(The) effectiveness of hypertonic saline and pentoxifylline (HTS-PTX) resuscitation in hemorrhagic shock and sepsis tissue injury : comparison with LR, HES, and LR-PTX treatments

김호중 Graduate School, Yonsei University 2012 국내박사

The effectiveness of hypertonic saline and pentoxifyline(HSPTX) resuscitation in hemorrhagic shock and sepsis tissue injury comparison with the LR, HES-HTS, LR-PTX treatments Purpose: to compare the organs (lung and liver) injury and laboratory results of hemorrhagic shock and sepsis models with various treatments; HTSPTX, RLPTX, HESHTS and RL. Methods: Male Sprague-Dawley rats (200?290g, Charles-River, St. Constant, Canada) were used this study and they were randomly assigned to one of the four groups (n = 16 per group) to receive the following treatments: (1) lactated Ringer's solution group (LR); (2) 7.5% hypertonic saline with hydroxyethyl starch group (HTS-HES); (3) lactated Ringer's (LR) solution with PTX group (LR-PTX); and (4) 7.5% hypertonic saline with PTX group (HTS-PTX) and each group was divided to one of the two following event models; (1) hemorrhagic shock (n = 8); (2) sepsis (n = 8). The venous catheter was utilized for injection of resuscitative fluids, and the arterial catheter was used to withdraw blood and monitor the mean arterial pressure (MAP) by MacLab? (PowerMac, AD Instruments, Australia). Organ (lung and liver) histologiy study, Bronchoalveolar lavage (BAL) and Cytokine test were performed. Results: Mean lung injury score is 1.7. Total leukocyte count in the BAL 24 h after treatment was significantly higher in LR treated sepsis model (10 x 106 ±0.8) as compared to other sepsis treatment models (HTS-HES; 6 x 106 ± 1.2, LR-PTX; 5 x 106 ± 1.5, HTS-PTX; 5 x 106 ± 0.6)(p<0.05). The higher total leukocyte count in the LR sepsis model (17 ±1.5 %) is due to an increased number of neutrophils, as compared to other sepsis treatment models (HTS-HES; 6 ± 0.8 %, LR-PTX; 10 ± 1.3 %, HTS-PTX; 5 ± 0.4 %%). The total hepatic injury score in sepsis model was significantly greater in the LR group (9.9 ± 0.5) than either the other treatment groups (HTS-HES; 6.7 ±0.8, LR-PTX; 5.6 ± 0.7, HTS-PTX; 3.1 ± 0.9, respectively; p < 0.05). Also, in shock model, LR group (10.6 ± 2.1) was significantly higer (HTS-HES; 5.8 ±0.9, LR-PTX; 7.3 ±0.9, HTS-PTX; 3.5 ± 0.9, respectively; p < 0.05). HTSPTX resuscitation resulted in a 49% decrease in TNF-α, a 29% decrease in IL-1β, and a 58% decrease in IL-6 at 24 hours when compared with RL in shock model (p < 0.05) and, in sepsis model, in a 45% derease in TNF-α, a 24% decrease in IL-1β, and a 35% decrease in IL-6 at 24 hours when compared with LR (p < 0.05). Conclusion: HTS-PTX may have most advantage over other proposed resuscitation strategies and LR-PTX or HTS-HES was better results than LR therapy.

Prediction models for severely injured occupants using machine learning analytics based on oversampling class imbalanced data

