The Trial Bank initiative to promote trustworthy living evidence for medication safety (PROMISE) of children: design and pilot validation

Chang Xu; Suhail A Doi; Sunita Vohra; Luis Furuya-Kanamori; Yingxia Wang; Lijun Tang; Yuan Tian; Shiqi Fan; Zhen Peng; Evan Mayo-Wilson; Lifeng Lin; Haitao Chu; M Hassan Murad; Tianqi Yu; Sheyu Li; Yaolong Chen; Su Golder; Yoon Loke; Justin Clark; Paul Glasziou; Liang Du; Houwen Lin; Zhichun Gu; PROMISE Collaborative Research Network

doi:10.1186/s12916-026-04709-y

. 2026 Feb 21;24:181. doi: 10.1186/s12916-026-04709-y

The Trial Bank initiative to promote trustworthy living evidence for medication safety (PROMISE) of children: design and pilot validation

Chang Xu ^1,^2,^✉, Suhail A Doi ³, Sunita Vohra ⁴, Luis Furuya-Kanamori ⁵, Yingxia Wang ⁶, Lijun Tang ⁶, Yuan Tian ¹, Shiqi Fan ¹, Zhen Peng ¹, Evan Mayo-Wilson ⁷, Lifeng Lin ⁸, Haitao Chu ^9,¹⁰, M Hassan Murad ¹¹, Tianqi Yu ¹², Sheyu Li ¹³, Yaolong Chen ¹⁴, Su Golder ¹⁵, Yoon Loke ¹⁶, Justin Clark ¹⁷, Paul Glasziou ¹⁷, Liang Du ^18,^✉, Houwen Lin ^2,^19,^20,^✉, Zhichun Gu ^2,^19,^20,^✉; PROMISE Collaborative Research Network

¹Proof of Concept Center, Eastern Hepatobiliary Surgery Hospital, Third Affiliated Hospital, Second Military Medical University, Naval Medical University, No.225 Changhai Road, Shanghai, China

²Department of Pharmacy, Punan Branch of Ren Ji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

³Department of Population Medicine, College of Medicine, QU Health, Qatar University, Doha, Qatar

⁴Department of Pediatrics and Department of Psychiatry, Faculty of Medicine & Dentistry, University of Alberta, Edmonton, AB Canada

⁵UQ Centre for Clinical Research, Faculty of Medicine, University of Queensland, Brisbane, Australia

⁶School of Pharmaceutical Sciences, Anhui Medical University, Anhui, China

⁷Department of Epidemiology, UNC Gillings School of Global Public Health, University of North Carolina-Chapel Hill, North Carolina, USA

⁸Department of Epidemiology and Biostatistics, University of Arizona, Tucson, AZ USA

⁹Statistical Research and Data Science Center, Data Sciences and Analytics, Pfizer Inc, New York, NY USA

¹⁰Division of Biostatistics and Health Data Science, University of Minnesota School of Public Health, Minneapolis, MN USA

¹¹Mayo Clinic Evidence-Based Practice Center, Rochester, MN USA

¹²Université Paris Cité, Research Center of Epidemiology and Statistics (CRESS-U1153), INSERM, Paris, France

¹³Department of Endocrinology and Metabolism, Division of Guideline and Rapid Recommendation, Cochrane China Centre, MAGIC China Centre, Chinese Evidence-Based Medicine Centre, West China Hospital, Sichuan University, Chengdu, China

¹⁴Evidence-Based Social Science Research Centre, School of Public Health, Lanzhou University, Lanzhou, China

¹⁵Department of Health Sciences, University of York, York, UK

¹⁶Norwich Medical School, University of East Anglia, Norwich, UK

¹⁷Institute for Evidence-Based Healthcare, Faculty of Health Sciences and Medicine, Bond University, Queensland, Australia

¹⁸Innovation Institute for Integration of Medicine and Engineering, Chinese Evidence-Based Medicine Center, West China Hospital, Sichuan University, Chengdu, China

¹⁹Department of Pharmacy, Ren Ji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

²⁰College of Clinical Pharmacy, Shanghai Jiao Tong University School of Medicine, Shanghai, China

^✉

Corresponding author.

PMCID: PMC13032608 PMID: 41723462

Abstract

Background

While medications are essential for preventing and treating disease, they can also cause harm. Evidence synthesis has been widely adopted for evaluating harms, but traditional methods are resource-intensive and may constrain timely decision-making. This study aims to validate a Trial Bank approach towards rapid evidence synthesis.

Methods

A Trial Bank consisting of 13,650 RCTs of pharmaceutical or biopharmaceutical agents for children was established using artificial intelligence (AI) and humans, based on five databases (e.g., PubMed) up to February 14, 2023. The Trial Bank approach for evidence synthesis was validated in two ways: First, the percentage of trials within 1,996 Cochrane meta-analyses of drug safety in children that were also available in the Trial Bank was reported as the Trial Bank coverage (TBC). Second, the agreement of pooled effects from trials limited to those in the Trial Bank was compared to the full Cochrane meta-analyses in terms of their direction and statistical significance.

Results

Of 1,020 trials included in the Cochrane meta-analyses, there was an overall 80.2% TBC, with an average TBC of 85.7% per meta-analysis (n = 1,996). With regards to agreement of meta-analytical results, use of only the Trial Bank trials achieved an agreement of 93.0% (95% confidence interval [CI]: 90.8% to 94.8%) in the direction, 95.8% (95%CI: 94.0% to 97.2%) in significance, and 89.1% (95%CI: 94.0% to 97.2%) in both direction and significance to Cochrane meta-analytical results for meta-analyses that had 2 or more trials (n = 668). Sensitivity analysis by removing unpublished trials from Cochrane meta-analyses showed slightly higher agreement (e.g., 90.7% in both direction and significance).

Conclusions

The Trial Bank approach demonstrated considerable coverage and agreement with Cochrane meta-analyses, suggesting it is a potentially feasible and efficient strategy for supporting living evidence synthesis of medication safety in children.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04709-y.

Keywords: Trial Bank, Medication harms, Living evidence, Rapid evidence synthesis

Background

Medications provide essential benefits for preventing and treating illnesses, but they can also cause harm. Medication-related harm refers to any unfavorable or adverse health outcome—whether expected or unexpected—linked to the use of medical products [1]. Harms can be serious, including hospitalization, disability, or even death [2]. According to estimates from the Global Burden of Disease project, medication-related harms contributed to 62.79 disability-adjusted life years (DALYs) and 58.33 years of life lost (YLL) per 100,000 population in 2017, especially for children under 4 years old [3]. Each year, approximately 134 million adverse events result in 2.6 million deaths globally [3–5]. In response to this substantial burden, the WHO Global Patient Safety Action Plan 2021–2030 emphasizes the goal of achieving ‘the maximum possible reduction in avoidable harm due to unsafe health care globally’ [6–8].

Randomized controlled trials (RCTs) are used to evaluate medication-related harms [9, 10]. However, individual RCTs often fall short of providing robust harm data due to low event rates, limited sample sizes, short follow-up durations, and patient selection prioritizing those with lower risk of harm [11–13]. Synthesizing evidence from multiple trials can improve the precision of harm estimates [14]. While traditional methods such as systematic reviews (including living systematic reviews) [15] have been adopted for this purpose, there exist two major challenges: the reviews are highly resource-intensive [16] and they can become outdated very quickly [17]. A rigorous systematic review typically takes 6 months to 2 years to complete [18]. Meanwhile, new trial data continue to emerge, and delays in identifying and incorporating new evidence can hinder timely, evidence-based decisions [19].

The concept of a Trial Bank presents a promising solution to these challenges [20]. First proposed by Ida Sim in 1995, followed by the launch of the Global Trial Bank initiative in 2005, the idea was to establish a continuously updated repository of structured trial data for rapid evidence synthesis [21]. However, progress has been limited by the intensive labor required to collect and format structured trial data. Despite the U.S. Food and Drug Administration’s (FDA) 2008 launch of the ClinicalTrials.gov “results database”, the vision remains unfulfilled [22]. As of April 15, 2025, only 55,268 of 229,107 completed interventional trials (24.1%) have reported results on ClinicalTrials.gov.

