MCTT HCE model을 이용한 안점막자극시험의 검증연구를 위해 다양한 통계방법을 통한 재현성 및 예측력 평가

Abstract

본 연구의 목적은 인체 각막 모델인 MCTT HCE model을 이용한 안점막자극시험의 검증연구 결과를 바탕으로 재현성을 평가하고자 한다. 3개 lab에서 실험한 20개 물질에 대한 결과를 다양한 통계 방법을 사용하여 실험실내·실험실간 반복 실험한 결과 간에 차이가 없음을 입증하도록 한다. 또한 추가적으로 실험을 실시하여 총 119개 물질에 대한 예측력 평가를 통해 이상적인 cut-off 값을 산출한다. 3개 기관에서 시행하여 나온 실험 data를 Descriptive statistics 방법과 Inferential statistics 방법으로 나누어 재현성을 평가하였다. Descriptive statistics 방법으로는 Graph with Mean±SD, Scatter plot with Mean±SD, Agreement 방법을 사용하였다. Inferential statistics 방법으로는 correlation of coefficient 방법인 Pearson correlation coefficient, Spearman correlation coefficient, Intraclass correlation, Fleiss’ Kappa 방법으로 trial 간 상관관계와 lab 간 상관관계를 구해 실험실내·실험실간 구하였다. 또한 lab 간 결과의 평균이 통계적으로 유의성이 있는지 검증하기 위해 ANOVA와 Kruskal-Wallis로 분석하였다. 예측력 평가에서는 ROC curve를 그려 민감도와 특이도를 비교해 이상적인 cut-off 값을 산출하였다.;Statistical evaluation of within- and between-laboratory reproducibility of eye irritation test employing a new reconstructed human corneal epithelial model MCTT HCE® Diverse alternative in vitro eye irritation test methods are being developed to replace animal experimentation such as the rabbit eye Draize test. One of the new experimental models developed is the in vitro eye irritation test employing reconstructed human corneal epithelial model (MCTT HCE® model) prepared from primary-cultured human limbal epithelial cells. Using this model, reproducibility was evaluated in the tests with 20 chemicals conducted in three laboratories. Chemicals were judged to be irritant, or non-irritant according to 50% cell viability cut-off. The appropriate statistical methods were employed to analyze two types of data: the continuous data of cell viability and the binary data of decision. The within- and between-lab reproducibility was examined based on descriptive as well as inferential statistics; for descriptive statistics, we presented mean±SD and scatter plot data. For inferential statistics, we considered parametric methods, such as ANOVA and pearson correlation coefficient, and non-parametric methods, namely Kruskal-Wallis and spearman correlation coefficient. Descriptive statistics illustrated that within-and between-lab variance was minimal, and strong correlation coefficients (>0.9) were observed for within-and between lab reproducibility; the correlation was stronger for the continuous outcome (cell-viability data) than the binary data (decision of being irritant or non-irritant). Since cell viability data of some chemicals was near the cut-off, it is not surprising that stronger correlation was observed when outcomes were analyzed as the continuous data rather than the binary data. Yet, ANOVA and Kruskal-Wallis tests showed that reproducibility varied based on tested chemicals.