裂区设计能够灵活地增加试验处理和进行误差分级控制,在农业试验中应用广泛。但数据的统计分析较复杂,目前有效的相关统计软件十分缺乏。为了建立操作简单、实用性强、计算结果无误的统计分析手段,采用SAS广义线性混合模型(GLIMMIX)程...裂区设计能够灵活地增加试验处理和进行误差分级控制,在农业试验中应用广泛。但数据的统计分析较复杂,目前有效的相关统计软件十分缺乏。为了建立操作简单、实用性强、计算结果无误的统计分析手段,采用SAS广义线性混合模型(GLIMMIX)程序模块进行裂区设计数据的统计和分析,并通过实例分析说明GLIMMIX相较于传统一般线性模型(general linear model,GLM)程序模块的优缺点。结果发现,与GLM相比,GLIMMIX能够自动选用正确的误差项方差和自由度进行统计量计算,克服了GLM在某些情况下难于计算所需统计量的问题。实例验证说明,采用GLM进行裂区设计统计分析的不足及GLIMMIX分析的其他优点,并与混合程序模块(MIXED)进行了比较,认为GLIMMIX是裂区设计数据统计分析的首选模块。展开更多
重复测量试验对同一受试对象进行多次测量,各时间点数据间存在自相关性,进行方差分析和均值比较时需要进行特殊处理。虽然此方法在农业等研究领域运用十分广泛,但目前有效地相关统计方法鲜见。为了建立操作简单、实用性强、结果可靠的...重复测量试验对同一受试对象进行多次测量,各时间点数据间存在自相关性,进行方差分析和均值比较时需要进行特殊处理。虽然此方法在农业等研究领域运用十分广泛,但目前有效地相关统计方法鲜见。为了建立操作简单、实用性强、结果可靠的统计分析方法,本研究采用SAS的广义线性混合模型(Generalized Linear Mixed Models,GLIMMIX),以随机区组重复测量试验资料为例,说明了协方差结构筛选、方差分析和均值比较的具体方法。结果表明,用传统的裂区设计、多变量统计等方法会造成资料信息浪费,统计功效降低,缺区无法处理等问题,甚至会导致错误的结论。GLIMMIX能很好地处理自相关问题,功能强大,结果可靠,使用简单,允许缺区,是进行重复测量试验资料方差分析和均值比较的理想方法。目前在国内将其运用到农学类试验数据的统计分析的相关报道鲜见,该文在本领域具有很强的实用性和创新性。展开更多
Background: Factors associated with hospital mortality are usually identified and their effects are quantified through statistical modeling. To guide the choice of the best statistical model, we first quantify the pre...Background: Factors associated with hospital mortality are usually identified and their effects are quantified through statistical modeling. To guide the choice of the best statistical model, we first quantify the predictive ability of each model and then use the CIHI index to see if the hospital policy needs any change. Objectives: The main purpose of this study compared three statistical models in the evaluation of the association between hospital mortality and two risk factors, namely subject’s age at admission and the length of stay, adjusting for the effect of Diagnostic Related Groups (DRG). Methods: We use several SAS procedures to quantify the effect of DRG on the variability in hospital mortality. These procedures are the Logistic Regression model (ignoring the DRG effect), the Generalized Estimating Equation (GEE) that takes into account the within DRG clustering effect (but the within cluster correlation is treated as nuisance parameter), and the Generalized Linear Mixed Model (GLIMMIX). We showed that the GLIMMIX is superior to other models as it properly accounts for the clustering effect of “Diagnostic Related Groups” denoted by DRG. Results: The GLM procedure showed that the proportional contribution of DRG is 16%. All three models showed significant and increasing trend in mortality (P < 0.0001) with respect to the two risk factors (age at admission, and hospital length of stay). It was also clear that the CIHI index was not different under the three models. We re-estimated the models parameters after dichotomizing the risk factors at the optimal cut-off points, using the ROC curve. The parameters estimates and their significance did not change.展开更多
文摘裂区设计能够灵活地增加试验处理和进行误差分级控制,在农业试验中应用广泛。但数据的统计分析较复杂,目前有效的相关统计软件十分缺乏。为了建立操作简单、实用性强、计算结果无误的统计分析手段,采用SAS广义线性混合模型(GLIMMIX)程序模块进行裂区设计数据的统计和分析,并通过实例分析说明GLIMMIX相较于传统一般线性模型(general linear model,GLM)程序模块的优缺点。结果发现,与GLM相比,GLIMMIX能够自动选用正确的误差项方差和自由度进行统计量计算,克服了GLM在某些情况下难于计算所需统计量的问题。实例验证说明,采用GLM进行裂区设计统计分析的不足及GLIMMIX分析的其他优点,并与混合程序模块(MIXED)进行了比较,认为GLIMMIX是裂区设计数据统计分析的首选模块。
文摘重复测量试验对同一受试对象进行多次测量,各时间点数据间存在自相关性,进行方差分析和均值比较时需要进行特殊处理。虽然此方法在农业等研究领域运用十分广泛,但目前有效地相关统计方法鲜见。为了建立操作简单、实用性强、结果可靠的统计分析方法,本研究采用SAS的广义线性混合模型(Generalized Linear Mixed Models,GLIMMIX),以随机区组重复测量试验资料为例,说明了协方差结构筛选、方差分析和均值比较的具体方法。结果表明,用传统的裂区设计、多变量统计等方法会造成资料信息浪费,统计功效降低,缺区无法处理等问题,甚至会导致错误的结论。GLIMMIX能很好地处理自相关问题,功能强大,结果可靠,使用简单,允许缺区,是进行重复测量试验资料方差分析和均值比较的理想方法。目前在国内将其运用到农学类试验数据的统计分析的相关报道鲜见,该文在本领域具有很强的实用性和创新性。
文摘Background: Factors associated with hospital mortality are usually identified and their effects are quantified through statistical modeling. To guide the choice of the best statistical model, we first quantify the predictive ability of each model and then use the CIHI index to see if the hospital policy needs any change. Objectives: The main purpose of this study compared three statistical models in the evaluation of the association between hospital mortality and two risk factors, namely subject’s age at admission and the length of stay, adjusting for the effect of Diagnostic Related Groups (DRG). Methods: We use several SAS procedures to quantify the effect of DRG on the variability in hospital mortality. These procedures are the Logistic Regression model (ignoring the DRG effect), the Generalized Estimating Equation (GEE) that takes into account the within DRG clustering effect (but the within cluster correlation is treated as nuisance parameter), and the Generalized Linear Mixed Model (GLIMMIX). We showed that the GLIMMIX is superior to other models as it properly accounts for the clustering effect of “Diagnostic Related Groups” denoted by DRG. Results: The GLM procedure showed that the proportional contribution of DRG is 16%. All three models showed significant and increasing trend in mortality (P < 0.0001) with respect to the two risk factors (age at admission, and hospital length of stay). It was also clear that the CIHI index was not different under the three models. We re-estimated the models parameters after dichotomizing the risk factors at the optimal cut-off points, using the ROC curve. The parameters estimates and their significance did not change.