공준석 Graduate School, Yonsei University 2023 국내박사

병원 전 단계에서 교통사고 환자의 인체상해 예측은 환자의 중증도분류에 대한 정확한 의사결정과 적절한 이송체계를 통해 인명피해를 경감시키는 효과가 있다. 최근 사고 현장에서 즉각적인 상해유형 판별을 위해 텔레메틱스를 기반한 자동검출 시스템의 법제화가 각 국에서 도입되고 있으며, 이를 위한 외상환자의 상해예측 모델에 대한 요구가 부각되고 있다. 그러나 환자의 상해예측 모델은 데이터의 클래스 불균형(Class imbalance)에 따라 실제 왜곡된 예측과 성능저하를 초래할 수 있다. 또한, 아직까지 교통사고 환자의 상해를 판별하기 위한 임상자료의 균등화(balancing)를 통한 최적화된 모델의 부재로 다양한 모델간의 성능 비교가 필요하다. 본 연구는 국내 5개 지역의 응급의료센터에 내원한 차대차 탑승자 교통사고 환자를 대상으로 상해중증도 판별을 개선하기 위해 최신의 기계학습 모델의 성능을 평가하고자 한다. 본 연구는 2011년 1월부터 2021년 4월까지 한국형 자동차사고-인체상해 구축 (Korea In-Depth Accident Study, KIDAS) 데이터베이스에 등록된 1,417명의 교통사고 환자를 대상으로 선정하였다. 상해중증도에 대한 분류는 손상중증도점수(Injury Severity Score, ISS) 기준 15점 이상을 중상해 환자로 선별하였다. 다양한 사고유형에 따라 보다 정밀한 예측성능 확보를 위해 전복사고를 제외한 평면충돌 사고를 고려하였다. 또한 차대차 사고에서 두 차량 간의 충돌 부조화(crash incompatibility)을 고려하여 서로 다른 차량 세그먼트 구성을 분류하였다. 탑승환자의 중증도분류 결과에 따른 데이터 불균형성을 극복하기 위해 네 가지의 데이터 샘플링 기법(i.e., class-weighting, resampling, synthetic minority oversampling, and adaptive synthetic sampling)을 사용하였다. 교통사고 환자의 상해예측 판별을 위한 기계학습 모델은 logistic regression, extreme gradient boosting (XGBoost), 그리고 multilayer perceptron (MLP)로 선정하였다. 모델 성능을 향상시키기 위해 하이퍼파리미터를 조정하고 5겹 교차검증을 통해 각 모델의 과적합을 방지하였다. 외상환자의 상해예측은 과소분류 10% 이하의 수준을 지닌 모델을 기반으로 모델의 성능을 평가하였다. 본 연구에서 데이터 샘플링 기법을 적용한 SMOTE와 ADASYN 모델이 클래스 불균형 데이터 보다 예측 성능이 높았다. 특히 SMOTE 기반 XGBoost 모델에서 가장 우수한 예측 성능을 보였다. 해당 모델을 활용한 특성중요도에서 두 차량간의 속도변화량(Delta-V)이 교통사고 탑승자의 상해 예측에 기여한 주요 요인으로 확인되었다. 이러한 결과는 환자의 중증도분류에 따른 클래스 불균형을 데이터 샘플링 기법을 구현하여 상해 심각도에 대한 개선된 예측 성능을 기대할 수 있다. 따라서, 교통사고 환자의 상해 예측을 위해 활용되는 변수의 유형에 따른 샘플링 기법과 학습모델 선정이 고려되어야 한다. Injury prediction models improve trauma outcomes for motor vehicle occupants with accurate decision-making and early transport to appropriate trauma centers. This study aimed to investigate the injury severity prediction (ISP) capability of machine-learning analytics based on five-different regional Level 1 trauma center-enrolled patients in Korea. We studied car crash-related injury data from 1,417 patients enrolled in the Korea In-Depth Accident Study database from January 2011 to April 2021. Severe injury classification was defined as an Injury Severity Score ≥ 15. Planar collisions were considered by excluding rollovers which would compromise an accurate prediction. Furthermore, dissimilarities of the collision partner component based on vehicle segmentation were assumed for crash incompatibility. To handle class-imbalanced clinical datasets, we used four data-sampling techniques (i.e., class-weighting, resampling, synthetic minority oversampling, and adaptive synthetic sampling). Machine-learning analytics based on logistic regression, extreme gradient boosting (XGBoost), and a multilayer perceptron model were used for the evaluations. Each model was executed using five-fold cross-validation to solve overfitting consistent with the hyperparameters tuned to improve model performance. The area under the receiver operating characteristic curve was 0.896. Additionally, the present ISP model showed an under-triage rate of 6.1%. The Delta-V, age, and Principal Direction of Force (PDOF) were significant predictors. The results demonstrated that the data-balanced XGBoost model achieved a reliable performance on injury severity classification of emergency department patients. This finding considers ISP model selection, which affected prediction performance based on overall predictor variables.

