zombie trials

Summary: 'Zombie trials'—problematic or fabricated studies that remain in the scientific record—are estimated to be prevalent in clinical research, with some estimates suggesting up to 44% of trials in specific fields may contain false data. These untrustworthy trials significantly distort the medical literature by exaggerating treatment effects, often by over 50%, and lead to biased clinical guidelines that are rarely corrected even after the underlying trials are retracted.

Prevalence of Zombie Trials

The term 'zombie trials' refers to published or unpublished randomized controlled trials (RCTs) with serious questions regarding the trustworthiness of their data or findings, regardless of whether they have been formally retracted (Direct, High; PMID: 37372725).

• General Estimates: Some experts suggest that hundreds of thousands of zombie randomized trials are currently circulating in the medical literature (Direct, High; PMID: 37372725).
• Specific Cohort Data: An analysis of individual participant data from 153 RCTs submitted to the journal Anaesthesia between 2017 and 2020 revealed that 44% contained false data and 26% featured fabricated data (Direct, High; PMID: 39628711).
• Self-Reported Misconduct: A meta-analysis of survey data found that, on average, 1.97% of scientists admitted to fabricating, falsifying, or modifying data at least once, while 14.12% reported personal knowledge of a colleague committing such misconduct (Derived, Medium; PMID: 19478950).
• Author-Specific Prevalence: In an investigation of 35 RCTs from a single author group, approximately 86% of the studies reported baseline characteristics that were mathematically unlikely to result from proper randomization (Direct, High; PMID: 39628711).
• Field-Specific Concerns: Formal trustworthiness assessments of Cochrane reviews for prenatal nutritional interventions led to the removal of 25% of included RCTs due to integrity concerns (Direct, High; PMID: 39628711).

Impact on Evidence Synthesis and Guidelines

Untrustworthy RCTs corrupt the findings of systematic reviews and meta-analyses, which are the highest levels of evidence informing clinical practice and health policy (Direct, High; PMID: 37372725, PMID: 36054583).

• Exaggeration of Treatment Effects: In a study of 32 unique systematic reviews on spinal pain, the removal of untrustworthy "index trials" reduced treatment effect sizes by a median of 58% (IQR 40–74) (Derived, Medium; PMID: 37453533).
• Alteration of Conclusions: An investigation into 32 evidence syntheses that included retracted RCTs found that the conclusions of 13 (40%) would have changed if the fraudulent data had been excluded (Direct, High; PMID: 37372725, PMID: 31666272).
• Impact on COVID-19 Research: Several systematic reviews initially concluded that ivermectin reduced COVID-19 mortality; however, after the removal of fabricated studies characterized by repeating blocks of data, the observed significant benefits could not be sustained (Direct, High; PMID: 37372725, PMID: 37873409, PMID: 36054583).
• Influence on Clinical Practice Guidelines (CPGs): In a cohort of spinal pain research, 9 out of 10 CPGs made positive recommendations for interventions based on evidence from untrustworthy trials (Derived, Medium; PMID: 37453533).
• Distortion of Economic Analyses: The inclusion of two untrustworthy trials in a NICE 2016 guideline drove a cost-effectiveness analysis that favored a specific rehabilitation program, a recommendation that might not have been appropriate without the compromised data (Derived, Medium; PMID: 37453533).

Challenges in Correction and Integrity Oversight

Current scientific and publishing systems are often slow or ineffective at identifying and removing untrustworthy research (Derived, Medium; PMID: 37453533).

• Low Correction Rates: Only 5% of systematic reviews and 11% of clinical practice guidelines subsequently corrected their findings after the RCTs they cited were retracted (Direct, High; PMID: 37372725; DOI: 10.1101/2022.01.30.22270124).
• Continued Citation: A meta-epidemiological study found that 90% of systematic reviews published after an RCT had been retracted cited the study without caution or mention of the retraction (Derived, Medium; DOI: 10.1101/2022.01.30.22270124).
• Limitations of Quality Assessment: Traditional tools like the Cochrane Risk of Bias (RoB) tool are not designed to detect fraud; many fabricated trials describe sound methodologies and receive low risk of bias ratings while containing false data (Direct, High; PMID: 37453533, PMID: 37873409, PMID: 36054583).
• Integrity Screening Gaps: In a large international guideline for Polycystic Ovary Syndrome (PCOS), 45% of originally identified studies were excluded due to integrity concerns only after the implementation of a specific new framework (RIGID) (Direct, High; PMID: 39628711).

