AimsUnderstanding the pattern of disease progression in chronic heart failure (HF) may inform patient care and healthcare system design. We used a four‐state Markov model to describe the disease trajectory of patients with HF.Methods and resultsConsecutive patients (n = 4918) were enrolled (median age 75 [67–81] years, 61.3% men, 44% with HF and reduced ejection fraction). We generated a model by observing events during the first 2 years of follow‐up. The model yielded surprisingly accurate predictions of how a population with HF will behave during subsequent years. As examples, the predicted transition probability from hospitalization to death was 0.11; the observed probabilities were 0.13, 0.14, and 0.16 at 3, 4, and 5 years, respectively. Similarly, the predicted transition intensity for rehospitalization was 0.35; the observed probabilities were 0.38, 0.34, and 0.35 at 3, 4, and 5 years, respectively. A multivariable model including covariates thought to influence outcome did not improve accuracy. Predicted average life expectancy was approximately 10 years for the unadjusted model and 13 years for the multivariable model, consistent with the observed mortality of 41% at 5 years.ConclusionsA multistate Markov chain model for patients with chronic HF suggests that the proportion of patients transitioning each year from a given state to another remains constant. This finding suggests that the course of HF at a population level is more linear than is commonly supposed and predictable based on current patient status.

中文翻译：

---

慢性心力衰竭的疾病进展是线性的：多状态模型的见解


目的了解慢性心力衰竭 (HF) 的疾病进展模式可以为患者护理和医疗保健系统设计提供信息。我们使用四状态马尔可夫模型来描述心力衰竭患者的疾病轨迹。方法和结果连续患者（ n = 4918 人）入组（中位年龄 75 [67-81] 岁，61.3% 为男性，44% 患有心力衰竭且射血分数降低）。我们通过观察随访前两年的事件生成了一个模型。该模型对心力衰竭人群在接下来的几年中的行为做出了令人惊讶的准确预测。例如，从住院到死亡的预测转移概率为 0.11； 3 年、4 年和 5 年观察到的概率分别为 0.13、0.14 和 0.16。同样，再住院的预测转变强度为 0.35； 3 年、4 年和 5 年观察到的概率分别为 0.38、0.34 和 0.35。包含被认为影响结果的协变量的多变量模型并没有提高准确性。未经调整的模型预测的平均预期寿命约为 10 年，多变量模型的预测平均预期寿命约为 13 年，与观察到的 5 年死亡率 41% 一致。从一个给定状态到另一个状态的年份保持不变。这一发现表明，在人群水平上，心力衰竭的病程比通常假设的更加线性，并且可以根据当前患者状况进行预测。