Recent advancements in artificial intelligence (AI), particularly large language models, offer new opportunities to overcome these barriers. Proof-of-concept studies have demonstrated that AI tools can achieve good accuracy in tasks such as literature screening and data extraction—critical components for building a Trial Bank [23–25]. In this context, we introduce the PROMISE (Promoting Medication Safety) project, a new Trial Bank initiative by our group focused on assessing medication-related harms in children via aggregated structured trial data. We focused on children because current evidence about medication harms in children is limited, as most trials continue to focus on adults, with only 14.2% being in children [26]. This study presents the design of an AI-supported, updatable trial bank infrastructure. As part of the evaluation, a pilot validation was performed to compare the system’s adverse event outcome outputs with corresponding findings from Cochrane systematic reviews.

Methods

Design and settings

The project involves four main aspects: 1) a systematic literature search 2) data repository, 3) terminology standardization, and 4) evidence synthesis (Fig. 1). To date, the project has completed the identification and collection of relevant RCTs, along with the extraction of their baseline characteristics. We used the definition of adverse events defined in the eligible trials as the definition of adverse events in the PROMISE project, which generally refers to ‘any untoward medical occurrence in a patient or subject in clinical practice’ [27]. The overall project is based on a pre-defined protocol (see Additional file 1: protocol).

Fig. 1 — The Trial Bank paradigm for rapid evidence synthesis in children

Establishment of the Trial Bank

Details regarding the procedures for setting up our Trial Bank are documented in the protocol (Additional file 1: protocol). First, a comprehensive search was conducted on February 14, 2023, across five electronic databases: PubMed, EMBASE, Scopus, Web of Science, and the Cochrane Central Register of Controlled Trials (CENTRAL). The search, conducted by an information specialist, aimed to identify RCTs of drug interventions in children, with no restrictions on publication date or language. We included trials that enrolled only children as well as trials that included both children and adults. The United Nations Convention on the Rights of the Child defines a “child” as a person under 18 years of age [28], and we also considered trials that used other age thresholds (e.g., under 21 years).

We included trials in which at least one study arm involved a pharmaceutical compound or biopharmaceutical agent, regardless of the composition, formulation, dosage, or administration route. Trials were excluded if all active intervention arms consisted solely of non-pharmaceutical interventions. We excluded retracted trials due to concerns over data quality and the risk of evidence contamination [29]. We imposed no restrictions on trial outcomes.

We first used the RobotSearch, a machine learning model with an estimated diagnostic accuracy of 0.987, for preliminary classification of the literature automatically, as potential RCTs or not [30]. Then, 10 well-trained team members (CX, XY, RZ, XY.Z, YZ, XL.Z, FF.H, MY.Z, HR.Z, ZS.G) screened titles and abstracts of these potential RCTs for population and intervention. Further full-text screening was conducted by 62 well-trained medical student volunteers, two of whom read each report independently. Finally, our team members reviewed the full text of the publications marked as ‘maybe’ and ‘conflicts’ by the student volunteers and made final decisions. Literature screening by humans used the Rayyan online tool.

We extracted data from full-text reports and, where possible, registry platforms (e.g., ClinicalTrials.gov). The following baseline characteristics of eligible trials were collected: name of the first author, digital object identifier (DOI) number, year of publication, contact information for the corresponding author, registration status, registration number, geographic region, number of trials centers, population age, the name of intervention and comparison(s), design of the trial, sample size, source of funding, data sharing statements, and phase of the trial. Claude 2 was used for the data extraction, followed by iterative human verification process by well-trained student volunteers, namely, those who reached a minimal agreement of 90% (calculated per item over all studies correctly extracted) during training. For each round of human verification, a random sample of 5% was selected for a further checking to estimate the percentage of agreement. The procedure was repeated until the agreement rate surpassed 95%, at which point the verification process was discontinued. The prompts of the Claud 2 were presented in Additional file 1.

Validation of the Trial Bank approach

The validation study was conducted by comparing Trial Bank results to Cochrane reviews on the same question. We chose Cochrane reviews as the current best approach [31] and therefore the current gold standard. We included Cochrane reviews up to January 7, 2025 that evaluated drug interventions in children. From the eligible reviews, we collected those meta-analyses (including subgroup analyses) that focused on children. We then conducted two pilot validation assessments: 1) the coverage of our Trial Bank by comparing the trials included in the corresponding Cochrane meta-analyses with those available in the Trial Bank and 2) Accuracy of the estimates produced by the trial bank using the same analytic settings as in the original Cochrane analyses (i.e., synthesis method, effect measure, analytic model, and trial data). To conduct the trial bank meta-analysis, we re-conducted each meta-analysis after restricting the included trials to those only available in the Trial Bank, thereby obtaining Trial Bank–based effect estimates. We extracted trial data from the Cochrane meta-analyses directly because outcome data collection and verification of our Trial Bank has not yet finished. Trial bank estimates were compared with those from the original Cochrane meta-analyses to assess the accuracy of the Trial Bank approach. Since the Trial Bank collected trials up to February 14, 2023 during the validation study, we excluded trials in Cochrane reviews published after this date.

We filtered Cochrane reviews by “Intervention” and “Child health” topic without any other search strategies or restrictions. Then, two authors (ZP and YT) screened search records independently for eligibility using Rayyan. We manually screened the records and included systematic reviews with randomized trials of drugs and at least one meta-analysis focused on adverse events in children. Only those excluded by both reviewers were excluded in the first stage; the remaining records were screened by full texts and any disagreements were solved by discussion. Data from the eligible meta-analyses were downloaded directly from the Cochrane review of the ‘rm5’ file. We compared the titles of studies included in Cochrane reviews and the Trial Bank using the ‘MATCH’ function in Excel (Microsoft, USA) by two authors, which a third author (SQ.F) then verified.

Our primary outcome was the percentage of studies within each eligible Cochrane meta-analysis that were similarly found in the Trial Bank (i.e., Trial Bank coverage; TBC). A second primary outcome was the proportion of meta-analyses in which the pooled effect estimates using the Trial Bank had the same direction and statistical significance as those reported in the corresponding Cochrane reviews (i.e., agreement). The secondary outcomes were the absolute differences in the magnitude of the pooled effect sizes as well as the P values for the Trial Bank versus Cochrane meta-analyses.

Statistical analysis

Baseline information for eligible trials in the Trial Bank were summarized as proportions or median values with interquartile range (IQR), depending on the type of data. For the main analysis, the TBC was summarized for each Cochrane meta-analysis separately. To better reflect the distribution of the TBC, we categorized it into the following four groups: [0.0%, 50%), [50%, 80%), [80%, 100%), and 100%. Because our current search does not include unpublished trials, we conducted a sensitivity analysis by excluding unpublished trials within Cochrane meta-analyses from the denominator.

With regards to the accuracy of the Trial Bank approach, the proportions and its 95% confidence interval (CI) of meta-analyses without a change in terms of the direction of the pooled effect and the significance of P value were estimated for those with 2 or more included studies. As the number of studies included in a meta-analysis may impact the robustness of the final effects, especially for those with few included studies, we further summarized the proportions by three categories: 2 to 4 studies, 5 to 9 studies, and 10 or more studies. Again, we conducted a sensitivity analysis by excluding unpublished trials.

We conducted all data analyses in Stata/SE 16.0 (Stata Corp LCC, College Station, TX), and visualization was undertaken using Excel 2016 (Microsoft, Washington). The code used for the analysis is presented in the Additional file 1.

Results

From the five databases, we obtained 947,877 records in total, with 229,110 identified as duplicates. The RobotSearch RCT classifier identified 216,258 records as potential RCTs. Based on titles and abstracts, we excluded 150,679 records; based on full-text screening, we identified 16,993 randomized controlled trials involving children in our Trial Bank, of which 13,650 focused on children only and 1,519 involved children and adults. Figure 2 presents details of the screening process for the Trial Bank.

Fig. 2 — Flow chart of literature screening for pediatric trials. Examples: 1) Study excluded due to wrong design (Dalla et al. Epilepsia. 1995;36(7):687–691.); 2) Study excluded due to wrong intervention (Pediatrics. 2015;136(5):885–894.)

