Transcript Estimating the Undiagnosed Fraction

Estimating the Undiagnosed Fraction:
Disseminating the
“Testing History” Method
Jeanette Birnbaum
presenting on behalf of
Martina Morris, Ian Fellows and Matt Golden
10/7/2014
SPRC PHC
1
Overview
The Method
• Summary and assumptions
• KC & WA state analyses
Dissemination
• Prototype “Rshiny” web app
Your Feedback
• Refining the method
• Working with stakeholders
10/7/2014
SPRC PHC
2
The Testing History Method
10/7/2014
SPRC PHC
3
Approach: Backcalculation
• New diagnoses today…
• …are based on infections that occurred in the
past
2 infections
3 infections
1 infection
3 infections
1 infection
Time
2010
10/7/2014
2011
2013
2012
SPRC PHC
2014
10
diagnoses
observed
4
Time from Infection to Diagnosis (TID)
TID = 0 years
TID = 4 years
2 infections
3 infections
1 infection
3 infections
1 infection
Time
2010
2011
2012
2013
2014
10
diagnoses
observed
10/7/2014
SPRC PHC
5
Backcalculation’s 3 Components
OBSERVED
Cases diagnosed at T
=
BACKCALCULATE
Sum of ( Incidence at X yrs ago *
TID probability for X yrs ago
)
NEED AN ESTIMATE
10/7/2014
SPRC PHC
6
TID from Testing History
Infection window = [Date of last negative test -> Date of diagnosis]
1) Get date of last negative test
“Have you
ever had a
last negative
test?”
Yes
Use date
provided
No
Assign age 16
or age-18 years
Missing
2 data inclusion
options
2) Determine time of infection within the infection window
10/7/2014
SPRC PHC
7
Impact of TID Assumptions on TID
DATA INCLUSION
OBSERVED ONLY
TIME OF INFECTION
UNIFORM
TID DISTRIBUTION
“BASE CASE”
% undiagnosed
# infections
Density of last negative tests
Last neg test
Diagnosis
AT LAST NEG TEST
0 years
18 years
10/7/2014
“WORST CASE (OBS)”
% undiagnosed
# infections
Last neg test
18 years
Diagnosis
SPRC PHC
18 years
8
DATA INCLUSION
OBSERVED ONLY
TIME OF INFECTION
TID DISTRIBUTION
UNIFORM
“BASE CASE”
# infections
% undiagnosed
Density of last negative tests
Last neg test
Diagnosis
AT LAST NEG TEST
# infections
0 years
INCLUDING MISSING
Density of last negative tests
0 years
“WORST CASE (OBS)”
% undiagnosed
18 years
Last neg test
10/7/2014
18 years
AT LAST NEG TEST
# infections
Last neg test
18 years
Diagnosis
18 years
“WORST CASE (MISS)”
% undiagnosed
Diagnosis
SPRC PHC
18 years
9
Applications to KC and WA State
10/7/2014
SPRC PHC
10
KC MSM: Incidence ~ Constant
15% missing testing history
11/4/2013
Morris/Fellows/Golden SPRC PHC
11
KC MSM: Undiagnosed Varies by Race
10/7/2014
SPRC PHC
12
WA: Incidence is Declining
MSM
35%
missing
Hetero
60%
missing
10/7/2014
SPRC PHC
13
White
40%
missing
Black
49%
missing
Hispanic
48%
missing
10/7/2014
SPRC PHC
14
WA: Undiagnosed Counts by Group
Hetero vs MSM
10/7/2014
Race
SPRC PHC
15
Dissemination to
Public Health Departments
10/7/2014
SPRC PHC
16
Web Portal
• Goals:
– Provide self-guided access to model for local use
– Minimal learning curve
– Minimal pre-processing of data
– Outputs presented in most policy-relevant form
• Mechanism
– “Rshiny” interface
10/7/2014
SPRC PHC
17
Protoype Rshiny App
https://hivbackcalc.shinyapps.io/HIVBackCalc_A
pp/
10/7/2014
SPRC PHC
18
Your Feedback
10/7/2014
SPRC PHC
19
Issues
• Data privacy requirements
– Local, protected upload, unprotected upload
• Detail in outputs
– Race, mode, sex subgroup results?
– Impact of missing data (suggestions for better
assumptions?)
• Training for users
10/7/2014
SPRC PHC
20
Thank you!
10/7/2014
SPRC PHC
21
Impact of TID Assumptions
DATA INCLUSION
10/7/2014
TIME OF INFECTION
SPRC PHC
TID DISTRIBUTION
22
DATA INCLUSION
10/7/2014
TIME OF INFECTION
SPRC PHC
TID DISTRIBUTION
23