In any clinical trial, the collection and reporting of adverse events is not only a regulatory requirement but also important clinically.1 Clear guidance and regulations related to the reporting of adverse events exist.2,3 However, what constitutes an adverse event for measures of cell concentrations with inherent biological variability (such as CD4 cell count) is less well described. This is particularly true for novel vaccines designed for the prevention of HIV (HIV-1). There are currently several vaccines and biologics that contain the CD4 molecules, which are being evaluated in preclinical or clinical studies.4–9 Beyond this, any HIV-1 vaccine that contains gp120 protein (of which there are many in clinical trials currently) has the potential to bind to cell surface or soluble CD4. In both instances (CD4 or gp120-based vaccines/therapies), a theoretical concern would be anti-CD4 responses leading to CD4 cell decline.
Researchers at the Institute of Human Virology (IHV) of the University of Maryland School of Medicine have developed a full-length single-chain gp120-CD4 complex vaccine that demonstrated promise in simian animal models.7 This novel immunogen will raise responses against the conserved structure that appears in gp120 after HIV-1 binds to CD4. It is expected that antibodies to these epitopes will act collectively to block virus entry and mediate protection from infection. However, it is not known whether the inclusion of a small portion of CD4 within the vaccine might elicit anti-CD4 antibody responses. To understand whether a vaccine candidate such as the full-length single chain might induce antibodies against CD4 and cause CD4 decline, it is important to know what the natural variability of CD4 cell counts are over time in a healthy population. Here, we study a volunteer population that would be eligible to participate in a phase I trial of this vaccine candidate.
Most studies to date related to the variability of CD4 cell counts in healthy adults are focused on circadian changes within a day10,11 or reference levels, which take variability based on nonmodifiable factors such as age, race, and sex into account.12–17 The few studies that have focused on temporal variation in CD4 cell counts were short term and did not quantify temporal changes in a manner that would facilitate identifying an average fold change by which an adverse event could be defined.18,19
The purpose of this study was to evaluate the temporal variability CD4 levels to better quantify the magnitude of CD4 decline, which could indicate a true adverse event. We examined the mean fold change in CD4 cell counts and the proportional change in CD4% over the course of 48 weeks among healthy adults who would be eligible to participate in a Phase I vaccine trial.
Design and Study Population
A prospective observational cohort study was conducted at the IHV in Baltimore, MD. One hundred healthy adult volunteers without HIV-1 infection were recruited for inclusion in the study from throughout the Baltimore region. To be eligible for inclusion in the study, volunteers had to be 18–45 years of age, have a documented HIV-1–seronegative ELISA test result, in general, good health without clinically significant medical history or physical examination findings or clinically significant laboratory results, absence of identifiable risk factors for acquisition of HIV infection such as active injecting drug use or unprotected sexual activity with multiple partners, available for at least 48 weeks of follow-up, and willing to provide informed consent to participate in the study. Volunteers also had to have a negative hepatitis B surface antigen screening test, and a CD4+ cell count and CD4 percentage in the normal range as determined by the laboratory performing the study assays. In addition, female patients had to have a negative b-HCG pregnancy test. Volunteers were excluded if they had active tuberculosis, a history of immunodeficiency, chronic illness, autoimmune disease, or use immunosuppressive medications, were under current treatment for malignancy, received an HIV or DNA vaccine previously, received any live, attenuated vaccine except rabies vaccine in the last 60 days, used experimental therapeutics in the last 30 days, received blood products or immunoglobulins in the last 3 months, abnormal blood chemistry, liver, or kidney panels, used immune modulators or suppressors in the last 45 days, or be breastfeeding and/or unwilling to use an effective barrier contraceptive method if female.
All volunteers who cleared all physical and laboratory evaluations at the screening visit and attending a first study visit were included in the study. Volunteers who consented and enrolled were scheduled for study visits every 8 weeks for 48 weeks. Physical examinations including vital signs and blood draws for immunologic laboratory panels were performed at each study visit. Pregnancy tests for female volunteers were performed every third study visit. ELISA and Western Blot tests for HIV were performed at the screening visit and at visit 6.