Trial characteristics

We extracted baseline data for the 13,650 trials of children only, with an estimated agreement of 93.5% after the first round of human verification and 96.1% after the second round, from a random sample of 763 and 674 trials separately. We then stopped here as this agreement reached our expectation. We are not sure if a third round of verification would increase the agreement further. The original data for the random sample test has been shared in OSF [32].

Among the 13,650 trials, data was unavailable for 185 due to lack of full-text or blurred PDF (Portable Document Format) file. The number of trials has increased in the past six decades, from 1 in 1956 to 560 in 2022 (Additional file 1: Fig. S1). With regard to geography, the top three countries that contributed the most trials were the United States (1,653, 12.3%), Iran (527, 3.9%), and China (515, 3.8%), see Additional file 1: Fig. S2. The three most frequently investigated conditions were asthma (962, 7.1%), pain (773, 5.7%), and attention deficit hyperactivity disorder (570, 4.2%), see Additional file 1: Fig. S3.

There were 44.7% (6,014) single-center trials, and 29.8% (4,006) were multi-center. With regard to the design, most (11,923, 88.5%) were parallel group trials, followed by cross-over trials (1,441, 10.7%), then cluster trials (90, 0.6%). The sample size ranged from 2 to 190,238, with a median value of 70 (IQR: 40 to 141); When stratified by the design type, the median sample size was 80 (IQR: 45 to 151) for parallel trials, 25 (IQR: 16 to 44) for cross-over trials, and 692 (IQR: 123 to 3368) for cluster trials.

Almost 40.0% of trials failed to report funding information (5,249, 39.0%). For those that reported funding, 57.5% (n = 4,724) were supported by institutional or government funding bodies and 25.5% (n = 2,097) were supported by industry. Only 27.6% of the trials were registered in total, while an increasing proportion of trials were registered since 2006, with 68.2% registered after 2018. For data sharing, 673 (5.0%) trials provided data sharing statements in the full text, and most commonly upon request (59.0%, 397/673) or via a provided weblink (34.9%, 235/673) (Table 1).

Table 1.

Characteristics of the included trials for children

Characteristics	1956–1999 (N = 3,605)	2000–2005 (N = 2,031)	2006–2017 (N = 5,194)	2018–2023 (N = 2,635)	Overall (N = 13,465)
Sample size (IQR)	53 (28–106)	63 (36–142)	80 (44–156)	85 (53–153)	70 (40–141)
Study design
Parallel	2,957 (82.02%)	1,794 (88.33%)	4,703 (90.55%)	2,469 (93.70%)	11,923 (88.55%)
Cross-over	630 (17.48%)	233 (11.47%)	449 (8.64%)	129 (4.90%)	1,441 (10.70%)
Cluster	14 (0.39%)	1 (0.05%)	40 (0.77%)	35 (1.33%)	90 (0.67%)
Parallel and cross-over	4 (0.11%)	3 (0.15%)	2 (0.04%)	2 (0.08%)	11 (0.08%)
Center
Single-center	1,380 (38.28%)	804(39.59%)	2,413 (46.46%)	1,417 (53.78%)	6,014 (44.66%)
Multi-center	821 (22.77%)	650 (32.00%)	1,687 (32.48%)	848 (32.18%)	4,006 (29.75%)
Not reported	1,404 (38.95%)	577 (28.41%)	1,094 (21.06%)	370 (14.04%)	3,445 (25.58%)
Phase
I	2 (0.06%)	1 (0.05%)	46 (0.89%)	44 (1.67%)	93 (0.69%)
II	2 (0.06%)	2 (0.10%)	226 (4.35%)	198 (7.51%)	428 (3.18%)
III	2 (0.06%)	4 (0.20%)	522 (10.05%)	485 (18.41%)	1,013 (7.52%)
IV	0 (0.00%)	0 (0.00%)	303 (5.83%)	271 (10.28%)	574 (4.26%)
Others ^a	0 (0.00%)	3 (0.15%)	154 (2.96%)	172 (6.53%)	329 (2.44%)
No information	3,599 (99.83%)	2,021 (99.51%)	3,943 (75.91%)	1,465 (55.60%)	11,028 (81.90%)
Trial registration
Yes	6 (0.17%)	7 (0.34%)	1,908 (36.73%)	1,798 (68.24%)	3,719 (27.62%)
No	3,599 (99.83%)	2,024 (99.66%)	3,286 (63.27%)	837 (31.76%)	9,746 (72.38%)
Funding source
Industry/Industry-employer	508 (14.09%)	378 (18.61%)	859 (16.54%)	352 (13.36%)	2,097 (15.57%)
Non-profit	1,002 (27.79%)	616 (30.33%)	2,008 (38.66%)	1,098 (41.67%)	4,724 (35.08%)
Industry and non-profit	163 (4.52%)	104 (5.12%)	202 (3.89%)	72 (2.73%)	541 (4.02%)
No funding	36 (1.00%)	42 (2.07%)	373 (7.18%)	403 (15.29%)	854 (6.34%)
Not reported	1,896 (52.59%)	891 (43.87%)	1,752 (33.73%)	710 (26.94%)	5,249 (38.98%)
Data sharing statement
Yes	15 (0.42%)	8 (0.39%)	77 (1.48%)	573 (21.75%)	673 (5.00%)
No	3,590 (99.58%)	2,023 (99.61%)	5,117 (98.52%)	2,062 (78.25%)	12,792 (95.00%)

Open in a new tab

Others ^a: Clinical trials reported different phases (e.g., Phase Ⅰ/Ⅱ) trials in one article

Data sharing statement: a statement for outlining the availability of research data. This often reported by the research authors at end of their paper

Trial Bank approach for medication harms: Trial Bank coverage (TBC)

From the Cochrane Library, we obtained 2,808 Cochrane reviews. After screening, 255 Cochrane reviews with 1,020 drug trials (after removing duplicates) that were used to generate 1,996 meta-analyses that focused on adverse events in children (Additional file 1: Fig. S4, Tables S1 and S2). Of 1,996 meta-analyses, 1,256 included one study and 740 included 2 or more studies. And for the 1,020 trials, 30 (2.9%) were marked as unpublished trials in Cochrane reviews.

Figure 3A presents the Trial Bank coverage (TBC). Of the trials in Cochrane meta-analyses, 80.2% (n = 818/1020) were included in our Trial Bank. The average TBC per meta-analysis was 85.7%. and 1521/1996 (76.2%) Cochrane meta-analyses had all included trials in the Trial Bank. By category, 1582/1996 (79.3%) had 80% or more trials in the Trial Bank, 215/1996 (10.8%) had 50% to 80% of trials in the Trial Bank, and 199/1996 (10.0%) had less than 50% of trials in the Trial Bank.

Fig. 3 — Trial Bank coverage (TBC) of included trials in Cochrane meta-analyses (Sensitivity analysis was done by removing unpublished trials). Trial Bank coverage: the percentage of studies within an eligible Cochrane meta-analysis that were similarly found in the Trial Bank

When Cochrane meta-analyses were stratified by the number of included studies as described earlier, the average TBC for each category were generally consistent, such that the percentages with coverage of all trials (TBC = 100%) decreased as the number of included studies within a meta-analysis increased: for those meta-analyses with only 1 included study (n = 1,256), 88.2% of trials were in the Trial Bank; whereas when 10 or more studies were included in a meta-analysis (n = 68), only 19.1% had all trials covered by the Trial Bank. The proportion of meta-analyses with less than 50% of included trials in the Trial Bank was low in all categories (accounting for less than 11.8%). Sensitivity analysis by removing unpublished trials showed a slight increase in the average TBC, see Fig. 3B.

Trial Bank approach for medication harms: agreement

For the 1,256 meta-analyses with only one study, 88.2% of the trials were recorded in our Trial Bank and therefore the outputs were identical in 88.2% between Trial Bank and Cochrane. For the 740 meta-analyses with 2 or more included studies, 72 were excluded during the replication of the results, as the covered trials were not synthesized in the original meta-analyses due to double-zero events. Ultimately, 668 meta-analyses of drug safety for children were used for the analysis.