(A) predictive model to analyze factors affecting the presence of mild whiplash-associated disorders in minor motor vehicle crashes based on the Korean in-depth accidents study (KIDAS) database

이희영 Graduate School, Yonsei University 2019 국내박사

연구목적: 본 연구에서는 경증 목 손상의 중증도에 영향을 미치는 요인을 분석하고 경미한 교통사고에서 탑승자의 경증 목 손상 여부를 예측할 수 있는 모델을 개발하고자 한다. 연구방법: 2011년 1월부터 2017년 6월 사이에 네 곳의 권역응급의료센터에서 수집한 한국형 자동차사고 심층분석 조사자료 데이터베이스를 사용하였다. 차량의 파손 정보는 사고차량을 실사고 조사하면서 촬영한 후 collision deformation classification (CDC) code를 이용하여 획득하였고, 차량탑승자의 상해 정보는 약식상해등급(Abbreviated Injury Scale, AIS), 최대약식상해등급(Maximum AIS), 손상중증척도(Injury Severity Score, ISS)를 이용하여 획득하였다. 환자의 통증 호소와 의료진의 진단을 바탕으로 Quebec Task Force 분류법을 사용하여 목 손상에 대해 총 다섯 단계(QTF 0-4)로 분류할 수 있다. 본 연구에서는 QTF 1에 속하는 군을 경증 목 손상으로 정의하였고, QTF 0을 손상이 없는 군으로 정의하였다. 2011년에서 2016년까지의 KIDAS 데이터를 이용하여 경증 목 손상 발생에 영향을 주는 요인을 분석하였고 로지스틱 회귀 분석을 통해 예측모델을 도출하였다. 전체 데이터의 90%를 사용하여 예측모델을 도출하고 나머지 10%의 데이터로 검증하는 bootstrapping 방법을 무작위로 300번 수행하여 내적타당도를 분석하였고, 예측을 위한 조건을 충족시키는 2017년 KIDAS 데이터베이스의 13명의 탑승자 데이터에 적용하여 외적타당도를 분석하였다. 결과: 2011년부터 2016년까지 수집된 KIDAS 데이터베이스의 2,629명의 탑승자 중 제외조건들을 적용한 후 459명의 탑승자 데이터를 이용하여 예측모델을 개발하였다. 로지스틱 회귀 분석을 이용하여 성별, 나이, BMI, 차량종류, 안전벨트 착용여부, 차량탑승위치, 충돌유형, 차량함입정도를 선택한 후 예측모델을 도출하였다. 이 예측모델은 실제 경증 목 손상을 예측하는데 있어 65.5%의 설명력을 가졌다 (C-statistics: 0.655). 2017 KIDAS 데이터베이스를 사용하여 예측모델의 외적타당도 분석 결과, 민감도는 0.500, 특이도는 0.692, 정확도는 0.692이었다. 결론: 본 연구에서 도출된 예측모델은 경증 목 손상 여부를 판단할 수 있다는 점에서 의미 있는 결과를 제공하였다. 그러나 경미한 교통사고에서 경증 목 손상에 영향을 주는 요인이 더 있을 것으로 판단하기 때문에, 본 연구의 결과를 바탕으로 공학적인 영향요인, 사회학적인 영향요인 등을 추가하여 좀 더 정확도 높은 예측모델을 발전시킬 수 있다. Objectives: We aimed to analyze factors affecting the severity of mild whiplash-associated disorders (WAD) and to develop a predictive model to evaluate the presence of mild WAD in minor motor vehicle crashes (MVCs). Methods: From 2011 to 2017, we used Korean In-Depth Accident Study (KIDAS) database which was collected from four regional emergency centers. The Collision Deformation Classification (CDC) code was obtained as vehicle’s damage information, and AIS, MAIS, and ISS were used as occupant’s injury information. The degree of WAD was determined using the Quebec Task Force (QTF) classification, comprised of 5 stages (QTF 0–4), depending on the occupant’s pain and the physician’s findings. QTF 1 was defined as mild WAD, and we used QTF 0 to define those uninjured. For KIDAS data between 2011 and 2016, a logistic regression model was used to identify factors affecting the occurrence of mild WAD and a predictive model was constructed. Internal validity was estimated using random bootstrapping, and the external validity of the derived predictive model was assessed using the 13 MVC occupants from the 2017 KIDAS database meeting our inclusion criteria. Among the 137 MVC occupants from 2017 KIDAS database for the analysis of the external validity of the derived predictive model, the predictive model was verified for 13 MVC occupants. Results: Of the 2,629 occupants in the KIDAS database collected from 2011 to 2016, after applying several exclusion conditions, 459 occupants were used for developing the predictive model. Logistic regression analysis was used to derive a predictive model based on sex, age, body mass index, type of vehicle, belt status, seating row, crush type, and crush extent. This predictive model had an explanatory power of 65.5% to determine an actual QTF of 0 and 1 (C-statistics: 0.655). As a result of the external validity analysis of the predictive model using data from the 2017 KIDAS database (N=13), sensitivity, specificity, and accuracy were 0.500, 0.857, and 0.692, respectively. Conclusions: Using the predictive model, the results of the external validity analysis showed low sensitivity but high specificity. This predictive model provided meaningful results, with a high success rate for determining no injury to an occupant. Given our study results, future research is needed to create a more accurate predictive model that includes relevant technical and sociological factors.