The study was approved by the Institutional Review Board of the University of Maryland, Baltimore, and all volunteers provided informed consent.
CD4+ absolute count (cells/mm3) and CD4 percentage (CD4%) were both obtained through flow cytometry for lymphocyte analysis using a single platform assay comprising of Becton Dickinson FACSCount II flow cytometer platform, BD Multitest CD3/CD8/CD45/CD4 reagents, and Trucount tubes. The assays were performed by the University of Maryland Medical Center Immunology/Flow Cytometry Laboratory.
The primary outcome for this study was change in CD4 levels, measured as both CD4+ absolute count (CD3+ CD4+ CD8− cells/mm3) and CD4 percentage (CD4%) defined as percentage of CD4+ CD8− cells among CD3+ cells. CD4 change outcomes were modeled as fold change for absolute CD4+ cell count and CD4 percentage difference for CD4 percentage between adjacent time points. For visit time points where laboratory measurement was missing, differences using that time point were coded as missing.
Age, sex, and race were captured for each volunteer. Additional laboratory measurements for secondary analysis included white blood cell count (WBC, 103 cells/μL), red blood cells (RBC, 106 cells/μL), platelet cell count (109 cells/L, log10 transformed), hemoglobin concentration (grams/dL), and hematocrit ratio (percent, defined as RBC volume/whole blood volume). Each demographic and laboratory variable was assessed as a predictor of CD4+ counts and percentage by fitting a linear mixed-effect model for each variable with a random intercept by participant to account for repeated measures.
Correlations were calculated using Spearman's rank coefficient. To test whether outcome variation was significantly affected by between participant variability (ie, whether mean outcomes are participant-specific), we fit intercept-only linear models with and without a participant-level random effect, and then tested the significance of the random effect using a log-likelihood ratio test. The coefficient of variation (CV%) was a measure of relative variability, which was defined as the ratio of the SD to the mean (average). The CV% was calculated for both CD4 outcomes to measure the variability over time for each subject. For each CD4 outcome, the SD was estimated from a linear mixed-effect model with random intercept by participant. CD4 count fold change was log-transformed in the model.
To model trends in changes in CD4 outcomes using demographic or laboratory predictors, linear mixed models were fit with the change in CD4 levels as the outcome and with a random intercept to control for variation between subjects. Stepwise variable elimination was performed to select predictors for a final model by adding one additional explanatory variable at a time. The P value less than 0.05 had to be included in the model.
One hundred volunteers completed the consenting procedure and attending a baseline study visit (Table 1). Of the 100 volunteers, 39 were female and 61 were male. All volunteers were between 20 and 45 years old (mean age 32 years). Seventy-one volunteers self-reported their race as African American, 14 white, and 15 were classified as other race (self-reported racial/ethnic breakdown: 5 Hispanic, 2 Asian, 2 African Asian, 1 Afro Latin, 1 Afro Caribbean, 1 American Indian, 1 Moorish American, 1 Biracial, and 1 other race). Overall, 40 volunteers had complete visit data. The mean CD4+ cell count at baseline was 936 cells/mm3 (range 442–1961). The mean CD4 percentage at baseline was 46.7 (range 32–60) (Table 1). CD4+ cell count and percentage measurements were also significantly, positively correlated (r = 0.23, P < 0.001). Although only 40 participants completed all 7 visits, 52 participants were available at the final visit (week 48) (see Table 1, Supplemental Digital Content, http://links.lww.com/QAI/B437). The final distributions were not dramatically different than the baseline characteristics. Within sex, males tended to leave more often and within race, African American tended to drop out more often. However, at the end of the study, these groups still represented a majority of the population. As our stability measurement relies on frequency of measurements, we also compared demographic total visits. In this study, females completed an average 5.1 visits to 4.5 for men. Among race groups, African Americans completed an average of 4.6 visits, whites completed an average of 5.6 visits, and other races completed 4.8 visits.
