Colais Paola, Di Martino Mirko, Fusco Danilo, Davoli Marina, Aylin Paul, Perucci Carlo Alberto
Department of Epidemiology, Regional Health Service, Lazio Region, Via Santa Costanza 53, Rome, 00198, Italy.
BMC Health Serv Res. 2014 Oct 23;14:495. doi: 10.1186/s12913-014-0495-3.
Hospital discharge records are an essential source of information when comparing health outcomes among hospitals; however, they contain limited information on acute clinical conditions. Doubts remain as to whether the addition of clinical and drug consumption information would improve the prediction of health outcomes and reduce confounding in inter-hospital comparisons. The objective of the study is to compare the performance of two multivariate risk adjustment models, with and without clinical data and drug prescription information, in terms of their capability to a) predict short-term outcome rates and b) compare hospitals' risk-adjusted outcome rates using two risk-adjustment procedures.
Observational, retrospective study based on hospital data collected at the regional level.Two cohorts of patients discharged in 2010 from hospitals located in the Lazio Region, Italy: acute myocardial infarction (AMI) and hip fracture (HF). Multivariate logistic regression models were implemented to predict 30-day mortality (AMI) or 48-hour surgery (HF), adjusting for demographic characteristics and comorbidities plus clinical data and drug prescription information. Risk-adjusted outcome rates were derived at the hospital level.
The addition of clinical data and drug prescription information improved the capability of the models to predict the study outcomes for the two conditions investigated. The discriminatory power of the AMI model increases when the clinical data and drug prescription information are included (c-statistic increases from 0.761 to 0.797); for the HF model the increase was more slight (c-statistic increases from 0.555 to 0.574). Some differences were observed between the hospital-adjusted proportion estimated using the two different models. However, the estimated hospital outcome rates were weakly affected by the introduction of clinical data and drug prescription information.
The results show that the available clinical variables and drug prescription information were important complements to the hospital discharge data for characterising the acute severity of the patients. However, when these variables were used for adjustment purposes their contribution was negligible. This conclusion might not apply at other locations, in other time periods and for other health conditions if there is heterogeneity in the clinical conditions between hospitals.
在比较医院间的健康结局时，医院出院记录是重要的信息来源；然而，它们包含的急性临床状况信息有限。对于添加临床和药物消费信息是否会改善健康结局预测并减少医院间比较中的混杂因素，仍存在疑问。本研究的目的是比较两个多变量风险调整模型的性能，一个包含临床数据和药物处方信息，另一个不包含，比较它们在以下方面的能力：a）预测短期结局发生率；b）使用两种风险调整程序比较医院的风险调整后结局发生率。
基于在区域层面收集的医院数据进行观察性、回顾性研究。2010年从意大利拉齐奥地区医院出院的两组患者：急性心肌梗死（AMI）和髋部骨折（HF）。实施多变量逻辑回归模型来预测30天死亡率（AMI）或48小时内手术率（HF），并对人口统计学特征、合并症以及临床数据和药物处方信息进行调整。在医院层面得出风险调整后的结局发生率。
添加临床数据和药物处方信息提高了模型预测所研究的两种疾病结局的能力。当纳入临床数据和药物处方信息时，AMI模型的辨别力增加（c统计量从0.761增加到0.797）；对于HF模型，增加幅度较小（c统计量从0.555增加到0.574）。使用两种不同模型估计的医院调整比例之间存在一些差异。然而，引入临床数据和药物处方信息对估计的医院结局发生率影响较小。
结果表明，可用的临床变量和药物处方信息是医院出院数据的重要补充，有助于描述患者的急性严重程度。然而，当将这些变量用于调整目的时，它们的贡献微不足道。如果医院之间的临床状况存在异质性，这一结论可能不适用于其他地点、其他时间段和其他健康状况。
