Comparison of the Framingham Heart Study Hypertension Model With Blood Pressure Alone in the Prediction of Risk of Hypertension
The Multi-Ethnic Study of Atherosclerosis
A prediction model, developed in the Framingham Heart Study (FHS), has been proposed for use in estimating a given individual’s risk of hypertension. We compared this model with systolic blood pressure (SBP) alone and age-specific diastolic blood pressure categories for the prediction of hypertension. Participants in the Multi-Ethnic Study of Atherosclerosis, without hypertension or diabetes mellitus (n=3013), were followed for the incidence of hypertension (SBP ≥140 mm Hg and/or diastolic blood pressure ≥90 mm Hg and/or the initiation of antihypertensive medication). The predicted probability of developing hypertension among 4 adjacent study examinations, with a median of 1.6 years between examinations, was determined. The mean (SD) age of participants was 58.5 (9.7) years, and 53% were women. During follow-up, 849 incident cases of hypertension occurred. The c statistic for the FHS model was 0.788 (95% CI: 0.773 to 0.804) compared with 0.768 (95% CI: 0.751 to 0.785; P=0.096 compared with the FHS model) for SBP alone and 0.699 (95% CI: 0.681 to 0.717; P<0.001 compared with the FHS model) for age-specific diastolic blood pressure categories. The relative integrated discrimination improvement index for the FHS model versus SBP alone was 10.0% (95% CI: −1.7% to 22.7%) and versus age-specific diastolic blood pressure categories was 146.0% (95% CI: 116.0% to 181.0%). Using the FHS model, there were significant differences between observed and predicted hypertension risks (Hosmer-Lemeshow goodness of fit: P<0.001); recalibrated and best-fit models produced a better model fit (P=0.064 and 0.245, respectively). In this multiethnic cohort of US adults, the FHS model was not substantially better than SBP alone for predicting hypertension.
Hypertension is a major risk factor for the incidence of cardiovascular and kidney diseases.1–4 The identification of adults at high risk for incident hypertension is important for the cost-effective implementation of interventions. Randomized, controlled trials have demonstrated the benefits of lifestyle modification and pharmaceutical therapy on the reduction of hypertension incidence.5–9
To assist in the identification of individuals at high risk for hypertension, a prediction model was developed using data from white adult participants from the Framingham Heart Study (FHS).10 This model includes multiple factors and was designed for use in clinical practice to identify patients at increased risk for the incidence of hypertension. Although good performance of this model was noted within the FHS population, this equation has not been externally validated, which is necessary before it can be recommended for widespread use. Often, there is substantial overestimation or underestimation of event rates when risk prediction models are applied in populations different from those in which they were developed.11,12 Therefore, we sought to determine the performance of the FHS hypertension risk prediction model using standard criteria in a contemporary, multiethnic population of US adults.13 Additionally, we compared this model with the predictive ability of systolic blood pressure (SBP) categories alone and age-specific categories of diastolic blood pressure (DBP). Age-specific categories were modeled rather than DBP alone, because the association between DBP and hypertension incidence is not uniform across age groups. These analyses were conducted using longitudinal data on the incidence of hypertension among adults ≥45 years of age from the Multi-Ethnic Study of Atherosclerosis (MESA) cohort.
The MESA enrolled 6814 community-dwelling adults who were 45 to 84 years of age at baseline.14 Participants from 4 race-ethnicity groups (white, black, Hispanic, and Asian–primarily of Chinese descent) were recruited from 6 US communities: Baltimore, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles County, California; northern Manhattan, New York; and St Paul, Minnesota. Participants were excluded if they had a history of clinically evident cardiovascular disease; were under treatment for cancer; were pregnant; weighed >300 lb; had significant cognitive deficits; were living in a nursing home or on the waiting list for a nursing home; had plans to leave the community within 5 years; only spoke a language other than English, Spanish, Cantonese, or Mandarin; had a chest computed tomography scan in the previous year; or had any serious medical condition that would prevent long-term participation in the MESA.
