Baseline Body Mass Index Category and Incident Treated Hypertension in Australian Men: A Linked Prospective Cohort Study Using PBS/MBS Data Authors: Ebenezer Afrifa‐Yamoah Published: 2026 ABSTRACT ABSTRACTBackgroundExcess adiposity is a major modifiable risk factor for hypertension; however, the incremental prognostic value of body mass index (BMI) beyond standard predictors remains unclear. We examined the associations between baseline BMI category and incident treated hypertension, and assessed whether BMI improved risk prediction.MethodsData were obtained from the Ten to Men longitudinal study (2013–2024) linked to Pharmaceutical Benefits Scheme (PBS) and Medicare Benefits Schedule (MBS) records. The cohort comprised 12,742 men aged ≥ 18 years without treated hypertension at baseline, followed for 133,038 person‐years. Adjusted Cox proportional hazards models were fitted. Discrimination, calibration, decision curve analysis, E‐values, and BMI × age interactions were evaluated using bootstrap 95% confidence intervals (CIs).ResultsDuring follow‐up, 1755 men were initiated on antihypertensive therapy (13.2 per 1000 person‐years). The 10‐year cumulative incidence increased across BMI categories: 7.5% (95% CI 6.6–8.3) in normal‐weight men, 10.3% (8.8–11.8) in underweight men, 12.0% (11.1–12.9) in overweight men, and 20.5% (18.9–22.0) in obese men. Fully adjusted hazard ratios (HRs) versus normal weight were 1.43 (95% CI 1.16–1.77) for underweight, 1.42 (1.22–1.67) for overweight, and 1.88 (1.59–2.22) for obesity; multiple‐imputation estimates were similar. The BMI × age interaction was significant (p < 0.01), with the obesity effect stronger at ages 30–44 years (HR 2.39, 1.91–2.98) than at ages ≥ 45 years (HR 1.57, 1.28–1.94). Adding BMI to age improved discrimination (ΔC = 0.028), and the fully adjusted model achieved C = 0.754.ConclusionBMI is independently associated with incident treated hypertension and provides a modest additional prognostic value beyond age, with obesity having the strongest relative effect in early midlife.Keywords: body mass index, decision curve analysis, hypertension, longitudinal cohort, men's health, obesity, pharmaceutical dispensing records, risk prediction Introduction Hypertension is the leading modifiable contributor to global cardiovascular morbidity and mortality, affecting an estimated 1.28 billion adults worldwide and accounting for approximately 10.4 million deaths annually [1, 2, 3, 4]. Excess body weight is among the most influential modifiable risk factors, with prospective studies consistently demonstrating that each kilogram of weight gain is associated with a 0.5–1.0 mmHg increase in systolic blood pressure [5, 6, 7]. This association reflects the convergence of sympathetic nervous system activation, stimulation of the renin–angiotensin–aldosterone system, renal sodium retention, and endothelial dysfunction [8]. Adipose tissue functions as an active endocrine organ, secreting leptin and other adipokines that exert important cardiovascular effects. Leptin increases sympathetic nervous system activity through central melanocortin pathways, contributing to sustained blood pressure elevation [9, 10, 11] despite resistance to the appetite‐suppressing effects of leptin that develop in obesity (selective leptin resistance) [12]. Insulin resistance amplifies these effects through hyperinsulinemia, endothelial dysfunction, and vascular stiffness, consistent with the established cluster of obesity, diabetes, and hypertension described as metabolic syndrome [13, 14]. The cross‐sectional association between body mass index (BMI) and hypertension risk is well established across diverse populations [15, 16, 17, 18]. BMI cutoff points are not, however, interchangeable across populations: World Health Organization (WHO) expert consultation has concluded that cardiometabolic risk increases at lower absolute BMI values in Asian populations, with the observed‐risk cut point varying between 22 and 25 kg/m2 across Asian groups, while recommending that the standard WHO cut points be retained as the international classification [15]. Therefore, comparisons of effect sizes across cohorts recruited from different populations must be interpreted with the applicable classification in mind. International cohort evidence has demonstrated that higher BMI and weight gain from early to middle adulthood are associated with a substantial increase in cardiovascular disease [19, 20]. Evidence from Australian male populations using objectively ascertained outcomes is relatively sparse. Most prior Australian studies have relied on a self‐reported hypertension diagnosis or a single‐occasion clinical blood pressure measurement, both of which are subject to substantial misclassification [21]. The use of nationally linked Pharmaceutical Benefits Scheme (PBS) dispensing records permits a more reliable longitudinal assessment of pharmacologically treated hypertension at the population scale, but introduces a different set of interpretational challenges around detection bias that have rarely been examined empirically in Australian cohorts. Rationale for a Male‐Specific Analysis Three considerations motivated an analysis restricted to men. First, the design of the source cohort is male‐specific: Ten to Men, established as a national study of male health, does not enroll women; therefore, a sex‐comparative analysis is not possible within these data and would require a separate cohort with equivalent linkage [22, 23]. Second, men carry a higher age‐specific burden of hypertension than women until approximately the sixth decade of life, and men have lower rates of awareness, treatment, and sustained primary care engagement; thus, the pathway from elevated blood pressure to recorded pharmacological treatment differs between sexes [1]. Consequently, an analysis of treatment‐based outcomes is not transportable between sexes without explicit re‐estimation. Third, Australian evidence on untreated hypertension indicates that a substantial fraction of the hypertensive population does not receive pharmacotherapy despite recent contact with primary care [21], and quantifying how adiposity relates to treatment initiation in the group with the poorest engagement is directly relevant to service planning. We make no claim that adiposity matters more in men than in women; the restriction is a feature of the data source and the male‐specific evidence gap, and not an assertion of differential importance. Two methodological considerations warrant emphasis in interpreting BMI–hypertension associations in administrative data. First, hypertension outcomes ascertained through pharmaceutical dispensing records capture treated hypertension rather than the underlying disease, and the probability of treatment depends not only on the presence of elevated blood pressure but also on detection by health services and the clinician's decision to prescribe rather than manage by lifestyle modification alone [21, 24]. Second, the strength and shape of BMI–hypertension associations are expected to vary with age, as the absolute risk of hypertension increases steeply with age, whereas the relative effect of adiposity may attenuate as competing predictors become more important [25]. Quantifying these phenomena directly is essential for clarifying the clinical interpretability of BMI‐based risk stratification in routine practice. An important and largely unresolved question is whether the baseline BMI category significantly improves discrimination and clinical utility beyond standard clinical predictors. If BMI is a strong proxy for unmeasured cardiometabolic processes, its incremental contribution may be small once age and diabetes are considered. Conversely, if BMI captures distinct information about long‐term cardiovascular risk, repeated BMI measurements may remain a core component of hypertension screening. Resolving this question requires an explicit comparison of nested risk models using both discrimination metrics (with appropriate uncertainty bounds) and net benefit metrics that translate the model performance into clinically interpretable terms. Using prospectively linked data from the Ten to Men Australian longitudinal study on male health and the PBS, we aimed to (1) estimate the association between baseline BMI category and incident treated hypertension in Australian men using both complete‐case and multiple‐imputation approaches; (2) formally test the BMI × age interaction and quantify