共找到 20 条结果
The limited availability of diverse and representative training data poses a critical barrier to the development of clinically relevant computational tools for intraoperative surgical decision support. Surgical procedures are not routinely recorded, and data annotation requires domain expertise, resulting in a scarcity of open-access surgical video datasets with high-quality annotations. Existing datasets are typically limited to single institutions and specific procedures, such as cholecystectomy, and rarely comprise patient-level metadata like demographic characteristics, disease history, or laboratory parameters. The Appendix300 dataset comprises 330 laparoscopic surgery recordings, including 325 full-length laparoscopic appendectomies and 5 control recordings from non-appendectomy procedures in pediatric and adult patients treated at five German centers. The dataset includes patient-level clinical metadata (demographics, medical history, clinical symptoms, preoperative laboratory parameters, and histopathologic findings, as well as standardized expert annotations of the laparoscopic grade of appendicitis). This dataset enables novel validation tasks for computer vision in laparoscopic surgery and facilitates simulation of decentralized learning approaches, overall enhancing the breadth and translational relevance of AI-based surgical video analysis.
This letter to the editor offers a constructive commentary on a recently published multivariate analysis of predictors of complications in abdominoplasty. The authors commend the original study for its rigorous surgical standardization, consecutive patient enrollment, and single‑surgeon design, and agree that high body mass index and active smoking are independent risk factors, while preservation of Scarpa's fascia appears protective. However, three methodological limitations are identified. First, the sample size and number of events barely meet the minimum recommended events‑per‑variable ratio for multivariate regression. Second, treating seroma, necrosis, and dehiscence as independent outcomes ignores competing risks among these complications, which may bias risk estimates. Third, the single‑center, single‑surgeon design limits external validity, and the finding that diabetes is not a risk factor contradicts some existing literature. The letter suggests that future analyses adopt competing‑risk models and include multicenter external validation. It also notes that no new patient data are provided, so the proposed statistical refinements await empirical verification.Level of Evidence V This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .
Spatial-temporal outlier detection is frequently carried out using model-free, spatial-only methods (e.g., Local Moran's I) that disregard temporal dependencies, or through univariate per-location criteria that neglect spatial dependencies, both of which may misclassify brief, spatially coherent episodes or overlook events obscured by these dependencies. This research proposes a dynamically aware, residual-based outlier detection method for the GSTAR model, which operates directly on the space-time residuals. The process involves (i) computing the residuals of the GSTAR model, (ii) identifying outliers based on these residuals, and (iii) incorporating an outlier factor into the model, followed by estimating the parameters of the revised model and reverting to step (i). This technique is iteratively performed until no outliers are identified and the model satisfies the white noise condition. This approach utilizes weekly forest fire hotspot counts from 13 districts in West Kalimantan (June 2023-March 2025), employing a row-standardized Queen Contiguity matrix for spatial weighting. The outcomes of outlier detection utilizing the proposed technique are thereafter compared with those derived from the spatial detection method (Local Moran's I) for each observation period. The findings indicate a more accurate detector, fewer alerts but better calibrated to temporally anomalous events, and a constant decrease in the MSE of GSTAR forecasts one step following the introduction of the indicator. This approach produces a localized and interpretable spatial-temporal signal with minimal computational expense. It is compatible with conventional spatial weight building, offering a viable and reproducible framework for detecting spatial-temporal anomalies in environmental monitoring.
暂无摘要（点击查看详情）
Interpretation of ethanol results in postmortem blood samples has well-known limitations. After death, ethanol can form in the blood if sufficient glucose and microorganisms are present to facilitate the fermentation process. The potential for postmortem formation of ethanol has led to the widespread practice of verifying blood ethanol concentrations in a secondary matrix, typically vitreous humor when available. This study examined underlying trends and causes of postmortem ethanol formation in vitreous humor. Postmortem casework from 1 January 2022 to 31 December 2024 was investigated for cases where ethanol was reported in vitreous humor but not detected in a paired blood sample. Volatile analysis was performed by headspace gas chromatography on a dual column with a flame ionization detector. Ethanol, methanol, isopropanol, and acetone were screened in both blood and vitreous humor at reporting limits of 10, 10, 5.0, and 5.0 mg/dL, respectively. During this timeframe, 858 cases were identified with vitreous humor ethanol concentrations ranging from 10 to 624 mg/dL (mean 49 mg/dL, median 20 mg/dL). Of those identified, 683 cases (79%) had low-range vitreous ethanol concentrations (10-49 mg/dL). Possible biomarkers of ketoacidosis were also assessed in the 858 cases, with 464 cases (54%) reporting positive acetone and/or isopropanol. Additionally, elevated levels of beta-hydroxybutyric acid (BHB) (≥ 250 µg/mL) in either the blood or vitreous humor, were detected in 92% of cases (n = 140) tested for BHB; elevated levels of glucose (≥ 200 mg/dL) in vitreous humor were detected in 37% of cases (n = 458) tested for glucose. The results of this study highlight the importance of testing multiple matrices for ethanol in postmortem casework and the potential for postmortem ethanol formation in unpreserved specimens.
