Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data Analytics & AI · Cohort studies

Cohort Studies: Definition, Design Types and Measures of Effect

A cohort study identifies a group by exposure and follows it forward to outcome. This structure determines the principal properties of the design: it explains why cohort studies can estimate incidence, why person-time is the natural denominator, why loss to follow-up is the most important threat, and why the design is unsuitable for rare outcomes unless a further design is nested within it.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time9 minutes Sources30 peer-reviewed

The defining feature of a cohort study is the direction of the logic, not the direction of the calendar. The investigator defines the group, classifies exposure and then observes who develops the outcome, and this holds whether the data were collected before the study began or will be collected afterwards. A correct understanding of this distinction resolves most of the confusion around prospective, retrospective and ambidirectional cohorts, and it explains why a cohort study answers questions a case-control study cannot address.

Learning objectives

  • What makes a study a cohort study, and what the classification depends on
  • The difference between prospective, retrospective and ambidirectional cohorts
  • Which measures of effect a cohort design can estimate, and why person-time matters
  • Why the nested case-control variant exists and what it costs
  • The three threats that most compromise validity in cohort studies, in order of seriousness
  • Which reporting items a cohort study must address

Common misconceptions

  • “A prospective cohort is one that follows people forward and a retrospective one looks backward.” All cohort studies follow people forward from exposure to outcome. The prospective and retrospective labels describe whether the outcomes had already occurred when the study began, not the direction of the inference.
  • “A bigger cohort means a better study.” Precision comes from events, not from participants. A cohort of 25,802 that yields 74 cases of the outcome has 74 cases’ worth of information about that outcome.
  • “Cohort studies are observational, so they cannot show causation.” This statement is an oversimplification. Cohort studies cannot rely on randomisation to handle confounding, which is a specific limitation requiring specific remedies. Temporal order, dose-response and magnitude are all available to a well-designed cohort, and some causal questions have never been answered any other way.
  • “Loss to follow-up just reduces the sample size.” It reduces it non-randomly. People lost to follow-up differ systematically from those retained, usually in ways related to both exposure and outcome, which converts a sample-size problem into a bias problem.
  • “A cohort study is impractical for rare diseases.” As a primary design, it usually is. However, the nested case-control and case-cohort variants exist to make expensive measurement affordable within a cohort, and they are the standard solution.

Defining structure of a cohort study

A cohort study assembles a group of people who do not yet have the outcome of interest, classifies them by exposure, and follows them forward in time to observe who develops it 18. The comparison is between exposed and unexposed members of the same defined group, over a shared period of observation.

The etymology of the term reflects the structure of the design. A Roman cohort was a military unit of between 300 and 600 men, ten of which made up a legion — a defined body of people identified together and moving together. That is the sense intended: a cohort in epidemiology is a group whose membership is fixed by a definition at entry, which is what makes the denominator meaningful and what makes attrition a threat rather than an inconvenience.

Three requirements follow from the structure. The exposure must be ascertained before the outcome occurs; otherwise the temporal claim fails 18. The cohort must be free of the outcome at entry; otherwise the study addresses prognosis rather than incidence. Follow-up must also be comparable between exposure groups, because unequal observation produces unequal event counts for reasons unrelated to the exposure. See our page on epidemiological designs for how this compares with the alternatives.

Prospective, retrospective and ambidirectional

Cohort studies can be prospective, retrospective or ambidirectional, and the label refers to the timing of data collection relative to the study’s inception rather than to the logic of the inference 18.

Prospective. The cohort is assembled, exposure is measured, and outcomes are then observed as they occur. Advantages: exposure measurement can be designed for the question, confounders can be measured at the right time, and outcome ascertainment can be standardised. Disadvantages: duration and cost, and the inability to address a question that becomes relevant only after the measurements were taken.