effect modification by age; (3) quantify robustness to unmeasured confounding using E‐values; (4) quantify the incremental discrimination provided by BMI over age and other clinical predictors using C‐indices bounded by bootstrap confidence intervals (CIs); (5) assess calibration and quantify net clinical benefit using decision curve analysis with bootstrap confidence ribbons; and (6) test the robustness of findings to age stratification, latency exclusion, alternative outcome definitions, missing‐data treatment, and competing‐risks adjustment. Methods Study Design and Population This study used a longitudinal cohort design using data from the Ten to Men Australian longitudinal study on male health, a national prospective cohort study established in 2013 to investigate male health across the life course [22, 23]. Ten to Men recruited 16,021 males aged 10–55 years in Wave 1 (2013–2014) using a stratified, multi‐stage, cluster random sampling design. Full details of the sampling frame, selection probabilities, and response rates are reported in the cohort profile and methods papers [22, 23]. Data were collected through self‐completion questionnaires covering sociodemographics, physical and mental health, health behaviors, social determinants, and health service utilization. The present analysis used five waves of Ten to Men data collection, linked at the individual level with PBS dispensing records from March 2011 to February 2025 and contemporaneous Medicare Benefits Schedule (MBS) claims data. Linkage was performed by the Australian Institute of Health and Welfare (AIHW) using encrypted unique identifiers. The underlying PBS and MBS collections are administered by Services Australia. The PBS linkage enabled the objective ascertainment of antihypertensive medication dispensing, and the MBS linkage provided complementary information on primary care engagement through the General Practitioner Management Plan (GPMP) and Team Care Arrangement (TCA) item claims. Analytic Cohort The analytic cohort comprised men who were free of treated hypertension at baseline (defined as no PBS dispensing of antihypertensive medication before January 1, 2014) and who had a valid baseline BMI category recorded in Wave 1. Of 16,021 men in Wave 1, 978 were excluded because of pre‐baseline antihypertensive dispensing and a further 194 lacked a baseline BMI category, yielding 14,849 participants for the secondary all‐ages analysis. The primary analysis was restricted a priori to adult men aged ≥ 18 years at baseline to avoid conflating adolescent growth‐related BMI variation with adult adiposity; this primary cohort comprised 12,742 men, contributing 133,038 person‐years of follow‐up and 1,755 incident treated hypertension events. Sample Size and Precision The analysis used a fixed, pre‐existing cohort, and the available sample was determined by enrollment and linkage rather than by design. Precision was evaluated post hoc against the accepted requirements for survival modelling. The fully adjusted model estimated 12 regression parameters from 1755 events (complete‐case analysis: 1317 events, 11 parameters), yielding between 110 and 146 events per variable, far above the conventional minimum of 10 events per variable. The narrowest 95% CI width for a BMI contrast was 0.45 on the hazard ratio (HR) scale, indicating adequate precision for the primary comparisons. The smallest age stratum (18–29 years, 202 events) was the least precisely analyzed and was interpreted accordingly. Follow‐Up and Loss to Follow‐Up Follow‐up for the primary outcome was passive and continuous rather than interval‐based. Once a participant consented to linkage, PBS dispensing was ascertained continuously from January 1, 2014 to the censoring date of February 28, 2025, a maximum of 11.2 years. Loss to follow‐up for the primary outcome is therefore near zero and is limited to a small number of participants who withdrew consent to linkage or whose records could not be linked. This is a material advantage of the design over survey‐based follow‐up. Survey attrition across Waves 2 to 5 does affect the availability of time‐updated covariates and of the self‐reported blood pressure measure used in the combined‐outcome sensitivity analysis. Because all exposures and covariates in the primary model are measured at baseline, survey attrition does not induce loss to follow‐up in the primary analysis. Exposure: Baseline BMI Category BMI was calculated as self‐reported weight in kilograms divided by the square of self‐reported height in meters, recorded at Wave 1, and categorized using the standard WHO cut points: underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25.0–29.9 kg/m2), and obese (≥ 30.0 kg/m2) [15]. Normal weight was selected as the reference category based on clinical grounds, consistent with established practices in cardiovascular epidemiology. Standard international cutoff points were applied uniformly; population‐specific thresholds were not used because ethnicity was not available in the linked analytic data set at a resolution that would support reliable reclassification. Given that cardiometabolic risk emerges at lower BMI values in some populations [15], this uniform classification may misclassify a minority of men from Asian backgrounds toward lower risk categories, biasing the overweight and obese contrasts toward the null; this is addressed in the limitations presented in Section 4.8. Outcome: Incident Treated Hypertension The primary outcome was incident treated hypertension, defined as the first PBS dispensing of antihypertensive medication recorded on or after January 1, 2014. Antihypertensive medications were identified using the Anatomical Therapeutic Chemical (ATC) classification codes C02 (antihypertensives), C03 (diuretics), C07 (beta‐blocking agents), C08 (calcium channel blockers), and C09 (agents acting on the renin–angiotensin system). Throughout the manuscript, we refer to “treated hypertension” rather than “hypertension” to make explicit that the outcome captures pharmacologically managed cases ascertained through dispensing records. The PBS dispensing date was used as the event date; participants without an antihypertensive dispensing record after baseline were censored on February 28, 2025. This outcome definition has two consequences that shape interpretation, which we explicitly address. First, elevated blood pressure in men that is managed by non‐pharmacological means alone (sodium restriction, energy‐restricted diets, increased physical activity, alcohol reduction, or weight loss) does not contribute to an event, even though such measures produce clinically meaningful blood pressure reductions [5, 26, 27, 28]. Lifestyle modification is recommended as first‐line management for stage 1 hypertension and is frequently tested before pharmacotherapy; therefore, these men are misclassified as event‐free. Australian data indicate that untreated hypertension is common: in the Australian Diabetes, Obesity and Lifestyle Study (AusDiab), the prevalence of untreated hypertension was 15.2 per 100 adults (95% CI 13.2–17.2) against a total hypertension prevalence of 28.6 per 100, and at least one modifiable lifestyle factor was present in 71.7% of individuals with untreated hypertension [21]. Because adiposity is associated with both the probability of lifestyle‐first management and eventual pharmacotherapy, this misclassification is unlikely to be independent of exposure, and its expected net direction is examined empirically in the combined‐outcome sensitivity analysis described in Section 2.6‐5. Second, several PBS‐listed drug classes have indications beyond hypertension, for example, beta‐blockers prescribed for arrhythmia or agents acting on the renin–angiotensin system prescribed for heart failure or proteinuria, and dispensing records do not record the indication. Therefore, misclassification of indications is possible and tends to inflate the apparent event count, particularly among men with established cardiovascular or renal disease. The combined‐outcome and landmark analyses bound the influences of both forms of outcome misclassification. Covariates The covariates were selected a priori based on established epidemiological evidence. They are defined in full below. Age at baseline was included as a continuous variable in years. Diabetes treatment status was classified into three mutually exclusive categories based on PBS dispensing records: no diabetes treatment, non‐insulin diabetes treatment (ATC A10B, oral glucose‐lowering agents), and insulin‐treated diabetes (ATC A10A). Physical activity was a binary indicator of meeting the Australian physical activity guidelines at baseline, derived from the self‐reported frequency and duration of moderate and vigorous activity. Smoking status was defined as a three‐level variable (never, former, or current). Self‐reported high cholesterol levels (reported by a physician) and sleep difficulties were binary indicators. A GPMP/TCA flag, derived from MBS claims, indicated whether a chronic disease management plan had been claimed and was included as a proxy for sustained primary care engagement and hence as a partial control for detection bias. Dietary intake was not measured in a form suitable for adjustment, which is addressed in the limitations presented in Section 4.8. Statistical Analysis Descriptive Statistics and Cox Proportional Hazards Models Continuous variables were assessed for normality using histograms, normal quantile–quantile plots, and the skewness statistic. Approximately normally distributed continuous variables are presented as mean ± standard deviation (SD). Had any continuous variable shown a skewed distribution, it would have been summarized as median (Q1, Q3) and analyzed using rank‐based methods. Age, the only continuous variable in the analysis, was approximately normally distributed and presented as mean (SD). Categorical variables are presented as counts and column percentages. Consistent with STROBE guidance, no significance tests were performed on the baseline characteristics, shown in Table 1. The table describes the composition of the exposure groups rather than testing a hypothesis; in a cohort of this size, such tests reflect sample size rather than clinical relevance [29]. All formal inferences were reported from the regression models, where test statistics were accompanied by each p‐value. Table 1Baseline characteristics of the primary adult cohort by body mass index category (n = 12,742).VariableNormalUnderweightOverweightObeseTotal n (%)3743 (29.4)1557 (12.2)4909 (38.5)2533 (19.9)12,742 (100)Age, years, mean (SD)34.9 (10.9)34.0 (11.3)39.2 (9.8)40.0 (9.6)37.5 (10.6)Married/de facto, n (%)2197 (59.2)778 (51.1)3564 (73.3)1827 (73.0)8366 (66.4)Never married, n (%)1338 (36.1)655 (43.0)968 (19.9)493 (19.7)3454 (27.4)Meets physical activity guidelines, n (%)2378 (63.5)685 (44.0)2986 (60.8)1319 (52.1)7368 (57.8)Current smoker, n (%)745 (28.0)460 (38.7)875 (24.7)479 (24.9)2559 (27.5)High cholesterol, n (%)329 (8.9)163 (11.0)795 (16.5)536 (21.6)1823 (14.6)Non‐insulin‐treated diabetes, n (%)34 (0.9)42 (2.7)113 (2.3)217 (8.6)406 (3.2)Insulin‐treated diabetes, n (%)28 (0.7)19 (1.2)33 (0.7)41 (1.6)121 (0.9)GPMP/TCA claim recorded, n (%)485 (13.0)232 (14.9)804 (16.4)579 (22.9)2100 (16.5)Self‐reported high blood pressure at Wave 1, n (%)208 (5.7)161 (10.9)561 (11.6)564 (22.7)1494 (12.0)Incident treated hypertension, n (%)321 (8.6)181 (11.6)685 (14.0)568 (22.4)1755 (13.8)Person‐years of follow‐up40,17016,37451,39925,096133,038Incidence rate per 1,000 person‐years8.011.113.322.613.2 Note: Body mass index categories follow World Health Organization cutoff points: underweight < 18.5 kg/m2; normal weight 18.5–24.9 kg/m2 (reference); overweight 25.0–29.9 kg/m2; and obese ≥ 30.0 kg/m2. Percentages are column percentages calculated among participants with non‐missing data for each variable; therefore, denominators differ between rows. Smoking status was missing for 3,433 men (26.9%), marital status was missing for 142 (1.1%), and self‐reported high cholesterol levels were missing for 251 (2.0%). All other variables were complete. No significance tests were applied to this descriptive table (see Section 2.6‐1).Abbreviations: GPMP, General Practitioner Management Plan; SD, standard deviation; TCA, Team Care Arrangement. Baseline characteristics of the primary adult cohort by body mass index category (n = 12,742). Note: Body mass index categories follow World Health Organization cutoff points: underweight < 18.5 kg/m2; normal weight 18.5–24.9 kg/m2 (reference); overweight 25.0–29.9 kg/m2; and obese ≥ 30.0 kg/m2. Percentages are column percentages calculated among participants with non‐missing data for each variable; therefore, denominators differ between rows. Smoking status was missing for 3,433 men (26.9%), marital status was missing for 142 (1.1%), and self‐reported high cholesterol levels were missing for 251 (2.0%). All other variables were complete. No significance tests were applied to this descriptive table (see Section 2.6‐1). Abbreviations: GPMP, General Practitioner Management Plan; SD, standard deviation; TCA, Team Care Arrangement. Time at risk was calculated from January 1, 2014 to the earliest incident treated hypertension dispensing, last known follow‐up, or the censoring date of February 28, 2025. Cox proportional hazards models were fitted using three progressively adjusted specifications: Model 1 (unadjusted, BMI category only), Model 2 (BMI category plus age and diabetes treatment status), and Model 3 (Model 2 plus physical activity, smoking, high cholesterol, sleep difficulties, and the GPMP/TCA flag). The Cox model, chosen over parametric alternatives, requires no assumption about the shape of the baseline hazard, which is unknown for treatment initiation and is likely to be non‐monotonic, given the influence of health service contact. The proportional hazards assumption was assessed using scaled Schoenfeld residuals with global and per‐covariate tests, supplemented by time‐varying coefficient plots with locally weighted regression smoothers for visual inspection. Effect Modification and Unmeasured Confounding Effect modification by age was formally tested using a likelihood‐ratio test comparing the additive model (Model 2) with a model including a BMI category × continuous age interaction; the likelihood‐ratio χ 2 statistic and degrees of freedom were reported alongside the p‐value. To assess robustness to unmeasured confounding, E‐values were computed for each fully adjusted BMI HR using the closed‐form expression of VanderWeele and Ding [30], with an analogous expression applied to the confidence limit nearest the null. Missing Data Missing covariate data in the fully adjusted model, predominantly smoking status (n = 3433; 26.9%), with smaller proportions missing for marital status and self‐reported high cholesterol, were addressed by multiple imputation using chained equations (mice package, R Foundation for Statistical Computing, Vienna, Austria), generating 20 imputed data sets with 10 iterations each. The imputation model included all analysis covariates, follow‐up time, the event indicator, and the Nelson–Aalen cumulative hazard estimator, following the recommendations of White and Royston for imputation compatible with a Cox substantive model [31]. Cox estimates were pooled using Rubin's rules. Data were assumed to be missing at random, conditional on the variables in the imputation model; complete‐case and multiple‐imputation results were compared as a missing‐data sensitivity analysis. Their agreement is reported in Section 3.3. Discrimination, Calibration, and Clinical Utility Discrimination was quantified using Harrell's C‐index [32], with 500‐replicate non‐parametric bootstrap percentile 95% CIs for each model and for the difference in the C‐index between nested models. The prespecified threshold for a clinically meaningful incremental gain was a C‐index difference greater than 0.01, consistent with conventions in cardiovascular risk prediction [33]. Calibration at 5 years was assessed by partitioning the cohort into deciles of predicted risk and computing the observed Kaplan–Meier survival within each decile, with Greenwood‐based 95% CIs for the observed risk. Clinical utility was quantified using decision curve analysis [34], computing the net benefit across the plausible range of decision thresholds (2%–30%) for prophylactic intervention, with 150‐replicate bootstrap percentile 95% confidence ribbons around each net benefit curve. Discrimination and net benefit metrics were chosen in preference to reporting the C‐index alone. A small improvement in discrimination may still carry clinical value, and decision curve analysis expresses this value on a scale that can be interpreted against a decision threshold. The cumulative incidence of treated hypertension at 5 and 10 years was estimated separately by BMI category as the complement of the Kaplan–Meier