(The) effect of post-treatment use of fluvastatin in hemorrhagic shock in rats

김오현 Graduate School, Yonsei University 2015 국내박사

(The) anti-inflammatory effects of ulinastatin in traumatic patients with a hemorrhagic shock

박경혜 Graduate School, Yonsei University 2007 국내석사

Background: Ulinastatin, a glycoprotein from human urine, inhibits the proteolytic action and has an anti-inflammatory effect on tissues. Ulinastatin reduces the renal dysfunction associated with the ischemia-reperfusion of the kidney as well as the blood transfusion-induced Polymorphonuclear Leukocyte Elastase (PMNE) which may injure a variety of tissues and organs. However, the effect of ulinastatin on traumatic hemorrhagic shock has rarely been reported.Purpose: The aim of this study was to investigate the use of ulinastatin in association with the suppression of plasma proinflammatory cytokine and PMNE and the good prognosis in the patients with traumatic hemorrhagic shock.Subjects and Methods: Nineteen patients who were admitted to the emergency department for trauma with hemorrhagic shock from June 2006 to October 2006 were enrolled. Eleven patients received ulinastatin at random. Ulinastatin 100,000 IU was intravenously administered every 8 hours for a total of 300,000 IU. Measurements of serum PMNE, Tumor Necrosis Factor Alpha (TNF-α) and Interleukin 6 (IL-6) were taken before ulinastatin treatment, at 24 hours, 2 days, 3 days and 7 days after admission. We compared the Systemic Inflammatory Response Syndrome (SIRS) score, the Multiple Organ Dysfunction Syndrome (MODS) score and the Acute Physiology, Age, Chronic Health Evaluation (APACHE) III between the control group and the ulinastatin group..Results: There were no significant baseline differences between the control group and the ulinastatin group. Furthermore, there were no significant differences in laboratory data, treatment and mortality between the control group and the ulinastatin group. The serum PMNE levels of the ulinastatin group were lower than the control at the second hospitalized day (11.58±5.57 vs 4.33±1.21, p=0.19). Serum TNF-α and IL-6 levels of the ulinastatin group decreased 24 hours after admission and were lower than the control, however, there were no significant differences.Conclusion: Ulinastatin 300,000 IU leads to decrease the serum PMNE in traumatic patients with a hemorrhagic shock on the second day of hospitalization 배경 및 목적: Ulinastatin은 사람의 소변에서 분리 정제된 당단백질로서 단백 분해 효소를 저해하고 항염증 작용이 있다. 또한 신장의 허혈 손상을 줄이고 수혈 후 주요 장기에 손상을 유발하는 Polymorphonuclear Leukocyte Elastase (PMNE)를 억제하는 효과가 있다고 알려져 있다. 그러나 외상에 의한 출혈성 쇼크에서는 ulinastatin의 효과는 잘 알려져 있지 않다. 본 연구에서는 출혈성 쇼크를 동반한 외상 환자에서 ulinastatin 투여가 환자의 혈청 전염증 시토카인 (cytokine)과 PMNE의 발현을 억제하여 환자의 예후에 좋은 영향을 주는지 알아보고자 한다.대상 및 방법: 2006월 6월부터 10월까지 응급실에 내원한 출혈성 쇼크를 동반한 외상 환자를 대상으로 하였다. 응급실 내원 당시에 출혈성 쇼크가 진단된 환자 중 무작위로 정하여 실험군에 해당하는 환자에게 ulinastatin을 1회에 10만 단위 씩8시간 간격으로 총 3회 투여하였다. 투여 전, 투여 후 24시간, 2일 째, 3일 째, 7일 째에 PMNE, Tumor Necrosis Factor Alpha (TNF-α), Interleukin 6 (IL-6)를 측정하였고, 내원 당시와 내원 후의 Systemic Inflammatory Response Syndrome (SIRS) score, Multiple Organ Dysfunction Syndrome (MODS) score와 Acute Physiology, Age, Chronic Health Evaluation (APACHE) III를 비교하였다.결과: 대상 환자는 모두 19명으로 대조군은 8명, 실험군은 11명이었으며, 두 군의 Injury Severity Scale은 차이가 없었다 (p=0.091). 그리고 두 군의 혈액학적 검사 소견, 치료, 사망률 등에도 유의한 차이가 없었다. PMNE의 농도는 내원 2일 째 대조군이 11.58±5.57 ng/ml, 실험군이 4.33±1.21 ng/ml로 두 군 간의 유의한 차이를 보였다 (p=0.19). TNF-α, IL-6는 내원 1일 째부터 감소하면서 대조군보다 낮은 값을 보였으나 의미 있는 차이는 없었다.결론: Ulinastatin 30만 단위는 출혈성 쇼크를 동반한 외상 환자의 혈청 PMNE 농도를 내원 2일 째에 ulinastatin을 투여하지 않은 환자보다 의미 있게 낮춘다

Effects of the characteristics of patient and emergency medical technician on the use intention of telemetry by EMT in ambulances

황지영 Graduate School, Yonsei University 2014 국내박사

This study aimed to verify the effects of patient factors perceived by emergency medical technicians (EMTs) as well as their social and organizational factors on pre-hospital telemetry Use Intention based on the technology Use Intention and elaboration likelihood models. This is a retrospective empirical study. Questionnaires were developed on the basis of clinical factors of 72,907 patients assessed by pre-hospital telemetry from January 1, 2009 to April 30, 2012 by reviewing their pre-hospital medical care records and in-hospital medical records. Questionnaires regarding the social and organizational factors of EMTs were created on the basis of a literature review. In total, 136 EMTs who had experience in utilizing pre-hospital telemetry were surveyed from April 1 to April 7, 2013. Reliability, validity, hypotheses, and the model goodness of fit of the study tools were tested. Factors that influence the use of telemetry by EMT in ambulances included patients’ clinical factors, as well as complex organizational and environmental factors surrounding their occupational environments. This suggests that the rapid Use Intention and dissemination of such systems require EMTs to be supported at both the technical and organizational levels.

