Dataset Information

Zero problems with compositional data of physical behaviors: a comparison of three zero replacement methods.

ABSTRACT: BACKGROUND:Researchers applying compositional data analysis to time-use data (e.g., time spent in physical behaviors) often face the problem of zeros, that is, recordings of zero time spent in any of the studied behaviors. Zeros hinder the application of compositional data analysis because the analysis is based on log-ratios. One way to overcome this challenge is to replace the zeros with sensible small values. The aim of this study was to compare the performance of three existing replacement methods used within physical behavior time-use epidemiology: simple replacement, multiplicative replacement, and log-ratio expectation-maximization (lrEM) algorithm. Moreover, we assessed the consequence of choosing replacement values higher than the lowest observed value for a given behavior. METHOD:Using a complete dataset based on accelerometer data from 1310 Danish adults as reference, multiple datasets were simulated across six scenarios of zeros (5-30% zeros in 5% increments). Moreover, four examples were produced based on real data, in which, 10 and 20% zeros were imposed and replaced using a replacement value of 0.5?min, 65% of the observation threshold, or an estimated value below the observation threshold. For the simulation study and the examples, the zeros were replaced using the three replacement methods and the degree of distortion introduced was assessed by comparison with the complete dataset. RESULTS:The lrEM method outperformed the other replacement methods as it had the smallest influence on the structure of relative variation of the datasets. Both the simple and multiplicative replacements introduced higher distortion, particularly in scenarios with more than 10% zeros; although the latter, like the lrEM, does preserve the ratios between behaviors with no zeros. The examples revealed that replacing zeros with a value higher than the observation threshold severely affected the structure of relative variation. CONCLUSIONS:Given our findings, we encourage the use of replacement methods that preserve the relative structure of physical behavior data, as achieved by the multiplicative and lrEM replacements, and to avoid simple replacement. Moreover, we do not recommend replacing zeros with values higher than the lowest observed value for a behavior.

SUBMITTER: Rasmussen CL

PROVIDER: S-EPMC7542467 | biostudies-literature | 2020 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Zero problems with compositional data of physical behaviors: a comparison of three zero replacement methods.

Rasmussen Charlotte Lund CL Palarea-Albaladejo Javier J Johansson Melker Staffan MS Crowley Patrick P Stevens Matthew Leigh ML Gupta Nidhi N Karstad Kristina K Holtermann Andreas A

The international journal of behavioral nutrition and physical activity 20201006 1

<h4>Background</h4>Researchers applying compositional data analysis to time-use data (e.g., time spent in physical behaviors) often face the problem of zeros, that is, recordings of zero time spent in any of the studied behaviors. Zeros hinder the application of compositional data analysis because the analysis is based on log-ratios. One way to overcome this challenge is to replace the zeros with sensible small values. The aim of this study was to compare the performance of three existing replac ...[more]

PMID: 33023619

Similar Datasets

Project description:We take a functional data approach to longitudinal studies with complex bivariate outcomes. This work is motivated by data from a physical activity study that measured 2 responses over time in 5-minute intervals. One response is the proportion of time active in each interval, a continuous proportions with excess zeros and ones. The other response, energy expenditure rate in the interval, is a continuous variable with excess zeros and skewness. This outcome is complex because there are 3 possible activity patterns in each interval (inactive, partially active, and completely active), and those patterns, which are observed, induce both nonrandom and random associations between the responses. More specifically, the inactive pattern requires a zero value in both the proportion for active behavior and the energy expenditure rate; a partially active pattern means that the proportion of activity is strictly between zero and one and that the energy expenditure rate is greater than zero and likely to be moderate, and the completely active pattern means that the proportion of activity is exactly one, and the energy expenditure rate is greater than zero and likely to be higher. To address these challenges, we propose a 3-part functional data joint modeling approach. The first part is a continuation-ratio model to reorder the ordinal valued 3 activity patterns. The second part models the proportions when they are in interval (0,1). The last component specifies the skewed continuous energy expenditure rate with Box-Cox transformations when they are greater than zero. In this 3-part model, the regression structures are specified as smooth curves measured at various time points with random effects that have a correlation structure. The smoothed random curves for each variable are summarized using a few important principal components, and the association of the 3 longitudinal components is modeled through the association of the principal component scores. The difficulties in handling the ordinal and proportional variables are addressed using a quasi-likelihood type approximation. We develop an efficient algorithm to fit the model that also involves the selection of the number of principal components. The method is applied to physical activity data and is evaluated empirically by a simulation study.

