Publications by authors named "Hripcsak G"
Objective: Propose a framework to empirically evaluate and report validity of findings from observational studies using pre-specified objective diagnostics, increasing trust in real-world evidence (RWE).
Materials And Methods: The framework employs objective diagnostic measures to assess the appropriateness of study designs, analytic assumptions, and threats to validity in generating reliable evidence addressing causal questions. Diagnostic evaluations should be interpreted before the unblinding of study results or, alternatively, only unblind results from analyses that pass pre-specified thresholds.
Using administrative claims and electronic health records for observational studies is common but challenging due to data limitations. Researchers rely on phenotype algorithms, requiring labor-intensive chart reviews for validation. This study investigates whether case adjudication using the previously introduced Knowledge-Enhanced Electronic Profile Review (KEEPER) system with large language models (LLMs) is feasible and could serve as a viable alternative to manual chart review.
View Article and Find Full Text PDFThe primary practice of healthcare artificial intelligence (AI) starts with model development, often using state-of-the-art AI, retrospectively evaluated using metrics lifted from the AI literature like AUROC and DICE score. However, good performance on these metrics may not translate to improved clinical outcomes. Instead, we argue for a better development pipeline constructed by working backward from the end goal of positively impacting clinically relevant outcomes using AI, leading to considerations of causality in model development and validation, and subsequently a better development pipeline.
View Article and Find Full Text PDFArticle Synopsis
- This study compares the cardiovascular effectiveness of different second-line antihyperglycemic agents (SGLT2 inhibitors, GLP-1 receptor agonists, DPP-4 inhibitors, and sulfonylureas) in patients with type 2 diabetes and cardiovascular disease.
- Using data from over 1.4 million patients across multiple databases, the researchers analyzed the risk of major adverse cardiovascular events (MACE) over a follow-up period of several years.
- Results indicated that SGLT2 inhibitors and GLP-1 receptor agonists had significantly lower risks of MACE compared to DPP-4 inhibitors and sulfonylureas, pointing to their potential superiority as treatment options for
Article Synopsis
- The study explores using patient portal messaging to recruit underrepresented individuals for the All of Us Research Program, aiming to improve diversity in biomedical research.
- A large-scale outreach was conducted at Columbia University, where over 59,000 patients were messaged, resulting in a 15.1% response rate and showing varying engagement levels among different racial and ethnic groups.
- While the method increased outreach efficiency, the researchers found that underrepresented groups struggled more with initial consent and message engagement, highlighting the need for targeted strategies to enhance participation.
Objective: Healthcare continues to grapple with the persistent issue of treatment disparities, sparking concerns regarding the equitable allocation of treatments in clinical practice. While various fairness metrics have emerged to assess fairness in decision-making processes, a growing focus has been on causality-based fairness concepts due to their capacity to mitigate confounding effects and reason about bias. However, the application of causal fairness notions in evaluating the fairness of clinical decision-making with electronic health record (EHR) data remains an understudied domain.
View Article and Find Full Text PDFPropensity score adjustment addresses confounding by balancing covariates in subject treatment groups through matching, stratification, inverse probability weighting, etc. Diagnostics ensure that the adjustment has been effective. A common technique is to check whether the standardized mean difference for each relevant covariate is less than a threshold like 0.
View Article and Find Full Text PDFArticle Synopsis
- Translational research needs data from different levels of biological systems, but combining that data is tough for scientists.
- New technologies help gather more data, but researchers struggle to organize all the information effectively.
- PheKnowLator is a tool that helps scientists create customizable knowledge graphs easily, making it better for managing complex health information without slowing down their work.
Article Synopsis
- Patients with drug-resistant epilepsy (DRE) need thorough neurodiagnostic evaluations, but there are significant delays in referrals and underutilization of surgery, particularly in diverse US settings.
- This study seeks to analyze the rates and factors influencing neurodiagnostic evaluations for DRE patients across three different US cohorts using extensive medical data.
- The findings reveal low rates of comprehensive evaluations among DRE patients, with only about 4.5% in the Medicaid cohort, 8.0% in the commercial insurance cohort, and 14.3% at Columbia University Medical Center.
Article Synopsis
- Ventilator dyssynchrony (VD) can increase lung injury, and detecting its variability is complex, but machine learning offers potential solutions for automating detection in ventilator waveform data.
- A systematic framework was developed to quantify features in ventilator signals, which allows for stratifying the severity of dyssynchronous breaths.
- The study analyzed over 93,000 breaths, achieving a predictive accuracy of over 97% for identifying flow-limited VD breaths, and established a computational approach for understanding the severity and impact of VD in clinical settings.
Article Synopsis
- The study aimed to evaluate how often kidney failure occurs in patients receiving intravitreal anti-VEGF treatments and to compare the risks associated with three specific drugs: ranibizumab, aflibercept, and bevacizumab.
- Researchers conducted a retrospective cohort study, analyzing data from 12 databases within the OHDSI network, focusing on patients over 18 with retinal diseases receiving these treatments.
- Results showed an average incidence of kidney failure of 678 per 100,000 persons, and no significant differences in risk were found among the three anti-VEGF drugs, indicating similar safety profiles regarding kidney health.
Article Synopsis
- Previous research showed better safety and effectiveness outcomes for thiazides compared to ACE inhibitors (ACEi), but allowed addition of a second medication after a week.
- In this study, the definition of monotherapy was changed, requiring participants to exit if they started another antihypertensive medication.
- The results demonstrated that the significant differences in effectiveness were no longer present, although thiazides still had a more favorable safety profile.
