Hanie Eskandari

박찬영

김홍범

최수호

DAHS Lab's research addresses problems at the interface of artificial intelligence and clinical practice, particularly focusing on precision medicine and data-driven decision support. The overarching theme of our lab is "Beyond Data: Actionable Health AI". We aim to infuse technology to support personalized, predictive, and proactive decision makings. By partnering with hospitals and healthcare providers, we focus on actionable health AI with high impact and practicality.

PULSE-ICU: a Pretrained Unified Long-Sequence Encoder for multi-task prediction in Intensive Care Units

Frictional Q-Learning

Mind the Missing: Variable-Aware Representation Learning for Irregular EHR Time Series using Large Language Models

Machine learning-driven subphenotyping of septic shock for predicting mortality and organ dysfunction

Use of machine learning to predict peak cardiorespiratory fitness in patients with cardiovascular disease

Safety-Aware Explainable Deep Reinforcement Learning for Nephrotoxic Medication Management in	Critical Care

Advancing continuous prediction for acute kidney injury via multi-task learning: towards better clinical applicability

CLEAR-Shock: Contrastive LEARning for Shock

Development of continuous warning system for timely identification of septic shock

Medication Recommendation for Parkinson’s Disease based on Dynamics of Symptom Progression

Sequential anomaly detection for continuous prediction of unexpected ICU admission from emergency department

Multimodal reinforcement learning for embedding networks and medication recommendation in parkinson's disease

Multivariate time-series cluster analysis for multiple functional domains to	 identify recovery patterns of patients with fragility hip fracture after surgery

Drug-drug interaction and acute kidney injury development: A correlation-based network analysis

A decision support system on postoperative care management of total joint replacement patients using clinical outcome measurement

Optimal intervention policies for TJR postoperative care process

A prediction and interpretation framework of acute kidney injury in critical care

Assessing the global burden of hemorrhage: The global blood supply, deficits, and potential solutions

Workforce allocation in motorcycle transmission assembly lines: a case study on modeling, analysis, and improvement

Reducing COPD readmissions through predictive modeling and incentive-based interventions

An analytical framework for TJR readmission prediction and cost-effective intervention

Workload balancing: staffing ratio analysis for primary care redesign

Bottleneck Analysis to Improve Multidisciplinary Rounding Process in Intensive Care Units at Mayo Clinic

Joint visit in primary care clinics: Modeling, analysis, and an application study

A queueing network model for analysis of patient transitions within hospitals

Pragmatic trial of a multidisciplinary lung cancer care model in a community healthcare setting: study design, implementation evaluation, and baseline clinical results

Reducing Bottlenecks to Improve the Efficiency of the Lung Cancer Care Delivery Process: A Process Engineering Modeling Approach to Patient-Centered Care

Modeling and analysis of postoperative intervention process for total joint replacement patients using simulations

Incentive-based optimal intervention policy to reduce hospital readmissions for COPD patients

Modeling and analysis of patient transitions in community hospitals: A systems approach

From production systems to health care delivery systems: a retrospective look on similarities, difficulties and opportunities

A system-theoretic method for modeling, analysis, and improvement of lung cancer diagnosis-to-surgery process

An iterative method for analysis of joint visit model at Dean east clinic

A markov chain model to evaluate patient transitions in small community hospitals

Staffing ratio analysis in primary care redesign: a simulation approach

Computer modeling of lung cancer diagnosis-to-treatment process

Assessing the global burden of hemorrhage: the global blood supply, deficits, and potential solutions

Multimodal reinforcement learning for embedding networks and medication recommendation in Parkinson’s disease

Workforce allocation in motorcycle transmission assembly lines: A case study on modeling, analysis, and improvement

Reducing bottlenecks to improve the efficiency of the lung cancer care delivery process: a process engineering modeling approach to patient-centered care

Bottleneck analysis to improve multidisciplinary rounding process in intensive care units at Mayo Clinic

Wolf: Wide-scope large language model framework for CXR understanding

Optimal intervention policies for total joint replacement postoperative care process

Jingshan Li

Xiang Zhong

Feng Ju

Cong Zhao

Yue Dong, MD, FSSH

Hyo Kyung Lee

There is a critical shortage of blood available for transfusion in many low- and middle-income countries. The consequences of this scarcity are dire, resulting in uncounted morbidity and mortality from trauma, obstetric hemorrhage, and pediatric anemias, among numerous other conditions. The process of collecting blood from a donor to administering it to a patient involves many facets from donor availability to blood processing to blood delivery. Each step faces particular challenges in low- and middle-income countries. Optimizing existing strategies and introducing new approaches will be imperative to ensure a safe and sufficient blood supply worldwide.

