Accuracy of a mixed effects model interpolation technique for the estimation of pregnancy weight values.
methods
missing data
pregnancy
weight
Journal
Journal of epidemiology and community health
ISSN: 1470-2738
Titre abrégé: J Epidemiol Community Health
Pays: England
ID NLM: 7909766
Informations de publication
Date de publication:
08 2019
08 2019
Historique:
received:
25
05
2018
revised:
17
04
2019
accepted:
04
05
2019
pubmed:
4
6
2019
medline:
28
11
2020
entrez:
2
6
2019
Statut:
ppublish
Résumé
Interpolation of missing weight values is sometimes used in studies of gestational weight gain, but the accuracy of these methods has not been established. Our objective was to assess the accuracy of estimated weight values obtained by interpolating from the nearest observed weight values and by linear and spline regression models when compared with measured weight values. The study population included participants enrolled in the LIFECODES cohort at Brigham and Women's Hospital. We estimated weights at 28 (n=764) and 40 (n=382) weeks of gestation using participants' two nearest observed weights and subject-specific slopes and intercepts derived from repeated measures mixed effects models. In separate models, gestational age was parameterised as a linear and restricted cubic spline variable. Mean differences, absolute error measures and correlation coefficients comparing observed and estimated weights were calculated. Mean differences and mean absolute error for weights derived from the 28-week linear model (0.18 lbs (SD 6.92), 2.73 lbs (SD 6.35)) and 40-week linear model (-0.40 lbs (SD 5.43) and 2.84 lbs (SD 4.65)) were low. Mean differences were somewhat greater at 28 weeks for weight values derived from the nearest two observed values (mean difference -1.97 lbs (SD 8.74)) and from spline models (mean difference -2.25 lbs (SD 7.13)). Results were similar at 40 weeks. Overall, weight values estimated using this interpolation approach showed good agreement with observed values. When repeated measures of weight are available, mixed effects models may be used to interpolate of missing weight values with minimal error.
Sections du résumé
BACKGROUND
Interpolation of missing weight values is sometimes used in studies of gestational weight gain, but the accuracy of these methods has not been established. Our objective was to assess the accuracy of estimated weight values obtained by interpolating from the nearest observed weight values and by linear and spline regression models when compared with measured weight values.
METHODS
The study population included participants enrolled in the LIFECODES cohort at Brigham and Women's Hospital. We estimated weights at 28 (n=764) and 40 (n=382) weeks of gestation using participants' two nearest observed weights and subject-specific slopes and intercepts derived from repeated measures mixed effects models. In separate models, gestational age was parameterised as a linear and restricted cubic spline variable. Mean differences, absolute error measures and correlation coefficients comparing observed and estimated weights were calculated.
RESULTS
Mean differences and mean absolute error for weights derived from the 28-week linear model (0.18 lbs (SD 6.92), 2.73 lbs (SD 6.35)) and 40-week linear model (-0.40 lbs (SD 5.43) and 2.84 lbs (SD 4.65)) were low. Mean differences were somewhat greater at 28 weeks for weight values derived from the nearest two observed values (mean difference -1.97 lbs (SD 8.74)) and from spline models (mean difference -2.25 lbs (SD 7.13)). Results were similar at 40 weeks.
CONCLUSIONS
Overall, weight values estimated using this interpolation approach showed good agreement with observed values. When repeated measures of weight are available, mixed effects models may be used to interpolate of missing weight values with minimal error.
Identifiants
pubmed: 31152073
pii: jech-2018-211094
doi: 10.1136/jech-2018-211094
doi:
Types de publication
Journal Article
Research Support, N.I.H., Extramural
Langues
eng
Sous-ensembles de citation
IM
Pagination
786-792Subventions
Organisme : NICHD NIH HHS
ID : T32 HD052458
Pays : United States
Informations de copyright
© Author(s) (or their employer(s)) 2019. No commercial re-use. See rights and permissions. Published by BMJ.
Déclaration de conflit d'intérêts
Competing interests: None declared.