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID:36054583 — The term 'zombie trials' refers to published or unpublished randomized controlled trials (RCTs) with serious questions r...
  Failed: conclusion — The paper defines 'problematic studies' using the claim's definition but does not use or define the specific term 'zombie trials'.
• PMID:31666272 — 5 mg daily risedronate was supported in CPGs solely by evidence from a group of trials that were later retracted
  Failed: conclusion — The paper identifies the dose as 2.5 mg, whereas the claim specifies 5 mg.
  Possible alternatives (unverified): PMID:37453533 (36% topic match)

The scientific landscape regarding clinical trial integrity has evolved from an era of assuming data veracity to a critical period of skepticism, prompted by the recognition of "zombie trials"—problematic or fabricated randomized controlled trials (RCTs) that persist in the evidence base (Tier 1, High; PMID: 37372725). This synthesis explores the maturation of integrity oversight, the structural impact of untrustworthy data on evidence syntheses, and the emerging technological frameworks designed to safeguard patient care.

1. Phases of Evidence Evolution

The corpus reveals three distinct phases in the scientific response to research misconduct.

• Early Phase: Defining Misconduct and Initial Prevalence (Median Year: 2009)
  Early efforts focused on establishing the "unholy trinity" of misconduct: fabrication, falsification, and plagiarism (FFP) (Tier 1, High; PMID: 37372725). Landmark investigations highlighted the "tip of the iceberg" phenomenon, where discovered frauds represented only a fraction of actual occurrences. A meta-analysis of surveys during this period estimated that 2% of scientists admitted to data fabrication at least once (Tier 1, High; PMID: 19478950). Representative examples include the William Summerlin case and the establishment of the Office of Research Integrity (ORI) (Tier 2, High; PMID: 36923394).

• Stable Phase: Quantifying Impact on Evidence Synthesis (Median Year: 2019)
  This phase marked a transition from identifying individual fraudsters to measuring the systematic distortion of medical literature. Research demonstrated that untrustworthy "index trials" permeate systematic reviews (SRs) and clinical practice guidelines (CPGs), leading to exaggerated effect sizes (Tier 1, High; PMID: 37453533, PMID: 31666272). For instance, investigations into retracted osteoporosis trials showed they dominated fracture-prevention literature for high-risk patients (Tier 1, High; PMID: 31666272).

• Emerging Phase: Integrative Frameworks and Automation (Median Year: 2023)
  The current phase is characterized by the development of proactive screening tools (RIA, TRACT) and formal frameworks (RIGID, INSPECT-SR) to identify problematic trials before they influence outcomes (Tier 1, High; PMID: 36054583, PMID: 39628711, PMID: 38471680). This phase also explores the utility of blockchain for immutable audit trails (Tier 2, Medium; PMID: 30796226) and Large Language Models (LLMs) for automated paper screening (Tier 2, Medium; PMID: 38214966).

2. Network Structure and Relationships

The Research Landscape Analysis identifies a network characterized by specific hubs of misinformation and the emergence of "integrity bridges."

• Hubs and Propagation: High-impact journals often serve as unintentional hubs for the distribution of fraudulent data. Once a fabricated study is published, it acts as a "zombie," continuing to receive citations for decades post-retraction (Tier 1, High; PMID: 37372725). In specific clusters like spinal pain research, a single author group can influence 32 unique systematic reviews and 10 CPGs (Tier 1, High; PMID: 37453533).
• Bridges and Integration: Frameworks like RIGID bridge the gap between research integrity specialists and clinical guideline developers (Tier 1, High; PMID: 39628711). By forming "integrity committees," these frameworks integrate specialized statistical checks (e.g., John Carlisle methods) into standard evidence synthesis pipelines.
• Network Density and Redundancy: The evidence suggests high redundancy in the citation of untrustworthy work. A meta-epidemiological study found that 90% of SRs citing retracted RCTs failed to mention the retraction, reinforcing the "zombie" status of the data (Tier 2, High; DOI: 10.1101/2022.01.30.22270124).

