When scientists want to establish a definitive truth, they typically turn to meta-analysis. Traditionally, this means conducting an aggregate-level review: gathering published papers, extracting summary statistics (such as mean treatment effects or odds ratios), and mathematically combining them.
As shown by Yue Li et al. in Statin uses and mortality in colorectal cancer patients: An updated systematic review and meta‐analysis, this approach is highly effective at pooling massive numbers of patients to find broad, macro-level trends. However, aggregate meta-analysis suffers from a fundamental limitation: it compresses hundreds or thousands of individual participant journeys into a few static metrics. This mathematical averaging erases individual variation, subtle patient-level interactions, and phenotypic nuances—a problem highlighted by Junfeng Huang et al. in Phenotypic heterogeneity in mortality and prognosis of pulmonary alveolar proteinosis: a large-scale, global pooled analysis of individual-level data.
To overcome this loss of resolution, researchers are increasingly shifting to Individual Participant Data (IPD) meta-analysis (also known as mega-analysis or raw data pooling). Instead of synthesizing the conclusions of prior studies, IPD meta-analysis pools the raw, de-identified data of every single participant across multiple independent cohorts into a single framework.
Shifting the Unit of Analysis
The core intuition behind why IPD meta-analysis works lies in changing the fundamental unit of analysis. In a traditional meta-analysis, the unit of analysis is the study. If a study did not explicitly publish a breakdown of how an intervention affected a specific subpopulation, that information is permanently lost to the meta-analyst.
In an IPD meta-analysis, the unit of analysis shifts back to the individual participant. By combining raw data rows from disparate sources, researchers create a unified master dataset. As Ellis Voerman et al. note in Maternal body mass index, gestational weight gain, and the risk of overweight and obesity across childhood: An individual participant data meta-analysis, this "one-stage" pooling enables unified statistical modeling that is impossible when dealing with published summary statistics. It allows researchers to maintain the full granularity of longitudinal data across diverse geographic and temporal contexts, a benefit demonstrated by Andrew Mertens et al. in Causes and consequences of child growth faltering in low-resource settings.
Harmonization and Noise Reduction
One of the greatest challenges in multi-site research is "site noise." Different cohorts often use different equipment, distinct clinical measures, or varied protocols. When summary statistics are generated locally, this noise is baked into the results.
Pooling raw data allows researchers to perform centralized data harmonization and apply uniform processing pipelines. For example, in large-scale neuroimaging, Sean R. McWhinney et al. utilized a standardized multi-site mega-analysis framework in Obesity and brain structure in schizophrenia – ENIGMA study in 3021 individuals to detect subtle structural brain changes that smaller, localized studies lacked the power to isolate.
By controlling the raw metrics directly, investigators can account for site-specific effects and individual covariates (such as age, sex, or scanning variations) simultaneously within the same model. As argued by Ashley A. Huggins et al. in Smaller total and subregional cerebellar volumes in posttraumatic stress disorder: a mega-analysis by the ENIGMA-PGC PTSD workgroup, this rigorous control resolves the inconsistent or contradictory findings that frequently plague small-sample imaging studies. Furthermore, the authors of Bridging big data in the ENIGMA consortium to combine non-equivalent cognitive measures (Eamonn Kennedy et al.) demonstrated that advanced psychometric techniques like Item Response Theory can be applied to raw data to map entirely different clinical tests onto a single, harmonized scale, drastically reducing site-to-site variance.
Unlocking Subgroups and Complex Interactions
Why is this granular control so vital? The primary clinical justification is the pursuit of personalized medicine. To understand which specific patient characteristics influence treatment success, models must evaluate complex, non-linear interactions between interventions and participant traits.
In an aggregate review, these interactions are obscured. By contrast, an IPD approach allows researchers to uncover hidden clinical moderators. For instance, Jim M Smit et al. leveraged an IPD meta-analysis in Predicting benefit from adjuvant therapy with corticosteroids in community-acquired pneumonia: a data-driven analysis of randomised trials to discover that a patient's baseline C-reactive protein (CRP) level determines whether they will benefit from corticosteroid therapy. Standard meta-analyses simply lack the resolution to detect these individual-level thresholds, a challenge discussed by Rebecca B. Price et al. in International pooled patient-level meta-analysis of ketamine infusion for depression: In search of clinical moderators.
Maintaining a consistent unit of analysis across disparate sources also grants the statistical power required to model rare clinical events and long-term trends. This capability is reflected in several domains:
- Rare Diseases and Complications: Nishant K. Mishra et al. outlined a protocol in Clinical characteristics and outcomes of patients with post-stroke epilepsy: protocol for an individual patient data meta-analysis from the International Post-stroke Epilepsy Research Repository (IPSERR) to predict post-stroke epilepsy, demonstrating how centralized repositories can capture interactions that are invisible in isolated studies.
- Longitudinal Trajectories: Holly Tibble et al. pooled clinical trial data in Prevalence, Trajectory, and Predictors of Poststroke Pain: Retrospective Analysis of Pooled Clinical Trial Data Set to map the long-term persistence of post-stroke pain, identifying predictors that smaller, localized cohorts could not reliably confirm.
- Confounding Control: In epidemiological tracking, Elise M. Philips et al. used IPD pooling in Changes in parental smoking during pregnancy and risks of adverse birth outcomes and childhood overweight in Europe and North America: An individual participant data meta-analysis of 229,000 singleton births to disentangle the independent impacts of maternal and paternal smoking on birth outcomes, achieving a level of covariate control that aggregate data cannot support.
Beating the Curse of Dimensionality
In high-dimensional fields like genomics and metabolomics, researchers routinely face the "large $p$, small $n$" problem: thousands of features ($p$) are measured, but only a handful of subjects ($n$) exist within any single study.
Traditional meta-analysis struggles here because the individual screenings themselves suffer from low statistical power and high false-positive rates. Pooling raw data across multiple independent studies—often called integrative analysis—fundamentally changes the math. As Rui Song 0006 et al. demonstrate in Integrative prescreening in analysis of multiple cancer genomic studies, screening pooled raw data identifies significant genetic markers far more effectively than meta-analyzing the results of isolated screenings. By capturing the underlying granular data structures directly, these models achieve significantly higher accuracy and reproducibility in biomarker identification (Qingzhao Zhang et al. in Penalized integrative analysis under the accelerated failure time model).
The Future: Prospective Alignment
Because the benefits of raw data pooling are so pronounced, clinical research is shifting from retrospective data gathering to prospective alignment. Instead of waiting for trials to publish and then attempting to extract their raw data, international consortia now align their independent study protocols beforehand.
As detailed by Anouk Pels et al. in STRIDER (Sildenafil TheRapy in dismal prognosis early onset fetal growth restriction): an international consortium of randomised placebo-controlled trials, this collaborative framework ensures that raw data can be merged seamlessly the moment the trials conclude. By designing studies with subsequent IPD meta-analysis in mind, the scientific community can systematically overcome the sample size limitations of underpowered trials, transforming isolated streams of research into a unified, high-resolution engine for discovery.
Key Foundational Work
STRIDER: Sildenafil therapy in dismal prognosis early-onset intrauterine growth restriction – a protocol for a systematic review with individual participant data and aggregate data meta-analysis and trial sequential analysis (2014) — contributed 2 concept elements to this direction.
Integrative prescreening in analysis of multiple cancer genomic studies (2012) — contributed 1 concept elements to this direction.