Of the 6814 MESA participants enrolled, those with prevalent hypertension at baseline (SBP ≥140 mm Hg, DBP ≥90 mm Hg, current antihypertensive medication use, or a self-report of a diagnosis of hypertension; n=3337) were excluded from the current analyses. Additionally, among those without hypertension at baseline, we excluded participants with diabetes mellitus (fasting glucose ≥126 mg/dL or use of hypoglycemic drugs or insulin; n=223), missing information needed to calculate their hypertension risk (n=6), and those who did not attend 2 sequential MESA visits (n=235). Participants with diabetes mellitus were excluded to correspond with the population used in the development of the FHS hypertension risk prediction model. After these exclusions, 3013 participants who attended the baseline visit (years 2000–2002) and examination 2 study visits (years 2002–2004) were included in the current analysis. There were 2476 MESA participants who were free of hypertension and diabetes mellitus at examination 2 who attended examination 3 (years 2004–2006), and 2130 MESA participants who were free of hypertension and diabetes mellitus at examination 3 who attended examination 4 (years 2006–2008). Those excluded, primarily for having hypertension at baseline, were older (mean age: 65 versus 59 years), more likely to be black (34% versus 20%), and had a higher mean body mass index (BMI; 29.2 versus 27.2 kg/m2). Written informed consent was obtained from all of the participants, and the study was approved by the institutional review boards of all of the participating sites.
Risk Factor Measures
For the current analysis, risk factors for the incidence of hypertension are based on data collected at baseline and MESA exams 2 and 3. During each examination, standardized questionnaires were used to obtain information on demographics, cigarette smoking, medical conditions, and prescribed medication use. Body height and weight were measured with participants wearing light clothing and no shoes. BMI was calculated as weight in kilograms divided by height in meters squared. Parental history of hypertension was only assessed at examination 2.
Blood Pressure Measurements and Hypertension Ascertainment
Blood pressure measurements from all 4 of the MESA examinations conducted to date were used in the current analysis. After resting for 5 minutes in the seated position, participant blood pressure was measured 3 times at 2-minute intervals using an automated oscillometric device (Dinamap Monitor Pro 100, GE Healthcare). Appropriately sized cuffs were used for blood pressure assessment. Blood pressure was defined as the average of the second and third readings. Participants were asked about their previous diagnoses of hypertension, and those responding affirmatively were also asked about their use of prescribed antihypertensive medications. Incident hypertension was defined as the first study visit, subsequent to baseline, at which the participant had SBP ≥140 mm Hg or DBP ≥90 mm Hg or the initiation of treatment with antihypertensive medications. This is the same definition for hypertension used in developing the FHS hypertension risk prediction model.10 Prehypertension was defined, for those who were free of hypertension, as an SBP of 120 to 139 mm Hg and/or a DBP of 80 to 89 mm Hg.
Hypertension Risk Prediction Models
The FHS hypertension risk prediction model has the following components: SBP, sex, parental history of hypertension, BMI, cigarette smoking, and the interaction between DBP and age.10 These variables, with the exception of parental history of hypertension, which was only assessed once, were updated at each study visit for the prediction of hypertension in the current analysis. The FHS hypertension risk prediction model applied an exponential function incorporating a linear combination of each of the aforementioned variables to estimate the probability of developing hypertension over the median 1.6 years of follow-up between each examination (Appendix S1, available in the online Data Supplement at http://hyper.ahajournals.org). The SBP-alone model included 7 SBP categories, whereas the model composed of age-specific levels of DBP included 20 categories (Appendix S1). The SBP-alone and age-specific categories of DBP models used the same blood pressure categories published in the original report of the FHS hypertension risk prediction model.10
The prevalence or mean and SD for each component of the hypertension risk prediction model at baseline was calculated overall and by race-ethnicity. The percentage of participants developing hypertension at 1.6 years of follow-up, the median time between MESA study visits, was calculated overall, by race-ethnicity, and for the individual components of the FHS risk prediction model. Participants remaining free of hypertension could contribute up to 3 periods at risk (ie, between baseline and visit 2, visit 2 and visit 3, and visit 3 and visit 4). Additionally, using a repeated-measures Poisson regression model and updating components at each examination, the relative risks for developing hypertension associated with race-ethnicity and components of the FHS risk prediction model were calculated. Next, the c statistic was calculated for the FHS model, the SBP alone model, and the model of age-specific DBP categories.15 C statistics were calculated for the overall population and for each race-ethnicity, separately. The statistical significance of differences in c statistics for each model was compared using the method of DeLong et al16 for correlated data.
