聚类复发事件加性模型的分段常数化估计

Estimation of an Additive Rate Model for Clustered Recurrent Events Based on the Piecewise Constant Method

摘要: 研究大型的医疗数据具有重要意义，但大量数据会导致计算复杂度变高，尤其涉及到反复发生的疾病数据时。本文对大型数据的聚类复发事件提出了一个半参数边际加性比率函数模型。在进行参数估计时，随着数据量级的增加而导致未知参数增加，为了提高计算效率，对基准比率函数进行分段常数化。利用估计方程方法给出了模型参数的估计量的表达式。并证明估计量满足相合性和渐近正态性等大样本性质。通过数值模拟，验证了提出的估计方法，模拟结果表明对参数部分的估计结果较好。与对基准比率函数未做分段常数化的方法进行比较，发现分段常数化的方法偏差更小，而且计算时间显著缩短。最后将模型和方法应用到慢性肉芽肿病的数据中，将病人按照医院进行分类，找到影响各医院病人病情复发的显著因素和因素的影响方式。

Abstract: Studying large medical databases is important. However, the use of large databases may introduce computational difficulties, particularly when the event of interest is recurrent. A semiparametric marginal additive rate function model is proposed for large databases clustered recurrence events. In the parameter estimation, the number of baseline rate functions increases as the number of classes increases, leading to an increase in the parameters. The baseline rate functions were re-garded as piecewise constant during the estimation, which improved the computational efficiency. The expressions of the estimate of the model parameters were given using the estimating equation method. The estimators were proved consistent and asymptotic normality. The proposed estima-tion method is verified by numerical simulation, and the simulation results show that the estima-tion results of the parameter part are good, compared with the method without piecewise constant of the risk ratio function. It is found that the bias of the piecewise constant method is smaller, and the calculation time is significantly shortened. Finally, the model and method were applied to the data of chronic granulomatous disease, and the patients were classified by hospital to find out the significant factors affecting patients' disease recurrence in each hospital and the influencing ways of the factors.