Is it time to stop sweeping data cleaning under the carpet? A novel algorithm for outlier management in growth data

Abstract

All data are prone to error and require data cleaning prior to analysis. An important example is longitudinal growth data, for which there are no universally agreed standard methods for identifying and removing implausible values and many existing methods have limitations that restrict their usage across different domains. A decision-making algorithm that modified or deleted growth measurements based on a combination of pre-defined cutoffs and logic rules was designed. Five data cleaning methods for growth were tested with and without the addition of the algorithm and applied to five different longitudinal growth datasets: four uncleaned canine weight or height datasets and one pre-cleaned human weight dataset with randomly simulated errors. Prior to the addition of the algorithm, data cleaning based on non-linear mixed effects models was the most effective in all datasets and had on average a minimum of 26.00% higher sensitivity and 0.12% higher specificity than other methods. Data cleaning methods using the algorithm had improved data preservation and were capable of correcting simulated errors according to the gold standard; returning a value to its original state prior to error simulation. The algorithm improved the performance of all data cleaning methods and increased the average sensitivity and specificity of the non-linear mixed effects model method by 7.68% and 0.42% respectively. Using non-linear mixed effects models combined with the algorithm to clean data allows individual growth trajectories to vary from the population by using repeated longitudinal measurements, identifies consecutive errors or those within the first data entry, avoids the requirement for a minimum number of data entries, preserves data where possible by correcting errors rather than deleting them and removes duplications intelligently. This algorithm is broadly applicable to data cleaning anthropometric data in different mammalian species and could be adapted for use in a range of other domains.

Screening & diagnosing errors in longitudinal measures of body size

“…Such simulation studies should mimic the types of errors that occur in datasets, otherwise results may have little generalisability when applied to datasets that have different error structures (49). Merely simulating normally distributed errors without age (date) errors, keystroke errors (29), duplications, internally inconsistent values etc is unrealistic, and such studies are unlikely to be a useful test of the performance of the method.…”

“…Logic checks based on data entry error mechanisms are another screening/diagnostic method and can be cross-sectional eg; digit errors (29), or longitudinal eg; measures carried forward (9). These checks are rarely reported, which might reflect their status as a routine part of quality control, but it may also reflect a lack of systematic rigor and transparency in data handling processes.…”

“…Lawman et al's., 2016 (8) review of approaches to identify height and weight BIVs was used as a starting to identify the different approaches that have been used for error detection. The 12 studies reported in their Table 1 (11-22) were screened along with all subsequent citations of Lawman et al (8) up to Oct 2020 (n=12) (3,9,(23)(24)(25)(26)(27)(28)(29)(30)(31)(32). The reference lists of these citing papers were also screened to identify methodological papers describing cleaning algorithms or approaches (n=5) (7, [33][34][35][36].…”

“…Longitudinal approaches can estimate the internal (within-subject) consistency of repeated values which means the entire distribution is screened and erroneous inliers detected. Comparisons with cross-sectional approaches showed improved sensitivity and specificity (28,29). Approaches include the use of change scores (13,18) and % change (3), residuals from OLS regression (28,32), multilevel models (MLM) (23,35) and non-parametric smoothing routines (9), conditional growth scores (36) and ratios of Euclidean distances between a set of 3 measures (3).…”

Plausible Outliers and Implausible Inliers

“…There is little evidence to support any standardized guidance for identifying implausible values in adults. More research is available for pediatric data, and the advantages and limitations of existing methods for children could be presented in guidelines while further advancement of these methods continues (2,(4)(5)(6)(7)(8)(9). Any guidelines regarding plausible outliers should reflect that the treatment of these values depends on the research question.…”