서브메뉴
검색
Applying Systems Thinking and Machine Learning Techniques to Identify Leverage Points for Intervening in Perioperative Opioid Use and Developing Risk Score Tools to Guide Perioperative Opioid Prescription
Applying Systems Thinking and Machine Learning Techniques to Identify Leverage Points for Intervening in Perioperative Opioid Use and Developing Risk Score Tools to Guide Perioperative Opioid Prescription
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211151149
- ISBN
- 9798382303260
- DDC
- 614.4
- 저자명
- Huang, Yongmei.
- 서명/저자
- Applying Systems Thinking and Machine Learning Techniques to Identify Leverage Points for Intervening in Perioperative Opioid Use and Developing Risk Score Tools to Guide Perioperative Opioid Prescription
- 발행사항
- [Sl] : Columbia University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 199 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-10, Section: B.
- 주기사항
- Advisor: Jacobson, Judith S.
- 학위논문주기
- Thesis (Dr.P.H.)--Columbia University, 2024.
- 초록/해제
- 요약Study Background and Objectives:Excessive perioperative opioid prescribing has been detrimental to public health, contributing to the elevated prevalence of opioid use disorder. Since 2016, rigorous regulation of opioid prescribing has reduced over-prescription, but has also led to opioid-phobia. The 2022 CDC guideline promotes person-centered decisions on pain management by relaxing restrictions on opioid prescription. The determination of opioid requirements for surgical pain management is influenced by various factors and stakeholders. Despite extensive research, the mechanisms underlying perioperative pain management and the persistence of opioid use after surgery remain unclear. Clinicians currently lack tools to guide opioid prescription in clinical settings, and patients often face a dearth of information regarding expected pain levels, proper opioid use, and options for surgical pain management. The main objective of my doctoral project is to disentangle the intricate relationships among patients, healthcare providers, and policy changes in perioperative opioid prescription for pain management and to identify key intervention points to balance the beneficial effects of proper opioids use against the risks of addition. Another objective is to develop a risk score algorithm for perioperative opioid requirements to help with decision-making in clinical practice.Materials and Methods:In chapter 1, I undertook a systematic review and meta-analysis, and investigated the percentage of adult patients scheduled for general surgeries who received opioid analgesia for perioperative pain management, the quantities of opioids prescribed to patients, the actual quantities consumed, the percentage of patients without prior opioid exposure experiencing prolonged opioid use, and the evolution of perioperative opioid prescription patterns since the policy changes. A causal loop diagram was used to visualize the complex conceptual framework of perioperative pain management and post-surgical prolonged use of opioids based on insights derived from the systematic review and meta-analysis. In chapter 2 and 3, data from patients aged 18-64 years undergoing one of 12 commonly performed procedures (e.g., laparoscopic cholecystectomy) from 2015 to 2018 at a single institution were analyzed. Perioperative opioid requirements (none/low, medium, high) were determined based on patients' self-reported pain scores and opioid prescription/administration from 30 days before to 2 weeks after surgery. Patients' clinical and procedure-related factors were collected as potential predictors. Random forest, the Least Absolute Shrinkage and Selection Operator (LASSO), and multinomial logistic regression were used to develop prediction models. Models' performance, including discrimination, calibration, classification measures were evaluated. A nomogram based on multinomial logistic regression was generated as a score tool, and decision curve analysis was used to examine the clinical utility of the final prediction model dichotomizing the opioid prescription as none/sparing versus medium/high requirements.Results: My systematic review and meta-analysis revealed that around 85% of surgical patients received opioids perioperatively. The pooled mean total amount of opioids dispensed was 210 MME per patient per surgical procedure. Notably, only approximately 44% of the prescribed opioids were consumed. Among opioid-naive patients who initiated opioid use perioperatively, 7.1% persisted in opioid use beyond the conventional three-month postoperative recovery timeframe. Intervention programs (such as setting up maximum limits of opioids prescription, providing trainings to health providers, monitoring opioids prescription behaviors, providing health education to patients, etcetera) reduced perioperative opioid prescription by 38% and opioid consumption by 63.2%. The causal loop diagram illustrates a balancing feedback loop between policy and over-prescription, highlighting the pivotal role of a decision tool in reducing the over-prescription of perioperative opioids while ensuring the fulfillment of opioid needs for effective perioperative pain management.To develop a decision-aid tool based on prediction