EClinicalMedicine

Background: How post-COVID-19 condition (PCC) differs from post-acute infection syndromes (PAIS) caused by other respiratory viruses remains uncertain. Comparing these conditions may clarify whether post-acute symptoms reflect specific consequences of SARS-CoV-2 infection or broader post-viral mechanisms. Methods: We conducted a systematic review and meta-analysis of cohort studies comparing persistent symptoms or conditions in adults after SARS-CoV-2 infection with those following other acute respiratory viral infections. PubMed, Embase, and Scopus were searched. Random-effects models were used to estimate pooled risks. Results: Among 9,371 records screened, 22 studies were included and 14 contributed to the meta-analysis. Increased risk after SARS-CoV-2 infection was observed for pulmonary embolism, abnormal breathing, fatigue, hemorrhagic stroke, memory loss/brain fog, and palpitations; heart rate abnormalities showed borderline significance. For most other outcomes pooled estimates were inconclusive. Conclusions: Only a subset of outcomes appears more frequent after SARS-CoV-2 infection, suggesting many symptoms attributed to PCC may reflect broader post-viral syndromes.

Background: Long COVID presents with one or multiple symptoms or diagnosable conditions after SARS-CoV-2 infection. To study whether use of the antiviral remdesivir in persons hospitalized with acute COVID-19 is associated with reduced Long COVID, we created a computational phenotype for Long COVID. Methods: In electronic health records (EHR) from a multistate healthcare system (US), hospital admissions from 5/1/20 - 9/30/22 were reviewed. The study group was hospitalized with acute COVID-19 and the control group was hospitalized for other reasons without prior SARS-CoV-2 infection. The populations were balanced with overlap weights based on a high-dimensional propensity score of pre-specified variables and the top 100 comorbidities differing between the groups. Hazard ratios (HR) were calculated for the combined primary outcome: U09.9 (Post-Covid Conditions) or any incident secondary outcome from 90 to 365 days after admission. Secondary outcomes included 27 individual incident diagnoses, corrected for multiplicity with Holm-Bonferroni. Results: Admissions included 45,540 with, and 409,186 without COVID-19 during the study period, evaluable for the primary outcome. After weighting, standardized difference was < 0.01 for all measured confounders including demographic and clinical features. In the COVID+ and non-COVID groups 38.0% and 29.3% met the combined primary outcome, respectively. Weighted HR (95%CI) for the primary outcome was 1.37 (1.35, 1.40), p < 0.0001. All secondary outcomes were associated with the COVID+ group, when adjusted for multiplicity. Incident diagnoses with strong associations (HR > 2) included thromboembolism, hair loss, diabetes mellitus, obesity, and hypoxia. Anosmia/dysgeusia was associated with COVID, but wide confidence intervals reflected few charted diagnoses. Conclusions: Manifestations of Long COVID at population scale are detectable as part of routine symptoms and clinical diagnoses in the EHR after admissions for COVID-19, compared with all other hospital admissions. This a prior computational phenotype for Long COVID will be used to assess whether remdesivir use is associated with decreased Long COVID.

BackgroundTo date, accessible diagnostic tools to identify whether a patients pneumonia is a bacterial, or viral infection, are not accurate or timely enough to prevent preemptive antibiotic administration. Relying on single biomarkers or clinical presentations has been insufficient. We aimed to incorporate a wide range of novel biomarkers and clinical presentations in a multivariable model and validate its capacity to differentiate cases of bacterial and viral pneumonia. MethodsData from 457 children aged 2-59 months, admitted to Kilifi County Referral Hospital, Kenya, with bacterial (n = 229) and viral (n = 228) infections, were used to develop and validate a predictive multivariable Poisson regression model to differentiate pneumonia etiology. The Receiver Operating Characteristic curve was used to assess biomarker performance and validate the model internally. ResultsSixty-three percent (63%) of the children presented with severe pneumonia. 72% with viral pneumonia had severe pneumonia, compared to 54% with bacterial pneumonia who had severe pneumonia. In crude analyses, chest-wall indrawing, cough, convulsions, crackles, angiotensinogen, and Serpin Family A Member 1 were significantly associated with pneumonia etiology, controlling for age. However, only chest-wall indrawing remained significant in multivariable analyses after controlling for age. The model demonstrated fair, but inadequate, discrimination, with an Area Under the Curve of 0.61. ConclusionAmong the children admitted to hospital with WHO defined pneumonia, a wide range of biomarkers and clinical presentations still failed to distinguish bacterial from viral pneumonia.

