Aim 1b
Dan Leibel
3/11/2019
Significant 2-way interaction of Trails A x Age with HgbA1C, indicating that Trails A predicts change in A1C over time.
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: HgbA1C ~ LogTrailsA * cenAge + cenSex + cenRace + cenPovStat +
## cenEducnum + (cenAge | HNDid)
## Data: Tasneem_1b
## Control: lmerControl(check.nobs.vs.nRE = "ignore")
##
## REML criterion at convergence: 2893.1
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3166 -0.4914 -0.1867 0.3468 5.6995
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## HNDid (Intercept) 2.100003 1.44914
## cenAge 0.001254 0.03542 -1.00
## Residual 2.352845 1.53390
## Number of obs: 687, groups: HNDid, 345
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 8.68053 0.85942 328.83490 10.100 < 2e-16 ***
## LogTrailsA -0.45783 0.54829 330.81544 -0.835 0.40432
## cenAge 0.18450 0.08475 524.38377 2.177 0.02992 *
## cenSex 0.67565 0.20476 291.52716 3.300 0.00109 **
## cenRace 0.25691 0.20588 299.04994 1.248 0.21306
## cenPovStat 0.26803 0.20978 298.28714 1.278 0.20237
## cenEducnum -0.05831 0.04102 292.88434 -1.422 0.15623
## LogTrailsA:cenAge -0.12471 0.05435 532.62751 -2.295 0.02214 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) LgTrlA cenAge cenSex cenRac cnPvSt cnEdcn
## LogTrailsA -0.993
## cenAge -0.178 0.195
## cenSex 0.136 -0.131 0.051
## cenRace 0.168 -0.169 0.040 0.121
## cenPovStat 0.104 -0.100 0.008 0.036 -0.009
## cenEducnum -0.160 0.161 -0.087 -0.022 -0.110 0.179
## LgTrlsA:cnA 0.196 -0.217 -0.992 -0.042 -0.046 0.006 0.089This one is a little weird…Plotting the interaction effect shows that slower Trails A performance at baseline is associated with the declining HgbA1C over time, whereas faster Trails A performance is associated with rising A1C. I’ll do a little more probing of this effect. 
Next, we have a significant 3-way interaction of CES x EF (composite) x Age with HgbA1C, indicating that the CES and EF interact to predict change in glucose over time
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: HgbA1C ~ CES * EF * cenAge + cenSex + cenRace + cenPovStat +
## cenEducnum + (cenAge | HNDid)
## Data: Tasneem_1b
## Control: lmerControl(check.nobs.vs.nRE = "ignore")
##
## REML criterion at convergence: 2908.1
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3143 -0.4968 -0.1811 0.3516 5.6816
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## HNDid (Intercept) 2.130678 1.45968
## cenAge 0.001859 0.04312 -0.75
## Residual 2.290015 1.51328
## Number of obs: 680, groups: HNDid, 341
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 7.667e+00 1.894e-01 3.078e+02 40.492 < 2e-16 ***