Next, the predicted probability of developing hypertension over 1.6 years was calculated using the FHS model, SBP alone, and DBP with age, and the relative integrated discrimination improvement index was calculated for the FHS model compared with models defined by SBP alone and the age-specific DBP categories. The relative integrated discrimination improvement is a measure of the average increase in sensitivity and specificity when comparing 2 prediction models.17
To assess calibration, the predicted probability of hypertension from the FHS model was divided into deciles. By decile of predicted risk, the observed and mean predicted probabilities of developing hypertension over 1.6 years were calculated. Next, the predicted probability of developing hypertension was calculated after recalibrating the FHS equation using the method described by D’Agostino et al.11 Finally, using the variables in the FHS equation, a best-fit model was generated. The predicted risk of hypertension before and after recalibration and using the best-fit model, separately, were compared with the observed incidence of hypertension via a modified Hosmer-Lemeshow goodness-of-fit χ2 test.18
Three additional analyses were conducted. First, the c statistic for prehypertension, as well as the relative integrated discrimination improvement index for the FHS model versus prehypertension, was assessed. Second, the c statistic and relative integrated discrimination improvement index were evaluated for the FHS equation and SBP alone for developing the incidence of hypertension over 4.8 years of follow-up (ie, without updating components of the equation). Finally, an analysis was conducted including individuals with diabetes mellitus. Statistical analyses were conducted with SAS 9.2 (SAS Institute).
Overall, the mean age of participants was 58.5±9.7 years; 53% were women, and 45%, 20%, 22%, and 13% were white, black, Hispanic, and Asian, respectively (Table 1). There were 849 incident cases of hypertension, 360 between baseline and examination 2, 268 between examinations 2 and 3, and 221 between examinations 3 and 4. The cumulative incidence of hypertension at 1.6 years was 11.1% (Table 2⇓). In a multivariable adjusted model including race-ethnicity and the FHS model components, black race-ethnicity, SBP, age-specific levels of DBP, and BMI were associated with hypertension incidence.
The c statistic for the incidence of hypertension was 0.788 (95% CI: 0.773 to 0.804) for the FHS model and 0.768 (95% CI: 0.751 to 0.785; P=0.096 compared with the FHS) for the SBP model and 0.699 (95% CI: 0.681 to 0.717; P<0.001 compared with the FHS) for the model of age-specific DBP categories (Table 3). The c statistic was highest for Asians and lowest for whites and Hispanics for each model. The relative integrated discrimination improvement for the FHS model compared with SBP alone was 10.0% (95% CI: −1.7% to 22.7%) and compared with the DBP with age model was 146.0% (95% CI: 116.0% to 181.0%).
The FHS prediction model underestimated the risk of hypertension in all of the deciles of predicted risk (P<0.001; Figure). The predicted risk for hypertension after recalibration and using a best-fit model was markedly closer to the observed risk of hypertension (P=0.062 and P=0.245, respectively).
Prehypertension and Hypertension Incidence
At baseline, 1092 MESA participants (36% of those without hypertension) had prehypertension. Among participants with and without prehypertension, 23.6% and 5.3%, respectively, developed hypertension during follow-up. The c-statistic for prehypertension predicting incident hypertension was 0.701 (95% CI: 0.684 to 0.717), which was significantly lower than the c statistics for the FHS model and SBP alone (each P<0.001). The c statistics for prehypertension predicting hypertension were 0.685 (95% CI: 0.658 to 0.711), 0.690 (95% CI: 0.658 to 0.722), 0.685 (95% CI: 0.649 to 0.721), and 0.778 (95% CI: 0.732 to 0.824) among whites, blacks, Hispanics, and Asians, respectively. The relative integrated discrimination improvement index was 78% (95% CI: 58% to 99%) for the FHS model compared with prehypertension.
Including Individuals With Diabetes Mellitus
During follow-up, 122 incident cases of hypertension over 497 intervals of follow-up (24.6%) occurred among MESA participants with diabetes mellitus. The c statistic was markedly similar including individuals with diabetes mellitus and was 0.782 (95% CI: 0.768 to 0.797), 0.763 (95% CI: 0.747 to 0.778), and 0.689 (95% CI: 0.672 to 0.706) for the FHS model, SBP alone, and DBP and age, respectively.
