Cross-sectional analysis of safety reporting in sleep disorder clinical trials

Abstract

Study objectives

Adverse event reporting is critical for evaluating the safety of interventions in sleep disorders. Inconsistent reporting between trial registries and publications may alter clinical decision-making and patient safety. We assessed the concordance of adverse event reporting between ClinicalTrials.gov and peer-reviewed publications for interventional trials in sleep disorders. Primary outcomes included the consistency of reporting serious adverse events, other adverse events and patient/event counts. The location and detail of adverse event reporting within publications were also evaluated.

Results

Among 168 trials, 143 (85.1%) showed discrepancies in serious adverse event counts between registry and publication. Patient count variations occurred in 73.2% and 35.1% of publications reported no adverse event data. Post-Final Rule trials demonstrated improved registry-based reporting, though discordance remained.

Conclusions

Substantial inconsistencies in adverse event reporting between registries and publications persist in sleep disorder trials. Improved standardisation and adherence to reporting guidelines are necessary to protect patient safety.

Trial registration details

Registered on Open Science Framework.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study is the first registry-to-publication evaluation of adverse event reporting consistency in sleep disorder clinical trials and employed duplicate blinded extraction with consensus review to enhance data reliability.

• Findings may not be generalisable beyond ClinicalTrials.gov-registered studies, and some adverse event classifications required interpretation of narrative reporting despite standardised extraction procedures.

Introduction

Sleep disorders affect millions of adults worldwide, with up to a 12% prevalence of insomnia.¹ However, they remain a widely under-recognised public health issue, despite their strong links to increased morbidity and mortality and their role as a critical marker for other medical conditions.^{2 3} As prevalence rises, clinical management is expanding, driven by greater public awareness and the growing burden of conditions such as obstructive sleep apnoea, insomnia and narcolepsy. Treatment options include pharmacotherapy, positive airway pressure (PAP) devices, neuromodulation and behavioural therapies, allowing for personalised treatments based on patient needs and preferences.^4–10 As treatment options continue to grow, so does the incidence of associated adverse events (AEs).^{11 12} It is essential that physicians understand these risks. Accurate AE reporting enables informed clinical decision-making, while inconsistent reporting can lead to preventable harm, especially with higher-acuity treatments.^13–15

Previous studies across a number of specialties have found that AE data reports are often incomplete or are inconsistent between sources.^16–21 Since implementation of the Food and Drug Administration Amendments Act 801 (FDAAA 801) and the FDAAA Final Rule of 2017, trialists have been required to report detailed AE outcomes on ClinicalTrials.gov.^{22 23} However, subsequent research has found persistent inadequacies in AE reporting.^{16 17 20 21} Such discrepancies compromise research credibility, hinder decision-making and threaten patient safety. To our knowledge, no systematic study has examined the concordance of AE reporting between ClinicalTrials.gov and their related publications within sleep medicine. The aim of this study is to characterise AE reporting in sleep disorder treatment clinical trials (CTs), identify any patterns of reporting discrepancies and provide targeted recommendations to enhance transparency and inform evidence-based practice.

Methods

Study design

We conducted a registry-to-publication comparison study to evaluate AE reporting in sleep disorder CTs. Given our use of systematic review methods to search, screen and extract trial data, we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.²⁴ Our methodology was adapted from previous studies.^{20 21} The protocol was preregistered on PROSPERO.^{25 26} This study did not qualify as human subjects research under 45 CFR 46.102(d) and (f) and was not subject to further oversight by the Oklahoma State University Center for Health Sciences Institutional Review Board.²⁷

Search strategy

On 28 June 2025, we conducted a systematic search of ClinicalTrials.gov for Phase 2–4 or ‘Not Applicable’ trials with results posted between 27 September 2009 and 31 December 2024.²⁸ This time frame aligns with the Final Rule, which mandates results submission within 1 year of trial completion (42 CFR §11.44).²⁹ We included only trials with publicly available results, excluding those with delayed or withheld reporting.