models, I included 2733 patients in the training dataset and 1081 in the testing dataset, all of whom underwent general surgeries. All prediction models demonstrated moderate discrimination in the testing dataset. The null hypothesis of perfect calibration intercepts and calibration slopes was rejected. In analyses restricted to patients undergoing laparoscopic cholecystectomy, model discrimination remained similar while model calibration improved. The revised LASSO model had an accuracy of around 65% in the testing dataset, classifying future cases correctly into opioid requirements groups in laparoscopic cholecystectomy cohort. Features in the final laparoscopic cholecystectomy model included the use of opioid/NSAID/anti-depressant before surgery, emergency surgery, anesthesia type, and surgical indication for cholelithiasis/cholecystitis. A nomogram was created to guide perioperative opioids use among laparoscopic cholecystectomy patients, and the decision curve analysis demonstrated the clinical utility of the prediction model; it generated higher net benefits than the strategy of prescribing no opioids or opioid sparing to surgical patients and the strategy of prescribing medium or high opioids doses to all patients, with a broad threshold probability from 18% to 92%.Conclusions:In summary, this dissertation described the historically high levels of perioperative opioid prescriptions and highlighted their adverse impacts: persistent opioid use and community diversion. Although the implementation of guidance and policies has significantly reduced nationwide over-prescriptions of opioids, it is essential to recognize the potential benefits of appropriate opioid use in perioperative pain management. The incorporation of a machine-learning approach with subject-matter knowledge may achieve more accurate predictions of opioid requirements than employing machine-learning techniques alone and increase the interpretability of the prediction model. Notably, the surgery-specific model demonstrated superior performance than the model for general surgeries. Future studies should further validate the conceptual model of perioperative opioid prescription and misuse in real-world scenarios, enhance model discrimination, extend external validation efforts, and develop electronic applications tailored to contemporary medical practices.
- 일반주제명
- Epidemiology
- 일반주제명
- Public health
- 일반주제명
- Surgery
- 일반주제명
- Health sciences
- 키워드
- Machine-learning
- 키워드
- Opioid use
- 키워드
- Prediction model
- 키워드
- Risk score tool
- 키워드
- System thinking
- 기타저자
- Columbia University Epidemiology
- 기본자료저록
- Dissertations Abstracts International. 85-10B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017161011
■00520250211151149
■006m o d
■007cr#unu||||||||
■020 ▼a9798382303260
■035 ▼a(MiAaPQ)AAI31235335
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a614.4
■1001 ▼aHuang, Yongmei.
■24510▼aApplying Systems Thinking and Machine Learning Techniques to Identify Leverage Points for Intervening in Perioperative Opioid Use and Developing Risk Score Tools to Guide Perioperative Opioid Prescription
■260 ▼a[Sl]▼bColumbia University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a199 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-10, Section: B.
■500 ▼aAdvisor: Jacobson, Judith S.
■5021 ▼aThesis (Dr.P.H.)--Columbia University, 2024.
■520 ▼aStudy Background and Objectives:Excessive perioperative opioid prescribing has been detrimental to public health, contributing to the elevated prevalence of opioid use disorder. Since 2016, rigorous regulation of opioid prescribing has reduced over-prescription, but has also led to opioid-phobia. The 2022 CDC guideline promotes person-centered decisions on pain management by relaxing restrictions on opioid prescription. The determination of opioid requirements for surgical pain management is influenced by various factors and stakeholders. Despite extensive research, the mechanisms underlying perioperative pain management and the persistence of opioid use after surgery remain unclear. Clinicians currently lack tools to guide opioid prescription in clinical settings, and patients often face a dearth of information regarding expected pain levels, proper opioid use, and options for surgical pain management. The main objective of my doctoral project is to disentangle the intricate relationships among patients, healthcare providers, and policy changes in perioperative opioid prescription for pain management and to identify key intervention points to balance the beneficial effects of proper opioids use against the risks of addition. Another objective is to develop a risk score algorithm for perioperative opioid requirements to help with decision-making in clinical practice.Materials and Methods:In chapter 1, I undertook a systematic review and meta-analysis, and investigated the percentage of adult patients scheduled for general surgeries who received opioid analgesia for perioperative pain management, the quantities of opioids prescribed to patients, the actual quantities consumed, the percentage of patients without prior opioid exposure experiencing prolonged opioid use, and the evolution of perioperative opioid prescription patterns since the policy changes. A