For meta-analyses from the Trial Bank compared to the Cochrane meta-analyses, the overall agreement on the direction of pooled effects was 93.0% (621/668; 95%CI: 90.8% to 94.8%), on significance was 95.8% (640/668; 95%CI: 94.0% to 97.2%), and on both direction and significance was 89.1% (595/668; 95%CI: 86.5% to 91.3%).

In stratified analysis, there were slight fluctuations based on the number of included studies with regards to agreement: for those meta-analyses with 2 to 4 included studies, the agreement was 92.5% (470/508; 95%CI: 89.9% to 94.7%) with regards to the direction of pooled effects, 96.1% (488/508; 95%CI: 94.0% to 97.6%) for significance, and 88.8% (451/508; 95%CI: 85.7% to 91.4%) for both; For those meta-analyses with 5 to 9 included studies, the agreement was 94.7% (89/94; 95%CI: 88.0% to 98.3%) for direction of pooled effects, 95.7% (90/94; 95%CI: 89.5% to 98.8%) for significance, and 90.4% (85/94; 95%CI: 82.6% to 95.5%) for both; For those meta-analyses with 10 or more included studies, the agreement was 93.9% (62/66; 95%CI: 85.2% to 98.3%) for direction of the pooled effects, 93.9% (62/66; 95%CI: 85.2% to 98.3%) for significance, and 89.4% (59/66; 95%CI: 79.4% to 95.6%) for both (Fig. 4A). Sensitivity analysis via removing unpublished trials showed slightly increased agreements; see Fig. 4B.

Fig. 4 — Agreement of the Trial Bank approach with Cochrane meta-analysis results (Sensitivity analysis was done by removing unpublished trials)

The comparison of the magnitude of the effect sizes and P values from the Trial Bank against those from Cochrane showed that there were no systematic differences (median differences were 0 for both, IQR: 0 to 0); See Additional file 1: Fig. S5.

Trial Bank approach for medication harms: reasons for trial omissions

In order to explore the underlying reasons why 202 trials were not captured in the Trial Bank, we conducted a post hoc reverse search of these trials in the aforementioned five databases. Our results showed that 155 (76.7%, 155/202) were indexed in these databases, and only 47 (23.3%, 47/202) were not. This means that the loss of the 202 trials was mainly due to errors during the screening process. A further examination of the falsely excluded 155 trials showed that 16 (10.3%, 16/155) were excluded by RobotSearch, and 139 (89.7%, 139/155) by humans.

Discussion

We have established the first Trial Bank of medication safety for children, and examined the use of this resource for evidence synthesis through a proof-of-concept study. Our AI assisted Trial Bank framework was able to identify/include the majority of trials included in Cochrane reviews for medication safety. Majority of analyses using the Trial Bank agreed with Cochrane reviews in both direction and significance; exclusion of the unpublished trials recorded in Cochrane reviews led to a slight increase in the level of agreement. These findings demonstrate that the Trial Bank paradigm is potentially feasible option to represent a continuously updateable source for rapid evidence synthesis, which represents a novel framework for the next generation of the science of evidence synthesis. Next, we aim to perform the retrieval of unpublished trials, develop a web system to support open access to the database, and implement regular (monthly) updates.

We observed that the coverage decreased as the number of trials included on a given topic increased. This pattern may reflect the fact that those Cochrane reviews with larger numbers of included trials may have conducted a more comprehensive literature search (e.g., across multiple databases), thereby amplifying the likelihood of missing trials through the Trial Bank. Although the reduction in coverage appeared to have minimal impact on agreement in the pooled effect estimates, it nonetheless increases uncertainty in the underlying evidence base. Therefore, results derived from the Trial Bank should be interpreted with caution, and a downgrade of the evidence quality is recommended when such outputs are used to inform decision-making.

In spite of the considerable accuracy, we still noticed that there were some trials from the Cochrane reviews that failed to be included in our Trial Bank. The study losses were attributed to three reasons: limited databases searched (23.3%), human error (68.8%), as well as AI error (7.9%). Although this indicates that expanding the search to more databases and (or) strengthening human efforts could largely improve the accuracy, it is not realistic. Each of these would require a substantial additional workload and more complex processes, and perhaps more errors. The continuous progress of AI technology may be a more feasible solution. One more effective method could be the use of citation searching for included trials as well as related systematic reviews [33]. There are existing tools for capturing the citations of publications automatically, such as the SpiderCite [34]. In our next phase, we will use the latter to address this issue.

There are several other important roles for a Trial Bank for the scientific and medical communities. First, it enables healthcare professionals to easily access pertinent trial results directly from a single source. In addition, researchers will also benefit from the capability to conduct between-trial analyses, facilitate comparisons, generate novel insights, and streamline their research process. Perhaps most importantly, the establishment of a specific Trial Bank is expected to provide a resource for the public to access living evidence for emerging clinical questions, and contribute to rapid evidence synthesis methods.

Some limitations of the current initiative should be highlighted. First, language and database coverage biases exist as this is the first attempt to establish a mega-structured Trial Bank so we were not able, for feasibility reasons, to search all of the numerous current medical databases; instead, the main medical databases were searched for a population particularly at risk, accepting that there may be a small proportion of trials that would not be identified from our Trial Bank; for the same reason (proof-of-concept stage), observational studies—another important source for assessing adverse effects—were not considered in the current study. Second, exclusion of unpublished trials. The current Trial Bank did not contain unpublished trial sources, which was a modest source of trial loss. Third, insufficient accuracy of AI technology, despite the huge current advances, makes establishment of a Trial Bank labor-intensive and therefore error-prone (e.g., misclassification from automated screening); human intelligence is still essential in almost all of the steps, for example, screening for full texts and verifying the data accuracy. With the assistance of AI tools, our group spent one year finishing the literature screen and another year checking extracted data. Even so, this approach is expected to be more effective than the traditional methods. With the rapid revolution of AI technologies, there is sufficient reason to believe that there will be a paradigm shift in the traditional evidence synthesis model in the near future. Having developed and tested our approach with a smaller dataset (i.e., pediatric medication trials), we are well-positioned to develop a larger dataset for other populations and interventions. Fourth, our validating study did not measure the agreements in terms of clinical significance, which could be considered a better method than statistical significance [35]. While considering the subjective nature of the definition of clinical significance, we believe the use of clinical significance does not bring gains for the rigor of the methodology of the current study. Moreover, the validation largely relies on the overlapping of the trials in Trial Bank and Cochrane reviews, while not all Cochrane reviews are of a high standard that ensures accurate estimation of the true effect, and the current validation could only be treated as a type of ‘proof-of-concept’ study. Finally, the trainer was also the validator and this may (unconsciously) confirm that students applied what was taught, rather than independently assess whether the extraction is truly correct and this could have inflated apparent agreement somewhat.

Despite these limitations, there would be some other challenges in the next stages of the PROMISE project. One major challenge would be the standardization of the terms and definitions of the reported adverse outcomes— for an adverse event, different research teams may adopt different criteria for definition, and the terminology used can also vary. Although different terms can be standardized according to the CTCAE (Common Terminology Criteria for Adverse Events) standards. How this could be handled has not yet been clarified as well as one of the main sources of heterogeneity in evidence synthesis. Another major challenge would be the identification of potentially problematic trials that have not been retracted, especially those involving research misconduct. These trials constitute a primary source of evidence contamination when the evidence from these trials is incorporated into evidence syntheses (e.g., meta-analyses, clinical practice guidelines) [29]. All these challenges need to be addressed in the near future.

Conclusions

In conclusion, the Trial Bank paradigm is a potentially feasible approach for living rapid evidence syntheses to promote the development of evidence-based medicine. Even though the establishment of the Trial Bank is time- and labor-intensive as well as costly, it comes with considerable benefit as highlighted previously. Some practical problems (e.g., information collection) during the establishment of the Trial Bank also bring huge challenges. We believe our first attempt to develop a Trial Bank will provide valuable experience and a solid foundation for future initiatives.

Supplementary Information

12916_2026_4709_MOESM1_ESM.docx^{(1.3MB, docx)}

Additional file 1: Fig. S1. Number of trials from 1956 to 2022. Fig. S2. Geographical distribution of trials. Multiple: multi-center trials; Single: single-center trials. Fig. S3. Distribution of conditions studied in 100 or more trials. Fig. S4. Flow chart of literature screening for Cochrane reviews of drug intervention for children. Fig. S5. Absolute differences of the magnitude of the effect sizes and P values for emulated meta-analyses between Trial bank and Cochrane reviews.

