865 
Development of a Rule to Predict HIV+ Status in a Cervical Cancer Prevention Program
Jennifer Hallock*1,2,3, M Mwanahamuntu2,4,5, A Westfall2,3, K King2,3,6, S Chisele2,4,5, J Stringer2,3, and G Parham2,3,4
1Columbia Univ, New York, NY, US; 2Ctr for Infectious Disease Res in Zambia, Lusaka; 3Univ of Alabama at Birmingham, US; 4Univ Teaching Hosp, Ministry of Hlth, Lusaka, Zambia; 5Univ of Zambia, Lusaka; and 6Univ of Michigan, Ann Arbor, US
Background: Cervical cancer is an AIDS-defining
illness. Histologically proven invasive cervical cancer should prompt
provider-initiated HIV testing and counseling. Certain morphologic criteria
detected at the time of visual-based cervical cancer screening may be more
predictive of HIV serostatus than invasive cervical cancer or other
AIDS-defining diseases.
Methods: In a recently established Cervical Cancer
Prevention Program in Zambia, nurses take a digital photograph (cervigram)
after application of acetic acid to the cervix. Based on clinical expertise, a
set of 10 criteria were developed based on the morphology of cervical
aceto-white lesions that predict HIV+ status. First, 2 gynecologists
reviewed 50 cervigrams of HIV+ women to determine inter-rater
agreement for each criterion. Next, morphologic criteria with k statistics >0.8 were used to determine
the association between each criteria and HIV status in crude logistic
regression models using 100 cervigrams of HIV+ and HIV–
women. A prediction rule was created using the criteria that were significant
predictors of HIV status. The rule predicts HIV+ status if 1 or more
of the individual criteria is identified. The rule was then applied to 100
cervigrams of both HIV+ and HIV– women, and sensitivity,
specificity, and positive and negative predictive value were calculated.
Results: All 10 morphologic criteria of aceto-white
lesions had k statistics >0.8. Four
morphologic criteria were significant for predicting positive HIV status in
crude logistic regression models, including cervicitis (OR 3.58; 95%CI 1.2 to
10.66), dense opacity (OR 4.6, 95%CI 1.95 to 11.04), extension onto at least
one-third of the ectocervix (OR 5.26, 95%CI 2.04 to 13.07), and occupation of
at least three-quarters of the transformation zone (OR 7.37, 95% CI 2.31 to
23.44). Applying the newly developed rule, we predicted HIV status with a
sensitivity of 0.75, specificity of 0.43, positive predictive value of 0.51,
and negative predictive value of 0.69.
Conclusions: In low-income, high-HIV-prevalent settings
like Zambia, visual-based cervical cancer screening programs may be able to identify
morphologic criteria that predict HIV+ status.
|
