Screening and Enrollment by Sex in Human Immunodeficiency Virus Clinical Trials in the United States

Abstract

Background

Women are underrepresented in human immunodeficiency virus (HIV) research in the United States. To determine if women screening for HIV clinical trials enrolled at lower rates than men, we performed a retrospective, cross-trial analysis.

Methods

We conducted an analysis of screening and enrollment during 2003–2013 to 31 clinical trials at 99 AIDS Clinical Trials Group network research sites in the United States. Random-effects meta regression estimated whether sex differences in not enrolling (“screen out”) varied by various individual, trial, or site characteristics.

Results

Of 10 744 persons screened, 18.9% were women. The percentages of women and men who screened out were 27.9% and 26.5%, respectively (P = .19); this small difference did not significantly vary by race, ethnicity, or age group. Most common reasons for screening out were not meeting eligibility criteria (30–35%) and opting out (23%), and these did not differ by sex. Trial and research site characteristics associated with variable screen-out by sex included HIV research domain and type of hemoglobin eligibility criterion, but individual associations did not persist after adjustment for multiple testing.

Conclusions

In the absence of evidence of significantly higher trial screen-out for women, approaching more women to screen may increase female representation in HIV trials.

A retrospective cross-protocol analysis of 31 AIDS network trials did not find evidence that women screening to trials subsequently enrolled at significantly lower rates than men. Efforts to recruit more women to HIV trials are needed to close knowledge gaps.

While 24% of adults and adolescents living with human immunodeficiency virus (HIV) in the United States are female [1], women remain underrepresented in US HIV research [2–8]. Sex is an important factor in many aspects of HIV disease, including treatment response [9, 10]. US federal law requires research funded through the National Institutes of Health to include women as participants in clinical research as appropriate to the scientific goals, and to ensure that confirmatory trials of interventions report a valid analysis of whether treatment effects differ by sex [11]. Despite these regulations, the median proportions of women among 395 HIV treatment and 132 HIV “cure” trials were 19.2% and 9.9%, respectively, and treatment studies conducted in high-income countries were less likely (than those conducted more globally) to enroll high proportions of women [12]. While regulations require a valid, or unbiased, assessment of whether effects differ by sex, they do not mandate enrolling a sufficient proportion of women, or enough participants in total, to ensure high statistical power to identify these sex differences [11]. Thus, even after trial results are published, it often remains unknown if women have significantly different treatment outcomes than men.

Individuals interested in trial participation must complete a formal, in-person screening process to assess whether they meet trial eligibility criteria and to provide informed consent. These persons may fail to enroll (often called screen-out or screen failure), due to not meeting entry criteria, choosing not to participate after learning about trial requirements or other reasons.

Treatment research may involve medications with potential teratogenicity [13] or such information may be inadequate or unknown [14]. Thus, many treatment trials for persons living with HIV (PLWHs) have eligibility criteria prohibiting pregnancy, breastfeeding, and requiring use of reliable contraception. Such requirements, which apply only (or apply differently) to participants who can become pregnant, may create a higher barrier to recruitment, and impose burdens during trial participation due to contraceptive cost, side effects, or inconvenience [10].

We hypothesized that underrepresentation of women in HIV intervention trials may be due to higher screen-out rates in women than in men, for reasons including trial ineligibility or refusal to participate. If true, then general recruitment campaigns aimed at increasing numbers of women approached for HIV clinical trials may not have the desired effect. To explore this hypothesis, we conducted a cross-trial, retrospective analysis of PLWHs in the United States who actively screened for enrollment to clinical trials in a publicly funded HIV treatment research network. We compared the rate of trial screen-out by sex (ie, sex assigned at birth rather than gender). We characterized the major reasons for not enrolling and explored whether these reasons varied by sex. We also examined associations of individual, trial, and clinical site characteristics with differences in trial screen-out rates by sex.

METHODS

Study Design

This was a retrospective cross-trial analysis of formal screening and enrollment by sex assigned at birth across a subset of trials conducted by the AIDS Clinical Trials Group (ACTG). The ACTG is a National Institute of Allergy and Infectious Diseases’ funded international research network, which has been conducting trials of therapeutics for HIV/acquired immune deficiency syndrome and associated infections in adults since 1987 [15]. Intervention trials (both randomized and nonrandomized) without enrollment exclusion by sex, and open to enrollment at all qualifying US network sites between late 2003 (when the network initiated a centralized screening database) through 2013, were included. The cutoff date was chosen so that this analysis did not impact the conduct or integrity of trials while in progress. One trial (early-phase experimental intervention) was limited to enrollment from 5 targeted ACTG sites, and none were trials of HIV prevention. During this time, no network trials in the United States excluded women, and the few trials excluding men (eg, contraceptive interventions) were not included in this analysis.