Acknowledgements

None.

PROMISE Collaborative Research Network: Bichun Huang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Changchang Chen, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Changchun He, Undergraduate Student (School of Health Management, Inner Mongolia Medical University, Hohhot, China); Chaoneng He, Postgraduate Student (Department of Pharmacy, Zhejiang Cancer Hospital, Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences, Hangzhou, China); Chenkai Xu, Undergraduate Student (School of Mechanical and Electrical Engineering, Luoyang Vocational＆Technical College, Luoyang, China); Chenxia Hao, Research Assistant (Shanghai children's medical center, Shanghai Jiao Tong University School of Medicine, Shanghai, China); Chunhui Li, Undergraduate Student (Xinglin School, Nantong University, Nantong, China); Chunxiao Wang, Associate Chief Pharmacist (Pharmacy Department, Shanghai Children's Medical Center Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China); Chunyan Mo, Postgraduate Student (School of medicine, Kunming University of Science and Technology, Kunming, China); Chunyu Zhang, Undergraduate Student (School of Health Management, Inner Mongolia Medical University, Nei Mongol, China); Dandan Xu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Danni Xia, Postgraduate Student (School of Public Health, Lanzhou University, Lanzhou, China); Fali Zhang, Chief Pharmacist (Department of Pharmacy, Huashan Hospital, Fudan University, Shanghai); Fan Gao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Fang Xie, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Fangfang Zuo, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Fangyi Qian, Undergraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Feifei Han,Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Guifang Jin, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Haian Diao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Hailan Wu,Postgraduate Student (School of Health Management, Anhui Medical University, Hefei, China); Haixing Wang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Hang Xin, Undergraduate Student (First School of Clinical Medicine, Anhui Medical University, Hefei, China); Haocheng Wang,Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Haoning Zhu, Postgraduate Student (West China School of Pharmacy, Sichuan University, Chengdu, China); Hong Cao, Research Assistant (Shanghai Eastern Hepatobiliary Surgery Hospital, Navy Medical University, Shanghai, China); Hong Liu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Hongtao Yang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Hongting Chen, Pharmacist-in-charge (Department of Pharmacy, Shanghai Children’s Medical Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China); Hongxin Shu,Postgraduate Student (The Second Clinical Medical College, Nanchang University, Nanchang, China); Hui Zou, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Huoran Zhou, Undergraduate Student (First School of Clinical Medicine, Anhui Medical University, Hefei, China); Jiajie Huang, Postgraduate (School of Nursing, Gansu University of Chinese Medicine, Lanzhou, China); Jiali Wu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jianzhou Liu,Postgraduate Student (School of Management, Shanxi Medical University, Jinzhong, China); Jiaojiao Kou, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jiaojiao Zhao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jiazhen Wang, Postgraduate Student (Anhui Engineering University, Hefei, China); Jie Wang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jihong Hu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jin Wang, Undergraduate Student (First School of Clinical Medicine, Anhui Medical University, Hefei, China); Jincheng Zhang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jing Ming, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jingxi Feng, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Jingyi Wang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Kai Liu, Postgraduate Student (Wannan Medical College, Wuhu, China); Ke Zhou, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Keer Jiang, Postgraduate Student (School of Public Health, Wenzhou Medical University, Wenzhou, China); Kexin Fu, Undergraduate Student (School of Health Administration, Inner Mongolia Medical University, Hohhot, Inner Mongolia, China); Kexin Gao, Postgraduate Student (Inner Mongolia Medical University, Inner Mongolia, China); Langyu Xiong, Postgraduate Student (School of Nursing, Chengdu University of Traditional Chinese Medicine, Chengdu, China); Lei Qian,Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Li Zhu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Lijie Zhang, Postgraduate Student (Inner Mongolia Medical University, Inner Mongolia, China); Lijun Tang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Limin Li, Master Student (Anhui Medical University, Hefei, China); Ling Wang, Undergraduate Student (Clinical College, Anhui Medical University, Hefei, China); Lizi Hu, Postgraduate Student (Chengdu University of TCM, Chengdu, China); Long Dao, Postgraduate Student, (School of public health, Tianjin Medical University, Tianjin, China); Manyu Zhang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Mengxia Yan, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Mengxia Zhang, Postgraduate Student (School of Basic Medicine and Clinical Pharmacy, China Pharmaceutical University, Nanjing, China); Menya Zhao, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Mianhui Hong, Clinical Pharmacist (Department of Pharmacy, Hainan General Hospital, Hainan, China); Mili Shi, Chief Pharmacist (Department of Pharmacy, Sixth Affiliated Hospital of Kunming Medical University, Yunnan, China); Min Gao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Minhong Li, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Na He, Postgraduate Student (Anhui Medical University, Hefei, China); Ning Wang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Qiang Zhou, Clinical Pharmacist (Department of Clinical Pharmacy, Jinling Hospital, Medical School of Nanjing University, Nanjing, China); Qiannan Wang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Qianqian Ji, Postgraduate Student (Anhui Medical University, Hefei, China); Qingwei Meng, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Ran Zhao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Rui zhang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Ruoxi Wang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Ruxin Zhang, Undergraduate Student (Chizhou University, Chizhou, China); Shanshan Xu, Chief Pharmacist (Department of Pharmacy, Shanghai Deji Hospital, Shanghai, China); Shizhao Shi, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Shizhong Zhou, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Shouzhen Liu, Master Student (School of Pharmaceutical Sciences, Anhui Medical University, Hefei, China); Shuting Xing, Postgraduate Student (Inner Mongolia Medical University, Inner Mongolia, China); Shuya Sun, Postgraduate Student (Anhui Medical University, Hefei, China); Si Gao, research assistant (Department of Pharmacy, The People's Hospital of Guangxi Zhuang Autonomous Region, Nanning, China); Sizhe Li, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Songhong Xie, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Teng Yao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Tongtong Miao, Chief Pharmacist (Department of Pharmacy, Nantong First People's Hospital, Second Affiliated Hospital of Nantong University, Jiangsu, China); Wanying Zhang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Wei Du, research assistant (Department of Pharmacy, The Second People's Hospital of Changzhou, the Third Affiliated Hospital of Nanjing Medical University, Changzhou, China); Weilong Zhao, Postgraduate Student (School of Public Health, Lanzhou University, Lanzhou, China); Weina Li, Undergraduate Student (Medical School, Xi 'an Jiaotong University, Xi 'an, China); Wenjing Qiang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Wenqian Zhu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xi Yang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Xianglin Zhang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Xiangyun Li, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xiaonuo Han, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xiaowei Yang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xiaoya Sun, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xiaoyu Wang, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xiaoyuan Zheng, Associate Chief Pharmacist (Department of Pharmacy, Affiliated Hospital of Jiangnan University, Wuxi, China); Xin He, Postgraduate Student (School of Public Health, Lanzhou university, Lanzhou, China); Xin Qiao, Undergraduate Student (School of Health Administration, Inner Mongolia Medical University, Hohhot, Inner Mongolia, China); Xinke Ding, Postgraduate Student (School of Management, Shanxi Medical University, Taiyuan, China); Xinke Gao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xinxin Wu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xinyan Bu, Undergraduate Student (College of Science, China Pharmaceutical University, Nanjing, China); Xinyu Gao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Xinyu Zhang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); XinYue Wang, Undergraduate Student (School of Public Health, An Hui Medical University, Hefei, China); Yafang Hao, Undergraduate Student (School of Health Management, Inner Mongolia Medical University, Hohhot, China); Yafang Zheng, Undergraduate Student (Department of Clinical Medicine, The Third School of Clinical Medicine, Guangzhou Medical University, Guangzhou, China); Yanan Zheng, Undergraduate Student (Department of Clinical Medicine, The Third School of Clinical Medicine, Guangzhou Medical University, Guangzhou, China); Yanyan Dong, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yanyan Pan, Master Student (School of Pharmaceutical Sciences, Anhui Medical University, Hefei, China); Yi Zhu, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Yifei Wang, Undergraduate Student (School of Pharmacy, Tianjin University of Traditional Chinese Medicine, Tianjin, China); Ying Zhao, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yingjia Xu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yingxia Wang, Master Student (School of Pharmaceutical Sciences, Anhui Medical University, Hefei, China); Yingying Zuo, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Yiwei Peng, Postgraduate Student (School of Pharmaceutical Sciences, Anhiu Medical University, Hefei, China); Yong Hong, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yongjin Zhai, Postgraduate Student (School of Management, Shanxi Medical University, Taiyuan, China); Yuan Tian, First School of Clinical Medicine, Anhui Medical University, Hefei, China; Yuanyuan Wang, Junior college Student (School of Textile and Clothing, Quanzhou Textile and Clothing Vocational College, Fujian, China); Yueyue Wang, Undergraduate Student (School of Mechanical and Electrical Engineering, Hefei City University, Hefei, China); Yueyue Zhang, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Yuhua He, research assistant (Department of Pharmacy, The Second affiliated Hospital, Zhe Jiang University School of Medicine, Zhe Jiang, China); Yujuan Zhai, Undergraduate Student (School of Mechanical and Electrical Engineering, Hefei City University, Hefei, China); Yukai Xu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yun Yin, Research assistant (Department of Pharmacy, Tongji Hospital, School of Medicine, Tongji University, Shanghai, China); Yunyan Chen, Associate Chief Pharmacist (Department of Pharmacy, Xiamen Hospital of Traditional Chinese Medicine, The office of Drug Clinical Trial Institution, Xiamen, China); Yuqing Yang, Research Assistant (Shanghai Eastern Hepatobiliary Surgery Hospital, Navy Medical University, Shanghai, China); Yutong Cheng, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Yuxuan Guo, Postgraduate Student (School of Public Health, Xi'an Jiaotong University Health Science Center, Xi'an, China); Zhengyang Pan, Undergraduate Student (First School of Clinical Medicine, Anhui Medical University, Hefei, China); Zhenrong Ran, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Zhenshan Gao, Postgraduate Student (School of Public Health, Anhui Medical University, Hefei, China); Zhicheng Guo, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Zhihui Jia, Undergraduates Student (School of Digital Intelligence Industry, Inner Mongolia University of Science and Technology, Baotou, China); Zhiyun Hu, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Zhuangzhuang Tang, Junior pharmacist (Department of Pharmacy, Nantong First People's Hospital, Second Affiliated Hospital of Nantong University, Jiangsu, China); Zhuo Deng, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Zhuzhu Qin, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China); Zihan Shen, Postgraduate Student (School of Pharmacy, Anhui Medical University, Hefei, China).

