The study cohort (Dynamic ECG & Blood Pressure for Hypertension at Sanxiang) consisted of managed care hypertension cases within the primary care community of Zhongshan City, Guangdong Province, China, from November 1, 2022, to January 1, 2025 [33]. The inclusion criteria for this study were based on meeting the diagnostic criteria for hypertension in the Chinese Hypertension Prevention and Treatment Guidelines (2024 Revised Edition) [34]: SBP ≥140 mm Hg and/or DBP ≥90 mm Hg measured on 3 non–consecutive days within a week. Patients with hypertension who were included in the management of the National Basic Public Health Service Programme were included in the study. Inclusion criteria included baseline age ≥35 years; clear consciousness, unimpeded communication, and no cognitive impairment; and completion of 4 ABPM sessions (at baseline, 3 months, 6 months, and 9 months after baseline; with 3-month intervals). Exclusion criteria were patients with secondary hypertension (such as renovascular hypertension, endocrine-induced hypertension, etc); individuals with gestational hypertension; those complicated with severe acute or chronic diseases (eg, malignant tumors, end-stage renal disease, and acute decompensated heart failure) or severe physical disabilities; those with incomplete baseline data (eg, missing key variables) or inability to cooperate with follow-up; and those who withdrew voluntarily or were lost to follow-up.

To ensure data quality for the managed patients with hypertension, multiple strategies were implemented. All follow-up staff were professionally trained and certified. They strictly adhered to standardized operating procedures during data collection to ensure the consistency and objectivity of raw data.

During data cleaning and validation, abnormal values were defined as clinically implausible data (eg, SBP <80 or >220 mm Hg) or statistical outliers beyond the mean (SD 3). These values were first cross-checked against original records, such as monitoring device logs and paper questionnaires. Confirmed measurement errors were corrected, while unverifiable values were marked as missing.

Missing data were subsequently handled based on variable criticality. Participants missing ≥20% of critical measurements were excluded according to predefined criteria. For noncritical variables and partially missing critical variables (eg, single-time RWT), multiple imputation (MI) was used to generate 5 complete datasets, and the results were pooled for analysis.

To guarantee standardization and accuracy, the processes of data cleaning, merging, and handling missing values were independently repeated twice and cross-compared. Additionally, research variables were classified and assigned values based on high-quality literature. Furthermore, a complete-case analysis (CCA), strictly excluding participants with any missing covariate data, was performed as a sensitivity analysis to verify the robustness of the primary imputed findings.

The study was approved by the Biomedical Research Ethics Review Committee of the School of Public Health, Sun Yat-sen University (approval number 2021-No.081). Written informed consent was obtained from all participants prior to enrollment. All study data were deidentified before analysis to ensure confidentiality. Participants received complimentary 24-hour ABPM and echocardiography assessments as compensation for their involvement in this research. No images or supplementary materials containing identifiable information are included in this manuscript.

The relevant parameters during the fitting trajectory are shown in Table 1. Since the proportion of group members in the group number of 5 was less than 5%, the optimal number of groups for the trajectory model was chosen as 4. Then the number of longitudinal trajectory groups in patients with hypertension was determined to be 4, the order was adjusted, and the order of the final trajectory curve was determined by combining the morphology of the trajectory curves in the graphs and the P value, which were all determined to be 0th order (straight lines). All trajectory groups satisfied the following criteria [25]: absolute values of Bayesian Information Criterion and Akaike Information Criterion are minimum, proportion of group members >5%, entropy >0.80, and average posterior probability >0.90, indicating excellent model classification accuracy.

As described in Table 2, the 4 trajectory groups have the following characteristics: group 1: BP (123/68 mm Hg) and HR (67 bpm) were both at the lowest levels, but a diastolic pressure below 70 mm Hg may lead to inadequate organ perfusion (eg, myocardial ischemia and cerebral ischemia), particularly in older adult patients or those with concomitant coronary artery disease or carotid artery stenosis [45]; group 2: BP was well controlled (128/77 mm Hg), but the persistently elevated HR (80 bpm) suggested increased sympathetic nervous system activity, which may increase cardiovascular risk; group 3: BP was within the stage 1 hypertension range (134/82 mm Hg), with a normal HR (70 bpm); and group 4: BP was within the stage 2 hypertension range (142/89 mm Hg), with high HR (84 bpm), indicating a decompensated state and the highest cardiovascular risk. The median pulse pressure across the 4 study groups ranged from 51 to 55 mm Hg. Although these group-level mean values were below the 60 mm Hg risk threshold specified by the European Society of Hypertension [45], this macro-level observation does not rule out the presence of elevated pulse pressure-related risk in individual patients.

The individual trajectories of BP and HR in 213 patients with hypertension (Figures 1A-1C) showed strong alignment with the joint trajectories identified by the GBMTM (Figures 2A-2C). Moreover, the empirical distributions of BP and HR at 9 months after baseline closely matched the model-predicted values for each trajectory group, confirming that the GBMTM reliably identified subgroups exhibiting distinct hemodynamic progression patterns. As illustrated in Figure 3, the 3D distribution of joint trajectories quantitatively distinguishes the spatiotemporal differentiation among the 4 trajectory groups within the multidimensional parameter space. This representation offers clinically valuable insights into the heterogeneous dynamic profiles across these physiological dimensions.

The baseline characteristics of the study population are shown in Table 3. A total of 213 patients with hypertension were included in the study, 110 (51.64%) males and 103 (48.36%) females; their age ranged from 35 to 80 years, with a median age of 57 (IQR 52‐65) years. The results of the Kruskal-Wallis H test indicated that the 4 factors of trajectory group, wake-up time, dietary pattern, and history of diabetes mellitus met the prespecified threshold for inclusion in the multivariable model (P<.10).

A multiple linear regression analysis was performed to model ΔRWT using the independent variables identified via univariate analysis. The corresponding results are presented in Table 4. Trajectory group, wake-up time, dietary pattern, and history of diabetes were converted to dummy variables, with group 4, wake-up time after 7 AM, balanced diet, and no diabetes set as reference categories. Multicollinearity diagnostics showed all VIFs <2.0.

Results indicated that trajectory group 2 (b=−0.059, 95% CI −0.112 to −0.007) and diabetes with medications (b=−0.067, 95% CI −0.116 to −0.018) were independently associated with a reduction in ΔRWT. Conversely, a wake-up time before 7 AM (b=0.045, 95% CI 0.007-0.083) was associated with an increase in ΔRWT (Table 4). The fully adjusted multiple linear regression model demonstrated a significant overall fit (F₈₂₀₄=3.392; P=.001), explaining 12.7% of the variance in ΔRWT (R²=0.127, adjusted R²=0.089). This indicates that group 2, medication history, and wake-up time are independently associated with ΔRWT, although the overall variance explained by the model remains modest.

To address potential variable selection and imputation biases, sensitivity analyses were performed by forcing age and sex into the regression models. In the sensitivity analysis using the fully adjusted model (Table 5), the association for group 2 remained statistically significant (P=.04). Furthermore, antidiabetic medication use (P=.009) and waking before 7 AM (P=.01) remained independently associated with ΔRWT. In this imputed model, age (P=.09) and sex (P=.66) did not reach statistical significance.

To further verify the robustness of these findings, a CCA was conducted (Table 6). In this model, the association for group 2 was attenuated to marginal significance (P=.06). This shift is primarily attributable to the reduction in statistical power when cases with missing data were excluded (n=196). Meanwhile, early waking (P=.004) and antidiabetic medication use (P=.01) robustly maintained their statistical associations. Furthermore, age emerged as an independent factor associated with ΔRWT (P=.047), whereas sex remained nonsignificant (P=.82).

