Michael Edelstein

BACKGROUND Systematic reviews (SRs) are a resource-intensive cornerstone of evidence-based medicine and public health. Their structured, rule-based workflow makes them amenable to large language models (LLMs) streamlining the process. Most evaluations benchmark a single model, task or corpus, leaving unresolved whether performance is valid and consistent across models, steps, review topics and workflow designs. OBJECTIVE To evaluate the validity and consistency of ChatGPT and Gemini across successive stages of SR screening, in order to determine whether LLMs can be safely integrated into SR workflows. METHODS We evaluated ChatGPT-4 and Gemini 3 against three public health SRs conducted by our research group. Human decisions served as the reference standard. Title, abstract and full-text screening were compared under two workflows: teacher forcing (TF), in which the model was constrained to the human-retained denominator at each stage, and non-teacher forcing (nTF), in which model decisions determined which records advanced. We measured sensitivity, specificity, positive and negative predictive values and F1 for each model, stage and review; interrater agreement (Cohen's and Fleiss's kappa); and within-model heterogeneity across reviews (Cochran's Q, I²). Discordant human–LLM decisions were audited, described and categorized. RESULTS Performance varied by context, stage and workflow. In nTF analyses, ChatGPT was more sensitive (76.7%–100% vs 75%–79.7% for Gemini) and Gemini more specific (82.7%–98.4% vs 83.2%–94.3% for ChatGPT). ChatGPT's sensitivity was heterogeneous across reviews (Q=15.91, p<.0005, I²=87%) whereas Gemini's was stable (I²≈0%); specificity was heterogeneous for both (I²=75% for ChatGPT and 98% for Gemini). Precision remained low throughout. Discordance analysis indicated systematic rather than random misclassification. Human–model agreement was modest to moderate (Cohen's kappa 0.255–0.442) and lower than inter-model agreement (up to 0.781). CONCLUSIONS LLMs can improve efficiency mainly as high-sensitivity assistants that reduce screening burden while leaving final adjudication to humans. Their safest role is within audited, human-guided hybrid workflows rather than fully autonomous pipelines, and potential efficiency gains should be formally quantified against possible losses in quality and completeness.

Background: The increasing use of fluorine-18 fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) in patients with breast cancer has led to a growing number of incidentally detected thyroid lesions. Thyroid incidentaloma refers to a focal area of increased metabolic activity within the thyroid gland. Up to 35% of focal thyroid incidentalomas are malignant. Early detection of thyroid carcinoma may allow less invasive surgical management and reduce the need for more extensive treatment associated with advanced disease, including total thyroidectomy, neck dissection, and radioactive iodine therapy. The objective of this study was to estimate the prevalence of focal thyroid incidentaloma and incidental thyroid carcinoma detected on PET/CT scans performed in breast cancer patients at a single tertiary medical center serving a highly ethnically diverse population. Methods: Medical records and PET/CT scans were retrospectively reviewed at a single medical center. Data collected included the presence of thyroid incidentaloma, maximum standardized uptake value (SUVmax), thyroid cytology, primary malignancy type, breast cancer status, ethnicity, and age. A total of 1233 patients with cancer who underwent PET/CT imaging and received treatment at ZIV Medical Center between August 2018 and December 2024 were included. Forty-two patients with primary thyroid carcinoma or head and neck carcinoma were excluded. Patients were categorized into breast cancer and non-breast cancer groups and compared regarding the prevalence of thyroid incidentaloma, referral rates for further evaluation by head and neck specialists, and the rate of incidental thyroid carcinoma. Results: Among 330 patients with breast cancer, 16 (4.8%) had a focal incidental thyroid finding, compared with 24 (2.8%) of 861 patients with non-breast malignancies who underwent PET/CT imaging. Among the 16 breast cancer patients with thyroid incidentaloma, thyroid carcinoma was subsequently confirmed in 4 patients (25%) who underwent further evaluation. The proportion of patients with confirmed thyroid carcinoma among those with incidentalomas was 1.2% (4/330) in the breast cancer group compared with 0.46% (4/861) in the non-breast cancer group. Approximately half of the patients in both groups were referred for further thyroid evaluation. Among breast cancer patients, non-significant trends toward higher referral rates were observed in patients with non-advanced disease, higher SUVs, and Jewish ethnicity. Nearly all patients evaluated by head and neck specialists underwent fine-needle aspiration (FNA). The mean SUV among patients diagnosed with papillary thyroid carcinoma (PTC) on FNA (n = 8; breast and non-breast cancer combined) was 15.2 (IQR: 7.3–17.0), compared with 5.2 (IQR: 5.0–6.0) among patients with benign cytology (n = 5; p = 0.011). Conclusions: Thyroid incidentalomas identified on PET/CT scans in breast cancer patients were relatively common in this cohort, and a proportion of evaluated lesions were subsequently diagnosed as thyroid carcinoma. Referral and diagnostic follow-up rates were variable, with approximately half of patients referred for further evaluation and only one-third attending a head and neck clinic. Given the retrospective design, limited number of biopsy-confirmed cases, and potential verification bias, these findings should be interpreted with caution. Nevertheless, they highlight the importance of clinical awareness and support individualized assessment of PET/CT-detected thyroid incidentalomas rather than broad management recommendations.

