatment led to a specific enrichment of the genus Dialister, suggesting targeted and potentially transient effects without disrupting the overall microbial landscape. Long COVID is characterised by a significant restructure of the gut ecosystem. This qualitative dysbiosis reflects a shift in homeostatic balance, where the core microbial community remains present, but its proportions are altered. Short-term montelukast treatment shows a minimal impact on the microbial landscape, suggesting treatment does not further destabilise the gut environment. These findings highlight the specific and targeted nature of gastrointestinal involvement in long COVID.

Background/Objectives: The study’s goal is to assess the association between long-term statin therapy and influenza incidence, influenza severity, and all-cause mortality. Methods: Two population-based dynamic cohorts (exposed and unexposed to statins) were followed from 2010 to 2019. Participants were 60 years or older; frail patients were excluded. The primary outcomes were influenza incidence, influenza-related intensive care unit (ICU) admission as a proxy for severity, and all-cause mortality. The exposed cohort comprised new statin users with a minimum of two pharmacy invoices within 90 days of enrollment. Adjusted risk ratios (aRRs) for influenza incidence, ICU admission, and mortality rate were calculated using Poisson regression. Results: The initial study population of 639,564 individuals was evenly split into exposed (319,782) and unexposed (319,782) cohorts; mean age was 71 years (standard deviation: 8 years), and 57% were women. Compared to non-users, new statin users showed a higher influenza incidence [9.39 (95% confidence interval: 9.36–9.42) vs. 7.64 (7.61–7.66) per 1000 person-years], ICU admission [1.65 (1.65–1.66) vs. 1.36 (1.35–1.36) per 1000 person-years], and overall mortality rate [97.09 (96.75–97.43) vs. 94.15 (93.82–94.47) per 1000 person-years]. Adjusted analysis revealed no significant association between statin use and influenza incidence [aRR: 1.04 (0.98–1.10)] or influenza-related ICU admission [aRR: 1.03 (0.89–1.19)] and shifted the effect on mortality from harmful to beneficial [aRR: 0.88 (0.87–0.89)]. Conclusions: Despite new users’ greater vulnerability at the start of treatment, our findings indicate that statins do not influence influenza incidence or severity but reduce all-cause mortality, warranting further exploration of their anti-inflammatory properties.

Background: Combating the rise of drug-resistant organisms and minimizing side effects requires a shift in how we approach the duration of antibiotic therapy. A promising strategy involves tailoring the length of antibiotic therapy to patients’ needs, allowing discontinuation once patients feel better. Objective: This study aims to assess whether shortening antibiotic therapy based on patients’ recovery time is as effective as completing the full course in treating acute respiratory tract infections (RTIs). Methods: We plan to enroll a minimum of 474 outpatients ranging from 18 to 75 years of age with clinical features of acute RTIs across Spanish health centers. Patients diagnosed with acute lower RTIs or acute rhinosinusitis, deemed by clinicians to require a β-lactam course, will be randomized to either usual care, involving a full-course antibiotic therapy based on current guidelines, or a tailored approach. In the intervention group, patients will be advised to visit the center as soon as they feel better and are afebrile for a clinical assessment and C-reactive protein rapid testing. Treatment will be discontinued if these clinical results are normal. The primary outcome will be assessment of clinical resolution at day 14, while secondary outcomes include antibiotics given and duration; adherence to the prescribed antibiotic; other therapies given; early clinical assessment at day 7; days of severe, moderate, and any symptom, complications and reattendance within the first month; drug-related adverse events; absenteeism; and health-related quality of life. All participants will be given a symptom diary, recording their symptoms each evening. Additionally, a cost-effectiveness study and qualitative studies involving clinicians and patients aimed at exploring the strategy’s pros, cons, uptake, and satisfaction levels will be carried out. Results: The final results will be published by the end of 2027. Conclusions: We will examine whether adults who present with symptoms of RTI who are treated with antibiotic courses until they feel better are as effective as longer standard courses. It is highly important that a possible reduction in the antibiotic course as soon as the patient feels better does not compromise patients’ recovery. This comprehensive approach aims to shed light on the practicality and impact of tailoring antibiotic duration in RTIs. Trial Registration: ClinicalTrials.gov NCT06581367; https://clinicaltrials.gov/study/NCT06581367 International Registered Report Identifier (IRRID): PRR1-10.2196/75453