Project description:Congenital heart defects (CHD) is one of the most common birth defects in China. Many studies have examined risk factors for CHD, but their predictive abilities have not been evaluated. In particular, few studies have attempted to predict risks of CHD from, necessarily unbalanced, population-based cross-sectional data. Therefore, we developed and validated machine learning models for predicting, before and during pregnancy, women's risks of bearing children with CHD. We compared the results of these models in a large-scale, comprehensive population-based retrospective cross-sectional epidemiological survey of birth defects in six counties in Shanxi Province, China, covering 2006 to 2008. This contained 78 cases of CHD among 33831 live births. We constructed nine synthetic variables to use in the models: maternal age, annual per capita income, family history, maternal history of illness, nutrition and folic acid deficiency, maternal illness in pregnancy, medication use in pregnancy, environmental risk factors in pregnancy, and unhealthy maternal lifestyle in pregnancy. The machine learning algorithms Weighted Support Vector Machine (WSVM) and Weighted Random Forest (WRF) were trained on, and a logistic regression (Logit) was fitted to, two-thirds of the data. Their predictive abilities were then tested in the remaining data. True positive rate (TPR), true negative rate (TNR), accuracy (ACC), area under the curves (AUC), G-means, and Weighted accuracy (WTacc) were used to compare the classification performance of the models. Median values, from repeating the data partitioning 1000 times, were used in all comparisons. The TPR and TNR of the three classifiers were above 0.65 and 0.93, respectively, better than any reported in the literature. TPR, wtACC, AUC and G were highest for WSVM, showing that it performed best. All three models are precise enough to identify groups at high risk of CHD. They should all be considered for future investigations of other birth defects and diseases.

Project description:BackgroundThe assessment of methods for analyzing over-dispersed zero inflated count outcome has received very little or no attention in stratified cluster randomized trials. In this study, we performed sensitivity analyses to empirically compare eight methods for analyzing zero inflated over-dispersed count outcome from the Vitamin D and Osteoporosis Study (ViDOS) - originally designed to assess the feasibility of a knowledge translation intervention in long-term care home setting.MethodForty long-term care (LTC) homes were stratified and then randomized into knowledge translation (KT) intervention (19 homes) and control (21 homes) groups. The homes/clusters were stratified by home size (<250/> = 250) and profit status (profit/non-profit). The outcome of this study was number of falls measured at 6-month post-intervention. The following methods were used to assess the effect of KT intervention on number of falls: i) standard Poisson and negative binomial regression; ii) mixed-effects method with Poisson and negative binomial distribution; iii) generalized estimating equation (GEE) with Poisson and negative binomial; iv) zero inflated Poisson and negative binomial - with the latter used as a primary approach. All these methods were compared with or without adjusting for stratification.ResultsA total of 5,478 older people from 40 LTC homes were included in this study. The mean (=1) of the number of falls was smaller than the variance (=6). Also 72% and 46% of the number of falls were zero in the control and intervention groups, respectively. The direction of the estimated incidence rate ratios (IRRs) was similar for all methods. The zero inflated negative binomial yielded the lowest IRRs and narrowest 95% confidence intervals when adjusted for stratification compared to GEE and mixed-effect methods. Further, the widths of the 95% confidence intervals were narrower when the methods adjusted for stratification compared to the same method not adjusted for stratification.ConclusionThe overall conclusion from the GEE, mixed-effect and zero inflated methods were similar. However, these methods differ in terms of effect estimate and widths of the confidence interval.Trial registrationClinicalTrials.gov: NCT01398527. Registered: 19 July 2011.

Project description:Background/objectivesEvidence on the associations between lifestyle movement behaviors and obesity has been established without taking into account the time-constrained nature of categorized, time-based lifestyle behaviors. We examined the associations of sleep, sedentary behavior (SED), light-intensity physical activity (LPA), and moderate-to-vigorous PA (MVPA) with body mass index (BMI) using Compositional Data Analysis (CoDA), and compared the associations between a report-based method (24-h Physical Activity Recall; 24PAR) and a monitor-based method (SenseWear Armband; SWA).Subjects/methodsReplicate data from a representative sample of 1247 adults from the Physical Activity Measurement Survey (PAMS) were used in the study. Participants completed activity monitoring on two randomly selected days, each of which required wearing a SWA for a full day, and then completing a telephone-administered 24PAR the following day. Relationships among behavioral compositional parts and BMI were analyzed using CoDA via multiple linear regression models with both 24PAR and SWA data.ResultsUsing 24PAR, time spent in sleep (γ = -3.58, p = 0.011), SED (γ = 3.70, p = 0.002), and MVPA (γ = -0.53, p = 0.018) was associated with BMI. Using SWA, time spent in sleep (γ = -5.10, p < 0.001), SED (γ = 8.93, p < 0.001), LPA (γ = -3.12, p < 0.001), and MVPA (γ = -1.43, p < 0.001) was associated with BMI. The SWA models explained more variance in BMI (R2 = 0.28) compared with the 24PAR models (R2 = 0.07). The compositional isotemporal substitution models revealed reductions in BMI when replacing SED by MVPA, LPA (not with 24PAR) or sleep for both 24PAR and SWA, but the effect estimates were larger with SWA.ConclusionsFavorable levels of relative time spent in lifestyle movement behaviors were, in general, associated with decreased BMI. The observed associations were stronger using the monitor-based SWA method compared with the report-based 24PAR method.

Dataset Information

Zero problems with compositional data of physical behaviors: a comparison of three zero replacement methods.

Publications

Zero problems with compositional data of physical behaviors: a comparison of three zero replacement methods.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure