背景：
高血压作为心血管病的主要危险因素之一，已成为全球公共卫生的重大挑战，其患病率持续上升，尽管防控体系不断完善，但高血压的知晓率、治疗率和控制率仍然较低。在高血压的治疗中，降压目标值设定的相关讨论一直存在，且强化降压策略的随机对照试验（Randomized Controlled Trial, RCT）证据也在持续积累。但其在临床实践中的应用效果仍然存在较大争议。尽管RCT在真实世界中干预效果验证的方法学框架已逐步完善，但策略类研究与常规RCT在干预界定上存在差异，通常依赖于一系列综合措施来确保干预策略的实施，使得真实世界研究中难以直接明确特定的干预策略。
目的：
强化降压策略与常规RCT干预的主要区别在于其以特定的血压控制目标作为干预策略。然而，由于个体药物耐受性、依从性等因素的影响，实际的血压控制往往无法维持在预设的目标范围内，从而导致RCT各组间存在一定程度的血压分布范围重叠，即强化降压策略真实世界效果验证中的决策灰区。在既往的相关研究中，通常基于预设的血压控制目标值进行分组，往往会高估强化或非强化中一组的疗效。因此，本研究旨在通过构建完整的分类方法，将血压值位于决策灰区的个体划归到合适的分组，从而为策略类研究的真实世界效果验证提供更为准确的分组基础和方法学框架。
方法：
本研究旨在验证强化降压策略的真实世界效果，将RCT研究终点显著的随访时刻血压分布作为真实世界中干预策略识别的替代方法，针对血压重叠区域的决策灰区个体共对比分析了四种分类方法的分组效果：分类方法一以预设的血压控制目标为界值，将个体划分为不同组别；分类方法二对决策灰区个体采用均等概率随机分配组别；分类方法三应用多项RCT研究，构建强化与非强化组研究终点时刻血压的概率密度曲线，计算决策灰区个体血压分别在强化和非强化曲线下的概率密度，进而形成双重概率，再分别根据引入随机性和直接按照概率大小的确定性分配来对灰区个体进行分组；分类方法四则以决策灰区内人群为对照，将血压低于和高于灰区界值的人群分别设为处理组，并依据基线特征构建灰区个体划归两侧强化和非强化组的倾向性，并评价是否引入随机性的效果差异。
在模拟研究部分，首先通过多项强化降压RCT分别构建研究场景。根据RCT的主要终点报告数据，提取不同策略组研究对象在主要终点疗效显著的随访时间下，实际收缩压均值、标准差等参数，进而模拟出与各组别收缩压分布特征一致的强化与非强化组模拟数据。随后，通过混合两组数据来构建带有原始组别标识的模拟研究队列，并建立分组判别体系，识别出收缩压位于“决策灰区”的研究人群。最后采用不同的分类方法进行分组，通过各方法预测的分组与原始组别标识的一致性分析，系统评估不同分类方法的分组效果及适用性。
在实证分析部分，拟基于PURE-China的社区人群，分别构建不同的目标人群进行强化降压策略的效果验证，目标人群1为患有高血压且自述服用降压药物的研究对象，目标人群2为仅以高血压作为纳入标准的研究对象，目标人群3则为包含全部社区人群的更广泛群体。通过提取多项强化降压RCT研究主要终点显著的随访时刻数据，估计强化组与非强化组的实际血压控制水平的加权平均值，用于识别血压位于决策灰区的个体。再通过应用不同的分类方法结合Cox共享脆弱模型，系统评估强化降压策略在不同目标人群中的真实世界效果。本研究以非致命性心血管事件、心血管疾病死亡和主要不良心血管复合事件作为主要结局指标。其中非致死性心血管事件包括急性心肌梗死、卒中及心力衰竭在内的复合终点，心血管疾病死亡是指致死性的心血管事件，主要不良心血管复合事件定义为随访期间首次发生的非致死性心血管事件复合结局或致死性心血管事件。
结果：
在模拟研究部分，基于预设血压控制目标的分类方法一具有场景敏感性，其预设的收缩压控制目标与实际强化和非强化组血压概率密度曲线交点之间距离的差异，导致了结果的稳定性较差；分类方法二在所有场景中的分类效果均显著较差；尽管分类方法四在相同场景中的效果显著优于分类方法三，但分类方法四对强化与非强化的组间的人群特征异质性敏感，同样导致结果的稳定性较差。 
在实证分析部分，PURE-China社区人群中共纳入5602例服用降压药物的高血压患者作为目标人群1，18192例高血压患者作为目标人群2，42103例社区人群作为目标人群3。经过中位随访时间10.76年，模拟研究中表现较好且稳定的分类方法三在结合确定性分配策略后，三种目标人群中强化降压策略均能显著降低心血管死亡风险，并且随着高血压治疗人群限定标准的逐步明确，心血管疾病死亡的风险比点估计值也有下降的趋势，在目标人群1、目标人群2和目标人群3中(HR) ̂(95%CI)分别为0.352(0.187-0.665)，0.405(0.253-0.649)和0.424(0.352-0.511)。同样在该方法下，以社区人群为代表的目标人群3中，心血管病和主要心血管病不良事件均有显著获益，(HR) ̂(95%CI)分别为0.690(0.645-0.738)和0.694(0.649-0.742)。