To improve search accuracy, we manually built a query using sleep disorder-specific keywords, exact phrases and Boolean operators, avoiding reliance on ClinicalTrials.gov’s automatic synonym expansion via the Unified Medical Language System.³⁰ The full search protocol is available on the Open Science Framework (OSF).³¹ The complete, line-by-line search strategy, including all keywords, exact phrases, Boolean operators and the filters and limits applied, is provided in online supplemental file 1.

Two reviewers (ML and CF) independently matched trials to publications by reviewing ClinicalTrials.gov-linked citations and performing additional manual searches in PubMed (via National Center for Biotechnology Information) and Google Scholar. Prior to full data collection, all reviewers completed structured training and participated in a calibration exercise using a pilot set of 10 studies.

We used ClinicalTrials.gov exclusively for AE data, as its standardised, FDAAA 801-mandated format enables reliable comparison with publications. In line with prior safety reviews, registries like the EU Clinical Trials Register and World Health Organization International Clinical Trials Registry Platform were excluded due to inconsistent AE reporting.^{32 33} Data were cleaned and de-duplicated in Google Sheets using built-in filters, and a full list of records is available on the OSF.³⁴ The complete list of included ClinicalTrials.gov records and their corresponding matched publications is also provided in online supplemental eTable3.

Eligibility criteria

Eligible studies were registered on ClinicalTrials.gov with posted results and investigated a primary intervention for at least one sleep disorder. Interventions included pharmacological agents (eg, sedatives, hypnotics and orexin receptor antagonists), medication adherence interventions, behavioural interventions and device-based treatments (eg, PAP, implantable medical devices). We excluded studies that: (1) focused on unrelated comorbidities, (2) lacked an English-language peer-reviewed results publication, (3) were Phase 1 or protocol-only reports, (4) were pooled analyses of multiple trials or (5) included paediatric populations.

Dataset criteria and selection

Our analysis included two trial groups: those subject to mandatory FDA reporting and those exempt but still relevant for evaluating AE reporting in sleep disorder research. Since applicable clinical trial (ACT) status is not publicly disclosed, we inferred eligibility using a rules-based method. Trials starting on or after 18 January 2017, were evaluated using the ACT checklist criteria defined in 42 CFR §11.22(b).³⁵ For trials initiated before the Final Rule, we referred to the NIH’s 2009 ACT guidance.³⁶

Eligible ACTs were interventional studies of FDA-regulated drugs or devices with a US nexus (eg, a US site, a product exported from the USA or conducted under an investigational new drug/investigational device exemption).²⁹ Drug trials had to be Phase 2–4, and device trials had to be randomised, non-feasibility studies. Inclusion also required an English-language publication and, for mixed-population studies, clearly reported sleep disorder-specific AE data.

The non-FDA-regulated dataset included trials not subject to reporting under 42 CFR Part 11 because they did not meet ACT criteria. This group included ACT-ineligible trials and expanded access studies (21 U.S.C. §360bbb), which are exempt from both federal registration and reporting requirements.³⁷

General data extraction

Two reviewers (ML and CF) independently extracted data from ClinicalTrials.gov and matched publications in a blinded, duplicate manner using a standardised Google Form. Discrepancies were resolved by consensus with a third reviewer (KK). To ensure consistency, reviewers completed a training session and calibrated extraction methods on a pilot sample of 10 studies.

We collected general trial information, including the National Clinical Trial identifier, key dates (start, completion, results posting), trial phase, sponsor and funding source. Harms data included counts of serious adverse events (SAEs), other adverse events (OAEs), AE-related withdrawals and all-cause mortality. We recorded discrepancies between registry and publication, such as differences in OAE categorisation, reporting thresholds or selective omission of harms.

