The immunity conferred by SARS-CoV-2 vaccines and infections reduces the transmission of the virus. To answer how the effect of immunity is shared between a reduction of infectiousness and an increased protection against infection, we examined >50,000 positive cases and >110,000 contacts from Geneva, Switzerland (June 2020 to March 2022). We assessed the association between secondary attack rate (i.e. proportion of new cases among contacts) and immunity from natural infection and/or vaccination, stratifying per four SARS-CoV-2 variants and adjusting for index cases and contacts’ socio-demographic characteristics and the propensity of the contacts to be tested. Here we show that immunity protected contacts from infection, rather than reducing infectiousness of index cases. Natural infection conferred the strongest immunity. Hybrid immunity did not surpass recent infection. Although of smaller amplitude, the reduction in infectiousness due to vaccination was less affected by time and by the emergence of new SARS-CoV-2 variants than the susceptibility to infection. These findings support the role of vaccine in reducing infectiousness and underscore the complementary role of interventions reducing SARS-CoV-2 propagation, such as mask use or indoor ventilation.

Objectives We propose and test a framework to detect disease diagnosis using a recent large language model (LLM), Meta’s Llama-3-8B, on French-language electronic health record (EHR) documents. Specifically, it focuses on detecting gout (‘goutte’ in French), a ubiquitous French term that has multiple meanings beyond the disease. The study compares the performance of the LLM-based framework with traditional natural language processing techniques and tests its dependence on the parameter used. Methods The framework was developed using a training and testing set of 700 paragraphs assessing ‘gout’ from a random selection of EHR documents from a tertiary university hospital in Geneva, Switzerland. All paragraphs were manually reviewed and classified by two healthcare professionals into disease (true gout) and non-disease (gold standard). The LLM’s accuracy was tested using few-shot and chain-of-thought prompting and compared with a regular expression (regex)-based method, focusing on the effects of model parameters and prompt structure. The framework was further validated on 600 paragraphs assessing ‘Calcium Pyrophosphate Deposition Disease (CPPD)’. Results The LLM-based algorithm outperformed the regex method, achieving a 92.7% (88.7%–95.4%) positive predictive value, a 96.6% (94.6%–97.8%) negative predictive value and an accuracy of 95.4% (93.6%–96.7%) for gout. In the validation set on CPPD, accuracy was 94.1% (90.2%–97.6%). The LLM framework performed well over a wide range of parameter values. Conclusion LLMs accurately detected disease diagnoses from EHRs, even in non-English languages. They could facilitate creating large disease registers in any language, improving disease care assessment and patient recruitment for clinical trials.

Objective To develop an automatic gout register from electronic health records (EHRs) data. Methods We analysed the EHR of all patients >18 years old from a tertiary academic hospital (2013–2022) based on six criteria: International Classification of Diseases 10 gout diagnosis, urate-lowering therapy prescription, monosodium urate crystals in joint aspiration and gout-related terms in problem lists, clinical or imaging reports. We assessed the positive and negative predictive value (PPV and NPV) of the query by chart reviews. Results Of 2 110 902 outpatients and inpatients, 10 289 had at least one criterion for gout. The combination of joint aspiration OR diagnostic in the problem list OR≥2 other criteria created a register of 5138 patients, with a PPV of 92.4% (95% CI 88.5% to 95.0%) and an NPV of 94.3% (95% CI 91.9% to 96.0%). PPV and NPV were similar among outpatients and inpatients. Incidence was 2.9 per 1000 person-year and dropped by 30% from the COVID-19 pandemic onward. Patients with gout were on average 71.2 years old (SD 14.9), mainly male (76.5%), overweight (69.5%) and polymorbid (mean number of comorbidities of 3, IQR 1–5). More than half (57.4%) had received a urate-lowering treatment, 6.7% had a gout that led to a hospitalisation or ≥2 flares within a year and 32.9% received a rheumatology consultation. Conclusion An automatic EHR-based gout register is feasible, valid and could be used to evaluate and improve gout management. Interestingly, the register uncovered a marked underdiagnosis or under-reporting of gout since the COVID-19 pandemic.