结论：
本研究通过模拟与实证分析评估了四种分类方法在强化降压策略验证中的效果。结果表明，基于概率密度函数的分类方法在模拟研究中表现出稳定且可靠的分类效果，能够有效区分决策灰区的人群分组；再结合实证分析验证了强化降压策略的获益，尤其在社区人群中效果显著。相较之下，其他分类方法的效果较差或敏感性较强。因此，基于概率密度的分类方法是较为稳健的决策灰区人群分组工具。

Backgrounds:

Hypertension, as one of the main risk factors of cardiovascular disease, has become a major challenge to global public health. Its prevalence continues to rise. Although the prevention and control system is constantly improved, the awareness rate, treatment rate and control rate of hypertension are still low. There is ongoing discussion of blood pressure goal setting in the treatment of hypertension, and evidence from RCTs of intensive blood pressure strategies continues to accumulate. However, there is still a great controversy about its application effect in clinical practice. Although the methodological framework of RCT intervention validation in the real world has been gradually improved, there are differences in intervention definition between strategic studies and conventional RCTs, which usually rely on a series of comprehensive measures to ensure the implementation of intervention strategies, making it difficult to directly identify specific intervention strategies in real world studies.

Objectives:

The main difference between intensive antihypertensive strategies and conventional RCT interventions is the use of specific blood pressure control targets. However, due to factors like drug tolerance and compliance, actual blood pressure often deviates from the target, creating overlap in blood pressure ranges among RCT groups and strengthening the decision gray area in real-world verification. Previous real-world validations of intensive strategies often overestimated efficacy by grouping based on preset targets. This study aims to develop a classification method to accurately categorize individuals in the gray zone, providing a more precise framework for real-world effect verification.

Methods:

This study aimed to verify the effect of intensive antihypertensive strategy. Using the blood pressure distribution at the endpoint of RCT study as an alternative method for intervention strategy identification in the real world, the grouping effect of four classification methods was compared for individuals in the decision gray area of blood pressure overlap: Classification method 1 divides individuals into different groups by using preset blood pressure control targets as boundary values; Classification method 3: Using multiple RCT studies, probability density curves of blood pressure at endpoint of intensive and non-intensive groups were constructed, probability densities of individual blood pressure under intensive and non-intensive curves were calculated in decision grey areas, and then double probabilities were formed, and individuals were grouped according to random and deterministic allocation directly according to probability. The fourth classification method takes the population in the gray area of decision making as the control, sets the population whose blood pressure is lower than and higher than the boundary value of the gray area as the treatment group respectively, and constructs the tendency of individuals in the gray area to be classified into the enhanced and non-enhanced groups on both sides according to the baseline characteristics, and also evaluates whether to introduce the effect difference of randomization.