Effect of mild hypothermia on coagulatory function and survival in sprague-dawley rats exposed to uncontrolle hemorrhagic shock

박경혜 Graduate School, Yonsei University 2012 국내박사

Background: Acute coagulopathy, hypothermia, and acidosis are the lethal triad of conditions manifested by major trauma patients. Recent animal studies have reported that hypothermia improves survival in animals subjected to controlled hemorrhagic shock. Purpose: The objective of this study was to investigate the effect of hypothermia on coagulation in rats subjected to uncontrolled hemorrhagic shock. Subjects and Methods: Thirty-two male Sprague?Dawley rats were randomly divided into four groups: normothermia (control, group N), hypothermia (group H), hypothermic hemorrhagic shock (group HS), and normothermic hemorrhagic shock (group NS). Hemorrhagic shock was induced by splenic laceration. Capacity for coagulation was measured by rotation thromboelastometry (ROTEM?), and was measured at baseline as well as the end of the shock and resuscitation periods. Survival was observed for 48 hours post-trauma. Results: Baseline parameters were not different among the groups. Rats exposed to hypothermia alone did not differ in coagulation capacity compared to the control group. Clot formation time (CFT) and maximal clot firmness (MCF) in group HS decreased as the experiment progressed. Maximal clot firmness time (MCFt) in groups H and HS was significantly prolonged during shock and resuscitation compared with that in group NS. In group NS, MCF did not change significantly, but MCFt was reduced compared with baseline. Group HS had poor survival when compared with normovolemic groups. Conclusion: Blood clotted less firmly in traumatic hemorrhagic shock, and hypothermia prolonged clotting. However, clot firmness maximized rapidly under normothermic hemorrhagic shock. Hemorrhage would continue for a longer time in hypothermic hemorrhagic shock. Survival of hypothermic shock was not significantly different compared to that of normothermic hemorrhagic shock.

Simple criteria that predict major injury of front-seat passenger in frontal collision of passenger car

김상철 Graduate School, Yonsei University 2014 국내박사

Background: A frontal motor vehicle collision is the most common type of crash that results in fatalities. In this study, we suggested simple criteria that predict major injury to the frontseat occupant in the frontal collision of a passenger car. Subjects and Methods: From January 2011 to December 2013, we collected data from front-seat occupants admitted to one of two emergency centers by ambulance following a frontal collision accident. We surveyed the cause of the accident, vehicle damage, information on the occupant, and severity of injury. Vehicle damage was assessed according to the collision deformation classification code through evaluation of photographs of the actual accident vehicle, and the patient’s injury severity was evaluated by the injury severity score (ISS). Bivariate logistic regression models were formulated, and the cutoff point of deformation extent (DE) was inferred by receiver operating characteristic (ROC) curve analysis. Results: Of the 192 subjects, 113 were males and 52 were major injury patients whose ISS exceeded 15. Gender, seat belt status, extent of vertical crash, and DE were significantly different between major and minor injuries (p < 0.05). After adjusting for confounds, not fastening the seat belt doubled the risk of major injury (OR = 2.2, 95% CI 1.061?4.390), and a cutoff value of three DE tripled the risk of major injury (OR = 3.2, 95% CI 1.382?7.343). In ROC curve analysis, DE 3 in the seat-belt-unfastened group and DE 5 in the seat-beltfastened group predicted major injury (area under the curve: 0.740 [95% CI, 0.627?0.834], sensitivity: 89.3%, specificity: 52.1%; and area under the curve: 0.696 [95% CI, 0.604?0.778], sensitivity: 41.7%, specificity: 94.6%, respectively). Conclusions: At the scene of a frontal collision, emergency personnel can consider seat belt nonuse and DE ≥ 3 as criteria to transport front-seat occupants to trauma center.