Importance The use of large language models (LLMs) in medicine is increasing, with potential applications in electronic health records (EHR) to create patient cohorts or identify patients who meet clinical trial recruitment criteria. However, significant barriers remain, including the extensive computer resources required, lack of performance evaluation, and challenges in implementation. Objective This study aims to propose and test a framework to detect disease diagnosis using a recent light LLM on French-language EHR documents. Specifically, it focuses on detecting gout (“goutte” in French), a ubiquitous French term that have multiple meanings beyond the disease. The study will compare the performance of the LLM-based framework with traditional natural language processing techniques and test its dependence on the parameter used. Design The framework was developed using a training and testing set of 700 paragraphs assessing “gout”, issued from a random selection of retrospective EHR documents. All paragraphs were manually reviewed and classified by two health-care professionals (HCP) into disease (true gout) and non-disease (gold standard). The LLM’s accuracy was tested using few-shot and chain-of-thought prompting and compared to a regular expression (regex)-based method, focusing on the effects of model parameters and prompt structure. The framework was further validated on 600 paragraphs assessing “Calcium Pyrophosphate Deposition Disease (CPPD)”. Setting The documents were sampled from the electronic health-records of a tertiary university hospital in Geneva, Switzerland. Participants Adults over 18 years of age. Exposure Meta’s Llama 3 8B LLM or traditional method, against a gold standard. Main Outcomes and Measures Positive and negative predictive value, as well as accuracy of tested models. Results The LLM-based algorithm outperformed the regex method, achieving a 92.7% [88.7-95.4%] positive predictive value, a 96.6% [94.6-97.8%] negative predictive value, and an accuracy of 95.4% [93.6-96.7%] for gout. In the validation set on CPPD, accuracy was 94.1% [90.2-97.6%]. The LLM framework performed well over a wide range of parameter values. Conclusions and Relevance LLMs were able to accurately detect disease diagnoses from EHRs, even in non-English languages. They could facilitate creating large disease registries in any language, improving disease care assessment and patient recruitment for clinical trials. Question How accurate and efficient are large language models (LLMs) in detecting diseases from unstructured electronic health records (EHR) text compared to traditional natural language processing techniques? Findings This study proposes a framework based on Meta’s Llama 3 8B, a recent public LLM, outperforming traditional natural language processing techniques in detecting gout and calcium pyrophosphate deposition disease in unstructured text. It achieves high positive and negative predictive values and accuracy. Performance was robust over a wide range of parameters. Meaning The proposed framework can ease the use of LLMs in effectively detecting disease in EHR data for various clinical applications.

Background/Objectives: Frank’s sign (FS), an acquired diagonal earlobe crease, has been associated with underlying cardiovascular (CV) disease. Pneumonia, a leading cause of hospitalization, may trigger CV events. FS may help identify patients at increased risk of CV complications following pneumonia. We aimed to evaluate the association between FS and CV events following pneumonia. Methods: We conducted a single-center prospective cohort study of internal medicine patients hospitalized with community-acquired pneumonia at a secondary hospital in Switzerland, followed for 18 months. FS was assessed at inclusion by trained investigators. Baseline characteristics were recorded. The primary outcome was a composite of CV endpoints: new-onset atrial fibrillation/flutter, myocardial infarction, hospitalization for heart failure, peripheral arterial ischemic events, or stroke. Events were identified through hospital records and general practitioner data between 2020 and 2023. Results: Among 203 patients (median age 76.0 years [IQR 65.0 to 83.0]; 41.4% female), 71.2% had FS. During follow-up, 103 CV events occurred. In univariate Cox analysis, FS was associated with an increased hazard of CV events (p = 0.042). After adjustment for confounders, FS was not significantly associated with CV events (HR 0.97; 95% CI 0.59–1.59). Only a history of CV disease and CV risk factors were associated with an increased hazard of subsequent cardiovascular events following pneumonia. Conclusions: In this population of patients hospitalized with pneumonia, Frank’s sign was not independently associated with cardiovascular events and appears to reflect underlying cardiovascular risk burden. It should not be interpreted as an independent prognostic marker.

Background: Contact tracing was one of the central non-pharmaceutical interventions implemented worldwide to control the spread of SARS-CoV-2, but its effectiveness depends on its ability to detect contacts. Aim: Evaluate the proportion of secondary infections captured by the contact tracing system in Geneva. Methods: We analysed 166,892 concomitant infections occurring at the same given address from June 2020 until February 2022 using an extensive operational database of SARS-CoV-2 tests in Geneva. We used permutation to compare the total number of secondary infections occurring at the same address with that reported through manual contact tracing. Results: Contact tracing captured on average 41% of secondary infections, varying from 23% during epidemic peaks to 60% during low epidemic activity. People living in wealthy neighbourhoods were less likely to report contacts (odds ratio (OR): 1.6). People living in apartment buildings were also less likely to report contacts than those living in a house (OR: 1.1–3.1) depending on the SARS-CoV-2 variant, the building size and the presence of shops. This under-reporting of contacts in apartment buildings decreased during periods of mandatory wearing of face masks and restrictions on private gatherings. Conclusion: Contact tracing alone did not detect sufficient secondary infections to reduce the spread of SARS-CoV-2. Campaigns targeting specific populations, such as those in wealthy areas or apartment buildings, could enhance coverage. Additionally, measures like wearing face masks, improving ventilation and implementing restrictions on gatherings should also be considered to reduce infections resulting from interactions that may not be perceived as high risk.