Cumulative Blood Pressure Measurement for Cardiovascular Disease Prediction

Promise and Pitfalls

Donald M. Lloyd-Jones, MD, SCM


J Am Coll Cardiol. 2022;80(11):1156-1158. 

Blood pressure (BP) elevated above optimal levels remains the most prevalent chronic risk factor for cardiovascular disease (CVD) worldwide, and CVD remains the leading cause of death and disability.[1] There is a continuous, log-linear relationship among BP levels and CVD and mortality events extending down to as low as 90 mm Hg.[2,3] Much of the data describing these associations derives from population- and community-based observational cohort studies, and has focused on single BP measurements from a specific examination cycle or age. In recent years, however, there has been increasing interest in measures that reflect aspects of exposure to BP over time, rather than at 1 point in time.

The story of how best to represent BP and its associations with risk for CVD events has evolved with the availability of increasing durations of repeated measures from longitudinal cohort studies that have measured BP in the same individuals over decades. Early studies linked single measures of BP relatively close in time to subsequent CVD events.[4] Next, observations from population studies emerged that single measures of BP earlier in life were stronger predictors of events than later measures. Studies of BP over time noted that BP tended to track, meaning that individuals tended stay in similar strata of BP across their life course;[5] although systolic blood pressure (SBP) tended to rise across adulthood in most people, individuals on the whole stayed within the same stratum (eg, top quartile) over time as BP rose. More recent analyses have examined mean levels of BP averaged over time, change in BP over time, cumulative BP (area under the BP exposure curve [AUC], expressed as mm Hg × years),[6–8] trajectories in BP (encompassing groups with diverse patterns of BP change curves),[9] and BP variability (visit-to-visit variability independent of underlying BP levels),[10–12] among other measurements. These measures have taken on increasing clinical relevance given the potential availability of longitudinal BP measures for patients within electronic health records (EHRs).

In general, each of these approaches has added incremental value to our understanding of BP exposures, how BP may cause cardiac and vascular changes that are only partially reversible, and, importantly, why earlier maintenance of optimal BP (through primordial prevention), or earlier restoration of low BP (through more intensive primary prevention) is more successful in reducing long-term CVD risk than later restoration of BP to low levels after onset of hypertension. A study performed by my colleague, Prof Kiang Liu,[13] in which we examined middle-aged and older individuals from the MESA (Multi-Ethnic Study of Atherosclerosis) and younger participants from the CARDIA (Coronary Artery Risk Development in Young Adults) study, illustrates these issues. In this analysis, MESA participants with hypertension who were treated with medication and controlled to optimal levels (<120/<80 mm Hg) still had twice the risk for CVD as those who always had untreated BP <120/<80 mm Hg. Data from the CARDIA participants followed over decades through young adulthood suggested the reason: those whose BP rose through young adulthood and was treated back down to levels of <120/<80 mm Hg had higher cumulative exposure to BP (mm Hg × years) and had greater left ventricular mass, more coronary artery calcification, and worse renal function. In other words, there was a price to be paid (in subclinical target organ damage) for cumulative BP exposure and time spent with BP elevation above optimal that could not be fully reversed by restoration of optimal BP levels with medication.[13] Other supportive data for this concept come from Mendelian randomization studies,[14] suggesting that lower lifelong BP levels are associated with lower risk than treatment reductions, and legacy effect studies, in which earlier treatment of BP reduces risk more than later treatment.[15]