estimator, with log–log‐transformed 95% CIs. Sensitivity Analyses Six prespecified sensitivity analyses were performed: (1) an all‐ages analysis including participants aged 10–17 years, to assess the influence of restricting to adults; (2) age‐stratified analyses (18–29, 30–44, and ≥ 45 years), to evaluate effect heterogeneity by life stage; (3) a landmark analysis restricted to participants event‐free at 2 years after baseline, with the survival clock reset, to address potential reverse causation from undiagnosed prevalent hypertension; (4) exclusion of events occurring within 90 days of baseline; (5) substitution of a combined outcome of PBS dispensing or self‐reported high blood pressure at follow‐up, which recovered men managed without pharmacotherapy and therefore bounded the influence of treatment‐based detection bias; and (6) a competing‐risks framework using both cause‐specific Cox regression and the Fine–Gray sub‐distribution hazard model [35]. Because direct ascertainment of death was not available in the linked analytic data set, competing death events were imputed at the AIHW estimated rate for Australian men aged 18–55 years over an 11‐year horizon (approximately 1.5%) and stress‐tested at 3% as an upper bound. Therefore, this competing‐risks approach constitutes a quantitative sensitivity analysis rather than a definitive estimate and is interpreted as such. Reporting Conventions Decision curve analysis was implemented from first principles following the framework of Vickers and Elkin [34], and E‐values were obtained by direct application of the closed‐form expressions of VanderWeele and Ding [30]. All the tests were two‐sided. Differences between comparison groups were regarded as statistically significant at p < 0.05. p < 0.01 is reported where applicable; exact p‐values are given to three decimal places, where p ≥ 0.001. Test statistics are reported along with all p‐values. Results Cohort Characteristics Of the 16,021 men enrolled in Ten to Men Wave 1, 978 had at least one antihypertensive dispensing before January 1, 2014, and were excluded; an additional 194 men lacked a valid baseline BMI category. Therefore, the all‐ages analytic cohort comprised 14,849 men, contributing 156,035 person‐years of follow‐up and 1853 incident treated hypertension events (overall incidence rate 11.9 per 1000 person‐years). The primary adult‐restricted cohort (≥ 18 years) comprised 12,742 men, contributing 133,038 person‐years and 1755 events (overall incidence rate 13.2 per 1000 person‐years; Table 1). Baseline characteristics differed substantially across BMI categories (Table 1). Men in the normal‐weight category had a mean age of 34.9 (SD 10.9) years, compared with 40.0 (SD 9.6) years for obese men. The prevalence of diabetes treatment ranged from 1.7% in normal‐weight men to 10.2% in obese men, and self‐reported high cholesterol levels increased from 8.9% to 21.6% across the same gradient. The incidence of treated hypertension increased monotonically across BMI categories, from 8.0 per 1,000 person‐years in normal‐weight men to 11.1 (underweight), 13.3 (overweight), and 22.6 (obese). Cumulative Incidence by BMI Category Kaplan–Meier hypertension‐free survival analysis showed a clear, monotonic separation between BMI categories from approximately 2 years after baseline (Figure 1). The cumulative incidence at 5 and 10 years (Table 2) demonstrated substantial absolute differences across the BMI categories. At 10 years, the cumulative incidence in obese men was approximately 2.7‐fold higher than that in normal‐weight men (20.5% vs. 7.5%). The absolute risk difference of 13.0 percentage points corresponds to the number needed to screen of approximately eight obese men to identify one additional treated‐hypertension case beyond the normal‐weight baseline rate over a decade. Kaplan–Meier hypertension‐free survival by baseline body mass index category, primary adult cohort (men aged ≥ 18 years, n = 12,742). Time since baseline in years is plotted on the horizontal axis and the probability of remaining free of treated hypertension is plotted on the vertical axis. Numbers at risk are shown beneath the plot. Table 2Cumulative incidence of treated hypertension at 5 and 10 years by baseline body mass index category, primary adult cohort (n = 12,742).Body mass index category n 5‐year cumulative incidence, % (95% CI)10‐year cumulative incidence, % (95% CI)Normal weight (reference)37433.1 (2.6–3.7)7.5 (6.6–8.3)Underweight15574.8 (3.7–5.8)10.3 (8.8–11.8)Overweight49095.0 (4.4–5.6)12.0 (11.1–12.9)Obese253310.0 (8.8–11.1)20.5 (18.9–22.0) Note: Estimates are the complement of the Kaplan–Meier survivor function (1 − Ŝ(t)) with log–log‐transformed 95% confidence intervals.Abbreviation: CI, confidence interval. Cumulative incidence of treated hypertension at 5 and 10 years by baseline body mass index category, primary adult cohort (n = 12,742). Note: Estimates are the complement of the Kaplan–Meier survivor function (1 − Ŝ(t)) with log–log‐transformed 95% confidence intervals. Abbreviation: CI, confidence interval. BMI Category and Incident Treated Hypertension Table 3 presents the Cox proportional hazards estimates from the three progressively adjusted models alongside the multiple‐imputation pooled estimates for the fully adjusted specification. In unadjusted analyses (Model 1), all non‐normal BMI categories were associated with elevated hazards of incident treated hypertension: underweight HR 1.39 (95% CI 1.15–1.66), overweight HR 1.67 (95% CI 1.46–1.91), and obese HR 2.85 (95% CI 2.49–3.27). After adjusting for age and diabetes treatment status (Model 2), the HRs attenuated as expected, most notably for obese men, where the age and diabetes confounding pathway accounted for approximately one‐third of the unadjusted association. Table 3Cox proportional hazards models for incident treated hypertension, primary adult cohort.VariableModel 1, unadjusted HR (95% CI)Model 2, +age and diabetes HR (95% CI)Model 3, complete case HR (95% CI)Model 3, MI pooled HR (95% CI)Normal weight (reference)1.001.001.001.00Underweight1.39 (1.15–1.66)** 1.36 (1.13–1.63)** 1.43 (1.16–1.77)** 1.37 (1.14–1.64)** Overweight1.67 (1.46–1.91)** 1.37 (1.20–1.57)** 1.42 (1.22–1.67)** 1.34 (1.17–1.54)** Obese2.85 (2.49–3.27)** 1.92 (1.67–2.21)** 1.88 (1.59–2.22)** 1.82 (1.58–2.10)** Age, per year—1.05 (1.04–1.05)** 1.04 (1.03–1.05)** 1.04 (1.04–1.05)** Diabetes, non‐insulin‐treated—5.04 (4.37–5.80)** 2.39 (2.02–2.84)** 2.42 (2.08–2.82)** Diabetes, insulin‐treated—5.09 (3.95–6.55)** 1.77 (1.31–2.41)** 2.09 (1.61–2.71)** Sleep difficulties——1.16 (1.04–1.29)* 1.19 (1.08–1.31)** GPMP/TCA claim recorded——3.78 (3.35–4.26)** 3.63 (3.27–4.03)** C‐index (bootstrap 95% CI)0.605 (0.591–0.618)0.700 (0.686–0.715)0.754 (0.740–0.769)—Schoenfeld test, body mass index category—— χ 2 = 5.90, df = 3, p = 0.117— n/events12,742/1,75512,742/17559309/131712,742/1755 Note: Covariates not statistically significant in Model 3 (physical activity, smoking status, and self‐reported high cholesterol) are omitted from the table for clarity; the full model output is provided in the Supporting Material.Abbreviations: CI, confidence interval; df, degrees of freedom; GPMP, General Practitioner Management Plan; HR, hazard ratio; MI, multiple imputation by chained equations (m = 20); TCA, Team Care Arrangement.