Hypertension Incidence Over 4.8 Years of Follow-Up
The c statistic was similar when evaluating the incidence of hypertension over 4.8 years of follow-up (0.792 [95% CI: 0.775 to 0.807] for the FHS model, 0.773 [95% CI: 0.775 to 0.791] for SBP alone, and 0.691 [95% CI: 0.671 to 0.711] for DBP and age). Additionally, the relative integrated discrimination improvement for incident hypertension over 4.8 years was 9.1% (95% CI: −1.6% to 20.8%) for the FHS model compared with SBP alone.
In the current analysis of data from a multiethnic, community-based sample ≥45 years of age, we evaluated the FHS hypertension risk prediction model for the incidence of hypertension over 1.6 years of follow-up. We report that this model provides good discrimination and, after recalibration, good fit compared with the observed incidence rates. However, using SBP alone provided similar discrimination when compared with the FHS model, which was derived using multiple variables. Specifically, the FHS model provided small, and not statistically significant, improvements in the c statistic over using SBP alone. Also, the relative improvement discrimination index suggests that the FHS model does not have substantially improved sensitivity and specificity when compared with SBP alone.
In the FHS, as well as in the MESA, the level of SBP was identified as a major risk factor for incident hypertension. In the current analysis, the percentage of MESA participants developing hypertension at 1.6 years of follow-up was <4% among individuals with an SBP <110 mm Hg and increased in a graded relation to >40% among those with an SBP of 135 to 139 mm Hg at baseline. In conjunction with our finding that SBP alone was similar to a multivariable model for the prediction of hypertension, this strong association suggests that SBP alone may be sufficient for identifying individuals at high risk for developing hypertension within the next few years. This finding highlights the importance of blood pressure elevations among individuals without hypertension. Previous studies have demonstrated a continuous graded association between SBP and the incidence of cardiovascular and renal diseases2,19,20 down to levels well below 140 mm Hg.
The high prevalence of hypertension (29.3% in 1999–200421) among US adults and its key role in cardiovascular disease and chronic kidney disease suggest that the ability to identify high-risk individuals is critically important. SBP is incorporated into global cardiovascular disease risk prediction models. In the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure guidelines, prehypertension has been defined to help clinicians identify individuals at high risk for hypertension.22 In the current analysis, the predictive ability of prehypertension was low when compared with either the FHS model or SBP alone. This suggests that a more refined risk stratification approach, perhaps using SBP categories, may be worthwhile to identify individuals for aggressive interventions to prevent hypertension.
Among adults with an SBP of 130 to 139 mm Hg, a randomized trial of candesartan versus placebo showed a 16% relative risk reduction in the incidence of hypertension.9 Additionally, lifestyle changes have demonstrated strong benefits in the prevention of hypertension.5,23 Given the high incidence of hypertension and graded association between SBP and hypertension incidence, population-based strategies for lowering blood pressure may yield large risk-reduction benefits.24 Also, aggressive interventions for individuals with elevated SBP may be warranted.
The current analysis highlights the need for external validation of risk prediction models before their implementation into clinical practice. Before risk prediction models are widely disseminated, their properties need to be evaluated in the populations in which they will be used. Without validation, physicians cannot be confident that the absolute risk of disease outcomes provided by the model is appropriate for using when making clinical decisions. Although the FHS model performed reasonably well in the current study, SBP alone also maintained a high predictive value for hypertension.
In the current study, the FHS risk prediction model substantially underestimated the risk for hypertension. However, this discrepancy can be corrected through the process of recalibration. This finding suggests that if one is to apply the FHS hypertension risk prediction model, the model should be recalibrated. To do this, cross-sectional data on the components of the risk prediction model and data on the hypertension incidence are needed. Details of this process have been described previously by D’Agostino et al.11
In addition to SBP, black ethnicity, higher BMI levels, and age-specific DBP levels were associated with an increased risk for hypertension. However, these factors did not substantially or statistically significantly improve the discrimination above SBP alone. Although these factors may be important in the pathogenesis of hypertension, they add little information when considered from a risk prediction perspective. Other factors not included in the FHS hypertension risk prediction model (eg, dietary factors, physical activity, and chronic kidney disease) have been associated with an increased risk for hypertension.6,25–27 It may be worthwhile to consider these factors in future risk prediction models.