Screening for participation in ACTG trials has 2 stages. The first, an informal stage that varies across sites, includes eligibility assessment from information available from referring providers or records or verbally provided by the participant. The second, or formal, screening stage occurs following informed consent for trial screening and includes an in-person eligibility assessment including laboratory tests. Screening data in the network database are available only from the second stage. Thus, data analyzed include only the subset of persons who agreed to formally screen, via informed consent, and met a subset of criteria assessed during the first informal screening stage.

Demographic data collected at screening included self-reported sex, race, ethnicity, and age. When these trials were enrolling, sex collected at screening was not specified as sex assigned at birth. To confirm that the screening variable represented sex assigned at birth, we analyzed a subset of trials, which separately collected sex assigned at birth among enrollees. This subset also reported current gender identity, allowing an estimate of representation of transgender persons.

Multiple reasons for an individual not enrolling could be reported. Because full ascertainment of eligibility may not have occurred, reasons analyzed include only what data were available and reported at the time of this decision. Due to sponsor policies, identification of individuals screening across trials was not possible, and thus persons may be included more than once if they screened for multiple trials.

Trial characteristics, extracted from each protocol document, included design elements of treatment allocation, phase, duration, intensity score reflecting number and complexity of visits, objectives relating to sex or women’s health, and HIV research domain. Eligibility characteristics included contraceptive requirements, hemoglobin and renal function criteria, and exclusion for active hepatitis B or C. Other targeted characteristics included pregnancy testing schedule and study medication Food and Drug Administration drug pregnancy categories. Clinical research site characteristics including state (and US census region), number of network studies conducted during time of this analysis, number of years in network, and type of research site (main versus subsite, or satellite) were extracted from a central network table.

Statistical Methods

The main outcome measure was screen-out, defined as an individual formally screening to a trial who did not subsequently enroll. Persons classified by sex and enrollment were pooled over trials. The association between sex and screen-out was estimated by the absolute difference in screen-out rate by sex (and its 95% confidence interval [CI]) and by Fisher’s exact test. Screen-out rates by sex were stratified by other demographic variables, and Breslow-Day tests assessed statistical significance of variations within these variables. These analyses used SAS/STAT software (SAS Institute, Inc) [16].

Trial and site characteristic analyses used data summarized over the unit of interest (trial or site). Random-effects meta-analysis tested if sex differences in screen-out significantly varied over group levels, where groups were defined by each trial or site characteristic. These models used the method of DerSimonian and Laird, with heterogeneity estimates taken from an inverse-variance fixed-effects model [17]. In order to provide multiple covariates in a model, the meta-analysis was extended to a random-effects meta-regression. Monte-Carlo permutation tests adjusted for multiple covariates [18]. These analyses used STATA software (StataCorp) [19].

RESULTS

A total of 10 744 screenings from 31 clinical trials at 99 network-affiliated clinical research sites in 28 states, the District of Columbia, and Puerto Rico were included in analyses [Supplementary Table 1]. The research topics reflected the portfolio of the network: 11 trials of optimizing antiretroviral therapies, 8 trials of inflammation, 4 trials of end-organ diseases, 2 trials each about viral hepatitis and HIV coinfections and malignancies, and 1 neurologic trial. A majority of the studies were relatively small: 9 (29%) trials enrolled fewer than 50 participants; 8 (26%) enrolled 50–99 participants; 6 (19%) enrolled 100–249 participants; and the remaining 8 (26%) enrolled more than 250 participants. The median (25th percentile, 75th percentile) female representation in these 31 trials was 15.7% (9.6%, 21.4%). A total of 12 (39%) trials included a prespecified secondary or tertiary objective relating either to sex differences or to a woman-specific outcome.

Table 1 summarizes screening and enrollment pooled across trials and sites by sex assigned at birth. Of 10 744 screenings, 18.9% were women. A total of 2871 (or 26.7%) people formally screened did not subsequently enroll. The percentage of women screening out was 27.9% compared to 26.5% of men. Thus, the overall screen-out rate among women was 1.4 percentage points higher than men (95% CI, −.72 percentage point, +3.6 percentage point; P = .19, Fisher’s exact test).

Table 1.

Screen-out Outcomes by Sex Among Individuals Completing Screening for 31 AIDS Clinical Trials Group Trials, 2003–2013

Screened Out	n (row %)	Enrolled		Total screened
		n	%	n	%
Women	567 (27.9)	1466	18.6	2033	18.9
Men	2304 (26.5)	6407	81.4	8711	81.1
Total	2871 (26.7)	7873	100	10 744	100.0

N = 10 744. Difference in screen-out: 27.9%–26.5% = 1.4 percentage points (95% confidence interval, -0.72 percentage point, +3.6 percentage point; P = .19, Fisher’s exact test). Abbreviation: AIDS, Acquired Immune Deficiency Syndrome.

Of 5294 enrollees from 6 trials where specific “sex assigned at birth” was available (67% of enrolled participants), there were only 2 cases where the screening variable sex did not match sex assigned at birth. The percentage of participants identifying as transgender (anyone not cis-gender) among this subset of enrollees was 0.25% (n = 13).

The reasons for not enrolling are summarized in Table 2. A total of 35% persons screening out had an HIV eligibility criterion reason (eg, CD4 cell count, plasma HIV-1 viral load, or HIV resistance testing requirements), 30% of persons had another eligibility criterion reason (eg, hemoglobin or creatinine clearance requirements), 23% had a personal reason (eg, they or their primary care provider opted out, or didn’t return to enroll), 13% had a health reason (eg, contraindication to treatment, use of prohibited medications, comorbidity, substance abuse, or died), and 6% of persons had an administrative reason (eg, trial, arm, or site closed to accrual; allowed screening period expired). Pregnancy, breastfeeding, and contraceptive requirements were rare reasons (<10 among all 2871 screen-outs). With the exception of administrative reasons, reasons for not enrolling did not differ by sex.

Table 2.

Reasons Reported for Not Enrolling (Screen-out) by Sex

	Overall		Women		Men
Reason	n	%	n	%	n	%	P
HIV disease–related eligibility criteria	995	35	192	34	803	35	.69
Non-HIV–related eligibility criteria	862	30	154	27	708	31	.10
Personal choice	649	23	136	24	513	22	.37
Comorbidity	368	13	77	14	291	13	.53
Administrative reasona	158	6	45	8	113	5	.002

Multiple reasons could be reported for screen-out and full ascertainment of eligibility may not have occurred.Abbreviation: HIV, human immunodeficiency virus.

^aExamples of administrative reasons include cohort/group closed to accrual, site closure.

Results of stratified analyses of sex differences in screen-out by race, ethnicity, and age are summarized in Table 3. Screen-out did not vary by race (P = .99) or ethnicity (P = .15). Nonsignificant variations in sex differences were observed between age groups (P = .06). Across both sexes, the percentages of screen-out increased with age, and among women this increase was observed starting in the fourth decade (30–39 years) compared to the fifth decade (40–49 years) for men.

Table 3.

Enrollment Outcomes by Sex Stratified by Race, Ethnicity, and Age (Pooled Across Studies)

Characteristic	Enrolled		Screened Out				Sex Difference Screen-out, Percentage Points	P a
	Women, n	Men, n	Women		Men
			n	%	n	%
Raceb								.99
White	496	3864	170	26	1286	25	0.6	.78*
African American/black	860	2097	369	30	875	29	0.6	.71*
Ethnicityb								.15
Not Hispanic	1126	5050	476	29	1897	27	2.4	.06*
Hispanic	331	1332	90	21	396	23	−1.5	.51*
Age (years)								.06
<30	218	1187	59	21	283	19	2.0	.46*
30–39	372	1631	137	27	436	21	5.8	.005*
40–49	523	2102	209	29	882	30	−1.0	.62*
≥50	353	1487	162	32	702	32	−0.6	.83*

^a P values for each characteristic (eg, race) are from Breslow-Day tests; P-values for individual levels within characteristic (*) are from Fisher’s exact tests.

^bRace and ethnicity based on self-identification; other groups excluded due to small numbers.

When examined one at a time, most trial and research site characteristics did not exhibit significant variation in sex differences for screen-out (Supplementary Table 2). However, we did observe variation in each of research area and type of hemoglobin eligibility criterion (ie, 1 minimum hemoglobin threshold versus different minimums by sex versus no criterion) (Table 4). The lowest allowed hemoglobin level for women (7–8 g/dL in 12 trials, 9–11 g/dL in 17 trials) was not significantly associated with differential screen-out (Supplementary Table 2). Some subgroups with the largest observed sex differences in screen-out were composed of only a few trials; for example, only 2 trials did not have a hemoglobin eligibility criterion, and some research domains contained fewer than 5 trials. Thus, inference may be influenced by the small size of these subgroups. A model including both research domain and type of hemoglobin criterion that adjusted for multiple covariates yielded nonsignificant results (Table 4).

Table 4.

Screen-out by Sex for 2 Trial Characteristics

	Sex Difference in Screen-out		Screen-outs/Total Screened
Trial Characteristic	Percentage Points	95% CI	Women	Men	P a
Overall	2.4	(−.30, +5.2)	567/2033	2304/8711
Research domains					.01 (0.07*)
ART optimization (11 trials)	3.8	(+1.6, +6.0)	301/1442	1065/5669
Inflammation (8 trials)	−6.3	(−13.3, +.7)	120/209	595/1046
End-organ diseases (4 trials)	5.3	(−4.2, +14.8)	39/100	258/810
Coinfections and malignancies (2 trials)	7.6	(−1.1, +16.2)	72/152	280/702
Viral hepatitis (3 trials)	8.2	(−1.0, +17.4)	33/114	81/378
Cure (2 trials)	−23.0	(−38.0, −.08)	0/12	20/89
Neurology (1 trial)	26.5	(−26.5, +79.5)	2/4	4/17
Type of hemoglobin eligibility criterion					.01 (0.16*)
One criterion for all participants (21 trials)	4.1	(+2.0, +6.1)	445/1753	1800/7655
Sex-specific criteria (8 trials)	−1.1	(−12.7, +10.5)	113/237	470/957
No minimum eligibility requirement (2 trials)	−14.2	(−29.4, +1.0)	9/43	34/99

Abbreviation: ART, antiretroviral therapy.

^aCovariate P values from F test with Knapp-Hartung modification (df = k-1, 31-k) where k = number of levels of covariate. *Adjusted P value includes both covariates in a single model and Monte-Carlo permutation test with 10 000 permutations.

The overall sex difference in screen-out as shown in Table 4 (2.4%) was slightly larger than the pooled difference estimate, but the 95% CI on the trial estimated difference also contained the zero value of no difference.

DISCUSSION

In this retrospective cross-trial analysis of almost 11 000 persons screening over a 10-year period to 31 HIV ACTG trials at 99 clinical research sites across the United States, we observed a pooled rate of trial screen-out of 26.7% and a small, and statistically nonsignificant, higher screen-out rate among women compared with men. We observed no significant evidence that this small sex difference varied among race, ethnic, or age groups. Overall, women represented 18.6% of trial participants. There were no differences by sex observed in the eligibility or opt-out reasons. None of various trial or research site characteristics explored in this analysis, after adjustments for multiplicity, were associated with significantly varying rates of screen-out by sex.

Across the 31 trials, the pooled estimate of female participants of 18.6% and the median female representation among trials of 15.7% are similar to the rates of female participation reported elsewhere [12]. These rates are lower than the 31% among published studies from North America and Europe reported in Westreich et al [20], but the majority of those data were from observational studies rather than clinical trials. In their review, of the 25 clinical trials published in high-impact journals in early 2011 with research questions relevant to persons of reproductive potential, 4 (16%) of those trials enrolled no women.

Notably, few individual trial (or cross-trial) publications report screen-out by sex (eg, CONSORT [Consolidated Standards of Reporting Trials] and ICMJE [International Committee of Medical Journal Editors] requirements mandate reporting enrollment by sex, but not pretrial screen-out by sex), so the estimates reported here represent novel information regarding potential mechanisms for underrepresentation of women in HIV clinical trial research.

In over 1500 screenings among women under 50 years old, fewer than 10 cases of screen-out were reported due to pregnancy, breastfeeding, or contraceptive requirements. Since these eligibility criteria are easily assessed without further testing or evaluation, it is likely that these criteria were assessed during the informal screening stage and therefore are not reflected in the current analysis.