Acute kidney injury (AKI) is a common clinical condition with high mortality and resource consumption. Early identification of high-risk patients to achieve an appropriate allocation of limited clinical resources and timely interventions is of significant importance, which has attracted substantial research to develop prediction models for AKI risk stratification. However, most available AKI prediction models have moderate performance and lack of interpretability, which limits their applicability in supporting care intervention. In this paper, a machine learning-based framework for AKI prediction and interpretation in critical care is presented. First, an ensemble model is developed to predict a patient’s risk of AKI within 72 h of admission to the intensive care units. Next, the model is interpreted both globally and locally. For the global interpretation, the important predictors are pinpointed and the detailed relationships between …

This paper introduces an analytical framework for assessing the cost-effectiveness of intervention strategies to reduce total joint replacement (TJR) readmissions. In such a framework, a machine learning-based readmission risk prediction model is developed to predict an individual TJR patient's risk of hospital readmission within 90 days post-discharge. Specifically, through data sampling and boosting techniques, we overcome the class imbalance problem by iteratively building an ensemble of models. Then, utilizing the results of the predictive model, and by taking into account the imbalanced misclassification costs between readmitted and nonreadmitted patients, a cost analysis framework is introduced to support decision making in selecting cost-effective intervention policies. Finally, using this framework, a case study at a community hospital is presented to demonstrate the applicability of the analysis.

Manufacturing systems have attracted substantial research attentions during the last 50 years. In recent years, there has been growing interest in health care systems research to improve efficiency, safety and care quality. The similarities identified between manufacturing systems and health care delivery systems heighten the importance of transferring the experience and knowledge in manufacturing to health care. In this paper, based on the lessons we learned and the experience we obtained during our journey from production systems research to health care delivery systems study, we discuss the similarities between production systems and health care delivery systems in system modelling, design, performance evaluation and continuous improvements and investigate the differences and difficulties that stem from variability, constraints, dynamics and human behaviour. Building upon these, the opportunities …

This paper introduces a case study at a community hospital to develop a predictive model to quantify readmission risks for patients with chronic obstructive pulmonary disease (COPD), and use it to support decision making for appropriate incentive-based interventions. Data collected from the community hospital’s database are analyzed to identify risk factors and a logistic regression model is developed to predict the readmission risk within 30 days post-discharge of an individual COPD patient. By targeting on the high-risk patients, we investigate the implementability of the incentive policy which encourages patients to take interventions and helps them to overcome the compliance barrier. Specifically, the conditions and scenarios are identified for either achieving the desired readmission rate while minimizing the total cost, or reaching the lowest readmission rate under incentive budget constraint. Currently, such …

We introduce an example of a rigorous, quantitative method for quality improvement in lung cancer care-delivery. Computer process modeling methods are introduced for lung cancer diagnosis, staging and treatment selection process. Two types of process modeling techniques, discrete event simulation (DES) and analytical models, are briefly reviewed. Recent developments in DES are outlined and the necessary data and procedures to develop a DES model for lung cancer diagnosis, leading up to surgical treatment process are summarized. The analytical models include both Markov chain model and closed formulas. The Markov chain models with its application in healthcare are introduced and the approach to derive a lung cancer diagnosis process model is presented. Similarly, the procedure to derive closed formulas evaluating the diagnosis process performance is outlined. Finally, the pros and cons of …

Background Drug-induced nephrotoxicity is a relatively common preventable cause of acute kidney injury (AKI), providing early recognition and management. The pharmacokinetics or pharmacodynamics of drug-drug interactions may lead to additive or synergistic toxicity. The influx of new medications or off-label use of medications in the critical care setting can lead to additional nephrotoxicities, often challenging to predict or detect. This study evaluates the patterns of medication utilization, their combinations, and the related associations with AKI. Methods We utilized correlation-based network analysis (CNA) to investigate the relationship between medications or their combinations with AKI in a large cohort of critically ill patients in a tertiary medical center between 2007 and 2018. Pairwise medication-AKI correlation analysis was performed to evaluate drug synergistic or additive effects. To investigate the inherent nephrotoxicity of medications, we further analyzed medications that were not paired with any other medications within 24 hours before or after their administration time (isolated medication analysis). Results Among 147,289 ICU admissions, we identified 244 associations among 1,555 unique medication types. In pairwise analysis, 233 significant correlations were found among 13,150,198 medication pair instances. In isolated medication analysis, ten significant AKI associations were noted. When stratified by eGFR level, substantial differences between eGFR<90 vs. eGFR≥90 patients were observed. This highlights a need to determine eGFR as a risk factor for nephrotoxicity assessment when drug interactions are considered …

Background Responsible for 25% of all US cancer deaths, lung cancer presents complex care-delivery challenges. Adoption of the highly recommended multidisciplinary care model suffers from a dearth of good quality evidence. Leading up to a prospective comparative-effectiveness study of multidisciplinary vs. serial care, we studied the implementation of a rigorously benchmarked multidisciplinary lung cancer clinic. Methods We used a mixed-methods approach to conduct a patient-centered, combined implementation and effectiveness study of a multidisciplinary model of lung cancer care. We established a co-located multidisciplinary clinic to study the implementation of this care-delivery model. We identified and engaged key stakeholders from the onset, used their input to develop the program structure, processes, performance benchmarks, and study endpoints (outcome-related process measures, patient …

Early diagnosis and treatment of lung cancer are of significant importance. In this paper, a system-theoretic method is introduced to analyze the diagnosis-to-treatment process for lung cancer patients who receive surgical resections. The complex care delivery process is decomposed into a collection of serial processes, each consisting of combinations of various tests and procedures. Closed formulas are derived to estimate the mean and coefficient of variation of waiting time during the diagnosis-to-surgery process. Simple indicators based on the data collected on the clinic/hospital floor are derived to identify the bottlenecks, i.e., the waiting times that impede the whole delivery process in the strongest manner. In addition, by approximating waiting times using Gamma distributions, an algorithm is introduced to evaluate the waiting-time performance, i.e., the probability to finish the diagnosis-to-surgery process within …

The objective of this paper is to investigate the staffing composition of chief care providers (e.g., physician (MD)) and support staff (e.g., medical assistant (MA)) under various task assignment settings to achieve the optimal operational efficiency. Specifically, we examine the effects of workload shifting and identify the proper ratio of MDs to MAs to attain an effective and efficient service level. Based on a Markov chain based framework that characterizes care providers’ activities during patients’ primary care clinic visits, analytical investigation and numerical experiments are conducted. The results articulate that the optimal staffing ratio is achieved when the workloads of MDs and MAs are balanced. To validate the findings under generic primary care clinic settings, discrete event simulation models are developed and extensive experiments are carried out. The sensitivity study elucidates that the balanced-workload …

This study presents a comprehensive framework for analyzing Parkinson’s disease (PD) progression and optimizing medication recommendations based on multimodal data integration and reinforcement learning (RL) techniques. The research framework integrates diverse biomarkers from structured clinical assessments, imaging data, and biospecimen information, creating various single, bimodal, and trimodal datasets. Deep learning-based embedding networks are employed to transform high-dimensional data into lower-dimensional representations, capturing essential patterns related to PD progression. The learned representations are evaluated using visualization methods and predictive modeling to differentiate clinically relevant outcomes. Subsequently, medication recommendation models are developed based on the learned representations, assisting clinicians in making informed decisions to manage …

Safe and efficient patient transitions are of critical importance to ensure patient safety and care quality. To study patient transitions, this paper presents a queueing network model-based iteration method to model and analyze transitions between emergency department, intensive care unit, and general ward within a hospital. Routings with feedback flows are considered under general arrival and service processes, and the effects of blocking on performance measures are presented for both the mean and variability. It is shown that the iteration procedure is convergent and leads to acceptable accuracy of estimation, for both small- and large-sized hospitals. In addition, the impacts of bed capacity, admission rate, as well as arrival and service time variabilities are discussed. Such a method provides an efficient way to study patient transitions within hospitals.

This letter presents a case study of workforce allocation in a gear assembly line at a motorcycle powertrain manufacturing plant. Through structural modeling of operators' work flow and part flow, assembly system models with different configurations are developed. Using iterative aggregation techniques, the performances of such systems are analyzed, and the associated workforce allocation strategies are investigated. By identifying and mitigating system bottlenecks, improvement recommendations are obtained. Such a study provides an effective method for operator assignment in assembly systems.

The process of lung cancer care from initial lesion detection to treatment is complex, involving multiple steps, each introducing the potential for substantial delays. Identifying the steps with the greatest delays enables a focused effort to improve the timeliness of care-delivery, without sacrificing quality. We retrospectively reviewed clinical events from initial detection, through histologic diagnosis, radiologic and invasive staging, and medical clearance, to surgery for all patients who had an attempted resection of a suspected lung cancer in a community healthcare system. We used a computer process modeling approach to evaluate delays in care delivery, in order to identify potential ‘bottlenecks’ in waiting time, the reduction of which could produce greater care efficiency. We also conducted ‘what-if’ analyses to predict the relative impact of simulated changes in the care delivery process to determine the most efficient …

A patient's stay at a hospital may encompass various departments or units. Since many critical and complex problems occur at the interfaces of healthcare delivery systems, safe and efficient transitions between the departments within a hospital has significant importance. This paper presents a Markov chain-based model to study patient transitions between emergency department, intensive or critical care unit, and hospital ward in small and medium-sized community hospitals. To make the analytical study tractable, an iteration method is introduced to approximate the system performance during transitions, including direct transferring probabilities without waiting, average patient occupancy in each department, and average patient length of stay. In addition, system properties, such as monotonicity and sensitivity, are analyzed. It is shown that such a method has a high accuracy in performance estimation and can be …

To improve patient flow and reduce provider workload, joint visit has been proposed and implemented in many primary care clinics. In such systems, the provider and medical assistant (MA) visit the patient jointly; while the provider diagnoses and interacts with the patient, the MA handles necessary documentation work. This article introduces Markov chain models of patient flow with joint visits, and investigates the system behavior under different scenarios. To reduce the state space dimension, convergent iterative procedures are proposed. The study is extended to the non-Markovian case by introducing empirical formulas based on the mean and coefficient of variation (CV) of service times. The results have been validated with good accuracy. To illustrate the applicability of the methods, an application study at Dean East Clinic is presented.

