The Ulutas Medical Journal p-ISSN 2149-0430 e-ISSN 2149-388X

Planning a retrospective study

Last updated:

A retrospective study uses information that already exists, usually patient files, imaging archives or laboratory records. It can look like the easiest project for a busy resident, because the data are already in the hospital system. But "the data are already there" is also the main weakness: they were collected for patient care, not for your question. Careful planning before you open the first file is what turns a chart review into a publishable observational study.

The research question

Write your question down in one sentence before you collect anything. A useful format names the population, the exposure or characteristic, the comparison and the outcome, for example: "In adults admitted with community-acquired pneumonia, is a high admission CRP associated with 30-day mortality compared with a normal CRP?"

The STROBE checklist, the reporting guideline for observational studies, asks authors to state specific objectives, including any prespecified hypotheses, and to define clearly every outcome, exposure, predictor and potential confounder. Doing this at the planning stage has two benefits: you know exactly which variables to extract, and you avoid the temptation to search the data for any "significant" result afterwards. SAMPL, a guideline for statistical reporting, asks that unplanned subgroup analyses be identified as exploratory, so decide your main analysis in advance and write it down.

Also check whether your records can actually answer the question. If the outcome is rarely recorded, or the exposure is measured in different ways in different years, change the question now rather than after months of work.

Inclusion and exclusion criteria

STROBE asks you to give the eligibility criteria and the sources and methods of selection of participants, and to describe the setting, locations and relevant dates, including the study period. Write these as rules that another person could apply to the same archive and arrive at the same list of patients:

  • the hospital or department and the exact date range;
  • how patients were identified (for example diagnosis codes, operation lists or laboratory searches);
  • inclusion criteria (who counts as a case);
  • exclusion criteria, with a reason for each one.

Count the patients at every step: found by the search, screened, excluded (with reasons) and analysed. STROBE asks for these numbers and suggests a flow diagram, so keep the counts as you go.

Data collection and bias

Design a data collection form before extraction and test it on a few files. Define each variable precisely (units, how a missing or unclear entry is recorded, which value to take if several exist). If more than one person extracts data, agree on the rules in writing.

STROBE asks authors to describe any efforts to address potential sources of bias, to explain how the study size was arrived at, to explain how missing data were handled, and to discuss the direction and size of possible bias in the limitations. In record-based studies, think in particular about:

  • missing data – report the number of patients with missing data for each variable, as STROBE asks, and state how you handled them;
  • selection – whether the patients whose files you could find differ from those you could not;
  • confounding – which other factors could explain the association, and how your analysis adjusts for them;
  • study size – explain how you decided how many patients to include, even if the answer is "all eligible patients in the study period".

Store the extracted data in a de-identified file, keep the key that links study numbers to patients separately and securely, and collect only the variables your question needs.

Ethics approval for record reviews

Using existing records still involves identifiable patient data. The Declaration of Helsinki states that where consent is impossible or impracticable to obtain, research on stored data may be done only after consideration and approval by a research ethics committee. This journal requires the name of the ethics committee, the approval number and date in every manuscript, and, where the committee waived the need for consent (as often happens in retrospective studies), a statement saying so together with the committee's name and approval number (see the Informed Consent page).

In Türkiye, several universities have a separate non-interventional clinical research ethics committee (girişimsel olmayan klinik araştırmalar etik kurulu). Istanbul Medipol University's directive, for example, lists retrospective reviews of file and image records among non-interventional research. Ask your own institution which committee is responsible. Apply before you start collecting data: the Declaration of Helsinki says the protocol must go to the ethics committee before the research begins. Our guide on ethics committee approval gives more detail.

Reporting with STROBE

This journal asks authors of observational studies to follow STROBE (see the Instructions for Authors). STROBE has separate checklists for cohort, case-control and cross-sectional studies, and a combined version. Use it while writing, not after:

  • name the design with a commonly used term in the title or abstract (for example "retrospective cohort study");
  • in the methods, cover setting, participants, variables, data sources, bias, study size and statistical methods;
  • in the results, report the numbers at each stage, descriptive data and both unadjusted and, where relevant, adjusted estimates with their precision;
  • in the discussion, cover key results, limitations, interpretation and generalisability;
  • state the funding source and the role of the funder.

Studies based on routinely collected health data may also use RECORD, an extension of STROBE developed for such data and listed by the EQUATOR Network. Submit the completed checklist with the page numbers for each item.

Sources

All guides

WhatsApp Tell the Editor
Submit manuscript Contact