Abbreviations

PROMISE: Promote Trustworthy Living Evidence for Medication Safety
AI: Artificial intelligence
TBC: Trial Bank coverage
CI: Confidence interval
IQR: Interquartile range
CTCAE: Common Terminology Criteria for Adverse Events

Authors’ contributions

CX conceived and designed the study and supported the project; SQ.F, YX.W, LJ.T, YT, ZP contributes the project management and involved in some of the data extraction; CX drafted the manuscript; DL, ZC.G and HW.L contributes the project management and provides funding support; SD and YT analyzed the data; HW.L, ZC.G, LFK, LF.L, JC, HT.C, TQ.Y, SG, YL, SD, LD, SV, HM, SY.L, YL.C, EM, PG provided advice on the study design, methodological guidance, and revision of the draft manuscript. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. All authors read and approved the final manuscript.

Funding

This study was mainly (80%) supported by the principal investigator (Chang Xu); The National Natural Science Foundation of China (72204003, 72574229) provide supports for some labour costs (10%). This study also received institutional funding from National Key Research and Development Program of China (2023YFC3605005), the Academic leader training program of Pudong New Area Health Commission (PWRd2023-02), the Talent Project established by Chinese Pharmaceutical Association Hospital Phamacy department (CPA-Z05-ZC-2023–003). Luis Furuya-Kanamori was supported by the University of Queensland’s Amplify Initiative. Suhail Doi was supported by Program Grant #NPRP-BSRA01-0406–210030 from the Qatar National Research Fund. The funding bodies had no role in any process of the study (i.e., study design, analysis, interpretation of data, writing of the report, and decision to submit the article for publication). The findings herein reflect the work, and are solely the responsibility of the authors.

Data availability

The data used in current study can be obtained via [https://osf.io/up7ag/overview].

Declarations

Ethics approval and consent to participate

Not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Chang Xu, Email: xuchang2016@runbox.com.

Liang Du, Email: duliang0606@vip.sina.com.

Houwen Lin, Email: linhouwenrenji@163.com.

Zhichun Gu, Email: guzhichun213@163.com.

PROMISE Collaborative Research Network:

Yingxia Wang, Lijun Tang, Yuan Tian, Bichun Huang, Changchang Chen, Changchun He, Chaoneng He, Chenkai Xu, Chenxia Hao, Chunhui Li, Chunxiao Wang, Chunyan Mo, Chunyu Zhang, Dandan Xu, Danni Xia, Fali Zhang, Fan Gao, Fang Xie, Fangfang Zuo, Fangyi Qian, Feifei Han, Guifang Jin, Haian Diao, Hailan Wu, Haixing Wang, Hang Xin, Haocheng Wang, Haoning Zhu, Hong Cao, Hong Liu, Hongtao Yang, Hongting Chen, Hongxin Shu, Hui Zou, Huoran Zhou, Jiajie Huang, Jiali Wu, Jianzhou Liu, Jiaojiao Kou, Jiaojiao Zhao, Jiazhen Wang, Jie Wang, Jihong Hu, Jin Wang, Jincheng Zhang, Jing Ming, Jingxi Feng, Jingyi Wang, Kai Liu, Ke Zhou, Keer Jiang, Kexin Fu, Kexin Gao, Langyu Xiong, Lei Qian, Li Zhu, Lijie Zhang, Limin Li, Ling Wang, Lizi Hu, Long Dao, Manyu Zhang, Mengxia Yan, Mengxia Zhang, Menya Zhao, Mianhui Hong, Mili Shi, Min Gao, Minhong Li, Na He, Ning Wang, Qiang Zhou, Qiannan Wang, Qianqian Ji, Ran Zhao, Rui Zhang, Ruoxi Wang, Ruxin Zhang, Shanshan Xu, Shizhao Shi, Shizhong Zhou, Shouzhen Liu, Shuting Xing, Shuya Sun, Si Gao, Sizhe Li, Songhong Xie, Teng Yao, Tongtong Miao, Wanying Zhang, Wei Du, Weilong Zhao, Weina Li, Wenjing Qiang, Wenqian Zhu, Xi Yang, Xianglin Zhang, Xiangyun Li, Xiaonuo Han, Xiaowei Yang, Xiaoya Sun, Xiaoyu Wang, Xiaoyuan Zheng, Xin He, Xin Qiao, Xinke Ding, Xinke Gao, Xinxin Wu, Xinyan Bu, Xinyu Gao, Xinyu Zhang, XinYue Wang, Yafang Hao, Yafang Zheng, Yanan Zheng, Yanyan Dong, Yanyan Pan, Yi Zhu, Yifei Wang, Ying Zhao, Yingjia Xu, Yingying Zuo, Yiwei Peng, Yong Hong, Yongjin Zhai, Yuanyuan Wang, Yueyue Wang, Yueyue Zhang, Yuhua He, Yujuan Zhai, Yukai Xu, Yun Yin, Yunyan Chen, Yuqing Yang, Yutong Cheng, Yuxuan Guo, Zhengyang Pan, Zhenrong Ran, Zhenshan Gao, Zhicheng Guo, Zhihui Jia, Zhiyun Hu, Zhuangzhuang Tang, Zhuo Deng, Zhuzhu Qin, and Zihan Shen

