The perils of using predicted values in place of observed covariates: an example of predicted values of body composition and mortality risk

Abstract

Background: Several studies have assessed the relation of body composition to health outcomes by using values of fat and lean mass that were not measured but instead were predicted from anthropometric variables such as weight and height. Little research has been done on how substituting predicted values for measured covariates might affect analytic results.

Objectives: We aimed to explore statistical issues causing bias in analytical studies that use predicted rather than measured values of body composition.

Methods: We used data from 8014 adults ≥40 y old included in the 1999-2006 US NHANES. We evaluated the relations of predicted total body fat (TF) and predicted total body lean mass (TLM) with all-cause mortality. We then repeated the evaluation using measured body composition variables from DXA. Quintiles and restricted cubic splines allowed flexible modeling of the HRs in unadjusted and multivariable-adjusted Cox regression models.

Results: The patterns of associations between body composition and all-cause mortality depended on whether body composition was defined using predicted values or DXA measurements. The largest differences were observed in multivariable-adjusted models which mutually adjusted for both TF and TLM. For instance, compared with analyses based on DXA measurements, analyses using predicted values for males overestimated the HRs for TF in splines and in quintiles [HRs (95% CIs) for fourth and fifth quintiles compared with first quintile, DXA: 1.22 (0.88, 1.70) and 1.46 (0.99, 2.14); predicted: 1.86 (1.29, 2.67) and 3.24 (2.02, 5.21)].

Conclusions: It is important for researchers to be aware of the potential pitfalls and limitations inherent in the substitution of predicted values for measured covariates in order to draw proper conclusions from such studies.

Keywords: Berkson error; all-cause mortality; bias; body composition; epidemiology; prediction equations; research methods.

FIGURE 1

Density plots showing the distribution of TF and TLM in 1999–2004 NHANES data. Results are stratified by gender (male: n = 3142; female: n = 3236). SR, self-reported; TF, total body fat; TLM, total body lean mass.

FIGURE 2

Association of TF, predicted TF, and predicted TF based on SR variables with all-cause mortality. (A) An unadjusted model; (B) a multiple-variable model adjusting for age, height, ethnicity, alcohol consumption, physical activity, and smoking status; (C) an additional multiple-variable model mutually adjusting for total body lean mass as a categorical variable in quintiles (in addition to the covariates in model B). TF was modeled using a restricted cubic spline fit with knots at the 5^th, 35^th, 50^th, 65^th, and 95^th centiles. Results are stratified by gender (male: n = 3142; female: n = 3236). SR, self-reported; TF, total body fat.

FIGURE 3

Association of TLM, predicted TLM, and predicted TLM based on SR variables with all-cause mortality. (A) An unadjusted model; (B) a multiple-variable model adjusting for age, height, ethnicity, alcohol consumption, physical activity, and smoking status; (C) an additional multiple-variable model mutually adjusting for total body fat as a categorical variable in quintiles (in addition to the covariates in model B). TLM was modeled using a restricted cubic spline fit with knots at the 5^th, 35^th, 50^th, 65^th, and 95^th centiles. Results are stratified by gender (male: n = 3142; female: n = 3236). SR, self-reported; TLM, total body lean mass.

FIGURE 4

Plot showing transportability biases for TF and TLM when predicting these values for adults aged 50–75 y. One line shows a fitted prediction function fit using adults aged 18–75 y from the 1999–2000 NHANES (Cohort 1: males: n = 1368; females: n = 1635) and the other a fitted prediction function for adults restricted to ages 50–75 y from the 2005–2006 NHANES (Cohort 2: males: n = 353; females: n = 330). TF, total body fat; TLM, total body lean mass.