Objective: To determine predictors of loss to follow-up (LTFU) in trials of candidate HIV vaccines.
Methods: Data were obtained from trials of candidate preventive HIV vaccines conducted by the AIDS Vaccine Evaluation Group (AVEG) and HIV Network for Prevention Trials (HIVNET) that enrolled HIV-negative volunteers. Analytic models included multiple logistic regression and generalized estimating equations.
Results: Of 3033 volunteers enrolled in 48 trials, 282 (9.3%) persons did not complete follow-up. In univariate analyses, age, trial duration, and number of immunizations were associated with LTFU. In a multivariate logistic model, age (per year) (adjusted odds ratio [AOR] = 0.96, 95% confidence interval [CI]: 0.95, 0.98) and study duration (per month) (AOR = 1.04, 95% CI: 1.01, 1.08) remained significantly associated with LTFU.
Conclusions: Younger age and increasing trial duration predicted LTFU. Limiting enrollment in trials of novel products to those less than 40 years of age may exclude participants shown to have improved retention. Trials should be designed to last only as long as required to address the scientific question. Retention efforts in future trials should especially address younger persons.
From the *HIV Vaccine Trials Network Core Operations, Seattle, WA; †Perinatal HIV Research Unit, University of the Witwatersrand, Soweto, South Africa; ‡Statistical Center for HIV/AIDS Research and Prevention, Fred Hutchinson Cancer Research Center, Seattle, WA; §Department of Biostatistics, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC; and ∥Behavioral and Clinical Surveillance Branch, Division of HIV/AIDS Prevention, Centers for Disease Control and Prevention, Atlanta, GA.
Received for publication June 23, 2004; accepted October 12, 2004.
Supported by the National Institutes of Health (AI-46747), HIV Vaccine Trials Network ([HVTN] AI-46703), HVTN Statistical Data Management Center (AI-45200), HIVNET Fred Hutchinson Cancer Research Center, Statistical and Clinical Coordinating Center (AI-45202), Central HIVNET Laboratory ([HIVNET] AI-45205 and N01 AI-45202 and [HVTN] U01 AI-46725), Emmes Corporation (AI-45206 and AI-48017), University of Washington/Fred Hutchinson Cancer Research Center (AI-45207 and AI-47976), Johns Hopkins University (AI-45208 and AI-47980), University of Rochester (AI-45209 and AI-47996), University of Alabama at Birmingham (AI-45210 and AI-47985), Vanderbilt University (AI-45211 and AI-48021), St. Louis University (AI-48022), San Francisco Department of Public Health (AI-48023), and Harvard University.
Presented in part at the AIDS Vaccine 2003 Conference, New York, September 18-21, 2003 [abstract 55].
Reprints: Guy de Bruyn, Perinatal HIV Research Unit, University of the Witwatersrand, Box 114, Diepkloof, Johannesburg, 1864, South Africa (e-mail: firstname.lastname@example.org).
The pressing need for a means to interrupt the global HIV epidemic continues to drive the search for an HIV vaccine. For any candidate vaccine, clinical trials are the necessary pathway from the laboratory to clinical use. It is vitally important that trials be designed and conducted to produce a valid result, and a critical determinant of the internal validity of any clinical trial is the adequate retention of trial participants. Substantial attrition leads to a loss of power of the trial. If loss to follow-up is differential between the arms of a trial, the validity of trial effect estimates is seriously compromised.1 Accurate estimates of retention are also useful for planning future trials. Understanding factors associated with retention may help to inform future screening processes, and knowledge of rates of retention among populations being targeted for recruitment would assist in determining sample size.
No prior studies have investigated retention among HIV vaccine trial participants. Several studies have investigated factors associated with retention in HIV vaccine preparedness cohorts or in trials of other interventions designed to prevent HIV or other sexually transmitted infections.2-4 In those studies, sociodemographic variables such as sex, race/ethnicity, or educational attainment were not consistently associated with retention. Rather, loss to follow-up (LTFU) was most consistently associated with younger age, unstable housing or moving frequently, and not having health insurance. Drug use or high-risk sexual behaviors such as exchanging sex for money or drugs have also been associated with lower retention.2,5 The trials examined in previous retention studies, however, may be fundamentally different from vaccine trials.
To investigate predictors of retention, we examined data from trials of candidate HIV vaccines conducted in the United States by the AIDS Vaccine Evaluation Group (AVEG), the predecessor to the HIV Vaccine Trials Network, and by the HIV Network for Prevention Trials (HIVNET).
Participants were healthy, adult, HIV-1-seronegative volunteers who comprehended the nature of the study and consented to participate. We excluded participants in studies that enrolled HIV-1-positive subjects or infants of HIV-infected mothers, were follow-on studies, or examined only vaccine adjuvants.
We examined the retention of participants in phase 1 and phase 2 clinical trials conducted between 1988 and 1999 by AVEG and HIVNET investigators. These trials tested a number of candidate HIV vaccine approaches: DNA alone, recombinant viral vectors (eg, vaccinia or canarypox [ALVAC, Aventis Pasteur, Swiftwater, PA]), recombinant bacterial vectors (eg, Salmonella), recombinant protein subunit vaccines (envelope glycoprotein gp120 or gp160), other peptide vaccines, virus-like particles, and lipopeptides.6 HIV immunogens were tested alone, together with alum or other experimental adjuvants, or coadministered with immune modulators, such as granulocyte macrophage colony-stimulating factor (GM-CSF). Accelerated vaccination schedules were tested in certain protocols. More than 1 vaccine was administered to some subjects. Most commonly, a subunit vaccine was given as a boost after a priming vaccination with another construct. Control preparations included adjuvant alone, vaccinia, recombinant hepatitis B virus vaccine, recombinant canarypox bearing rabies glycoprotein genes, and placebo.
For the purposes of analysis, we classified vaccines into 7 categories: DNA, ALVAC, peptide, vaccinia, subunit envelope, recombinant viral vectors, or control. Participants who received more than 1 vaccine type were classified to a vaccine category based on the priming vaccine type. Route of administration was classified as intramuscular, intradermal/subcutaneous, scarification, oral, rectal, nasal, or vaginal.
Retention was defined as attendance at the final scheduled protocol visit. Sites classified participants as completed, lost, dead, or other. For 1 trial (HVTN 201), a change in the study schema was made after enrollment of volunteers; the change extended the study duration from 18 months to 4 years. We therefore considered the last scheduled protocol visit under the schema in place at the time of enrollment (ie, 18 months) to be the last scheduled visit.
We defined the level of risk regarding sexual or drug use behaviors according to the enrollment criteria used in the relevant trial. For trials that included high-risk participants, enrollment was generally stratified by risk group. Persons considered at higher risk of HIV acquisition included men who have sex with men practicing high-risk sexual behavior, injection drug users active within the past 3 years, heterosexual partners of HIV-seropositive individuals, or persons meeting the eligibility criteria for the HIVNET Vaccine Preparedness Study.4
We analyzed univariate associations between chosen covariates and retention by calculating unadjusted odds ratios. Covariates of interest included age, sex, race/ethnicity, risk status, sexual orientation, type of vaccine being tested, route of administration, protocol duration, number of protocol-defined study visits, number of vaccinations, study site, and calendar year. Factors that were at least marginally significant (P < 0.10) at a univariate level were then combined in a multiple logistic regression model (PROC LOGISTIC; SAS Institute, Cary, NC). We subsequently used generalized estimating equations (PROC GENMOD; SAS Institute) to conduct additional analyses addressing our concerns of possible correlation within sites or protocols.
The 3033 volunteers enrolled in the 48 AVEG and HIVNET HIV vaccine trials included in the analysis were predominantly white (84%), male (60%), heterosexual (55%), and not considered to be at high risk of HIV infection (78%). Of these volunteers, 282 (9.3%) did not complete follow-up.
In univariate analyses, the covariates that were significantly or marginally associated with LTFU included age, trial duration, and number of immunizations (Table 1). Factors that were found not to be significant at the univariate level included sex, race/ethnicity, sexual preference, risk of HIV infection, trial phase, type of vaccine, route of vaccine administration, and calendar year.
In a multivariate logistic model that included all marginally significant (unadjusted P < 0.1) covariates, we observed significant associations between LTFU and age and study duration. Results of a generalized estimating equations model that clustered participants by site and protocol were similar to the estimates obtained in the multivariate logistic model. The final model indicated that age (adjusted odds ratio [AOR] = 0.96 per year increase in age, 95% confidence interval [CI]: 0.95, 0.98) and study duration (AOR = 1.04 per month increase in duration, 95% CI: 1.01, 1.08) were both significantly associated with LTFU (Table 2). No statistically significant 2-way interactions were observed.
Our analysis documented an overall rate of LTFU of approximately 9% in a large series of HIV vaccine trials in the United States conducted by the AVEG and HIVNET. We observed that younger age and longer trial duration were significantly associated with LTFU.
This study has certain limitations. Our data are all derived from trials conducted in the United States and did not include any participants in phase 3 trials. Measures of risk behavior were determined based on risk enrollment criteria for the trials rather than on individually administered risk assessments. Lastly, potentially important social phenomena, such as personal motivation for trial participation or exposure to news events related to ongoing clinical trials, are not captured by our data.
Despite these weaknesses, these data have a number of important strengths. Our analysis includes a large number of participants, and the data were essentially complete with regard to covariates that were examined. Our findings corroborate factors associated with retention in other prevention trials.2-5 The overall rate of loss to follow-up in this study was lower than that in prior studies of hepatitis B vaccine in high-risk populations7,8 but similar to the rate observed in a longitudinal study of healthy minority women.9 The results of this study should be useful for the design and conduct of future HIV vaccine trials.
Our data suggest that restricting enrollment to those younger than 40 years of age may reduce trial retention. Moreover, these findings suggest that retention efforts in trials should be especially focused on younger participants. Our data also reinforce the importance of designing trials that are as short as possible to answer the scientific questions at hand. Future research needs to continue monitoring trial retention to ensure that retention efforts are informed by the true target population.
The authors acknowledge the assistance of Elizabeth Noonan and Erik Schwab.
1. Schulz KF, Grimes DA. Sample size slippages in randomised trials: exclusions and the lost and wayward. Lancet
2. Scheer S, Douglas JM, Jr, Vittinghoff E, et al. Feasibility and suitability of targeting young gay men for HIV vaccine efficacy trials. J Acquir Immune Defic Syndr
3. Screening, recruiting and predicting retention of participants in a multisite HIV prevention trial. NIMH Multisite HIV Prevention Trial. AIDS
. 1997;11(Suppl 2):13-19.
4. Seage GR III, Holte SE, Metzger D, et al. Are US populations appropriate for trials of human immunodeficiency virus vaccine? The HIVNET Vaccine Preparedness Study. Am J Epidemiol
5. Marmor M, Titus S, Wolfe H, et al. Preparations for AIDS vaccine trials. Retention, behavior change, and HIV-seroconversion among injecting drug users (IDUs) and sexual partners of IDUs. AIDS Res Hum Retroviruses
. 1994;10(Suppl 2):207-213.
6. Graham BS. Clinical trials of HIV vaccines. Annu Rev Med
7. Francis DP, Hadler SC, Thompson SE, et al. The prevention of hepatitis B with vaccine. Report of the Centers for Disease Control multi-center efficacy trial among homosexual men. Ann Intern Med
8. Szmuness W, Stevens CE, Harley EJ, et al. Hepatitis B vaccine: demonstration of efficacy in a controlled clinical trial in a high-risk population in the United States. N Engl J Med
9. Gilliss CL, Lee KA, Gutierrez Y, et al. Recruitment and retention of healthy minority women into community-based longitudinal research. J Womens Health Gend Based Med
10. Neter J, Kutner M, Nachtsheim C, et al. Applied Linear Statistical Models. 4th ed. Columbus: McGraw Hill-Irwin; 1996.
The analysis included data from the AVEG and HIVNET trials listed below. Trials are listed by AVEG number: 002, 002A, 003A, 003B, 004, 004A, 004B, 005B, 005C, 006, 007A, 007B, 007C, 008, 009, 010, 011, 012A, 012B, 013A, 013B, 014A, 014C, 015, 016, 016A, 016B, 017, 018, 019, 020, 021, 022, 022A, 023, 024, 026, 027, 028, 029, 031, 032, 033, 034, 034A, 036, 201, and 202. Outlines of the study goals, specific inclusion and exclusion criteria, and other details are available at www.clinicaltrials.gov (accessed September 17, 2004).