Decision Tree-Based Classification for Maintaining Normal Blood Pressure Throughout Early Adulthood and Middle Age: Findings From the Coronary Artery Risk Development in Young Adults (CARDIA) Study

Abstract

Background

For most individuals, blood pressure (BP) is related to multiple risk factors. By utilizing the decision tree analysis technique, this study aimed to identify the best discriminative risk factors and interactions that are associated with maintaining normal BP over 30 years and to reveal segments of a population with a high probability of maintaining normal BP.

Methods

Participants from the Coronary Artery Risk Development in Young Adults study aged 18–30 years with normal BP level at baseline visit (Y0, 1985–1986) were included in this study.

Results

Of 3,156 participants, 1,132 (35.9%) maintained normal BP during the follow-up period and 2,024 (64.1%) developed higher BP. Systolic BP (SBP) within the normal range, race, and body mass index (BMI) were the most discriminative factors between participants who maintained normal BP throughout midlife and those who developed higher BP. Participants with a baseline SBP level ≤92 mm Hg and White women with baseline BMI < 23 kg/m² were the two segments of the population with the highest probability for maintaining normal BP throughout midlife (69.2% and 59.9%, respectively). Among Black participants aged >26.5 years with BMI > 27 kg/m², only 5.4% of participants maintained normal BP throughout midlife.

Conclusions

This study emphasizes the importance of early life factors to later life SBP and support efforts to maintain ideal levels of risk factors for hypertension at young ages. Whether policies to maintain lower BMI and SBP well below the clinical thresholds throughout young adulthood and middle age can reduce later age hypertension should be examined in future studies.

Graphical Abstract

Graphical Abstract.

Despite global efforts to maintain normal blood pressure (BP) across the lifespan, the prevalence of elevated BP among US adults is high.¹ Hypertension is a multifactorial disease, which involves a broad array of risk factors² and leads to target organ injury and cardiovascular events.³ The Coronary Artery Risk Development in Young Adults (CARDIA) study and many others have shown that a wide range of demographic, behavioral, and clinical characteristics are associated with increases in BP and the incidence of hypertension.^4–8 Beyond the independent association of each factor of these identified risk factors, interactions among risk factors should also be considered. From a public health perspective, identifying the characteristics associated with the maintenance of normal BP could inform the development of future policies to preserve normal BP (<120/80 mm Hg) throughout the lifetime.

Traditional regression models allow for the assessment of the association of a priori defined interactions with the outcome of interest. In a decision tree analysis,^9–11 multiple independent variables are analyzed to reveal multilevel interactions. When using decision tree analysis, the entire study population is split into subgroups based on the most discriminative predictors, the best cutoff points of each predictor, and the best combinations of these predictors. Eventually, homogenous subgroups, with similar probability for having an outcome, are defined and branches of these mutually exclusive subgroups are presented in a graphical way.

In this study, we utilized the decision tree analysis technique to identify the best discriminative risk factors and interactions associated with maintaining normal BP from young adulthood to middle age and to reveal segments of a population with the greatest likelihood of maintaining normal BP.

