Can Research Registries Make Rare-Outcome Studies Feasible?

Can an Existing Registry Solve the Problem of Having Too Few Rare Cases?

Suppose you want to investigate an uncommon clinical complication, an unusual educational outcome, or a condition affecting a relatively small population. Your institution identifies only a few eligible cases each year, making conventional recruitment difficult.

You discover that a regional or national registry has accumulated records of the outcome across several institutions and many years. Instead of recruiting participants individually, you could potentially use those records to identify cases, reconstruct exposure histories, and investigate relevant outcomes.

The opportunity is attractive. A registry may contain hundreds of cases that would otherwise take years to recruit. However, its records were not necessarily collected to answer your particular research question.

Does the registry represent the population you want to study? Are cases identified consistently? Does it contain the exposure, confounder, and follow-up information required for the intended analysis?

A registry can make a rare-outcome study operationally feasible without necessarily making its conclusions scientifically defensible.

Registries Can Improve Feasibility When Their Data Support the Research Question

How Research Registries Can Support Studies of Rare Outcomes

What is a research registry?

A registry is an organized system that collects and maintains information about individuals, conditions, events, exposures, or other defined characteristics according to specified procedures.

Registries may be established for clinical care, public health surveillance, administrative reporting, quality improvement, or research. Their purposes influence which participants are included and which variables are recorded.

For example, a disease registry may collect diagnostic information, treatments, and follow-up outcomes. An educational registry might document enrollment, progression, completion, or uncommon institutional events across participating institutions.

Not every database is a research registry. A collection of records becomes useful for registry-based research when its population, eligibility rules, ascertainment procedures, and data structure can be understood sufficiently for the intended investigation.

The Agency for Healthcare Research and Quality's registry guidance emphasizes the importance of defining registry purposes, populations, data elements, and quality procedures before interpreting registry information.

Why are registries especially useful for rare outcomes?

Rare outcomes produce relatively few cases within any one institution or short observation period.

A registry may aggregate eligible events across a much larger population or accumulate observations over several years.

For example, a complication occurring in approximately 1 of every 5,000 eligible individuals may yield only a few cases annually at one hospital. A registry covering hundreds of thousands of individuals may identify substantially more cases.

Registries can therefore reduce the time and expense required to locate uncommon events.

However, a larger case count does not automatically mean that enough information is available for the intended statistical analysis. Exposure distributions, missing confounders, outcome definitions, and the number of events within relevant subgroups remain important.

What kinds of registries can support rare-outcome research?

Different registries offer different forms of evidence.

These categories can overlap. A registry may be both population-based and longitudinal, for example.

Researchers should evaluate the registry's actual design rather than relying exclusively on its name.

Does a registry need to be population-based?

Not for every research question.

A registry that collects cases from selected specialist centers may be useful for describing clinical characteristics, treatment experiences, or outcomes among registered patients.

However, it may not support incidence estimation for the general population because the probability of entering the registry is unknown or varies across groups.

A population-based registry aims to capture eligible cases within a defined geographic or administrative population.

Even then, researchers must assess ascertainment completeness and the accuracy of population denominators.

The distinction becomes particularly important when estimating prevalence, incidence, or population-level risks rather than describing registered cases.

How can researchers determine whether registry coverage is adequate?

Coverage concerns which eligible individuals or events can enter the registry and how completely they are captured.

Researchers should examine the geographic area, participating institutions, eligibility rules, reporting obligations, registration procedures, and periods covered.

For example, a registry containing cases from major referral hospitals may miss patients treated in community facilities.

Similarly, an educational registry involving only institutions that voluntarily submit data may exclude schools with different student populations or administrative resources.

Where appropriate, researchers can compare registry counts with external administrative statistics, independent ascertainment systems, or other relevant sources.

However, agreement with another source does not necessarily establish complete coverage if both sources omit similar groups.

The appropriate assessment depends on the registry's purpose and the population to which findings will be applied.