(The) effect of mobile and wearable device intervention on increased physical activity to prevent metabolic syndrome

김희진 Graduate School, Yonsei University 2023 국내박사

There are currently few studies on whether wearable device and mobile application-based interventions as a means of increasing physical activity (PA) can effectively prevent metabolic syndrome (MetS) among middle-aged and elderly people living in rural areas. The study aimed to determine whether interventions using mobile and wearable devices in metabolic syndrome risk groups in rural Korea are effective in improving PA and improving physical health indicators. In this study, conducted from December 2019 to June 2020, participants used a wearable device (Galaxy watch Active1, Korea) and an additional mobile application (YONSEI HEALTH, Korea), depending on the group. At the beginning of the study and six months later, each participant completed a clinical examination, the International Physical Activity Questionnaire-long form (IPAQ-LF), the EQ-5D-3L questionnaire, and the Korean-style eating habits questionnaire. The number of step counts completed by each participant was monitored via a website. The main outcome of intervention was differences in PA and clinical measurements within and between groups before and after clinical trials in the enhanced intervention (EI) group and the standard intervention (SI) group after intervention. At the secondary endpoint, reductions in MetS risk factors that improved with changes in PA were assessed. 267 participants were selected according to the randomized study registration procedure and finally 221 completed the study after 6 months. As a result of the study, only the EI group had improved weight and body mass index (BMI) (P<0.001). There were statistically significant differences between groups after clinical trials of these variables. There was a statistically significant difference between groups in weight (t=-2.26; P<0.05), and there was also a statistically significant difference in BMI (t=-2.04; P<0.05). In both the groups, the systolic blood pressure (SBP), diastolic blood pressure (DBP), waist circumstance (WC), and glycated hemoglobin A1c (HbA1c) levels decreased (P<0.001). The total PA in the EI group was 124.36 (SD, 570.0), an increase of about 2.8 times over the SI group (44.47 (SD, 224.85)). In addition, only the EI group achieved moderate to moderate physical activity (MVPA) at the recommended level (EI: 188 min/wk; SI: 118 min/wk). The quality of life index scores did not change before or after the clinical trial, either