I am interested in examining whether my protein of interest is associated with cognitive decline over a period of time. In our study, participants were followed longitudinally for 8 years. At each visit (Baseline, 3 months, 12m, 24m, 36m,48m,60m,72m, 84m) participants underwent cognitive testing. Test scores are treated as continuous, repeated-measures. However, protein levels were only measured once at baseline. Therefore my independent variable would be my protein whilst cognitive test scores would be my dependent variable. However, I would like to control for covariates/ confounders such as age, gender and years of education. Finally, some of the participants missed cognitive testing at certain months – I hope this won’t affect the analyses. I spoke with a statistician and he recommended a linear mixed-effects model, however I am new to this type of modelling. I will use SPSS (V23) to run my analyses, however I have a few questions:
Google Marie Davidian NCSU longitudinal models course notes. These are excellent and contain examples including code . She developed those models following HIV drug treatment. Everything you need should be in these notes. Best wishes David Booth
The answers to this question may help
https://www.researchgate.net/post/Can_anyone_recommend_an_online_introduction_course_about_mixed_models_for_the_analysis_of_longitudinal_repeated_measure_data_in_R
Your protein of interest is only measured once so it is a time-invariant variable, as such it can only account for between person differences and not within-individual between occasions variation. I would start with time (month) as a fixed effect (month and month squared, month cubed ) and allow time to be random at level 2 (a random intercepts and slopes model). This will allow each person to have their own trajectory around the overall (potentially non-linear trend). Having established what is going on over time I would then put in the protein variable to see its overall effect. After modelling, you can then do some predictions to see what happens to people with low, medium and high protein levels - is it clinically important? You can then include other covariates to see what effect they have, and what effect there is on the estimated protein relation - the exact form of the model and its interpretation depends on the conceptual status of the variables be it confounder, moderator or mediator. The models are about calibrating relations.
This a very useful tutorial on the roles of covariates http://dagitty.net/learn/graphs/roles.html
The consequences are important:
"the covariates in a statistical analysis can have a variety of different roles from a causal inference perspective: they can be mediators, confounders, proxy confounders, or competing exposures. If a suitable set of covariates can be identified that removes confounding, we may proceed to estimate our causal effect using a multivariable regression model. In linear regression models, there are only two types of variables: the dependent variable (DV) and independent variables (IVs, or predictors). No further distinction is made between the IVs – specifically, the exposure is by no means a "special" IV and is treated just like any other IV. Thus, as you can see, there is a conceptual mismatch between causal theory (DAG) that leads us to formulate a multivariable regression model (that highlights the exposure-outcome relationship and associated statistical adjustment for confounding) and the regression model itself. This conceptual mismatch can easily lead to misinterpretation of the results from a multivariable regression model. "
(source: http://dagitty.net/learn/graphs/table2-fallacy.html)
The ability of the mixed model to analyse missing data refers to the response variable not being measured on each and every occasion. The model will give unbiased results providing the response is not NMAR. Finally, I would recommend that you contact someone who is familiar with these models.
Another possibility might be Cox regression and a Kaplan Meier survival curve. See the attached for more information. Best wishes David Booth
BTW some graphs may really be helpful here
Try some spaghetti plots for example. David Booth
If you wish to use R, or just understand these models in more detail then this is a useful resource: see chapter 4 Book Lme4: Mixed-Effects Modeling With R
It contains a sleep study dataset example which may be useful to go through.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA