【作者】 孙宏鹏；

【导师】 刘美娜；

【作者基本信息】 哈尔滨医科大学 ， 流行病与卫生统计学， 2012， 博士

【摘要】 目的：急性心肌梗死是一种高发病率和高病死率的疾病，已经严重危害中国人的健康和生命。提高该病的治疗质量水平是减少疾病负担和经济负担最重要的手段之一，但目前中国还没有完整的治疗质量评价机制。本研究结合中国的医疗环境，针对质量评价中能遇到的问题，对整个评价过程进行充分研究。建立急性心肌梗死疾病治疗质量评价指标体系，对每个指标进行详细的定义，分析治疗过程之间、治疗过程与结局之间相关性，计算医院间公平比较的标准化指标使用率，探讨基于贝叶斯估计的多水平潜变量模型在综合评价中的应用。方法：查阅相关文献和急性心肌梗死治疗指南，制定候选指标；利用Delphi专家评分筛选指标，确定最终评价指标；收集20家综合医院的临床数据，利用风险调整的思想，使用多水平logistic回归模型计算指标的风险调整使用率；对多个指标加权求和，计算医院综合质量指数；使用贝叶斯多水平潜变量模型进行综合评价：假定医院的各观测指标的使用率由一个潜变量决定，利用潜变量和各指标的关系构建潜变量模型，在模型中加入随机效应，使用贝叶斯推断估计模型的后验参数，计算急性心肌梗死综合质量指数。结果：主要研究结果如下：共查阅并制定85个急性心肌梗死指标，经第一轮专家咨询后，确定60个候选指标。经过专家评分，有23个指标进入下一轮筛选。最后，在专家面对面讨论后，第二轮的23个指标具有很好的可靠性和可行性，故都纳入指标体系中，共包含3个医院结构指标，15个治疗过程指标，5个病人结局指标。20家三级甲等医院共收集急性心肌梗死患者2203名，可以实际获得数值的有10过程指标和1个结局指标，氯吡格雷和他汀类药物有较高的使用率，溶栓药物和冠脉支架使用率较低；医院的治疗过程指标的使用率和院内病死率都有较大的差异，经过风险调整后，院内病死率最高为11.62%，最低为4.54%；不同的治疗方式之间有相关性，阿司匹林与β受体阻滞剂（r=0.61）、阿司匹林与ACEI（r=0.50）、氯吡格雷与溶栓（r=-0.53）、氯吡格雷与PCI（r=0.47）、氯吡格雷与造影（r=0.45）、他汀类药物与造影（r=0.50）、PCI与造影（r=0.91）的相关性有统计学意义；加权的综合指数显示：第3家医院的治疗质量最好，第17家医院的治疗质量最差。选9个指标进行综合评价，贝叶斯多水平潜变量模型有更好的收敛诊断标准。模型中参数后验分布显示：急性心肌梗死的院内死亡率和溶栓与治疗质量负相关，其余指标与治疗质量为正相关；9个指标的后验回归系数95%可信区间显示：阿司匹林、β受体阻滞剂和ACEI与治疗质量的相关性没有统计学意义，冠脉支架、冠脉造影、溶栓药物、氯吡格雷和他汀类药物与治疗质量的相关性有统计学意义，其中冠脉支架和冠脉造影与治疗质量之间的相关性最大；阿司匹林、β受体阻滞剂、ACEI和他汀类药物4个指标有较大的随机效应；利用模型计算的治疗质量指数的95%可信区间将医院分为三级，编号为3、19和16的医院高于平均质量，编号为2、10和17的医院低于平均质量。结论：本研究建立的急性心肌梗死治疗质量评价指标体系包含有23个指标，实践表明有11个指标可以通过医院病案数据收集；医院间治疗过程指标的使用率和院内病死率有较大的差异，风险调整的方法能够减少由混杂因素导致的差异，使医院间比较更为公平；考虑了随机效应的贝叶斯多水平潜变量模型更适合于医院治疗质量综合评价，不但提供一个可以将多指标合成为一个综合指标的统计学方法，而且能对综合指数和综合指数排序做出统计推断，更好的解释了医院间指标的变异来源。综上，多水平贝叶斯潜变量模型更适合用于治疗质量的综合评价，该方法为综合质量指数的计算奠定了统计理论基础。更多还原