Hand, foot, and mouth disease (HFMD), particularly EV-A71, has caused large-scale epidemics in the Asia-Pacific region, with China alone reporting over 2 million annual cases and causing a substantial morbidity and mortality. While EV-A71 vaccination was introduced in 2016 and COVID-19 non-pharmaceutical interventions (NPIs) were implemented from 2020-2022, comprehensive evidence on their independent and combined effects on HFMD epidemiology remains scarce. We aimed to quantify the impact of vaccination, NPIs, and post-pandemic policy relaxation on HFMD incidence and pathogen distribution. We conducted an interrupted time series segmented regression analysis using a 14-year of surveillance dataset (January 2011-December 2024) from Beijing's National Notifiable Disease Reporting System and citywide pathogen surveillance network. The study population included all HFMD cases reported in Beijing. We defined four intervention periods: baseline (pre-vaccination), vaccination introduction (August 2016), stringent NPIs (January 2020), and reopening (January 2023). Primary outcomes were HFMD incidence rates and incidence rate ratios (IRRs). We estimated prevented cases under counterfactual scenarios. Among 324,623 reported HFMD cases across the 14-year study period, interrupted time series regression demonstrated that vaccination was associated with a 33% reduction in incidence (IRR 0.67, 95% CI 0.49-0.91); cumulatively over 8 years (2016-2024), this was estimated to have prevented 23.8% of cases that would have occurred without vaccination. Stringent NPIs were associated with an 84% reduction (IRR = 0.16, 95% CI: 0.06-0.41); cumulatively, during the intervention period (2020-2022), NPIs were estimated to have prevented 82.6% of cases. Following policy relaxation, incidence rebounded 2.53-fold (95% CI 0.64-9.92), although this increase did not reach statistical significance. In absolute terms, monthly mean case counts declined from 3,001 (baseline) to 1,886 following vaccination introduction (37.2% decrease), and further to 299 during stringent NPIs (90.0% decrease from baseline), before surging to 1,503 post-reopening, a fivefold increase from the NPI period. Pathogen surveillance revealed EV-A71 prevalence declined from 30.5% (baseline) to 4.5% (vaccination period), while coxsackievirus A6 emerged as the dominant serotype, from 9.5% (baseline period) to 56.2% (vaccination period). EV-A71 vaccination and NPIs were each associated with reduced HFMD transmission, but policy relaxation was followed by a substantial resurgence. The observed serotype replacement with non-vaccine-type enteroviruses highlight critical gaps in current prevention strategies. Sustaining HFMD control requires optimizing vaccine design, developing multivalent vaccines, maintaining high coverage, and integrating targeted NPIs during high-transmission seasons.
Ethiopia has not only one of the largest livestock populations in Africa, but also a substantial growing human population, increasing the risk of zoonotic disease transmission. Anthrax is a priority zoonosis in Ethiopia due to its risks to both human and animal health. Utilising 13 years of retrospective data (2008-2020) from 92 zones, this study investigated the spatio-temporal distribution of livestock anthrax cases in Ethiopia. High variability in annual incidence was observed, with the lowest number of cases (n=2357, 3.73 %) in 2018 followed by the highest number of cases (n=8210, 12.98 %) in 2019. Cattle contributed the majority of cases (n=36,104; 57.09 %) with relatively even distribution across the other species (sheep, goats, camels, and equines). Kruskal-Wallis tests found that incidence of anthrax per 100,000 animals varied significantly by species, year, region and zone, with the largest effect size (0.20) observed between zones. Getis-Ord General G analysis found significant clustering of anthrax cases per 100,000 animals in multiple years for cattle (2013, 2018 and 2019), goats (2009, 2010 and 2012), and sheep (2013, 2015, 2016, and 2019). Getis-Ord Gi* hotspot analysis of data from these years identified zones contributing to these hotspot clusters, but none of the hotspots occurred in multiple species in the same location and year. Cattle hotspots were limited to zones in the Somali region and goat hotspots to neighbouring zones of the Amhara and Oromia regions. Sheep hotspots were not restricted in geographic distribution. Understanding these patterns is vital to coordinating anthrax control and prevention efforts in Ethiopia.