We also noted whether publications referenced the Consolidated Standards of Reporting Trials (CONSORT) harms extension and whether sleep disorder-specific safety outcomes (eg, somnolence) were predefined in either source. AEs were defined per 42 CFR §11.10(a) as any unfavourable medical occurrence during the trial, regardless of causality.²⁹ Terms like ‘side effects’ and ‘complications’ were treated as equivalent to AEs. Reporting ambiguities and other relevant contextual findings were also documented. The complete data extraction form is publicly available on our OSF repository, and a blank copy is provided in online supplemental file 2.

Defining adverse events

AEs were defined according to 42 CFR §11.10(a) as any unfavourable or unintended sign, symptom or disease temporally associated with participation in a clinical trial, regardless of attribution. SAEs were defined according to ClinicalTrials.gov reporting requirements as events resulting in death, life-threatening illness, hospitalisation or prolongation of hospitalisation, persistent disability or incapacity, congenital anomaly or other medically important events. OAEs were defined as non-SAEs reported within the ClinicalTrials.gov AE module. When publications used alternative terminology such as ‘side effects’, ‘complications’, ‘harms’ or ‘treatment-emergent adverse events’, these reports were reviewed and coded according to the corresponding AE category. Definitions and classification approaches were guided by FDAAA reporting requirements and CONSORT harms recommendations.

Data analysis

We used descriptive statistics (frequencies, percentages or medians with IQRs, as appropriate for data distribution) to characterise trial attributes (eg, phase, sponsor, enrolment) and summarise AE reporting patterns across ClinicalTrials.gov and publications. To visualise the concordance of SAE counts between these sources, we generated Bland-Altman plots. We visually inspected a funnel plot (proportion of SAEs vs enrolment size) for asymmetry to explore trial size-related reporting bias. χ² tests were used to compare categorical variables (eg, presence/absence of AE reporting) across sponsor types and study phases. For non-normal outcomes like SAE counts, we applied the Mann-Whitney U test. We performed both unadjusted and adjusted linear regression analyses to identify trial-level predictors of AE reporting, including sponsor type, study phase, sample size, blinding status and trial duration. To assess changes in reporting practices over time, we also conducted segmented regression analysis using 18 January 2017, the effective date of the Final Rule, as the breakpoint.^{38 39} The dependent variable for all models was a composite AE reporting score, ranging from 0 to 7, with one point assigned for reporting (including zero values) in each of seven required domains: (1) all-cause mortality, (2) number of participants with SAEs, (3) total number of SAE events (by system and study arm), (4) number of participants with OAEs occurring in ≥5% of participants, (5) total number of OAE events (≥5%), (6) number at risk for OAEs (≥5%) and (7) the specified OAE frequency threshold. Following prior studies, entries such as ‘0’ or ‘rnc’ (reported, not calculated) were considered indicative of reporting, whereas non-mandated fields, such as the Medical Dictionary for Regulatory Activities, were excluded from the score.^{14 40} Each of the seven domains was weighted equally because all are independently mandated under the Final Rule, and no regulatory or empirical basis exists for prioritising one domain over another; equal weighting therefore avoided arbitrary, subjective weighting decisions and is consistent with the scoring approach used in prior registry-to-publication studies. Segmented regression was used to evaluate changes in AE reporting before and after implementation of the Final Rule. The Durbin-Watson statistic indicated no significant autocorrelation in either the pre-2017 (1.650) or post-2017 (1.785) periods, supporting the validity of the time-series analysis. However, trends observed after 2020 were interpreted with caution due to smaller sample sizes during that period.

All statistical analyses were conducted in R (V.2024.09.1) and Python (V.3.13.5), with significance defined as p<0.05.^41–43

Risk of bias and confidence in findings

We did not apply traditional bias assessment tools, as the study focused on AE reporting consistency rather than clinical outcomes. Instead, trials were categorised by reporting pattern: complete, partial or discordant. Confidence in our findings is supported by a rigorous methodology, including systematic searches, dual independent data extraction and careful matching of registry records to publications.