On average, over 48 weeks of follow-up, women had mean CD4 cell count 115.8 cells higher than men (95% CI: 4.2 to 227.0; P = 0.0447), African Americans had a mean CD4 cell count 48.1 cells higher than whites (95% CI: −114.4 to 210.7); however, the difference was not significantly different from 0 (Table 2). For every year increase in age, CD4 cell count increased by 10.0 cells (95% CI: 2.5 to 17.4; P = 0.0103). No demographic variables were significant predictors of CD4+ percentage. In addition, several laboratory measurements were significant predictors of both outcomes.
The overall slope of participants' CD4+ cell count and percentage was unchanged (Fig. 1A), and no trend was detected for either outcome (P = 0.121 and 0.671, respectively). Mean CD4 + cell counts had a significant positive correlation to their SD (r = 0.44, P < 0.001), implying that participants with higher average CD4+ counts have a higher variability (Fig. 1B). This correlation was not observed for CD4+ percentage (r = −0.04, P = 0.70). In addition, the CV% between the 2 outcomes were positively correlated (r = 0.25, P = 0.029) indicating that high variation is correlated between outcomes (Figs. 1C, D). We next tested whether variation in average outcomes was more attributable to general measurement variability or between participant variation. For both CD4+ cell count and percentage, participant variation was significant (P < 0.001 for both). Overall, determining unusual declines for either measurement would require subject-specific cutoffs where higher mean CD4+ counts are generally more variable.
Because flagging unusual declines is challenging when individuals have unique averages and variation, we next examined changes in these measurements over time. Overall, there were 354 total measurements of change between adjacent 8-week time points. In terms of stability, the CD4 percentage difference centered around 0, and CD4 count fold change (difference in log CD4 count) was centered around 1 (Fig. 2A). For CD4 count, we use a relative scale (fold change) rather than absolute change to account for the scale of the outcome where the absolute variation of one participant's count could not be meaningful applied to another participant (eg, see Fig. 1B, for cases where some participants' SDs are greater than other's mean values). On the relative scale, a variation can be described uniformly across participants as fold change. The mean CD4+ count fold change was 1.03 (IQR 0.85–1.15), and the mean of the CD4 percentage differences was 0.02% (IQR −2.0 to 2.0). No trend in time was detected for either measurement. Between participant variation no longer contributed any information to the overall variation (P = 1 for both outcomes), and there was no clear trend between mean and SD for changes in either CD4+ count or percentage (P = 0.09 and 0.79, respectively) indicating stable variation across various magnitudes of these outcomes (Fig. 2B). The estimated SD for the change in CD4 count was 1.3-fold and for CD4% was 4.23%. Correlations of the difference outcomes were similar to the correlation of absolute measurements (r = 0.26, P < 0.001).
Overall, changes in CD4 outcomes were stable among different sex, race, and age groups (Table 2). We next tested whether changes in laboratory measurements predicted changes in CD4+ levels. For CD4+ count fold change, a one unit increase in the change of hemoglobin predicts a 1.04-fold decline (P = 0.041) and a one unit decrease in the change of WBC predicts a 1.03-fold decline (P = 0.0003). For CD4 percent, a one unit increase in the change of hematocrit predicts a −0.19% decline (P = 0.0456). Although these are statistically significant predictors, they may not be clinically significant as a maximum change in these measurements generally predicts less than SD change in the CD4+ outcomes (see Figures 1 and 2, Supplemental Digital Content, http://links.lww.com/QAI/B437). Therefore, for ease of interpretation and implementation, we next propose cutoffs using the unadjusted change measurements from the study population.