To determine the incidence of placental chorangioma and evaluate its associated perinatal complications and mortality. A retrospective observational study was conducted at Fernandez Hospitals, Hyderabad, India, from March 2011 to March 2025. All singleton pregnancies with structurally normal fetuses diagnosed antenatally with placental chorangioma and delivered at our center, with postnatal confirmation by histopathological examination (HPE), were included in the study. The diagnosis was made using B-mode ultrasound examination and color Doppler flow imaging, demonstrating a well-circumscribed hypo or hyperechoic placental mass with a clearly identifiable vascular supply. Demographic details, prenatal sonographic findings, perinatal and neonatal outcomes were collected from electronic medical records. A total of 20 cases were included, with an incidence of 9.8 per 100 000 pregnancies. Large chorangiomas (≥ 4 cm) accounted for 90% of cases, while small chorangiomas (< 4 cm) accounted for 10%. The mean gestational age at diagnosis was 27.65 ± 5.27 weeks. Sixteen cases (80%) were managed conservatively, while four cases (20%) required prenatal intervention. Polyhydramnios was observed in 12 (60%) fetuses and cardiomegaly in 5 (25%). Fifteen fetuses (75%) were appropriate for gestational age, four (20%) had fetal growth restriction, and one (5%) was large for gestational age secondary to hydrops. Eleven pregnancies (55%) resulted in preterm birth. Nine neonates (45%) required admission to the neonatal intensive care unit (NICU), and eight (40%) experienced adverse neonatal outcomes. No neonatal deaths occurred despite the predominance of large chorangiomas in the study population. Serial ultrasound surveillance with meticulous fetal assessment enabled early identification of asymptomatic chorangiomas and timely intervention when required, resulting in favorable perinatal outcomes.
Throughout the pandemic, COVID-19 cases tended to organize into spatial clusters. In the absence of individual and/or network information on cases, however, surveillance systems face challenges in identifying the pathways through which these clusters emerge. Using Québec COVID-19 surveillance data from 2020 to 2022, this study identified characteristics associated with the likelihood of being among the first cases in a cluster. Our results show that young adults were consistently at higher risk of constituting the first cases of large community clusters across all waves, suggesting that closer monitoring of COVID-19 incidence in this age group could be pivotal for early detection of emerging clusters or new waves. Overall, our study provides valuable insights for public health authorities by helping to identify priority groups at the forefront of cluster emergence and by supporting the development of tailored and adaptive interventions.
Laparoscopic common bile duct exploration (LCBDE) remains an effective single-stage strategy for choledocholithiasis, but declining utilization has raised concerns about diminishing operative experience and training exposure among contemporary surgeons. We evaluated institutional trends in LCBDE, intraoperative cholangiography (IOC), and resident participation and identified predictors of successful duct clearance. This retrospective single-center cohort study (2012-2021) included adults undergoing laparoscopic or robotic cholecystectomy. Patients were grouped as single-stage LCBDE with cholecystectomy or two-stage management with ERCP and endoscopic sphincterotomy followed by cholecystectomy. Cases were identified using CPT and ICD procedure codes cross-referenced with operating room logs. Data were abstracted from the electronic medical record. Primary outcomes were temporal trends in LCBDE, IOC utilization, and resident participation. Secondary outcomes were predictors of successful LCBDE. LCBDE succeeded in 240/291 cases (82.4%). Failures occurred exclusively during transcystic exploration, most commonly due to an inability to cannulate the cystic duct (43%). Median length of stay was shorter after successful LCBDE (1 vs 3 days, p < 0.001). Surgeons in successful cases had greater experience (median 17.5 vs 12.9 years, p = 0.03). On multivariable analysis, surgeon experience independently predicted success (OR 1.04 per year, 95% CI 1.004-1.078), whereas surgeon specialty was not significant after adjustment. Of 303 LCBDE patients, 291 met inclusion criteria; among 339 two-stage patients, 257 met inclusion criteria. Institutionally, annual LCBDE volume declined (trend p < 0.05), IOC utilization decreased (66% in 2018 to 54% in 2021), and resident involvement fell (91.8% in 2014 to 69.6% in 2021), with per-resident exposure declining from 2.2 to 0.9 cases per year. LCBDE achieves high duct-clearance rates and shorter hospitalization when successful. Surgeon experience, rather than specialty, independently predicts success, underscoring a clinically meaningful learning curve. Declines in LCBDE volume, IOC utilization, and resident exposure highlight the need for training strategies including simulation, standardized workflows, and improved case access to preserve competency.
