Inter-Rater Reliability of Ohio School-Based Overweight and Obesity Surveillance Data

Abstract

Measurement of height and weight in large studies may force the use of multiple measurers. The purpose of this study was to evaluate the reliability of height, weight, and body mass index (BMI) measures collected by multiple measurers in a large, statewide BMI surveillance program. A random subsample of schools (n=30) was selected from schools that participated in the 2009 to 2010 Ohio third-grade Oral Health/BMI surveillance program. Children (n=1,189) were measured by multiple volunteer health professional measurers and again by a trained researcher, who was standard across all schools. Mean differences for height, weight, and BMI percentiles were calculated for BMI category classifications. Agreement was estimated by the reliability coefficient, McNemar's test, and Kappa statistic. Sensitivity, specificity, and positive and negative predictive values were estimated using the trained researcher measures as the reference. Overall mean differences (95% confidence interval) were 0.45 (0.41-0.48) cm for height, 0.07 (−0.01-0.15) kg for weight, and 1.37 (1.20-1.53) for BMI. The correlation coefficient for all three measures was over 0.9 (P<0.01), indicating a strong positive association between measures. BMI category classifications showed substantial reliability (Kappa range: 0.94-0.96). Percentage agreement ranged from 98% to 99% for all BMI categories, as did sensitivities and specificities. Positive predictive values for all BMI categories were approximately 97%, and close to 100% for negative predictive values. Reliability for height, weight, BMI percentile, and BMI classification was very high, supporting the use of multiple trained measurers in a statewide BMI surveillance program. Similar methods can be applied to other public health and clinical settings to improve anthropometric measurement reliability.