The protocol for this trial and supporting CONSORT checklist are available as supporting information (see Checklist S1
and Protocol S1
This Phase III, double-blind, randomized, placebo-controlled trial was conducted in Lagos and Port Harcourt, Nigeria between November 2004 and March 2007. Three Institutional Review Boards approved the study: those of the College of Medicine, University of Lagos; the University of Port Harcourt Teaching Hospital; and Family Health International (FHI).
To be enrolled in the study women had to be HIV-seronegative, non-pregnant, 18–35-years-old and have, on average, three or more acts of intercourse per week and more than one sexual partner in the last three months. We excluded women who were injection drug users, were currently participating in another microbicide trial, were less than three months since their last pregnancy, or desired pregnancy in the next 12 months. Most study participants were low-income women who exchanged sex for money to supplement their incomes, although they did not self-identify exclusively as sex workers. All participants signed written informed consent form before screening and enrollment. Measures were put in place to ensure that the informed consent process was adequate for illiterate participants.
Each of the two study sites had two clinics with up to 15 outreach posts located in areas with densely concentrated, low-income populations where HIV transmission was thought to be high. The informal results of previous research among high risk populations as well as condom distribution programs conducted by local investigators and non-government organizations helped to identify such areas in Lagos and Port Harcourt. Outreach workers recruited women from bars, markets, and other common gathering areas and referred potential study participants to the study clinics for screening. After receiving detailed information about the study and signing the screening informed consent, women were screened for eligibility. An interview, general physical examination, pelvic examination, and testing for HIV, other sexually transmitted and reproductive tract infections (STIs and RTIs) and pregnancy were performed. During the interview baseline demographic data and information on medical history, contraceptive practices and sexual behavior were also collected. Risk reduction counseling, condom demonstration and free lubricated latex condoms not coated with nonoxynol-9 were provided to all participants. Eligible women were asked to return within 30 days for enrollment.
On return, participants signed an enrollment consent form and were then tested for HIV, syphilis, gonorrhea, chlamydial infection and pregnancy. Women who were pregnant or HIV positive were not enrolled. Women diagnosed with gonorrhea, chlamydial infection, trichomoniasis, syphilis, candidiasis or bacterial vaginosis during screening or enrollment were treated per the Centers for Disease Control and Prevention (CDC) treatment guidelines 
and admitted to the study. Upon confirmation of eligibility, participants were randomized to the CS or placebo group and supplied with study gels and condoms. Study staff instructed participants to insert the contents of one full applicator of their assigned study gel into the vagina immediately prior to each act of sexual intercourse throughout the 12 months of study participation and reapply the gel if intercourse did not take place within one hour after application. Participants were instructed to use condoms for all acts of sexual intercourse regardless of gel use, not to douch after sex, not to use any other vaginal products, and not to use the study gel for anal intercourse. Participants were provided with referral information for local family planning clinics if they expressed interest in using contraceptives other than condoms.
Follow-up visits were held at the outreach post most conveniently located for each participant. Procedures included an interview, testing for HIV, gonorrhea, chlamydial infection and pregnancy, re-supply and demonstration - if required - of gel and condom use, and risk reduction and adherence counseling. As part of the interview process participants were asked about their health, any adverse experiences and concomitant medication use since their last visit, and coital frequency, gel and condom adherence in the last 7 days. Participants were encouraged to return for re-supply of condoms and gels if they ran out between their scheduled visits. Women presenting with an adverse event were referred to the study clinic for evaluation and treatment. Due to the investigational nature of the gel women who became pregnant stopped using product until the pregnancy had ended. To avoid social stigma, women that seroconverted were not discontinued from the study, nor did we require them to stop gel use so they could continue contributing to the STI and safety outcomes. All HIV-infected participants were referred to appropriate local facilities for social support and clinical management, including antiretroviral drugs if indicated.
Both CS and placebo gels were identical in packaging and labeling and were administered in a 3.5 ml dose via a plastic single-use applicator. Each 3.5 ml application of 6% CS gel contained 231 mg of the active ingredient, sodium cellulose sulfate. The CS gel had a pH of 7.5. The placebo gel contained hydroxyethylcellulose (HEC) as a gelling agent, had no cell toxicity or anti-HIV properties, and had a pH of 4.4. The HEC placebo was previously deemed safe and sufficiently inactive for use in clinical studies of investigational microbicides 
The primary outcome was incident HIV-1 or HIV-2 infection, as determined by antibodies in oral mucosal transudate using OraQuick® Advance Rapid HIV-1/2 Antibody test (OraSure Technologies, Inc., Bethlehem, PA, USA) and confirmed by Western blot (Genetic Systems™ HIV-1 Western Blot, Bio-Rad, Hercules, CA, USA). Western blot testing for HIV was carried out by the study laboratories in Nigeria. For women who seroconverted during first three months of follow-up, qualitative RNA-based polymerase chain reaction (PCR) testing for HIV with the AmpliScreen HIV-1 test (Roche Diagnostics, Branchburg, NJ, USA) was performed on stored enrollment plasma to assess whether the infection was pre-existing. PCR testing for HIV was also conducted on final visit plasma samples to identify recent infections in the absence of antibodies. HIV PCR testing was conducted by a laboratory of the Institute of Tropical Medicine (ITM, Antwerp, Belgium).