Migraine is a neurological disorder affecting approximately 12% of the population, more frequent in women, causing disability. Preventive treatment is recommended to prevent chronification and analgesics’ abuse and to improve quality of life, but not all candidates receive it. Common preventive drugs include amitriptyline, flunarizine, propranolol, and topiramate. Their effectiveness and safety have not yet been assessed in our setting. The objective of this study is to evaluate the effectiveness of these drugs in reducing the monthly migraine days (MMD) and to evaluate their safety. Phase IV, pragmatic randomized clinical trial of four parallel groups, open and multicentric, in Primary Care in Catalonia, Spain. We will include adults with migraine eligible for preventive treatment. They will be randomized to the following: (1) propranolol, (2) amitriptyline, (3) flunarizine, or (4) topiramate. The main efficacy measure will be the change in the mean MMD at week 12 in comparison to the mean MMD at baseline. Other secondary efficacy and safety measures will be assessed. The main non-inferiority analysis will be conducted in the per protocol population at week 12, comparing topiramate versus propranolol as the main analysis, and amitriptyline and flunarizine versus propranolol as exploratory analyses. There are epidemiological, clinical and pharmacological sex-differences in migraine which may result in differences in the use of migraine preventive drugs. However, the effectiveness of the most frequent preventive treatment and the differences in effectiveness and safety between women and men have not yet been studied. Moreover, as new expensive drugs have been authorized for preventive treatment, indicated when the usual drugs have failed, assessing the effectiveness and safety of the usual oral preventive drugs through a clinical trial is especially important. We also plan to evaluate their cost-effectiveness, and the clinical trial will count on an active contribution from patients through an app designed for this purpose, which may help in the management of the disease in the future. ClinicalTrials.gov ID: NCT06499116 ( https://clinicaltrials.gov/study/NCT06499116 ). Trial registration date: July 12, 2024. EU CT Number: 2024–513597-22. Protocol ID: IJG-PREMI-2024. Sponsor: Fundació d’Investigació en Atenció Primària Jordi Gol i Gurina. Gran Via de les Corts Catalanes 587, àtic. 08007, Barcelona, Spain.

Background: This paper focuses on women’s subjective experiences with UTIs and antibiotic use. Uncomplicated urinary tract infections (UTI) are among the most common infections treated in primary care, however, there is limited evidence regarding the experiences and their antibiotic treatment preferences. Objective: The aim of this study is to gather experiences, beliefs, and attitudes regarding the use of antibiotics and the participation in the clinical trial. Methods: This qualitative sub-study is part of the SCOUT project, which, through a clinical trial, aims to assess the effectiveness of four short antibiotic regimens (a single 3g dose of fosfomycin; 3g of fosfomycin for two days; 400mg of pivmecillinam for three days; and 100mg of nitrofurantoin for five days). A total of 68 semi-structured interviews were conducted in four regions of Spain (Catalonia N=19, Aragón N=19, Madrid N=15, and the Balearic Islands N=15). A purposive and theoretical sampling approach was employed. Participants were recruited after taking part in the clinical trial. Data collection took place between March and July 2024 and analysed using Thematic Analysis. Results: UTI did not cause excessive concern, except in cases where the antibiotic was ineffective or when participants had experienced multiple infections in the past. Fosfomycin was the most widely recognized antibiotic, and the two-dose regimen was generally well-regarded for being easy to take while also being more effective than the single-dose treatment. Longer regimens could cause some side effects, but in certain cases, they were perceived as more effective. Participation in the clinical trial was viewed positively, as it contributed to increasing overall knowledge about UTI and the antibiotics used to treat them, as well as ensuring closer monitoring of each specific case. Conclusion: This study gathers experiences related to UTI, antibiotic treatment regimens, and participation in the SCOUT clinical trial. It highlights the need to document the realities of UTI in order to improve future clinical trials.

