Dear Stata experts,
I have a repeated measures study design where subjects were followed over 5 visits and their HbA1c was measured throughout. There are some missing HbA1c values. I am able to generate the mixed commands, and plotted the graphs. Below is what I've obtained thus far.
What I'm struggling with is to obtain the slopes between each visit. I read from the Stata manual that the simple effects correspond to the slopes of each of the lines. So I attempted the following:
Looking at the Contrast values, it seems to be reflect the direction of the graph, but I'm not sure if these values are indeed the slope. However, the p-values are not what I'm after. What I'm interested in comparing is if the slope at each time point differs from the previous time-point and if this difference is statistically significant. Is it possible for this to be done?
Thanks so much!
I have a repeated measures study design where subjects were followed over 5 visits and their HbA1c was measured throughout. There are some missing HbA1c values. I am able to generate the mixed commands, and plotted the graphs. Below is what I've obtained thus far.
Code:
. mixed hba1c_ j##track || record_id:, var reml
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -1063.9138
Iteration 1: log restricted-likelihood = -1063.9138
Computing standard errors:
Mixed-effects REML regression Number of obs = 547
Group variable: record_id Number of groups = 145
Obs per group:
min = 2
avg = 3.8
max = 5
Wald chi2(9) = 327.99
Log restricted-likelihood = -1063.9138 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
hba1c_ | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
j |
1 | -2.531488 .3016913 -8.39 0.000 -3.122792 -1.940183
2 | -3.01946 .4786132 -6.31 0.000 -3.957525 -2.081395
3 | -2.524065 .4322982 -5.84 0.000 -3.371354 -1.676776
4 | -2.323791 .4422862 -5.25 0.000 -3.190656 -1.456926
|
track |
yes | .3795789 .3109946 1.22 0.222 -.2299593 .9891171
|
j#track |
1#yes | .2095378 .3739649 0.56 0.575 -.5234199 .9424954
2#yes | .4563514 .5540576 0.82 0.410 -.6295816 1.542284
3#yes | -.2356258 .4859038 -0.48 0.628 -1.18798 .7167281
4#yes | -.6247749 .4981435 -1.25 0.210 -1.601118 .3515685
|
_cons | 10.732 .2517275 42.63 0.000 10.23862 11.22538
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
record_id: Identity |
var(_cons) | .9210646 .1912886 .613071 1.383787
-----------------------------+------------------------------------------------
var(Residual) | 2.247272 .1607321 1.953327 2.585451
------------------------------------------------------------------------------
LR test vs. linear model: chibar2(01) = 48.43 Prob >= chibar2 = 0.0000
. margins, over(j track)
Predictive margins Number of obs = 547
Expression : Linear prediction, fixed portion, predict()
over : j track
------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
j#track |
0#no | 10.732 .2517275 42.63 0.000 10.23862 11.22538
0#yes | 11.11158 .1826223 60.84 0.000 10.75365 11.46951
1#no | 8.200512 .2539557 32.29 0.000 7.702768 8.698257
1#yes | 8.789629 .1867404 47.07 0.000 8.423625 9.155634
2#no | 7.71254 .4500516 17.14 0.000 6.830455 8.594625
2#yes | 8.54847 .2528818 33.80 0.000 8.052831 9.04411
3#no | 8.207935 .4004468 20.50 0.000 7.423073 8.992796
3#yes | 8.351888 .1877782 44.48 0.000 7.983849 8.719926
4#no | 8.408209 .4112091 20.45 0.000 7.602254 9.214164
4#yes | 8.163013 .196392 41.56 0.000 7.778092 8.547935
------------------------------------------------------------------------------
. marginsplot, x(j) title("Overall HbA1c trend split by Tracked group") ///
> xtitle("Visit") ytitle("Mean HbA1c")
Variables that uniquely identify margins: j track
What I'm struggling with is to obtain the slopes between each visit. I read from the Stata manual that the simple effects correspond to the slopes of each of the lines. So I attempted the following:
Code:
. contrast ar.j@track, pveffects
Contrasts of marginal linear predictions
Margins : asbalanced
-------------------------------------------------
| df chi2 P>chi2
--------------+----------------------------------
hba1c_ |
j@track |
(1 vs 0) no | 1 70.41 0.0000
(1 vs 0) yes | 1 110.41 0.0000
(2 vs 1) no | 1 1.03 0.3105
(2 vs 1) yes | 1 0.73 0.3937
(3 vs 2) no | 1 0.75 0.3855
(3 vs 2) yes | 1 0.48 0.4866
(4 vs 3) no | 1 0.16 0.6935
(4 vs 3) yes | 1 0.66 0.4160
Joint | 8 327.98 0.0000
-------------------------------------------------
------------------------------------------------------
| Contrast Std. Err. z P>|z|
--------------+---------------------------------------
hba1c_ |
j@track |
(1 vs 0) no | -2.531488 .3016913 -8.39 0.000
(1 vs 0) yes | -2.32195 .2209798 -10.51 0.000
(2 vs 1) no | -.4879724 .4811204 -1.01 0.310
(2 vs 1) yes | -.2411588 .2827586 -0.85 0.394
(3 vs 2) no | .4953947 .5709149 0.87 0.386
(3 vs 2) yes | -.1965825 .2825823 -0.70 0.487
(4 vs 3) no | .2002747 .5081325 0.39 0.693
(4 vs 3) yes | -.1888744 .2322142 -0.81 0.416
------------------------------------------------------
Thanks so much!
Comment