METHOD

The current analysis was based on the CARDIA study,¹² a multicenter longitudinal US cohort that began in 1985–1986 (year 0, Y0) with 5,115 individuals aged 18–30 years, and continued in follow-up examinations at Y2, Y5, Y7, Y10, Y15, Y20, Y25, and Y30 after baseline (2015–2016). Each site’s Institutional Review Board has approved the study protocol. Participants were included in the current analysis if they met the following inclusion criteria at baseline (Y0): systolic BP (SBP) < 120 mm Hg, diastolic BP (DBP) < 80 mm Hg, not taking antihypertensive medication, and no history of hypertension diagnosis (self-report). Additionally, the analysis was restricted to participants who attended ≥3 follow-up examinations during which their BP was measured. Participants who were missing baseline modifiable health behaviors (n = 38) were excluded from the analysis. For each participant, at each visit, sitting SBP and DBP were measured three times by trained research staff according to standard protocol.¹² BP at each visit was defined as the average of the second and third measurements. Since the method for BP measurements was changed at Y20, calibrated measurements were used for Y20, Y25, and Y30. Having normal BP (yes/no) throughout midlife was the outcome.¹³ Individuals with SBP < 120 mm Hg and DBP < 80 mm Hg at all follow-up exams without the use of antihypertensive medication and with no self-report of previous diagnosis of hypertension at all available follow-up visits were considered to maintain normal BP throughout midlife. Differences between individuals who maintained normal BP during follow-up and those who developed high BP were evaluated using the χ² test for categorical variables and the unpaired t test and Mann–Whitney U test for continuous variables. A decision tree analysis allows us to capture unknown multiway interactions and to define combinations of variables with the highest actual percentage of individuals with the outcome of interest. Therefore, we used this technique to identify the best discriminative risk factors and interactions associated with maintaining normal BP from young adulthood to middle age and to reveal segments of a population with a high likelihood of maintaining normal BP. All predictors were assessed at Y0 and included sociodemographic variables: sex (male/female), age (years), education (years of school completed), race (Black participants; White participants), and marital status (married, other); medical history: self-report of high cholesterol, heart problem, diabetes, kidney problem, thyroid problem, stomach or duodenal ulcer, liver problem, cancer, nervous/emotional/mental disorder, gallstones/gall bladder disease, and sickle cell trait (yes/no for each item); family history of hypertension (yes/no based on presence of hypertension in at least one of the biological parents, as reported by each participant); modifiable health behaviors: measured body mass index (BMI; kg/m²) and self-report of alcohol use (yes/no), dietary sodium (mg), cigarette smoking status (never, former, and current), cardiorespiratory fitness (minutes, duration of a treadmill test), and total intensity score (exercise units of heavy and moderate recreational sport categories¹⁴); four elements of social support: instrumental support (four questions, weighted mean ranging from 0 to 3), emotional support (one question ranging from 0 to 3), net adequacy (four questions, weighted mean ranging from 1 to 4), and social network index (two questions, mean ranging from 0 to 2)¹⁵; and baseline BP: SBP and DBP. Of all these potential predictors, the most discriminative predictors and their interactions are visually displayed in the shape of a tree diagram.

RESULTS

Of 5,115 individuals included in CARDIA study, 3,156 (61.7%) met the inclusion criteria for the current analysis and had normal BP levels at baseline. Compared to those excluded, these individuals were more likely to be female (62.6% vs. 41.4%) and White participants (52.8% vs. 41.5%), and less likely to have a family history of hypertension (67.6% vs. 72.6%), diabetes (0.6% vs. 1.3%), and high cholesterol (1.6% vs. 2.8%).

Of the 3,156 participants, 1,132 (35.9%) maintained normal BP during a median follow-up of 29.9 (IQR, 29.2–30.2) years and 2,024 (64.1%) developed higher BP (prehypertension/hypertension). Main characteristics of the total study cohort, stratified by maintaining normal BP groups, are described in Supplementary Table S1. Utilizing a decision tree analysis, subgroups with different probability for maintaining normal BP throughout midlife were revealed (Supplementary Table S2), and the best discriminative variable for maintaining normal BP throughout midlife was baseline SBP (Figure 1a). Specifically, 69.2% of participants with baseline SBP ≤ 92 mm Hg maintained normal BP levels (Figure 1a, node 1a). This proportion decreased as the baseline SBP increased, 17.5% of participants with a baselined SBP of >113 and <120 mm Hg maintained normal BP.

Figure 1.

