gms | German Medical Science

Introduction: LIFE child as a part of the "Leipzig Research Centre for Civilization Diseases" is a longitudinal cohort study aiming, inter alia, at monitoring normal development in children and adolescents from fetal life to adulthood. Extentensive assessments result in a wide range of discrete and continuous measurement values from "clinical examinations, questionnaires, and interviews and includes a collection of several types of biological materials" [1]. As an important part of the study, anthropometric dimensions are measured via classic methods, e.g. stadiometer or tape measure (ca. 15 items), but also via 3D body scanner technology (ca. 150 items).

Typically anthropometric data measured in children depend on age and gender. The age-value relationship is often non-linear and associated with the occurence of heterogeneity of variance.

Therefore it is difficult to detect outliers or systematic errors bases on analysis of the measured values. As a consquence reference value from large population based studies has been used to transform raw into standardized data. However there is only a limited number of reference values publicly available such as for height, weight, or BMI.

Methods: We address the problem of absent reference values by using the data itself as a reference sample.

Applying the LMS-method on a reference sample which is large enough is results in standard deviation or Z-scores. These scores are recommended by the WHO as as the "best system for analysis and presentation of anthropometric data" [2]. They can be used to reassess measurements age- and sex-independently and therefore allow intrapersonal (over time) and interpersonal comparisons. Therefore their usage is beneficial in quality control and data cleansing.

The VGAM package [3] provides functions which can be used for a first approximate solution using quantile regression. The functions return separate values for l, m, and s according the age and gender. These are the input values for the transformation by Cole generating standard devation scores. The l, m, and s values can be iteratively revised after each data cleansing cycle or new data were added.

Results: In LIFE child the current reference sample consists of more than 3000 visits. The l, m, and s values are generated by dedicated R-routines and stored in a relational database system.The transformation algorithm by Cole [4] is implement as database function which can we dynamically on all associated raw data. Hence conspiciuous values can be automatically reported. Subsequently reported values are corrected manually.

We applied this proposed method for quality control to classic anthropometric and 3D body scanner data. In the first run on the classic anthropometric values two third of the detected data were true positives. The before-and-after comparison of the associated scatterplots shows a remarkable improvement of data quality.

Discussion: The use of self created reference values provides a feasible way for data correction of non-standard age- and gender-dependent measurements. In case of an outlier the value should be reviewed in association with other values of the same or earlier visits of the participant.