within or between groups. In the EI group, the anxiety/depression levels improved marginally, but it was not statistically significant. As for eating habits, in the EI group, the frequency of intake of fatty, instant, and fast food decreased, but it was not statistically significant. In the SI group, fruit intake increased statistically (P<0.01), but the intake of fatty food, instant, and fast food intake remained unchanged. In the logistic regression analysis using improvement in PA levels as the dependent variable, the EI group had a 2.27 times higher probability of improving PA than the SI group (OR: 2.27, 95% CI: 1.27–4.05). MetS was effectively prevented in rural Korea by utilizing wearable devices for PA monitoring and intervention. BP, WC, and HbA1c were improved in both groups and effectively reduced MetS factors. However, it was only the EI group that continuously increased the number of MVPA and steps to prevent MetS above the recommended level. Weight and BMI were further improved, and the proportion of participants without MetS risk factors increased. 현재 농촌 지역에 거주하는 중년 및 노년층에서 신체 활동을 개선하는 방법으로 웨어러블 기기 및 모바일 애플리케이션을 활용한 중재가 대사 증후군을 효과적으로 예방할 수 있는지에 대한 연구는 거의 없다. 본 연구는 한국 농촌 지역의 대사증후군 위험군에 모바일 및 웨어러블 기기를 사용한 중재가 PA 개선 및 임상적 지표 개선에 효과적인지 여부를 확인하는 것을 목표로 했다. 2019 년 12 월부터 6 개월동안 진행된 연구에서 참여자들은 그룹에 따라 웨어러블 기기(Galaxy watch Active1, 한국)와 추가적으로 모바일 애플리케이션(YONSEI HEALTH, 한국)을 사용했다. 임상시험 후, 각 참가자는 임상검사, 국제신체활동 설문지, 삶의 질 평가 설문지, 한국식 식습관 설문지를 작성했다. 각 참가자가 완료한 걸음 수는 웹 사이트를 통해 모니터링 되었다. 개입의 주요 결과는 개입 후 강화중재군(EI)과 일반중재군(SI)에서 임상 시험 전후 그룹 내 및 그룹 간 신체활동 및 임상 측정값 차이였다. 2 차 방문에서 신체활동량의 변화에 따라 개선된 대사증후군 위험 인자의 개선정도를 평가했다. 무작위 연구 등록 절차에 따라 267 명의 참가자가 선발되었고 최종적으로 221 명이 6 개월 후에 연구를 완료했다. 연구 결과, EI 군만이 체중과 체질량지수(BMI)가 개선되었다(P<0.001). 이러한 변수에 대한 임상 시험 후 그룹 간에 통계적으로 유의미한 차이가 있었다. 그룹간 체중(t=-2.26; P<0.05)에서 통계적으로 유의한 차이가 있었고 BMI(t=-2.04; P<0.05)에서도 통계적으로 유의한 차이가 있었다. 두 그룹 모두에서 수축기 혈압(SBP), 이완기 혈압(DBP), 허리둘레(WC), 당화혈색소 A1c(HbA1c) 수치가 감소했다(P<0.001). EI 군의 총 신체활동량은 124.36(SD, 570.0)으로 SI 군(44.47(SD, 224.85))보다 약 2.8 배 증가했다. 삶의 질 점수는 임상 시험 전후, 그룹 내 또는 그룹 간에 변화되지 않았다. EI 군에서 불안/우울이 다소 감소하고 식습관에서 기름진 음식, 즉석식품, 패스트푸드 섭취 빈도가 감소했지만 통계적으로 유의하지 않았고, SI 군에서는 과일 섭취량이 통계적으로 유의하게 증가했다(P<0.01). 기름진 음식, 인스턴트, 패스트푸드 섭취량은 변화가 없었으며, 신체활동량 수준의 개선을 종속변수로 한 로지스틱 회귀분석에서는 EI 군의 신체활동 개선 확률이 SI 그룹보다 2.27 배 더 높았다 (OR: 2.27, 95% CI: 1.27–4.05). 한국 농촌 인구의 웨어러블 기기를 활용한 신체활동 모니터링 및 중재는 대사증후군을 효과적으로 예방되었다. 혈압, 허리둘레 및 당화혈색소는 두 그룹 모두에서 개선되었고 대사증후군 요인도 효과적으로 감소시켰다. 그러나 대사증후군 예방을 위해 권장 수준 이상으로 중고강도 신체활동 및 걸음수를 지속적으로 늘린 것은 EI 군 뿐이었다. 그 결과, 체중과 BMI 가 더욱 개선되었고 대사증후군 위험 요인이 없는 참가자의 비율이 증가했다.