Multimodal foundation models have shown compelling but conflicting performance in medical image interpretation. However, the ways in which these models integrate and prioritize different data modalities, including images and text, remain poorly understood. Here we evaluate 8 proprietary and open-source multimodal foundation models using 1090 multimodal medical cases. We show that image predictions are largely driven by text, with accuracy increasing monotonically with the amount of informative text. Exploitation of text is a double-edged sword; even mild suggestions of an incorrect diagnosis in text diminish image-based classification, dramatically reducing performance in cases the model could previously answer using images alone-o3 accuracy fell from 84% to 28% when a misleading clinical vignette was introduced. In physician evaluations of long-form cases, adding images reduces or does not improve performance when text is highly informative (e.g., GPT-4V showed decreased accuracy when images were added to highly informative text across 69 clinicopathological conferences). Our results suggest that multimodal AI models may be useful in medical diagnostic reasoning but that their accuracy is largely driven, for better and worse, by text.
Lynch syndrome is an inherited cancer predisposition syndrome caused by germline pathogenic variants in mismatch repair (MMR) genes and is primarily associated with colorectal and endometrial cancers. Classically, Lynch syndrome-associated tumors exhibit microsatellite instability (MSI) and loss of MMR protein expression on immunohistochemistry (IHC); therefore, MSI and IHC are routinely used in screening algorithms to support diagnostic evaluation. ; Methods: In this study, we present the molecular and pathological characteristics of four cases from three unrelated families carrying the same MSH2 variant [c.70C > T (p.Gln24Ter)], interpreted as pathogenic, who were diagnosed with breast and colorectal cancer. MSI analysis and MMR protein expression were evaluated in available tumor tissues. ; Results: Preserved MMR protein expression was observed in three cases, and two cases were microsatellite stable (MSS), whereas only one case demonstrated loss of MMR protein expression and an MSI-high phenotype consistent with the classical Lynch syndrome profile. These findings suggest that individuals carrying the same MSH2 variant may exhibit heterogeneous tumor-level MMR/MSI phenotypes. Early truncating variants in MSH2 may allow partially functional protein production via alternative translation initiation, potentially limiting complete loss of MMR function. Variability in somatic second-hit mechanisms and tumor-specific molecular pathways may also contribute to this heterogeneity. ; Conclusion: In conclusion, Lynch syndrome tumor biology may be more heterogeneous than expected, and MSI and IHC should not be interpreted in isolation as exclusionary tests for an underlying germline MMR variant, but rather as markers of tumor-level biological consequences. Therefore, germline findings, tumor characteristics, and family history should be evaluated together in Lynch syndrome diagnosis and risk assessment.
This review provides an overview of analytical techniques that have been published for detecting the (fraudulent) addition of reconstituted bovine milk to liquid bovine milk and identifies which of these techniques are most suitable for identifying partial replacement of liquid bovine milk by reconstituted bovine milk at levels that are economically meaningful. Evaluation of these analytical techniques was done against a detection limit of 10% inclusion of reconstituted milk. In total, 30 studies were included and categorized to deal with type II heat indicators, spectroscopic techniques, omics techniques and other techniques, respectively. The variation in the lowest detectable level of added reconstituted milk powder to liquid milk is very high, ranging from 0.5 to 50%, highly depending on the fat content and temperature treatment of both reconstituted milk and liquid milk. In cases that liquid milk and reconstituted milk with similar fat content and comparable processing histories are combined, the sensitivity of most techniques decreases. In many cases the sensitivity even may be insufficient to reliably identify adulteration at economically relevant levels. Many recent publications describe omics-based techniques which show high potential for detecting fraudulent addition of reconstituted milk in liquid milk, even though these techniques in many cases need further exploration of their potential by applying the techniques to a sample set reflecting the typical adulteration practices. In conclusion, the detection of fraudulent addition of reconstituted bovine milk in bovine liquid milk remains challenging and relevant up to today.
