Multiple Imputation for Transmission Category and the Care Continuum

Georgia, 2012

Adults and adolescents living with HIV, Georgia, 2012

Diagnosed only 0

20

40

60

80

100 100%

54%38% 34% 39%

Diagnosed Engaged Retained ART Viral suppression

N=46,495

Adults and adolescents >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 46,495Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Prescribed ART derived from MMP sample Viral suppression (VS) = VL<200 copies/ml Estimated undiagnosed based on CDC projections for proportion undiagnosed nationally

Transmission category definitions

• MSM = Male to male sexual contact• IDU = Injection drug use• MSM/IDU = Male to male sexual contact and injection drug

use• HET = Heterosexual contact with a person known to have, or

to be at high risk for, HIV infection• Other = hemophilia, blood transfusion, perinatal exposure,

and risk factor not reported or not identified

Multiple Imputation• Multiple imputation (MI) methods are used to

assign transmission categories to those persons whose diagnoses are reported without a risk factor

• MI is a statistical approach in which missing transmission categories are replaced with plausible values

• CDC uses MI for the national HIV dataset*• Georgia uses the same methodology as CDC

*Harrison KM, Kajese T, Hall HI, Song R. Risk factor redistribution of the national HIV/AIDS surveillance data: an alternative approach. Public Health Rep 2008;123:618–27.

More details on MI• MI draws a random sample of missing values• Then, MI replaces each missing value with a

set of values according to the relationship with other variables

• These multiply-imputed datasets are analyzed

• Final estimates are made based on these results

• MI is considered a sound approach and the best choice for large datasets*

* Rubin DB. Multiple imputation for nonresponse in surveys. New York: John Wiley & Sons; 1987.

Assumptions used in MI• The distribution of risk factors

among cases initially submitted with no reported risk factor (NRR) does not change during the period used in calculating weights

• Cases reclassified as NRR are representative of all NRR cases

• The data are missing at random

Adult and adolescent males living with HIV, by transmission category with multiple imputation, Georgia

2012

MSM IDU MSM/IDU HET Other0

20406080

100

5643

52 53 5539

31 37 34 394031 36 37 42

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 24,935 N= 2,583 N= 2,428N= 2,022 N= 2,541

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 34,510Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Adult and adolescent males living with HIV, by transmission category,

no multiple imputation, Georgia 2012

MSM IDU MSM/IDU HET Other0

20406080

100

57

3951 52 53

4129

37 32 364129 36 36 38

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 17,533 N= 1.739 N= 1,580N= 1,490 N= 12,168

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 34,510Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/ml*MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Adult and adolescent females living with HIV, by transmission category with

multiple imputation, Georgia 2012

HET IDU Other0

20

40

60

80

100

57 5445

37 37 3139 37 31

Diagnosed EngagedRetained Viral suppression

Perc

ent

N= 8,420 N= 2,452 N= 842

Adult and adolescent females >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 11,715Engaged in care >= 1 CD4 or VL in 2012 Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/ml Multiple imputation is used to estimate number of persons in each transmission category. *IDU = Injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Adult and adolescent females living with HIV, by transmission category, no

multiple imputation, Georgia 2012

HET IDU Other0

20

40

60

80

100

55 51 56

35 34 3839 33 38

Diagnosed EngagedRetained Viral suppression

Perc

ent

N= 3,048 N= 1,056 N= 7,611

Adult and adolescent females >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 11,715Engaged in care >= 1 CD4 or VL in 2012 Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/ml *IDU = Injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Black MSM living with HIV, by age (years), with multiple imputation,

Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

55 54 57 55 49

29 33 38 40 3725 30

38 40 39

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 1,083 N= 4,307 N= 6,796N= 4,977 N= 1,824

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 18,987Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Black MSM living with HIV, by age (years), no multiple imputation,

Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

60 58 59 56 4932 36 41 42 38

27 3240 41 40

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 644 N= 2,729 N= 3,115N= 2,786 N= 1,127

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 10,401Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

White MSM living with HIV, by age (years), with multiple imputation,

Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

60 56 60 56 5139 40 43 44 4235 39

47 47 45

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 82 N= 592 N= 3,622N= 1,732 N= 2,043

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 8,071Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

White MSM living with HIV, by age, no multiple imputation, Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

6757 61 57

494739 45 45 4242 40

48 49 45

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 43 N= 370 N= 2,507N= 1,182 N= 42

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 4,144Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Hispanic/Latino MSM living with HIV, by age (years), with multiple imputation,

Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

6353 47 52 4644

34 3038 33

4435 34 40 36

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 52 N= 355 N= 559N= 607 N= 97

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 1,670Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Hispanic/Latino MSM living with HIV, by transmission category, no multiple

imputation, Georgia 2012

13-24 25-34 35-44 45-54 55+0

20406080

100

7163 56 54 5450

42 38 40 4243 42 41 41 43

Diagnosed Engaged Retained Viral suppression

Perc

ent

N= 28 N= 192 N= 286N= 328 N= 72

Adult and adolescent males >= age 13, diagnosed by 09/30/2011, living 12/31/2012, Georgia = 906Engaged in care >= 1 CD4 or VL in 2012Retained in care >= 2 CD4 or VL at least 3 months apart in 2012Viral suppression (VS) = VL<200 copies/mlMultiple imputation is used to estimate number of persons in each transmission category. *MSM = Male to male sexual contact IDU = Injection drug useMSM/IDU = Male to male sexual contact and injection drug useHET = Heterosexual contact with a person known to have, or to be at high risk for, HIV infectionOther = hemophilia, blood transfusion, perinatal exposure, and risk factor not reported or not identified

Conclusions• Overall, the Care Continuum varied little whether

MI is or is not used in the Georgia 2012 estimates • Use of MI underestimates engagement, retention

and viral suppression (VS) among MSM compared to analysis without MI in Georgia

• This underestimation is greater in groups with a smaller N (e.g., Hispanic/Latino MSM by age)

• Among males, MI overestimates engagement, retention and viral suppression by 1-2% points for IDU, HET and MSM/IDU males

• Among females, MI overestimates engagement, retention and viral suppression by 0-4% points for both IDU and HET

Limitations• Cannot assess the validity of assumptions made in

MI• The pattern of risk factors has changed since the

beginning of the epidemic• Cases missing risk may be fundamentally different

from those for whom risk is reported.• Results for smaller subgroups may differ

statistically with or without use of MI• Under- and overestimates found using MI in the

Georgia 2012 prevalent population are not generalizable to new diagnoses or to future years’ analyses

• Despite limitations, MI is considered by CDC to be the best currently available method to adjust for missing (not reported) risk factor information

References• Barnard J, Meng XL. Applications of multiple

imputation in medical studies: from AIDS to NHANES. Stat Methods Med Res. 1999 Mar;8(1):17-36

• Harrison KM, Kajese T, Hall HI, Song R. Risk factor redistribution of the national HIV/AIDS surveillance data: an alternative approach. Public Health Rep 2008;123:618–27

• Rubin DB. Multiple imputation for nonresponse in surveys. New York: John Wiley & Sons; 1987