When Are Medical Records, Administrative Records, or Routine Data Suitable for Research?

Can Records Collected for Healthcare or Administration Be Used as Reliable Research Data?

Hospitals maintain electronic health records, universities maintain student information systems, and government agencies collect administrative information as part of their regular operations. These systems may contain thousands or millions of observations accumulated over several years.

For researchers, such records offer opportunities to investigate patterns that would be difficult, expensive, or impractical to study through new data collection.

But routine records were generally created to support care delivery, service administration, reporting, reimbursement, or operational decisions. Their contents may reflect those purposes rather than the measurement requirements of a particular research question.

When can these records support credible scientific conclusions? What must researchers establish before treating them as evidence, and when might their limitations make another data source necessary?

Routine Records Are Suitable When Their Data-Generation Processes Support the Research Question

Evaluating Medical, Administrative, and Routinely Collected Data for Research

What Are Routinely Collected Data?

Routinely collected data are information generated during regular activities rather than collected primarily to answer a particular research question.

They may arise from clinical care, institutional administration, service delivery, financial transactions, regulatory reporting, or digital systems.

Examples include:

• Electronic health records documenting clinical encounters, diagnoses, prescriptions, and laboratory results.
• Hospital discharge records and healthcare claims databases.
• Government administrative records relating to public services and program participation.
• University enrollment, course completion, attendance, and academic performance records.
• Learning management system logs documenting recorded platform interactions.
• Institutional human resource and professional-development records.

These sources differ substantially in their purposes, structures, and measurement properties.

Researchers should avoid treating "routine data" as a single homogeneous category. A laboratory result generated through a documented clinical procedure presents different measurement issues from a free-text administrative note or an automatically recorded login event.

How Are Medical Records Different from Administrative Records?

Medical records primarily document information generated during healthcare encounters, although they may also serve administrative, legal, and reimbursement functions.

They can include clinical observations, diagnoses, treatments, laboratory findings, imaging reports, medication histories, and professional notes.

Administrative records generally document activities required to operate institutions, deliver services, allocate resources, or meet reporting obligations.

Examples include hospital billing records, insurance claims, student registration records, and public program enrollment databases.

The categories overlap. A diagnosis code may appear in both an electronic health record and a claims database, but its documentation purpose and accuracy may differ.

Researchers should identify the original function of each data element rather than assuming that similarly named variables are equivalent across systems.

Why Are Routine Data Attractive for Research?

Routine data can provide access to large populations, extended observation periods, and information about events occurring in ordinary institutional settings.

They may reduce the need for new recruitment and enable investigation of uncommon outcomes or long-term patterns that would otherwise require substantial resources.

For example, a healthcare database may support examination of recorded service utilization over several years. An educational administrative system may permit analysis of student progression across academic terms.

Such data can also reduce certain recall problems because some events are recorded near the time they occur.

However, a timestamp does not establish that every relevant event was captured accurately, and large numbers of records do not guarantee freedom from systematic error.

The advantages of routine data are conditional on their fitness for the intended research purpose.

The Original Purpose of Data Collection Matters

Routine information is shaped by the processes that generate it.

A healthcare professional may record information needed for clinical decision-making, while an insurance claims system may prioritize information required for reimbursement.

A university information system may accurately document enrollment status but contain limited evidence about students' learning experiences.

These differences affect what researchers can infer.

For example, a diagnosis recorded for reimbursement may not have the same evidentiary basis as a diagnosis established through a standardized research assessment.

Similarly, a learning management system may record that a student opened a resource without establishing whether the student read, understood, or learned from it.

Researchers should examine whether the original purpose of the dataset aligns with the research question before deciding that its variables are suitable.

Population Coverage Determines Whom the Findings Describe

Routine records generally cover people who interact with the system producing the data.

A hospital database contains information about patients who receive care at that institution, not necessarily everyone with a particular condition in the surrounding population.

A university learning platform records interactions by users whose activities are captured by that platform, not necessarily all learning activities undertaken by enrolled students.

Researchers must therefore identify the source population, eligibility rules, observation period, and circumstances under which records enter the database.

Selection into the system may be related to the outcome under investigation.

For example, people with limited healthcare access may be underrepresented in clinical records, while students with limited connectivity may appear less active in online learning logs.

These limitations can affect generalizability and, under some analytical designs, introduce selection bias into estimated relationships.

Data Completeness Must Be Evaluated at Several Levels

Completeness is not simply the percentage of nonblank cells in a dataset.

A database may contain complete records for every registered patient while omitting encounters that occurred outside the institution.

Likewise, a student information system may have complete enrollment histories but incomplete records of extracurricular learning activities.

Researchers should distinguish:

• Population completeness: whether relevant individuals or units are represented.
• Event completeness: whether relevant encounters, activities, or outcomes are recorded.
• Variable completeness: whether required information is available within included records.
• Temporal completeness: whether records cover the periods needed for the research question.