Tuberculosis (TB) remains a major public health challenge in Brazil, strongly influenced by social and territorial inequalities. Although the Federal District has a relatively low overall incidence, pronounced internal heterogeneity may conceal localized areas of sustained transmission. This study aimed to analyze temporal trends, spatiotemporal distribution, and demographic and epidemiological determinants of TB incidence in the Federal District, Brazil, from 2010 to 2024. We conducted an ecological study including all newly reported TB cases. Temporal trends were assessed using time-series decomposition and Joinpoint regression. Spatial and space-time clusters were identified using scan statistics, and spatial variation in temporal trends was assessed. A decomposition analysis was conducted to quantify the contributions of population growth, aging, and epidemiological changes in TB incidence. Between 2010 and 2024, 5149 TB cases were reported. The incidence rate increased from 8.5 to 13.1 cases per 100,000 inhabitants, with an initial rise, a prolonged decline until 2022, and a recent resurgence, resulting in an overall increasing trend. Seasonal patterns were stable, with peaks in April and August. Spatial analyses identified persistently high-incidence areas, with a primary space-time cluster in the Administrative Region of Sobradinho during 2012-2018, with a relative risk of 5.39. Localized acceleration of transmission was observed in areas such as Plano Piloto, SIA, and Varjão. Decomposition analysis indicated that changes in TB incidence were driven predominantly by epidemiological factors. In conclusion, TB in the Federal District exhibits dynamic temporal behavior, marked spatial heterogeneity, and concentration of risk in specific territories, largely driven by epidemiological determinants.
Breast abscesses are localised infections that can be lactational or non-lactational. Non-lactational abscesses are more often associated with less common bacteria. We describe a case of Klebsiella pneumoniae as the causative bacteria of a non-lactational breast abscess. Resistance to amoxicillin was noted, and recurrence occurred despite multiple treatments with appropriate antibiotics. Definitive management required incision and drainage of the abscess. Our literature review shows 22 total cases of K. pneumoniae being isolated from breast abscesses, with only 9.1% of cases confirming full susceptibility to all tested antibiotics. Resistance to at least one antibiotic was seen in 45.5% of cases. The literature shows that rates of antibiotic resistance in K. pneumoniae are rising, and therefore, monitoring of antibiotic resistance in these patients may be indicated. Culture and sensitivity should be encouraged as this may guide management options for recurrent breast abscesses.
Malignant intestinal obstruction (MIO) is a severe complication of advanced cancer. Traditional static assessment models struggle to capture its dynamic pathological mechanisms, limiting their clinical value. To address this, this study developed a multimodal machine learning framework. Core features (including the dynamic tumor enhancement ratio TER) were extracted via Lasso-Boruta dual-modality screening, and risk prediction was performed using an XGBoost-Random Survival Forest (RSF) cascade model. Results demonstrated an AUC of 0.84 ± 0.03 and Brier score of 0.19 in the internal validation cohort, with robust external validation performance. Clinical translation reduced mechanical ventilation duration by 41% and lowered antibiotic resistance rates from 37 to 14%. This approach ultimately provides dynamic, interpretable decision support for precise MIO diagnosis and treatment. This study enrolled 320 MIO patients, randomly divided into a training set (192 cases), internal validation set (64 cases), and external validation set (64 cases) at a 6:2:2 ratio.
DNM1 encephalopathy is a rare autosomal dominant genetic condition characterized by a range of neurological and developmental manifestations. The typical phenotype is severe, including profound intellectual disability, treatment-resistant epilepsy, ataxia, and structural brain abnormalities. However, milder presentations have increasingly been reported, suggesting that the full phenotypic spectrum remains incompletely defined. We describe the case of a female patient who presents only with well-controlled epilepsy and mild dysmorphic features. Trio exome sequencing identified a de novo likely pathogenic variant in the GTPase domain of DNM1, supporting the diagnosis of DNM1 encephalopathy. Historically, disease-causing variants in the GTPase domain were thought to confer a more severe presentation. However, our patient demonstrates a more attenuated phenotype than previously reported cases, including those viewed as mild, which have consistently included some degree of developmental impairment. This supports an expanded and more variable phenotypic spectrum of DNM1 encephalopathy than previously recognized. This additionally complicates domain-based prognostication; though further cases will aid in elucidating genotype-phenotype correlation.
