Transcript Partners in Prevention

Partners in Prevention
Jim Hughes
in collaboration with, and many slides stolen from …
Connie Celum
Mary Campbell
Jai Lingappa
Gerry Myers
Jim Mullins
HIV and Herpes
• HIV + HSV-2  increased HIV viral loads
HIV and Herpes
• HIV + HSV-2  increased risk of transmitting
the virus
HIV and Herpes
• Acyclovir (and related drugs) are very effective in suppressing
the HSV-2 virus
• Acyclovir has no direct effect on HIV but …
?
Acyclovir   HSV-2 replication   HIV viral load   HIV
transmission
 Can we reduce the risk of HIV transmission in HSV-2 + HIVpositive individuals by treating their HSV-2 infection with
acyclovir?
Partners in Prevention
HSV-2 Suppression to Prevent HIV Transmission
3408 HIV-discordant couples with
HIV+ partner also HSV 2-coinfected
Randomize HIV/HSV-2 + partner
Acyclovir 400 mg twice daily
Placebo twice daily
Follow couples for up to 2 years
1° endpoint: HIV infection in HIV-negative partner
Nairobi, Thika
Eldoret, Kisumu
Kenya (4)
Lusaka, Kitwe,
Ndola, Zambia (3)
Kampala, Uganda
Gaborone,
Botswana
Moshi, Tanzania
Kigali, Rwanda
Soweto, Orange
Farm, Cape Town
S. Africa (3)
Partners in Prevention
•
HIV-infected partner (also HSV-2 infected) =
“index”
•
HIV-uninfected partner = “partner”
•
Intervention: Daily acyclovir (400 mg, bid) given to
the index
•
Primary outcome: HIV infection measured on the
partner – compare infection rates between the two
arms
•
Two analyses of interest:
i.
All transmissions (“easy”)
ii. Only transmissions from the index (“hard”)
Is the index the source of the
partner’s HIV-1 infection?
• Phylogenetic Trees
• Genetic Distance
• Amino Acid Signature Pattern
Ou, et al 1992. Science 256:1165; Learn and Mullins. 2003. HIV Sequence Compendium 2003:22.
Phylogenetics
Basic Concepts
•
•
Trees are comprised of tips, branches and nodes
Tips represent the actual gene sequences used to
create the tree
•
A branch is a representation of the genetic distance
separating tip sequences.
•
A node represents the hypothetical ancestor of the
sequences on the branches stemming from it
•
Monophyletic group = Clade
•
•
terms used interchangeably
implies descent from a single common ancestor
Simple Phylogenetic Tree
monophyletic
sequence
pair
tip
branch
node
root node
Genetic Distance
Number of mismatched positions
Distance
=
______________________
Number of positions in alignment
•
•
the number of nucleotide changes needed to make one sequence the same as
another in an alignment
can be calculated from pairwise comparisons or estimated by tracing distances
between sequences on a phylogram
Phylogenetic
Linkage
Analysis
13.4%
3.3%
Is the index the source of the
partner’s HIV-1 infection?
•
Monophyletic pair
•
•
•
•
Gives yes/no answer, but no degree of uncertainty
Should intervening sequence automatically rule out
linkage?
Should lack of an intervening sequence automatically
confirm linkage?
Genetic Distance
•
•
“Linked” partners should have sequences that are
close; unlinked partners should have sequences that
are not close
Is there a “magic” distance which separates linked and
unlinked? Not quite but …
Smoothed probabilities for distances between sequence
pairs known to be linked or unlinked
linked
unlinked
0.0
0.1
0.2
Distance
0.3
0.4
Bayesian Classification of
partner linkage
P(D|linked)P(linked)
P(linked|D)
P(D|linked)P(linked)  P(D|not linked)P(not linked)
D = distance
P(D | linked) = probability of distance D assuming linkage (based on intraindividual sequence pairs and partners known to be linked)
P(D | not linked) = probability of distance D assuming no linkage (based on
sequence pairs from apparently unlinked individuals)
P(linked) = assumed probability of linkage before looking at D.
P(linked | D) = posterior probability of linkage; base decision on this.
Issues/Challenges
• At present, the algorithm is based on distances only. In principle,
additional quantitative data could be incorporated into the
information D.
• What should the prior – P(linked) – be?
Issues/Challenges
• At present, the algorithm is based on distances only. In principle,
additional quantitative data could be incorporated into the
information D.
• What should the prior – P(linked) – be?
We set the prior equal to the proportion of linked partnerships over
all PIP seroconverters
Most decisions not very sensitive to choice of prior
Issues/Challenges
• Classification may be based on individual sequence pairs (multiple
per partnership) or consensus sequence pairs (generally one per
partnership)
• Single consensus sequence may not represent all variants
• Combining information from multiple sequence pairs poses
methodologic challenges
• The transmitting partner may have multiple variants but only
one may match the variant found in the newly infected partner.
Should this be considered linked? What if the situation is
reversed?
Pair 58
Contrasts between Bayesian and phylogenetic
approaches
Bayesian classifier based on distances only and does not explicitly
consider context (i.e. monophyletic or not)
In this application, monophyletic approach tends to more conservative
(more likely to call the sequences unlinked); more appropriate for
courtroom?
Use of explicit prior assumption in Bayesian formulation allows control
of “conservativeness”
Bayesian classifier can incorporate additional information on
partnership (e.g. gender, reported number of sex partners) in through
prior
Questions
• What scientific question is the PIP trial trying
to answer?
• Why is it difficult to to determine if one
person has transmitted the HIV virus to
another (even if you have virus samples from
both individuals)?
• What genetic measurement does the Bayesian
classifier use to determine the probability of
transmission?