Jialin Mao

Background: Antibiotic strategies for drug use–associated infective endocarditis (DUA-IE) are evolving, but data are limited regarding antibiotic access after patient-directed discharge (PDD). We examined antibiotic fill patterns and 30-day readmission among DUA-IE patients with and without PDD. Methods: Using national Medicaid data (2016–2022), we identified adults aged 18–64 with new DUA-IE episodes. We measured antibiotic fill rates among patients discharged to nonfacility settings. We used a landmark design, requiring readmission-free survival for 3 days after discharge to observe antibiotic fills. We used multivariable logistic regression to assess associations between PDD and antibiotic fill with 30-day readmission, overall and stratified by PDD status. Results: Among 19 808 patients included, 5519 (27.9%) left via PDD. Compared with non-PDD patients, PDD patients were younger, had fewer comorbidities, and had more polysubstance use. Only 17.1% of PDD patients had an antibiotic fill within 3 days vs 44.2% of non-PDD patients (adjusted odds ratio [aOR] 0.25; 95% confidence interval [CI], .20–.33). Patient-directed discharge was associated with increased readmission (aOR 1.91; 95% CI, 1.65–2.21), while antibiotic fill was associated with reduced readmission (aOR 0.45; 95% CI, .40–.50). In stratified analyses, antibiotic fill was associated with a larger magnitude of readmission reduction among PDD patients (aOR 0.27; 95%CI, .21–.34) than among non-PDD patients (aOR 0.50; 95% CI, .46–.55; P for interaction <.001). Conclusions: These findings highlight the need for proactive strategies that ensure treatment continuity including planning for potential PDD, bedside medication delivery, and expanded use of oral and long-acting intravenous antibiotics.

Background: Prior evidence indicate that differences in treatment settings between patients with colorectal cancer (CRC) from high-poverty areas (HPA, ≥ 20% residents living under poverty level) and low-poverty areas (LPA) might have contributed to disparities in their health outcomes. We sought to determine whether certain hospitals predominantly provided surgical care for patients with CRC from HPAs and examine associated patient outcomes. Patients and Methods: We identified patients undergoing surgery for nonmetastatic CRC diagnosed during 1/1/2009–12/31/2019 from SEER-Medicare. We defined poverty-area-serving (PAS) hospitals as hospitals with ≥ 50% patients from HPAs. We compared in-hospital adverse events, 30 day readmission, and long-term mortality between patients from HPAs and LPAs treated at PAS and non-PAS hospitals using logistic and Cox regression. Results: Our cohort included 81,992 patients with CRC (median age = 78 years, 53.8% female, 15.9% in HPAs) treated by 991 hospitals. The 180 (18.2%) PAS hospitals treated 64.2% of patients from HPAs versus 2.6% from LPAs. Compared with patients from LPAs treated at non-PAS hospitals, patients from HPAs treated at PAS hospitals had more frequent in-hospital adverse events (OR[95%CI] = 1.17[1.07–1.29]), 30-day readmission (OR[95%CI] = 1.33[1.20–1.47]), worse all-cause (HR[95%CI] = 1.16[1.10–1.22]), and cancer-specific mortality (HR[95%CI] = 1.23[1.15–1.32]). Conclusions: A group of PAS hospitals treated a significant proportion of patients with CRC from HPAs and few from LPAs and was associated with worse short- and long-term patient outcomes. These findings highlight the presence and negative impact of healthcare segregation by area-level poverty and systemic inequities faced by individuals from HPAs. Multilevel resources are needed to address quality of care and other healthcare-associated needs for individuals from disadvantaged areas.