Protocol amendments and deviations

The protocol was preregistered with PROSPERO and posted to OSF prior to data extraction, with all subsequent modifications and their rationales documented both in this manuscript and on OSF.^{26 31}

Patient and public involvement

Patients and members of the public were not involved in the design, conduct, reporting or dissemination of this study.

Results

We initially identified 670 records through our ClinicalTrials.gov search. After screening, 456 were excluded, leaving 214 for publication matching and full-text review. An additional 46 were excluded, resulting in 168 trials included in the final analysis (see eFigure 1 for details).

Of these 168 sleep disorder trials, most (68%) began before the FDAAA Final Rule. Drug interventions were most common (51%), followed by devices (28%) and behavioural therapies (16%). Most trials were randomised (89%) and masked (72%). Funding came from multiple sources (47%) or industry (35%), while 8.3% did not report funding. Full characteristics are in online supplemental eTable1.

AE reporting in studies is outlined in table 1. There was no significant difference in SAE reporting between ClinicalTrials.gov and publications, before or after the Final Rule (p=0.053, p=0.118). Non-reporting of SAEs remained high overall, particularly in publications (71% pre, 65% post). Median SAE event counts were low across all groups (typically 0–1). However, publications reported significantly more patients affected by SAEs pre-Final Rule (median 2 vs 0; p=0.009), despite low event counts, suggesting selective reporting of patient-level harms.

Table 1. Adverse event reporting in studies before and after Final Rule implementation.

Adverse event domain	Study start before Final Rule (N=114)			Study start after Final Rule (N=54)
	ClinicalTrials.gov	Publication	P value	ClinicalTrials.gov	Publication	P value
Number of studies reporting AEs, by type
Serious adverse events	48/114 (42%)	33/114 (29%)	0.053*	27/54 (50%)	18/54 (33%)	0.118*
Other adverse events	61/114 (54%)	28/114 (25%)	<0.001†*	27/54 (50%)	17/54 (31%)	0.078*
Deaths	73/114 (64%)	21/114 (18%)	<0.001†*	54/54 (100%)	6/54 (11%)	<0.001†*
Studies reporting zero AE, by type
Serious adverse events	20/114 (18%)	19/114 (17%)	1.000*	18/54 (33%)	12/54 (22%)	0.283*
Other adverse events	19/114 (17%)	7/114 (6%)	0.022†*	3/54 (6%)	8/54 (15%)	0.202*
Deaths	64/114 (56%)	14/114 (12%)	<0.001†*	50/54 (93%)	5/54 (9%)	<0.001†*
Studies not reporting AEs, by type
Serious adverse events	66/114 (58%)	81/114 (71%)	0.053*	27/54 (50%)	35/54 (65%)	0.173*
Other adverse events	53/114 (46%)	77/114 (68%)	0.002†*	27/54 (50%)	35/54 (65%)	0.173*
Deaths	41/114 (36%)	92/114 (81%)	<0.001†*	0/54 (0%)	48/54 (89%)	<0.001†*
Median number of AE per trial (IQR/range)
Serious adverse events	1 (0.0–4.0/0–60)	0 (0.0–2.0/0–107)	0.238‡	0 (0.0–1.0/0–41)	0 (0.0–1.0/0–9)	0.989‡
Other adverse events	8 (0.0–64.0/0–1978)	4.5 (0.8–20.5/0–430)	0.502‡	8 (1.0–16.5/0–647)	1 (0.0–15.0/0–78)	0.086‡
Deaths	0 (0.0–0.0/0–7)	0 (0.0–1.0/0–29)	0.036†‡	0 (0.0–0.0/0–2)	0 (0.0–0.0/0–2)	0.411‡
Median number of affected patients per trial (IQR/range)
Serious adverse events	0 (0.0–1.0/0–74)	2 (0.0–3.0/0–44)	0.009†‡	0 (0.0–0.8/0–30)	0 (0.0–2.5/0–30)	0.052‡
Other adverse events	5 (0.0–33.8/0–344)	29.5 (1.0–113.0/0–526)	0.001†‡	6.5 (0.0–28.8/0–210)	22.5 (1.0–51.5/0–304)	0.069‡
Deaths	0 (0.0–0.0/0–70)	0 (0.0–1.0/0–29)	0.036†‡	0 (0.0–0.0/0–2)	0 (0.0–0.0/0–2)	0.411‡

Data are presented as n/N (%) for categorical outcomes, where N is the number of trials in each period (N=114 before the Final Rule; N=54 after). Median rows are presented as median (IQR/range).

