For this study, we used the MEPS, which is a set of large-scale, nationally representative surveys of families and individuals. MEPS collects data on the specific health services that families and individuals use, how frequently they use them, the cost of these services, and how they are paid. Also, patient socioeconomic and demographic information such as age, sex, family income, language spoken in the home, and self-reported recent health status are available.23 The MEPS data used in this study are patient- or a patient’s parent-reported information, which has limitations that we discuss below.
We combined the most recently available data (from 2007) with 2006 data to create a pooled sample of 11,798 unique patients who had clinic or/and hospital visits (except emergency department visits). We considered only patients within the age range of 3 through 17 years to reflect current guidelines for BP readings. MEPS does not provide whether patients have had or received at this examination a hypertension diagnosis. However, MEPS provides whether the examiner measured BP, and this decision is the focus of our study. The report of a BP measurement comes from the patient or the patient’s parent or guardian. The MEPS question was: “Doctor checked blood pressure.” Respondents had the option of answering “yes,”“no,” or “don’t know,” and among those answering yes or no, 34% said no. If this answer had come from medical professionals who did the examination themselves and were relying on both their memory and medical records, there would be little concern regarding the accuracy of their responses. Because the MEPS responses are from patients or parents relying on recall, the survey responses risk inaccuracy associated with three factors. First, the child, parent, or guardian may not know what a BP measurement is or involves. Second, the parent or other reporting adult may not have been present when the BP was taken. Third, the most recent clinical visit may have been sufficiently in the distant past to raise recall issues.
These potential sources of inaccuracy are mitigated by the survey option of answering “don’t know.” In constructing our sample of 11,798 respondents, we discarded 737 respondents (5.9% of total) who stated “don’t know,” thus reducing the number of potentially inaccurate responses due to the aforementioned factors. This narrows the notional confidence interval banding our estimate of 34% of children were unmeasured. However, an unknown number of respondents remain in our sample who answered “yes” or “no” when in fact they were wrong. These false-positive and false-negative responses, in themselves, generate uncertainty around our 34% estimate, but the presence of inaccurate responses, per se, does not bias our estimate unless there is a predominance of false-negative over false-positive responses or vice versa.
Using the MEPS, family histories of hypertension and diabetes variables were created by linking parents’ records to their children. Poverty level was classified using “family income as percent of poverty line” (poor, near poor, low income, middle income, and high income), and health insurance status was based on “health insurance coverage indicator” (private, public, and no insurance) in the dataset. The health status variable was generated using self-reported health status (ie, “less healthy than other children”). Because less-healthy children might be more likely to have the BP measured or clinic/hospital visits than healthy children, we included this variable as a proxy variable. We used BMI as a continuous variable because classifying BMI by degree of obesity (healthy weight: <85th percentile for age and sex, overweight: BMI of 85th–94th percentile, obese: BMI of ≥95th percentile) was not possible due to a lack of height and weight values for children.
Below we construct a logistic regression to model the behavior of examiners in deciding whether to measure a child’s BP. In the MEPS, BMI, an important indicator of the risk of hypertension,3–12 is reported only for children 6 years and older. Consequently, our regression examines a 6- through 17-year-old subsample of our overall sample spanning ages 3 through 17. Our smaller subsample consists of 7242 children and adolescents. Within this subsample of older children, 71.1% had their BP measured (as opposed to 66% for the whole sample) (Table I provides descriptive statistics for the regression subsample). This reflects examiners’ tendency to measure BP in older children. Within the regression subsample, the average age of children whose BP was measured was 12.37 years while the average age of children with unmeasured BP was a statistically significantly lower 11.05 years (P<.001). Similarly, the BMI of the measured group was statistically significantly higher than the unmeasured group (20.98 vs 19.72, P<.001). However, the difference in BMI between two groups may not be clinically significant. While there was no statistically significant difference between males and females in the percentage measured, both family histories of hypertension and diabetes were associated with BP measurement: 32.9% of those with measured BP had family histories of hypertension compared with 26.5% of those with unmeasured BP had family histories of hypertension (P<.001). Similarly, 9.9% of those with measured BP had family histories of diabetes compared with 7.4% in those with unmeasured BP (P<.026). Finally, among those with measured BP, 7.3% reported a recent history of illness, while 4.9% of those with unmeasured BP reported a recent history of illness (P<.001). Thus, it appears that examiners are selecting to measure the BP of patients at greater risk of hypertension, but these examiners also omit measuring BP in children in whom there is a surprising proportion with personal and family risk factors.
