This study examined the accuracy of self-reported weight, height and resultant BMI in the assessment of the prevalence of overweight in Chinese adolescents. The LOAs between the reported and measured values in this study were unacceptably wide. When reported values were used to classify individuals into BMI categories, the low sensitivity indicated that reported data may not be appropriate to screen for overweight adolescents. The degree of the discrepancy was not affected by gender, but it was associated with area of residence, age and BMI category.
On average, the adolescents' reported weights were underestimated, their heights were overestimated, and their resultant BMI was underestimated. These findings were similar to those in previous studies in adults [24
] and adolescents [16
]. The magnitude of the discrepancies in our study was moderate compared to existing studies in adolescents. In our study, the mean differences were -2.35 kg for weight, 1.36 cm for height and -1.23 kg/m2
for BMI. In comparison, a review of previous studies in adolescents showed a mean difference of -4.0 to 1.5 kg of weight, -1.1 to 6.9 cm of height and -3.0 to 0.2 of BMI value [30
The Pearson's correlation coefficient between the reported and measured values was high for weight, height and BMI. The Pearson's correlations for weight, height and resultant BMI were consistent with previous studies [16
]. The weighted Kappa statistics revealed a high level of agreement for weight and height, and substantial level for BMI, a pattern similar to that reported for the correlation coefficients [17
]. We also need to mention that quadratic weighted kappa coefficients tend to increase with the number of categories. But, after all, as the number of categories increases, so does the proportion of the variability in the true variable captured by the imperfect ordinal variable[46
]. Self-reported data could be considered for use in surveillance systems and large epidemiology studies, given the ease of data collection, its less resource-intensive nature, and high linear correlation and kappa statistics between reported and measured data. But we need to be cautious of the error of reported data, and correlation is a measure of association not agreement. The high correlation could not infer that reported data may be used interchangeably [26
In the present study, the sensitivity (56.1%) and specificity (98.6%) and positive predictive value (86.5%) in screening for overweight individuals were similar to those in a study of American adolescents [17
]. The lower sensitivity was also consistent with previous studies in children from Wales [21
], Greece [12
] and the Netherlands [14
]. These values were lower than those in found studies in three other American adolescent samples [18
]. Sensitivities in these three studies were about 70%, specificities were above 88%, and positive predictive values were above 80%.
The present study found no significant differences by gender between the self-reported and measured weight and height, however there were differences in BMI. The results for weight and height are consistent with several previous studies, but inconsistent in regards to BMI [12
]. However, an earlier study has observed a gender difference in reporting bias for weight and resultant BMI values [19
], and another in the correlation between self-reported and measured values for weight and resultant BMI [45
]. Age was associated with differences between self-reported and measured values in our study, with older adolescents more likely to exhibit bias than younger ones. This trend is consistent with some previous findings [12
], although one earlier study found no difference according to age [14
], while other research found the same trend for height but the opposite trend for weight [16
]. Children categorized as overweight or obese were more likely to underestimate their weight than normal children. This result is consistent with all previous reports [12
]. Household economic status was not associated with differences between the self-reported and measured values for weight, but it was associated with bias for height and resultant BMI. One previous study found results similar to our findings [16
], but another study reported that household economic status was not associated with the difference between self-reported and measured values [19
]. We found adolescents living in suburban areas had more bias in their self-reported anthropometric values than those living in urban areas of the city. Previous studies have no information about the effect of area of residence on the report error. It is beyond the scope of the current study to interpret the effects of household economic status and area of residency. Previous studies have indicated that adults [48
] and adolescents [49
] with a higher socioeconomic status are more concerned about body shape or other peoples' perceptions of their weight. Prior research also shows that rural students are less concerned about weight [50
]. The difference in concerns about weight may partly explain why household economic status and area of residency were associated with difference between reported and measured values in our study.
Most studies conclude that overweight and/or obese adolescents underreport their body weight and, thus, their resultant BMI, compared to adolescents of normal weight [30
]. We had similar findings. We also hypothesized that a population with a relatively thinner body shape or a population with fewer obese people might more accurately report their weight and height. However, the reported errors in our study were moderately to slightly larger than previous studies conducted in Western countries, where the rates of overweight and obesity were higher than in our study. Unexpectedly, we found that adolescents from the suburban areas of the city, a population with a lower prevalence of overweight and obesity [52
], reported their body weight and height less accurately compared to their urban counterparts, after adjustment for gender, age, household economic status and BMI status in multivariable linear regression models.
By adjusting the self-reported BMIs for socioeconomic variables, the sensitivity of screening for overweight individuals was increased from 56.1% to 86.6%, and the specificity decreased from 98.6% to 96.4%. Thus, the application of an adjusted formula results in a more accurate identification of overweight adolescents. Nevertheless, the sensitivity does not seem to be sufficient for the identification of overweight individuals even if the reported BMI is adjusted in this way. In addition, the use of the correction formula in this study, or other studies, is limited because the characteristics may differ in different populations or change over time.
One shortcoming of the present study was that our sample was drawn from schools, and adolescents who did not attend a school were not included. The results of the present study will not reflect this relatively small section of the adolescent population. In addition, there was a time interval of about one week between when the students answered the questionnaire and when they were measured. The height and weight of the adolescents may have changed during this week, although this change is likely minimal. Xi'an is a city located in central China. Since China is a vast nation characterized by social, economic, cultural and environmental diversity, the result of this study cannot be generalized to the whole country. However, it may be generalized to several neighbouring big cities that demonstrate similar qualities and patterns.