Background: Interventions based on testing and communication training have been developed to reduce antibiotic prescribing in primary healthcare (PHC) for the treatment of acute lower respiratory infections (ALRTIs). However, research based on the experiences of PHC clinicians participating in ALTRIs interventions to reduce antibiotic prescribing in Barcelona is scanty. Aim: This study aimed to explore the perceptions and experiences of clinicians (physicians and nurses) on an intervention to reduce antibiotic prescription in PHC in Barcelona (Spain). This intervention was a randomised controlled study (cRCT) based on three arms: 1) use of a C-reactive protein (CRP) rapid test; 2) enhanced communication skills; and 3) combination of CRP rapid test and enhanced communication skills. In addition, the study aimed to explore the impact of COVID-19 on the detection of ALRTIs. Methods: This qualitative study used a socio-constructivist perspective. Sampling was purposive. Participants were selected based on age, sex, profession, intervention trial arm in which they participated, and the socioeconomic area of the PHC where they worked. They were recruited through the healthcare centres participating in the study. Nine participants (7 women and 2 men) participated in two focus groups, lasting 65–66 min, in September–October 2022. Framework analysis was used to analyse the data. Findings: Three themes were identified: ‘(The intervention) gave us reassurance’: intervention experiences among health professionals. This theme includes accounts of clinicians’ satisfaction with the intervention, particularly with CRP testing to support clinical diagnoses; ‘We don’t have time in primary healthcare’: structural and community resources in healthcare services. This theme encompasses clinicians’ experiences on healthcare pressures and PHC organisational structures barriers to PHC interventions; and ‘I only did three CRP’: impact of COVID-19 pandemic on the intervention. The last theme focuses on the impact of the COVID-19 pandemic on the intervention’s implementation. Conclusions: CPR testing and promoting communication skills can be useful tools to support clinical decisions for ALRTIs. Structural barriers (e.g., healthcare pressures) and social inequities amongst service users were acknowledged as the main barriers for the implementation of ALRTIs interventions.

Objectives: This study aims to evaluate the impact of chronic statin therapy on the incidence of community-acquired pneumonia (CAP) and the rate of intensive care unit (ICU) admissions associated with CAP. Methods: Two population-based dynamic cohorts, consisting of individuals exposed and unexposed to statins, were followed from 2010 to 2019. Participants were older than 60 years, with frail patients excluded. The primary outcomes were the incidence of CAP and ICU admissions due to CAP, serving as a proxy for complicated cases. The exposed cohort included new statin users with at least two pharmacy invoices within 90 days of the recruitment period. Adjusted risk ratios (aRRs) for CAP incidence and CAP-associated ICU admissions were calculated using Poisson regression. Results: This study analyzed a sample of 639,564 individuals, evenly divided into exposed (319,782) and unexposed (319,782) groups, with a mean age of 71 years (standard deviation of 8 years) and 57% women. New statin users had a higher incidence of CAP [42.1 (95% confidence interval: 41.9–42.2) vs. 36.6 (36.5–36.8) per 1000 person-years] and ICU admissions [11.5 (11.5–11.6) vs. 10.1 (10.0–10.1) per 1000 person-years] compared to non-users. The adjusted analysis indicated that statin treatment reduced CAP risk by 6% [aRR: 0.94 (0.91–0.96)] and ICU admission by 7% [aRR: 0.93 (0.88–0.98)]. Conclusions: Prior statin therapy was associated with a clinically significant reduction in the incidence of CAP and ICU admissions due to CAP, despite the greater vulnerability of new users at the start of treatment compared to non-users.