References

1.Zorzela L, Loke YK, Ioannidis JP, et al. PRISMA harms checklist: improving harms reporting in systematic reviews. BMJ. 2016;352:i157. [DOI] [PubMed] [Google Scholar]
2.Junqueira DR, Zorzela L, Golder S, et al. CONSORT harms 2022 statement, explanation, and elaboration: updated guideline for the reporting of harms in randomised trials. BMJ. 2023;381:e073725. [DOI] [PubMed] [Google Scholar]
3.Khan MA, Soteriades ES, King J, et al. Global trends and forecast of the burden of adverse effects of medical treatment: epidemiological analysis based on the global burden of disease study. Cureus. 2020;12(3):e7250. [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Koyama T, Iinuma S, Yamamoto M, et al. International Trends in Adverse Drug Event-Related Mortality from 2001 to 2019: An Analysis of the World Health Organization Mortality Database from 54 Countries. Drug Saf. 2024;47(3):237–49. [DOI] [PubMed] [Google Scholar]
5.Panagioti M, Khan K, Keers RN, et al. Prevalence, severity, and nature of preventable patient harm across medical care settings: systematic review and meta-analysis. BMJ. 2019;366:l4185. [DOI] [PMC free article] [PubMed] [Google Scholar]
6.World Health Organization (WHO). Global Patient Safety Report 2024. Geneva (2024). Available at: https://iris.who.int/bitstream/handle/10665/380067/B09204-eng.pdf? sequence =1.
7.Kearns GL, Abdel-Rahman SM, Alander SW, Blowey DL, Leeder JS, Kauffman RE. Developmental pharmacology--drug disposition, action, and therapy in infants and children. N Engl J Med. 2003;349(12):1157–67. [DOI] [PubMed] [Google Scholar]
8.World Health Organization (WHO). Global patient safety action plan 2021–2030: towards eliminating avoidable harm in health care. Geneva (2021). Available at: https://iris.who.int/bitstream/handle/10665/343477/9789240032705-eng.pdf?sequence=1.
9.Egger M, Smith GD, Phillips AN. Meta-analysis: principles and procedures. BMJ. 1997;315(7121):1533–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Naidoo N, Nguyen VT, Ravaud P, Young B, Amiel P, Schanté D, et al. The research burden of randomized controlled trial participation: a systematic thematic synthesis of qualitative evidence. BMC Med. 2020;18(1):6. [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Bhaumik DK, Amatya A, Normand SL, et al. Meta-Analysis of Rare Binary Adverse Event Data. J Am Stat Assoc. 2012;107(498):555–67. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Noyes J, Booth A, Flemming K, et al. Cochrane Qualitative and Implementation Methods Group guidance series-paper 3: methods for assessing methodological limitations, data extraction and synthesis, and confidence in synthesized qualitative findings. J Clin Epidemiol. 2018;97:49–58. [DOI] [PubMed] [Google Scholar]
13.Li L, Bai O, Wang X. An integrative shrinkage estimator for random-effects meta-analysis of rare binary events. Contemp Clin Trials Commun. 2018;10:141–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;339:b2535. [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Elliott JH, Synnot A, Turner T, et al. Living systematic review: 1. Introduction-the why, what, when, and how. J Clin Epidemiol. 2017;91:23–30. [DOI] [PubMed] [Google Scholar]
16.Créquit P, Trinquart L, Yavchitz A, et al. Wasted research when systematic reviews fail to provide a complete and up-to-date evidence synthesis: the example of lung cancer. BMC Med. 2016;14:8. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Bastian H, Glasziou P, Chalmers I. Seventy-five trials and eleven systematic reviews a day: how will we ever keep up? PLoS Med. 2010;7(9):e1000326. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Shojania KG, Sampson M, Ansari MT, et al. How quickly do systematic reviews go out of date? A survival analysis. Ann Intern Med. 2007;147(4):224–33. [DOI] [PubMed] [Google Scholar]
19.Garner P, Hopewell S, Chandler J, et al. When and how to update systematic reviews: consensus and checklist. BMJ. 2016;354:i3507. [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Sim I, Rennels G. A trial bank model for the publication of clinical trials. Proc Annu Symp Comput Appl Med Care. 1995; 863–7. [PMC free article] [PubMed]
21.Sim I, Detmer DE. Beyond trial registration: a global trial bank for clinical trial reporting. PLoS Med. 2005;2(11):e365. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Zarin DA, Tse T, Williams RJ, et al. The ClinicalTrials.gov results database--update and key issues. N Engl J Med. 2011; 364(9):852-60. [DOI] [PMC free article] [PubMed]
23.Peng Z, Wu X, Qin Z, et al. Accuracy of large language models in data extraction from randomized controlled trials in sleep medicine: A proof-of-concept study. Sleep Med Rev. 2025;84:102192. [DOI] [PubMed] [Google Scholar]
24.Gartlehner G, Kahwati L, Hilscher R, et al. Data extraction for evidence synthesis using a large language model: a proof-of-concept study. Res Synth Methods. 2024;15(4):576–89. [DOI] [PubMed] [Google Scholar]
25.Khraisha Q, Put S, Kappenberg J, et al. Can large language models replace humans in systematic reviews? Evaluating GPT-4’s efficacy in screening and extracting data from peer-reviewed and grey literature in multiple languages. Res Synth Methods. 2024;15(4):616–26. [DOI] [PubMed] [Google Scholar]
26.Groff ML, Offringa M, Emdin A, et al. Publication trends of pediatric and adult randomized controlled trials in general medical journals, 2005-2018: a citation analysis. Children. 2020;7(12):293. [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Zorzela L, Loke YK, Ioannidis JP, Golder S, Santaguida P, Altman DG, et al. PRISMA harms checklist: improving harms reporting in systematic reviews. BMJ. 2016;352:i157. [DOI] [PubMed] [Google Scholar]
28.United Nations International Children's Emergency Fund. The Convention on the Rights of the Child: The children’s version (1989). Available: https://www.unicef.org/child-rights-convention/convention-text-childrens-version.
29.Xu C, Fan S, Tian Y, et al. Investigating the impact of trial retractions on the healthcare evidence ecosystem (VITALITY Study I): retrospective cohort study. BMJ. 2025;389:e082068. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Marshall IJ, Noel-Storr A, Kuiper J, et al. Machine learning for identifying Randomized Controlled Trials: An evaluation and practitioner’s guide. Res Synth Methods. 2018;9(4):602–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Moseley AM, Elkins MR, Herbert RD, Maher CG, Sherrington C. Cochrane reviews used more rigorous methods than non-Cochrane reviews: survey of systematic reviews in physiotherapy. J Clin Epidemiol. 2009;62(10):1021–30. [DOI] [PubMed] [Google Scholar]
32.Xu C. PROMISE Project. OSF https://osf.io/up7ag/overview (2025).
33.Wright K, Golder S, Rodriguez-Lopez R. Citation searching: a systematic review case study of multiple risk behaviour interventions. BMC Med Res Methodol. 2014;14:73. [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Clark J, Glasziou P, Del Mar C, et al. A full systematic review was completed in 2 weeks using automation tools: a case study. J Clin Epidemiol. 2020;121:81–90. [DOI] [PubMed] [Google Scholar]
35.Hultcrantz M, Schünemann HJ, Mustafa RA, et al. GRADE Certainty Ratings: Thresholds Rather Than Categories of Contextualization (GRADE Guidance 41). Ann Intern Med. 2025;178(8):1183–6. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

12916_2026_4709_MOESM1_ESM.docx^{(1.3MB, docx)}

Data Availability Statement

The data used in current study can be obtained via [https://osf.io/up7ag/overview].

[CR1] 1.Zorzela L, Loke YK, Ioannidis JP, et al. PRISMA harms checklist: improving harms reporting in systematic reviews. BMJ. 2016;352:i157. [DOI] [PubMed] [Google Scholar]

[CR2] 2.Junqueira DR, Zorzela L, Golder S, et al. CONSORT harms 2022 statement, explanation, and elaboration: updated guideline for the reporting of harms in randomised trials. BMJ. 2023;381:e073725. [DOI] [PubMed] [Google Scholar]

[CR3] 3.Khan MA, Soteriades ES, King J, et al. Global trends and forecast of the burden of adverse effects of medical treatment: epidemiological analysis based on the global burden of disease study. Cureus. 2020;12(3):e7250. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR4] 4.Koyama T, Iinuma S, Yamamoto M, et al. International Trends in Adverse Drug Event-Related Mortality from 2001 to 2019: An Analysis of the World Health Organization Mortality Database from 54 Countries. Drug Saf. 2024;47(3):237–49. [DOI] [PubMed] [Google Scholar]

[CR5] 5.Panagioti M, Khan K, Keers RN, et al. Prevalence, severity, and nature of preventable patient harm across medical care settings: systematic review and meta-analysis. BMJ. 2019;366:l4185. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR6] 6.World Health Organization (WHO). Global Patient Safety Report 2024. Geneva (2024). Available at: https://iris.who.int/bitstream/handle/10665/380067/B09204-eng.pdf? sequence =1.

