Change point models for cognitive tests using semi-parametric maximum likelihood

Abstract

Random-effects change point models are formulated for longitudinal data obtained from cognitive tests. The conditional distribution of the response variable in a change point model is often assumed to be normal even if the response variable is discrete and shows ceiling effects. For the sum score of a cognitive test, the binomial and the beta-binomial distributions are presented as alternatives to the normal distribution. Smooth shapes for the change point models are imposed. Estimation is by marginal maximum likelihood where a parametric population distribution for the random change point is combined with a non-parametric mixing distribution for other random effects. An extension to latent class modelling is possible in case some individuals do not experience a change in cognitive ability. The approach is illustrated using data from a longitudinal study of Swedish octogenarians and nonagenarians that began in 1991. Change point models are applied to investigate cognitive change in the years before death.

Cognitive Decline in a Colombian Kindred With Autosomal Dominant Alzheimer Disease

“…The likelihood approach was used to compare the models 23 based on the goodness-of-fit index as the Akaike information criterion and the Bayesian information criterion. 26 The lowest Bayesian information criterion was accepted as appropriate. The effect of sex, SES, educational level, vascular factors, tobacco or alcohol use, and APOE ε4 on cognition was analyzed using the MMSE score, CERAD subtest scores, and CERADts.…”

Joint Models for Discrete Longitudinal Outcomes in Aging Research

“…Investigating this behaviour with age as the timescale is problematic as it requires death as a reference point; see van den Hout et al . () for change point modelling using years to death as the timescale.…”

Distribution-Free Estimation of Local Growth Rates Around Interval Censored Anchoring Events

“…The 95% Wald confidence intervals from the midpoint imputation method actually had decreasing coverage probabilities when the sample size increased. In comparison with the normal models, (van den Hout et al, ) the proposed method tended to produce estimates with smaller bias and variance estimates, due to its increased robustness. As one would expect, parameter estimation performed best in the hypothetical situation of known .…”

“…The third method we tested was a model assuming the true anchoring event times follows a normal distribution with unknown mean and variance . The parameters were jointly estimated as described by van den Hout et al, (). Finally, we fitted the model assuming the true anchoring event time distribution was known, i.e., we used instead of in Step 2 to obtain parameter estimates.…”

“…More important, many of the imputation methods lack proper accommodation of the uncertainty associated with the estimate , and thus are prone to produce questionable inferences. An alternative approach is to conduct a joint modeling analysis by assuming a parametric distribution for T (Robinson et al, ; van den Hout et al, ). Such assumptions, however, are impossible to verify, and misspecified distributions could lead to biased estimation.…”