BACKGROUND: The accuracy of contemporary administrative claims codes to discriminate between different phenotypes of peripheral artery disease is not well defined. We aimed to validate a predefined set of International Classification of Diseases, Tenth Revision , codes used to distinguish between claudication and chronic limb-threatening ischemia (CLTI) and to optimize their diagnostic accuracy using a supervised machine-learning approach. METHODS: We included all patients who underwent a peripheral vascular intervention for claudication or CLTI in the US Medicare-matched VQI-VISION (Vascular Quality Initiative Vascular Implant Surveillance and Interventional Outcomes Network) registry database between January 2016 and December 2019. Gold standard claudication and CLTI diagnoses were determined using VQI (Vascular Quality Initiative) registry data. These diagnoses were compared with a predetermined set of International Classification of Diseases, Tenth Revision , codes in the Medicare-matched data set. We used traditional logistic regression modeling and 6 machine-learning models to distinguish claudication from CLTI. We evaluated the sensitivity, specificity, total agreement, and area under the curve for all models, implementing grid search cross-validation to boost machine-learning model performance. RESULTS: Of 54 180 patients who underwent a peripheral vascular intervention (mean age, 71.9±10.0 years; 41.0% female; 74.2 non-Hispanic White), 20 769 (38.3%) had claudication and 33 411 (61.7%) had CLTI per gold standard registry definitions. The predefined set of International Classification of Diseases, Tenth Revision , codes had high sensitivity (80.9%), specificity (81.9%), and total agreement (81.3%) for distinguishing claudication versus CLTI. Traditional logistic regression improved sensitivity to 96.2%, but with a substantial drop in specificity (41.8%) and an area under the curve of 0.785. Of the machine-learning models, gradient boosting classifier performed the best (area under the curve, 0.892), improving sensitivity to 88.6% and total agreement to 84.2% with minimal drop in specificity (77.1%). CONCLUSIONS: International Classification of Diseases, Tenth Revision , codes can be used to discriminate between claudication and CLTI in claims data. Our defined set of claims codes can be used by investigators to accurately distinguish between these 2 peripheral artery disease phenotypes.

Objectives: To determine whether neural network models based on electronic health record (EHR) data can match and augment the performance of models based on clinical registry data in predicting readmission after peripheral vascular intervention (PVI). Design: Observational cohort study. Setting: Vascular Quality Initiative registry and INSIGHT Clinical Research Network EHR data from multiple academic institutions in New York City. Participants: Patients undergoing PVI during January 1, 2013 to September 30, 2021. Main outcome measures: Our outcome variable was 90-day readmission. We developed logistic regression (LR), multilevel perceptron (MLP), and recurrent neural network (RNN) models using registry alone, EHR data alone, and combined registry-EHR data. EHR data were evaluated using derived variables to match registry variables (EHR-derived data) and clinically meaningful code aggregation (EHR-direct data). Models were evaluated using area under the curve (AUC) for discrimination, Spiegelhalter z score for calibration, and Brier score for overall performance. Results: The analytical cohort included 2348 patients undergoing PVI (mean age: 69.9±11.5 years). 832 (35%) patients were readmitted within 90 days. LR to predict 90-day readmission based on registry data alone had an AUC of 0.710, Spiegelhalter z score of 1.021, and Brier score of 0.211. MLP based on registry data alone had similar performance. MLP and RNN based on EHR-direct data (MLP: AUC=0.742, Spiegelhalter z=0.933, Brier=0.204; RNN: AUC=0.737, Spiegelhalter z=1.026, Brier=0.206) and registry+EHR-direct data (MLP: AUC=0.756, Spiegelhalter z=0.794, Brier=0.199; RNN: AUC=0.751, Spiegelhalter z=1.057, Brier=0.200) had improved performances. LR based on EHR-direct data and combined registry+EHR-direct data had worse performances. Conclusions: EHR data, when used with neural network models, can be useful to establish readmission predictive models or augment clinical registry data. EHR-based models can be potentially embedded in the clinical workflow, but model performance may be constrained by the absence of certain information in clinical encounters, such as social determinants of health.

Changes in screening guidelines have influenced stage at diagnosis for prostate cancer, but it remains unclear whether these trends differ by neighborhood socioeconomic status (SES). Using cancer registry data from the Surveillance, Epidemiology, and End Results program from 2011 to 2020, we estimated age-standardized stage-specific incidence rates and annual percent changes for localized and distant prostate cancer incidence by neighborhood SES quintile and age group. Incidence of localized prostate cancer was highest in higher neighborhood SES areas, while distant prostate cancer rates were highest in areas with lowest neighborhood SES. Annual percent changes in localized prostate cancer incidence were similar by neighborhood SES over the decade, whereas the differences in distant prostate cancer incidence by neighborhood SES diminished over this period. The differing trends in localized and distant prostate cancer incidence by neighborhood SES highlight the importance of equitable access to screening among younger high-risk individuals and improved personalized screening strategies among older men based on health status.