Why is consistent case ascertainment essential?

Case ascertainment is the process of identifying individuals or events that satisfy the outcome definition.

For rare outcomes, relatively few classification errors may substantially affect estimated frequencies and associations.

Suppose a registry identifies an uncommon complication using diagnostic codes. If those codes have limited positive predictive value, some recorded cases may not meet the intended clinical definition.

Researchers should examine the registry's diagnostic or event criteria, validation procedures, and relevant coding changes.

Where feasible, a subset of cases may be verified against source records using an appropriate validation procedure.

Case definitions should also be examined across sites and years. A registry may use the same outcome label while applying different operational criteria over time.

These concerns are especially important when historical data are added to increase sample size.

Can a registry provide an appropriate comparison group?

Sometimes. A registry containing only individuals who experienced a rare outcome may support case descriptions but not necessarily a comparative study.

For a case-control design, researchers need controls sampled according to the population that produced the cases.

For a cohort study, the registry or linked databases must identify eligible individuals at risk, including those who did not experience the outcome.

For example, a registry containing only students who permanently withdrew cannot directly estimate withdrawal risk among all enrolled students without additional enrollment information.

Likewise, a disease registry containing cases but no suitable noncases may require linkage to population records or another defensible comparison source.

The comparison population should be established through the research design, not selected merely because another dataset is available.

Which research designs can registries support?

Registries may support several designs when their structure and coverage are appropriate.

A registry is a data infrastructure, not an automatic substitute for a research design.

The chosen design determines the required sampling information, causal assumptions, and appropriate estimators.

Can registries improve temporal information?

Yes, when exposure dates, eligibility periods, event dates, and follow-up information are recorded reliably.

For example, a registry may document when a treatment began and when a subsequent complication occurred.

This can support analyses requiring clear temporal ordering.

However, some registries record only the date of diagnosis or registration, which may differ from the actual onset of the condition.

Researchers should distinguish event occurrence, recognition, reporting, and registration dates.

Incorrect timing can affect incidence estimates, risk periods, and causal interpretations.

Registry data should not be assumed longitudinally valid merely because they contain dates.

What happens when registry definitions change over time?

Changes in diagnostic criteria, coding systems, reporting obligations, or institutional practices can affect which cases enter a registry.

For example, an apparent increase in an uncommon condition may reflect improved screening rather than a genuine increase in occurrence.

Similarly, a change in university withdrawal classifications may increase recorded events without a corresponding change in student behavior.

Researchers should review registry documentation and identify relevant changes before combining periods.

Where defensible, harmonization can establish a common definition. When that is impossible, period-specific analyses or restrictions may be necessary.

Statistical adjustment for calendar year does not automatically correct incompatible measurement definitions.

How does registry participation create selection bias?

Registry inclusion may depend on factors related to exposure, outcome severity, access to services, or participant characteristics.

For example, a voluntary rare disease registry may attract individuals with more severe symptoms or stronger engagement with patient organizations.

If those characteristics also relate to the exposures under investigation, associations among registered individuals may differ from those in the intended population.

Similarly, specialist-center registries may disproportionately include complicated cases.

Researchers should investigate how individuals enter the registry and whether participation conditions on variables that affect the intended comparison.

Statistical adjustment cannot automatically remove selection bias when relevant determinants of inclusion are unmeasured.

Why do missing data remain important in registry research?

Registries may contain many cases while lacking essential variables for particular analyses.

For example, an administrative registry may record diagnosis and treatment but omit socioeconomic characteristics, prior exposures, or detailed outcome measures.

Missingness may also differ across institutions, years, and participant groups.

Researchers should examine completeness at the variable and record level, distinguishing information never collected from measurements expected but unavailable.

Methods such as multiple imputation may be useful under suitable assumptions, but they cannot automatically recover variables that were systematically absent from entire historical periods.

Before selecting the registry, researchers should identify which essential variables are vulnerable to missing data.

Can registry linkage improve rare-outcome research?

