Early Discontinuation, Results Reporting, and Publication of Pediatric Clinical Trials

Publicly available records for all clinical trials submitted to ClinicalTrials.gov before March 9, 2020, were downloaded using the Aggregate Analysis of ClinicalTrials.gov database tool.¹³  We included trials with an interventional study status, maximum participant age <18 years, and an initial registration date on or after October 1, 2007, to align with the Food and Drug Administration Amendments Act (FDAAA), which have stipulated registration for non–phase 1 clinical trials of US Food and Drug Administration (FDA)–regulated drugs and devices. Although the statutory requirements of FDAAA do not apply to all trials, we limited our analysis to only after its enactment, given the significant expansion in functionality and utilization of ClinicalTrials.gov that followed.^14–19 

A multistep strategy was used to identify potential publications resulting from clinical trials. First, publication data were extracted from the PubMed MEDLINE bibliographic database and programmatically linked to the unique ClinicalTrials.gov identifier (National Clinical Trial, or NCT, number) recorded in the article title, abstract, and secondary identifier field.^18,20–22  Second, if an associated publication could not be found, we manually reviewed publication citations listed on the ClinicalTrials.gov record and searched Embase, PubMed, and Scopus with the clinical trial title and NCT number. The complete text of articles yielded from database searches was evaluated on the basis of matching trial characteristics, including study design, indications, primary aims, and lead investigator. The earliest publication date, or online access date, if applicable, was used for trials with multiple publications. Manual review accounted for 26.5% of the total number of publications identified. Because all the data obtained were from publicly available registries, our study was exempt from oversight by the Stanford University Institutional Review Board.

We assigned pediatrics trials to ≥1 of 29 specific therapeutic foci (Supplemental Table 4). Therapeutic foci and associated diseases were prospectively devised by the study team and adapted from previous publications and Medical Subject Headings.¹⁰  We assigned therapeutic foci based on review of study title, abstract, and detailed trial description.

Trial characteristics were selected from the ClinicalTrials.gov record as potential explanatory variables for early trial discontinuation, results reporting, and publication as follows: (1) funding source (government, industry, other), (2) intervention type (behavioral; device; drugs, biologics, and supplements; other), (3) trial phase (not applicable, phase 1, phase 1/2–2, phase 2/3–3, phase 4), (4) trial enrollment (ie, number of participants enrolled), (5) blinding (single, double, none), (6) use of randomization, (7) number of sites, (8) presence of a data monitoring committee, and (9) geographic region of sites.^17–19  Anticipated enrollment was reported for ongoing trials, whereas actual enrollment was reported for completed trials. All variables were included in the final regression analyses.

We determined the funding source on the basis of the trial sponsor (defined by the US National Library of Medicine as the organization that initiates and has primary control over the study) and collaborators (defined as organizations that provide support to the sponsor in activities related to trial conduct, funding, or analysis).²³  Clinicaltrials.gov has designated possible funding types as industry, the National Institutes of Health and other government agencies, and other.

Our approach for sponsor assignment has been described in previous studies.^{17–19,24,25}  Any trial for which industry was listed as either a sponsor or a collaborator was classified as industry. Among nonindustry trials, government was designated when a government agency was involved as a sponsor or collaborator. Nongovernment and nonindustry trials were categorized as academic if the funding source was an academic institution or academic medical center as defined by US legal code and The Joint Commission, respectively.^19,26,27  The remaining trials were labeled as other and represented predominately independent research groups, foundations, and community-based organizations.

We used The World Bank definitions of country income categories to classify trial location as low-income, middle-income, or high-income.²⁸  Low- and middle-income countries were grouped into a single category.

Early discontinuation was defined as a trial that was stopped early with the status of withdrawn, terminated, or suspended. We excluded trials with a duration of <1 day (from reported trial start date to primary completion date in ClinicalTrials.gov) and those that were either ongoing at the time of data collection or without a verified status. The reason for discontinuation was specified as free text on ClinicalTrials.gov. We manually reviewed entries, where provided, which were subsequently categorized on the basis of schemes from previous research as follows: accrual, funding/staff shortages, informative termination, project investigator departure, conduct, regulatory, company decision, other, or unclear/not specified.^{10,18,29–31} 

Completed trials were further analyzed for ClinicalTrials.gov registry results reporting and publication in a peer-reviewed journal within 3 years of completion. These outcomes were evaluated over a period of 3 years from the primary trial completion date in accordance with federal guidelines for a certified delayed submission.¹⁶  Only trials completed by March 8, 2017, were included to allow for 3 years of elapsed time between completion and this analysis.

To measure changes in trial outcomes over time, we divided the time periods by their midpoints and compared rates of early discontinuation, results reporting, and publication between October 2007 and June 2012 (early) and July 2012 and March 2017 (late). Trial assignment was based on trial submission date for early discontinuation and primary completion date for results reporting and publication. Trials submitted after March 2017 were excluded in the analysis of early discontinuation to limit the bias of comparatively shorter completed trials in the late period.

We used a 2-sided Pearson χ² test and Fisher’s exact test, as appropriate, to compare proportions. Differences in the proportions of trials that were prematurely discontinued, reported results, and published between the early and late periods were evaluated with the 2-proportion z-test.

We performed multivariable logistic regressions to determine associations between funding source and other trial characteristics with early discontinuation and results reporting and publication within 3 years of trial completion. Model covariates were selected a priori and included the aforementioned trial design features and therapeutic foci. Multiple imputation with chained equations were performed to account for missing data in multivariable analyses. We generated 30 imputed datasets that were subsequently pooled using Rubin’s rules.³²  All statistical analyses were 2-sided with a threshold of significance of α = 0.05 and performed using R version 3.5.0 statistical software.

There were 332 417 trials registered on ClinialTrials.gov over the study period, with 101 086 (30.4%) excluded because of either registration before 2007 or noninterventional study design. Among the 231 331 remaining, 13 259 (6.1%) met criteria for inclusion as a pediatric trial (Fig 1).

Drugs, biologics, and supplements (49.8%) were the most predominant intervention type across pediatric trials (Table 1). Academic funding was most common (57.5%), followed by industry (26.4%), government (8.3%), and other (7.7%). Most trials were randomized (72.6%) and nonblinded (50.8%) with a small enrollment of <100 participants (54.9%). The most common therapeutic foci were nutrition and metabolism (16.8%) and mental health and behavior (16%). Trial characteristics by trial phase are detailed in Supplemental Table 5.

A total of 903 pediatric trials (11.1%) were prematurely discontinued. The proportion of trials that were discontinued early significantly decreased from October 2007 to June 2012 and July 2012 to March 2017 (12.8% vs 10.8%, P < .001) (Supplemental Table 6).

Poor participant recruitment was the predominant reason for trial discontinuation overall (37.1%), particularly for academic trials (42.7%) (Table 2). For 15.9% of trials, the reason for discontinuation either was not stated or was unclear.

Government-funded (adjusted odds ratio [aOR], 0.72; 95% CI, 0.47–0.97) and academic (aOR, 0.64; 95% CI, 0.50–0.82) trials were significantly less likely to be discontinued than industry trials (Table 3). Higher odds of discontinuation was associated with randomization (aOR, 2.53; 95% CI, 1.94–3.31); smaller sample sizes (<100 participants); trials of drugs, biologics, or supplements (aOR, 1.51; 95% CI, 1.05–2.17); and the presence of a data monitoring committee (aOR, 1.29; 95% CI, 1.06–1.56). A stratified multivariable analysis of early discontinuation by trial phase is available in Supplemental Table 7.

By March 8, 2017, a total of 4657 pediatric trials had reached completion (Supplemental Table 8). Of these, investigators of 23.5% reported results to the ClinicalTrials.gov registry, and 38.8% of the trials were published in a peer-reviewed journal within 3 years of completion. The proportion of trials that either had reported results or were published was 51.2%. Investigators of industry and government trials most frequently reported results to ClinicalTrials.gov (43.6%) and published their findings (50.5%), respectively. Between October 2007 and June 2012 versus July 2012 and March 2017, the proportion of trials that were published (35% vs 40.5%) significantly increased (P < .001), whereas rates of results reporting remained unchanged (23.1% vs 23.8%) (Supplemental Table 9).