Retrospective. Exposure and outcome have both already occurred when the study starts, and both are reconstructed from existing records. The logic is still exposure-to-outcome; only the investigator’s point of entry is later. Advantages: speed and the ability to study long induction periods. Disadvantages: exposure measurement is limited to what the records captured, and required confounders may not be available in the data.

Ambidirectional. Some follow-up has already happened and more is collected going forward. This design is common in practice and frequently not labelled as such.

The practical consequence is that the strength of a cohort study is determined almost entirely by the quality of its exposure measurement and the completeness of its outcome ascertainment, and prospective designs are superior on the first, whereas retrospective designs are sometimes superior on the second where a complete registry exists.

Cohort study directions, and what each gains and loses

TypeWhen the outcomes occurredGainsLoses
ProspectiveAfter the study beganExposure and confounders measured for the question; standardised outcome ascertainmentTime and cost; cannot answer questions that became interesting later
RetrospectiveBefore the study beganSpeed; feasible for long induction periods; uses existing complete recordsExposure measured as the records happened to capture it; needed confounders may not exist
AmbidirectionalPartly before, partly afterExisting follow-up plus designed prospective measurementTwo measurement regimes in one dataset, which has to be handled explicitly
Nested case-control within a cohortEitherExpensive exposure measured in a small fraction of the cohortPrecision limited by case count; requires a defensible control sampling scheme

Directions as conventionally defined. In all of them the inference runs from exposure to outcome; the label describes when the data were collected relative to the study’s inception.

Measures of effect estimable from a cohort study

Because the design starts from a defined group and observes who develops the outcome, the frequency of the outcome can be estimated directly. This is the central advantage of the design over a case-control study, which cannot do so 17,18.

Cumulative incidence — the proportion of the cohort developing the outcome over a fixed period. It is appropriate when follow-up is essentially complete and equal, which is rare in practice.

Incidence rate — events divided by person-time at risk. This measure should be used whenever participants are followed for unequal periods, which is almost always: staggered entry, withdrawal, death from other causes and a common administrative close-out date all produce unequal observation 18.

Risk ratio and rate ratio — the ratio of those measures between exposed and unexposed. These are directly interpretable, unlike an odds ratio, which is one reason cohort evidence is easier to communicate.

Risk difference and attributable measures — the absolute versions, on which decisions are based and which depend on the baseline risk in the population.

One structural point about precision: the number of outcome events governs it, not the size of the cohort 5. A large cohort with few events is imprecise, and additional recruitment cannot compensate for an outcome that does not occur; see our page on imprecision.

Two worked examples

A long follow-up for a mortality question. A general-practice cohort of 46,000 women followed for 25 years was used to estimate mortality associated with oral contraceptive use 21. The design features that made this possible merit description: a stable primary-care record system allowing exposure to be ascertained as it occurred rather than recalled, and a follow-up period long enough for mortality differences to emerge 21. No randomised trial has ever been funded to follow a cohort that far on this question, which clearly illustrates the purpose of cohort designs.

A short follow-up for a prognostic question. In a cohort of 138 patients with oral Kaposi’s sarcoma lesions, median survival in the exposed group was about a third that of the comparison group 18. Two features of this example are instructive. The cohort is small, but the effect is large and the outcome common within it, so the design is adequate; this example counters the belief that cohort studies must be large. The measure is also a survival comparison rather than a risk ratio, which is the natural choice when the question concerns time to an event that most participants will experience.

These two examples span most of the useful range of the design: 46,000 people for 25 years to detect a mortality difference, or 138 people over a much shorter period to characterise a threefold difference in survival. The size follows from the question, not from a convention.

46,000women in a general-practice cohort
25years of follow-up for the mortality estimate
138patients in the oral Kaposi’s sarcoma cohort
about one thirdmedian survival in the exposed group relative to the comparison group

Two cohort studies at opposite ends of the scale. The first followed 46,000 women for 25 years to estimate mortality associated with oral contraceptive use; the second compared median survival in 138 patients with oral Kaposi’s sarcoma lesions. Size follows from the question and the event rate, not from convention.

