Predicting Preclinical Disease by Using The Mixed‐Effects Regression Model

Abstract

Most methods of disease prediction are generally restricted to logistic or Cox proportional hazards model regression using only baseline values of the risk factor, and where the individuals age and sex account for most of the predictive capacity of the method. This chapter presents a method of risk prediction appropriate for repeated measures or time‐dependent risk factors from longitudinal studies. The method is based on posterior probabilities calculated from mixed‐effects regression models to model intra‐ and inter‐individual differences in the serial risk factor measurements, to predict a disease‐related outcome or other future event according to age changes within gender groups. While aging is inevitable and the occurrence of disease increases with age, disease and aging are not synonymous. Disease is a selective event that occurs in only some members of the population; others maintain a greater degree of physiological function as they age and remain relatively disease free. Thus, longitudinal data that considers each individual as their own control and the use of an appropriate method of prediction can play an important role in promoting a healthy human life throughout the aging process.

Gender differences in the accuracy of time-dependent blood pressure indices for predicting coronary heart disease: A random-effects modeling approach

“…The method is appropriate for repeated measurements on the same individuals over an extended period of observation 22,23. Following a method for predicting preclinical disease using linear mixed-effects models, longitudinal trajectories were estimated for each of the four BP indices over all study participants.…”

“…Random-effects terms in the model included intercept to account for between-person variability as well as possibly for time, and time2 as tested for inclusion by a likelihood-ratio χ 2 test. 29 The marginal distributions from the mixed-effects model for being a CHD event and non-event for each individual are then computed and using Bayes theorem posterior probabilities of having CHD are obtained at each serial BP measurement for each individual 22,23. Since it was not possible to perform a mixed-effects regression analysis on the entire VHM&PP population because of the enormous amount of required computer memory, a random sample was chosen representing approximately 5% of the VHM&PP study population to obtain the mixed-effects estimates for the subsequent CHD predictions.…”

“…In the present study, we applied a method of risk prediction based on posterior probabilities calculated from a mixed-effects regression model22,23 to allow for differences between average values of covariates with regard to sex and age groups, and individual longitudinal changes within each group to estimate the predictive accuracy of serial BP measurements for CHD. We used data from two prospective, community-dwelling cohort studies from the U.S. and Europe.…”

Adaptive data-driven selection of sequences of biological and cognitive markers in pre-clinical diagnosis of dementia

“…Mixed-effects models were used in earlier studies to model univariate or multivariate repeated (longitudinal) clinical data. [20][21][22][23][24][25][26][27][28][29][30][31] Because of their ability to integrate irregular sampling intervals and varying sequence lengths, mixed-effects models were applied for medical diagnosis with longitudinal data in general [32][33][34][35][36][37][38][39][40][41] and in the field of neurodegeneration in particular 23,38,41 . In recent years, mixedeffects models were implemented to derive flexible predictions based on variable subsets of measurements or dynamically updating predictions in case new measurements were collected.…”

“…In this study mixed-effects model-based estimation was embedded into linear (or quadratic, see Supplementary Methods 3-4) discriminant models to account for inter-subject differences [32][33][34][35]40,41,47 (Fig. 9A-B).…”