* p < 0.05** p < 0.01, for comparison with the normal‐weight reference category or with the covariate reference level. Cox proportional hazards models for incident treated hypertension, primary adult cohort. Note: Covariates not statistically significant in Model 3 (physical activity, smoking status, and self‐reported high cholesterol) are omitted from the table for clarity; the full model output is provided in the Supporting Material. Abbreviations: CI, confidence interval; df, degrees of freedom; GPMP, General Practitioner Management Plan; HR, hazard ratio; MI, multiple imputation by chained equations (m = 20); TCA, Team Care Arrangement. p < 0.05 p < 0.01, for comparison with the normal‐weight reference category or with the covariate reference level. After full multivariable adjustment (Model 3, complete‐case n = 9309), the underweight (HR 1.43, 95% CI 1.16–1.77), overweight (HR 1.42, 95% CI 1.22–1.67), and obese (HR 1.88, 95% CI 1.59–2.22) categories remained associated with elevated risk relative to normal‐weight men (Figure 2). Forest plot of hazard ratios for body mass index category across progressively adjusted Cox proportional hazards models. The reference category is normal weight. Hazard ratios are plotted on a logarithmic scale; horizontal lines denote 95% confidence intervals. Abbreviations: CI, confidence interval; HR, hazard ratio. Pooled multiple‐imputation estimates (m = 20, n = 12,742) for the fully adjusted model were closely comparable with the complete‐case results: underweight 1.37 (1.14–1.64), overweight 1.34 (1.17–1.54), and obese 1.82 (1.58–2.10). The agreement between the two approaches indicates that the complete‐case estimates were not materially biased under the assumption that covariate data were missing at random, and that the missingness mechanism did not differentially affect the inference about BMI effects. Complete‐case estimates are reported in the primary results, with the multiple‐imputation comparisons presented in Table 3. The proportional hazards assumption was satisfied for BMI category (Schoenfeld test, p = 0.117 for the joint factor; individual‐dummy tests p = 0.16 for underweight, p = 0.06 for overweight, and p = 0.13 for obese). Time‐varying coefficient plots with locally smoothed regression curves (Figure 3) confirmed that β(t) was approximately constant over the follow‐up window for all three BMI contrasts. The global proportional hazards test was borderline (p = 0.019) and was driven by the GPMP/TCA covariate (p < 0.01) rather than by BMI category or other exposures of primary interest. Sensitivity re‐analysis stratified on GPMP/TCA produced BMI HRs that were essentially unchanged from the primary estimates. Scaled Schoenfeld residual plots for body mass index category with locally weighted regression smoothers (men aged ≥ 18 years, fully adjusted model). Time since baseline in years is plotted on the horizontal axis and the scaled Schoenfeld residual β(t) is plotted on the vertical axis for each pairwise contrast (underweight, overweight, and obese, each vs. normal weight). The smoothed regression curve lies approximately on the β(t) = 0 reference line across the follow‐up window for all three contrasts, with 95% confidence bands spanning the reference line, supporting the proportional hazards assumption. E‐Values for Unmeasured Confounding E‐values for the fully adjusted BMI HRs were 2.21 for underweight men (confidence‐limit E‐value 1.59), 2.19 for overweight men (1.74), and 3.17 for obese men (2.56). These values indicate that an unmeasured confounder would need to be associated with both BMI category and incident treated hypertension by risk ratios of at least 2.21 to 3.17, above and beyond all measured confounders, to fully explain the observed associations. For the obesity association, the magnitude of unmeasured confounding required to nullify the result would exceed the association observed for diabetes treatment, the strongest measured cardiometabolic confounder in this analysis (HR 2.39 in the fully adjusted model), which makes strong residual confounding less likely. BMI × Age Interaction Likelihood‐ratio testing confirmed an effect modification by age (χ 2 = 12.79, df = 3, p = 0.005). The interaction was driven primarily by the obese × age term: the obesity HR was attenuated by approximately 2% per additional year of baseline age (ratio of HRs per year 0.98, p < 0.01), consistent with a diminishing relative effect of adiposity as age‐related vascular aging and competing predictors became more important. Age‐stratified HRs are presented in Table 4. Table 4Sensitivity analyses for the association between baseline body mass index category and incident treated hypertension.Analysis n/eventsUnderweight HR (95% CI)Overweight HR (95% CI)Obese HR (95% CI)Primary (adults ≥ 18 years, Model 2)12,742/17551.36 (1.13–1.63)** 1.37 (1.20–1.57)** 1.92 (1.67–2.21)** All ages (10–55 years)14,849/18531.25 (1.05–1.48)* 1.34 (1.18–1.52)** 1.87 (1.64–2.15)** Age 18–29 years3357/2021.09 (0.71–1.67)1.51 (1.06–2.15)* 1.94 (1.30–2.90)** Age 30–44 years5497/7121.34 (0.98–1.83)1.38 (1.11–1.71)** 2.39 (1.91–2.98)** Age ≥ 45 years3888/8411.48 (1.14–1.93)** 1.31 (1.08–1.59)** 1.57 (1.28–1.94)** Landmark at 2 years12,167/15061.27 (1.04–1.56)* 1.36 (1.18–1.57)** 1.86 (1.59–2.16)** Excluding events < 90 days after baseline12,742/16891.35 (1.12–1.62)** 1.39 (1.22–1.60)** 1.93 (1.67–2.23)** Combined outcome (dispensing or self‐report)12,742/26141.56 (1.35–1.80)** 1.56 (1.40–1.74)** 2.43 (2.17–2.72)** Competing risks 1.5% (cause specific)12,742/17551.36 (1.13–1.64)** 1.38 (1.20–1.57)** 1.93 (1.67–2.22)** Competing risks 1.5% (Fine–Gray)12,742/17551.36 (1.14–1.63)** 1.37 (1.20–1.57)** 1.93 (1.67–2.22)** Competing risks 3.0% (Fine–Gray)12,742/17551.38 (1.15–1.66)** 1.40 (1.22–1.60)** 1.93 (1.67–2.23)** Multiple imputation (full model)12,742/17551.37 (1.14–1.64)** 1.34 (1.17–1.54)** 1.82 (1.58–2.10)** Note: All models are adjusted for age and diabetes treatment status; the reference category is normal weight.Competing‐risks rates correspond to the Australian Institute of Health and Welfare estimates of cumulative all‐cause mortality for Australian men aged 18–55 years over 11 years; the 3.0% rate is an upper‐bound stress test. Event counts for the stratified, landmark, and combined‐outcome analyses were those of the corresponding analytical subset.Abbreviations: CI, confidence interval; HR, hazard ratio.* p < 0.05** p < 0.01. Sensitivity analyses for the association between baseline body mass index category and incident treated hypertension. Note: All models are adjusted for age and diabetes treatment status; the reference category is normal weight. Competing‐risks rates correspond to the Australian Institute of Health and Welfare estimates of cumulative all‐cause mortality for Australian men aged 18–55 years over 11 years; the 3.0% rate is an upper‐bound stress test. Event counts for the stratified, landmark, and combined‐outcome analyses were those of the corresponding analytical subset. Abbreviations: CI, confidence interval; HR, hazard ratio. p < 0.05 p < 0.01. Incremental Discrimination and Clinical Utility Harrell's C‐index increased monotonically with model complexity (Figure 4). BMI category alone discriminated modestly (C = 0.605, 95% CI 0.591–0.618), and age alone slightly more (C = 0.644, 95% CI 0.629–0.658). Adding BMI category to age produced a small but precise increment in discrimination (ΔC = 0.028, 95% CI 0.020–0.038), yielding C = 0.672 for the BMI plus age model. The clinical model (BMI, age, and diabetes) achieved C = 0.700 (95% CI 0.686–0.715), and the fully adjusted model further increased discrimination to C = 0.754 (95% CI 0.740–0.769), an increment of ΔC = 0.054 (95% CI 0.045–0.064) over the clinical model. All bootstrap 95% CIs for between‐model ΔC excluded zero and exceeded the prespecified threshold of 0.01. Discrimination of nested risk models for incident treated hypertension. Harrell's C‐index is shown with 500‐replicate bootstrap percentile 95% confidence intervals. Body mass index category alone yields C = 0.605; adding age yields C = 0.672, adding diabetes treatment status yields C = 0.700, and the fully adjusted model yields C = 0.754. All between‐model differences exceed the pre‐specified threshold of 0.01, with confidence intervals excluding zero. Abbreviation: CI, confidence interval. Calibration at 5 years (Figure 5) was acceptable across the predicted risk distribution for both the BMI plus age model and the fully adjusted model. The fully adjusted model showed close agreement between the predicted and observed risk in the lowest seven deciles and slight underprediction in the top decile (predicted 23.0% vs. observed 24.9%). The BMI plus age model performed similarly but with greater variability across deciles, reflecting its coarser stratification of risk. Calibration of the body mass index plus age model and the fully adjusted model at 5 years, partitioned by decile of predicted risk. The mean predicted 5‐year risk of treated hypertension in each decile is plotted on the horizontal axis against the observed Kaplan–Meier 5‐year risk in the same decile on the vertical axis, with Greenwood‐based 95% confidence intervals on observed risk. The diagonal line denotes perfect calibration. Decision curve analysis with bootstrap percentile 95% confidence ribbons (Figure 6) showed that the fully adjusted clinical model provided a positive net benefit across the plausible range of decision thresholds (2%–25%) for triggering further hypertension surveillance, exceeding both the treat‐all and treat‐none reference strategies. The 95% confidence ribbon