基于ZIGP模型的阳性过敏原数影响因素研究

The Study about Influencing Factors of Positive Allergens' Number Based on ZIGP Model

目的利用零膨胀广义泊松模型,分析阳性过敏原数的分布及相关的影响因素,为过敏和过敏性疾病的预防提供科学依据。方法收集2782例过敏患者信息,利用过离散O检验、Vuong检验和AIC、BIC等信息准则选取最优模型——零膨胀广义泊松模型,应用该模型分析过敏患者阳性过敏原数的影响因素。结果在就诊患者中,阳性过敏原数为0的患者有1969例,占70.78%。过离散的O=52.13,P<0.001,说明因变量存在过离散现象;两次Vuong检验的值分别是22.728和4.648,值均小于0.05,说明零膨胀模型优于传统计数模型;结合AIC、BIC值确定零膨胀广义泊松模型。影响因素分析,文化程度高易出现阳性过敏原(OR=2.232,OR 95%CI=1.2067~4.1286);性别(OR=0.596,OR 95%CI=-0.7703^-0.2663)、文化程度(OR=0.869,OR 95%CI=-0.2759^-0.0043)、辛辣食物(OR=1.119,OR 95%CI=0.0052~0.2190)和接触宠物以及毛絮物品(OR=1.338,OR 95%CI=0.0868~0.4962)是影响阳性过敏原数多少的因素,男性相对于女性、文化程度低相对于文化程度高的人群阳性过敏原数更多,喜好辛辣食物和常接触宠物以及毛絮物品的人群是阳性过敏原数增加的危险因素。结论分析零膨胀数据时模型的选择至关重要,合适的统计模型有助于提高数据的拟合效果,使分析结果更全面可靠;过敏患者可以是单一或多种过敏原呈阳性反应,同时发生多种阳性过敏原更应关注男性、文化程度低、喜好辛辣食物和常接触宠物等人群,从而减少过敏的发生,达到预防过敏性疾病的目的。

Objective To determine the distribution and influencing factors of the number of positive allergen using Zeroinflated generalized Poisson model（ ZIGP）,and provide scientific evidence for preventing allergy and allergic disease. Methods2782 patients＇ information w as collected in hospital. Overdispersion O test,Vuong test and model fitting index value suggested that ZIGP model w as the best regression model to investigate positive allergens data. Results 1969 of 2782（ 70. 78%） patients w erepositive allergens number of 0. Overdispersion（ O = 52. 13,P 0. 001） suggested that response variables are discrete; Vuong test（ V =22. 728 and 4. 648,P〈 0. 05） suggested that zero-inflated model was better than traditional counting model; according to AIC and BIC values,ZIGP model w as determined. High education degree（ OR = 2. 232,OR 95% CI = 1. 2067 - 4. 1286） is prone to appear positive allergens; Gender（ OR = 0. 596,OR 95% CI =- 0. 7703 -- 0. 2663）,education degree（ OR = 0. 869,OR 95% CI =- 0. 2759 -- 0. 0043）,spicy food（ OR = 1. 119,OR 95% CI = 0. 0052 - 0. 2190） and touching pets（ OR = 1. 338,OR 95% CI =0. 0868 - 0. 4962） are factors affecting the number of positive allergens. Compared w ith female and high education degree,those being maleand low education degree w ill be likely to have more positive allergens. Spicy food and touching frequently petsw ere risk factor for increase of positive allergen. Conclusion Model selection for zero-inflated data is significant,appropriate statistical model is helpful to improve the data fitting effect,and make the results more comprehensive and reliable; allergies can be single or multiple allergens positive,the later conditionwas more likely for the male,the lower education degree,favor of spicy food and touching frequently pets. To pay more attention to these risk factors will reduce the occurrence of allergic and prevent allergic disease.