Article Synopsis
- * The study analyzed data from over 1.4 million patients treated with various second-line diabetes medications, using advanced statistical methods to compare outcomes and risks of heart issues.
- * Findings indicated that both SGLT2 inhibitors and GLP-1 receptor agonists reduce the risk of cardiovascular events compared to DPP-4 inhibitors and sulfonylureas, but no significant differences were found between SGLT2is and GLP1-RAs themselves regarding heart risks.
Article Synopsis
- Data-driven clinical prediction algorithms help clinicians, and understanding the factors that affect their performance and fairness is crucial for equitable healthcare.
- This study focused on creating a prediction algorithm for estimating glomerular filtration rate (GFR) using electronic health records (EHR) and compared three approaches: CKD-EPI equations, epidemiological models, and EHR-based models with various machine learning techniques.
- The findings revealed that while using more complex models and relevant clinical features reduced overall estimation error, the performance gap between African American and non-African American patients persisted, highlighting the need for better identification of objective patient characteristics to enhance algorithm fairness and effectiveness.
Article Synopsis
- The diversity of clinical notes in electronic health records (EHRs) highlights the need for standardization to improve data retrieval and integration, which is where the LOINC Document Ontology (DO) comes in, specifically designed for naming clinical documents.
- This study evaluated the LOINC DO by mapping clinical note titles from five institutions, categorizing them into three classes based on how similar they are to LOINC DO codes, and developed an automated pipeline for this mapping that doesn't require accessing note content.
- The automated mapping system, powered by various language models, demonstrated a high accuracy of 0.90, and the research compared its results with manual mappings to assess LOINC DO's effectiveness and identify opportunities for expanding
Article Synopsis
- OHDSI (Observational Health Data Sciences and Informatics) is a massive distributed data network with over 331 sources and 2.1 billion patient records, facilitating large-scale observational research through standardized data.
- The OHDSI Standardized Vocabularies, a crucial component of this network, include more than 10 million concepts from 136 vocabularies, allowing for better data harmonization and easier research execution.
- This open-source vocabulary system addresses challenges in observational research, enabling various analyses such as efficient phenotyping and patient-level predictions, and encourages researchers to utilize and contribute to its ongoing development.
Article Synopsis
- This study hypothesizes that polygenic factors contribute to the variability seen in these conditions, specifically looking at how genetic scores can aid in predicting CKD risk.
- Results show that individuals with a high polygenic risk score (GPS) who also carry ADPKD variants have significantly increased CKD risk, with similar findings for COL4A-AN carriers, highlighting the importance of polygenic assessment in managing monogenic kidney diseases.
Article Synopsis
- The study focuses on the importance of creating accurate phenotype definitions for reliable safety research, comparing different definitions to see how they affect background incidence rates of adverse events.
- Using data from 16 sources, the researchers analyzed 13 adverse events and discovered significant variations in incidence rates based on how phenotypes were defined, particularly with different modifications like inpatient settings.
- The results indicated that requiring an inpatient setting significantly increased the incidence rates, showing the need to carefully evaluate definitions before using them for background rate assessments in a global context.
Article Synopsis
- Postmarket safety surveillance is crucial for mass vaccination programs, but traditional methods face challenges like multiple testing issues and data biases, leading to the need for improved approaches.
- The researchers developed a Bayesian surveillance method that utilizes negative control outcomes to reduce bias and offers increased flexibility in analyzing vaccine-related adverse events.
- Their empirical evaluation, using data from over 360 million patients, showed that this new method significantly outperformed the existing MaxSPRT approach by reducing Type 1 errors and improving estimation accuracy, with all findings made publicly accessible via an R ShinyApp.
Article Synopsis
- The paper addresses the challenge of creating personalized, high-quality phenotypes based on complex physiological data from electronic health records (EHR), focusing on unmeasured physiological parameters.
- A new methodology is developed that applies advanced calculations to the glucose-insulin system for ICU patients, using data assimilation and optimization to estimate parameters like insulin secretion and resistance.
- The study analyzed 109 ICU patients, generating interpretable phenotypes that reflect individual patient physiology over three-day periods, resulting in multiple discrete phenotypes for each patient during their stay.
Article Synopsis
- Blind and deaf individuals, particularly from marginalized backgrounds, face significant health disparities and are underrepresented in precision medicine research, such as the All of Us Research Program (AoURP).
- Analysis of AoURP data from 2018-2023 reveals that blind and deaf participants, especially working-age and Asian or multi-racial individuals, are notably lacking in representation.
- This underrepresentation, particularly for Black or African American women with low education and income, raises concerns about the validity and applicability of research findings for these underserved communities.
Article Synopsis
- Researchers wanted to find out if using data from electronic health records is better than looking at patient charts for studying health conditions.
- They created a tool called KEEPER that organizes important health information to help understand these conditions faster and more clearly.
- The results showed that KEEPER helps doctors agree on patient diagnoses more often and does it in half the time compared to traditional chart reviews.
Article Synopsis
- - The study investigates factors that contribute to successful participant recruitment in randomized clinical trials (RCTs) by analyzing data from 393 completed treatment studies, focusing on accrual percentages as a measure of success.
- - A comparison of different predictive models revealed that the CatBoost regressor provided the best performance, identifying government funding and participant compensation as positive recruitment factors, while cancer-focused studies and unconventional recruiting methods were less successful.
- - The research concludes by emphasizing the importance of specific recruitment strategies, such as flexible infrastructure and adequate compensation, to enhance participant recruitment in future clinical trials.