As changes in CD4+ measurements were stable in the population, we determined cutoffs to flag unusual declines based on the SD estimates for each outcome. Most of the difference data points for either measurement were constrained within 2 SD boundaries (only 1.7% of measurements outside of both simultaneously). Looking for declines greater than 1.5 SDs, the cutoff for CD4+ count fold change was 1.5-fold and for CD4 percentage was 6.4%. For either outcome, 38/354 (10.7%) of measurements were flagged for an unusual decline, 22 measurements (6.2%) had a decline greater than 1.5-fold in CD4 count, 20 (5.7%) had a decline of greater than 6.4% in CD4 percentage, and 4 measurements (1.13%) were flagged for declines in both outcomes (Fig. 3A). At the participant level, 29/76 (38.2%) met either of the criteria at least once, 17 (22.4%) had a decline greater than 1.5-fold in CD4 count, 19 (25.0%) had a decline of greater than 6.4% in CD4 percentage, and 4 (5.26%) met both criteria simultaneously.
Given that these are healthy participants, declines for both outcomes greater than the estimated 1.5-SD correspond to a 1.13% false-positive rate per measurement, respectively. With repeated measurements per participant, there would be an expected increase in the false-positive rate at the participant level reaching up to 6.6% across 6 measurements, near the 5.3% participant-level false-positive rate observed in this study. Note that multiple, simultaneous drops in CD4 count and CD4% in the same participant are unlikely, and none were observed in this study. Thus, although we did rarely note simultaneous drops in CD4 count and CD4%, these did not carry over to the next time point.
In the clinical trial setting, looking for unusual changes in adjacent time points may not be feasible because the vaccine may affect CD4+ levels immediately after baseline. To address this, one could calculate changes by comparing postvaccination to prevaccination CD4+ levels. To assess that approach in these data, we recalculated changes at follow-up visits relative to a single baseline value (Fig. 3B). Using relative to baseline change measurements, similar frequencies were flagged for unusual declines (Fig. 3D); however, although only 3 participants were flagged for simultaneous declines, 2 of those participants were flagged twice, none of which carried over to the next measurement.
Using a single prevaccination measurement for comparison may be risky as an unusual prevaccine measurement may repeatedly inflate relative changes calculated at postvaccination time points. This may be addressed by averaging multiple prevaccination time points, if available, to stabilize the prevaccine estimate. In a clinical trial, these data may be available if there was a prevaccination screening and baseline visit. We test that here by recalculating the changes in follow-up visits relative to the average of the baseline and second visit measurements (Fig. 3C). At the participant level, this approach is slightly more conservative than the other 2 calculations and only flags 2 unique participants (once each) for simultaneous declines (Fig. 3D), none of which carried over to the next measurement.
As the SD estimated here may not be generalizable to other populations, multiple prevaccination measurements can also be used to estimate the cutoffs within sample using the same approach as above with adjacent time points. Using that approach here, we recalculated the SD using only changes between the second visit and the baseline visit. The recalculated 1.5 SDs cutoff was similar for CD4 count fold change (1.4- vs. 1.5-fold) and CD4 percentage (6.5%–6.4%).
In this observational cohort study of healthy adult volunteers who would be eligible to participate in a phase I HIV-1 vaccine trial, we prospectively followed CD4 count and CD4 percent in a mixed race and gender population. Previous studies have demonstrated that absolute CD4 count and percentage of CD4 cells differ by sex, with women having significantly higher counts and percentages compared with men.12,13 Our study did note a mean increase of CD4 counts in women compared with men, and no race-related differences were noted.