[CR7] 7.Kearns GL, Abdel-Rahman SM, Alander SW, Blowey DL, Leeder JS, Kauffman RE. Developmental pharmacology--drug disposition, action, and therapy in infants and children. N Engl J Med. 2003;349(12):1157–67. [DOI] [PubMed] [Google Scholar]

[CR8] 8.World Health Organization (WHO). Global patient safety action plan 2021–2030: towards eliminating avoidable harm in health care. Geneva (2021). Available at: https://iris.who.int/bitstream/handle/10665/343477/9789240032705-eng.pdf?sequence=1.

[CR9] 9.Egger M, Smith GD, Phillips AN. Meta-analysis: principles and procedures. BMJ. 1997;315(7121):1533–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Naidoo N, Nguyen VT, Ravaud P, Young B, Amiel P, Schanté D, et al. The research burden of randomized controlled trial participation: a systematic thematic synthesis of qualitative evidence. BMC Med. 2020;18(1):6. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Bhaumik DK, Amatya A, Normand SL, et al. Meta-Analysis of Rare Binary Adverse Event Data. J Am Stat Assoc. 2012;107(498):555–67. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Noyes J, Booth A, Flemming K, et al. Cochrane Qualitative and Implementation Methods Group guidance series-paper 3: methods for assessing methodological limitations, data extraction and synthesis, and confidence in synthesized qualitative findings. J Clin Epidemiol. 2018;97:49–58. [DOI] [PubMed] [Google Scholar]

[CR13] 13.Li L, Bai O, Wang X. An integrative shrinkage estimator for random-effects meta-analysis of rare binary events. Contemp Clin Trials Commun. 2018;10:141–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR14] 14.Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;339:b2535. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Elliott JH, Synnot A, Turner T, et al. Living systematic review: 1. Introduction-the why, what, when, and how. J Clin Epidemiol. 2017;91:23–30. [DOI] [PubMed] [Google Scholar]

[CR16] 16.Créquit P, Trinquart L, Yavchitz A, et al. Wasted research when systematic reviews fail to provide a complete and up-to-date evidence synthesis: the example of lung cancer. BMC Med. 2016;14:8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Bastian H, Glasziou P, Chalmers I. Seventy-five trials and eleven systematic reviews a day: how will we ever keep up? PLoS Med. 2010;7(9):e1000326. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Shojania KG, Sampson M, Ansari MT, et al. How quickly do systematic reviews go out of date? A survival analysis. Ann Intern Med. 2007;147(4):224–33. [DOI] [PubMed] [Google Scholar]

[CR19] 19.Garner P, Hopewell S, Chandler J, et al. When and how to update systematic reviews: consensus and checklist. BMJ. 2016;354:i3507. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Sim I, Rennels G. A trial bank model for the publication of clinical trials. Proc Annu Symp Comput Appl Med Care. 1995; 863–7. [PMC free article] [PubMed]

[CR21] 21.Sim I, Detmer DE. Beyond trial registration: a global trial bank for clinical trial reporting. PLoS Med. 2005;2(11):e365. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Zarin DA, Tse T, Williams RJ, et al. The ClinicalTrials.gov results database--update and key issues. N Engl J Med. 2011; 364(9):852-60. [DOI] [PMC free article] [PubMed]

[CR23] 23.Peng Z, Wu X, Qin Z, et al. Accuracy of large language models in data extraction from randomized controlled trials in sleep medicine: A proof-of-concept study. Sleep Med Rev. 2025;84:102192. [DOI] [PubMed] [Google Scholar]

[CR24] 24.Gartlehner G, Kahwati L, Hilscher R, et al. Data extraction for evidence synthesis using a large language model: a proof-of-concept study. Res Synth Methods. 2024;15(4):576–89. [DOI] [PubMed] [Google Scholar]

[CR25] 25.Khraisha Q, Put S, Kappenberg J, et al. Can large language models replace humans in systematic reviews? Evaluating GPT-4’s efficacy in screening and extracting data from peer-reviewed and grey literature in multiple languages. Res Synth Methods. 2024;15(4):616–26. [DOI] [PubMed] [Google Scholar]

[CR26] 26.Groff ML, Offringa M, Emdin A, et al. Publication trends of pediatric and adult randomized controlled trials in general medical journals, 2005-2018: a citation analysis. Children. 2020;7(12):293. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Zorzela L, Loke YK, Ioannidis JP, Golder S, Santaguida P, Altman DG, et al. PRISMA harms checklist: improving harms reporting in systematic reviews. BMJ. 2016;352:i157. [DOI] [PubMed] [Google Scholar]

[CR28] 28.United Nations International Children's Emergency Fund. The Convention on the Rights of the Child: The children’s version (1989). Available: https://www.unicef.org/child-rights-convention/convention-text-childrens-version.

[CR29] 29.Xu C, Fan S, Tian Y, et al. Investigating the impact of trial retractions on the healthcare evidence ecosystem (VITALITY Study I): retrospective cohort study. BMJ. 2025;389:e082068. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Marshall IJ, Noel-Storr A, Kuiper J, et al. Machine learning for identifying Randomized Controlled Trials: An evaluation and practitioner’s guide. Res Synth Methods. 2018;9(4):602–14. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR31] 31.Moseley AM, Elkins MR, Herbert RD, Maher CG, Sherrington C. Cochrane reviews used more rigorous methods than non-Cochrane reviews: survey of systematic reviews in physiotherapy. J Clin Epidemiol. 2009;62(10):1021–30. [DOI] [PubMed] [Google Scholar]

[CR32] 32.Xu C. PROMISE Project. OSF https://osf.io/up7ag/overview (2025).

[CR33] 33.Wright K, Golder S, Rodriguez-Lopez R. Citation searching: a systematic review case study of multiple risk behaviour interventions. BMC Med Res Methodol. 2014;14:73. [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Clark J, Glasziou P, Del Mar C, et al. A full systematic review was completed in 2 weeks using automation tools: a case study. J Clin Epidemiol. 2020;121:81–90. [DOI] [PubMed] [Google Scholar]

[CR35] 35.Hultcrantz M, Schünemann HJ, Mustafa RA, et al. GRADE Certainty Ratings: Thresholds Rather Than Categories of Contextualization (GRADE Guidance 41). Ann Intern Med. 2025;178(8):1183–6. [DOI] [PubMed] [Google Scholar]

PERMALINK

The Trial Bank initiative to promote trustworthy living evidence for medication safety (PROMISE) of children: design and pilot validation

Chang Xu

Suhail A Doi

Sunita Vohra

Luis Furuya-Kanamori

Yingxia Wang

Lijun Tang

Yuan Tian

Shiqi Fan

Zhen Peng

Evan Mayo-Wilson

Lifeng Lin

Haitao Chu

M Hassan Murad

Tianqi Yu

Sheyu Li

Yaolong Chen

Su Golder

Yoon Loke

Justin Clark

Paul Glasziou

Liang Du

Houwen Lin

Zhichun Gu

Abstract

Background

Methods

Results

Conclusions

Supplementary Information

Background

Methods

Design and settings

Fig. 1.

Establishment of the Trial Bank

Validation of the Trial Bank approach

Statistical analysis

Results

Fig. 2.

Trial characteristics

Table 1.

Trial Bank approach for medication harms: Trial Bank coverage (TBC)

Fig. 3.

Trial Bank approach for medication harms: agreement

Fig. 4.

Trial Bank approach for medication harms: reasons for trial omissions

Discussion

Conclusions

Supplementary Information

Acknowledgements

Abbreviations

Authors’ contributions

Funding

Data availability

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Footnotes

Contributor Information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases