Our analysis shows that in the practice we studied, clinical staff systematically overestimated length in children 0–23 months of age. We found that the magnitude of this bias increased with the length of the child. Overestimation included a component proportional to the length of the child (95.3%) plus a constant of 1.88 cm. This bias did not depend meaningfully on a particular clinician or research assistant, or on the age or sex of the child. Thus, clinicians may be underdiagnosing overweight in their practices, a potentially important issue in this era of rapidly rising rates of childhood obesity.
Obesity in childhood and adolescence is associated with both short- and long-term adverse outcomes, including both physical and psychosocial consequences. Children who are overweight tend to become overweight adults, and once present, obesity is notoriously hard to treat. Among school-age children and adolescents, Field and colleagues[12
] found that high normal weight status (BMI between the 50th and 84th percentile) in childhood predicted becoming overweight or obese and (among boys) having hypertension as an adult. Therefore, we believe that it is important to assess length accurately in all children, not only those above a set cut point.
Bias in clinical measurement may be due to several factors: The child may move on the paper after the clinician draws the tick mark at the top of the child’s head but before marking the feet; the paper may be wrinkled under the child but smoothed out before the clinician measures the difference between the points; the clinician may draw the tick marks in a diagonal; or the diameter of the pen may increase the length measurement. Also, clinicians measured length to the nearest quarter inch and researchers measured length to the nearest tenth of a centimeter. Such differences in precision might have added to the inaccuracy of clinical values. In this study we were not able to determine the cause of the observed measurement error.
To assess the degree to which clinical length measurement bias affects population estimates of weight-for-length, we applied the regression correction factor developed in this study to 6408 length measurements among children 0–23 months of age in the year 2000 participating in a height and weight surveillance system at 13 centers of a managed-care organization, of which the study center is one. Before application of the regression correction factor, 13.0% (832/6408) of children had weight-for-length less than the 5th percentile, and 4.7% (303/6408) were above the 95th percentile.[13
] After correction, 5.7% (368/6408) were below the 5th percentile while 9.3% (597/6408) were above the 95th percentile, proportions similar to recent national data.[14
Examining these data another way, one would detect only about half as many children in this population with weight-for-length exceeding the 95th percentile by using the paper-and-pencil method (303) compared with the gold standard (597, 303/597 = 51%). Conversely, one would “overdiagnose” about 2 times as many children in this population with weight-for-length below the 5th percentile by using the paper-and-pencil method (832) vs the gold standard (368, 832/368 = 2.3). These analyses demonstrate the marked biases in underestimating overweight, and in overestimating underweight, introduced by the paper-and-pencil measurement that is commonly used in pediatric practices.
A possible shortcoming of our study is that we recruited participants from a single health center. Because clinicians in other practices may measure with different protocols, generalizability may be limited. We note, however, that measurement bias did not depend on a particular clinician. Also, although the research assistants and medical assistants were blinded to each other’s measurements, the medical assistants knew that there was a length-measurement study in progress, which might have caused them to measure more accurately than usual. If this were the case, however, our regression equation would conservatively estimate the bias. Also, for the research measurement, 2 research assistants, with the help of the parent, measured length. The paper-and-pencil measurement involved only the medical assistant. It is possible that the paper-and-pencil method would be more accurate if more clinical staff were involved.
A further limitation of this study is that we did not explicitly assess interrater reliability. By adding a term for each research assistant to the model, however, we did assess the possibility that the 3 assistants had different measuring characteristics. This addition did not meaningfully improve the fit of the model, evidence that the observed bias did not depend on which research assistant did the measuring. Also, our data included children with clinical measurements from 47 cm to 91 cm; one should exercise caution in extrapolating the regression equation beyond these lengths.
Medical providers should be aware that using the paper-and-pencil method can lead to underestimates of overweight and exaggerated estimates of thinness. To improve the accuracy of length measurement, clinical practices should use standardized procedures with a recumbent length board to measure children under 2 years of age, at least for children whose initial paper-and-pencil measurement of length puts them at one extreme or the other.