Academic trials had the lowest odds of results reporting compared with industry trials (aOR, 0.42; 95% CI, 0.34–0.52) (Table 3). Double blinding (aOR, 1.27; 95% CI, 1.01–1.60) and multicenter design (aOR, 1.40; 95% CI, 1.16–1.69) were among trial features associated with increased reporting, whereas investigators were least likely to report randomized trial results (aOR, 0.79; 95% CI, 0.63–0.99).

Compared with industry, government-funded (aOR, 1.94; 95% CI, 1.52–2.48) and academic (aOR, 1.35; 95% CI, 1.35–1.92) trials had significantly greater odds of being published within 3 years of trial completion (Table 3). Other features associated with publication were the presence of a data monitoring committee (aOR, 1.24; 95% CI, 1.09–1.41) and a multicenter design (aOR, 1.19; 95% CI, 1.02–1.39). In general, trials with larger actual or target enrollment (>500 participants) were more likely to be published than smaller trials (<100 participants). Stratified multivariable analyses of results reporting (Supplemental Table 10) and publication (Supplemental Table 11) by trial phase along with cumulative rates of results reporting and publication (Supplemental Fig 2) are available in the Supplemental Information.

In this cross-sectional analysis of the ClinicalTrials.gov database between 2007 and 2020, we show that early discontinuation, reporting of results, and publication remain pervasive challenges for pediatric clinical trials. Government-funded and academic trials had lower odds of discontinuation compared with industry trials and were more likely to be published. Academic trials had the lowest odds of results reporting. To our knowledge, this study is the largest and most comprehensive evaluation of pediatric clinical trial outcomes. This study highlights prevailing trends in trial performance and areas for improvement toward advancing evidence-based pediatric medicine.

The 11.6% prevalence of early discontinuation encompasses a wide range of pediatric trial types and sizes, making it difficult to directly compare with other estimates, which have ranged from 19% to 40%.^9–11  Nonetheless, our broad inclusion criteria, extended study period, and robust multivariable models help us to identify key factors that affect trial outcomes and, by extension, reconcile some of the heterogeneity across previous studies. The modest improvement in trial completion between 2007 and 2020, for example, may be a partial explanation for the higher rates of premature termination in reviews spanning earlier time frames (2000–2002⁹  and 2008–2010¹⁰ ). Previous investigators also restricted their cohorts to specific trial subsets based on characteristics, such as industry sponsorship¹¹  or randomization,¹⁰  that were found to be independently associated with discontinuation in our analysis.

Poor participant accrual was the most common reason for discontinuation and is a widespread challenge among clinical trials.^10,18,33,34  The ethical standards for consenting children for biomedical research are more complex compared with adults.^3,35,36  Parents and/or legal guardians can have differing levels of perceived risk and understanding from investigators, which may negatively affect likelihood of enrollment and retention.^37–39  Other patient-level considerations include the minimization of risks and possible harms, participant compensation, and development of child-friendly study protocols.^4,40  These barriers are compounded by the overall lower burden of disease for many conditions compared with adults.⁴¹  Many trials default to smaller sample sizes (the majority of trials studied here enrolled <100 participants), which contributes to low statistical power and generalizability.^4,42 

Trials funded by industry were most likely to be discontinued early relative to other funding sources. Like other sponsors, industry trials struggled with meeting enrollment targets. A considerable proportion of trials were also stopped because of company decision. This association is concordant with previous analyses both within and outside pediatrics.^10,18,19,30  Authors of the Best Pharmaceuticals for Children Act, Pediatric Research Equity Act, and other policies have sought to stimulate investment in pediatric research; however, the small market size can make the incentive gap challenging to overcome.^43,44  Companies may thus have a low threshold to terminate a trial if it does not align with commercial interests. Another potential explanation lies with the nature of trials undertaken by industry sponsors. Indeed, the risk and complexity inherent in many pharmaceutical and device studies may predispose industry trials to higher rates of discontinuation.⁴⁵ 

In contrast to industry trials, we found that academic trials were more frequently ended because of personnel and budget shortages and project investigator turnover. Academic institutions are known to lack robust human and material resources, all of which may be more problematic in pediatrics because of stricter regulatory requirements and a limited funding environment.^2,46–48 

Although early discontinuation represents an important source of trial failure, the timely disclosure of trial results may pose an even greater challenge. Within 3 years of completion, investigators of only 23.5% of trials submitted results to the ClinicalTrials.gov registry and only 38.8% of trials were published in a peer-reviewed journal. Dissemination of findings has been found to be variable in previous analyses and dependent on trial type, sponsor, and specialty,^{17–19,30,49–52}  with publication rates ranging between 29% and 70%, even within exclusively pediatric trials.^10,53 

Our findings extend concerns about the reporting and publication bias in the pediatric medical literature, usually attributed to investigators and journals who preferentially withhold negative, unfavorable, or equivocal results. For example, publishers of articles that resulted from abstracts presented at pediatric academic society meetings significantly favored studies with positive outcomes.^54,55  Industry involvement in clinical research has come under increasing scrutiny for selective reporting and publication.^56–58  Consistent with previous work, that industry trials were least likely to publish, even after adjusting for intervention type, may implicate potential commercial conflicts of interest.^10,53  Of note, the comparatively higher rates of results reporting for industry trials might reflect the large number that fall under FDAAA disclosure requirements.^24,59  Despite heightened expectations of scholarly values and public benefit, challenges with reporting and publication were also common for government- and academic-funded trials.^49,52 

The reduction in discontinued trials and the increase in the proportion of trials that were published over the study period suggest a promising trend. However, it remains that clinical trial investigators who stop a study prematurely or fail to disseminate results compromise the integrity of evidence-based medicine and waste limited resources.⁶⁰  These outcomes lead clinical guideline authors, policymakers, and funding authorities to rely on an incomplete and biased knowledge base, a well-described challenge in pediatrics.^6–8  Furthermore, they reveal important ethical considerations.^4,5  Trial participants are voluntarily exposed to inherent risks and harms with the expectation of societal benefit. Failure to uphold this standard constitutes a violation of a commitment toward participants and their families. Numerous international mandates, such as the Declaration of Helsinki and those of the World Health Organization, have codified the ethical safeguards for clinical trials^61,62  Evolving regulatory frameworks,⁶³  dedicated interdisciplinary consortia,^64,65  and the growth of ClinicalTrials.gov^55,66  are among the multifaceted efforts to increase accountability toward such scientific and ethical imperatives.

Our study should be interpreted within the context of several limitations. First, ClinicalTrials.gov does not encompass all clinical trials globally.¹  However, it accounts for the largest share of studies registered on the World Health Organization International Clinical Trials Registry Platform, which includes 16 other international registries.⁶⁷  Inconsistencies with self-reported data entry and verification and modifications to the database structure over time are other constraints of ClinicalTrials.gov.^1,15  Second, there are potential biases in the methodology used for sponsor classification. Although our approach has been previously described, it did not allow us to reflect the degree of involvement when multiple funding types had contributed to a trial. As a result, we likely overestimated the size of the industry trial portfolio. Finally, our cohort and multivariable analyses are robust, but the exclusion of trials enrolling both children and adults and lack of funding data and other key trial features are additional limitations.

Although there are encouraging recent developments, challenges with the completion and timely reporting and publication of pediatric trial results persist. Our study outcomes are significantly associated with important trial characteristics, such as funding source. As the pediatric clinical trial ecosystem expands, so too must the mechanisms to address unique ethical, economic, and epidemiological barriers be prioritized. Ultimately, addressing ongoing gaps in pediatric trial completion and availability of research findings will ensure that all children receive the highest quality, evidence-based care.

aOR

adjusted odds ratio

CI

confidence interval

FDA

US Food and Drug Administration

FDAAA

Food and Drug Administration Amendments Act