Two Directions for Collecting Observational Data
In an observational study, researchers measure or record what happens without assigning participants to an exposure. They may begin collecting information before the outcomes of interest occur, or they may examine information that already exists. These are the central timing differences between prospective and retrospective observational studies.
The distinction concerns the researchers’ data collection, not whether exposure and outcome have a meaningful order. For example, a study may ask whether a workplace exposure is associated with a later health outcome. In either design, the researchers need to identify what counts as the exposure, what counts as the outcome, and the time period relevant to each. The earlier material on cohort and case-control studies provides related design categories; prospective and retrospective describe a different feature: when and how study information is gathered.
The starting point, sometimes called the index date, is the reference time from which a participant’s eligibility, exposure status, and follow-up are defined. In a prospective study, it is generally set before the outcome period begins. In a retrospective study, researchers set or identify a past starting point using records. Being retrospective does not mean that the outcome must have occurred before the exposure: the records can still describe exposure first and outcome later. What is retrospective is the researchers’ use of already-existing information.
Define Exposure and Outcome Before Classifying the Study
The exposure is the characteristic, behavior, or circumstance whose relationship with an outcome is being studied. It might be a naturally occurring behavior, an environmental condition, or a medical treatment received in ordinary care. The outcome is the event or measurement of interest, such as a diagnosis within a specified period. In an observational study, researchers do not assign the exposure as they would assign a treatment in an experiment.
A useful description states four things: who is studied, how exposure is defined, what outcome is measured, and the period in which that outcome is assessed. For instance, “adults who report working night shifts at enrollment, followed for a new diagnosis during the next year” identifies a population, exposure, outcome, and follow-up period. Precise definitions matter because a vague exposure or an unclear outcome window makes the study difficult to interpret or reproduce.
In a prospective design, researchers can define these features in advance and collect exposure information before the outcome period. In a retrospective design, they define them using records or recollections from the past. A health record might contain a reliable date of diagnosis but incomplete information about a past behavior. Conversely, a participant may remember a behavior but not its timing accurately. The available source affects how well each variable can be measured.
Worked Example: Following Night-Shift Workers Forward
Suppose researchers enroll 120 adults who have not had a particular sleep disorder at the start of a hypothetical study. They record each person’s current work schedule without changing it, then check health records for a new diagnosis over the next 12 months. At enrollment, 40 participants work regular night shifts and 80 do not. During follow-up, 8 night-shift workers and 8 other workers receive the diagnosis.
This is a prospective observational study. Researchers record the exposure—regular night-shift work—before the 12-month outcome period and then follow participants forward. The outcome is a new diagnosis of the sleep disorder during that period. Since researchers observe rather than assign work schedules, this is not a randomized experiment.
The observed outcome proportion in each group is:
In this hypothetical sample, 20% of the night-shift group and 10% of the other group received the diagnosis during follow-up. The difference is 10 percentage points. These descriptive results do not establish that night-shift work caused the outcome; the groups were observed, not created by random assignment, and may differ in other ways.
Prospective Studies: Measure Now, Follow Forward
The defining feature of prospective collection is that researchers organize measurement before the outcomes of interest have occurred. They can specify a follow-up period and decide in advance how exposure and outcome information will be recorded. A planned schedule of check-ins, standardized measurements, or regular record reviews can make data collection more consistent than relying on incomplete historical information.
Prospective studies can also require considerable time and resources. Outcomes may be uncommon, take years to appear, or occur in participants who are difficult to contact later. Some participants may leave the study or miss assessments. Researchers therefore need to describe who was followed and how much outcome information was obtained. A study’s prospective timing alone does not guarantee accurate measurement, complete follow-up, or a representative group.
Worked Example: Classifying a Planned Follow-Up Study
Imagine a hypothetical research team that identifies 300 adults who currently use a particular non-prescription supplement and 300 who do not. The team does not ask anyone to start or stop using it. At enrollment, researchers record supplement use, then review medical records for a first diagnosis of a specified condition over the following two years. The hypothetical results are 18 diagnoses among supplement users and 12 among non-users.
This is prospective because exposure status is recorded before the outcome period, and the team follows participants forward. It is observational because supplement use was not assigned. Here, the outcome is a first diagnosis during two years, not any diagnosis at any time. That definition keeps the outcome window explicit.
If all 300 people in each group have complete follow-up, the observed proportions are:
Thus, 6.0% of the supplement group and 4.0% of the non-user group had the specified diagnosis during the two-year period in this hypothetical study. The arithmetic uses each group’s number of diagnoses divided by its total number followed. These proportions describe the observed groups; by themselves, they do not show that supplement use changed the chance of diagnosis.
Retrospective Studies: Reconstruct the Past
A retrospective study uses information about the relevant period that is already available when researchers undertake the study. Sources might include medical charts, employment records, archived surveys, or participant interviews about past events. Researchers still need to specify exposure and outcome clearly; “the records already exist” is not a substitute for defining what values count.
