Retrospective chart review and data abstraction
Why this level?
Specify the research question: intervention, diagnostic accuracy, prognosis, aetiology, or screening intervention. Other question types are outside this hierarchy. NHMRC 2009 sourceChart review is a way of collecting data rather than a design, so the question it serves has to be named before any placement applies. A cohort, case-control study or case series built from charts takes that design's placement, and the completeness and consistency of documentation are appraised separately.
What question does this help answer?
Chart review is a method of data collection, not a design. It can supply a case series, a cross-sectional estimate, a cohort or a case-control study, and the design it supplies determines what the study can conclude.
When to use it
Use chart review when the information needed is recorded in clinical notes, results or orders, when it is not available as reliable coded data, and when a prospective study would be slower or impractical.
When not to use it
Avoid relying on charts for variables that are documented inconsistently, such as symptoms, functional status or behaviours recorded only when abnormal, or when documentation practice changed during the study period in a way linked to the exposure or outcome. If the key variable is rarely recorded, no abstraction protocol can recover it.
Build the abstraction protocol before reading charts
- Write an operational definition for every variable, including where in the record it is found and what to record when it is absent, contradictory or ambiguous.
- Pilot the abstraction form on charts outside the study sample and revise it before abstraction begins.
- Train abstractors on the definitions and the form, and record who abstracted each chart.
- Where feasible, keep abstractors unaware of the study hypothesis and, when abstracting exposures, of the outcome.
- Have a second abstractor independently re-abstract a sample of charts and report agreement for key variables, using a statistic suited to each variable type.
- Date the protocol and log later changes. A definition revised after reading charts must be applied to every chart, including those already abstracted.
Not documented is not the same as absent
Separate 'recorded as absent' from 'not documented'. Treating an undocumented finding as absent assumes documentation was complete, and missing entries are rarely missing completely at random: tests are ordered, and findings written down, for patients who looked different. State the missingness assumption and how it will be handled.
Identify charts without looking at outcomes
Define the source population and the inclusion window using something that does not depend on the outcome, such as an admission date, a procedure code or a clinic list, and describe how charts were found. Report how many were screened, excluded and abstracted, with reasons.
Clinical example
A team studies delirium after hip fracture surgery from charts. The protocol defines delirium as a documented diagnosis or a positive screening result, with a written rule for notes describing confusion without either. A second abstractor, unaware of the medication exposures, re-abstracts a random subset, and disagreements are resolved against the definitions. This example is illustrative.
What to watch out for
Charts were written to deliver care, not to answer the study question. A finding that was not documented is not the same as a finding that was absent, abstractors who know the hypothesis or the outcome can shift what they record, and a definition revised after reading some charts makes the data depend on the data.
How to report it
RECORD is the guideline for studies using routinely collected health data, and it extends STROBE rather than replacing it. RECORD asks for the codes and algorithms used to classify exposures, outcomes and confounders, the data cleaning and linkage methods, and how far the investigators could reach the database population the study was drawn from, all of which an abstraction from charts must state to be reproducible. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.
Before using this in your workspace
Record the design the chart review supplies, the source population and how charts were identified, each variable's definition and location, the abstraction form version, abstractor training and blinding, the re-abstraction sample and the missing-data assumption.
- A methodology for conducting retrospective chart review research in child and adolescent psychiatry (opens in a new tab) · J Can Acad Child Adolesc Psychiatry
- The retrospective chart review: important methodological considerations (opens in a new tab) · J Educ Eval Health Prof · doi 10.3352/jeehp.2013.10.12
- Looking through the retrospectoscope: reducing bias in emergency medicine chart review studies (opens in a new tab) · Ann Emerg Med · doi 10.1016/j.annemergmed.2014.03.025
- Advanced statistics: understanding medical record review (MRR) studies (opens in a new tab) · Acad Emerg Med
- The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement (opens in a new tab) · PLoS Med · doi 10.1371/journal.pmed.1001885
- Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies (opens in a new tab) · BMJ · doi 10.1136/bmj.39335.541782.AD