The prevalence of undiagnosed HIV infection among
the Kenyan population was approximated to be 2%in 2012 and
with the maturing of the epidemic, there is a need to focus
resources on ensuring those who are HIV positive are on ART,
andto identify the undiagnosed HIV infected persons. With an
average cost per test estimated to be between USD 2-8, the
present cost-benefit ratio ofuniversal HIV Testing and
Counselling (HTC) programs are close to intolerable limits.
However, HIV infected persons who are unaware of their status
continue to unknowingly transmit HIV to their sexual partners
who may also in-turn infect other partners and propagate the
spread of the infection. We propose to develop a screening
interview (prior to HIV testing) that would have the highest yield
in identifying new HIV diagnoses among ambulatory and in
patients at the JaramogiOgingaOdinga Teaching and Referral
hospital (JOOTRH) and families of HIV infected persons
enrolled at the JOOTRH HIV Clinic. A mixed methods study
design will be employed. A retrospective medical record review
will be conducted at JOOTRH to describe the yield of HIV
testing among various patient categories and family members of
HIV infected persons at the HIVclinic. We will review records of
1500in patients and out patients who were seen at JOOTRH
between January-December 2014 and HIV testing records of
approximately 400family members of HIV infected persons
enrolled at the JOOTRH HIV clinic. From this, a screening
interview to increase the yield of HIV testing will be
developed.Prospectively we will operationally compare the yield
of the newly developed algorithm to routine testing. This will be
conductedamong 831patients in 24 hospital departments and
among family members of 277index cases at the JOOTRH HIV
clinic.
Variables to be collected include patient demographics
(age , gender, occupation, residence), clinical characteristics
including presenting complaint, medical history, physical
examination findings, vital signs, admission notes, diagnoses,
treatment, whether it is a new or repeat visit and date of last visit.
Variables collected for index clients at the JOOTRH HIV clinic
will include all their clinical information, data on family size, age
of family members, history of previous testing of family
members and knowledge of HIV status, health status of family
members. Data from the JOOTRH HIV clinicpatient databases
will then be linked to the Provider Initiated Testing and
Counseling, PITC database. The JOOTRH HIV clinic database
contains all clinical information for the patient who are enrolled
at the HIV clinic; this will be merged to family members’
database.
Analysis of the prediction model will be performed using a
multivariable logistic regression model. Selection of the variables
will largely follow sequential manual steps based on knowledge
of the HIV epidemiology and hypothesized associations between
patient characteristics and HIV infection. We will then develop
the risk score by multiplying the final prediction model’s
regression coefficients by 10. A 10-fold cross-validation method
will be used to evaluate the internal validity of the model.
Receiver operating characteristic (ROC)curve will be constructed
and area under the curve calculated to assess discrimination. The
predicted HIV positivity with the observed HIV positivity will be
plotted fitting a linear regression line and calculating the slope
and R2to assess calibration.
We anticipate developing a screening interview that would
have a high sensitivity, specificity, negative predictive value and
positive predictive value in identifying persons who are in need
of a HIV test result. Consequently, HIV programs can achieve a
higher yield for HTC with a more cost-effective method.