Kashani Homa, Zeraati Hojjat, Mohammad Kazem, Goodarzynejad Hamidreza, Mahmoudi Mahmood, Sadeghian Saeed, Boroumand Mohammadali
Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran.
Tehran Heart Center, Tehran University of Medical Sciences, Tehran, Iran.
J Tehran Heart Cent. 2016 Apr 13;11(2):55-61.
Investigators frequently encounter continuous outcomes with plenty of values clumped at zero called semi-continuous outcomes. The Gensini score, one of the most widely used scoring systems for expressing coronary angiographic results, is of this type. The aim of this study was to apply two statistical approaches based on the categorization and original scale of the Gensini score to simultaneously assess the association between covariates and the presence and severity of coronary artery disease (CAD). We considered the data on 1594 individuals admitted to Tehran Heart Center with CAD symptoms from July 2004 to February 2008. The participants' baseline demographic and clinical characteristics were collected, and their coronary angiographic results were expressed through the Gensini score. The generalized ordinal threshold and two-part models were applied for the statistical analyses. Totally, 320 (20.1%) individuals had a Gensini score of zero. The results of neither the two-part model nor the generalized ordinal threshold model showed a significant association between Factor V Leiden and the occurrence of CAD. However, based on the two-part model, Factor V Leiden was associated with the severity of CAD, such that the Gensini score increased by moving from a wild genotype to a heterozygote (β = 0.44; 95% CI: 0.20-0.69 in logarithm scale) or a homozygote mutant (β = 0.70; 95% CI: 0.28- 1.12 in logarithm scale). The proportional odds assumption was not met in our data ([Formula: see text]= 54.26; p value < 0.001); however, a trend toward severe CAD was also observed at each category of the Gensini score using the generalized ordinal threshold model. We conclude that besides loss of information by sorting a semi-continuous outcome, violation from the proportional odds assumption complicates the final decision, especially for clinicians. Therefore, more straightforward models such as the two-part model should receive more attention while analyzing such outcomes.
研究人员经常遇到连续型结局，其中大量数值聚集在零处，这种情况被称为半连续型结局。广泛用于表达冠状动脉造影结果的Gensini评分就是这种类型。本研究的目的是应用基于Gensini评分分类和原始量表的两种统计方法，同时评估协变量与冠状动脉疾病（CAD）的存在及严重程度之间的关联。我们纳入了2004年7月至2008年2月期间因CAD症状入住德黑兰心脏中心的1594名个体的数据。收集了参与者的基线人口统计学和临床特征，并通过Gensini评分来表示他们的冠状动脉造影结果。采用广义有序阈值模型和两部分模型进行统计分析。总共有320名（20.1%）个体的Gensini评分为零。两部分模型和广义有序阈值模型的结果均未显示因子V莱顿与CAD的发生之间存在显著关联。然而，基于两部分模型，因子V莱顿与CAD的严重程度相关，从野生基因型转变为杂合子（对数尺度下β = 0.44；95% CI：0.20 - 0.69）或纯合子突变体（对数尺度下β = 0.70；95% CI：0.28 - 1.12）时，Gensini评分会升高。我们的数据不满足比例优势假设（[公式：见原文] = 54.26；p值 < 0.001）；然而，使用广义有序阈值模型在Gensini评分的每个类别中也观察到了CAD严重程度增加的趋势。我们得出结论，除了对半连续型结局进行分类会导致信息丢失外，违反比例优势假设会使最终决策变得复杂，尤其是对临床医生而言。因此，在分析此类结局时，像两部分模型这样更直接的模型应受到更多关注。
