Section for Medical Statistics, Center for Medical Statistics, Informatics and Intelligent Systems, Medical University Vienna, A-1090 Vienna, Austria.
Division Cardiac-, Thoracic-, Vascular Anaesthesia and Intensive Care, Medical University Vienna, A-1090 Vienna, Austria.
Clin Nutr. 2016 Apr;35(2):522-527. doi: 10.1016/j.clnu.2015.03.019. Epub 2015 Apr 7.
BACKGROUND & AIMS: A major problem occurring in cross-sectional studies is sampling bias. Length of hospital stay (LOS) differs strongly between patients and causes a length bias as patients with longer LOS are more likely to be included and are therefore overrepresented in this type of study. To adjust for the length bias higher weights are allocated to patients with shorter LOS. We determined the effect of length-bias adjustment in two independent populations.
Length-bias correction is applied to the data of the nutritionDay project, a one-day multinational cross-sectional audit capturing data on disease and nutrition of patients admitted to hospital wards with right-censoring after 30 days follow-up. We applied the weighting method for estimating the distribution function of patient baseline variables based on the method of non-parametric maximum likelihood. Results are validated using data from all patients admitted to the General Hospital of Vienna between 2005 and 2009, where the distribution of LOS can be assumed to be known. Additionally, a simplified calculation scheme for estimating the adjusted distribution function of LOS is demonstrated on a small patient example.
The crude median (lower quartile; upper quartile) LOS in the cross-sectional sample was 14 (8; 24) and decreased to 7 (4; 12) when adjusted. Hence, adjustment for length bias in cross-sectional studies is essential to get appropriate estimates.
横断面研究中存在一个主要问题，即抽样偏倚。住院时间（LOS）在患者之间存在很大差异，导致长度偏倚，因为 LOS 较长的患者更有可能被纳入研究，因此在这种类型的研究中过度代表。为了调整长度偏倚，对 LOS 较短的患者分配更高的权重。我们在两个独立的人群中确定了长度偏倚调整的效果。
长度偏倚校正应用于营养日项目的数据中，该项目是一项为期一天的多国家横断面审计，对接受住院病房治疗的患者的疾病和营养数据进行了采集，随访 30 天后进行右删失。我们应用了一种基于非参数最大似然法的估计患者基线变量分布函数的加权方法。使用 2005 年至 2009 年期间维也纳总医院所有入院患者的数据验证了结果，其中 LOS 的分布可以假定是已知的。此外，还在一个小患者示例上演示了估计 LOS 调整分布函数的简化计算方案。
横断面样本中 LOS 的粗中位数（下四分位数；上四分位数）为 14（8；24），调整后降至 7（4；12）。因此，在横断面研究中调整长度偏倚对于获得适当的估计值至关重要。