【Abstract】 Objective: Acute myocardial infarction has a high incidence and fatality rate,and has serious damage to health and life. Improve quality of care is one of the mostimportant means reduce the burden of disease and the economic burden of, but thereis not complete quality evaluation system in China. This research is combined withChinese medical environment, to solve the problems in the quality evaluation, studythe whole evaluation process. We try to build up a set of quality indicators for patientswith acute myocardial infarction, to define each indicator in detailed, analysiscorrelation between care processes and with the patient outcome, to calculate hospitalstandardized rates for fair comparison, study a composite measure of hospital qualitybased on a bayesian hierarchical latent variable model (BLVM).Methods: A literature review practice guidelines of acute myocardial infarctionidentified existing quality indicators for AMI care. A list of potential indicators wasassessed by a panel of clinicians from a variety of disciplines using amodified-Delphipanel process, form the final indicators; Collected clinical data form20hospitals, and,used multi-level logistic regression model to calculate risk adjusted rate by using theidea of risk adjustment; Calculated composite quality index by weight sum;Comprehensive evaluation used the bayesian hierarchical latent variable model:assume a latent variable that decide the hospital the utilization rate of each indicators,the latent variables is unobservable quality of care, build on a latent variables modelwith random effects, the use of bayesian inference estimates posterior parameters,estimated a composite measure of hospital quality based on BLVM.Results: The main research results as follows: Eighty-five potential indicators for AMI care were established. In the first roundof consultation,25indicators were deleted; the rest of the60were reviewed byexperts. The second round of consultation, experts rated each indicator according tothe following six criteria; there are23indicators into the next round process. The thirdround of consultation, after the expert face-to-face discussion, the second round of23indicators had good reliability and feasibility, including three structure indicators,fifteen process indicators and five outcome indicators.We collected2203patients with AMI from20first-class hospitals, can actuallyget data in10process indicators and1outcame indicators. Clopidogrel and statinshave higher utilization rates, thrombolysis drug and percutaneous coronaryintervention had high-usage. Rates for care process indicators between differenthospitals had a great difference; the mortality between hospitals also had a variance,risk-adjusted mortality from4.54%to11.62%,51%of the variation was from thequality of care,49%of the variation was caused by mixed factors; there is acorrelation between different indicators, aspirin and beta blockers (r=0.61), aspirinand angiotensin-converting enzyme inhibitors (ACEI)(r=0.50), clopidogrel andthrombolysis (r=0.53), clopidogrel and percutaneous coronary intervention (PCI)(r=0.47), clopidogrel and coronarography (r=0.45), statins and coronarography (r=0.50), PCI and coronarography (r=0.91) are significant; Weighted composite measureshows: the third hospital had best quality of care best, the17th hospital was the worst.We chose9indicators to build overall evaluation model. The compare betweenmodels showed that BLVM had better convergence than other models. The posteriordistribution of parameters in the model indicated that all the indicators had positivecorrelation with quality of care besides in-hospital mortality and thrombolytic.95%confidence interval (CI) of9indicators’ posterior regression coefficient showed thatthe relationship between PCI, coronarography, thrombolytic, clopidogrel, statins andquality of care was statistically significant, and the correlation between PCI, coronarography and quality of care was stronger; aspirin, beta blockers, ACEI andstatins had bigger random effect. All the hospitals were divided into three groupsbased on the95%CI of quality of care index which was calculated through BLVM.The quality of care of hospital3,19and16was higher than average, in contrast,hospital2,10and17was lower than average.Conclusion: We have developed a set of quality indicators for patients with AMI,including23indicators. It had been demonstrated that11of23indicators can becollected through medical record. There was large variation on the use of processindicators and in-hospital mortality between hospitals. The method of risk-adjustmentprovides equity to compare hospitals due to adjusting confounding factors. BLVMallowed for random effect is more suitable for overall evaluation of quality ofhealthcare. It not only provided statistical theoretical framework to integratemulti-dimension indicators into a synthetic indicator, but also inferred compositeindexes and their ranks and accounted for the sources of hospitals variation. Inconclusion, BLVM is more suitable for overall evaluation of quality of healthcare, andit lays statistical theory foundation for the calculation of composite index.更多还原