Multi-Institutional Sharing of Electronic Health Record Data to Assess Childhood Obesity

Abstract

Objective: To evaluate the validity of multi-institutional electronic health record (EHR) data sharing for surveillance and study of childhood obesity.

Methods: We conducted a non-concurrent cohort study of 528,340 children with outpatient visits to six pediatric academic medical centers during 2007-08, with sufficient data in the EHR for body mass index (BMI) assessment. EHR data were compared with data from the 2007-08 National Health and Nutrition Examination Survey (NHANES).

Results: Among children 2-17 years, BMI was evaluable for 1,398,655 visits (56%). The EHR dataset contained over 6,000 BMI measurements per month of age up to 16 years, yielding precise estimates of BMI. In the EHR dataset, 18% of children were obese versus 18% in NHANES, while 35% were obese or overweight versus 34% in NHANES. BMI for an individual was highly reliable over time (intraclass correlation coefficient 0.90 for obese children and 0.97 for all children). Only 14% of visits with measured obesity (BMI ≥95%) had a diagnosis of obesity recorded, and only 20% of children with measured obesity had the diagnosis documented during the study period. Obese children had higher primary care (4.8 versus 4.0 visits, p<0.001) and specialty care (3.7 versus 2.7 visits, p<0.001) utilization than non-obese counterparts, and higher prevalence of diverse co-morbidities. The cohort size in the EHR dataset permitted detection of associations with rare diagnoses. Data sharing did not require investment of extensive institutional resources, yet yielded high data quality.

Conclusions: Multi-institutional EHR data sharing is a promising, feasible, and valid approach for population health surveillance. It provides a valuable complement to more resource-intensive national surveys, particularly for iterative surveillance and quality improvement. Low rates of obesity diagnosis present a significant obstacle to surveillance and quality improvement for care of children with obesity.

Figure 1. Evaluable Population for Obesity Analyses.

Development of the dataset for obesity-related analyses, showing the number of evaluable children and visits at each step. Percentages at each step are calculated relative to totals in the prior step. Since patients may have both primary care and specialty visits, subject counts at this step do not sum to the prior total; these values are marked with an asterisk.

Figure 2. Comparison of EHR and NHANES 2007–8 Cohorts.

Average measured BMIs for children of both sexes at each month of age from 2–17 years in the multi-institutional EHR cohort and in the NHANES 2007–8 cohort. In addition to individual points, curves fitted to each dataset by cubic polynomial regression are shown.

Figure 3. Diagnosis of Obesity at Outpatient Visits.

Percentages of children who were obese at any time during the study period, and diagnosed as obese at any visit to the indicated specialty. All specialties with a diagnosis rate ≥4% are included.

Figure 4. Obesity-Related Co-Morbidities.

Standardized morbidity ratios (observed prevalence in obese children/expected prevalence from entire cohort) with 95% confidence intervals for diagnostic groups (EDCs) having SMR >1.5 and CI₉₅>1.0 among children with measured obesity. N = total number of children in cohort with diagnosis in that EDC.