In this issue of the Journal of the American College of Cardiology, Wang et al[16] add some new data regarding the importance of considering cumulative BP exposure in an especially high-risk group: individuals with type 2 diabetes. Using extended follow-up data from the ADVANCE randomized controlled trial of BP lowering (with a fixed-dose combination vs placebo) and intensive glycemic control (vs usual care), the investigators examined on-study BP levels over 2 years and calculated the cumulative BP exposure. They then estimated the cumulative SBP "load," defined as the proportion of AUC (mm Hg x years) of levels ≥130 mm Hg divided by the AUC of all BP measures, to effectively provide a quantification of time and amount of BP spent above 130 mm Hg during the 2 years. The cumulative SBP load was compared with mean BP, time spent with SBP controlled, and SBP variability from visit to visit. Participants were followed for up to 7.6 years for CVD events, including CVD death and nonfatal myocardial infarction or stroke, and secondary outcomes of all-cause mortality and specific CVD subtypes. Overall, the authors observed that cumulative SBP load was more informative regarding CVD risk, and discriminated risk and reclassified more patients' risk correctly, than the other measures. Differences in discrimination and reclassification were statistically significant, but very modest.[16] Strengths of this study include the large and well-characterized sample, rigorous measurement of BP and follow-up of participants during the trial, and analytic approaches. Some modest limitations to be considered include the relatively short duration of BP measures (2 years), incomplete and possibly informative follow-up for events after the formal trial was completed, incomplete adjustment for potential confounding variables (eg, smoking, alcohol use), and uncertain generalizability of the trial population to broader clinical samples. It would also have been informative to adjust for baseline/entry BP levels, or change in BP level during the 2 years, to assess whether any of the newer BP measures provided incremental value. Nonetheless, these data do provide insight into an important high-risk clinical population.

Based on these results, Wang et al[16] contend that cumulative SBP load and visit-to-visit SBP variability "should be used in conjunction in future cardiovascular risk prediction algorithms." (Presumably, these would replace current use of single BP measures.) Perhaps so; but before jumping to such a conclusion, these measures need to be considered through additional lenses. The addition of new measures or biomarkers to risk prediction algorithms does not happen merely because of demonstration of independent statistical association with events, or necessarily even after demonstration of statistically significant incremental benefit in discrimination, calibration, or reclassification in 1 patient sample. All of these are important, but insufficient, considerations, and there are others.[17]

First, in whom should we use these measures? The present study included patients with prevalent CVD (approximately 10% had prior stroke and 20% prior coronary disease), and we do not know if the "benefit" of cumulative BP was consistent among those with and without CVD; risk prediction algorithms are generally used in primary prevention clinical settings. Second, would longer-term cumulative BP values have provided better (or worse) information?

Third, do cumulative and variability BP values measured in routine clinical practice provide the same information as standardized measures performed in the setting of a clinical study? Ahmad et al[18] recently demonstrated only modest correlation between clinically reported BPs in a large EHR sample and standardized BP measures in the same individuals who were MESA participants. On average, the EHR data overestimated SBP by 6.5 mm Hg (95% CI: 4.2–7.8 mm Hg) compared with MESA values, but there was wide and unpredictable variability in differences. When noise is introduced into measurements, the value of predictor variables declines, and it may be amplified or ameliorated somewhat with use of repeated measures.

Fourth, do the new measures add incremental value to existing risk prediction equations? We have some information here as well. Using data from multiple cohorts, we recently demonstrated that cumulative SBP measures over 5 or 10 years, compared with single BP measurements, did not improve the C-statistic and only modestly improved the net reclassification of individuals for CVD risk estimation.[19]

Fifth, and crucially, although use of these measures would likely be cost-effective and relatively easy to program, will clinicians or health systems take the time to integrate these cumulative measures and present and interpret them correctly to assist risk prediction and preventive decision making? And finally, can we demonstrate that use of these measures leads to improved decision making and better outcomes? This last bar is one that is rarely even attempted for novel (or even existing) measures or biomarkers.[17] But, certainly, the next guidelines should reconsider all types of BP measures, and other potential predictors, to optimize risk estimation and identification of patients with greatest net benefit from risk-reducing therapies.

Ultimately, clinicians should leverage as much information on their patients as possible to understand their BP-related CVD risk, to identify those who may be more likely have occult or emerging subclinical target organ damage, and to identify those who may have particular net benefit from earlier or more-intensive treatment. These opportunities are more readily available with integration of data that allow for visualization of longer-term BP patterns and incorporation of home BP monitoring and ambulatory BP monitoring data to monitor out-of-office BP levels and control.