RT Journal Article
SR Electronic
T1 P1-S6.34 Evaluation of risk-score algorithms for the detection of HIV infection and syphilis in North Carolina county jails
JF Sexually Transmitted Infections
JO Sex Transm Infect
FD BMJ Publishing Group Ltd
SP A210
OP A210
DO 10.1136/sextrans-2011-050108.258
VO 87
IS Suppl 1
A1 L A Sampson
A1 W C Miller
A1 P A Leone
YR 2011
UL http://sti.bmj.com/content/87/Suppl_1/A210.1.abstract
AB Background In North Carolina, the funding environment for jail screening has shifted from a syphilis-centred model to one based on integrating HIV and STI testing. We developed and evaluated risk score algorithms for HIV screening with and without the addition of syphilis screening in North Carolina county jails. Methods This study included 3610 inmates screened for both syphilis and HIV in two North Carolina jails, 2002–2005. We verified reactive syphilis tests against surveillance records to identify new syphilis cases. We created logistic regression models to predict two different outcomes: HIV only and HIV or syphilis. We created risk scores by rounding the β coefficients from the final models. Cut-offs were chosen based on testing 50% of the available inmate population. We calculated the sensitivity and specificity for each of the risk scores for three outcomes: HIV only, syphilis only, HIV or syphilis. Analyses were conducted using Stata. Results The final model for the HIV-only outcome included sex, age, ever tested for HIV, and race/ethnicity. The lowest scoring individual type would be a heterosexual man, age 18–24, never tested for HIV, with a race/ethnicity in the referent group (total score=0). The highest scoring individual would be MSM, age 25 or older, previously tested for HIV, and of Hispanic or Black non-Hispanic race/ethnicity (total score=6). A risk score cutoff of three or above will lead to screening <50% of the available inmate population. The weighted risk scores from the HIV-only outcome model had better sensitivity for the detection of HIV (82.6, 95% CI 71.2—to 94.0) than the combined-outcome model (65.2 95% CI 50.9—to 79.5). If inmates are selected for screening based on the HIV model, the sensitivity for detection of new syphilis cases is also good (73.3, 95% CI 56.5—to 90.1) and is only slightly inferior to the HIV or syphilis model (80.0 95% CI 64.8—to 95.2). Conclusions We believe that the screening algorithm will perform well in the county from which the sample was drawn. Generalisation to other communities in the Southern USA with similar demographics and rates of HIV and syphilis is also possible. More generally, we recommend targeting HIV jail screening efforts based on HIV data. In communities with incident syphilis infections, we recommend adding syphilis screening to the HIV protocol.