For example, researchers might identify a past group of patients, use records to determine which patients received a particular medication before a specified date, and check those same records for an outcome during a later period. This is commonly described as a retrospective cohort design: the historical records supply information about exposure and subsequent outcomes for a defined group. The events may have unfolded in the same exposure-then-outcome order as in a prospective study, even though the researchers examine them after they occurred.
Retrospective studies can often be completed more quickly than studies that wait for future outcomes. They may also use large records covering many years. Their limitations depend on what was recorded and why. If exposure information was not collected consistently, if outcome definitions changed over time, or if records are missing, researchers may be unable to measure the variables as precisely as they would in a planned study. A record of “no entry” does not always mean that an event did not happen.
Worked Example: Using Past Medical Records
Consider a hypothetical team reviewing records from a hospital’s 2022 surgeries. The researchers identify 200 patients who had the same type of surgery. The records show that 60 received a specified medication before surgery and 140 did not. The team then checks for a documented infection within 30 days after surgery. There are 9 infections among the 60 medication recipients and 7 among the 140 others.
This is retrospective observational research: when the team conducts the study, the medication exposure and the 30-day outcomes have already occurred and are reconstructed from records. The exposure is receipt of the medication before surgery; the outcome is a documented infection within 30 days afterward. Researchers did not assign the medication for this study.
Assuming the records identify all 200 eligible patients and completely ascertain the stated outcome window, the observed proportions are:
In these hypothetical records, 15% of the medication group and 5% of the other group had a documented infection within 30 days. The comparison is descriptive, not proof of a medication effect. For instance, the records alone do not establish that the two groups were similar in the reasons for receiving medication.
Why Study Direction and Sampling Matter
Not every retrospective study begins by identifying an exposed group and then counting outcomes. In a case-control study, researchers select participants based on whether they have the outcome, then look back to compare their prior exposure histories. This is retrospective in its use of past exposure information, but the sample was assembled around outcome status rather than by following all members of an original group.
Worked Example: Looking Back From Cases and Controls
Suppose a hypothetical case-control study selects 50 people with a particular diagnosis (cases) and 100 people without it (controls). Researchers review past records to determine whether each person had a specified workplace exposure. The exposure is recorded for 30 cases and 20 controls.
The exposure proportions in the selected groups are:
The hypothetical sample contains a higher proportion with the recorded exposure among cases than among controls. Because researchers selected a fixed number of people based on outcome status, these fractions do not tell us the proportion of exposed people in the source population who developed the diagnosis. The study looks backward from outcome groups to exposure; it does not directly follow an exposed and unexposed cohort to count new outcomes.
This example shows why “retrospective” does not, by itself, describe how participants were selected or which numerical summaries can be calculated. A retrospective cohort can use past records to count outcomes in exposure groups. A case-control study can use past records to compare exposure histories in outcome groups. The study’s sampling structure determines what its observed counts represent.
A Short Justification: Why the Denominator Changes the Meaning
For a group followed over a defined period, the observed proportion with the outcome is the number of group members who experience the outcome divided by the number in that group at the start of the period. In the first example, the night-shift group’s denominator is 40 because 40 people were in that group and followed for the outcome. The calculation is:
In the case-control example, researchers instead chose 50 cases and 100 controls based on outcome status. Dividing exposed cases by all cases answers “what fraction of selected cases had the recorded exposure?” It does not answer “what fraction of exposed people developed the outcome?” Those are different questions because the denominators are different. A number should therefore be interpreted in light of how participants entered the study, not just how it was calculated.
Common Mistakes and Practical Checks
- Confusing event order with collection timing. A retrospective study can examine exposure that preceded outcome. “Retrospective” means researchers use information about past events, not that the outcome must have come first.
- Treating prospective as experimental. Following people forward does not mean researchers assigned their exposure. If participants’ existing behaviors or circumstances are observed rather than assigned, the study is observational.
- Assuming every retrospective study is case-control. Existing records can also support a retrospective cohort study. Ask whether participants were grouped by exposure or selected by outcome.
- Calling a percentage a risk without checking the denominator and period. A proportion based on people followed over a stated period differs from the proportion exposed among selected cases. State who is in the denominator and what time window is used.
- Assuming records are complete just because they are written down. Check whether exposure and outcome definitions were applied consistently and whether missing information could change the interpretation.
Check Your Understanding
Answer each question by identifying the timing, exposure, outcome, or interpretation involved.
- Researchers record current exercise habits and then check for a new diagnosis over the next three years. Is the study prospective or retrospective, and what is the exposure?
- A team reviews records from five years ago to compare outcomes in people who did and did not receive a treatment at that time. Why can this be retrospective even if treatment came before the outcome?
- Why does following participants forward not, by itself, make a study an experiment?
- In a case-control study, what does the fraction of cases with a prior exposure describe? Why does it not directly give the outcome proportion among all exposed people in the population?
- Name two details that should be specified when defining an outcome for either a prospective or retrospective study.