Pre-Final Rule trials with AE events reported in publication but not calculable: 0 SAE, 9 OAE, 1 death.

Post-Final Rule trials with AE events reported in publication but not calculable: 1 SAE, 2 OAE, 0 deaths.

^*Based on χ2 or Fisher’s exact test.

^†Statistically significant, p<0.05.

^‡Based on Mann-Whitney U test.

AE, adverse event; OAE, other adverse event; SAE, serious adverse event.

OAEs were reported significantly less in publications compared with ClinicalTrials.gov before the Final Rule (25% vs 54%; p<0.001), with a non-significant but notable gap after (31% vs 50%; p=0.078). Non-reporting was also consistently higher in publications (68% pre, 65% post) than in ClinicalTrials.gov (46% pre and 50% post). Median OAE event counts did not significantly differ, but interestingly, pre-Final Rule publications had reported significantly more affected patients than in the registry (median 29.5 vs 5; p=0.001). This pattern was also seen in the post-Final Rule comparison but was non-significant (p=0.069).

Reporting of death rate was more frequent on ClinicalTrials.gov than in publications across both periods, with large, significant gaps: 64% versus 18% pre-Final Rule and 100% vs 11% post-Final Rule (p<0.001 for both). Although median death counts and affected patient numbers were 0 across all groups, publications showed greater variability pre-Final Rule, with wider IQRs.

Concordance in SAE reporting between ClinicalTrials.gov and publications was generally low. Patient count discrepancies were present in 73.2% (123/168) of all included studies. Additionally, SAE event counts were commonly missing from both sources (75/143, 52.4% of discordant pairs) or quantitatively discrepant between the publication and registry (68/168, 40.5%), with slightly lower rates post-Final Rule compared with pre-Final Rule. Across both time periods, registry reporting was more complete than in publications. While all studies had AE tables on ClinicalTrials.gov, only 63.1% (106/168) of publications included AEs in the results section, either as narrative or in tables. Notably, 35.1% (59/168) of all publications did not report any AE data. This occurred in 25.0% (13/52) of pre-Final Rule ACTs and 14.3% (5/35) of post-Final Rule ACTs. Because several denominators shift across subanalyses as a result of missing publication data, counts and denominators are reported explicitly throughout table 2. More data can be found in table 2.

Table 2. Concordance of SAE reporting and documentation of AE.

	All studies (n=168)	FDA-regulated studies
		Pre-Final Rule start date (n=52)	Post-Final Rule start date (n=35)
Patient count mismatch (ClinicalTrials.gov vs publication)
No	45/168 (26.8)	15/52 (28.8)	14/35 (40.0)
Yes	123/168 (73.2)	37/52 (71.2)	21/35 (60.0)
More in publication	8/123 (6.5)	3/37 (8.1)	4/21 (19.0)
More in registry	30/123 (24.4)	10/37 (27.0)	5/21 (23.8)
Publication NR, registry 0	85/123 (69.1)	24/37 (64.9)	12/21 (57.1)
SAE event count mismatch (ClinicalTrials.gov vs publication)
No	25/168 (14.9)	7/52 (13.5)	8/35 (22.9)
Yes	143/168 (85.1)	45/52 (86.5)	27/35 (77.1)
More in publication	4/143 (2.8)	2/45 (4.4)	2/27 (7.4)
More in registry	27/143 (18.9)	9/45 (20.0)	5/27 (18.5)
Publication NR, registry 0	23/143 (16.1)	9/45 (20.0)	8/27 (29.6)
Registry NR, publication 0	14/143 (9.8)	5/45 (11.1)	3/27 (11.1)
Both NR	75/143 (52.4)	20/45 (44.4)	9/27 (33.3)
Location of AE reporting (publication)
Results (in-text narrative)	106/168 (63.1)	37/52 (71.2)	29/35 (82.9)
Table/figure	54/168 (32.1)	21/52 (40.4)	18/35 (51.4)
Discussion	46/168 (27.4)	19/52 (36.5)	12/35 (34.3)
Supplementary file	5/168 (3.0)	1/52 (1.9)	1/35 (2.9)
Not reported	59/168 (35.1)	13/52 (25.0)	5/35 (14.3)