3. Misconduct Mechanisms → Synthesis → Clinical Outcomes

The narrative maps how specific data-related misconduct translates into potentially harmful clinical practice recommendations.

• Fabrication Mechanisms: Specific indicators of fabrication include "excessive similarity" in baseline characteristics (Tier 1, High; DOI:10.1101/2022.01.30.22270124) and repeating blocks of data in participant spreadsheets (Tier 2, High; PMID: 37873409). In one cohort, 86% of trials from a single group had baseline distributions unlikely to occur through random allocation (Tier 1, High; PMID: 39628711).
• Synthesis Distortion: The inclusion of these trials in meta-analyses drastically inflates treatment efficacy. Removal of untrustworthy trials in the spinal pain field reduced effect sizes (Standardized Mean Difference) by a median of 58% (Tier 1, High; PMID: 37453533).
• Clinical/Operational Outcomes: These distortions lead to positive recommendations for interventions that may be futile or risky. In the 2016 NICE guidelines, untrustworthy data drove a cost-effectiveness analysis that recommended a specific multidisciplinary rehabilitation program (Tier 1, High; PMID: 37453533). Similarly, meta-analyses of ivermectin for COVID-19 showed significant mortality benefits that disappeared once fabricated studies were excluded (Tier 2, High; PMID: 37873409).

4. Biases and Reliability

The corpus emphasizes systemic biases that compromise the reliability of the scientific record.

• Correction Bias: There is a profound failure of the scientific community to self-correct. Only 5% of systematic reviews have corrected their results following the retraction of included studies (Tier 1, High; PMID: 37372725).
• Quality Assessment Blindness: A critical reliability gap exists because established Risk of Bias (RoB 2) tools assume data validity (Tier 1, High; PMID: 36054583). Fabricated trials often describe "perfect" methods that receive "low risk" ratings, masking their inauthenticity (Tier 1, High; PMID: 37453533).
• Publication Pressure: Systemic incentives, such as "publish or perish" and the requirement for positive results, are identified as primary drivers of questionable research practices (QRPs) (Tier 2, High; PMID: 35531936, PMID: 36888916).

5. Significance and Translational Impact

This landscape matters urgently because evidence syntheses of RCTs are ranked highest in the hierarchy of valid evidence (Tier 1, High; PMID: 37372725). If this "gold standard" is contaminated, the translation of research into practice becomes dangerous. The implementation of frameworks like RIGID represents a paradigm shift, where 45% of studies initially identified for an international Polycystic Ovary Syndrome guideline were excluded due to integrity concerns only after rigorous screening (Tier 1, High; PMID: 39628711). This highlights the necessity of shifting from a model of "assumed trust" to one of "verified integrity" in clinical research.

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID:36054583 — A meta-epidemiological study found that 90% of SRs citing retracted RCTs failed to mention the retraction, reinforcing t...
  Failed: conclusion — This paper cites Avenell et al. for finding evidence of distortion, but does not contain the specific '90% of SRs' quantitative finding which comes from Kataoka et al. (Paper 1).
• PMID:31462531 — ** Quality Assessment Blindness: A critical reliability gap exists because established Risk of Bias (RoB 2) tools a...*
  Failed: conclusion — This paper (index 4) describes the application of RoB 2 in a meta-analysis, but does not discuss the 'blindness' or the limitation that the tool assumes data validity; it merely uses the tool.

Meta-analyses built on "zombie data"—untrustworthy, fabricated, or fatally flawed randomized controlled trials (RCTs)—suffer from significant systematic distortion, primarily characterized by the massive exaggeration of treatment effects and a failure of the scientific record to correct these errors even after retractions occur (Direct, High; PMID: 37372725, PMID: 37453533) «✓ PMID:37372725» «✓ PMID:37453533».

Systematic Distortion of Effect Sizes

The primary impact of including zombie data in meta-analyses is the artificial inflation of the estimated treatment benefit.

• Magnitude of Exaggeration: In a systematic exploration of 55 meta-analytic comparisons in spinal pain research, the removal of untrustworthy "index trials" reduced treatment effect sizes by a median of 58% (IQR 40–74) (Derived, Medium; PMID: 37453533) «✓ PMID:37453533».
• Case of Ivermectin: Several systematic reviews initially concluded that ivermectin significantly reduced COVID-19 mortality; however, these findings could not be sustained once studies withdrawn due to medical fraud and repeating blocks of data were excluded (Direct, High; PMID: 37372725, PMID: 37873409, PMID: 36054583) «✓ PMID:37372725» «✓ PMID:37873409» «✓ PMID:36054583».
• Outlier Influence: Zombie trials are often extreme outliers, diverging significantly from the wider literature by reporting implausibly large effect sizes while maintaining unremarkable "low risk" profiles on traditional bias assessment tools (Derived, Medium; PMID: 37453533, PMID: 37873409) «✓ PMID:37453533» «✓ PMID:37873409».

Shifting Statistical Significance

The inclusion of untrustworthy data often creates a false appearance of clinical efficacy that vanishes upon sensitivity analysis.

• Loss of Significance: In the spinal pain cohort, 12 out of 40 (30%) statistically significant effects (p < 0.05) became non-significant once the untrustworthy trials were removed (Derived, Medium; PMID: 37453533) «✓ PMID:37453533».
• Fracture Prevention: A meta-analysis of vitamin K for fracture prevention initially reported a clinically significant reduction in hip fractures (OR 0.23; 95% CI 0.12–0.47). A sensitivity analysis excluding three affected trials changed the result to a non-significant effect with wide confidence intervals (OR 0.30; 95% CI 0.05–1.74) (Direct, High; PMID: 31666272).

Persistence and Failure to Self-Correct

Despite the identification of fraudulent or unreliable data, meta-analyses and the guidelines they inform rarely undergo the necessary updates to remove the "zombie" influence.

• Low Correction Rates: A meta-epidemiological study found that only 5% of systematic reviews (6/130) corrected their results after it was discovered they had included RCTs that were later retracted (Derived, Medium; DOI: 10.1101/2022.01.30.22270124).
• Continued Citation: Approximately 90% of systematic reviews published after an underlying RCT had already been retracted cited that trial without caution or acknowledgment of its retracted status (Derived, Medium; DOI: 10.1101/2022.01.30.22270124) «✓ DOI:10.1101/2022.01.30.22270124».
• Lag Times: Investigations into research misconduct in specific fields, such as women's health, have been shown to last a median of over 11 years, during which time the zombie data continues to influence evidence syntheses (Direct, High; PMID: 39628711) «✓ PMID:39628711».

Impact on Clinical Recommendations

Meta-analyses are the "gold standard" for Clinical Practice Guidelines (CPGs); when they are built on zombie data, they drive inappropriate medical decisions.

• NICE Guidelines: In 2016, two untrustworthy trials dominated the evidence base for a de novo cost-effectiveness analysis that led the UK National Institute for Health and Care Excellence (NICE) to recommend a specific multidisciplinary rehabilitation program for low back pain (Derived, Medium; PMID: 37453533) «✓ PMID:37453533».
• Osteoporosis Management: A group of 12 osteoporosis trials from a single Japanese group dominated fracture-prevention literature for high-risk patients for over 15 years, influencing guidelines in the US, Scotland, and Japan before they were retracted for fabrication and plagiarism (Direct, High; PMID: 31666272) «✓ PMID:31666272».

Which statistical methods are most effective at identifying fabricated data in a meta-analysis?

What specific features differentiate a zombie trial from an honestly conducted but low-quality trial?

How do frameworks like RIGID or INSPECT-SR propose to prevent untrustworthy data from reaching clinical guidelines?

Summary: Statistical identification of fabricated data in meta-analyses primarily relies on assessing the mathematical plausibility of baseline characteristics, detecting outlier treatment effects via funnel plots, and employing machine learning algorithms to identify anomalous data patterns. The "Carlisle method" is widely recognized as a core statistical tool for determining whether the distribution of baseline variables is likely to have occurred through genuine randomization.

1. Assessment of Baseline Data (Carlisle Method)

The statistical evaluation of baseline characteristics is a central method for identifying potentially fabricated randomized controlled trials (RCTs) (Direct, High; PMID: 39628711).

• Plausibility of Randomization: This method, often referred to as the Carlisle or Carlisle-Shen method, tests whether the means and standard deviations of baseline variables across treatment groups are mathematically likely under the assumption of random allocation (Direct, High; PMID: 39628711, PMID: 37453533).
• Success Rate: In one evaluation of 35 RCTs from a single author group, this statistical approach determined that 86% of the studies had baseline distributions that were highly improbable in a real-world randomized context (Direct, High; PMID: 39628711).
• Common Flags: Key indicators include "perfectly balanced" groups across many variables or distributions that are "too narrow" or "too wide" to be genuine (Direct, High; PMID: 39628711).

2. Outlier Identification and Effect Size Analysis

Meta-analysts use visual and quantitative tools to identify trials that deviate uncharacteristically from the broader literature.

• Funnel Plot Analysis: Funnel plots are used to inspect small studies that report implausibly large treatment effects, which may indicate data fabrication or publication bias (Direct, High; PMID: 37372725).
• Divergence Screening: Untrustworthy trials often appear as extreme outliers in meta-analyses, reporting effect sizes that are drastically larger than those found in honestly conducted trials (Derived, Medium; PMID: 37453533).
• Integrity-Based Sensitivity Analysis: Once outliers are identified, researchers perform sensitivity analyses to measure the impact of excluding those specific trials. A massive reduction in the pooled effect size (e.g., a median reduction of 58%) often confirms that the meta-analysis conclusion was heavily driven by a few untrustworthy sources (Derived, Medium; PMID: 37453533).

3. Machine Learning and Anomaly Detection

Advanced machine learning algorithms have been developed to automate the detection of anomalous patterns across large clinical datasets.

• Clustering-Based Detection: Algorithms can use clustering techniques where data objects (participant records) are analyzed for their distance from a central centroid. A record is flagged as anomalous if it exceeds established distance thresholds (Direct, High; PMID: 33851576).
• Distance Metrics: The most effective combination for detecting simulated data anomalies includes the Mahalanobis, Manhattan, and Canberra distance metrics, which achieve a sensitivity greater than 85% (Direct, High; PMID: 33851576).
• Central Statistical Monitoring (CSM): CSM uses thousands of statistical tests across multiple variables to calculate a "data inconsistency score" for clinical sites, identifying those with atypical patterns that may signify fabrication (Direct, High; PMID: 33851576).

4. Individual Participant Data (IPD) Analysis

When the underlying datasets are available, several forensic statistical checks can be applied directly to the raw data.

• Repeating Blocks: A hallmark of fabricated spreadsheets discovered in COVID-19 research was the presence of repeating blocks of data, where identical participant data rows were duplicated multiple times (Derived, Medium; PMID: 37873409).
• Anomalous Patterns: Statistical checks can identify "excessive similarity" or "excessive difference" between groups that would not occur naturally, as well as calculation errors where data in figures and tables do not add up (Direct, High; PMID: 36054583).

What are the specific steps of the RIGID framework for incorporating statistical integrity checks into meta-analyses?

How do distance metrics like Mahalanobis and Canberra help in identifying fabricated results in clinical registries?

What role does the inspection of individual participant data (IPD) play in verifying the findings of a suspicious randomized controlled trial?

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID:39628711 — The most frequently cited method for identifying potentially fabricated randomized controlled trials (RCTs) is the stati...
  Failed: conclusion — The paper lists statistical techniques as the most common class of detection method but does not state that the specific 'statistical evaluation of baseline characteristics' is the most frequently cited method.
  Possible alternatives (unverified): PMID:36054583 (37% topic match)
• PMID:39628711 — , age, weight, height) across treatment groups are mathematically likely under the assumption of random allocation
  Failed: conclusion — The paper discusses baseline characteristics being unlikely results of randomization but does not list or confirm the specific entities 'age, weight, height' in this context.
• PMID:37453533 — , age, weight, height) across treatment groups are mathematically likely under the assumption of random allocation
  Failed: entities — The paper mentions baseline variables being unlikely under random allocation but does not explicitly name 'age, weight, height' as the variables used or verified in this context.