There have been a number of studies that have looked at longitudinal changes in CD4 cell count and CD4 percentage, but these mostly focused on establishing reference ranges for lymphocyte counts and have noted that sex, age, and race should be taken into consideration.14,16,18,20–23 However, few studies to date have attempted to quantify total variability required to indicate the amount of variability needed for 2 longitudinal lymphocyte measurements to be considered significant different from one another. We found that declines in CD4+ cell counts of 1.5-fold were common, and declines of 1.76-fold were within 2 SDs of the average cell change. For CD4 percentage, declines of up to 6.4% were within 1.5 SD from the mean, and declines up to 8.5% were within 2 SDs. Although it was uncommon among these study participants to observe a decline of greater than 1.5 SDs for either measurement, the probability of observing a decline greater than 1.5 SDs increases with repeated testing over time. For example, if a participant averaged 7 visits, they would have a 6.6% chance that at least one CD4 cell count or percentage count measurement would exceed the 1.5 SD decline from their previous measurement. Therefore, when considering clinically relevant thresholds for adverse events in future studies, incorporating a conservative false-positive rate could help ensure increased sensitivity to detect adverse events. If adverse events are noted in future studies, they should also help refine the false-positive rate while maintaining sensitivity.
In the presence of prevaccination and postvaccination measurements, we established that these cutoffs similarly flag unusual declines comparing postvaccinations changes relative to the prevaccination measurements. If multiple prevaccinations are available, averaging those values helps stabilize those estimates and could potentially reduce the false-positive rate. Although the cutoffs calculated here may be generalizable to other study populations, we also show that recalculating the SD using only the changes of prevaccination measurements (if multiple exist) is an alternative but similar estimate to establish a cutoff that could be applied within sample. In a clinical trial including placebos, changes in placebo measurements could also be used to recalculate the SD within sample using this approach. In either case, recalculation in small sample studies should be applied cautiously as precision of the SD estimate will be lower. In a study without repeated prevaccination measurements or placebo data, these cutoffs should be applied with the appropriate clinical discretion, particularly for border cases or in the presence of other pathology.
Few studies exist in the literature by which the findings of this study further can be further contextualized. One published study focused on the impact of biological variability has yielded similar results. A study by Tosato et al24 estimated that a 67% change in the overall mean CD4 cell count would be required for a change in serial values for a change to be considered significant. Although the study used a fairly homogeneous population, our study among a more diverse population agrees with their finding and further demonstrates that large lymphocyte changes occur due to normal biological variability and should be taken into account when thresholds for adverse changes in a clinical trial are established. The inherent variability in CD4 cell count and percentage make selecting a cutoff for an adverse event in a Phase I trial targeting CD4 cells difficult. However, this study provides evidence that the thresholds for CD4 and CD4% decline should be about 1.5 SDs from these data (50% relative change in absolute count and 6.4% difference in CD4 percentage) to be considered a potential adverse event attributable to a vaccine intervention.
The authors thank all the volunteers who participated in this study.
1. Friedman L, Furberg C, DeMets D. Fundamentals of Clinical Trials. 4th ed. New York, NY: Springer Science+Business Media, LLC; 2010.
2. US Government Electronic Code of Federal Regulations. Office of the Federal Register Title 21, Food and Drugs (series). Available at: https://www.ecfr.gov/cgi-bin/text-idx?SID=3af41b5d19bb0bc2eed323e7053836ff&mc=true&tpl=/ecfrbrowse/Title21/21cfr314_main_02.tpl
. Accessed November 20, 2018.
3. FDA. Regulations: Good Clinical Practice and Clinical Trials 2018 [updated 9/21/201811/20/2018]. Available at: https://www.fda.gov/ScienceResearch/SpecialTopics/RunningClinicalTrials/ucm155713.htm#FDARegulations
. Accessed November 20, 2018.
4. Fouts TR, Tuskan R, Godfrey K, et al. Expression and characterization of a single-chain polypeptide analogue of the human immunodeficiency virus type 1 gp120-CD4 receptor complex. J Virol. 2000;74:11427–11436.
5. Vu JR, Fouts T, Bobb K, et al. An immunoglobulin fusion protein based on the gp120-CD4 receptor complex potently inhibits human immunodeficiency virus type 1 in vitro. AIDS Res Hum Retroviruses. 2006;22:477–490.
6. Gardner MR, Kattenhorn LM, Kondur HR, et al. AAV-expressed eCD4-Ig provides durable protection from multiple SHIV challenges. Nature. 2015;519:87–91.