Linking registry records with other authorized sources may provide exposure histories, population denominators, follow-up outcomes, or confounding information unavailable in the original registry.

For example, a disease registry might be linked to authorized treatment records or vital statistics.

An educational registry might be linked to enrollment histories or standardized assessment records.

However, linkage can introduce false matches and missed matches. Linkage quality may also differ across population groups.

Researchers should document identifiers, matching procedures, validation methods, and privacy protections.

Linkage is useful only when it adds valid information and complies with applicable ethical and legal requirements.

Does a registry automatically provide enough statistical power?

No. The number of registered individuals is not equivalent to the number of informative outcome events or usable comparisons.

A registry containing 100,000 participants may have only 40 cases of the outcome under investigation.

Similarly, a registry containing 500 rare cases may have very few cases exposed to a particular risk factor.

Statistical information depends on the event count, exposure distribution, confounding structure, clustering, and planned estimator.

Researchers should use design-specific sample size or precision calculations, and simulation where appropriate.

The registry's advertised size should not substitute for an analytical feasibility assessment.

Can registries support causal inference?

Potentially, but registry size does not establish exchangeability or eliminate confounding.

Individuals receiving particular treatments or experiencing particular exposures may differ systematically from comparison participants.

For example, patients receiving an uncommon treatment may have more severe disease than those receiving standard care.

Researchers should define the causal estimand, establish temporal ordering, evaluate eligibility and comparison groups, and measure relevant confounders.

They should also examine selection processes, positivity, and measurement validity.

Approaches informed by target trial emulation may help structure certain observational comparisons, but they still depend on appropriate data and identification assumptions.

Registry-based observational associations should not automatically be described as causal effects.

How should researchers assess registry quality before obtaining data?

Begin with the registry protocol, data dictionary, eligibility criteria, population coverage, and documentation of collection procedures.

Determine whether essential exposures, outcomes, dates, and confounders are available and sufficiently complete.

Examine validation studies where available, particularly for rare-event definitions.

Review changes in data collection over time and across participating institutions.

Researchers should also establish whether the registry permits the required sampling, linkage, and analysis under applicable access conditions.

Where documentation is insufficient, a feasibility extraction or data-quality assessment may be necessary before committing to the full study.

When should researchers avoid relying on a registry?

A registry may be unsuitable when its population differs substantially from the intended target population, case ascertainment is unreliable, or essential exposure and confounding information is unavailable.

It may also be inadequate when the desired comparison group cannot be identified or the outcome definition changes in ways that cannot be reconciled.

Researchers should consider whether new multisite recruitment, a purpose-built cohort, or another design would provide more defensible evidence.

In some circumstances, a registry may still support a valuable descriptive study even when it cannot answer the originally proposed causal question.

The appropriate response is to align the research objective with the evidence the registry can actually provide.

Using a University Registry to Investigate an Uncommon Withdrawal Outcome

The calculation is hypothetical and depends on assumptions that must be evaluated rather than presumed.

If institutions contribute students with different follow-up durations, the simple proportion may not estimate a common-period risk. An incidence-rate or time-to-event analysis may be more appropriate.

The registry's value lies in providing a defined population and potentially reliable outcome information, not merely in containing hundreds of rare cases.

Common Misconceptions About Rare-Outcome Registries

Evaluate the Registry as a Research Design Resource, Not Just a Source of Cases

Before requesting registry access, establish the research question and determine which population, exposure, outcome, and confounding information is required.

Then examine the registry's documentation, ascertainment procedures, coverage, and analytical limitations.

Where the registry cannot support the intended study, compare its limitations with other research designs for rare outcomes.

The most useful registry is not necessarily the largest one. It is the one whose population, measurements, and observation process align with the research question.

Before Using a Registry for Rare-Outcome Research

Frequently Asked Questions About Registries and Rare Outcomes

Registries Improve Rare-Outcome Feasibility When Their Data Support Valid Inference

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.