To investigate differences in the use of antihypertensive medications across trajectory groups, this study categorized patients into 5 groups based on their medication regimens over a 9-month period: ARBs, CCBs, β-blockers, combination therapy (≥2 medications), and unrecorded medications. Given the elevated HR in group 2 and the potential chronotropic effects of CCBs, we focused on comparing CCB usage rates across the 4 trajectory groups. Chi-square analysis revealed no statistically significant differences in overall antihypertensive drug distribution (χ²₉=8.22; P=.51) or CCB usage rates (χ²₃=3.30; P=.35) across groups (Table 7). Notably, group 4 demonstrated the highest CCB adoption rate (85.7%, predominantly amlodipine), exceeding group 2’s rate (78.7%) by 7.0 percentage points, although this difference did not reach statistical significance. Amlodipine is primarily indicated for patients requiring BP reduction without sympathetic activation or metabolic abnormalities, and its HR-lowering effect is weaker than that of specific CCBs such as cilnidipine [46]. Furthermore, the VALUE trial demonstrated that amlodipine may induce reflex sympathetic activation via peripheral vasodilation, leading to a slight HR increase [47]. While this mechanism supports BP stability, it compromises effective HR control. The prevalent use of amlodipine in Zhongshan’s 3 communities may contribute to sustained HR ≥80 bpm in these patients.

In this community-based cohort, GBMTM identified 4 distinct joint trajectory groups of BP and HR. Regarding the clinical implications, group 2 (characterized by normal BP and high HR) was significantly associated with a reduction in ΔRWT over 6 months, suggesting a potential improvement in LV geometry. BP in this group was well controlled and below the current guideline-recommended standard for hypertension of 130/80 mm Hg [45,48,49].

Furthermore, the use of antidiabetic medications in patients with hypertension with comorbid diabetes was significantly associated with a reduction in ΔRWT. This suggests that ameliorating insulin resistance may inhibit myocardial fibrosis and facilitate the repair of microvascular dysfunction [50-52], thereby providing evidence for the potential beneficial structural associations of these medications. Conversely, a wake-up time before 7 AM was significantly associated with an increase in ΔRWT, indicating that circadian dysregulation (manifested as early awakening) may elevate the risk of hypertensive cardiac remodeling by disrupting metabolic, endocrine, and BP regulation [42,53,54].

In this study, unlike previous studies focusing solely on BP trajectories, we applied GBMTM to identify 4 distinct joint trajectories of 24-hour mean SBP, DBP, and HR in patients with hypertension. Previous studies have used the product of BP and HR (BP × HR), known as the double product, as an indicator of myocardial oxygen consumption. The double product serves as an indicator that simultaneously integrates BP and HR. It not only reflects the risk of angina pectoris episodes but also independently predicts all-cause mortality and cardiovascular mortality at rest, highlighting the necessity of simultaneously monitoring BP and HR in clinical assessment and risk prediction [55,56]. This metric increases with age in untreated populations and shows a U-shaped trend in treated populations. The dynamic changes of this index suggest that the synergistic elevation of BP and HR increases myocardial oxygen consumption, while their combined control may reduce related risks. This indicates that there is still significant room for improvement if the focus is solely on BP control, and combined management with other indicators (such as HR) may enhance control effectiveness [57].

Group 1 exhibited the lowest BP and HR levels; however, their DBP fell below 70 mm Hg. Consistent with the “J-curve” phenomenon in cardiovascular risk, clinical management for this phenotype should prioritize systolic BP control while closely monitoring DBP to avoid potential hypoperfusion [58].

Group 2 maintained optimal BP control (below the 130/80 mm Hg threshold) alongside improved ventricular structure, which aligns with findings that good BP control is associated with the potential reversal of LV hypertrophy [59]. However, their elevated resting HR (approximately 80 bpm) suggests persistent sympathetic activation. The structural improvement in ΔRWT in these patients may have preceded the reduction in HR, similar to the phenomenon of delayed autonomic modulation (HR variability) after structural improvement observed in patients with chronic thromboembolic pulmonary hypertension [60]. Despite this short-term structural improvement, long-term high HR may still pose potential risks to myocardial tissue, warranting caution. This study, through GBMTM, is the first attempt to reveal the cardiac remodeling characteristics of normal BP with high HR, providing a new perspective for risk stratification in patients with hypertension. These findings suggest that neglecting HR may lead to an underestimation of the potential beneficial associations with cardiac geometry in some patients. Therefore, a combined assessment of BP and HR is necessary for evaluating risk in patients with hypertension.

Group 3 and group 4 represent escalating stages of pressure overload and autonomic decompensation, requiring aggressive, combined targeted therapies. Previous studies have shown that BP trajectories, by integrating long-term BP levels and the slope of changes, can predict cardiovascular events and mortality, providing important evidence for the risk assessment and intervention of CVDs. SBP trajectories have a stronger association with cardiovascular outcomes; furthermore, high-increasing SBP trajectories are also significantly correlated with elevated LVMI [61,62]. Previous studies have confirmed that elevated RWT is an independent predictor of CVD in patients with hypertension and diabetes [63]. This study reveals the synergistic association of dynamic interaction patterns between BP and HR with cardiac remodeling. Furthermore, regarding sleep patterns, the observed risk associated with early awakening aligns with chronobiological evidence that the morning BP surge and cortisol awakening response peak during the early morning window. Awakening prematurely triggers a shift toward sympathetic dominance and amplifies neuroendocrine stress, which are mechanistically linked to elevated vascular resistance. Finally, the observed potential beneficial associations of antidiabetic medications suggest that ameliorating insulin resistance may be linked to the inhibition of myocardial fibrosis and the improvement of microvascular dysfunction.

Our findings also highlight the clinical implication of 0th-order trajectories. The GBMTM algorithm identified 0th-order trajectories as the optimal fit for all 4 groups, indicating that the joint BP and HR profiles remained stable over the 9-month follow-up. This stability aligns with the clinical characteristics of the study cohort, which consisted of patients with chronic hypertension maintained on established medication regimens in a community setting, rather than newly diagnosed patients undergoing initial dose titration. Consequently, the 0th-order trajectories demonstrate that the identified hemodynamic phenotypes, such as the persistently elevated HR in group 2, represent chronic physiological states rather than transient measurements. Establishing this longitudinal stability provides the necessary methodological basis for assessing the cumulative longitudinal associations of these sustained profiles with continuous cardiac structural remodeling.

Our study possesses several key strengths. First, methodologically, this study applies GBMTM to a community-based hypertensive cohort to construct joint trajectories of SBP, DBP, and HR. This approach overcomes the limitations of traditional single-variable trajectory analyses, capturing the population heterogeneity of joint hemodynamic changes and identifying subgroups with distinct pathophysiological profiles. Second, by evaluating the progression of early cardiac remodeling across these joint trajectories, we explored the longitudinal determinants of structural changes. In contrast to previous studies that predominantly relied on cross-sectional associations, demonstrating the link between specific trajectory phenotypes and cardiac remodeling progression deepens our understanding of BP-HR interactions and provides longitudinal evidence for early intervention. Third, based on a Chinese community-dwelling cohort, this study highlights the optimal BP control range, the critical need for concurrent HR management, and the clinical significance of sleep interventions. These findings support the implementation of more precise, individualized hypertension management and cardiovascular risk reduction strategies within primary care settings.