(A) prediction model for prevention and management of metabolic syndrome based on machine learning

이정훈 Graduate School, Yonsei University 2023 국내박사

Digital health-based lifestyle interventions (e.g., mobile applications, short message services, wearable devices, social media, and interactive websites) are widely used to manage metabolic syndrome (MetS). This study aimed to confirm the usefulness of digital health-based lifestyle interventions using healthcare devices and propose a novel prediction model of prevention and management for MetS. Participants with one or more MetS risk factors were recruited from December 2019 to September 2020, and finally, 106 participants were analyzed. Participants were provided with five healthcare devices and applications. Characteristics were compared at baseline and follow-up, and lifelog data that were collected during the clinical trial were analyzed. With these results, the frequency of use of healthcare devices for continuous self-care was quantified, and a novel prediction model for the prevention and management of MetS was developed. The model predicts persistence in continuous engagement as well as abbreviated risk factors for self-care effects. Representative machine-learning classifiers were used and compared. In both models, the random forest classifier showed the best performance, and feature selection was optimized through random forest-recursive feature elimination. As a result, the prediction model for persistence showed recall of 83.0%, precision of 92.4%, an F1-score of 0.874, a Matthews correlation coefficient (MCC) of 0.844, and accuracy of 94.9%. The prediction model for abbreviated risk factors showed a recall of 79.8%, a precision of 87.2%, an F1-score of 0.834, and an MCC of 0.797 for increased abbreviated risk factors, and a recall of 75.1%, a precision of 85.5%, an F1-score of 0.800, and an MCC of 0.747 for decreased abbreviated risk factors. The prediction model proposed showed high performance. Based on self-care with digital health-based lifestyle interventions, prediction models could be helpful for the prevention and management of MetS.