The inspection of individual participant data (IPD) serves as a forensic tool to detect fabrication and systematic errors that are often invisible in published summary reports. By reviewing raw datasets rather than aggregate results, investigators can identify internal inconsistencies, such as repeating blocks of data or mathematically improbable distributions of baseline characteristics (Direct, High; PMID: 37873409, PMID: 39628711) «✓ PMID:37873409» «✓ PMID:39628711».

Detection of Fabrication and False Data

• Identification of False Submissions: A comprehensive review of individual participant data from 153 randomized controlled trials (RCTs) submitted to the journal Anaesthesia revealed that 44% contained false data and 26% featured fabricated data (Direct, High; PMID: 39628711) «✓ PMID:39628711».
• ** spreadsheet Forensics:** In investigations of ivermectin for COVID-19, IPD inspection identified participant spreadsheets containing "repeating blocks of data," where identical rows were duplicated, confirming the data were partially or wholly fabricated (Tier 1, High; PMID: 37873409, PMID: 38471680) «✓ PMID:37873409» «✓ PMID:38471680».
• Anomalous Patterns: IPD allows for the detection of "anomalous patterns" in underlying numerical data that signify forgery, fabrication, or falsification (Direct, High; PMID: 37372725) «✓ PMID:37372725».

Verification of Statistical and Randomization Integrity

• Validation of Summary Results: Access to IPD allows regulators to independently run statistical analyses to ensure that the raw data actually support the reported results, preventing the overinflation of treatment effects or p-values (Direct, High; PMID: 30796226) «✓ PMID:30796226».
• Mathematical Plausibility of Randomization: IPD inspection allows researchers to test whether baseline characteristics (e.g., group balance) are mathematically consistent with genuine random allocation; in one cohort, 86% of trials from a single group had baseline distributions unlikely to result from proper randomization (Direct, High; PMID: 39628711) «✓ PMID:39628711».
• Data Inconsistency Scoring: Central statistical monitoring (CSM) utilizes IPD to calculate "data inconsistency scores" derived from thousands of statistical tests to identify clinical sites with atypical patterns that may indicate fraud or carelessness (Direct, High; PMID: 33851576) «✓ PMID:33851576».

Forensic Role in Evidence Synthesis Frameworks

• Step for Major Concerns: The Research Integrity in Guidelines and evIDence synthesis (RIGID) framework and the Trustworthiness in Randomized Controlled Trials (TRACT) checklist designate IPD assessment as the definitive step for investigating trials flagged with major integrity concerns (Direct, High; PMID: 39628711) «✓ PMID:39628711».
• Establishing Authenticity: Established Risk of Bias (RoB) tools are predicated on the assumption that data are true; IPD inspection is required to establish the "authenticity" of a trial before traditional quality appraisal can be considered valid (Direct, High; PMID: 36054583) «✓ PMID:36054583».
• Standardized Tools: The INSPECT-SR project is developing a specialized domain (working name INSPECT-IPD) specifically for forensic checks of underlying datasets in systematic reviews (Derived, Medium; PMID: 37873409) «✓ PMID:37873409».

Operational Challenges

• Resource Intensity: In-depth checks of individual participant data are highly time-consuming and require specific statistical training, often making them prohibitive for routine use by most systematic reviewers (Direct, High; PMID: 36054583) «✓ PMID:36054583».
• Lack of Transparency: Verification is often hindered by the widespread lack of data sharing; in many instances, researchers are unable to clarify concerns because authors do not respond to requests for IPD (Direct, High; PMID: 39628711, PMID: 37453533) «✓ PMID:39628711» «✓ PMID:37453533».

What specific statistical tests are used in Central Statistical Monitoring (CSM) to identify clinical trial fraud?