A dataset may perform well on one dimension and poorly on another.

The consequences of incomplete documentation in routine records depend on which information is missing and how that information relates to the study's objectives.

Recorded Diagnoses and Events Require Careful Interpretation

Medical and administrative databases frequently use diagnostic, procedure, or event codes.

These codes can be useful for identifying conditions and outcomes, but their meaning depends on the coding system, institutional procedures, and reasons for documentation.

For example, a diagnosis code may indicate a confirmed condition, a provisional assessment, a billing classification, or another documented status depending on the system.

Researchers should examine coding definitions, clinical or administrative validation studies, and the procedures through which codes were assigned.

Algorithms combining codes, laboratory results, prescriptions, or repeated encounters may sometimes improve case identification, but their performance must be evaluated for the intended context.

The absence of a recorded diagnosis should not automatically be interpreted as the absence of disease.

The distinction between an unrecorded diagnosis and the absence of a medical condition is particularly important when estimating prevalence or defining comparison groups.

Routine Data May Contain Measurement Error and Misclassification

Measurement error occurs when recorded values differ from the quantities or attributes researchers intend to measure.

Misclassification occurs when observations are assigned to incorrect categories, such as identifying someone as having or not having a particular condition.

Routine data may contain these problems because of documentation errors, inconsistent definitions, system limitations, or the operational purpose of recording.

For example, a medication prescription does not necessarily establish that a patient obtained or took the medication.

Similarly, a recorded course attendance event may not establish active participation in learning activities.

Researchers should determine whether the available variables measure the intended exposures and outcomes with sufficient accuracy.

Where appropriate, validation against clinical review, independent assessments, or other credible reference information may be necessary.

How Consistently Were the Data Collected?

Routine records may be produced by different professionals, departments, institutions, or information systems.

Documentation practices can vary even when the same nominal variables are used.

For example, one hospital may routinely record a particular clinical assessment while another records it only for selected patients.

In education, one institution may define course withdrawal differently from another.

Researchers should examine whether data definitions, recording procedures, and quality controls are sufficiently consistent for the intended comparisons.

The question of whether routine data were collected consistently enough for research should be addressed before combining institutions or interpreting longitudinal trends.

Coding Practices May Change Over Time

Longitudinal routine datasets may span changes in classification systems, institutional policies, software platforms, or documentation requirements.

These changes can affect the meaning and frequency of recorded events.

For example, a hospital may adopt a revised diagnostic coding system, or a university may change how course completion is recorded.

An apparent increase in recorded cases may reflect improved documentation rather than a genuine increase in the underlying phenomenon.

Researchers should review historical codebooks, system migration records, and relevant institutional documentation.

When coding practices change over time, harmonization or sensitivity analyses may be needed, and some comparisons may remain infeasible.

Institutional Differences Can Complicate Multi-Site Research

Records collected across institutions may contain differences in coding practices, service populations, data completeness, and operational procedures.

For example, hospitals may differ in admission policies, laboratory testing practices, and diagnostic documentation.

Universities may differ in grading systems, enrollment classifications, or the configuration of learning management platforms.

Combining these records without examining their comparability can create misleading patterns.

Researchers should assess institutional differences in coding practices and consider whether common definitions can be established.

Statistical adjustment for institution may address some variation, but it cannot automatically correct incompatible measurement definitions.

Temporal Information Is Valuable but Requires Validation

Routine records often contain dates and timestamps that permit longitudinal analysis.

However, researchers should establish what each timestamp represents.

A recorded date may indicate when an event occurred, when information was entered, when a claim was processed, or when a record was subsequently updated.

These distinctions matter when defining exposure periods, outcomes, and follow-up.

For example, a diagnosis entered after treatment began may not establish the date when the condition first developed.

Researchers should also consider delayed entry, incomplete historical records, and observation periods that differ across individuals.

The procedures for defining exposures and outcomes using routine records should reflect the actual temporal meaning of the recorded information.

Missing Data and Missing Care Are Different Problems

In healthcare research, the absence of a recorded service or procedure may have several explanations.

The patient may not have received the service, may have received it elsewhere, or may have received it without the event being captured in the available database.

These possibilities have different implications.

For example, no recorded laboratory test does not necessarily establish that testing was clinically unnecessary or that the patient received inadequate care.

Researchers should examine the data system's coverage and the processes through which services are documented.

The distinction between missing care and missing data in health records is essential when evaluating service delivery or healthcare quality.

Record Linkage Can Expand Research Possibilities

Some research questions require information distributed across multiple databases.

For example, a researcher may need to link hospital encounters with mortality records, or university enrollment records with learning platform activity.

Record linkage can improve the completeness of longitudinal information and enable analyses unavailable within a single dataset.

However, linkage introduces additional concerns about identifiers, permissions, matching accuracy, and privacy.

False matches may combine records belonging to different individuals, while missed matches may leave relevant information disconnected.

Researchers should establish when record linkage is necessary and assess how linkage errors may affect the research design.

Routine Data Can Support Descriptive and Associational Research

Routine records are often useful for describing recorded events, service utilization, institutional outcomes, and temporal patterns.

They may also support analyses of associations among exposures, characteristics, and outcomes.

For example, administrative education records might be used to investigate associations between recorded course participation and subsequent academic progression.

However, associations may reflect confounding, selection, measurement error, or other mechanisms.

The researcher should distinguish what the records document from what the analysis seeks to infer.

A large dataset does not automatically establish causal identification.

Can Routine Data Support Causal Inference?

Yes, under suitable research designs and assumptions.

Routine data may support quasi-experimental evaluations, natural experiments, target trial emulation, or other observational causal analyses when the required information and identification conditions are available.

Hernán and Robins' work on causal inference emphasizes the importance of clearly defining the causal question, treatment strategies, eligibility, follow-up, outcomes, and assumptions.

For example, evaluating a healthcare policy using routine records may require a credible comparison group and information about relevant changes over time.

However, missing confounders, inconsistent exposure definitions, or incomplete follow-up may undermine causal conclusions.

Researchers should not treat the routine origin of the data as either sufficient or disqualifying for causal inference. The strength of the inference depends on the design and available evidence.

Data Quality Should Be Evaluated for the Intended Use

Data quality is multidimensional and purpose-dependent.

Weiskopf and Weng (2013) reviewed approaches to evaluating electronic health record data quality and identified dimensions such as completeness, correctness, concordance, plausibility, and currency.

These concepts provide useful starting points for assessing routine data, although their operational definitions may differ across systems and research contexts.

No single data-quality score establishes suitability for every study. A dataset may be adequate for one outcome and unsuitable for another.

Ethical and Legal Requirements Remain Important

Routine records frequently contain sensitive or identifiable information.

Researchers must determine whether the proposed use is permitted under applicable law, institutional policy, data-use agreements, and research ethics requirements.

De-identification may reduce privacy risks but does not automatically eliminate re-identification concerns, particularly when records contain unusual events or combinations of characteristics.

Data linkage can increase these risks by combining information from multiple sources.

Researchers should apply appropriate access controls, data minimization, secure storage, retention procedures, and disclosure safeguards.

Whether consent is required, can be waived, or is otherwise addressed depends on the jurisdiction and circumstances. Researchers should not assume that routine or previously collected data are automatically exempt from ethical oversight.

How Should Routine Data Studies Be Reported?

Transparent reporting should explain the source and purpose of the data, population eligibility, observation period, record selection, variable definitions, linkage procedures, and analytical methods.

The RECORD statement extends the STROBE reporting framework for observational studies using routinely collected health data.

It emphasizes issues such as database characteristics, code-based definitions, linkage, and the processes used to select study participants.

Although RECORD was developed for health research, its attention to data provenance and operational definitions can inform reporting in other fields when applied appropriately.

Researchers should report limitations that arise from the original data-generation process rather than treating routine records as though they were collected under a purpose-built research protocol.

When Should Routine Records Not Be Used?

Routine records may be unsuitable when essential constructs were never measured, the relevant population is inadequately represented, or documentation practices prevent meaningful comparison.

They may also be unsuitable when the intended analysis requires temporal or causal information that the records cannot provide.

For example, a dataset containing only platform logins may not adequately measure students' learning engagement, while billing codes alone may be insufficient to identify a clinical condition with the accuracy required for a particular study.

Researchers should evaluate whether the limitations of the dataset make the proposed secondary analysis infeasible before committing to the study.

When a meaningful narrower question remains, routine records may still be valuable. When the central inference cannot be supported, another data source or research design may be necessary.

Using University Administrative Records to Study Student Progression

The administrative records are suitable for a carefully defined study of institutional progression. They would be less suitable for a broader investigation of students' personal reasons for leaving unless additional evidence were collected.

Common Misconceptions About Research Using Routine Records

How to Decide Whether Routine Records Are Appropriate for Your Study

Begin by reconstructing how the records were generated. Identify why information was collected, which individuals and events enter the system, and how variables are defined.

Then evaluate whether those processes support the research question and intended conclusions.

Routine records should be selected because their documented characteristics make them suitable evidence, not merely because they are already available.

Where limitations remain, define the study's conclusions accordingly and explain which claims the records cannot support.

Check Routine Data Suitability Before Beginning Analysis

Frequently Asked Questions About Research Using Routine Records

Routine Records Are Valuable When Their Original Purpose and Research Use Are Compatible

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.