However, several limitations of this study must be acknowledged. First, selection bias cannot be completely ruled out, as the sample was derived from a single community. Second, the specific roles of different classes of antidiabetic medications were not explored. Third, the follow-up period was relatively short. Although this duration was sufficient to observe early changes in ΔRWT, it is inadequate to fully evaluate long-term target organ damage and hard clinical end points. Fourth, echocardiographic measurements were obtained at the 3-month and 9-month follow-ups, rather than at baseline. Baseline imaging was omitted to reduce the logistical burden on primary care facilities during initial community recruitment. Consequently, the first 3 months functioned as a run-in period, during which approximately 200 participants were lost to follow-up. This attrition may have introduced selection bias, restricting the final cohort to patients with higher adherence. Furthermore, the shortened 6-month observation window (from month 3 to month 9) restricted the detectable magnitude of changes in ΔRWT. Because cardiac structural remodeling progresses slowly, this reduced time frame increases the likelihood that true biological changes are masked by the inherent measurement error of echocardiography, which may have attenuated the observed associations in our analysis. Nonetheless, the fact that significant correlations were identified despite these constraints suggests that the associations between high HR, BP trajectories, and cardiac remodeling are clinically noteworthy and warrant attention even within a limited observation window. Fifth, the utilization rate of β-blockers within our cohort was extremely low. Because β-blockers are the primary pharmacological agents for HR reduction, their underuse in this community setting likely contributed to the prevalence of the “normal BP, high HR” phenotype observed in group 2. Consequently, our findings primarily reflect the hemodynamic progression of patients with hypertension lacking specific HR-targeted interventions. This restricts the generalizability of our results to advanced clinical settings where β-blocker prescription rates are substantially higher. Sixth, regarding medication adjustments, we characterized each patient’s pharmacological exposure based on their most frequently prescribed regimen over the 9-month period. While this “predominant exposure” approach effectively captures the primary long-term maintenance therapy typical of community management, it inevitably simplifies the longitudinal pharmacological history. Transient dose titrations or brief medication switches were not continuously modeled as time-varying covariates in the GBMTM. Therefore, the acute hemodynamic impacts of short-term prescription fluctuations cannot be entirely ruled out. Seventh, the relatively modest sample size of our community cohort renders the multiple linear regression models somewhat sensitive to specific data-handling techniques. This is evidenced by the minor discrepancies observed between the sensitivity analysis using imputed data and the strict CCA. Specifically, the association for group 2 attenuated from statistical significance to marginal significance in the CCA. This shift is primarily attributable to the reduction in statistical power when cases with missing data were excluded. Conversely, age emerged as an independently associated factor in the CCA but did not reach full statistical significance in the imputed model. This suggests that in a limited sample, the additional variance introduced by the MI process may dilute certain subtle associations, whereas complete-case exclusion may slightly alter the overall sample composition. These fluctuations highlight the necessity of interpreting the statistical associations cautiously and underscore the need for larger cohort studies to confirm these findings.

Future large-scale, multicenter prospective cohort studies with extended follow-up periods are needed to fully evaluate long-term cardiac target organ damage and hard clinical end points. Therefore, future research should extend the follow-up period beyond 2 years to better elucidate the temporal relationship between HR optimization and structural improvements, particularly for patients in group 2. Expanding recruitment across diverse community populations will help verify the external validity and generalizability of our findings. Additionally, using objective sleep measurement tools, such as polysomnography, would minimize self-reporting bias and improve the accuracy of sleep parameter assessments.

In this study, we applied GBMTM to longitudinal data on SBP, DBP, and HR in community-dwelling patients with hypertension, identifying 4 distinct joint trajectory groups of BP and HR. Notably, group 2 (characterized by normal BP and high HR) was significantly associated with a reduction in ΔRWT over 6 months, suggesting that achieving stable BP control is linked to a potential short-term attenuation of adverse LV geometric remodeling. However, it should be noted that this beneficial association attenuated to marginal significance in the CCA due to the reduction in statistical power. Additionally, concurrent antidiabetic medication use in patients with hypertension with comorbid diabetes was significantly associated with a reduction in ΔRWT, further supporting that these medications showed a beneficial association with the relative change in RWT. Conversely, waking up before 7 AM was associated with an increase in ΔRWT (Table 4), indicating that circadian rhythm disruption is involved in the progression of hypertensive cardiac remodeling.

These findings suggest the necessity to incorporate multidimensional BP and HR metrics to refine risk stratification and cardiac management in hypertension. Targeted interventions need to be explored in the future through long-term follow-up and the integration of multidimensional cardiovascular measurements.