## CES 1.757e-02 1.079e-02 3.115e+02 1.629 0.10442
## EF 4.610e-02 9.166e-02 2.833e+02 0.503 0.61534
## cenAge 3.679e-03 1.931e-02 1.361e+02 0.191 0.84919
## cenSex 6.839e-01 2.069e-01 2.651e+02 3.305 0.00108 **
## cenRace 3.059e-01 2.184e-01 2.797e+02 1.401 0.16236
## cenPovStat 2.159e-01 2.166e-01 2.834e+02 0.997 0.31971
## cenEducnum -5.076e-02 4.472e-02 2.872e+02 -1.135 0.25734
## CES:EF 7.924e-04 4.872e-03 2.920e+02 0.163 0.87091
## CES:cenAge -9.578e-04 1.092e-03 1.309e+02 -0.877 0.38217
## EF:cenAge 1.220e-02 8.622e-03 1.213e+02 1.415 0.15951
## CES:EF:cenAge -1.060e-03 4.825e-04 1.343e+02 -2.198 0.02968 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) CES EF cenAge cenSex cenRac cnPvSt cnEdcn CES:EF
## CES -0.823
## EF -0.043 -0.038
## cenAge -0.276 0.177 0.249
## cenSex -0.019 0.063 -0.059 0.037
## cenRace -0.059 0.100 0.143 -0.028 0.095
## cenPovStat 0.137 -0.127 0.058 0.087 0.023 -0.003
## cenEducnum -0.131 0.133 -0.221 -0.033 0.012 -0.172 0.107
## CES:EF -0.015 0.173 -0.788 -0.176 0.082 0.065 0.011 0.008
## CES:cenAge 0.180 -0.111 -0.180 -0.810 -0.014 0.037 -0.044 0.035 0.183
## EF:cenAge 0.248 -0.183 -0.211 -0.155 0.035 0.032 -0.004 -0.081 0.140
## CES:EF:cnAg -0.187 0.187 0.121 0.068 -0.034 -0.024 -0.023 0.056 -0.067
## CES:cA EF:cnA
## CES
## EF
## cenAge
## cenSex
## cenRace
## cenPovStat
## cenEducnum
## CES:EF
## CES:cenAge
## EF:cenAge 0.062
## CES:EF:cnAg 0.069 -0.828## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## -5.25085 -1.64835 -0.05227 -0.15709 1.27476 6.00450 76## [1] 2.16346## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## 0.00 6.00 13.00 14.91 22.00 46.00 46## [1] 10.4319This one is also weird. The following three graphs represent age-related change in A1C at low, average, and high levels of EF, with the graphs split at different levels of depression.
Low Depression
Mean Depression
High Depression
Next, we have a 3-way interaction of CES x Depression x Verbal Fluency with A1C
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula:
## HgbA1C ~ CES * FluencyWord * cenAge + cenSex + cenRace + cenPovStat +
## cenEducnum + (cenAge | HNDid)
## Data: Tasneem_1b
## Control: lmerControl(check.nobs.vs.nRE = "ignore")
##
## REML criterion at convergence: 2998.6
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3192 -0.5038 -0.1953 0.3490 5.7415
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## HNDid (Intercept) 2.145611 1.46479
## cenAge 0.001415 0.03761 -0.93
## Residual 2.287940 1.51259
## Number of obs: 701, groups: HNDid, 351
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 7.919e+00 6.857e-01 3.049e+02 11.549 < 2e-16
## CES -7.398e-03 3.681e-02 3.215e+02 -0.201 0.84087
## FluencyWord -1.691e-02 3.475e-02 2.835e+02 -0.486 0.62702
## cenAge -1.061e-01 6.679e-02 1.609e+02 -1.589 0.11411
## cenSex 6.264e-01 2.022e-01 2.702e+02 3.098 0.00216
## cenRace 2.038e-01 2.043e-01 2.838e+02 0.997 0.31950
## cenPovStat 1.266e-01 2.109e-01 2.938e+02 0.600 0.54892
## cenEducnum -2.758e-02 4.253e-02 2.944e+02 -0.648 0.51726
## CES:FluencyWord 1.461e-03 1.933e-03 3.052e+02 0.756 0.45038
## CES:cenAge 6.957e-03 3.931e-03 2.292e+02 1.770 0.07808
## FluencyWord:cenAge 5.714e-03 3.263e-03 1.495e+02 1.751 0.08203
## CES:FluencyWord:cenAge -4.016e-04 2.004e-04 2.160e+02 -2.004 0.04632
##
## (Intercept) ***
## CES
## FluencyWord
## cenAge
## cenSex **
## cenRace
## cenPovStat
## cenEducnum
## CES:FluencyWord
## CES:cenAge .
## FluencyWord:cenAge .
## CES:FluencyWord:cenAge *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) CES FlncyW cenAge cenSex cenRac cnPvSt cnEdcn
## CES -0.832
## FluencyWord -0.964 0.806
## cenAge -0.304 0.230 0.244
## cenSex 0.079 -0.044 -0.089 -0.030
## cenRace -0.102 0.038 0.090 -0.058 0.066
## cenPovStat 0.052 -0.036 -0.017 0.006 0.023 -0.047
## cenEducnum 0.137 0.004 -0.182 0.040 0.036 -0.100 0.164
## CES:FlncyWr 0.786 -0.960 -0.823 -0.178 0.059 -0.019 -0.005 0.044
## CES:cenAge 0.212 -0.175 -0.168 -0.828 0.052 0.082 0.056 -0.014
## FlncyWrd:cA 0.251 -0.191 -0.207 -0.962 0.045 0.047 0.014 -0.050
## CES:FlncW:A -0.168 0.133 0.135 0.778 -0.061 -0.077 -0.069 0.026
## CES:FlW CES:cA FlnW:A
## CES
## FluencyWord
## cenAge
## cenSex
## cenRace
## cenPovStat
## cenEducnum
## CES:FlncyWr
## CES:cenAge 0.126
## FlncyWrd:cA 0.151 0.809
## CES:FlncW:A -0.094 -0.964 -0.818
## fit warnings:
## Some predictor variables are on very different scales: consider rescalingPlots split by level of depression:
Low depression
Mean Depression
High Depression
Finally, there was a significant 3-way interaction of CES x EF (composite) x Age on Glucose
## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: Glucose ~ CES * EF * cenAge + cenSex + cenRace + cenPovStat +
## cenEducnum + (cenAge | HNDid)
## Data: Tasneem_1b
## Control: lmerControl(check.nobs.vs.nRE = "ignore")
##
## REML criterion at convergence: 7752.4
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5437 -0.4867 -0.1716 0.3046 3.6979
##
## Random effects:
## Groups Name Variance Std.Dev. Corr
## HNDid (Intercept) 2119.345 46.036
## cenAge 1.912 1.383 -0.91
## Residual 3478.933 58.982
## Number of obs: 683, groups: HNDid, 343
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 155.93877 6.49157 298.90219 24.022 < 2e-16 ***
## CES 0.46313 0.37007 305.13179 1.251 0.211723
## EF 1.51617 3.12145 299.18046 0.486 0.627518
## cenAge -0.08610 0.68028 163.06403 -0.127 0.899438
## cenSex 27.01572 7.06619 258.76074 3.823 0.000165 ***
## cenRace -10.65947 7.46644 269.41856 -1.428 0.154550
## cenPovStat 8.21437 7.40291 275.95139 1.110 0.268131
## cenEducnum -1.67701 1.53268 278.88815 -1.094 0.274826
## CES:EF 0.05477 0.16619 299.11730 0.330 0.741973
## CES:cenAge -0.06334 0.03838 157.68194 -1.650 0.100851
## EF:cenAge 0.14859 0.30143 146.79076 0.493 0.622776
## CES:EF:cenAge -0.03456 0.01688 163.33734 -2.047 0.042280 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr) CES EF cenAge cenSex cenRac cnPvSt cnEdcn CES:EF
## CES -0.823
## EF -0.062 -0.025
## cenAge -0.295 0.192 0.255
## cenSex -0.018 0.063 -0.065 0.038
## cenRace -0.059 0.099 0.144 -0.035 0.095
## cenPovStat 0.137 -0.133 0.061 0.092 0.026 -0.005
## cenEducnum -0.131 0.133 -0.215 -0.034 0.008 -0.175 0.110
## CES:EF -0.002 0.163 -0.789 -0.179 0.086 0.066 0.004 0.002
## CES:cenAge 0.195 -0.128 -0.182 -0.811 -0.012 0.043 -0.052 0.036 0.178
## EF:cenAge 0.254 -0.186 -0.225 -0.159 0.037 0.030 -0.006 -0.084 0.153
## CES:EF:cnAg -0.190 0.184 0.133 0.072 -0.035 -0.020 -0.028 0.059 -0.086
## CES:cA EF:cnA
## CES
## EF
## cenAge
## cenSex
## cenRace
## cenPovStat
## cenEducnum
## CES:EF
## CES:cenAge
## EF:cenAge 0.067
## CES:EF:cnAg 0.059 -0.829Plots split by level of depression:
Low Depression
Mean Depression
High Depression
