Baneshi Mohammad Reza, Mishra Gita, Dobson Annette
School of Public Health, The University of Queensland Faculty of Medicine, Herston, Queensland, Australia.
BMJ Public Health. 2024 Nov 29;2(2):e000963. doi: 10.1136/bmjph-2024-000963. eCollection 2024 Dec.
Studies investigating the relationship between patterns of multimorbidity and risk of a new condition have typically defined the patterns at a baseline time and used Kaplan-Meier (KM) or Cox proportional hazards regression. These methods do not consider the competing risk of death or the changes in the patterns of conditions over time. This study illustrates how these methodological limitations can be overcome in the setting of progression from cardiometabolic conditions to dementia.
Data from 11 930 women who participated in the Australian Longitudinal Study on Women's Health were used to define patterns of diabetes, heart disease and stroke and estimate the cumulative incidence or HRs of subsequent dementia. Seven methods were compared. For cumulative incidence these were KM method, cumulative incidence function (CIF) (to account for the competing risk of death) and multistate model with Aalen-Johansen estimates (to account also for the progression of conditions over time). For HRs, the corresponding methods were Cox model and Fine and Gray model (for sub-HRs) with the cardiometabolic patterns treated as time-invariant (from baseline) or as time-varying predictors.
The estimated cumulative incidence of dementia using the KM method declined when the competing risk of death was considered. For example, for women with no cardiometabolic condition at baseline, the KM and CIF estimates were 35.7% (95% CI 34.6%, 36.8%) and 27.3% (26.4%, 28.2%) but these women may have developed cardiometabolic conditions during the study which would increase their risk. The Aalen-Johansen multistate estimate for women with no cardiometabolic condition over the whole study period was 11.0% (10.4%, 11.7%). Comparing models to estimate HRs, the estimates in the Fine and Gray models were lower than those in the Cox models.
Multistate and time-varying survival analysis models should be used to study the natural development of multimorbidity.
研究多种疾病共存模式与新疾病风险之间关系的研究通常在基线时间定义这些模式，并使用Kaplan-Meier（KM）或Cox比例风险回归。这些方法没有考虑死亡的竞争风险或疾病模式随时间的变化。本研究说明了在从心脏代谢疾病发展为痴呆症的背景下，如何克服这些方法学上的局限性。
来自11930名参与澳大利亚女性健康纵向研究的女性的数据被用于定义糖尿病、心脏病和中风的模式，并估计随后患痴呆症的累积发病率或风险比（HRs）。比较了七种方法。对于累积发病率，这些方法是KM法、累积发病率函数（CIF）（用于考虑死亡的竞争风险）和具有Aalen-Johansen估计值的多状态模型（也用于考虑疾病随时间的进展）。对于风险比，相应的方法是Cox模型和Fine and Gray模型（用于亚风险比），将心脏代谢模式视为时间不变（从基线开始）或随时间变化的预测因子。
当考虑死亡的竞争风险时，使用KM法估计的痴呆症累积发病率下降。例如，对于基线时没有心脏代谢疾病的女性，KM和CIF估计值分别为35.7%（95%CI 34.6%，36.8%）和27.3%（26.4%，28.2%），但这些女性在研究期间可能发展出了心脏代谢疾病，这会增加她们的风险。在整个研究期间，对于没有心脏代谢疾病的女性，Aalen-Johansen多状态估计值为11.0%（10.4%，11.7%）。比较估计风险比的模型，Fine and Gray模型中的估计值低于Cox模型中的估计值。
应使用多状态和随时间变化的生存分析模型来研究多种疾病共存的自然发展过程。