The results of the current analysis should be considered in the context of certain limitations. As is common in large epidemiological studies, the incidence of hypertension was based on blood pressure measurements taken on a single date. Also, the FHS risk prediction model was developed in a younger population (ages 20.0 to 69.0 years; mean age: 42.0 years) than the MESA (ages ≥45.0 years; mean age: 58.5 years). This may have influenced the calibration of the FHS model. Furthermore, the role of risk factors for hypertension may be age dependent. For example, among young adults, blacks have a higher incidence of hypertension than whites. Although the performance of the FHS hypertension risk prediction model, as well as SBP alone, may be different in adults <45 years of age, we did not have data available to evaluate this question. Almost 50% of MESA study participants had developed hypertension by the time of their baseline study visit. Despite these limitations, the data set used for the current analyses has many strengths. These include the large ethnically diverse population enrolled, detailed clinical and metabolic characterization of the cohort, and active follow-up for incident hypertension in the MESA. A substantial number of MESA participants (n=849) developed hypertension, providing ample power for the current analysis. Also, all of the data were collected following standardized protocols by trained study staff.
In this contemporary multiethnic study of US adults ≥45 years of age, the FHS hypertension risk prediction model demonstrated good discrimination. However, this model’s performance was not substantially or statistically significantly better than SBP modeled alone. Future studies in other populations, especially adults <45 years of age, are needed to further validate this hypertension risk prediction model. If appropriate, the development and validation of new hypertension risk models may be warranted. Given the high incidence of hypertension among US adults, as well as the strong association between SBP, <140 mm Hg, and the subsequent incidence of hypertension, individualized and population-wide approaches to lower SBP are needed.
Sources of Funding
This research was supported by contracts N01-HC-95159 through N01-HC-95169 from the National Heart, Lung, and Blood Institute.
We thank the other investigators, the staff, and the participants of the MESA for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org.
- Received December 27, 2009.
- Revision received January 11, 2010.
- Accepted April 6, 2010.
Xin X, He J, Frontini MG, Ogden LG, Motsamai OI, Whelton PK. Effects of alcohol reduction on blood pressure: a meta-analysis of randomized controlled trials. Hypertension. 2001; 38: 1112–1117.
Bild DE, Bluemke DA, Burke GL, Detrano R, Diez Roux AV, Folsom AR, Greenland P, Jacob DR Jr, Kronmal R, Liu K, Nelson JC, O'Leary D, Saad MF, Shea S, Szklo M, Tracy RP. Multi-ethnic study of atherosclerosis: objectives and design. Am J Epidemiol. 2002; 156: 871–881.
Reynolds K, Gu D, Muntner P, Kusek JW, Chen J, Wu X, Duan X, Chen CS, Klag MJ, Whelton PK, He J. A population-based, prospective study of blood pressure and risk for end-stage renal disease in China. J Am Soc Nephrol. 2007; 18: 1928–1935.
Ong KL, Cheung BM, Man YB, Lau CP, Lam KS. Prevalence, awareness, treatment, and control of hypertension among United States adults 1999–2004. Hypertension. 2007; 49: 69–75.
Chobanian AV, Bakris GL, Black HR, Cushman WC, Green LA, Izzo JL Jr, Jones DW, Materson BJ, Oparil S, Wright JT Jr, Roccella EJ. Seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure. Hypertension. 2003; 42: 1206–1252.
Whelton PK, He J, Appel LJ. Treatment and prevention of hypertension. In: Manson JE, Ridker PM, Gaziano JM, Hennekens CH, eds. Prevention of Myocardial Infarction. New York, NY: Oxford University Press; 1996: 154–171.
Rose G. Strategy of prevention: lessons from cardiovascular disease. BMJ (Clin Res Ed). 1981; 282: 1847–1851.
Whelton PK, Buring J, Borhani NO, Cohen JD, Cook N, Cutler JA, Kiley JE, Kuller LH, Satterfield S, Sacks FM. The effect of potassium supplementation in persons with a high-normal blood pressure: results from phase I of the Trials of Hypertension Prevention (TOHP)–Trials of Hypertension Prevention (TOHP) Collaborative Research Group. Ann Epidemiol. 1995; 5: 85–95.
Sacks FM, Svetkey LP, Vollmer WM, Appel LJ, Bray GA, Harsha D, Obarzanek E, Conlin PR, Miller ER III, Simons-Morton DG, Karanja N, Lin PH. Effects on blood pressure of reduced dietary sodium and the Dietary Approaches to Stop Hypertension (DASH) diet: DASH-Sodium Collaborative Research Group. N Engl J Med. 2001; 344: 3–10.