Assessment of a Comprehensive Early Childhood Education Program and Cardiovascular Disease Risk in Midlife

Key Points Question Is a large-scale early childhood program providing comprehensive services from ages 3 to 9 years associated with midlife Framingham Risk Scores (FRSs), and is this association explained by educational attainment? Findings In a matched-group cohort study that followed 1060 Black and Hispanic children from high-poverty communities to age 37 years, Child-Parent Center preschool participation was associated with a 20% reduction in cardiovascular disease risk, as measured by 30-year general FRS and hard FRS. The number of years of education by age 34 years partially mediated program-FRS associations, accounting for 23% of these observed and adjusted differences. Meaning These findings suggest that a comprehensive and established multilevel early childhood program may promote cardiovascular health in midlife, which is associated with long-term risk of cardiovascular disease, the leading cause of death in the US.

.7% (G-FRS) and 11.3% (H-FRS). The normal and optimal values, respectively, for the ages of study members are 11.0% and 8.9% (G-FRS) and 5.6% and 4.3% (H-FRS). For women and men, the probability values for the CLS sample in comparison to FRS norms were as follows: A large literature describes the background, development, findings, and use of the FRS in many populations. [13][14][15][16][17][18][19][20][21][22][23] For example, scores predict equally for White and Black adults. 13 The calculated scores correlated highly with in-person FRS exam scores (r = .85; N = 286) regardless of whether cholesterol or BMI was included. The in-person exam had a total sample of 301 participants and was completed at the Northwestern University Department of Preventive Medicine (Feinberg School of Medicine) in Chicago from March 24, 2017 to December 21, 2019. We further note that self-reported BMI was highly corrected with inperson exam measurement (r = .85) for the 286 participants with both sets of scores. Body composition and waist circumference were also highly correlated with BMI reports.

Educational Attainment Mediators at Age 34
Our primary measure for assessing mediation was years of completed education (range = 7 to 22) by May 2014 (mean age 34.1 years). The value 12 denotes a high school diploma or equivalent credential; and 14, 16, 18, and 20 Associate's through Doctorate degree (Post-Doctorate is possible). The two alternative indicators were (a) high school completion by diploma or equivalent (GED) and any college attendance (versus high school dropout), and (b) earned Associate's or Bachelor's degree or higher).
Education is one of the most comprehensively measured indicators in the CLS. 4 4 For high school completion status, 114 participants reported postsecondary education attendance, but they are missing on whether they completed high school via diploma or GED. Their types (high school graduation or GED) of high school completion were estimated based on available information from other data sources, including ISEG, NSC, CPS, and DCFS. For years of education, 2 participants are missing last grade they completed before they dropped out of school.
For the present study 1,025 of the 1,060 study sample members had a valid value for educational attainment at age 34. In the total CLS cohort, 1,473 participants had available data on any level of educational attainment. Of these, 1,397 were defined as being in the educational attainment sample at age 34. They met the following criteria:

Mediation Procedure
Following existing methodological practice 24,25 and prior CLS research, 1,6,8 we used the difference-indifference model of mediation--referred to as the "percentage reduction" (PR) approach--to assess whether educational attainment accounted for at least part of the observed group differences. In this method, the adjusted group difference with the hypothesized mediator ( _ ) is subtracted from the adjusted group difference without the mediator B ("main effect model") and then expressed as a percentage reduction over the main effect model. Ranging from 0 to 100 percent, values larger than 100 percent are taken as 100 percent. Negative calculated values, often do to suppressor effects, are reported as 0. Thus the calculation is as follow: For example, if the "main effect" for G-FRS is 3 points and then is reduced to 2 points under the mediation model (e.g., educational attainment added), PR would be 33%. This would denote a sizeable reduction in the observed difference or partial mediation but far short of complete or full mediation. Of course, any one mediator would not be expected to make a large contribution given that multiple mediators are most likely and influences are both indirect and direct. The descriptive PR method is often a first step to more complex (but with many assumptions) methods ranging from path analysis to structural equation modeling via full information maximum likelihood estimation. 1,8 Baseline Covariates for Estimating Impacts Based on administrative records from multiple sources and parent surveys, 6,8,10 17 variables were included as model covariates. They were measured primarily from birth to age 3 as baseline characteristics. Two significant differences were detected between groups: CPC participants grew up in higher poverty neighborhoods and their parents had higher rates of high school completion (but not college attendance). Sample attrition from the original cohort is shown in eTable 1. The covariates were as follows (all dichotomous with one exception):

Inverse Propensity Score Weighting
Following previous CLS reports, 4,6,7,9 Inverse Propensity Score Weighting (IPW) was used to adjust for potential attrition bias. About one quarter of the cohort were missing on FRS either because they did not complete the survey or had insufficient information to calculate a score. IPW methods can reduced attrition bias arising from measurable factor influencing sample recovery status. 26 The regression models included the following weight variable: The predicted probabilities of sample recovery (SR; age 37 survey) were estimated by logit regression (linear regression yielded similar estimates) with 31 input predictors hypothesized or known to be important. These included birth outcomes and demographics (BD), home environment (HE), program (PR), school, and neighborhood factors (SN) such as poverty and the share of those aged . In the outcome regressions, this weight was applied such that individuals with higher weights were counted more heavily in program estimates, as they have lower probabilities of responding to the adult survey. The lower weighted were counted less. Standard errors are adjusted to account for these weights. Robustness analyses with this and alternative model are shown in eTable 2.