causal loop diagram was used to visualize the complex conceptual framework of perioperative pain management and post-surgical prolonged use of opioids based on insights derived from the systematic review and meta-analysis. In chapter 2 and 3, data from patients aged 18-64 years undergoing one of 12 commonly performed procedures (e.g., laparoscopic cholecystectomy) from 2015 to 2018 at a single institution were analyzed. Perioperative opioid requirements (none/low, medium, high) were determined based on patients' self-reported pain scores and opioid prescription/administration from 30 days before to 2 weeks after surgery. Patients' clinical and procedure-related factors were collected as potential predictors. Random forest, the Least Absolute Shrinkage and Selection Operator (LASSO), and multinomial logistic regression were used to develop prediction models. Models' performance, including discrimination, calibration, classification measures were evaluated. A nomogram based on multinomial logistic regression was generated as a score tool, and decision curve analysis was used to examine the clinical utility of the final prediction model dichotomizing the opioid prescription as none/sparing versus medium/high requirements.Results: My systematic review and meta-analysis revealed that around 85% of surgical patients received opioids perioperatively. The pooled mean total amount of opioids dispensed was 210 MME per patient per surgical procedure. Notably, only approximately 44% of the prescribed opioids were consumed. Among opioid-naive patients who initiated opioid use perioperatively, 7.1% persisted in opioid use beyond the conventional three-month postoperative recovery timeframe. Intervention programs (such as setting up maximum limits of opioids prescription, providing trainings to health providers, monitoring opioids prescription behaviors, providing health education to patients, etcetera) reduced perioperative opioid prescription by 38% and opioid consumption by 63.2%. The causal loop diagram illustrates a balancing feedback loop between policy and over-prescription, highlighting the pivotal role of a decision tool in reducing the over-prescription of perioperative opioids while ensuring the fulfillment of opioid needs for effective perioperative pain management.To develop a decision-aid tool based on prediction models, I included 2733 patients in the training dataset and 1081 in the testing dataset, all of whom underwent general surgeries. All prediction models demonstrated moderate discrimination in the testing dataset. The null hypothesis of perfect calibration intercepts and calibration slopes was rejected. In analyses restricted to patients undergoing laparoscopic cholecystectomy, model discrimination remained similar while model calibration improved. The revised LASSO model had an accuracy of around 65% in the testing dataset, classifying future cases correctly into opioid requirements groups in laparoscopic cholecystectomy cohort. Features in the final laparoscopic cholecystectomy model included the use of opioid/NSAID/anti-depressant before surgery, emergency surgery, anesthesia type, and surgical indication for cholelithiasis/cholecystitis. A nomogram was created to guide perioperative opioids use among laparoscopic cholecystectomy patients, and the decision curve analysis demonstrated the clinical utility of the prediction model; it generated higher net benefits than the strategy of prescribing no opioids or opioid sparing to surgical patients and the strategy of prescribing medium or high opioids doses to all patients, with a broad threshold probability from 18% to 92%.Conclusions:In summary, this dissertation described the historically high levels of perioperative opioid prescriptions and highlighted their adverse impacts: persistent opioid use and community diversion. Although the implementation of guidance and policies has significantly reduced nationwide over-prescriptions of opioids, it is essential to recognize the potential benefits of appropriate opioid use in perioperative pain management. The incorporation of a machine-learning approach with subject-matter knowledge may achieve more accurate predictions of opioid requirements than employing machine-learning techniques alone and increase the interpretability of the prediction model. Notably, the surgery-specific model demonstrated superior performance than the model for general surgeries. Future studies should further validate the conceptual model of perioperative opioid prescription and misuse in real-world scenarios, enhance model discrimination, extend external validation efforts, and develop electronic applications tailored to contemporary medical practices.
■590 ▼aSchool code: 0054.
■650 4▼aEpidemiology
■650 4▼aPublic health
■650 4▼aSurgery
■650 4▼aHealth sciences
■653 ▼aMachine-learning
■653 ▼aOpioid use
■653 ▼aPerioperative pain management
■653 ▼aPrediction model
■653 ▼aRisk score tool
■653 ▼aSystem thinking
■690 ▼a0766
■690 ▼a0573
■690 ▼a0576
■690 ▼a0566
■71020▼aColumbia University▼bEpidemiology.
■7730 ▼tDissertations Abstracts International▼g85-10B.
■790 ▼a0054
■791 ▼aDr.P.H.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17161011▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
