서브메뉴
검색
Improving High-Stakes Decision Making with Statistical and Machine Learning Methods
Improving High-Stakes Decision Making with Statistical and Machine Learning Methods
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153055
- ISBN
- 9798346389378
- DDC
- 362.1
- 서명/저자
- Improving High-Stakes Decision Making with Statistical and Machine Learning Methods
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 114 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
- 주기사항
- Advisor: Baiocchi, Mike;Chen, Jonathan.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약Over the past twenty years, emergency department (ED) visits have increased about twice as fast as population growth, and each year, roughly two million ED visits resulted in Intensive Care Unit (ICU) admissions. Everyday, physicians triage patients to determine their acuity levels, the levels of care they need when being admitted to the hospital.These decisions largely depend on human judgment in a high-stakes environment with poor evidence, limited information, and intensive-time pressures. The difficulty of triaging coupled with inherent biases in decision-making highlights the need and opportunity for computer-aided clinical decision support, leveraging electronic health records (EHRs) to manage difficult triage decisions that otherwise place undue pressure on decision-makers with potentially dire consequences.The goal of my research is to develop a framework to support clinical decision making in high-stakes environments with a data-driven approach using statistical and machine learning methods. How can we help physicians to better prioritize patients and decide initial treatments in hospitals?First, I developed machine learning models for initial risk assessments. One model focuses on a specific issue of initial insulin dosing for patients admitted with high blood glucose, whether a patient needs low or higher insulin dose, and how many units, to guide initial interventions. Another model predicts general patient's risk for initial ICU admission.Second, I proposed an inductive and iterative framework for improving model development and assessment, named thick data analytics. Thick data analytics utilizes both quantitative and qualitative approach where experts review discordance cases from related prediction models (e.g. a patient admission with risks predicted at different times within the first 24 hours of admission). The goal is to understand the discordance, improve prediction, and reduce preventable transfers within a short time frame. Thick analytics facilitates model redesign to be more trustworthy in aiding triage.Lastly, I will discuss tie-breaker designs for a pragmatic clinical trial where patients with similar risks could be randomized for admission to ICUs vs non-ICUs. We aim to evaluate the effectiveness of interventions such as ICU admission on patient's health outcomes such as mortality and length of stay. When allowing overriding in such experiments, randomization is an encouragement, which is a form of instrumental variables. Our work will help understand medical decision biases to improve triage process for better patient outcomes.
- 일반주제명
- Health care access
- 일반주제명
- Patients
- 일반주제명
- Emergency medical care
- 일반주제명
- Success
- 일반주제명
- Clinical decision making
- 일반주제명
- Design
- 일반주제명
- Insulin
- 일반주제명
- Intensive care
- 일반주제명
- Hospitalization
- 일반주제명
- Clinical outcomes
- 일반주제명
- Drug dosages
- 일반주제명
- Health sciences
- 일반주제명
- Information science
- 일반주제명
- Pharmaceutical sciences
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 86-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164849
■00520250211153055
■006m o d
■007cr#unu||||||||
■020 ▼a9798346389378
■035 ▼a(MiAaPQ)AAI31643381
■035 ▼a(MiAaPQ)Stanfordsz697sp0293
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a362.1
■1001 ▼aNguyen, Minh Chau Thanh.
■24510▼aImproving High-Stakes Decision Making with Statistical and Machine Learning Methods
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a114 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-05, Section: B.
■500 ▼aAdvisor: Baiocchi, Mike;Chen, Jonathan.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aOver the past twenty years, emergency department (ED) visits have increased about twice as fast as population growth, and each year, roughly two million ED visits resulted in Intensive Care Unit (ICU) admissions. Everyday, physicians triage patients to determine their acuity levels, the levels of care they need when being admitted to the hospital.These decisions largely depend on human judgment in a high-stakes environment with poor evidence, limited information, and intensive-time pressures. The difficulty of triaging coupled with inherent biases in decision-making highlights the need and opportunity for computer-aided clinical decision support, leveraging electronic health records (EHRs) to manage difficult triage decisions that otherwise place undue pressure on decision-makers with potentially dire consequences.The goal of my research is to develop a framework to support clinical decision making in high-stakes environments with a data-driven approach using statistical and machine learning methods. How can we help physicians to better prioritize patients and decide initial treatments in hospitals?First, I developed machine learning models for initial risk assessments. One model focuses on a specific issue of initial insulin dosing for patients admitted with high blood glucose, whether a patient needs low or higher insulin dose, and how many units, to guide initial interventions. Another model predicts general patient's risk for initial ICU admission.Second, I proposed an inductive and iterative framework for improving model development and assessment, named thick data analytics. Thick data analytics utilizes both quantitative and qualitative approach where experts review discordance cases from related prediction models (e.g. a patient admission with risks predicted at different times within the first 24 hours of admission). The goal is to understand the discordance, improve prediction, and reduce preventable transfers within a short time frame. Thick analytics facilitates model redesign to be more trustworthy in aiding triage.Lastly, I will discuss tie-breaker designs for a pragmatic clinical trial where patients with similar risks could be randomized for admission to ICUs vs non-ICUs. We aim to evaluate the effectiveness of interventions such as ICU admission on patient's health outcomes such as mortality and length of stay. When allowing overriding in such experiments, randomization is an encouragement, which is a form of instrumental variables. Our work will help understand medical decision biases to improve triage process for better patient outcomes.
■590 ▼aSchool code: 0212.
■650 4▼aHealth care access
■650 4▼aPatients
■650 4▼aEmergency medical care
■650 4▼aElectronic health records
■650 4▼aSuccess
■650 4▼aClinical decision making
■650 4▼aDesign
■650 4▼aInsulin
■650 4▼aIntensive care
■650 4▼aHospitalization
■650 4▼aClinical outcomes
■650 4▼aDrug dosages
■650 4▼aHealth sciences
■650 4▼aInformation science
■650 4▼aPharmaceutical sciences
■690 ▼a0389
■690 ▼a0800
■690 ▼a0769
■690 ▼a0566
■690 ▼a0723
■690 ▼a0572
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164849▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