The strengths of the current analysis include uniformly collected data at a network level and the project’s breadth and size; this analysis included over 30 clinical trials in all phases and domains of HIV intervention research being conducted in the United States by the ACTG and included almost 11 000 screenings. While the screening process at the time of these trials did not specify sex assigned at birth (note: this has since been modified), availability of additional information from the subset of enrollees from 6 of the largest trials in this analysis supports our assertion that the variable captured during the screening process represented sex assigned at birth. Because gender identity was captured only among a subset of trial enrollees, we could not assess if gender was associated with differential screen-out. As data continue to emerge about clinically relevant differences for transgender women [21], it will be critical to examine gender in the trial screening and enrollment process.

There are a number of limitations to this analysis. Some individual-level factors identified by others as being important regarding trial participation, including socioeconomic status, caregiving status of dependents, education, and employment status [22, 23], were not available. Further information about resources provided by clinical research sites that might influence trial participation, such as transportation or childcare assistance, stipends for time spent at trial visits, or other participant support provided, was also not available [24, 25]. Another limitation is availability of information from only the second, formal, stage from the 2-stage screening process. Thus, selection bias by sex could potentially have occurred in the first, informal stage of screening and could not be identified by this analysis. Other limitations included that persons may have been counted more than once if they screened for multiple trials. Since this analysis was limited to trials enrolling at network clinic research sites in the United States, these results cannot be applied to HIV research being conducted internationally, where the demographics of HIV differ from that in the United States. Finally, as the screening process ends once a decision to not enroll has been made, full ascertainment of trial eligibility may not be complete and analyses of reported reasons for screen-out may be biased.

Continuing underrepresentation of women in HIV research has resulted in the persistence of gaps in knowledge regarding whether women have different treatment responses, adverse event reactions to medications, or other outcome differences. Reporting of the absence of significant sex differences in many trials may not be due to an actual lack of a difference, but instead, the inability (via lack of statistical power) of these trials to identify important sex differences due to low enrollment of women [26]. Strategies used to improve statistical power for sex differences include conducting trials in non-US locations [27] and combining data across multiple trials [28]. The latter option may result in complex analyses and a delay in reporting of sex differences (due to the retrospective nature of cross-protocol analyses). For instance, to confirm the anecdotal evidence and information from observational cohorts that women in the United States gain more weight than men following initiation of antiretroviral therapy, data had to be combined over 3 completed confirmatory trials [29]. When the research question involves sex differences by age, the number of trials required to elucidate age by sex effects is even larger [30].

In summary, in the absence of an observed significant difference of screen-out by sex, we hypothesize that improving enrollment of women requires increasing outreach activities that identify, invite, and eventually screen more women for trial participation. We are aware of a few such recruitment campaigns to trials within the ACTG network such as “Follow Your Heart,” a campaign to increase screening and enrollment of women to the REPRIEVE trial [31, 32], and the Women’s Outreach Worker (WOW) project, which aimed to develop strategies for enhancing recruitment and retention of women, and particularly women of color, in ACTG trials [33].

Notably, a few US-based HIV trials have enrolled higher proportions of women, especially when they have been planned with a female recruitment goal and other design features to support that goal [34]. A closer look at those trials’ eligibility criteria, study procedures, and other study characteristics may provide insight into how higher enrollment of women was achieved [35]. Qualitative research such as interviews or open-ended surveys with members of trial teams who have successfully enrolled higher numbers of women may provide additional contextual data to inform this issue. Further research is also needed to identify factors such as implicit bias that may influence the first, informal stage of screening.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Financial support. This work was supported by the Statistical and Data Management Center of the AIDS Clinical Trials Group, under the National Institute of Allergy and Infectious Diseases grant number UM1 AI068634 (to L. M. S.), and additional National Institute of Allergy and Infectious Diseases grant numbers AI069481 (to A. C. C.), AI27757 (to A. C. C.), and AI069494 (to S. L. K.). Research reported in this publication was also supported by the National Institute of Allergy and Infectious Diseases of the National Institutes of Health under award numbers UM1 AI068636 and UM1 AI106701.

Potential conflicts of interest. K. L. K. is involved in this work as a National Institutes of Health (NIH) employee, but the views expressed in this paper do not necessarily represent those of the NIH. S. L. K. is a co-investigator on several Gilead-sponsored studies. A. C. C. has unrelated conflicts including honoraria from Merck & Co. for data-monitoring committee membership and research funding from Bristol-Myers-Squibb to the University of Washington for clinical trial conduct but none related to the work. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.