Influenza and SARS-CoV-2 are the primary contributors to seasonal respiratory infections and frequently co-circulate, creating significant health challenges. The present respiratory surveillance study was conducted in Dibrugarh, Assam, India from January 2025 to August 2025 to investigate the genomic characteristics of circulating viruses and identify potential co-infections. Overall, 4,948 respiratory samples were screened using multiplex real-time PCR, followed by subtyping of Influenza A and Influenza B. Next-generation sequencing (NGS) was performed in selected positives of SARS-CoV-2 and Influenza A. Genomic analysis included mutational profiling, phylogenetic analysis and N-glycosylation site prediction using bioinformatics tools. Two co-infection cases were detected: one involving Influenza A (H3N2) with SARS-CoV-2 (Omicron XFG lineage) and another involving Influenza A (H3N2) with Influenza B (Victoria lineage). Both patients experienced mild illness without hospitalisation. NGS revealed that the Influenza A (H3N2) viruses belonged to clade 3C.2a1b.2a.2a.3a.1 while SARS-CoV-2 sequence was classified under the Omicron XFG lineage. Mutational analysis of the HA gene showed several amino acid differences compared to the reference vaccine strain A/Darwin/6/2021. N-glycosylation analysis predicted conserved sites at positions 79, 181, 262, and 301 in all strains along with an additional predicted site at position 110 in both co-infection cases. Although the co-infection cases presented with mild clinical manifestations, the observed genomic variations indicate a potential role of co-infecting viruses in shaping viral evolution. Given the limited genomic data available from Northeast India, the study underscores the need for sustained large scale follow up and genomic surveillance to monitor emerging mutations and target future vaccine strategies.
Type B aortic dissection (TBAD) and Type B intramural hematoma (IMH) are 2 pathologies that fall within acute aortic syndromes. Complication risk in TBAD is determined by false lumen hemodynamics, whereas IMH progression is driven more by morphologic factors. However, it is still debated as to how these high-risk features, in the absence of complications, should affect their management. To synthesize current evidence and evaluate whether TBAD-analogous treatment protocols are most appropriate for Type B IMH, a narrative review was conducted of guidelines, retrospective cohorts, registries, and studies assessing optimal medical therapy and thoracic endovascular aortic repair (TEVAR). The focus was on high-risk features, intervention timing, and guideline discordance. IMH demonstrates greater early-phase instability. Optimal medical therapy, for low-risk cases, and TEVAR, for complicated pathology, are effective for both IMH and TBAD. However, high-risk IMH assessment and course pose greater variability. There are low-quality evidence and reports for early prophylactic TEVAR in high-risk IMH and TBAD. Surveillance recommendations are similar but supported by weaker IMH evidence. Despite overlapping initial medical management strategies in low-risk cases, Type B IMH requires a management approach that is not fully analogous to TBAD. Its unique risk factors and unpredictable course require individualized decision making rather than protocolized care at the current state of evidence. Harmonized risk criteria, optimized intervention timing, and concrete prospective studies are needed to strengthen IMH-specific management.
Area-level livability indices are commonly used tools for policymakers to evaluate and improve built and social environments, but their relationship to mental health is not well-understood. In this study, we used geographically weighted Poisson regression (GWPR) to measure spatially varying associations between neighborhood livability and poor mental health across New York City (NYC) census tracts. A small-area livability index was previously constructed with publicly available data for 2101 census tracts in NYC using principal components analysis applied to 38 variables across six domains: socioeconomic, housing, resources, environment, safety, and accessibility & transportation. The outcome was cases of self-reported mental health not good for ≥14 days in past month among adults (CDC 500 Cities - 2016). GWPR models measured spatially varying associations of poor mental health cases, offset by adult population size, with the overall livability score and, separately, with each of the domain sub-scores. Estimated risk ratios were mapped to identify areas in NYC where livability domains acted as protective/risk factors for poor mental health, depending on the location. Socioeconomic livability was a geographically consistent protective factor (negative associations across all census tracts), while the overall livability score and other domain sub-scores showed geographically varying associations with poor mental health (associations ranging from negative to positive). Geographic areas showing protective associations between livability and poor mental health highlight opportunities to improve mental health outcomes through targeted domain-specific interventions. In areas where elements of livability were positively associated with poor mental health, tackling mental health disparities may require more multifaceted approaches.