In this design, researchers identify and select participants before the outcome of interest occurs. Exposed and non-exposed groups are defined and followed over time to assess the incidence of the disease. A key advantage is the ability to measure exposures accurately and establish the temporal sequence between exposure and disease, which reduces information bias¹.

A classic example is the Framingham study. It was one of the most influential cohort studies in cardiovascular epidemiology, initiated in 1948 in Framingham, Massachusetts. Key risk factors for cardiovascular diseases, such as smoking, hypertension, and dyslipidemia, were identified^1,3.

In this design, both the exposure and the outcome have already occurred by the time the study begins. Medical records, databases, or clinical histories are used to reconstruct the exposure history and assess the incidence of the disease. While this method is more time- and cost-efficient, it can be limited by the quality of available data and information bias¹.

An example is exposure to radiation and cancer. Retrospective studies in Hiroshima and Nagasaki have evaluated the incidence of cancer among atomic bomb survivors, using historical records to determine exposure and follow individuals over the years^1,4.

This design combines both prospective and retrospective elements, using historical data to assess past exposures and continuing to follow individuals in the present and future. This approach allows for the use of prior information, which can be complemented with prospective data.

An example is the British birth cohort of 1946. This study began with perinatal data and has followed participants into adulthood, gathering information on multiple exposures throughout their lives¹.

There are methodological variations that can enhance the efficiency and precision of cohort studies.

• Multiple cohort studies: These studies compare two or more groups of individuals with different levels of exposure. For example, workers exposed to a chemical agent can be compared to unexposed workers in the same industry¹. One example is the study on asbestos exposure. A study compared workers exposed to asbestos with office employees and evaluated the incidence of mesothelioma¹.

• Matched cohort studies: In this design, each exposed participant is matched with an unexposed individual with similar characteristics, such as age and sex, reducing the effect of confounding variables¹. One example is studies on smoking and lung cancer. Several studies have matched smokers and non-smokers based on factors such as age and socioeconomic status, improving comparability between the groups¹.

• Nested cohort studies: These studies combine the advantages of cohort and case-control studies. Within an established cohort, cases of the disease are identified and compared with a randomly selected subset of controls from the same cohort, reducing costs and improving study efficiency¹. An example is the Nurses’ Health Study. This study has used a nested cohort design to investigate risk factors for breast cancer and other chronic diseases, leveraging its extensive prospective database⁵.

Cohort studies can be subject to biases that affect the validity of their results, including the following:

• Selection bias: This occurs if the comparison groups are not representative or if there are systematic differences in the inclusion of participants¹.

• Information bias: This arises when the measurement of exposure or outcome is inaccurate. It is especially relevant in retrospective studies that rely on previous records¹.

• Confounding: This happens when an external variable is associated with both the exposure and the outcome, thus biasing the observed association¹.

Cohort studies have been fundamental in epidemiology and clinical research, as they provide crucial information on risk factors and preventive strategies in public health. Their implementation requires rigorous design to minimize biases and maximize the validity of findings. The choice between a prospective, retrospective, or nested design depends on the study’s objectives, the availability of data, and the resources available. Despite their challenges, the impact of these studies on risk identification and the formulation of public health policies is undeniable¹.

• Explicitly state that it is a cohort study and specify whether it is prospective, retrospective, or ambispective.

• Justify the choice of design based on the research question.

• Describe the hypothesis and the expected direction of the association.

Example of phrasing: “A prospective cohort study was conducted to evaluate the relationship between benzodiazepine exposure and the risk of falls in older adults, with a five-year follow-up.”

• Inclusion and exclusion criteria: clearly define the criteria used to select participants.

• Recruitment methods: indicate whether participants were identified from medical records, surveys or population sampling.

• Define the source of the cohort: explain whether it comes from a hospital database, epidemiological database or pre-existing cohort.

Example: “Patients aged ≥65 years treated in healthcare centers affiliated with the National Health System between 2015 and 2020 were included. Patients with prior history of fractures or neuromuscular disease were excluded.”

Definition of exposure and outcome

• Exposure measurement: describe how exposure was determined (clinical history, self-report, biomarkers, etc.).

• Outcome measurement: explain how the diagnosis or event of interest was established (radiological confirmation, laboratory tests, among others).

• Covariates and confounding: explain how confounding factors were addressed through matching, stratification, or statistical models.

Example: “The primary outcome was the incidence of severe falls, defined as those requiring hospitalization. Benzodiazepine exposure was defined as continuous use for ≥3 months according to the electronic medical record.”

STROBE recommends documenting strategies to minimize the following biases:

• Selection bias: use of propensity score matching or strict inclusion criteria.

• Information bias: application of standardized methods to measure exposure and outcome.

• Confounding: adjustment by multivariate models or sensitivity analysis.

Example: “to minimize confounding bias, the analysis was adjusted for age, sex, comorbidities, and polypharmacy using a Cox regression model.”

• Explain the calculation of the sample size based on the expected incidence of the outcome and the minimum detectable effect.

• Justify whether the sample is sufficient to detect statistically significant differences.

Example: “the sample size calculation determined that 8000 participants were required to detect a 20% difference in the risk of falling with a power of 80% and a significance level of 5%.”

• Specify the methods used to analyze the data.

• Indicate how missing data were handled.

• Include the measures of association used, such as risk ratio (RR), hazard ratio (HR) or odds ratio (OR).

Example: “a Cox regression model was used to estimate the adjusted hazard ratio, and a sensitivity analysis assessed the impact of missing data.”

The first step is to determine whether the study minimized biases that could affect internal validity.

Comparability of the groups:

• Did the groups start with the same risk of outcome?

• Were initial differences adjusted by statistical models?

• Were known confounding factors controlled for?

Example: “the benzodiazepine-exposed group had more comorbidities than the non-exposed group, so a propensity score adjustment was made.”

Evaluation of the outcome:

• Were the same methods used to assess outcome in both groups?

• Was there a risk of surveillance bias?

Example: “the outcome was determined by standardized hospital diagnoses, minimizing information bias.”

Complete follow-up:

• Was the loss rate less than 20%?

• Were there differences in follow-up between the groups?

Example: “the follow-up rate was 95% in both groups, which reduced the impact of loss-to-follow-up bias.”

The second step is to assess the magnitude and precision of the association between exposure and outcome.

Magnitude or strength of the association:

• Measures such as relative risk (RR), hazard ratio (HR) or odds ratio (OR) are reported.

• RR values close to 1 indicate a weak association, whereas values >2 suggest a stronger relationship.

Example: “prolonged use of benzodiazepines was associated with an increased risk of falls (adjusted HR: 1.52; 95% CI: 1.32-1.75).”

Estimation accuracy:

• It is evaluated by observing the confidence intervals (95% CI).

• A wide CI suggests uncertainty, while a narrow CI indicates greater precision.

Example: “the 95% CI did not include unity, indicating that the association is statistically significant”.

The third step is to determine whether the findings can be applied in medical practice.

Comparability with the clinical population:

• Are participants representative of patients in clinical practice?

Example: “the study included only hospitalized elderly adults, which limits the generalizability to outpatient populations.”

Follow-up duration:

• Was it enough to capture the outcome?

Example: “the mean follow-up was 6.2 years, which made it possible to evaluate long-term effects.”

Magnitude of absolute risk:

• The absolute risk is compared between groups.

Example: “the absolute risk of falls in benzodiazepine users was 12.3%, compared with 8.5% in the control group (number needed to treat [NNT]): 25).”