Can Generative AI Identify Duplicate or Overlapping Study Populations in a Systematic Review?

Could Two Different Articles Be Reporting Results From the Same Participants?

Two articles can have different titles, publication dates, outcomes, and even author lists while reporting data from the same underlying study. One publication might present the primary results of an educational intervention, while another examines student engagement or reports a later follow-up using the same participants.

This creates a problem for systematic reviewers. If the publications are treated as independent studies, the same participants may contribute more than once to the synthesis. The resulting evidence may appear larger or more precise than it actually is.

Generative AI can compare descriptions of participants, recruitment settings, study dates, interventions, and datasets. Could it help researchers recognize overlapping study populations that are not obvious from bibliographic records alone?

AI Can Flag Potential Overlap, but It Cannot Reliably Establish Independence on Its Own

Why Overlapping Study Populations Matter in Evidence Synthesis

What Is an Overlapping Study Population?

Population overlap occurs when two or more research reports include some or all of the same participants, observations, or underlying study data. The overlap may be complete, as when several publications report different outcomes from one trial, or partial, as when separate analyses use intersecting subsets of a larger cohort.

Overlap is not necessarily evidence of misconduct or inappropriate publication. Multiple reports may legitimately provide complementary information about a single investigation. The methodological concern arises when reviewers fail to recognize the relationship and treat dependent evidence as independent.

The Cochrane Handbook emphasizes that studies, rather than reports, are generally the units of interest in systematic reviews. Multiple reports of one study should be identified and linked so that relevant information can be collected without counting the study repeatedly.

How Is Population Overlap Different From Duplicate Search Records?

Duplicate bibliographic records refer to repeated database entries for the same publication. Population overlap concerns the relationship between the underlying participants or data represented in different reports.

Removing duplicate search records does not resolve all possible population overlap. Two genuinely distinct publications may still report evidence derived from the same participants.

What Types of Overlap Might Researchers Encounter?

The distinction between shared data sources and shared observations is important. Two articles using the same national database may analyze different years or nonoverlapping participant groups. Conversely, studies using different analytical methods may rely on exactly the same observations.

What Information Can Generative AI Compare?

AI may help organize evidence concerning study identifiers, trial registrations, recruitment sites, recruitment periods, sample sizes, eligibility criteria, participant characteristics, interventions, funding projects, and named datasets.

For example, two articles may report the same trial registration number despite having different titles and outcomes. That provides a stronger indication of a common underlying study than a shared university affiliation alone.

Other clues are less definitive. Similar sample sizes or recruitment dates may suggest a relationship, but they cannot independently establish overlap.

A useful AI output would therefore identify matching and conflicting details, cite their locations in the source documents, and classify the relationship as requiring verification rather than declaring duplication solely from resemblance.

Can AI Detect Overlap When the Authors Are Different?

Potentially. Different author lists do not rule out shared datasets or participants. Large collaborative projects, secondary analyses, and publicly accessible datasets may produce publications involving different investigators.

AI could compare dataset names, study identifiers, cohort descriptions, and recruitment periods to flag possible relationships. However, the absence of matching authors or identical study titles should not be interpreted as evidence that the underlying samples are independent.

Where publication details remain ambiguous, researchers may need to examine supplementary materials, study registrations, dataset documentation, or contact investigators.

Why Is Double-Counting Participants a Statistical Problem?

Many conventional meta-analytic methods assume that included effect estimates are statistically independent, unless the analysis explicitly models dependence.

If two estimates arise from the same participants and are treated as independent, the analysis may give that evidence excessive weight. Depending on the circumstances, this can affect pooled estimates, standard errors, confidence intervals, and heterogeneity statistics.

For example, including a trial's primary report and follow-up report as two independent trials could incorrectly suggest that evidence comes from two separately recruited samples.

However, overlap does not always require discarding one publication. Different reports may contribute distinct outcomes, time points, or methodological details. The appropriate synthesis strategy depends on the review question and statistical model.

How Should Researchers Handle Confirmed Overlap?

When several reports describe one study, researchers should link the publications under a common study identifier and extract relevant information from each source.

When samples partially overlap across otherwise distinct analyses, the appropriate response is more complicated. Possible approaches include selecting one eligible estimate for a particular analysis, accounting for dependence statistically, or conducting sensitivity analyses.

These choices should follow a justified analytical plan. Researchers should not simply delete every publication sharing participants, because doing so may discard information relevant to different outcomes or comparisons.

Generative AI may help identify the relationship, but it should not independently choose the statistical solution.

What If the Extent of Overlap Cannot Be Established?

Some reports provide insufficient information to determine whether participants overlap. Identical sample sizes, shared institutions, or similar recruitment periods may raise suspicion without resolving it.

Researchers should record the uncertainty and investigate authoritative sources where possible. This may include trial registries, study protocols, cohort documentation, supplementary materials, and correspondence with investigators.

Where uncertainty remains, its implications for synthesis should be considered explicitly. Sensitivity analyses may be appropriate if alternative assumptions about independence could materially change the findings.

What Does the Evidence Establish About Generative AI for Overlap Detection?

Generative AI has demonstrated capabilities in extracting and comparing research information, but that does not establish validated performance for detecting overlapping participant samples across publications.

Published evidence concerning AI-assisted screening and data extraction should not be treated as direct evidence of reliable population-overlap detection. The task requires establishing relationships between reports, sometimes using information unavailable in the articles themselves.

Researchers should therefore regard AI-based overlap detection as an exploratory assistance method unless its performance has been evaluated against independently verified study relationships.

Identifying Overlapping Participants in Educational Technology Research

What Would the Correct Interpretation Be?

The review has identified three publications, but they describe one underlying trial and its associated analyses. The publications may provide useful information about different outcomes and follow-up periods, yet they do not represent three independent samples.

For the original trial population, the number of unique participants is 120, not 320. Adding the reported sample sizes would double-count participants already represented in the earlier reports.

The appropriate treatment of the 80-person follow-up depends on the outcome and synthesis question. The report should not simply be discarded, but its participants should not be treated as newly recruited independent observations.

Common Misconceptions About Overlapping Studies

Use AI to Flag Relationships, Then Verify the Underlying Studies

Population-overlap detection is most useful when integrated into study identification and data extraction. Researchers should preserve study identifiers, trial registration numbers, recruitment information, and links between related publications.

AI can assist by comparing these details across reports and highlighting relationships that deserve investigation.

How Should AI-Based Overlap Detection Be Evaluated?

Compare AI-flagged relationships against independently verified study links. Evaluation should include confirmed overlapping reports, partially overlapping samples, and genuinely independent studies with similar characteristics.

Both missed overlaps and false overlap claims matter. Missing a relationship may lead to double-counting, while falsely identifying overlap may cause researchers to discard independent evidence.

Because the task differs from ordinary eligibility classification, validation should focus on study relationships rather than simply reusing screening accuracy measures without adaptation.

Material AI involvement should be documented through the review's methodological records and AI-use reporting.

Before Treating Publications as Independent Studies

Frequently Asked Questions About AI and Overlapping Study Populations

AI Can Identify Clues, but Study Independence Must Be Established

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.