Table I. Descriptive Statistics for Logistic Regression Sample
|Variable||Overall (N=7242)||BP Measured (n=5149)||BP Unmeasured (n=2093)||t Test/Chi-Square Test|
|Mean (SE)/%||Mean (SE)/%||Mean (SE)/%|
|BP measured=1, other=0||71.1%||71.1%||28.9%|| |
|Age, y||12.011 (0.0501)||12.373 (0.058)||11.054 (0.092)||<.001|
|BMI,||20.638 (0.076)||20.982 (0.090)||19.723 (0.135)||<.001|
|Family history of hypertension=1, other=0||31.12||32.87||26.46||<.001|
|Family history of diabetes=1, other=0||9.2||9.87||7.4||.026|
|Race/ethnicity|| || || ||<.001|
| Caucasian (reference)||60.7||62.82||58.38|| |
| Black=1, other=0||15.34||16.55||15.86|| |
| Hispanic=1, other=0||17.74||15.15||20.88|| |
| Asian=1, other=0||2.42||1.94||0.37|| |
| Others=1, other=0||3.8||3.54||4.51|| |
|Poverty level|| || || ||<.001|
| Poor=1, other=0||14.78||14.39||15.84|| |
| Near poor =1, other=0||4.64||4.27||5.62|| |
| Low income=1, other=0||15.16||14.31||17.38|| |
| Middle income=1, other=0||34.79||34.61||35.27|| |
| High income (reference)||30.63||32.42||25.89|| |
|Health insurance status|| || || ||<.001|
| Private insurance (reference)||65.5||67.04||61.43|| |
| Public insurance=1, other=0||27.06||26.15||29.44|| |
| No insurance=1, other=0||7.45||6.81||9.14|| |
|Language other than English at home=1, English=0||11.94||10.47||15.86||<.001|
Adverse economic factors discourage the measurement of BP. The unmeasured group was poorer and less frequently insured. Grouping patients into 5 income-and-wealth categories, 32.4% of the measured group was high-income compared with 25.9% of the unmeasured group. In contrast, 14.4% of the measured group were poor and 4.3% were near-poor compared with the unmeasured group, where 15.8% were poor and 5.6% were-near poor. Within the measured group, 14.3% were low-income, while in the unmeasured group, 17.4% were low-income. A chi-square test showed that the distribution of the measured group across 5 income/wealth categories statistically significantly differed from the distribution of the unmeasured group (P<.001). The unmeasured group was also less well insured. Within the measured group, 67.0% had private health insurance, while 61.4% of the unmeasured group had private health insurance; 26.2% of the measured group had public health insurance, while 29.4% of the unmeasured group relied on public health insurance; and 6.8% of the measured group had no insurance, while 9.1% of the unmeasured group was uninsured. A chi-square test showed that the distribution of the measured group across 3 insurance categories statistically significantly differed from the distribution of the unmeasured group (P<.001). Thus, for ability-to-pay reasons, examiners appear to have placed some children in the unmeasured group and, thus, the unmeasured group appears not to have been populated based solely on the basis that the child was at lower risk for hypertension.
Social and language factors also differ between the measured and unmeasured groups. Within the measured group, 10.5% did not speak English in the home compared with 15.9% in the unmeasured group (P<.001). Similarly, 15.2% of the measured group were Hispanic while 20.9% of the unmeasured group were Hispanic. The measured group was disproportionately white (62.8% vs 58.4%), disproportionately Asian (1.9% vs 0.4%), disproportionately black (16.6% vs 15.9%), and less other (3.5% vs 4.5%). A chi-square test showed that the distribution of the measured group across 5 racial/ethnic categories statistically significantly differed from the distribution of the unmeasured group (P<.001). While non-black Hispanics were at greater risk for hypertension,12,19,20 examiners nonetheless disproportionately placed them in the unmeasured group. This may reflect a combined effect associated with ability-to-pay and language as well as ethnicity.
To sort out the independent effects of health, economic, and social factors in determining whether a child’s BP was taken, we construct a logistic regression estimating the odds that BP was measured. We hypothesized that examiners were more likely to measure BP if the patient manifested risks of hypertension, but the examiner was less likely to measure BP if the patient manifested inability-to-pay. We also hypothesize that patient English-language handicaps make it less likely that BP will be measured. We do not believe there was bias regarding racial/ethnic characteristics. The literature indicates that minority status is a risk factor for hypertension.12,19,20 Thus, examiners may be more likely to measure minority BP. However, controlling for ability-to-pay, language handicaps, and health factors, minority status may be associated with a poorer quality of local health services or limited family information regarding the health risks associated with hypertension. Such adverse supply or demand factors might lead to a failure to measure BP. All analyses including summary statistics were conducted with Stata 11 software (Stata Version 11; College Station, TX) to reflect the study design, sampling, and complexity of the MEPS.