around the full model net benefit curve excluded zero at all thresholds between approximately 3% and 25%, providing statistical support for the clinical utility of the model. At a 10% threshold, a clinically reasonable trigger for enhanced surveillance, the fully adjusted model yielded a net benefit of 0.018 (95% CI: 0.015–0.022), corresponding to approximately 18 additional true‐positive cases identified per 1,000 men screened, with no net increase in false positives relative to a treat‐none strategy. The BMI plus age model also provided a positive net benefit relative to a treat‐none strategy at thresholds below 10%; however, its benefit was approximately half that of the fully adjusted model at clinically meaningful thresholds. Decision curve analysis for 5‐year incident treated hypertension with 150‐replicate bootstrap percentile 95% confidence ribbons. The threshold probability for triggering enhanced hypertension surveillance is plotted on the horizontal axis and net benefit is plotted on the vertical axis. The body mass index plus age model, the clinical model, and the fully adjusted model are compared against the treat‐all and treat‐none reference strategies. The fully adjusted model shows positive net benefit with confidence ribbons excluding zero between approximately 3% and 25%. Sensitivity Analyses Findings were robust across all six prespecified sensitivity analyses (Table 4). The inclusion of adolescents aged 10–17 years alongside adults produced essentially unchanged HRs. In age‐stratified analyses, the obesity–hypertension association was strongest in men aged 30–44 years (HR 2.39, 95% CI 1.91–2.98) and attenuated in men aged ≥ 45 years (HR 1.57, 95% CI 1.28–1.94), consistent with the BMI × age interaction reported in Section 3.5. The point estimate among men aged 18–29 years was similar in magnitude (HR 1.94, 95% CI 1.30–2.90) but was estimated with less precision, reflecting the smaller number of events in this stratum (n = 202). A landmark analysis restricted to participants who remained hypertension‐free at 2 years, with the survival clock reset, yielded a similar obese HR (1.86, 95% CI 1.59–2.16), arguing against substantial reverse causation from undiagnosed prevalent hypertension. Excluding events occurring within 90 days of baseline also produced consistent estimates (obese HR 1.93, 95% CI 1.67–2.23). The combined outcome of PBS dispensing or self‐reported high blood pressure at follow‐up produced larger HRs across all BMI categories (obese HR 2.43, 95% CI 2.17–2.72). This is consistent with the expectation that a treatment‐based outcome selectively undercounts hypertensive men who are not receiving pharmacotherapy and that such men are relatively more common in lower BMI strata. Competing‐risks sensitivity analyses imputing competing death at the AIHW‐estimated rate (1.5% over 11 years for Australian men aged 18–55 years), and stress‐tested at 3%, produced essentially identical BMI HRs under both cause‐specific Cox regression (obese HR 1.93–1.94) and the Fine–Gray sub‐distribution hazard model (obese HR 1.93). The negligible impact of competing‐risks adjustment reflects the substantial gap between the expected mortality rate at these ages (1%–2%) and the observed treated‐hypertension event rate (13.8%), indicating that informative censoring by mortality is unlikely to materially affect inference. Discussion In a large national cohort of Australian men prospectively linked with PBS dispensing records, the baseline BMI category showed a graded association with incident treated hypertension. Relative to normal‐weight men, the fully adjusted HRs were 1.43 for underweight men, 1.42 for overweight men, and 1.88 for obese men. The pattern was robust to age stratification, landmark restriction, alternative outcome definitions, multiple‐imputation treatment of missing covariates, and competing‐risks adjustment. The proportional hazards assumption held for BMI exposure throughout follow‐up, and time‐varying coefficient plots confirmed approximately constant log‐hazard ratios over the 11‐year horizon. The obese E‐value of 3.17 indicates that unmeasured confounding of considerable strength would be required to explain this association. Comparison With Previous Literature Our estimates were within the range established by long‐running cohort studies on adiposity and blood pressure. The Framingham Heart Study first documented the graded the relationship of adiposity to blood pressure and to the development of hypertension [20]. Subsequent cohort evidence has confirmed that weight gain from early to middle adulthood carries a substantial excess cardiovascular risk [19]. A network meta‐analysis of 18 cohorts comprising 89,094 participants reported relative risks of hypertension of 1.80 (95% CI 1.29–2.50) for a stable high BMI trajectory and 1.53 (1.27–1.83) for a sharply increasing trajectory relative to a stable normal trajectory [36]. Our fully adjusted obese estimate of 1.88 is closely comparable in magnitude, notwithstanding the difference between a single baseline categorization and a trajectory‐based exposure. Cohort evidence on BMI trajectories during young adulthood reports a similar direction of effect [37]. In Australia, AusDiab reported a hypertension prevalence of 28.6 per 100 adults and documented a strong association between adiposity and hypertension [21]. This is congruent with the 2.3‐fold difference in the crude incidence rate that we observe between obese and normal‐weight men. We emphasize one source of noncomparability that has sometimes been treated loosely in the literature. Studies differ in whether the outcome is a measured blood pressure threshold, self‐reported diagnosis, or pharmacotherapy initiation, and these definitions do not estimate the same quantity. A treatment‐based outcome is contingent on detection and a prescribing decision, and is therefore expected to yield attenuated relative effects compared to clinic‐measured blood pressure. Differences in BMI cutoff points across populations compound this [15]. A direct numerical comparison of HRs across studies using different outcome ascertainment is, in our view, of limited value, and we have restricted our comparisons to studies whose design supports them. Treated Versus Untreated Hypertension and the Interpretation of Dispensing‐Based Outcomes A central consideration is that PBS dispensing records capture treated hypertension, not all incident hypertension. This is not merely a measurement issue; the probability of a dispensing record depends on (1) the presence of elevated blood pressure, (2) detection by a clinician, and (3) the clinician's decision to prescribe pharmacotherapy rather than manage the patient through lifestyle modification alone. Each step might be differentially distributed across BMI categories. Obese men are more likely to undergo opportunistic blood pressure screening during clinical encounters [38], which may inflate the apparent incidence of hypertension in higher BMI strata relative to the true incidence. Conversely, normal‐weight men who rarely present to primary care may have undiagnosed and untreated hypertension that is never entered in the PBS records, thereby biasing comparisons in the opposite direction. The untreated fraction is not small. AusDiab estimated the prevalence of untreated hypertension among Australian adults at 15.2 per 100 (95% CI 13.2–17.2), slightly more than half of all hypertension cases in that population, and reported that 80.8% of those with untreated hypertension had had a blood pressure check within the preceding 12 months [21]. Therefore, untreated hypertension in Australia is predominantly a treatment decision phenomenon rather than a failure of detection. Consequently, two groups are absent from our event count: men with elevated blood pressure who have not been detected and men who have been detected but are being managed without medication. The second group is substantial because non‐pharmacological management produces clinically meaningful blood pressure reduction—modest sodium reduction lowers systolic blood pressure by several mmHg in a dose‐dependent manner [26, 28], aerobic exercise produces comparable reductions [27], and weight reduction lowers blood pressure in proportion to weight loss [5]—and is recommended as first‐line management before pharmacotherapy in uncomplicated stage 1 hypertension. The three analyses conducted in this study triangulated this concern. First, the GPMP/TCA flag in Model 3 