Purpose of research Long COVID (LC) risk varies across SARS-CoV-2 variants, but causal inference is complicated by rising vaccine coverage coinciding with viral evolution. We examined whether pre-infection vaccination status modifies the association between SARS-CoV-2 variant and LC risk, to disentangle the contributions of viral evolution and vaccine-induced immunity. Principal results In this prospective cohort study of Israeli adults with SARS-CoV-2 infections from July 2021–March 2023, participants completed baseline and follow-up online surveys ≥60 days post-infection. LC was defined as ≥3 symptoms within validated symptom clusters. Infections were assigned to dominant circulating variants (Wild type, Alpha, Delta, Omicron). Pre-infection vaccination was categorized as unvaccinated (0–1 doses), partially vaccinated (2 doses), or fully vaccinated (≥3 doses). Generalized estimating equations estimated multivariable risk ratios, adjusted for age, sex, and follow-up time. Among 1715 participants reporting 2018 infections, LC was reported following 30·4% of infections. In models adjusted for vaccination, LC risk did not differ between Delta and Omicron variants (aRR 1·01, 95% CI 0·80-1·26). Among partially vaccinated individuals, Omicron infections carried lower LC risk than Delta (aRR 0·62, 95% CI 0·40–0·98). LC prevalence did not differ by variant among unvaccinated or fully vaccinated individuals. Major conclusions Variant-specific differences in LC risk may be largely explained by concurrent increases in vaccine coverage rather than intrinsic changes in variant pathogenicity. Full pre-infection vaccination confers consistent protection against LC regardless of variant.

Research on under-vaccination often segments populations using demographic or administrative variables that are operationally useful but fail to capture identity dimensions relevant to vaccination decisions. Drawing on social identity theory, we propose an identity-landscape approach distinguishing identity membership, identity centrality, and multidimensional identity structure. Using a cross-sectional survey of 1,000 UK parents, we measured 65 identity indicators, identity-importance ratings, and their association with attitudinal and behavioural hesitancy toward childhood vaccination using validated scales. Beyond established socio-demographic predictors, alternative-medicine and natural-lifestyle identities, as well as affiliation with social media networks, were linked to greater hesitancy. Greater centrality of religion and political affiliation within personal identity was also associated with higher hesitancy. Principal component analysis suggested that individuals actively engaged across multiple societal issues were more hesitant, whereas stereotypically male-gendered engagement was associated with lower hesitancy. An identity-focused population segmentation may identify previously unrecognized undervaccinated groups and inform innovative tailored immunization campaigns.

Angiotensin converting enzyme 2 (ACE2), a component of the renin-angiotensin aldosterone system (RAAS), is the main receptor for SARS-CoV-2 entry into huma

Background The long-term impact of COVID-19 vaccination on post-acute COVID-19 symptoms and associated quality of life (QoL) changes remains incompletely described. This study aimed to explore the impact of the timing of COVID-19 priming and booster doses, on reporting long COVID symptoms and associated QoL changes. Methods Individuals who had PCR testing for SARS-CoV-2 processed in government hospitals in Northern Israel between 15^th March 2021 and 15^th June 2022 were invited to answer serial online surveys collecting information on SARS-CoV-2 infection, vaccination status and post-acute symptoms every 3-4 months for two years. Participants were categorized into groups based on the number of doses received prior to infection. We compared these groups over time in terms of reporting post-COVID symptom clusters and QoL, using population-average and mixed-effects regression models, respectively. Results A total of 4809 individuals are enrolled and respond to up to five follow-up surveys. Of these, 1377 (28.61%) report a positive SARS-CoV-2 test, while 3432 (71.39%) report a negative result. After adjustment for potential confounders, receiving at least three COVID-19 vaccine doses prior to infection is associated with a 34% reduction in the likelihood of reporting at least one long COVID symptom cluster compared to being unvaccinated (adjusted odds ratio [aOR] = 0.66, p = 0.022). Pre-infection vaccination is also associated with higher quality of life (QoL) scores (β = 0.07, p < 0.001). The estimated vaccine effectiveness of three pre-infection doses against long COVID over a two-year period is 26.5% (95% CI: 10.8–39.4). This protective effect remains stable over time. In contrast, vaccination received after infection shows no association with long COVID symptoms or QoL outcomes. Conclusions Receiving at least three COVID-19 vaccine doses prior to SARS-CoV-2 infection provides a sustained protective effect against long COVID and its negative impact on quality of life for at least two years. The longer-term durability of this protection, the role of reinfection, and the influence of emerging viral variants remains to be investigated.

Acute respiratory tract infections (ARI), including COVID-19, are a significant global health challenge with pandemic potential. Understanding factors influencing ARI susceptibility and severity in different populations is crucial for pandemic preparedness and mitigation. We investigated epidemiological risk factors for COVID-19 susceptibility and severity in the Israeli population. Participants who took a COVID-19 PCR test between July 2021 and August 2022 completed an online survey including demographics, comorbidities, infection and vaccination status. We identified factors associated with SARS-CoV-2 infection and disease severity using logistic regression and compared our results with the global literature. We included 2128 participants. Of these 823 participants were tested positive for SARS-CoV-2. Individuals without COPD and BMI > 30 had higher infection odds (aOR = 6.21, 95% CI:1.42–27.25 and aOR = 1.54, 95% CI:1.02–2.32; respectively). High school-level education was associated with increased susceptibility and severity compared to higher education (aOR = 1.55, 95% CI:1.1–2.18 and aOR = 2.80, 95% CI 1.62–4.86; respectively). Hospitalization was associated with older age (p < 0.001). Vaccination with 3 doses significantly reduced infection and hospitalization risk compared to unvaccinated (aOR = 0.024, 95% CI:0.016–0.04 and aOR = 0.37, 95% CI: 0.17–0.81; respectively, p < 0.05). Compared to Arab participants, Jewish ones were less likely to be infected (aOR = 1.48, 95% CI:1.02–2.15, p < 0.05). Identified risk factors are consistent with meta-analyses of studies from other countries. Similar to other countries, minority groups were found to be at higher risk, therefore risk models predicting ARI susceptibility and severity have to take this into account.