7. Schwartz JA, Prado I, Misamore J, et al. An HIV gp120-CD4 immunogen does not elicit autoimmune antibody responses in cynomolgus macaques. Clin Vaccin Immunol. 2016;23:618–627.
8. Davis-Gardner ME, Gardner MR, Alfant B, et al. eCD4-Ig promotes ADCC activity of sera from HIV-1-infected patients. PLoS Pathog. 2017;13:e1006786.
9. Fetzer I, Gardner MR, Davis-Gardner ME, et al. eCD4-Ig variants that more potently neutralize HIV-1. J Virol. 2018;92:e02011–17.
10. Miyawaki T, Taga K, Nagaoki T, et al. Circadian changes of T lymphocyte subsets in human peripheral blood. Clin Exp Immunol. 1984;55:618–622.
11. Carmichael KF, Abayomi A. Analysis of diurnal variation of lymphocyte subsets in healthy subjects in the Caribbean, and its implication in HIV monitoring and treatment. Afr J Med Med Sci. 2006;35:53–57.
12. Maini MK, Gilson RJ, Chavda N, et al. Reference ranges and sources of variability of CD4 counts in HIV-seronegative women and men. Genitourinary Med. 1996;72:27–31.
13. Reichert T, DeBruyere M, Deneys V, et al. Lymphocyte subset reference ranges in adult Caucasians. Clin Immunol Immunopathol. 1991;60:190–208.
14. Rungta A, Hooja S, Vyas N, et al. Enumeration of CD4 and CD8 T lymphocytes in healthy HIV seronegative adults of northwest India: a preliminary study. Indian J Pathol Microbiol. 2008;51:127–129.
15. Tollerud DJ, Clark JW, Brown LM, et al. The influence of age, race, and gender on peripheral blood mononuclear-cell subsets in healthy nonsmokers. J Clin Immunol. 1989;9:214–222.
16. Valiathan R, Deeb K, Diamante M, et al. Reference ranges of lymphocyte subsets in healthy adults and adolescents with special mention of T cell maturation subsets in adults of South Florida. Immunobiology. 2014;219:487–496.
17. Shete A, Thakar M, Abraham PR, et al. A review on peripheral blood CD4+ T lymphocyte counts in healthy adult Indians. Indian J Med Res. 2010;132:667–675.
18. Maloney EM, Brown LM, Kurman CC, et al. Temporal variability in immunological parameters: peripheral blood mononuclear cell subsets, serum immunoglobulins, and soluble markers of immune system activation. J Clin Lab Anal. 1997;11:190–195.
19. Kitcharoensakkul M, Bacharier LB, Yin-Declue H, et al. Temporal biological variability in dendritic cells and regulatory T cells in peripheral blood of healthy adults. J Immunol Methods. 2016;431:63–65.
20. Nag VL, Agarwal P, Venkatesh V, et al. A pilot study on observations on CD4 & CD8 counts in healthy HIV seronegative individuals. Indian J Med Res. 2002;116:45–49.
21. Bisset LR, Lung TL, Kaelin M, et al. Reference values for peripheral blood lymphocyte phenotypes applicable to the healthy adult population in Switzerland. Eur J Haematol. 2004;72:203–212.
22. Muller V, von Wyl V, Yerly S, et al. African descent is associated with slower CD4 cell count decline in treatment-naive patients of the Swiss HIV Cohort Study. AIDS. 2009;23:1269–1276.
23. Ngowi BJ, Mfinanga SG, Bruun JN, et al. Immunohaematological reference values in human immunodeficiency virus-negative adolescent and adults in rural northern Tanzania. BMC Infect Dis. 2009;9:1.
24. Tosato F, Bernardi D, Sanzari MC, et al. Biological variability of lymphocyte subsets of human adults' blood. Clinica Chim Acta Int J Clin Chem. 2013;424:159–163.