ObjectiveBiologically defined sepsis subtypes have been identified in low- and middle-income countries (LMICs), but limited access to molecular diagnostics challenges broader evaluation and implementation in resource-limited settings. We assessed whether models including bedside clinical and rapid microbiologic data could accurately stratify Ugandan adults with sepsis by molecular subtype. DesignSecondary analysis of two prospective observational sepsis cohorts, testing bedside-adaptable classifier models against transcriptomic and proteomic subtype assignments. SettingEntebbe Regional Referral Hospital (urban) and Tororo General Hospital (rural), Uganda. PatientsAdults ([&ge;]18 years) hospitalized with sepsis, with available transcriptomic (N=355) and/or proteomic (N=495) profiling enabling subtype assignment. InterventionsNone. Measurements and Main ResultsUsing data from two prospective cohorts (RESERVE-U-2-TOR and RESERVE-U-1-EBB), we evaluated bedside-adaptable models against Uganda-derived molecular sepsis subtypes, and, secondarily, against molecular subtypes and axes derived in high-income countries. In RESERVE-U-2-TOR, clinical models including demographics and bedside physiological variables demonstrated moderate discrimination for transcriptomic and proteomic subtype assignment (AUROC 0.75 [95% CI, 0.69-0.81] and 0.73 [0.66-0.80], respectively) with strong calibration (Integrated Calibration Index [Eavg] [&le;]0.015 for both models). Adding rapid diagnostic results for HIV, malaria, and tuberculosis produced similar performance (AUROC 0.76 and 0.74; Eavg [&le;]0.016). In RESERVE-U-1-EBB, discrimination for clinical and clinico-microbiological models was more variable (AUROC range 0.63-0.75) while calibration remained acceptable (Eavg [&le;]0.053). Performance was similar when models were evaluated against molecular sepsis frameworks derived in high-income countries, with acceptable calibration and moderate discrimination. ConclusionsBedside-adaptable clinical models, with or without rapid microbiologic testing, demonstrated acceptable calibration but only modest discrimination for molecular sepsis subtype assignment in Uganda. Expanding laboratory capacity and access to scalable, low-cost molecular biomarker assays will be necessary to advance precision sepsis care in LMIC settings. Key PointsO_ST_ABSQuestionC_ST_ABSAmong adults hospitalized with sepsis in a resource-limited setting, can bedside clinical variables, alone or combined with rapid pathogen diagnostics, accurately stratify molecular sepsis subtype assignments? FindingsIn two prospective Ugandan sepsis cohorts, bedside clinical and clinico-microbiologic models showed robust calibration but only modest discrimination for classifying Uganda-derived transcriptomic and proteomic subtypes. Models also achieved moderate performance for stratifying high-income-country-derived transcriptomic subtypes and immune dysfunction axes, suggesting bedside variables reflect illness severity but incompletely capture underlying molecular signatures. MeaningBedside-adaptable models can support reasonably calibrated risk estimation for molecular sepsis stratification in resource-limited settings but lack sufficient discriminatory power to serve as stand-alone tools. These findings support efforts to improve acute-care laboratory capacity and access to scalable molecular biomarker panels, with the goal of enabling precision sepsis care in low- and middle-income countries.

Background: The 2024-25 influenza season was the most severe in the United States (US) since 2017-18, with co-circulation of both influenza A virus subtypes (H1N1 and H3N2). Influenza vaccine effectiveness (VE) has varied by season, setting, and patient characteristics. Methods: Using electronic healthcare encounter data from eight US states, we evaluated influenza vaccine effectiveness (VE) against influenza-associated hospitalizations and emergency department or urgent care (ED/UC) encounters from October 2024-April 2025 among children aged 6 months-17 years and adults aged 18+ years. Using a test-negative, case-control design, we compared the odds of influenza vaccination between acute respiratory illness (ARI) encounters with a positive (cases) versus negative (controls) test for influenza by molecular assay, adjusting for confounders. Results: Analyses included 108,618 encounters (5,764 hospitalizations and 102,854 ED/UC encounters) among children and 309,483 encounters (76,072 hospitalizations and 233,411 ED/UC encounters) among adults. Among children across care settings, 17.0% (6,097/35,765) of cases versus 29.4% (21,449/72,853) of controls were vaccinated. Among adults, 28.2% (21,832/77,477) of cases versus 44.2% (102,560/232,006) of controls were vaccinated. VE was 51% (95% confidence interval [95% CI]: 41-60%) against influenza-associated hospitalizations and 54% (95% CI: 52-55%) against influenza-associated ED/UC encounters among children. VE was 43% (95% CI: 41-46%) against influenza-associated hospitalizations and 49% (95% CI: 47-50%) against influenza-associated ED/UC encounters among adults. Conclusions: Influenza vaccination provided protection against influenza-associated hospitalizations and ED/UC encounters among children and adults in the US during the severe 2024-25 influenza season. These findings support influenza vaccination as an important tool to reduce influenza-associated disease.

Background Outcome after stroke varies according to stroke subtype by location, but healthcare systems data studies do not include subtyping information. We linked natural language processing (NLP) of brain imaging reports to routinely collected data to estimate risk of death and other outcomes after stroke subtypes in a nationwide dataset. Methods We applied a previously validated NLP algorithm to all CT and MRI head scan reports in Scotland between 2010 and 2018. We linked the reports to hospital readmissions, prescriptions and death data to identify and characterize people with stroke, and to categorize into deep and cortical ischemic stroke, deep and lobar intracerebral hemorrhage (ICH), subarachnoid hemorrhage, and subdural hemorrhage. We used a matched cohort design, and age- and sex-matched four controls per case who never had a stroke. By subtype, we estimated rehospitalization with stroke, myocardial infarction (MI), cancer, dementia, epilepsy and death, accounting for confounders and competing risk of death. Results From 785,331 people with a head scan, we identified 64,219 with clinical stroke phenotypes (mean age 73.4yrs, 49.5% male), and subtyped 12,616 with deep ischaemic stroke; 14,103 with cortical ischaemic stroke; 1,814 with deep ICH; and 1,456 with lobar ICH. There was higher absolute rate of 1-year hospital readmission for lobar compared with deep ICH (4.9% [95%CI 3.9% - 6.1%] vs 3.4% [2.6% - 4.3%]), higher risk of dementia beyond 6 months after lobar ICH compared to controls than for other stroke subtypes (aHR 3.5 [2.3-5.3]); and higher risk of MI within 6 months of cortical ischemic stroke than for other stroke subtypes (aHR 4.6 [3.4-6.3]). Conclusions NLP of free-text reports linked to coded data successfully subtyped stroke at scale, and we estimated risk of clinically relevant outcomes. Future work should use free text to enable large-scale audit and epidemiology of people with stroke.

