Hardoon Sarah L, Lewsey James D, van der Meulen Jan H P
Clinical Effectiveness Unit, The Royal College of Surgeons of England, 35-43 Lincoln's Inn Fields, London WC2A 3PE, UK.
Stat Med. 2007 Dec 10;26(28):5081-99. doi: 10.1002/sim.2900.
The CUSUM continuous monitoring method could be a valuable tool in evaluating the performance (revision experience) of prostheses used in hip replacement surgery. The dilemma when applying the CUSUM in this context is the choice of statistical model for the outcome (revision). The Bernoulli model is perhaps the most straightforward approach but the Poisson model is a plausible, and could be argued, preferable alternative for long-term outcomes such as this, provided the rate of revision with time from surgery can be assumed to be constant. However, a rate (or hazard) varying according to the Weibull distribution appears to be a better representation of a prosthesis lifetime. We show how to adapt the Poisson approach to allow for the hazard to vary according to the Weibull model as well as other parametric survival models. Application to data on a known poorly performing prosthesis shows both the Poisson and Weibull CUSUMs could have given early warning of the poor performance, with the Weibull chart alerting before the Poisson. Simulation work to investigate the robustness of the Poisson and Weibull CUSUM to departures from the underlying survival model highlights the need for correct specification of the model for the outcome.
累积和（CUSUM）连续监测方法可能是评估髋关节置换手术中使用的假体性能（翻修经验）的一个有价值的工具。在这种情况下应用CUSUM时的困境在于选择用于结果（翻修）的统计模型。伯努利模型可能是最直接的方法，但泊松模型对于像这样的长期结果是一种合理的、甚至可以说是更优的替代方法，前提是可以假定从手术开始随时间的翻修率是恒定的。然而，根据威布尔分布变化的速率（或风险率）似乎能更好地描述假体寿命。我们展示了如何调整泊松方法，使其能考虑风险率根据威布尔模型以及其他参数生存模型而变化的情况。应用于已知性能不佳的假体数据表明，泊松累积和与威布尔累积和都能对性能不佳给出早期预警，威布尔图表比泊松图表更早发出警报。研究泊松累积和与威布尔累积和对偏离基础生存模型的稳健性的模拟工作突出了正确设定结果模型的必要性。