In the emergency department, accurately identifying patients who will require intensive care unit admission is crucial for effective care and resource allocation. Existing methods, such as triage judgments based on Emergency Severity Index, may not fully capture the dynamic nature of patient conditions, potentially overlooking critical risks. Therefore, it is essential to detect sequential deterioration in patients with initially underestimated triage scores. This study addresses the challenge of identifying patients who, despite being initially triaged as low-acuity, later require intensive care unit admission due to clinical deterioration. We propose an anomaly detection approach to continuously monitor patients for signs of deterioration. To achieve this, we employed a sequential anomaly detection model integrated with Long Short-Term Memory to capture temporal relationships within the data. Given the limitations of existing …

In a hospital's intensive care unit (ICU), multidisciplinary rounding (MDR) is a combination of care management with various healthcare providers from different clinical expertise meeting together to coordinate patient care, establish daily goals, and determine treatment plans. Such meetings require significant time and resource utilization from the providers. However, despite its significance, the workflow of MDR has not yet been rigorously studied. Using the data collected in ICUs at Mayo Clinic, this letter studies the MDR process by introducing a continuous time Markov chain model to systematically analyze the workflow and provide guidelines for efficiency improvement. In addition to evaluating current MDR process, a bottleneck analysis method is introduced to identify the task or activity whose improvement can lead to the largest gain in system performance. Based on the findings in bottleneck analysis, a potential …

Significant methodological strides have been made toward Chest X-ray (CXR) understanding via modern vision-language models (VLMs), demonstrating impressive Visual Question Answering (VQA) and CXR report generation abilities. However, existing CXR understanding frameworks still possess several procedural caveats. (1) Previous methods solely use CXR reports, which are insufficient for comprehensive Visual Question Answering (VQA), especially when additional health-related data like medication history and prior diagnoses are needed. (2) Previous methods use raw CXR reports, which are often arbitrarily structured. While modern language models can understand various text formats, restructuring reports for clearer, organized anatomy-based information could enhance their usefulness. (3) Current evaluation methods for CXR-VQA primarily emphasize linguistic correctness, lacking the capability to offer nuanced assessments of the generated answers. In this work, to address the aforementioned caveats, we introduce WoLF, a Wide-scope Large Language Model Framework for CXR understanding. To resolve (1), we capture multi-faceted records of patients, which are utilized for accurate diagnoses in real-world clinical scenarios. Specifically, we adopt the Electronic Health Records (EHR) to generate instruction-following data suited for CXR understanding. Regarding (2), we enhance report generation performance by decoupling knowledge in CXR reports based on anatomical structure even within the attention step via masked attention. To address (3), we introduce an AI-evaluation protocol optimized for assessing the …

Problem definition : Successful outcomes of total joint replacement (TJR) depend not only on the surgery but also on the patients’ postsurgical behaviors; inadequate postsurgical performance can lead to complications and/or readmissions. To prevent such events, patients go through postoperative intervention processes in which they receive physical therapy and/or some form of rehabilitative services. Although the costs of providing such interventions and readmission penalties are substantial, no methods exist that enable healthcare professionals to determine the optimal timing and target group of the interventions. Academic/practical relevance : To address these gaps, this paper investigates the decision problem faced by healthcare professionals: When should we provide interventions to which group of patients to minimize the total expense? We formulate the postdischarge intervention process as a finite-horizon …

Data Analytics and Healthcare Systems Lab

교수님 소개

이효경

연구실 소개

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

PULSE-ICU: a Pretrained Unified Long-Sequence Encoder for multi-task prediction in Intensive Care Units

Frictional Q-Learning

Mind the Missing: Variable-Aware Representation Learning for Irregular EHR Time Series using Large Language Models

연구실 보유 특허

대학별 입시끈

우리 연구실 모집

교수님 소개

이효경

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

PULSE-ICU: a Pretrained Unified Long-Sequence Encoder for multi-task prediction in Intensive Care Units

Frictional Q-Learning

Mind the Missing: Variable-Aware Representation Learning for Irregular EHR Time Series using Large Language Models

연구실 보유 특허

최근 3년간 연구실적 높은 연구실

대학별 입시끈

우리 연구실 모집