Search strategy and selection criteria

On April 1, 2018, we searched MEDLINE, EMBASE and Cochrane for neonatal RCTs of any intervention in LMICs or HICs published between January 1, 2013 and April 1, 2018. We used the following MEDLINE MeSH terms: 1) infant, 2) newborn OR infant. We also included the search terms randomized controlled trial(s) OR RCT(s). For the LMIC search, we paired descriptors related to economic status (e.g., low income, middle income, developing) with descriptors related to the population. We also searched for LMICs using individual country names per the 2017 World Bank criteria [11, 12]. For the HIC search, we paired descriptors related to economic status (e.g., high income, developed) with descriptors related to the population, and searched for HICs using individual country names (full search string included in S1 Table) [12]. In January 2021, we repeated the literature search using the same search strategy and retrieved a random sample of 20 neonatal RCTs (10 trials from LMICs and 10 trials from HICs) published between January 1, 2020 and January 1, 2021.

The search was conducted by a specialist librarian. Since the number of articles for HICs far exceeded those for LMICs, we used a random number generator to select a subset from the HIC group for screening. Two reviewers independently screened all articles for inclusion starting with the title and abstract, and proceeding to full-text; a third reviewer adjudicated discrepancies. We used Covidence systematic review software for screening (Veritas Health Innovation, Melbourne, Australia; www.covidence.org).

We included studies that were RCTs, implemented a newborn intervention and occurred in an LMIC or HIC. We defined an RCT as any trial that randomized participants, including individual and cluster randomization. We defined an intervention on the newborn as one initiated during the first 28 days after birth. We excluded studies from upper middle-income countries as well as studies that extended enrolment or allowed for initiation of the intervention beyond the newborn period. We did not include follow-up studies of primary trials, secondary analyses, studies published in languages other than English, conference abstracts, commentaries or study protocols.

Data analysis

We extracted the following study characteristics: 1) country where the RCT took place, 2) control type (no placebo, placebo), 3) intervention (drug/nutrition, medical device, other), 4) funding (public, private, both, none, not stated), 5) timing of consent (antenatal, postnatal, both), 6) publication year, 7) method of randomization (individual, cluster, quasi), and 8) number enrolled, displayed as trial size (very large [n>1000], moderate to large [n = 100–1000], small [n<100]). We defined a drug/nutrition intervention as any substance introduced into the body by enteral, nasal, inhaled, subcutaneous, topical or intravenous routes; this included nutrition interventions such as macronutrients, micronutrients, vitamins and food products (breastmilk, colostrum, formula, fortification). We defined medical device interventions as interventions involving medical equipment such as respiratory equipment, methods for obtaining vascular access and suction devices (of note, we included commercially available, approved devices in this category). All interventional trials that could not be classified as drug/nutrition or medical device were classified as other (e.g., delayed cord clamping). We defined public funding as any source of funds supported by taxes such as the National Institutes of Health, the Department for International Development, and the World Health Organization.

We also extracted data on the consent process including format of consent (written, written or thumbprint, strictly verbal, written or verbal, not stated), method of obtaining consent (informed consent, deferred informed consent, informed or deferred consent, not stated), and video recording of consent (yes, no or not stated). To understand the consent rate, we collected the number approached and consented as well as the number actually enrolled. To minimize publication bias with respect to consent rate reporting, we requested the consent rates from primary authors of articles that did not report it. To explore whether our findings remain relevant in newly published trials, we extracted and analyzed consent rates only from the random sample of trials obtained from our more recent search.

Our primary outcome was consent rate, defined as the percentage of participants who consented amongst those who were eligible and approached. If additional factors made a participant ineligible between screening and enrolment, we presumed they were not approached for consent.

We analysed the difference in study characteristics between LMIC and HIC trials using the Chi-squared test. In descriptive analysis, we compared distribution of consent rates between LMIC and HIC trials overall, as well as among the sub-samples of control and intervention types using a two-sample Wilcoxon rank-sum test. We also compared the difference in sub-sample consent rates within LMICs and within HICs by control type using a two-sample Wilcoxon rank-sum test and by intervention and funding using the Kruskal-Wallis Rank test. We analysed the odds of consenting in LMICs versus HICs by trial type using a generalized linear model for fractional dependent variables, which specifies a logistic link function, binomial family distribution, and heteroscedasticity-robust standard errors [13]. Given the limited literature on variables affecting informed consent, we included all study characteristics collected unless there was limited variation between the groups (e.g., method of randomization). We used the continuous variable of number enrolled as our marker of trial size in the model. Our final model adjusted for funding, timing of consent, publication year and number enrolled (as a natural log). In supplementary material, we analysed the difference in study characteristics between trials reporting a consent rate and those that did not. We also analysed the odds of reporting a consent rate in LMICs versus HICs by trial type using logistic regression with the adjusted model including the study characteristics detailed above. Finally, we conducted a post-hoc analysis of the consent rate for HIC trials by region using the Kruskal-Wallis Rank test. We used STATA version 16 (College Station, TX) for all analyses.

There was no funding source for this study. The final protocol can be obtained by emailing the primary author. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors approved the final version of the manuscript submitted for publication.

Disclosure

Institutional ethical review boards and research governance frameworks play an important role in ensuring that trials have well-defined and documented procedures for obtaining informed consent. This infrastructure is well-established in HICs, ensuring that research participants are informed and adequately protected while being recruited for and participating in clinical trials. In contrast, the infrastructure in LMICs is often ill-equipped and poorly managed by overworked clinical staff or volunteers without specialist expertise [10]. Thus, it is possible that poor research governance in LMICs frequently leads to inadequate disclosure of the details of a study, a problem identified by at least 35% of researchers with trials in LMICs [15].

Comprehension

Even if disclosure is adequate, individuals in LMICs may be at greater risk for poor comprehension. For example, participants may erroneously believe that they are consenting to receive treatment instead of to participate in research, a phenomenon referred to as therapeutic misconception [25–28]. Lower literacy levels and language barriers may also reduce comprehension in LMICs. Furthermore, poor comprehension may be more likely for the subset of clinical trials reviewed in this study, as prior work has demonstrated that comprehension of both randomization procedures and placebo controlled designs is generally lower than other aspects of trial comprehension [29–31]. However, this challenge in understanding complex trial designs is not unique to LMICs [29].

Voluntary choice

Voluntary participation may be a challenging concept for potential participants in LMICs to understand. For example, evaluation of the consent process for an HIV study in a South African antenatal clinic revealed that 84% of participants thought their participation was compulsory [32]. Even when potential participants understand that involvement in research is voluntary, they may feel pressured to participate due to their child’s illness [30]. This pressure may be disproportionately felt in LMICs where access to quality medical care is poor. Furthermore, even with standard of care or placebo control arms, participants may perceive additional incentives such as free health care, increased surveillance and reimbursement of travel costs as benefits to their participation. These incentives may hold greater importance as positive motivators for consent in LMICs.

Authorization

Global guidelines for informed consent emphasize the importance of individual authorization for participation in research. This recommendation is founded upon the principle of autonomy which is a normative underpinning of the developed world. In contrast, communal consciousness in LMICs may raise the importance of involving community leaders and families in decision-making, and ultimately influence consent at the individual level [33]. In some cases, this can lead to a decision to participate before the informed consent process is even begun [34, 35]. Furthermore, vulnerable populations within LMICs with low socio-economic status and limited access to health care may pre-emptively decide to participate based on information elicited from the community [34]. Pre-emptive decisions to participate in advance of the informed consent process may alter the utility of this process.

Our study has a number of important limitations. The majority of LMIC studies were from India, thus may not be broadly representative of consent rates in other LMICs. We did not search trial registries for unpublished studies. We cannot comment on other factors beyond trial characteristics that have been associated with informed consent rates such as parental characteristics, physicians approaching participants for consent, or incentives to participation [36]. Finally, the LMIC trials included in this systematic review were larger than the HIC trials, raising the possibility that the LMIC trials disproportionately represented later phase trials with a more favourable benefit to risk ratio. However, this is unlikely to account for the difference in consent rates as our adjusted model controlled for study size; furthermore, study phase has not been found to impact consent rates [16].

Consent rates are an important indicator of the informed consent process, yet their utility in assessing the quality of the consent process is limited without additional information. Only 67% of studies that met inclusion criteria for this review reported adequate information to be able to calculate consent rates. Few additional details about the consent process were included in the methods of these papers, with one third of studies failing to report basic details such as the format of their consent. Although audio-visual recording of the consent process is a legal requirement in India since 2009, only one trial from India explicitly reported audio-visual consent. Global guidelines for reporting on clinical trials should include recommendations for standard reporting of both consent rates and relevant details of the consent process.

Our data suggest that the informed consent process is flawed in LMICs. Multiple international guidelines exist to govern informed consent processes globally [37–41]. Despite these guidelines, the informed consent process in LMICs may be fraught with inadequate disclosure by researchers, as well as poor comprehension and pressured authorization by participants. We join others in calling for an improvement in governance as well as culturally appropriate adaptations to the informed consent process in LMICs [10, 33]. The uniformly high consent rates in LMICs demonstrated in this review suggest that the need for such change is urgent. In fact, the ethical conduct of research in LMICs depends upon it.