Objectives: To evaluate the feasibility for use of electronic health record (EHR) data in conducting adverse event surveillance among women who received mid-urethral slings (MUS) to treat stress urinary incontinence (SUI) in five health systems. Design: Retrospective observational study using EHR data from 2010 through 2021. Women with a history of MUS were identified using common data models; a common analytic code was executed at each site. A manual chart review was conducted in a per-site random patient subset to establish a reference standard. Automated text processing (Text Processed Integrated (TPI)) was developed and evaluated at each site to determine the surgical approach and synthetic mesh implantation. Patients were characterized and surgical outcomes were ascertained over 730 subsequent days. Setting: Five large tertiary care academic medical centers. Participants: Across five health systems, 9,906 eligible patients (mean age 57–60 per site) were identified. Main outcome measures: Determination of surgical approach, synthetic mesh implantation, and assessment of the duration of surveillance for mortality and reoperation rates following MUS implantation. Results: In the TPI cohort analysis, 3,331 patients were identified. Surgical approach per site was retropubic (42% to 77%), transobturator (6% to 44%), single incision (0% to 24%), and adjustable sling (0% to <4%). Concordance rates for TPI using chart review were 71%–90% at each site for the surgical approach and 28%–85% for synthetic mesh implantation. Patient follow-up observation rates for mortality and reoperation ranged from 22% to 36% at 90 days, 15% to 30% at 365 days, and 8% to 19% at 730 days. Conclusion: Using EHR data alone, identification of medical devices and surgical approaches was feasible among women with MUS surgery for SUI, but long-term follow-up ascertainment rates were low. Medical device surveillance using EHR data should be evaluated in the context of the clinical use case, as applicability may vary.

Objectives: Traditional methods for medical device post-market surveillance often fail to accurately account for operator learning effects, leading to biased assessments of device safety. These methods struggle with non-linearity, complex learning curves, and time-varying covariates, such as physician experience. To address these limitations, we sought to develop a machine learning (ML) framework to detect and adjust for operator learning effects. Materials and Methods: A gradient-boosted decision tree ML method was used to analyze synthetic datasets that replicate the complexity of clinical scenarios involving high-risk medical devices. We designed this process to detect learning effects using a risk-adjusted cumulative sum method, quantify the excess adverse event rate attributable to operator inexperience, and adjust for these alongside patient factors in evaluating device safety signals. To maintain integrity, we employed blinding between data generation and analysis teams. Synthetic data used underlying distributions and patient feature correlations based on clinical data from the Department of Veterans Affairs between 2005 and 2012. We generated 2494 synthetic datasets with widely varying characteristics including number of patient features, operators and institutions, and the operator learning form. Each dataset contained a hypothetical study device, Device B, and a reference device, Device A. We evaluated accuracy in identifying learning effects and identifying and estimating the strength of the device safety signal. Our approach also evaluated different clinically relevant thresholds for safety signal detection. Results: Our framework accurately identified the presence or absence of learning effects in 93.6% of datasets and correctly determined device safety signals in 93.4% of cases. The estimated device odds ratios’ 95% confidence intervals were accurately aligned with the specified ratios in 94.7% of datasets. In contrast, a comparative model excluding operator learning effects significantly underperformed in detecting device signals and in accuracy. Notably, our framework achieved 100% specificity for clinically relevant safety signal thresholds, although sensitivity varied with the threshold applied. Discussion: A machine learning framework, tailored for the complexities of post-market device evaluation, may provide superior performance compared to standard parametric techniques when operator learning is present. Conclusion: Demonstrating the capacity of ML to overcome complex evaluative challenges, our framework addresses the limitations of traditional statistical methods in current post-market surveillance processes. By offering a reliable means to detect and adjust for learning effects, it may significantly improve medical device safety evaluation.

Objectives To assess the feasibility of assessing long-term outcomes of peripheral vascular intervention (PVI) by linking data from a clinical registry to electronic health records (EHR) data from a clinical research network. Design Observational cohort study. Setting Vascular Quality Initiative registry linked to INSIGHT Clinical Research Network, which aggregated EHR data from multiple institutions in New York City. Participants Patients receiving PVI during 1 January 2013–30 November 2021 in four centres in New York City. Primary and secondary outcome measures We examined the proportion of registry patients retained in EHR over time and predictors of EHR retention after year 1. We evaluated the implications of EHR attrition by examining amputation-free survival (AFS) in the observed data and predicted data when patients discontinued in the EHR were hypothesised to have increased risks of events than the observed average. Results We included 1405 patients receiving PVI (age=70.8±11.2 years, 51.3% male). Among eligible patients, 75.2% were retained in EHR through year 3. Patients who aged 75 years or above (vs Conclusions Studies using registry-EHR data may benefit from the timeliness of the data but may be most appropriate to focus on short-term to intermediate-term outcomes of interventions and devices. Future research is needed to investigate the value of registry-EHR linkage in facilitating short-term to intermediate-term outcome assessment following vascular interventions and advanced statistical approaches to account for non-random missing long-term data.