Objectives: Among individuals attending stroke rehabilitation, we aimed to determine the proportion who participated in cardiorespiratory exercise, identify patient characteristics predicting participation, and describe exercise characteristics. Design, setting, and participants: This was an observational cohort study involving all patients admitted to four stroke rehabilitation centres in Ontario, Canada, during March or October 2019, or over 12 months starting in 2021. Main measures: Patient characteristics extracted during chart review included age, sex, marital status, employment status, date of stroke, time post-stroke at admission, length of stay for rehabilitation, past medical history that could affect exercise participation, Functional Independence Measure, Functional Ambulation Category, mobility aid use, Chedoke-McMaster Stroke Assessment, Montreal Cognitive Assessment, National Institutes of Health Stroke Scale, and details describing cardiorespiratory exercise completed. Results: 40.1% of stroke patients participated in cardiorespiratory exercise, with 26.4% having it included in their treatment plan. Diagnosed cardiac disease (OR=0.74), poor left ventricular function (OR=0.09), history of mental health conditions (OR=0.69), lower functional ambulation ability (OR=0.74), and wheelchair use at rehabilitation admission (OR=0.46) were associated with lower odds of participating in cardiorespiratory exercise after stroke (p-values<0.05). Use of a walker or rollator at rehabilitation admission (OR=3.22), having a cardiorespiratory exercise goal (OR=2.13), and longer lengths of stay (OR=1.01) were associated with higher odds of participating in cardiorespiratory exercise after stroke (p-values<0.05). Only 1.5% of patients (N=9/601) who participated in cardiorespiratory exercise completed it with recommended intensity and duration. Conclusion: Improving participation in cardiorespiratory exercise during stroke rehabilitation may require addressing cardiovascular, mental health, and mobility-related barriers.

Background: Current medical large language model (LLM) evaluations largely rely on small collections of cases, whereas rigorous safety testing requires large-scale, diverse, and complex cases with verifiable ground truth. Multiple Sclerosis (MS) provides an ideal evaluation model, with validated diagnostic criteria and numerous paraclinical tests informing differential diagnosis, investigation, and management. Methods: We generated synthetic MS cases with ground-truth labels for diagnosis, localisation, and management. Four frontier LLMs (Gemini 3 Pro/Flash, GPT 5.2/5 mini) were instructed to analyse cases to provide anatomical localisation, differential diagnoses, investigations, and management plans. An automated evaluator compared these outputs to the ground-truth labels. Blinded subspecialty experts validated 70 cases for realism and automated evaluator accuracy. We then evaluated LLM decision-making across 1,000 cases and scaled to 10,000 to characterise rare, catastrophic failures. Results: Subspecialist expert review confirmed 100% synthetic case realism and 99.8% (95% CI 95.5 to 100) automated evaluation accuracy. Across 1,000 generated MS cases, all LLMs successfully included MS in the differential diagnoses for more than 91% cases. However, diagnostic competence did not associate with treatment safety. Gemini 3 models had low rates of clinically appropriate steroid recommendations (Flash: 7.2% 95% CI 5.6 to 8.8; Pro: 15.8% 95% CI 13.6 to 18.1) compared to GPT 5 mini (23.5% 95% CI 20.8 to 26.1), frequently overlooking contraindications like active infection. OpenAI models inappropriately recommended acute intravenous thrombolysis for MS cases (9.6% GPT 5.2; 6.4% GPT 5 mini) compared to below 1% for Gemini models. Expanded evaluation (to 10,000 cases) probed these errors in detail. Thrombolysis was recommended in 10.1% of cases lacking symptom timing information and paradoxically persisted (2.9%) even when symptoms were explicitly documented as more than 14 days old. Conclusion: Automated expert-level evaluation across 10,000 cases characterised artificial intelligence clinical blind spots hitherto invisible to small-scale testing. Massive-scale simulation and automated interrogation should become standard for uncovering serious failures and implementing safety guardrails before clinical deployment exposes patients to risk.

