Assessment of regression models for adjustment of iron status biomarkers for inflammation in children with moderate acute malnutrition in Burkina Faso

TY - JOUR

T1 - Assessment of regression models for adjustment of iron status biomarkers for inflammation in children with moderate acute malnutrition in Burkina Faso

AU - Cichon, Bernardette

AU - Ritz, Christian

AU - Fabiansen, Christian

AU - Christensen, Vibeke Brix

AU - Filteau, Suzanne

AU - Friis, Henrik

AU - Kæstel, Pernille

N1 - CURIS 2017 NEXS 028

PY - 2017

Y1 - 2017

N2 - Background: Biomarkers of iron status are affected by inflammation. In order to interpret them in individuals with inflammation, the use of correction factors (CFs) has been proposed. Objective: The objective of this study was to investigate the use of regression models as an alternative to the CF approach. Methods: Morbidity data were collected during clinical examinations with morbidity recalls in a cross-sectional study in children aged 6-23 mo with moderate acute malnutrition. C-reactive protein (CRP), α1-acid glycoprotein (AGP), serum ferritin (SF), and soluble transferrin receptor (sTfR) were measured in serum. Generalized additive, quadratic, and linear models were used to model the relation between SF and sTfR as outcomes and CRP and AGP as categorical variables (model 1; equivalent to the CF approach), CRP and AGP as continuous variables (model 2), or CRP and AGP as continuous variables and morbidity covariates (model 3) as predictors. The predictive performance of the models was compared with the use of 10-fold crossvalidation and quantified with the use of root mean square errors (RMSEs). SF and sTfR were adjusted with the use of regression coefficients from linear models. Results: Crossvalidation revealed no advantage to using generalized additive or quadratic models over linear models in terms of the RMSE. Linear model 3 performed better than models 2 and 1. Furthermore, we found no difference in CFs for adjusting SF and those from a previous meta-analysis. Adjustment of SF and sTfR with the use of the best-performing model led to a 17% point increase and < 1% point decrease, respectively, in estimated prevalence of iron deficiency. Conclusion: Regression analysis is an alternative to adjust SF and may be preferable in research settings, because it can take morbidity and severity of inflammation into account. In clinical settings, the CF approach may be more practical. There is no benefit from adjusting sTfR.

AB - Background: Biomarkers of iron status are affected by inflammation. In order to interpret them in individuals with inflammation, the use of correction factors (CFs) has been proposed. Objective: The objective of this study was to investigate the use of regression models as an alternative to the CF approach. Methods: Morbidity data were collected during clinical examinations with morbidity recalls in a cross-sectional study in children aged 6-23 mo with moderate acute malnutrition. C-reactive protein (CRP), α1-acid glycoprotein (AGP), serum ferritin (SF), and soluble transferrin receptor (sTfR) were measured in serum. Generalized additive, quadratic, and linear models were used to model the relation between SF and sTfR as outcomes and CRP and AGP as categorical variables (model 1; equivalent to the CF approach), CRP and AGP as continuous variables (model 2), or CRP and AGP as continuous variables and morbidity covariates (model 3) as predictors. The predictive performance of the models was compared with the use of 10-fold crossvalidation and quantified with the use of root mean square errors (RMSEs). SF and sTfR were adjusted with the use of regression coefficients from linear models. Results: Crossvalidation revealed no advantage to using generalized additive or quadratic models over linear models in terms of the RMSE. Linear model 3 performed better than models 2 and 1. Furthermore, we found no difference in CFs for adjusting SF and those from a previous meta-analysis. Adjustment of SF and sTfR with the use of the best-performing model led to a 17% point increase and < 1% point decrease, respectively, in estimated prevalence of iron deficiency. Conclusion: Regression analysis is an alternative to adjust SF and may be preferable in research settings, because it can take morbidity and severity of inflammation into account. In clinical settings, the CF approach may be more practical. There is no benefit from adjusting sTfR.

KW - Faculty of Science

KW - Inflammation

KW - Correction factors

KW - C-reactive protein

KW - Iron deficiency

KW - Regression analysis

KW - Serum ferritin

KW - Soluble transferrin receptor

KW - Young children

U2 - 10.3945/jn.116.240028

DO - 10.3945/jn.116.240028

M3 - Journal article

C2 - 27881597

SN - 0022-3166

VL - 147

SP - 125

EP - 132

JO - Journal of Nutrition

JF - Journal of Nutrition

IS - 1

ER -