Data are presented as n/N (%). Denominators differ across subanalyses because of missing publication data and are shown explicitly for every cell.

For the mismatch subcategories (eg, ‘More in publication’, ‘More in registry’, ‘Both NR’), the denominator is the number of discordant (‘Yes’) trials in each column.

For ‘Location of AE Reporting’, the denominator is the total number of studies in each column; categories are non-mutually exclusive (a publication may report AEs in more than one location), so columns may sum to more than 100%.

FDA-regulated studies comprise ACTs; the pre–Final and post–Final Rule columns together total 87 ACTs.

ACTs, applicable clinical trials; AE, adverse event; FDA, Food and Drug Administration; NR, not reported; SAE, serious adverse event.

The regression analysis in online supplemental eTable2 examines the impact of the FDAAA Final Rule on AEs reporting scores. In both unadjusted (p=0.019) and adjusted models (p=0.014), reporting scores were significantly lower before the FDAAA Final Rule, indicating improved reporting after its implementation. The adjusted model controlled for intervention type, funding source and participant enrolment, but none of these additional variables were statistically significant predictors of reporting scores (all p>0.05). Effect estimates with 95% CIs for all model terms are reported in online supplemental eTable2.

Figure 1 represents a Bland-Altman plot comparing total SAE counts reported in clinical trial registries versus publications for the 33 studies that documented specific values for both variables. The mean difference between sources was 1, indicating that trials tended to report more SAEs on their registries than the corresponding publications. However, while many trials clustered near the concordance line, some studies exhibited marked discrepancies, resulting in a wide CI.

Figure 1. Bland-Altman plot comparing adverse event counts reported in ClinicalTrials.gov and corresponding publications. The Y-axis represents the difference in AE counts (registry—publication), while the X-axis shows the average AE count from both sources. The solid horizontal line represents zero difference (ie, equal AE counts in both sources). The dashed line indicates the mean difference across trials, and the dotted lines represent the 95% limits of agreement (±1.96 SD from the mean). Three trials with extreme differences were trimmed to improve visualisation and avoid distortion of the plot scale. Points above the upper confidence band represent trials in which the registry reported substantially more AEs than the publication, suggesting potential under-reporting in the manuscript. Points below the lower confidence band indicate the reverse. Arrows highlight the direction of discrepancies favouring either the registry or the publication. Each point represents a single clinical trial. Triangles represent trials post-Final Rule. Circles represent trials post-Final Rule. AE, adverse event; FDAAA, Food and Drug Administration Amendments Act; SAE, serious adverse event.

Figure 2A shows a funnel plot of SAE rates by enrolment size across sleep disorder trials. Most clustered around low SAE rates, consistent with low-risk interventions, though under-reporting may contribute. A few trials showed higher rates, possibly due to higher-risk therapies, different populations or stricter AE detection. Smaller trials showed more variability, while larger ones had more stable rates. Figure 2B, limited to FDA-regulated trials, showed a mean SAE rate of 2.1 versus 1.8 in the full sample, suggesting non-regulated trials may have slightly lowered the overall mean. A segmented regression analysis showed no significant inflection point in AE reporting scores following the 2017 FDAAA Final Rule (figure 3). Mean scores remained relatively stable before and after the policy, with only a slight, non-significant upward trend. Overlapping CIs suggest these differences likely reflect random variation rather than a policy effect. Notably, no trials after the Final Rule scored below 4. This gradual improvement may reflect incremental gains in reporting practices independent of regulatory changes.

Figure 2. (A) The funnel plot displays the reported SAE rates (% of affected participants) against trial enrolment size (n) for 162 included trials. The horizontal dotted line represents the mean SAE rate with solid lines indicating the 95% CI. Six trials were trimmed in order to enhance the symmetry and visual presentation of the figure (trials with over-enrolment far beyond others in the dataset were trimmed for both visual and statistical calculation of mean/CI). Trials are categorised by the regulatory period. Triangles represent clinical trials initiated pre-Final Rule. Circles represent studies initiated post-Final Rule. The distribution reflects heterogeneity in SAE reporting that may not be due to chance. (B) The funnel plot displays the reported SAE rates (% of affected participants) against trial enrolment size (n) for 84 included likely applicable clinical trials. The horizontal dotted line represents the mean SAE rate with solid lines indicating the 95% CI. Three were trimmed in order to enhance the symmetry and visual presentation of the figure (trials with over-enrolment far beyond others in the dataset were trimmed for both visual and statistical calculation of mean/CI). Trials are categorised by the regulatory period. Triangles represent studies started pre-Final Rule. Circles represent trials started post-Final Rule. The distribution reflects heterogeneity in SAE reporting that may not be due to chance. SAE, serious adverse event.

Figure 3. Segmented linear regression of AE reporting scores over time, stratified by trial start date (X-axis). The vertical dashed line marks the implementation of the FDAAA Final Rule on 18 January 2017. Each black dot represents the AE reporting score of an individual trial. Regression lines with 95% CIs depict trends in reporting scores before and after the Final Rule implementation. AE, adverse event; FDAAA, Food and Drug Administration Amendments Act.

Discussion

AE reporting remains an essential yet inconsistently executed aspect of clinical research, particularly in sleep medicine.^44–47 Our cross-sectional analysis of registry-to-publication AE reporting revealed persistent discrepancies between ClinicalTrials.gov and peer-reviewed publications. We assessed the frequency, completeness and alignment of serious and non-SAEs across 168 sleep disorder trials. Our results show that AE reporting was consistently more complete on ClinicalTrials.gov, especially after the implementation of the FDAAA Final Rule in 2017. For example, 100% of post-Final Rule trials reported mortality rates in registries, while only 11% did so in publications. Although registry reporting of OAEs and deaths improved modestly over time, publication reporting did not show similar progress. Across both pre-Final Rule and post-Final Rule periods, ClinicalTrials.gov consistently reported a higher number of AEs than corresponding publications. Additionally, 35.1% of publications reported no AE data, and over 85% of studies showed SAE mismatches between sources. Our analysis highlights a potential risk to both patient safety and the integrity of clinical evidence in the context of harms reporting.

These findings reflect widespread inconsistencies and align with previous studies that have shown similar gaps in reporting across other disciplines, including psychiatry, urology and oncology.^48–52 While examining a broad range of clinical trials, Earley et al found unclear and inconsistent death reporting.⁵³ It has been suggested that deaths may be omitted due to the absence of mortality risk; however, the lack of standardised reporting remains problematic.^{33 50} This inconsistency has been found to extend to SAEs as well. For example, Tang et al found that many studies from different medical specialties lacked group-level SAE data, and a small portion failed to mention SAEs.¹⁶ Our findings align with these studies, adding a focused analysis of AE reporting patterns in sleep medicine.

Such deficiencies threaten clinician decision-making and patient safety, highlighting the need for stricter publication standards and stronger alignment with registry transparency in sleep medicine. Inconsistencies may stem from low physician compliance, difficulty evaluating causality or participant misunderstanding.^54–57 This is particularly concerning in a field where patients often belong to vulnerable populations, including older adults, individuals with obesity and those with comorbid cardiopulmonary or neuropsychiatric conditions, who may be more susceptible to treatment-related harms.^58–61 While ClinicalTrials.gov generally provided more complete AE data, divergence from published results leaves an evidence gap that clinicians must navigate without clear guidance. Because many sleep therapies are employed long-term, the lack of reliable AE data may undermine benefit-harm analyses essential to care. Guideline developers such as the American Academy of Sleep Medicine (AASM) and the National Center on Sleep Disorders Research rely heavily on published trial data, meaning incomplete harms reporting may directly impact care recommendations.^{62 63} Clinicians should be cautious when interpreting safety claims based solely on these guidelines and may benefit from routinely consulting registry data when evaluating sleep disorder interventions.

To address these concerns, a multifaceted approach is needed. Key safety outcomes like SAEs and mortality remain under-reported in publications despite mandates such as FDAAA 801, the Final Rule and the CONSORT harms extension.^{23 64 65} Journals could improve alignment with registries by adopting standardised harms reporting formats to streamline cross-referencing and enhance transparency. Regulatory agencies could expand audits and link reporting completeness to trial registration or funding eligibility, promoting accountability through clear consequences. Additionally, the AASM could develop a structured Sleep Medicine Harms Reporting Framework to improve AE reporting clarity by requiring stratified harms data by intervention type (eg, pharmacologic, PAP, behavioural). Integrating registry data into clinical guidelines would further address evidence gaps when publication data is insufficient. Collectively, these strategies would strengthen AE reporting integrity, bridge registry-publication gaps and support more informed, patient-centred care.

This study has limitations that warrant consideration. By analysing both FDA-regulated and non-regulated trials, we captured a broad overview of AE reporting practices in sleep medicine. However, relying on ClinicalTrials.gov may limit generalisability to non-US trials, and some linked publications may have been missed despite manual searches. Differences in AE reporting may also reflect factors beyond our control, such as sponsor preferences or journal policies. Finally, interpretation of narrative AE data involved a level of subjectivity that was mitigated through blinded review and consensus. Additionally, our composite reporting score assigned equal weight to each reporting domain; although this approach minimises subjective weighting decisions, alternative weighting schemes could yield different estimates of reporting completeness.

Some discrepancies may also reflect differences in reporting timelines and adjudication processes between ClinicalTrials.gov and peer-reviewed publication. AEs reported to trial registries are often submitted to satisfy regulatory reporting requirements and may subsequently undergo additional review, adjudication or reclassification before manuscript publication. Furthermore, reporting deadlines may differ between registries, sponsors and journals, resulting in temporal differences between data sources. These processes could reasonably account for minor discrepancies in AE counts or classifications. However, they are unlikely to fully explain the substantial reporting gaps observed in our study, including complete omission of AE domains, large differences in reported participant counts and the absence of any AE data in more than one-third of publications. While timing and adjudication differences likely contribute to some discordance, they do not appear sufficient to explain the magnitude of discrepancies observed across many trial-publication pairs.

An important additional limitation of this study is that AE counts were sparse and highly zero-inflated across many reporting domains, with median event counts frequently equal to zero. Consequently, statistically significant differences should be interpreted cautiously, as p values alone may not fully reflect the magnitude or practical significance of reporting discrepancies. Although our analyses focused on identifying reporting inconsistencies between sources, future studies should incorporate effect size estimates and CIs to better quantify the extent of these differences.

Conclusion

In this registry-publication comparison of sleep disorder trials, AE reporting was more complete on ClinicalTrials.gov, especially for deaths, SAEs and OAEs. Despite regulatory efforts like the FDAAA Final Rule, many publications still omit key safety data, limiting clinicians’ ability to assess risks accurately. Our findings underscore the need for more consistent reporting standards and greater accountability across research in sleep medicine. Improving transparency across registries and journals would support safer, more informed clinical decision-making and help build trust in the evidence base guiding everyday practice.

Data availability statement

Data are available in a public, open access repository.