The secondary outcome was incident STI (gonorrhea or chlamydial infection), measured by detecting DNA material in self-administered vaginal swabs using the strand displacement amplification (SDA) BDProbeTec™ ET CT/NG assay (Becton Dickinson, Erembodegem, Belgium). The quality of SDA testing performed by the study laboratories in Nigeria was assured by repeat testing of all positive and 10% of all negative results from enrollment and quarterly follow-up visits at the ITM; in the event of a discrepancy the ITM result was used for data analysis. Storage and shipment of all biological samples were conducted in accordance with the ITM instructions, manufacturer's recommendations and requirements of a shipping company (World Courier, Allentown, PA, USA).
The trial and its reporting complied with the CONSORT Guidelines 
. We conducted the study under an Investigational New Drug application (IND) to the U.S. Food and Drug Administration and in accordance with Good Clinical Practice as established by the International Conference on Harmonisation 
. The trial was registered with the ClinicalTrials.gov registry under #NCT00120770. The National Agency for Food and Drug Administration and Control of Nigeria approved the study prior to implementation.
We aimed to enroll a total of 2,160 participants (1,080 in each treatment group) to observe 66 total HIV infections. This study size was designed to provide 80% power to detect a 50% reduction in the risk of HIV infection among CS gel users, controlling the type I error for falsely concluding a reduction in risk at the 0.025 level (an independent data and safety monitoring committee was to evaluate any potentially harmful effect of CS using less stringent criteria). A 50% reduction in typical use risk was considered to be a meaningful effect for impacting the epidemic. Such an assumption required that CS be more than 50% efficacious during consistent and correct use, since women would not use gel for all acts during the trial (e.g. due to withdrawal of product during pregnancy, missed product supply visits, and other non-adherence).
The sample size calculation assumed that loss to follow-up would not exceed 20% and that the incidence rate in the control group would be 5 per 100 woman-years. This rate of HIV infection was estimated based on HIV prevalence data available prior to study initiation 
, as well as incidence-to-prevalence ratios from previous research conducted by FHI among similar study populations in West Africa 
. To compensate for the lack of directly measured incidence, we planned to monitor the overall infection rate during the study to determine if the sample size had to be adjusted to achieve the target number of events (66).
Randomization and Blinding
Participants were randomly assigned to either the CS or placebo arm using a 1
1 allocation ratio. A statistician not otherwise involved in the study developed the allocation sequence using a stratified (by study site), randomly permuted block design with block sizes 12, 18, and 24. Six product label colors (3 for CS and 3 for placebo) were used to improve blinding (revealing one color would not un-blind the entire study). Sequentially numbered, sealed opaque envelopes were used to assign participants to one of the six color groups after they signed the enrollment consent form and were determined eligible for the study. There was no indication that any unblinding occurred during the study.
We compared the distribution of time to HIV infection between groups using an exact log-rank test, stratified by site. We calculated Kaplan-Meier estimates of HIV infection probabilities by treatment group, pooled across sites. Time to HIV infection, in days, was computed as the difference between the estimated date of HIV infection (based on the midpoint between the dates of the first confirmed positive HIV test visit and the preceding HIV negative visit) and the enrollment date, plus one. Data from participants who were lost to follow-up were included in the primary analysis but were censored on the date of their last HIV test visit. In secondary analyses, we used a proportional hazards regression model to estimate the hazard ratio (HR) for HIV infection, controlling for pre-specified baseline prognostic variables. The effect of CS gel in preventing transmission of gonorrhea and chlamydial infection was primarily evaluated by a proportional hazards model that controlled for site and other pre-specified baseline prognostic variables including age, history of pregnancy and anal intercourse, previous use of spermicides, number of male partners and sexual acts not protected by condoms, and positive results for gonorrhea or chlamydial infection at enrollment. We calculated exact confidence intervals for the relative risk of adverse events within system organ classes under a Poisson assumption for the event rates in each treatment group.
All primary analyses were performed on an intent-to-treat (ITT) basis, with the following modifications: randomized participants later found to be positive for HIV at enrollment (via HIV PCR testing) were excluded from analysis; and - for gonorrhea and chlamydial outcomes - women who were positive for STI at enrollment started their time in analysis on the date of their first negative SDA test following treatment.
We also performed pre-planned, exploratory on-product analyses of HIV, gonorrhea and chlamydial infection that excluded data collected from participants after their first documented interruption of product use (e.g. due to a positive pregnancy test, a lack of gel supplies following a missed visit, or safety concerns raised by the study clinician). Non-use of available product (e.g. choosing not to use available gel) was considered part of typical use and was not documented as a product interruption.
Data were collected at the sites on two-ply data collection forms and subsequently entered into FHI's 21 CFR Part 11 compliant Clintrial 4.5 database management system (Phase Forward, Inc., Waltham, MA, USA) by local data entry staff through a secure Internet server. Data analyses were implemented using version 9.1 of SAS statistical software (SAS Institute, Cary, NC, USA) and StatXact (Cytel Software, Cambridge, MA, USA).
Two interim analyses were planned. The first analysis took place after 16 HIV infections had occurred and focused exclusively on the safety profile of CS (i.e. there was no test of effectiveness). The second interim analysis was planned to occur after 33 infections to evaluate both safety and preliminary effectiveness data. We planned to use the Lan-Demets spending function 
with O'Brien-Fleming type boundaries 
to control the type I error for concluding effectiveness at the one-sided 0.025 level. In contrast, a fixed, one-sided p-value less than 0.10 in the direction of harm was used as the criteria to stop for potential harm or futility. Since the study was stopped prematurely due to external factors, no type I error adjustment was required when reporting the final results using two-sided p-values.