경동맥화학색전술 후 색전후증후군 관리를 위한 머신러닝기반 임상의사결정지원시스템 개발 및 효과

Abstract

본 연구는 간세포암종을 가진 환자가 경동맥화학색전술(transarterial chemoembolization: TACE)을 시행한 후 겪게 되는 색전후증후군(post-embolization syndrome: PES) 발생을 예측하고 치료적 중재안을 권고하는 머신러닝기반 임상의사결정지원시스템(post-embolization syndrome clinical decision support system CDSS: POEM CDSS)을 개발하고 그 효과를 평가하기 위해 시행된 비동등성 대조군 전후 시차설계를 적용한 유사실험연구이다. 연구는 POEM CDSS의 개발과 중재 적용 후 효과를 평가하는 과정을 거쳤다. 먼저 개발은 분석, 설계, 개발, 구현, 평가를 따르는 ADDIE 모형에 따라 진행되었고, 개발된 시스템은 모바일에 웹 애플리케이션의 방식으로 탑재하였다. 중재는 4개의 상급종합병원과 1개의 종합병원에서 TACE 환자를 간호하고 있는 간호사 40명을 대상으로(실험군 21명, 대조군 19명)하였고, 실험군에게 6주간 POEM CDSS를, 대조군에게는 TACE와 PES 관리 포켓북을 제공하였다. 효과평가는 간호사 실험군과 대조군에게 시스템 측면(간호사 변화)의 결과로 POEM CDSS 수용도, POEM CDSS 이용도, PES관련 간호지식을, 간호사결과 측면(안위 간호)에서는 증상관리, 환자중심기반 돌봄태도, 증상관리 장애요인, 안위간호역량을 측정하였다. 환자결과 측면(환자 안위)에서는 실험군의 병동에 입원한 환자를 대상으로 POEM CDSS 중재 전 26명, 중재 후 25명을 대상으로 PES 증상발생, 증상간섭, 입원기간, PES관련 추가약물의 투여횟수를 측정하였다. 수집된 데이터는 SPSS/MAC 27.0 소프트웨어를 이용하여, 빈도와 백분율, 평균과 표준편차, chi-square test, Fisher's exact test, Shapiro-Wilk test, independent t-test 및 Mann-Whitney U test로 분석하였다. 본 연구의 결과는 POEM CDSS의 개발과 효과평가로 다음과 같다. 1. POEM CDSS 개발은 1개의 상급종합병원에서 2011년부터 2022년까지 등록된 1,200명의 환자를 대상으로 전자간호기록에서 3,757건, 전자의무기록에서 1,496건을 추출하였고 머신러닝과 자연어처리를 활용하여 PES위험도를 예측하고 치료적 중재안을 권고하는 시스템을 개발하여, 예측의 정확도는 발열 83.9%, 오심/구토 90.3%, 복부 통증 70.6%, 전체 PES 72.2%로 나타나 높은 정확도를 보여주었다. POEM CDSS에 대한 전문가 내용타당도는 전체평균이 0.91점이었고, POEM CDSS 수용도는 3.79±0.61점로 긍정적이었으며, POEM CDSS 이용도는 1일 평균접속횟수 3.46회로 양호한 수준이었다. 2. POEM CDSS의 효과평가에서 시스템 측면(간호사 변화)의 결과에서 PES관련 간호지식 수준은 실험군과 대조군간 차이가 없었고, 간호사결과 측면(안위 간호)에서는 환자중심기반 돌봄태도 수준과 통증관리 장애요인 인식수준, 안위간호역량이 두 군간 유의하게 차이가 있었다. 실험군의 병동에 입원한 환자의 POEM CDSS 중재 전과 후의 환자결과 측면(환자안위)에서는 증상간섭 정도가 두 군간 유의한 차이가 있어 일부 가설이 채택되었다. 본 연구에서 개발한 POEM CDSS는 TACE 시술 후 발생가능한 PES의 위험을정확도 높게 예측하고 관련된 치료적 중재안을 권고하여, 간호의 질 향상과 근거기반 간호실무 향상에 기여하고, 환자에게 긍정적인 결과를 보여 임상현장에서 활용가능할 것이다.|This quasi-experimental study aimed to develop a machine learning-based post-embolization clinical decision support system called POEM CDSS and followed a non-equivalent control group non-synchronized design to evaluate its effectiveness in predicting post-embolization syndrome (PES) in patients with hepatocellular carcinoma after transarterial chemoembolization (TACE) and recommending therapeutic interventions. The study developed the POEM CDSS and evaluated its effectiveness after applying the intervention. First, the development was conducted according to the ADDIE model, i.e., analysis, design, development, implementation, and evaluation, and the developed system was uploaded as a mobile web application. The intervention targeted 40 nurses (experimental group: 21; control group: 19) who were caring for TACE patients in four tertiary general hospitals and one general hospital. The POEM CDSS was provided to the experimental group for 6 weeks, while an educational pocketbook was provided to the control group. To evaluate its effectiveness, the POEM CDSS acceptance, POEM CDSS utilization, and PES nursing knowledge were assessed in terms of the system aspect (nurse change) in the experimental and control groups. On the other hand, symptom management, patient-centered care attitude, symptom management barriers, and comfort care competence were assessed in the nurse outcome aspect (comfort care). In the patient outcome aspect (patient comfort), the PES symptoms, symptom interference, hospitalization period, and PES-related medications administered were assessed in 26 patients before the POEM CDSS intervention and in 25 patients after the POEM CDSS intervention for those in the experimental group. The collected data are presented as frequencies and percentages or means and standard deviations. The data were analyzed using chi-square test, Fisher's exact test, Shapiro-Wilk test, independent t-test, and Mann-Whitney U test using Statistical Product and Service Solutions/MAC 27.0 software. The results of this study detail the development of POEM CDSS and evaluation of its effectiveness. 1. To develop the POEM CDSS, 3,757 electronic nursing records and 1,496 electronic medical records were extracted from 1,200 patients registered at a tertiary general hospital between 2011 and 2022. A PES risk and therapeutic recommendation system was developed using machine learning and natural language processing. The prediction accuracy was 83.9% for fever, 90.3% for nausea/vomiting, 70.6% for abdominal pain, and 72.2% for the total PES, thus showing relatively high accuracy. As for the expert evaluation, overall average content validity index score was 0.91, the POEM CDSS acceptance was positive at 3.79±0.61 points, and the average frequency of accesses for POEM CDSS was 3.46 times/day, which was considered a good level. 2. When evaluating the effectiveness of POEM CDSS, there was no difference in the PES nursing knowledge of the system aspect (nurse change) between the experimental and control groups. However, a significant difference was observed in the patient-centered care attitude, pain management barriers, and comfort care competence of the nurse outcome aspect (comfort care). Some hypotheses were supported as there was a significant difference in the symptom interference of the patient outcome aspect (patient comfort) before and after POEM CDSS use in patients of the experimental group. The POEM CDSS developed in this study predicts the risk of PES after TACE with high accuracy and recommends related therapeutic interventions, contributing to the improvement in nursing quality and evidence-based nursing practice as well as positive results in patients. Therefore, it can be applied in the clinical field.