Dear all,

I fitted a mixed-effects models in stata for the longitudinal analysis of bmi (body weight index) after differnet type of surgery to compare the course of two different groups (case and control), with random intercepts and random slopes, after documenting, with a likelihood ratio test, that this model had better fit than simpler ones (see Figure below).

Regarding time, a quadratic term was added, to account for a non-linear association of the bmi course over time with 95% CI and predictor variables age, sex and type_of_surgery. I am interested on the interaction of time and beeing a case.

The command:

Performing EM optimization:

Performing gradient-based optimization:

Iteration 0: log likelihood = -4635.5813

Iteration 1: log likelihood = -4635.5812

Computing standard errors:

Mixed-effects ML regression, Number of obs = 1654

Group variable: pid, Number of groups = 277

Obs per group: min = 2

avg = 6.0

max = 7

Wald chi2(8) = 3109.11

Log likelihood = -4635.5812, Prob > chi2 = 0.0000

My questions is about the interpretation of the coefficients which I tried to transfer from this post https://www.statalist.org/forums/for...xtmixed-output :

1. Are these interpretations correct?

2. One further question was, if it is necessary to use "reml" (restricted maximum likelihood option) or if "mle" (default, maximum likelihood estimation) is fine. The advantage of "mle" is the use of likelihoodratio test to justify the used model over others (e.g. the quadratic term for time) as far I understood. Using the one or the other does not change much.

3. Is the Figure adequate and shows that cases have a different course than controls?

Hope you can help me. Thank you in advance!

Kind regards

Martin

I fitted a mixed-effects models in stata for the longitudinal analysis of bmi (body weight index) after differnet type of surgery to compare the course of two different groups (case and control), with random intercepts and random slopes, after documenting, with a likelihood ratio test, that this model had better fit than simpler ones (see Figure below).

Regarding time, a quadratic term was added, to account for a non-linear association of the bmi course over time with 95% CI and predictor variables age, sex and type_of_surgery. I am interested on the interaction of time and beeing a case.

The command:

**mixed c.bmi c.time i.case c.time#c.time i.case#c.time c.age i.sex i.type_surgery || pid: time, stddev base****The STATA Output is:**Performing EM optimization:

Performing gradient-based optimization:

Iteration 0: log likelihood = -4635.5813

Iteration 1: log likelihood = -4635.5812

Computing standard errors:

Mixed-effects ML regression, Number of obs = 1654

Group variable: pid, Number of groups = 277

Obs per group: min = 2

avg = 6.0

max = 7

Wald chi2(8) = 3109.11

Log likelihood = -4635.5812, Prob > chi2 = 0.0000

My questions is about the interpretation of the coefficients which I tried to transfer from this post https://www.statalist.org/forums/for...xtmixed-output :

**time:**The coefficient in the **controls** that describes the bmi course over time is different significantly different from zero, indicating that there was a change over time in the BMI of the controls independently of the course of the cases. (independantly of sex, type_of_surgery and age)**case:**There is no evidence against the null hypothesis (p=0.569) of difference between the BMI of cases and controls **at baseline** (time = 0 months) independantly of sex, age, type_of_surgery.**case*time:**There is an interaction effect between being a case and time in regard to bmi, e.g. after 24 months there is a difference in the bmi of cases compared to controls of **“0.3900 + 24 * 0.0902 = 2.554”** independantly for type_surgery and sex and age.**sex:**at baseline (0 months) in the control group there is no difference in bmi of woman compared to man**age:**at baseline (0 months) there is some evidence that the bmi in the control group reduces about 0.0608 kg/m^2 per 1 increase of age i.e. 10 years: 0.608 smaller BMI.**type_surgery:**At baseline in the control group people having a smaller mean BMI in the Bypass group compared to the Banding group (-5.3)****My questions:****1. Are these interpretations correct?

2. One further question was, if it is necessary to use "reml" (restricted maximum likelihood option) or if "mle" (default, maximum likelihood estimation) is fine. The advantage of "mle" is the use of likelihoodratio test to justify the used model over others (e.g. the quadratic term for time) as far I understood. Using the one or the other does not change much.

3. Is the Figure adequate and shows that cases have a different course than controls?

Hope you can help me. Thank you in advance!

Kind regards

Martin

## Comment