Enrollment, randomization, sample collection and follow-up in the Kisumu-based RCT
Detailed methods of the RCT are described elsewhere  (ClinicalTrials.gov, NCT00059371). Briefly, it was conducted in Kisumu district, western Kenya. Male participants were recruited via newspapers, radio, fliers and street shows and enrolled between 2002 and 2005. Inclusion criteria included being an uncircumcised, HIV-negative, sexually active, 18–24-year-old Kisumu resident. Exclusion criteria included contraindications to, and absolute indications for, surgical MC. An opaque envelope system was used to produce 1:1 allocation between arms. Plasma HIV testing was performed and self-reported HIV exposure data collected at 1, 3, and 6 months after randomization, then at 6-month intervals. The trial was stopped early in December 2006 due to efficacy. Free MC was then offered to all interested participants throughout the follow-up period, causing crossover from the control group. Semi-annual follow-up visits continued through September 2010. The dataset is publically available .
HIV testing and determination of seroconversion and infection dates
Two rapid tests were used at each visit, Determine HIV 1/2 (Abbott Diagnostic Division, Hoofddorp, Netherlands) and Unigold Recombigen (Trinity Biotech, Wicklow, Ireland). If one or both was positive, serum was sent for double enzyme-linked immunosorbent assay testing (Detect HIV 1/2, Adaltis Inc., Montreal, Canada and Recombigen HIV 1/2, Trinity Biotech, Wicklow, Ireland). If at least one ELISA was positive, final confirmatory testing was done by line immunoassay (INNO-LIA HIV 1/2, Immunogenetics NV, Ghent, Belgium). For participants confirmed as HIV positive, the first visit with at least one positive rapid test was designated the HIV seroconversion visit. We calculated the presumed infection date as the date midway between the final negative test and the HIV seroconversion visit.
Additional data collected at enrollment and in follow-up included MC procedure date, behavioral and other HIV risk factors, and sexually transmitted infection (STI) diagnoses. High-risk sexual behavior within the 6 months before a visit was defined as self-report of > 2 female sex partners, exchanging gifts or money for sex, or using a condom “less than half the time”.
STI testing included serum rapid plasma reagin (RPR) (Becton Dickinson) with T. pallidum particle agglutination (TPHA) confirmation for syphilis and HSV-2 serology; and urine polymerase chain reaction (PCR) for N. gonorrhoeae and C. trachomatis. Men with urethral discharge also had urethral swabs for N. gonorrhoeae culture and PCR, C. trachomatis PCR (Roche Diagnostics), and T. vaginalis culture (In pouch test). Any genital ulcers were swabbed for H. ducreyi culture, and PCR and HSV-2 PCR testing using the Roche Multiplex PCR system. Men were defined as having an STI if any tests were positive except a positive RPR with negative TPHA. The study did not collect data on antiretroviral therapy (ART) initiation among participants; between 2005 and 2011, the Kenyan National HIV Treatment guidelines set a CD4 threshold for ART of <=200/mm3.
Inclusion and exclusion criteria for HIV VL analysis
Individual inclusion criteria for this analysis were: seroconversion during the trial or follow-up; and having at least one available plasma sample drawn at least 6 months after infection date and, for those who became circumcised after infection, drawn 6 months after MC (to eliminate bias from the procedure causing transient VL increases). Individual exclusion criteria were: potential HIV exposures other than vaginal sex at any time (sex with men, intravenous drug use, or blood transfusion); a still-unhealed MC wound less than 3 weeks prior to infection; and for circumcised men, infection date fewer than 6 weeks after MC (unhealed wounds may have facilitated infection among men who resumed sex earlier).
Sample exclusion criteria were also applied. Samples drawn before Jan 1, 2007 were excluded from the final analysis because all but five (of 85) had undetectable or unquantifiable VLs, correlations described below between sample age and VL suggested older samples were subject to degradation, and none of these samples had second aliquots available for quality control. Finally, samples with undetectable or unquantifiable results were also ultimately excluded (see Data Analysis section for methods and reasoning). Among eligible men, the first eligible samples, up to three, were used.
The anticipated sample size based on total available aliquots was 123 men (82 circumcised and 41 uncircumcised), which provided 80% power to detect a between-group difference in mean VL of .54 standard deviations, equivalent to .93 log10 copies/mL . Power calculations were done using SAS 9.3 PROC POWER TWOSAMPLEMEANS with a two-sample t-test.
Sample storage and testing
Plasma samples were not tested at the time of collection, and were not thawed at any time before testing. They were stored at − 80 °C, initially at the Kenya Medical Research Institute in Kisumu, and permanently at the Chicago Developmental Center for AIDS Research. VL testing was performed by the International Laboratory Branch of the US Centers for Disease Control’s Division of Global HIV and Tuberculosis in Atlanta, Georgia, using the Abbott HIV-1 Real Time HIV-1 assay on the Abbott m2000 instrument, with a lower limit of detection of 150 copies/mL with a 200 μL plasma input volume. All samples with an available second aliquot were retested and results compared against initial values, with a difference threshold of ≤0.5 log10 copies/mL for concurrence.
In the primary (as-treated) analysis, men were classified as circumcised or uncircumcised based on whether they had been circumcised through the RCT’s services before infection. The secondary (per-protocol) analysis restricted the comparison to men remaining in their original randomization group. Analysis was performed using SAS 9.3 on log10-transformed VL results from the first round of testing.
Once a high proportion of samples with undetectable or unquantifiable (VL detected at a concentration below the quantifiability threshold, 2.18 log copies/mL) results was identified, methods for handling these values were devised. Prior to the second round of testing, it was pre-specified that all such samples would be retested if a second aliquot was available, and the values would be included in the analysis if at least 80% of their retested samples were again either undetectable or unquantifiable, increasing confidence that these represented true plasma values rather than degradation of the aliquots.
To compare characteristics between groups, we used Fisher’s exact Chi-square test for categorical variables, and t-tests for continuous variables. Pearson correlation coefficients were used to examine the relationship between sample age and log VL.
PROC MIXED was used to test for differences in log VL between groups, with a random effect for individual to account for the correlation between samples from the same man, This approach yields results similar to an analysis on log VL means computed for each man and then compared between groups, but also allows for the use of sample-level covariates such as sample age.