Classification tree analysis for maintaining normal blood pressure.^a,b(a) Classification tree including baseline blood pressure.^c,d (b) Classification tree without baseline blood pressure.^d,e Abbreviations: BP, blood pressure; HTN, hypertension; SBP, systolic blood pressure; DBP, diastolic blood pressure. ^aSPSS software version 26.0 was used for the analysis. Chi-square Automatic Interaction Detection (CHAID) was defined as the growing method. This method uses a multiway split and enables better segments and interpretation of the results. The decision graph, consisting of nodes which split and create branches, was created based on maximum three number of splits (a total of four layers), and minimum 100 number of cases in each node. ^bThe percentage of individuals who maintained normal blood pressure throughout midlife is displayed in gray. ^cAll predictors were assessed at Y0 and included sociodemographic variables: sex (male/female), age (years), education (years of school completed), race (Black participants; White participants), and marital status (married, other); medical history: self-report of high cholesterol, heart problem, diabetes, kidney problem, thyroid problem, stomach or duodenal ulcer, liver problem, cancer, nervous/emotional/mental disorder, gallstones/gall bladder disease, and sickle cell trait (yes/no for each item); family history of HTN (yes/no based on presence of HTN in at least one of the biological parents, as reported by each participant); modifiable health behaviors: measured body mass index (kg/m²) and self-report of alcohol use (yes/no), dietary sodium (mg), cigarette smoking status (never, former, current), cardiorespiratory fitness (minutes, duration of a treadmill test), and total intensity score (exercise units of heavy and moderate recreational sport categories); and four elements of social support: instrumental support (four questions, weighted mean ranging from 0 to 3), emotional support (one question ranging from 0 to 3), net adequacy (four questions, weighted mean ranging from 1 to 4), and social network index (two questions, mean ranging from 0 to 2).¹⁵ Participants (n = 68) who were missing ≥1 element of baseline BP, social support were included in the analysis: SBP and DBP (mm Hg). ^dRisk (Std. Error), proportion of cases incorrectly classified after adjustment for prior probabilities and misclassification costs was 0.30 (0.01) and 0.32 (0.01) for tree A and tree B, respectively. The simplest trees with the highest level of accuracy were selected. ^ePredictors included: all predictors included in this figure except of baseline SBP and DBP.

When the baseline SBP was removed from the decision tree analysis in an attempt to explore multiway interactions between other risk factors, race represented the next most discriminative variable (Figure 1b). Among Black participants, only 23.2% maintained normal BP throughout midlife, compared to 47.2% among White participants (P < 0.001). In both race groups, BMI improved the discrimination. Among White participants, a BMI of 22.9 kg/m² was the cutoff point that best discriminated maintaining normal BP; 55% of those with a baseline BMI of <22.9 kg/m² maintained normal BP compared to 37% of those with baseline BMI ≥ 22.9 kg/m². White women with baseline BMI < 23 kg/m² (Figure 1b, node 10) represent the second segment of the population with the highest probability for maintaining normal BP throughout midlife. Black participants aged >26.5 years with BMI > 27 kg/m² were the highest at-risk segment of a population, with 5.4% having maintained normal BP (Figure 1b, node 19). Among the modifiable risk factors, only BMI and smoking status (Figure 1b, nodes 15 and 16) have shown discriminative ability.

Discussion

Utilizing the decision tree analysis technique, we demonstrated that baseline SBP levels well in the normal range, race, BMI, and sex are the factors which best discriminate between CARDIA participants who maintained normal BP from young adulthood to middle age (36%) and those who developed high BP (64%). By defining a set of rules regarding these variables, we identified the segments of the CARDIA population with the highest probability for maintaining normal BP. These findings highlight the importance of primordial prevention at young ages as already by young adulthood (mean age 24 years) baseline SBPs well below clinical thresholds were the strongest predictor of an individual’s probability of maintaining normal BP throughout middle age. Importantly, even among individuals at normal baseline BP levels, fewer Black participants (23%) compared to White participants (47%) were able to maintain these normal levels throughout almost 30 years of follow-up. This is consistent with prior CARDIA studies which have demonstrated substantially higher risk for hypertension among Black participants compared with White participants regardless of BP level at baseline, and may be related to racial differences in risk factors for hypertension (i.e., parental history of hypertension and maintenance of health behavior),⁵ and be a marker for psychosocial factors resulting from societal disparities/racism and can influence SBP levels.

Among the modifiable risk factors, only BMI and smoking status provided additional discriminative ability within some subgroups of the population. Other traditional risk factors, for which the main effects for hypertension prevention have been previously shown, were not found to improve discrimination, presumably, due to lack of interaction with those variables included in the final trees.

While decision tree analysis is a powerful technique to reveal interactions between risk factors its limitations should be considered. Decision tree analysis tends to be unstable and sample dependent and to determine arbitrary thresholds for continuous factors. Accordingly, the thresholds revealed in this study may be considered as reference points rather than definitive demarcations of actual risk. The robustness of the resultant models should be examined in future studies.

This study emphasizes the importance of early life factors to later life SBP and support efforts to maintain ideal levels of risk factors for hypertension at young ages. Whether policies to maintain lower BMI and SBP well below the clinical thresholds throughout young adulthood and middle age can reduce later age hypertension remains to be determined.

Disclosure

The authors declared no conflict of interest.