In the simulation part, scenarios were first constructed using multiple intensive hypotensive RCTs. According to the primary endpoint report data of RCT, parameters such as mean and standard deviation of actual systolic blood pressure of study subjects in different strategy groups at the follow-up time with significant efficacy of primary endpoint were extracted, and then simulated data of intensive and non-intensive groups consistent with the systolic blood pressure distribution characteristics of each group were simulated. Then, by mixing the two sets of data to construct a simulated study cohort with the original group identification, and establish a group discrimination system to identify the study population with systolic blood pressure in the "decision gray zone". Finally, different classification methods are used to classify the groups. The classification effect and applicability of different classification methods are systematically evaluated by analyzing the consistency between the predicted groups and the original group identification.

In the empirical analysis part, it is proposed to construct different target populations based on PURE-China community population to verify the effect of intensive antihypertensive strategy. Target population 1 is the study subjects with hypertension and self-reported antihypertensive drugs, target population 2 is the study subjects with hypertension as inclusion criteria, and target population 3 is a broader group including all community populations. By extracting follow-up data at significant times for the primary endpoint of multiple intensive antihypertensive RCTs, the actual level of blood pressure control in the intensive and non-intensive groups was estimated to identify populations with blood pressure in the decision gray zone. The real-world effects of intensive antihypertensive strategies in different target populations were systematically evaluated by applying different classification methods and Cox shared vulnerability models. Nonfatal cardiovascular events, cardiovascular death, and major adverse cardiovascular composite events were used as primary outcome measures. Among them, non-fatal cardiovascular events include composite endpoints including acute myocardial infarction, stroke and heart failure. Cardiovascular death refers to fatal cardiovascular events. Major adverse cardiovascular events are defined as composite outcomes of non-fatal cardiovascular events or fatal cardiovascular events occurring for the first time during follow-up period.

Results:

In the simulation part, classification method I based on preset BP control target has scene sensitivity, and the difference between the distance between the preset systolic BP control target and the intersection point of the actual BP probability density curve of the intensive group and the non-intensive group may lead to instability of the results; Although classification method 4 is significantly better than classification method 3 in the same scene, classification method 4 is sensitive to heterogeneity of population characteristics between reinforced and unreinforced groups, which may also lead to unstable results.

In the empirical analysis part, PURE-China community population included 5602 hypertension patients taking antihypertensive drugs as target population 1, 18192 hypertension patients as target population 2, and 42103 community population as target population 3. After a median follow-up of 10.76 years, the classification method 3, which performed well and stably in the simulation study, significantly reduced the risk of cardiovascular death in all three target populations after combining with deterministic allocation strategy, and with the gradual clarification of the definition criteria for hypertension treatment population, the risk ratio point estimate of cardiovascular death also showed a downward trend. (95%CI) in target population 1, 2 and 3 were 0.352(0.187-0.665), 0.405(0.253-0.649) and 0.424(0.352-0.511), respectively. Under this method, cardiovascular disease and major cardiovascular adverse events were significantly benefited in the target population 3 represented by community population, with (95% CI) of 0.690(0.645-0.738) and 0.694(0.649-0.742), respectively.

Conclusions:

This study evaluated the effectiveness of four classification methods in the validation of intensive antihypertensive strategies through simulation and empirical analysis. The results show that the classification method based on probability density function shows stable and reliable classification effect in simulation study, and can effectively distinguish groups in grey decision area; combined with empirical analysis, it also verifies the benefit of strengthening blood pressure reduction strategy, especially in community population. In contrast, other classification methods are less effective or more sensitive. Therefore, the classification method based on probability density is a more robust decision-making tool for grey area population grouping.