Which forensic indicators in participant spreadsheets are most predictive of data fabrication in clinical research?

How does the RIGID framework manage the transition from suspected integrity risk to a formal individual participant data investigation?

Summary: The Research Integrity in Guidelines and evIDence synthesis (RIGID) framework manages the transition to formal individual participant data (IPD) investigation through a multi-stage triage process. Transition is triggered when independent reviewers flag a study as having "several items of major concern" using integrated tools like TRACT, followed by integrity committee authorization and formal author engagement to request the underlying datasets (Direct, High; PMID: 39628711).

1. Risk Identification and Triage (TRACT Tool)

The transition begins with Step 3 of the RIGID framework, where studies are screened for "forensic" indicators that cannot be explained by summary reports alone.

• Signals for Escalation: Studies are flagged for IPD investigation if they show major concerns in domains including baseline characteristics (e.g., "perfectly balanced" groups), implausible recruitment timeframes, or illogical methods (Direct, High; PMID: 39628711).
• Assessment of Trustworthiness: The framework utilizes the Trustworthiness in Randomized Controlled Trials (TRACT) checklist. This tool explicitly designates IPD assessment as the next step for randomized controlled trials (RCTs) that accumulate several items of major concern during the preliminary "Are the data true?" evaluation (Direct, High; PMID: 39628711).

2. Committee Authorization and Accountability

RIGID centralizes the decision to move from suspected risk to a formal investigation within a specialized administrative body.

• The Integrity Committee: Decisions to demand IPD are made by a multidisciplinary committee of at least five members, including clinicians, methodologists, and consumers. This committee evaluates the reviewers' initial risk ratings (Step 4: Discuss) to ensure requests for raw data are fair and objective (Direct, High; PMID: 39628711).
• Voting and Consensus: Final transition to the "High Risk" category, which necessitates IPD verification, is determined by committee vote. If a majority consensus cannot be reached, the committee chair makes the final decision based on the technical concerns presented (Direct, High; PMID: 39628711).

3. Formal Requests and Path Allocation

Once a risk is confirmed, the framework moves to Step 5 (Establish Contact), where the transition becomes operational through direct author engagement.

• Standardized Request: The committee sends a formal, diplomatic email to the corresponding authors requesting specific information or clarification. This explicitly includes a request for "sharing of individual participant data" or other supporting documentation like ethics applications or protocols (Direct, High; PMID: 39628711).
• Pathway Allocation (Step 6: Re-assess): The framework defines three investigation outcomes based on the IPD request:
  • Included: The transition is successful if authors provide IPD that addresses all concerns, and the committee validates the authenticity of the data (Direct, High; PMID: 39628711).
  • Awaiting Classification: If the investigation requires more time (e.g., establishing data-sharing agreements for IPD), the study is moved to a "living" category and excluded from the immediate synthesis to preserve timeline integrity (Direct, High; PMID: 39628711).
  • Not Included: If authors refuse to provide IPD or fail to respond to the request, the transition to investigation ends, and the study is removed to prevent the potential inclusion of unreliable data (Direct, High; PMID: 39628711).

4. Forensic Integration

When IPD is successfully obtained, the framework facilitates the application of specialized statistical monitoring techniques.

• Statistical Verification: Once raw data are transitioned to investigators, statistical checks for "anomalous patterns," such as repeating blocks of data or mathematical imbalances in group distributions, are applied to confirm or refute the initial suspicion of fabrication (Direct, High; PMID: 37372725, PMID: 37873409).

Unverified Citations

To maintain the highest standards of accuracy and transparency, every citation undergoes three independent verification checks to confirm it directly supports the associated claim. The references below did not satisfy all verification stages. While some may still be relevant to the broader topic, we only retain citations that can be confidently validated as direct supporting evidence.

• PMID:39628711 — ** Statistical Verification: Once raw data are transitioned to investigators, statistical checks for "anomalous pat...*
  Failed: conclusion — While the paper mentions IPD assessment, it does not describe specific statistical checks for 'repeating blocks of data' or 'mathematical imbalances' as findings/results of the framework.