BackgroundIn England, individuals with chronic liver disease (CLD) are among those with the lowest seasonal influenza vaccine uptake despite being at elevated risk of severe influenza. We examined the relationship between CLD severity and aetiology, and influenza vaccine uptake in England. MethodsA retrospective cohort study of adults (18-115 years) using Clinical Practice Research Datalink Aurum primary care data was conducted for five seasons (2019/20-2023/24). Poisson regression was used to estimate rates of uptake by CLD severity (clinical diagnoses categorised as low, moderate, or severe) and aetiology (alcohol-related, viral-related, and diagnoses in the Green Book guidelines). FindingsThere were 182,174-277,470 with CLD per cohort. Among those who were additionally age-eligible for vaccination, uptake was 71{middle dot}1-79{middle dot}7% compared to 30{middle dot}9-40{middle dot}5% in those not additionally age-eligible. Among individuals below age eligibility without other comorbidities, severity was associated with higher uptake (incidence rate ratio [IRR] moderate 1{middle dot}80, 95% CI 1{middle dot}69-1{middle dot}90; severe 1{middle dot}95, 95% CI 1{middle dot}84-2{middle dot}08 in 2023/24); there was no effect in those with at least one additional comorbidity (moderate 1{middle dot}05, 95% CI 0{middle dot}99-1{middle dot}10; severe 1{middle dot}05, 95% CI 1{middle dot}01-1{middle dot}09). Alcohol- and viral-related aetiology were also associated with increased uptake in those not additionally age-eligible. Among individuals meeting age eligibility without additional comorbidities, severity was associated with a reduced uptake (moderate 0{middle dot}81, 95% CI 0{middle dot}73-0{middle dot}90; severe 0{middle dot}79, 95% CI 0{middle dot}74-0{middle dot}85), with attenuation in those with additional comorbidities (moderate 0{middle dot}99, 95% CI 0{middle dot}94-1{middle dot}04; severe 0{middle dot}91, 95% CI 0{middle dot}89-0{middle dot}94). InterpretationCLD severity and aetiology were important determinants of uptake in the absence of additional indications for influenza vaccination. Future research should prioritise understanding facilitators and barriers to vaccine uptake in individuals with CLD, particularly for those at highest risk of severe infection. FundingNIHR Health Protection Research Unit in Vaccines and Immunisation (NIHR200929/NIHR207408). Research in contextO_ST_ABSEvidence before this studyC_ST_ABSWe searched PubMed up to June 2025 using the terms "chronic liver disease", "cirrhosis", "hepatitis", "influenza vaccination", "seasonal influenza", and "vaccine uptake". Previous research, including national data from England, has shown that people with chronic liver disease tend to have lower seasonal influenza vaccine uptake than individuals with other medical comorbidities which qualify for vaccination such as diabetes, chronic kidney disease or immunosuppression. The reasons for low influenza vaccine uptake in people with chronic liver disease are not well understood, and it is therefore difficult for vaccination providers, principally primary care services in England, to tailor interventions aimed to increase uptake. Qualitative research involving individuals aged less than 65 years living in England with clinical risk comorbidities, most commonly diabetes, found that chronic disease management pathways inconsistently provided information about the importance of influenza vaccination as part of chronic disease management. Individuals with long-term conditions reported low perceived risk of influenza infection and limited awareness of vaccine benefits as important reasons for non-uptake. We hypothesised that the severity and aetiology of chronic liver disease may be important determinants of uptake. Added value of this studyWe conducted a population-based study to examine how chronic liver disease severity and aetiology influence seasonal influenza vaccine uptake in adults in England. Using primary care electronic health record data from five consecutive influenza seasons (2019/20-2023/24), we found that more severe chronic liver disease was associated with a substantial increase in vaccine uptake in those without additional indications for seasonal influenza vaccination (age-based eligibility or other qualifying clinical risk comorbidities). Alcohol- and viral-related aetiology were also associated with increased uptake in those who were not additionally age-eligible for vaccination. In contrast, severity, alcohol- and viral-related underlying aetiology were associated with a modest reduction in uptake for individuals with chronic liver disease who also qualified for vaccination due to age. Implications of all the available evidenceDespite clear clinical vulnerability to infection and a substantially elevated risk of morbidity and mortality following infection, a large proportion of adults with chronic liver disease, particularly those aged under 65 years, remain unvaccinated against seasonal influenza each year. This study suggests that chronic liver disease severity and underlying aetiology are important determinants of uptake in individuals not meeting age-based vaccine eligibility, particularly in those without additional clinical risk comorbidities. This could be because of differing perceptions of influenza risk, or due to varying degrees of interaction with healthcare specialists as part of chronic disease management. In individuals who met age-based vaccination eligibility, the negative effect of severity on influenza vaccine uptake may reflect greater barriers to accessing vaccination services by those with more complex health needs, or competing medical priorities for long-term condition management during consultations. To inform targeted vaccination strategies, future research should aim to understand the specific facilitators and barriers to influenza vaccination experienced by individuals with chronic liver disease. This should include perspectives of individuals with different disease severity, across different age groups, in those with and without additional co-morbidities.

Objectives: In Latin America, up-to-date information to monitor UNAIDS 95-95-95 HIV targets in key populations, such as men who have sex with men, is limited. Elsewhere, structural homophobia restricts access to ART. Conceptual frameworks suggest that intersecting forms of violence and discrimination may negatively influence HIV care outcomes through psychosocial and structural pathways, although empirical evidence remains limited. The study aimed to assess whether sexual orientation outness and recent homophobic violence are associated with not being on ART among Latin American MSM living with HIV. Methods: This cross-sectional study is a secondary analysis of data from LAMIS-2018, including 7,609 MSM aged 18+ with an HIV diagnosis [&ge;]1 year prior from 18 Latin American countries. Participants self-reported ART status, sociodemographic characteristics, homophobic violence, and sexual orientation outness. Bivariate and multivariate logistic regressions identified those factors associated with not being on ART. Results: Nine percent of MSM with HIV were not on ART, 18% reported low sexual orientation outness, and 27% experienced homophobic violence, especially in Andean and Central American countries. Not being on ART was associated with recent homophobic violence (aPR=1.25), low outness (aPR=1.22), unemployment (aPR=1.27), and residence in the Andean subregion (aPR=1.87), Mexico (aPR=1.28), or the Southern Cone (aPR=1.45) versus Brazil. Protective factors included being older (25-39: aPR=0.72; >39: aPR=0.49), living in large cities (aPR=0.72), having a stable partner (aPR=0.78), and university education (aPR=0.74). Conclusions: Recent homophobic violence and low sexual orientation outness were associated with not being on ART among MSM in Latin America. While access varies across countries, structural factors such as stigma and violence may limit engagement in care. Addressing these barriers alongside strengthening health systems may be key to improving ART uptake and advancing progress toward the 95-95-95 targets.

BackgroundMulticancer early detection tests could be used for cancer screening, but may lead to harms, including false positive results and overdiagnosis of indolent tumours that would not have become clinically evident during that persons lifetime. We assessed the potential for these screening harms in the context of future population-based screening with a multicancer early detection test. MethodsWe used a microsimulation model to assess potential population-level impacts of screening at ages 50-75 years with a multicancer early detection test in Canada. We assumed high test specificity (97-99.1%) and test sensitivity increasing with cancer stage. The model includes latent indolent cancers that would not be diagnosed within that persons lifetime but can be overdiagnosed through screen-detection. We calculated the yearly and cumulative lifetime probabilities of screening overdiagnosis and false positive test results, assuming a range of preclinical screen-detectable periods (2-5 years). ResultsAn estimated 2.1-6.0% of all yearly screen-detected cancers with a multicancer screening test were predicted to be overdiagnoses across scenarios. The proportion of overdiagnosis varied by site, and strongly increased with age, going from 1% at age 50 to over 10% of screen-detected cancers by age 75. The test positive predictive value ranged from 15.9%-77.6%, meaning that there could be 0.3-5.3 false positives with no underlying cancer for every true cancer case detected by the test. ConclusionPopulation-level multicancer screening with a multicancer early detection test would likely not lead to substantial screen-related overdiagnosis. Healthcare systems should consider how screening false positives may increase their diagnostic service caseload.

Background Chronic subdural hematoma (cSDH) recurrence requiring reoperation occurs in 5-33% of cases, representing a substantial clinical and economic burden. The ability to predict recurrence could enable risk-stratified surveillance protocols, potentially reducing imaging burden in low-risk patients while maintaining close monitoring for high-risk individuals. We evaluated whether machine learning algorithms could achieve clinically actionable recurrence prediction using routinely available clinical and radiographic variables. Methods This retrospective single-center study included 564 consecutive patients who underwent surgical evacuation of cSDH between 2015 and 2023. Data were randomly divided into training (75%, n=422) and test (25%, n=142) sets. We developed and compared three machine learning models--regularized logistic regression, Random Forest, and XGBoost--using 31 predictor variables including demographics, comorbidities, medications, laboratory values, hematoma characteristics, and postoperative features. Model development and hyperparameter tuning were performed exclusively on the training set using 10-fold cross-validation. The best-performing model was selected and evaluated on the held-out test set. The primary outcome was postoperative recurrence requiring reoperation. Results Postoperative recurrence occurred in 170 patients (30.1%). Within the training set, XGBoost achieved the highest cross-validated ROC AUC of 0.713 (SE=0.024), outperforming regularized logistic regression (0.686) and matching Random Forest (0.713). Variable importance analysis identified hematoma volume, coagulation parameters (INR, platelets, aPTT), and disease severity markers (ICU admission, GCS) as the most influential predictors, though absolute effect sizes remained modest. On the held-out test set, the final XGBoost model achieved ROC AUC 0.688 (95% CI: 0.590-0.772) with excellent calibration. However, at the clinically relevant 90% sensitivity threshold, test set specificity was only 30.3%, allowing potential imaging reduction in approximately one-third of non-recurrence patients. The consistency between training and test performance confirmed that limitations stem from inherent predictor information content rather than overfitting. Conclusions Machine learning models using routinely available clinical and radiographic variables cannot achieve clinically actionable risk stratification for cSDH recurrence. Despite rigorous methodology and internal validation, discriminative capacity remained insufficient to identify a low-risk patient subgroup suitable for de-escalated surveillance. These findings suggest that recurrence is driven by factors not captured in standard clinical assessment, and support either uniform surveillance protocols or symptom-driven imaging strategies rather than risk-stratified approaches.

ObjectiveStudies of nutritional status and host responses during severe and critical illness have focused predominantly on obesity; in contrast, the relationship between undernutrition, host responses, and clinical outcomes in adults hospitalized with severe infection remains poorly defined. We sought to determine whether severe undernutrition is associated with distinct host responses and clinical outcomes in adults hospitalized with severe infection. DesignProspective cohort study. SettingTwo public referral hospitals in Uganda. PatientsNon-pregnant adults ([&ge;]18 yr) hospitalized with severe, undifferentiated infection. InterventionsNone. Measurements and Main ResultsWe analyzed clinical data and serum Olink proteomic data from 432 participants (median age, 45 yr [IQR, 31-57 yr]; 44% male). Overall, 213 participants (49%) met prespecified criteria for undernutrition, including 52 (12%) with severe undernutrition. Clinically, severe undernutrition was associated with HIV coinfection, microbiologically diagnosed tuberculosis, greater physiological instability, and higher mortality. After adjustment for age, sex, illness duration, study site, and HIV, malaria, and tuberculosis coinfection, severe undernutrition was associated with higher expression of proteins involved in pro-inflammatory immune signaling, endothelial and vascular remodeling, hypoxia and oxidative stress responses, and extracellular matrix remodeling, together with lower expression of proteins linked to growth signaling, anticoagulant regulation, and lipid homeostasis. ConclusionsSevere undernutrition is associated with a distinct high-risk clinical phenotype and biologic signature in adults hospitalized with severe infection. These findings suggest that undernutrition may potentiate key domains of sepsis pathobiology, with implications for strengthening nutritional support and informing host-directed treatment strategies in low- and middle-income countries where malnutrition is common. Key PointsO_ST_ABSQuestionC_ST_ABSHow does undernutrition influence immune, metabolic, and endothelial responses to severe infection in adults? FindingsIn this multicenter cohort study of 432 adults hospitalized with severe infection in Uganda, severe undernutrition was associated with greater physiologic instability, higher mortality, and a distinct proteomic host-response profile. Adults with severe undernutrition exhibited a proteomic signature characterized by pro-inflammatory immune signaling, endothelial and extracellular matrix remodeling, and hypoxia and oxidative stress responses, together with lower expression of proteins involved in growth signaling, anticoagulant regulation, and lipid homeostasis. MeaningSevere undernutrition is associated with a distinct high-risk clinical and biologic phenotype during severe infection, with implications for nutritional support, risk stratification, and host-directed therapeutic strategies, particularly in low- and middle-income countries.

Background: Many children under five die post hospital discharge in low-and middle-income countries (LMICs), particularly after treatment for severe infections. While some models exist, evidence on risk prediction for post-discharge mortality remains limited, with most relying solely on admission characteristics, overlooking in-hospital disease progression and discharge features. Methods: We used secondary data from prospective cohort studies in six Ugandan hospitals (2012-2021) to update models at discharge. Of 8,810 children included, 3,665 were aged <6 months and 5,145 were aged 6-60 months. Models were developed utilizing an elastic net regression approach, with admission variables selected a priori and discharge variables selected based on variable importance ranking. Performance was evaluated by applying 10-fold cross-validation, area under the receiver operating characteristic curve (AUROC), Brier score, and Net Reclassification Index (NRI). Results: Models augmented with discharge characteristics outperformed admission-only models. For children aged <6 months, the model AUROC improved by 5.1% (95% CI 3.0 - 7.3, P<0.001), achieving an AUROC of 0.81 and a Brier score of 0.06. In the 6-60m cohort, the model AUROC increased by 4.4% (95% CI 2.0 - 6.9, P<0.001), with an AUROC of 0.79 and a Brier score of 0.04. The NRI was 10.41% for children <6 months and 14.51% for those 6-60m and was achieved primarily through a reduction of false positive rates. Conclusion: Adding only three discharge characteristics to the post-discharge mortality model based on admission characteristics enhanced prediction accuracy, including model calibration, discrimination and risk stratification compared to admission-only models. Keywords: Post-discharge mortality, Risk prediction model, Elastic Net regression, Low-and middle-income countries, Child mortality, Critical illness.

IntroductionThere is increasing interest in the peripheral administration of vasopressors for two main reasons: (1) to expedite vasopressor initiation in patients with refractory shock and (2) to avoid the potential complications associated with central venous catheter placement. The current evidence on the use of peripheral vasopressor administration is primarily based on single-center observational studies. There are inconsistencies in the administration of peripheral vasopressors, including catheter gauge and location, monitoring practices, vasopressor concentrations, and duration of use. This has made it difficult for institutions to develop best practice guidelines. A randomized controlled trial is needed to address this knowledge gap. Methods and analysisThe Peripheral Use of Low-dose Vasopressors for Safety and Efficacy (PULSE) in the intensive care unit is a prospective, unblinded feasibility study. Eligible patients will be 18 years or older, have no existing central venous catheter or peripherally inserted central catheter and have the presence of shock requiring a minimum vasopressor dose of any of the following: norepinephrine 0.0625 mcg/kg/min, phenylephrine 0.625 mcg/kg/min, and epinephrine 0.0625 mcg/kg/min. Fifty patients will be randomized 1:1 into either the peripheral venous catheter or central venous catheter group. The primary outcome is feasibility, defined as (1) a recruitment rate of 4 participants per month, (2) a data capture rate of [&ge;]90%, and (3) a <50% conversion rate from peripheral to central access. The secondary outcomes include the safety of peripheral vasopressor use, alive and central-line-free days, the number of attempts needed to place a catheter, volume status, in-hospital mortality rate, ICU and hospital length of stay, and patient-centred important outcomes. ImplicationsThe data collected from this study will inform the design of a definitive randomized controlled trial to assess the safety and efficacy of protocol-driven peripheral vasopressor administration. Ethics and disseminationThis study received approval (6042888) from the Queens University Health Sciences/Affiliated Teaching Hospitals Research Ethics Boards. Results of this study will be presented at critical care conferences and submitted for publication. Trial registration numberNCT06920173 (https://clinicaltrials.gov/study/NCT06920173).

Abstract Background: Cardiovascular disease is the leading cause of mortality in Nigeria and across sub-Saharan Africa, with rising incidence attributable to urbanisation, sedentary lifestyles, and limited access to early detection tools. Concurrently, patient dropout from rehabilitation programs remains a critical operational challenge for Nigerian clinics, with many patients failing to return after their initial consultation. Methods: We developed CardioAI, an Explainable Artificial Intelligence system comprising two predictive modules. The cardiovascular risk module trained four machine learning models - Logistic Regression, Random Forest, Gradient Boosting (XGBoost), and a Multilayer Perceptron - on a combined UCI Heart Disease dataset of 1,025 patient records. A novel Lifestyle Risk Index was engineered from five modifiable clinical markers. SHAP (SHapley Additive exPlanations) was applied for per-prediction feature attribution. The patient retention module trained three classifiers on a synthetic dataset of 800 records, modelling 10 operational and behavioural dropout factors. An NLP and OCR pipeline using Tesseract v5.5 and spaCy was implemented for clinical document processing. Results: The cardiovascular risk module achieved an AUC-ROC of 0.999 (XGBoost), 0.998 (Random Forest), 0.994 (MLP), and 0.927 (Logistic Regression) on the held-out test set. Cross-validated AUC with constrained tree depth was 0.97, confirming generalisation. SHAP analysis identified the Lifestyle Risk Index, ST depression, resting blood pressure, exercise-induced angina, and cholesterol as the five most influential predictors. The retention module achieved AUC-ROC of 0.66 (Logistic Regression), demonstrating the difficulty of dropout prediction with synthetic data. Conclusions: CardioAI demonstrates that explainable machine learning can provide clinically actionable cardiovascular risk assessment and patient retention intelligence in a low-resource Nigerian healthcare context. The system is freely deployable, open-source, and designed for pilot validation in teaching hospitals across Lagos and Port Harcourt. Keywords: cardiovascular risk prediction, machine learning, explainable AI, SHAP, patient retention, clinical decision support, Nigeria, sub-Saharan Africa, XGBoost, random forest, digital health

Importance: Physicians routinely prognosticate to guide care delivery and shared decision making, particularly when caring for patients with critical illnesses. Yet, these physician estimates are prone to inaccuracy and uncertainty. Artificial intelligence, including large language models (LLMs), show promise in supporting or improving this prognostication. However, the performance of contemporary LLMs in prognosticating for the heterogeneous population of critically ill patients remains poorly understood. Objective: To characterize and compare the performance of LLMs and physicians when predicting 6-month mortality for hospitalized adults who survived critical illness. Design: Embedded mixed methods study with elicitation and comparison of prognostic estimates and reasoning from LLMs and practicing physicians. Setting: The publicly available, deidentified Medical Information Mart for Intensive Care (MIMIC)-IV v2.2 dataset. Participants: We randomly selected 100 hospitalizations of adult survivors of critical illness. Four contemporary LLMs (Open AI GPT-4o, o3- and o4-mini, and DeepSeek-R1) and 7 physicians provided independent prognostic estimates for each case (1,100 total estimates; 400 LLM and 700 physician). Main outcomes and measures: For each case, LLMs and physicians used the hospital discharge summary and demographics to predict 6-month mortality (yes/no) and provide their reasoning (free text). We assessed prognostic performance using accuracy, sensitivity, and specificity, and used inductive, qualitative content analysis to characterize reasonings. Results: Mean physician accuracy for predicting mortality was 70.1% (95% CI 63.7-76.4%), with sensitivity of 59.7% (95% CI 50.6-68.8%) and specificity of 80.6% (95% CI 71.7-88.2%). The top-performing LLM (OpenAI o4-mini) accuracy was 78.0% (95% CI 70.0-86.0%), with sensitivity of 80.0% (95% CI 67.4-90.2%) and specificity of 76.0% (95% CI 63.3-88.0%). The difference between mean physician and top-performing LLM accuracy was not statistically significant (p = 0.5). Qualitative analysis revealed similar patterns in LLM and physician expressed reasoning, except that physicians regularly and explicitly reported uncertainty while LLMs did not. Conclusion and Relevance: In this study, LLMs and physicians achieved comparable, moderate performance in predicting 6-month mortality after critical illness, with similar patterns in expressed reasoning. Our findings suggest LLMs could be used to support prognostication in clinical practice but also raise safety concerns due to the lack of LLM uncertainty expression.

Purpose: Neurogenic dysphagia is prevalent in neurological inpatients and associated with adverse outcomes, yet its independent economic impact after adjustment for frailty and functional status remains poorly quantified. We aimed to estimate the independent effect of dysphagia on hospital length of stay (LOS) and costs, to test whether this effect differs between geriatric and non-geriatric patients, and to quantify the probability and magnitude of cost savings from improvements in swallowing function. Methods: We analysed 10,375 neurological inpatient cases (2021-2024) at a German university hospital. Dysphagia was defined by fiberoptic endoscopic evaluation of swallowing (FEES) or ICD-10 R13 coding (n = 1,382; 13.3%). Bayesian Gamma-log regression with informative priors from historical data and published literature was used to model LOS and total case costs (German DRG), adjusted for age, sex, Hospital Frailty Risk Score (HFRS, R13-adjusted), self-care index ("Selbstpflege-Index", SPI), stroke status, and emergency admission. A geriatric cohort was defined as age >=70 and adjusted HFRS >=5 (n = 2,053; 19.8%). Posterior predictive simulation estimated cost savings for hypothetical improvements of 1-3 points on the Functional Oral Intake Scale (FOIS). Results: After comprehensive adjustment, dysphagia was independently associated with 46.5% longer LOS (posterior ratio 1.465; 95% credible interval [CrI] 1.397-1.537) and 28.2% higher total case costs (ratio 1.282; CrI 1.213-1.354). The dysphagia x geriatric interaction was small but credible and ran in opposite directions: slightly attenuated for LOS (interaction ratio 0.908, CrI 0.837-0.986) but slightly amplified for costs (1.096, CrI 1.012-1.185), consistent with complexity-driven DRG grouping in geriatric patients. The absolute economic burden remained larger in the geriatric cohort due to higher baseline costs. In the geriatric cohort, a one-point FOIS improvement yielded a 74.3% posterior probability of LOS-based savings (mean EUR 555/case); at three points, this rose to 84.2% (mean EUR 1,115/case). The direct cost model confirmed high benefit probabilities from the payer's perspective (82.6% at dFOIS = 3). Conclusions: Neurogenic dysphagia is an independent and substantial driver of hospital LOS and costs in neurological inpatients, even after adjustment for frailty and functional status. The proportional effect on costs is slightly larger in geriatric patients, while the LOS effect is slightly smaller, consistent with the mechanics of the G-DRG system. Bayesian simulation indicates that improvements in swallowing function carry a high probability of generating cost savings, supporting the characterisation of dysphagia as a modifiable economic target with particular relevance to geriatric neurology.

Nighttime agitation (NA) is a prevalent and challenging phenomenon affecting people with dementia (PwD), often resulting in premature institutionalization. Yet, informal caregivers' perspectives on this phenomenon remain underexplored. We conducted 15 in-depth interviews with informal caregivers to gain insight into their experiences and reactions to NA. Thematic analysis identified seven sub-themes related to carers' experience and eight sub-themes concerning their reactions. These themes emerged across three levels, namely, PwD, informal caregiver and the environment. Most phenomena occurred at a dyadic level between PwD and informal caregiver, highlighting the potential of interventions targeting dyadic coping. Informal caregivers feel insufficiently supported when sleep disturbances co-occur with NA. They primarily rely on self-initiated strategies and learn by experience. Caregivers mention the need for more advanced knowledge and skills in reacting to co-occurrence of sleep disturbances with NA or systemic support in terms of dealing with emergencies. Caregivers also reflect extensively on the impact of challenging behaviors during the night on their mental and physical well-being. Notably, no non-pharmacological interventions for NA adequately address the themes identified in this study, highlighting the urgent need for integrative approaches and recognition of caregiver wellbeing as a core outcome, not a secondary consideration in interventions.

Introduction: Current tuberculosis (TB) screening tools, such as the WHO four-symptom screen (W4SS), lack sufficient sensitivity and specificity for effective community-based active case finding, contributing to both missed diagnoses and unnecessary diagnostic evaluations. This study aimed to develop and validate a machine learning (ML) model to improve TB risk prediction among persons aged >=15 years in community settings of Zambia and South Africa. Methods: A large, harmonized dataset was created from four community-based TB prevalence surveys in South Africa and Zambia (N=169,813), restricted to individuals not under treatment at the time of survey. A binary reference outcome was defined based on available microbiological and radiographic data, grouping individuals as either 'Possible TB' or 'Unlikely TB'. An XGBoost model was trained on 80% (N=135,854) of the data using demographic, clinical, and socio-economic variables, and model interpretability was assessed using SHapley Additive exPlanations (SHAP) values. Internal validation was performed using a 20% hold-out test set (N=33,959). Model performance was assessed using discrimination, calibration, and clinical utility measures compared to the W4SS and against WHO's 2025 Target Product Profile (TPP) for a tool in a two-step screening algorithm. Results: Overall, 16,413 (9.7%) of individuals were labelled as 'Possible TB'. On the test set, the XGBoost model yielded an area under the curve (AUC) of 79.7% (95% CI: 78.7, 80.7), outperforming the W4SS (AUC 57.0%; 95% CI: 56.1, 57.8). The XGBoost model achieved 81.5% sensitivity (95% CI: 77.6, 84.9) at a 60% specificity threshold. This exceeded the W4SS, which achieved only 38.2% sensitivity (95% CI: 36.5, 39.9) on the same dataset. SHAP analysis identified age, previous TB treatment, times treated for TB and unemployment as the primary contributors to risk. Conclusion: The ML XGBoost model shows promise as a screening tool to support community-based active case finding activities prior to diagnostic testing. However, as performance remained below TPP targets, and adding variables, e.g. on geolocation, could be considered.