partially controlled for clinical engagement, and its adjustment did not abolish the association between BMI and hypertension. Second, the combined outcome incorporating self‐reported high blood pressure at follow‐up, which recovered men managed without medication, yielded uniformly larger HRs (obese HR 2.43 vs. 1.92 under Model 2). This is the direction predicted if a dispensing‐only outcome selectively undercounts hypertensive men not receiving pharmacotherapy, indicating that our primary estimates are conservative rather than inflated. Third, the 10‐year cumulative incidence in normal‐weight men (7.5%) was broadly consistent with population‐level estimates of treated hypertension at the corresponding ages, suggesting that the ascertainment in the reference category was reasonably complete. Residual detection bias cannot be excluded and should temper inference regarding absolute incidence comparisons across BMI categories. Linkage to primary care blood pressure records, where available, allows the untreated fraction to be estimated directly rather than bounded. Effect Modification by Age Likelihood‐ratio testing confirmed a statistically significant BMI × age interaction (χ 2 = 12.79, df = 3, p = 0.005), with the obesity effect attenuating by approximately 2% per additional year of baseline age. Age‐stratified analyses showed the strongest relative effect in men aged 30–44 years (obese HR 2.39, 95% CI 1.91–2.98), attenuating to 1.57 (1.28–1.94) in men aged ≥ 45 years despite the higher absolute event rate at older ages. This pattern is consistent with life‐course evidence indicating that the relative contribution of adiposity to cardiovascular risk diminishes with advancing age, as competing risk factors such as vascular aging, accumulated atherosclerosis, and increasing prevalence of diabetes, dyslipidemia, and chronic kidney disease become more important [25]. This implies that population‐level prevention may yield a greater absolute benefit when targeted at obese men in early midlife than at older ages, when irreversible vascular changes are more likely to have accrued. The Underweight Finding Underweight men in our cohort had an unexpectedly elevated hazard of incident treated hypertension (fully adjusted HR 1.43, 95% CI 1.16–1.77), a finding that warrants careful interpretation. Several explanations for this have been proposed. Reverse causation is the first: Men with chronic illness, including cardiovascular, malignant, or gastrointestinal, may have lower body weight through disease‐related weight loss and may simultaneously be more likely to receive cardiovascular medications, including agents with indications beyond hypertension. The 2‐year landmark analysis (underweight HR 1.27, 95% CI 1.04–1.56) attenuates but does not eliminate the association, suggesting a reverse causation contribution without fully explaining it. A second explanation is behavioral risk: Underweight men in this cohort had the highest prevalence of current smoking (38.7% vs. 24.7%–28.0% in other categories) and the lowest prevalence of meeting physical activity guidelines (44.0% vs. 52.1%–63.5%). Residual confounding by smoking intensity and other behavioral exposures not captured in a three‐level smoking variable cannot be ruled out. Third, a subset of underweight individuals may have low muscle mass with relatively preserved or elevated visceral adiposity, which is an adverse cardiometabolic phenotype that cannot be identified using BMI alone. Distinguishing between these explanations would require body composition data that are not available in Ten to Men. A fourth possibility that we cannot exclude is misclassification arising from self‐reported height and weight at the extremes of the distribution; the proportion of the analytic cohort classified as underweight (12.2%) is higher than the published national estimates for Australian adult men, and this warrants verification against the source anthropometric variables. Clinically, this finding underscores that low BMI is not protective against treated hypertension in Australian men and should not be a reason to defer routine blood pressure screening. Sleep Difficulties and Primary Care Engagement Sleep difficulties were independently associated with an increased hazard of incident treated hypertension (HR 1.16, 95% CI 1.04–1.29; multiple‐imputation pooled HR 1.19, 1.08–1.31), consistent with the literature linking sleep disorders, particularly obstructive sleep apnea, to elevated blood pressure [39]. The magnitude observed here probably underestimated the true effect; the available sleep difficulty variable was coded only as affirmative responses, with non‐responses treated as absent or missing. This conservative operationalization attenuated the apparent effect by misclassifying some affected men as unaffected. Strengthening sleep‐specific assessments in future waves and linking sleep‐clinic referral data, where available, would clarify the contribution of sleep to the risk of hypertension in this population. The strongest covariate association in the fully adjusted model was the GPMP/TCA flag (HR 3.78, 95% CI 3.35–4.26). GPMP and TCA item claims indicate sustained primary care engagement in chronic disease management and are not direct measures of cardiovascular risk. This large HR almost certainly reflects two distinct phenomena. First, GPMP/TCA recording is associated with the probability of receiving PBS antihypertensive dispensing at any given blood pressure level, a detection bias pathway operating through clinical contact. Second, GPMP/TCA recipients are a more comorbid subgroup with an elevated background cardiometabolic risk, which is incompletely captured by other covariates. We retained the variable in the fully adjusted model as a partial control for clinical engagement, while acknowledging that its inclusion contributed substantially to the apparent global proportional‐hazards violation (Schoenfeld p < 0.01 for this term). Stratifying on GPMP/TCA in sensitivity analyses produced essentially unchanged BMI HRs from the primary estimates, supporting the robustness of the BMI–hypertension inference to this complex covariate. Clinical Implications These findings have several implications for primary care practice and population health policy. They support the continued use of routinely collected BMI category as a simple, low‐cost component of opportunistic hypertension risk stratification in adult Australian men: the fully adjusted clinical model achieves a C‐index of 0.754, comparable to many established cardiovascular risk calculators, and demonstrates positive net benefit across clinically meaningful decision thresholds with CIs excluding zero. The demonstration of effect modification by age has an actionable implication in that BMI‐based screening intensity and weight management intervention should be greatest in men aged 30–44 years, where the relative effect of obesity is the largest. The absolute risk estimates are also directly communicable in a consultation; at 10 years, an obese Australian man has an approximately 20% probability of commencing antihypertensive pharmacotherapy compared with 7.5% in a normal‐weight peer. These implications are accompanied by two important qualifications. First, as the outcome of interest was treatment initiation, the model predicted the likelihood of being prescribed antihypertensive medication rather than the likelihood of developing elevated blood pressure. Accordingly, its predictions should be communicated to patients in those terms. Second, because lifestyle modification is effective and recommended as first‐line management before pharmacotherapy in uncomplicated cases [5, 27, 28], identifying a high‐risk man should trigger a structured lifestyle intervention and closer monitoring rather than creating an expectation that medication will necessarily be required. Methodological Contribution In addition to the substantive findings, this study contributes to the methodology of using PBS‐linked outcomes in Australian epidemiological research. Detection‐bias concerns inherent in pharmaceutical‐record outcomes can be partially addressed through three complementary strategies: inclusion of clinical‐engagement proxies such as GPMP/TCA recording as covariates, triangulation with self‐reported outcomes at follow‐up to bound the direction of bias, and competing‐risks adjustment to address informative censoring by mortality. The convergence of estimates across these strategies provides a