1 Appendix 1. Denominator models for computing of Inverse Probability of Death and Censoring Weights in Estimating Effects of Baseline Diabetes on Dementia Risks Mortality1 Censoring2 Estimate OR p Estimate OR p Baseline Covariates 1998 diabetes status .42 1.52 <.001 male .46 1.59 <.001 age .04 1.04 <.001 black -.52 0.60 <.001 Years of education .04 <.001 -.06 <.001 Baseline Wealth (reference=highest quintile) lowest quintile .03 .65 2nd -.12 .09 3rd .15 .01 4th -.02 .72 Year of Assessment (reference=2008) 2002 -.01 .90 .70 .23 <.001 2004 -.12 .80 .001 .01 .84 2006 .02 .92 .61 -.18 .004 Baseline BMI .05 1.05 <.001 Baseline Smoking .48 1.62 <.001 Time-varying covariates Marital Status (reference=married) Widowed .09 1.10 .09 Divorced/Separated .28 1.32 .002 Never married .39 1.48 .003 Currently working -.24 .79 .007 -.20 .05 Self-rated health .34 1.41 <.0001 .09 .005 Disability .70 2.01 <.0001 Wealth (reference=highest quintile) lowest quintile .16 1.48 <.001 2nd .05 1.33 0.61 3rd .02 1.28 0.67 4th -.01 1.24 0.78 BMI -.10 .91 <.001 Heart Problems .38 1.46 <.001 Prior wave memory -.63 .52 .002 -0.26 <.001 score Missing indicator 1 .26 1.30 .0 0.56 <.001 Missing indicator 2 0.51 .02 Note. Blank cells mean the effects were not significant at .10 level in numerator or denominator models. 1 Estimates were from logistic models of estimating a person’s probability of mortality based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person was alive and participated in the survey last wave. Participants were weighted by the inverse of the probability of not dying. 2 Estimates were from logistic models of estimating a person’s probability of being censored based on the person’s basic demographic variables and prior wave time-varying covariates, 2 given that this person participated last wave and was still alive this wave. Participants were weighted by the inverse of the probability of not censored. 3 Appendix 2. Denominator models for computing Inverse Probabiltiy of Death and Censoring Weights in Estimating Effects of Baseline Diabetes on Memory Scores mortality censoring Estimate OR p Estimate OR p Baseline Covariates 1998 diabetes status .42 1.52 <.001 male .46 1.58 <.001 age .04 1.04 <.001 black -.51 0.57 <.001 -.21 .81 .04 Years of education .04 <.001 -.04 .97 .002 Baseline Wealth (reference=highest quintile) lowest quintile .04 1.13 .62 2nd -.08 1.01 .21 3rd .17 1.29 .006 4th -.03 1.07 .71 Year of Assessment (reference=2008) 2002 -.01 0.89 .93 .24 1.42 <.001 2004 -.12 0.8 .002 .03 1.16 .58 2006 .02 0.91 .57 -.15 .96 .01 Baseline BMI .05 1.05 <.001 Baseline Smoking .47 1.6 <.001 Time-varying covariates Marital Status (reference=married) Widowed .09 1.09 .09 Divorced/Separated .29 1.33 .002 Never married .38 2.06 .005 Currently working -.24 0.8 .006 -.18 .84 .09 Self-rated health .34 1.39 <.001 .09 1.09 .007 Disability .70 2.05 <.001 Wealth (reference=highest quintile) lowest quintile .17 1.39 <.001 2nd .05 1.24 0.22 3rd .02 1.22 0.70 4th -.01 1.19 0.78 BMI -.10 0.9 <.001 Heart Problems .38 1.46 <.001 Prior wave memory score -.66 0.53 <.001 -.31 .74 <.001 Missing indicator 1 .25 1.28 .001 .42 1.52 <.001 Missing indicator 2 .63 1.89 .002 Note. Blank cells mean the effects were not significant at .10 level in numerator or denominator models. 1 Estimates were from logistic models of estimating a person’s probability of mortality based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person was alive and participated in the survey last wave. 2 Estimates were from logistic models of estimating a person’s probability of being censored based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person participated last wave and was still alive this wave. 4 Appendix 3. Denominator models for computing of Inverse Probability of Death, Censoring and exposure Weights in Estimating Effects of Incident Diabetes on Dementia Risks Mortali Censori Exposu ty1 ng2 re3 Estima Estima Estima OR p OR p OR te te te Baseline Covariates <.00 male .52 1.68 .40 1.49 1 <.00 age .04 1.04 -.03 .97 1 <.00 black -.50 .61 1 Years of <.00 <.00 .04 1.04 -.05 .95 education 1 1 Baseline Wealth (reference=highest quintile) lowest quintile 2nd 3rd 4th Year of Assessment (reference=2008) 2000 NA NA -.11 .87 <.00 2002 .01 .89 .73 .21 1.29 .15 1.13 1 <.00 2004 -.15 .75 .03 1.08 .65 -.03 .95 1 2006 .01 .88 .87 -.19 .87 .004 -.04 .94 <.00 Baseline BMI .05 1.05 .04 1.04 1 Baseline .35 1.42 hypertension Baseline <.00 .48 1.62 Smoking 1 Time-varying covariates Prior wave <.00 .35 1.41 diabetes status 1 Marital Status (reference=married) Widowed .04 1.04 .54 .08 1.09 Divorced/Separ .26 1.29 .01 -.01 .99 ated Never married .31 1.36 .03 .46 1.58 Currently -.23 .80 .02 -.18 .84 working Self-rated <.00 <.00 .35 1.42 .13 1.14 .16 1.18 health 1 1 <.00 Disability .69 2.00 1 Wealth (reference=highest quintile) lowest quintile .16 1.47 .001 2nd .08 1.35 .10 3rd .03 1.30 .48 p <.00 1 <.00 1 .14 .03 .70 .66 .03 .02 .40 .98 .04 .10 <.00 1 5 4th BMI -.05 1.20 -.11 .90 .36 <.00 1 Hypertension Heart Problems .37 1.45 <.00 1 <.00 1 .05 1.05 .002 .34 1.40 .03 .22 1.25 .008 Prior wave <.00 -.66 .52 -.29 .75 memory score 1 Missing <.00 .24 1.28 .004 .55 1.73 .39 1.47 .003 indicator 1 1 Missing .52 1.68 .02 .36 1.43 .04 indicator 2 Note. Blank cells mean the effects were not significant at .10 level in numerator or denominator models. 1 Estimates were from logistic models of estimating a person’s probability of mortality based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person was alive and participated in the survey last wave. 2 Estimates were from logistic models of estimating a person’s probability of being censored based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person participated last wave and was still alive this wave. 6 Appendix 4. Denominator models for computing of Inverse Probability of Death, Censoring and Exposure Weights in Estimating Effects of Incident Diabetes on Dementia Risks Mortali Censori Exposu ty1 ng2 re3 Estima Estima Estima OR p OR p OR te te te Baseline Covariates <.00 male .52 1.68 .41 1.51 1 <.00 age .04 1.04 -.03 .97 1 <.00 black -.49 .61 -.19 .82 .07 1 Years of <.00 <.00 .04 1.04 -.04 .96 education 1 1 Baseline Wealth (reference=highest quintile) lowest quintile -.07 .97 .33 2nd -.01 1.02 .80 3rd .02 1.05 .70 4th .09 1.13 .09 Year of Assessment (reference=2008) 2000 NA NA -.11 .85 <.00 2002 .02 .90 .70 .21 1.37 .15 1.10 1 <.00 2004 -.15 .76 .04 1.16 .52 -.03 .92 1 2006 .01 .90 .77 -.13 .98 .03 -.05 .91 <.00 Baseline BMI .05 1.05 .04 1.04 1 Baseline .35 1.42 hypertension Baseline <.00 .48 1.61 Smoking 1 Time-varying covariates Prior wave <.00 .35 1.43 diabetes status 1 Marital Status (reference=married) Widowed .04 1.04 .50 .09 1.10 Divorced/Separ .26 1.30 .01 -.01 .99 ated Never married .30 1.36 .04 .49 1.64 Currently -.23 .79 .01 -.17 .84 working Self-rated <.00 <.00 .35 1.42 .13 1.14 .17 1.19 health 1 1 <.00 Disability .68 1.98 1 Wealth (reference=highest quintile) lowest quintile .21 1.52 .001 2nd .08 1.33 .16 3rd .01 1.25 .85 p <.00 1 <.00 1 .12 .04 .66 .58 .03 .02 .35 .97 .03 .10 <.00 1 7 4th BMI -.08 1.15 -.11 .90 .16 <.00 1 Hypertension Heart Problems .36 1.44 <.00 1 <.00 1 -.02 .98 .02 .05 1.05 .002 .33 1.39 .04 .21 1.23 .01 Prior wave <.00 -.67 .51 -.31 .73 memory score 1 Missing <.00 .22 1.25 .01 .40 1.49 .40 1.49 .002 indicator 1 1 Missing <.00 .72 2.06 indicator 2 .31 1.36 .08 1 Note. Blank cells mean the effects were not significant at .10 level in numerator or denominator models. 1 Estimates were from logistic models of estimating a person’s probability of mortality based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person was alive and participated in the survey last wave. 2 Estimates were from logistic models of estimating a person’s probability of being censored based on the person’s basic demographic variables and prior wave time-varying covariates, given that this person participated last wave and was still alive this wave.