defensible basis for inference. We further show that multiple imputations produce nearly identical estimates to complete‐case analysis when missingness is plausibly random, which supports the use of either approach, provided that the comparison is reported. Finally, E‐values [30] provide a quantitative bound on the residual unmeasured confounding required to explain an observed association, and we suggest that they be reported routinely in observational cohort analyses of this type. Strengths and Limitations This study has several strengths. The cohort was recruited by random sampling and broadly representative of Australian men aged 10–55 years, with a prospective follow‐up of up to 11.2 years. Outcomes were objectively ascertained through PBS dispensing records, eliminating the recall bias inherent in self‐reported diagnoses; passive linkage meant that loss to follow‐up for the primary outcome was negligible. Analytical methods were prespecified and followed the current recommendations for survival modeling, discrimination assessment, calibration evaluation, decision curve analysis, missing‐data handling, and competing‐risks adjustment. The analysis code is openly available. The sensitivity analyses across multiple specifications converged to consistent conclusions. This study had several limitations. BMI was derived from self‐reported height and weight at Wave 1, and was available only as a four‐level category in the linked analytic dataset. Self‐reporting introduces measurement error, particularly under‐reporting of weight, and the resulting misclassification tends to attenuate the observed BMI–hypertension associations. Repeated BMI measurements were not available in continuous form, which precluded group‐based trajectory modeling or growth mixture modeling [40, 41]. Standard WHO cutoff points were applied uniformly because ethnicity was not available at usable resolution. Given that cardiometabolic risk emerges at a lower BMI in some populations [15], this may attenuate overweight and obese contrasts. The absence of detailed dietary and physical activity data is a substantial limitation requiring explicit treatment. Physical activity was measured, but only as a binary indicator of meeting the Australian guidelines derived from self‐reported frequency and duration. This coarse operationalization cannot capture activity intensity, sedentary time, or change in activity over follow‐up. Therefore, residual confounding by physical activity was likely even though the variable was included in Model 3. Dietary intake was not measured in a form suitable for adjustment at any wave, and no food frequency questionnaires, dietary recalls, or urinary sodium measurements were available. Sodium intake is an established determinant of blood pressure with a dose‐dependent effect [26, 28] and is positively correlated with the total energy intake and hence with adiposity; therefore, its omission is expected to bias the BMI–hypertension associations away from the null. Alcohol intake was available only in a crude categorization, and family history of hypertension was not collected. We did not attempt to adjust for what was not measured. Instead, the E‐values reported in Section 3.4 quantify the joint strength of association that unmeasured confounders of this kind would need to have with both exposure and outcome to explain the findings, and for the obese contrast that strength (3.17) exceeds the association observed for any measured confounder in this analysis. Cohorts with objective dietary‐ and accelerometer‐based activity measurement linked to dispensing records would be required to resolve this properly. Further limitations concern the outcome and analytic data set. The outcome captures only treated hypertension, with detection bias and lifestyle management considerations discussed in Section 4.2. Sleep difficulty was coded only as an affirmative response in the source data set; therefore, the absence of an affirmative response was treated as absent or missing, which attenuated the measured effect of this covariate. Direct ascertainment of mortality was not available, requiring competing‐risks sensitivity analysis to be implemented by imputing deaths at the AIHW‐estimated rate. Although a substantial gap between expected mortality (1%–2%) and observed events (13.8%) supports robustness, a definitive competing‐risks analysis would require linked death‐registry data. Finally, although the source cohort was recruited by random sampling, participation and consent to linkage were not random. Thus, the analytic sample may underrepresent men with the least engagement with health services, precisely the group in whom untreated hypertension is most likely. Conclusion Baseline BMI category is associated with incident treated hypertension in Australian men in a graded manner, with obese men at almost twice the risk of normal‐weight men after full adjustment. The effect of obesity is modified by age and is attenuated later in life. The incremental prognostic value of BMI over age alone is modest but precisely estimated, and a multivariable clinical model provides a positive net benefit across a plausible range of decision thresholds. These findings are robust to multiple‐imputation treatment of missing covariates, competing‐risks adjustment, latency restriction, age stratification, and detection‐bias triangulation, and the magnitude of unmeasured confounding required to explain the obesity association exceeds that of any measured cardiometabolic confounder. These findings support the use of routinely collected BMI category as a simple risk stratification tool in primary care for hypertension prevention, with value in identifying obese men in early mid‐life. As the outcome captures only pharmacologically treated hypertension, the estimates should be interpreted conservatively. Linked‐data work incorporating repeated continuous BMI, primary care blood pressure records, dietary and objective physical activity measurement, and death‐registry linkage would allow weight phenotypes to be characterized and the untreated fraction to be estimated directly rather than bounded. Funding The author has nothing to report. Ethics Statement The Australian Longitudinal Study on Male Health was approved by the University of Melbourne Human Research Ethics Committee (HREC 1237897 and 1237376). All participants provided written informed consent, including consent to linkage with administrative health records. Secondary analysis of the linked de‐identified data set was approved by the relevant data custodians. The study was conducted in accordance with the Declaration of Helsinki. Reporting follows the STROBE guideline for cohort studies. Conflicts of Interest The author declares no conflicts of interest. Use of Large Language Models, AI, and Machine Learning Tools A large language model assistant (Claude) was used to support language editing and manuscript formatting. The authors reviewed and verified all content and take full responsibility for the accuracy and integrity of the manuscript. Acknowledgments The authors thank the men who participated in the Ten to Men Australian Longitudinal Study on Male Health for their ongoing contribution, and the Australian Institute of Health and Welfare for undertaking the data linkage. Pharmaceutical Benefits Scheme and Medicare Benefits Schedule data were provided by Services Australia. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. Please note that Ten to Men data are available to approved researchers through the Australian Data Archive (https://ada.edu.au). PBS and MBS linked data are available subject to data custodian approval. Analysis code is publicly available at https://github.com/EAfrifa-Yamoah/ttm-bmi-hypertension and archived at Zenodo (https://doi.org/10.5281/zenodo.20471416) under the MIT licence; the linked data remain governed by Ten to Men and Services Australia agreements. References Associated Data Data Availability Statement The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. Please note that Ten to Men data are available to approved researchers through the Australian Data Archive (https://ada.edu.au). PBS and MBS linked data are available subject to data custodian approval. Analysis code is publicly available at https://github.com/EAfrifa-Yamoah/ttm-bmi-hypertension and archived at Zenodo (https://doi.org/10.5281/zenodo.20471416) under the MIT licence; the linked data remain governed by Ten to Men and Services Australia agreements.