The nested case-control design

Cohort studies become expensive when the exposure of interest is costly to measure — a stored biomarker, an assay, a record abstraction requiring hours per participant. The nested case-control design addresses this by measuring the expensive exposure only in the cases and a sample of the cohort 22.

The scale of the saving is the central consideration. In a published example, a nested case-control analysis used 74 cases and 147 controls drawn from a parent cohort of 25,802 — exposure was measured in under 1% of the cohort 18,21. The question was answered, whereas measuring 25,802 exposures would not have been funded.

Two points should be understood before the design is used. Precision is set by the number of cases, so a nested design within a very large cohort is no more precise than the case count allows. In addition, the control sampling scheme determines what the odds ratio estimates: controls sampled from those still at risk at the time each case occurred give an odds ratio that estimates a rate ratio, while controls sampled at baseline or at the end of follow-up estimate something else 26,29. Our pages on case-control studies and their statistical analysis cover the sampling and the matched analysis it requires.

Principal threats to validity, in order of seriousness

Loss to follow-up, the first and most serious threat. The people who leave a cohort are not a random subset 16,27. They are more mobile, often less well, sometimes more affected by the exposure, and their departure correlates with the outcome. Twenty per cent attrition that is differential by exposure group can bias an estimate more than twenty per cent attrition that is random would reduce its precision. Attrition should be reported by exposure group, the characteristics of those lost should be compared with those retained, and a sensitivity analysis should be run under a plausible adverse assumption about their outcomes.

Confounding, second. Without randomisation, exposure groups differ in ways that also affect the outcome. The remedies — measuring confounders well, adjusting, stratifying, restricting, matching — all depend on having identified and measured the right variables, which retrospective designs often cannot. Residual confounding should be discussed in terms of its likely direction and magnitude, not acknowledged in a single sentence.

Misclassification, third. Exposure measured with error attenuates the estimate toward no effect when the error is unrelated to the outcome, and can bias it in either direction when it is not 16. Outcome ascertained by a method that differs between exposure groups — for example, exposed participants under closer surveillance — produces a spurious association that no adjustment removes 16.

Competing risks sit alongside these: a participant who dies of another cause cannot experience the outcome of interest, and where such deaths are common, a naive cumulative incidence overstates the risk of that outcome. Biostatistics support is where that pre-specification happens.

The follow-up and attrition paragraph of a cohort study (figures below are illustrative)

Attrition hidden

Participants were followed for a mean of 4.2 years. A total of 312 participants were lost to follow-up and were excluded from the analysis. Baseline characteristics were similar between exposure groups, and multivariable regression was used to adjust for potential confounders.

Attrition is reported as a single number and then dismissed. “Excluded from the analysis” tells the reader nothing about whether exclusion was related to exposure or outcome.

No breakdown by exposure group. Differential loss is the mechanism by which attrition becomes bias, and this paragraph makes it impossible to see.

Nothing about how the lost differed from the retained. Without it, a reader has no basis to judge the direction of the likely bias.

A mean follow-up only. With staggered entry a mean conceals the distribution, and a mean cannot support a rate.

No person-time. With unequal follow-up, proportions are not comparable between groups.

Competing risks unaddressed. If deaths from other causes are common, censoring them overstates the cumulative incidence of the outcome.

Attrition reported

Of 2,840 participants enrolled, 2,528 contributed follow-up data, giving 10,604 person-years of observation; the median follow-up was 4.4 years with an interquartile range of 2.9 to 5.8 years, and the distribution is given in Supplementary Figure 2. Loss to follow-up was 11.0% overall but was unequal between exposure groups: 148 of 1,362 exposed participants (10.9%) and 164 of 1,478 unexposed participants (11.1%). Participants lost to follow-up were younger than those retained, more likely to live outside the catchment area, and had a higher baseline symptom score; their characteristics are tabulated against those retained in Supplementary Table 3. Because loss was associated with baseline severity, the primary analysis is complemented by a sensitivity analysis in which all participants lost from the exposed group are assumed to have experienced the outcome and all those lost from the unexposed group are assumed not to have, which bounds the estimate in the direction least favourable to our conclusion. Incidence rates are reported per 1,000 person-years rather than as cumulative proportions, because entry was staggered over 26 months and follow-up ended at a common administrative close-out date. Deaths from other causes were treated as competing events using a pre-specified subdistribution approach rather than being censored, since they occurred in 4.3% of the cohort.

Loss is given overall and by exposure group, so a reader can see immediately whether it was differential.

The characteristics of those lost are compared with those retained, which is what allows the direction of any bias to be reasoned about.

A bounding sensitivity analysis is run in the direction least favourable to the authors’ own conclusion, which is the strongest available response to attrition.

The follow-up distribution is reported, not just a mean, and person-time is used as the denominator because entry was staggered.

Competing events are handled with a pre-specified approach and their frequency is stated, so the cumulative incidence means what it appears to mean.

Every claim can be checked against a named supplementary table, which is what distinguishes a reported cohort from an asserted one.

Differential loss to follow-up is the most damaging threat in cohort studies and the one most easily concealed. The second version allows a sceptical reader to quantify the worst case independently, which is more persuasive than any assurance that the groups were similar.

Reporting of cohort studies

STROBE has a cohort-specific checklist, and the items most often missed are predictable from the threats above 9,10.

The eligibility criteria and the sources and methods of selection should be given, together with the number of participants at each stage (eligible, examined, included, followed, analysed), so that a reader can reconstruct the attrition 10. Follow-up duration should be reported with its distribution rather than only a mean, and person-time should be reported where rates are given. The measurement of exposure, outcome and confounders should be described, with an explicit statement of whether outcome ascertainment could have differed by exposure status. The handling of missing data should be reported, and residual confounding should be discussed with its likely direction.

The sentence most cohort papers lack is a statement of how many participants were lost to follow-up in each exposure group and how they differed from those retained. A reader who cannot find it will assume the worst, often correctly. See the STROBE cohort checklist and STROBE overview.

Key points

Key points

A cohort study defines a group, classifies exposure, and follows forward to outcome — and that structure is what lets it estimate incidence, risk ratios and rate ratios directly, which a case-control study cannot. Prospective, retrospective and ambidirectional describe when the data were collected relative to the study’s inception, not the direction of the inference, which always runs exposure to outcome. Person-time should be used whenever follow-up is unequal, which is almost always the case. Precision comes from events, not participants: a parent cohort of 25,802 yielding 74 cases carries 74 cases’ worth of information, which is also why the nested case-control variant exists and why measuring exposure in under 1% of a cohort can be the right design. Size follows from the question — 46,000 women followed for 25 years for a mortality difference, or 138 patients for a threefold difference in median survival. Attrition should be reported by exposure group with the characteristics of those lost, because differential loss to follow-up is the most damaging threat and the one most easily concealed.

Application to your study

Application to your study

  1. The cohort should be confirmed to be free of the outcome at entry. If some members already have the outcome, the study addresses prognosis, and the measures and interpretation change.
  2. Person-time should be used as the denominator unless follow-up is genuinely equal. Staggered entry, withdrawal and a common close-out date all make proportions non-comparable.
  3. The study should be powered on expected events, not on cohort size. It should then be established whether the event count will reach a level at which the interval can distinguish a worthwhile effect from no effect.
  4. Attrition should be reported by exposure group, with a comparison of participants lost and those retained. This is the most informative paragraph in a cohort paper and the one most often omitted.
  5. A bounding sensitivity analysis of attrition should be performed. The least favourable plausible outcome should be assumed for those lost, and the resulting estimate should be reported.
  6. The competing-risks approach should be pre-specified. Where death from other causes is common, censoring it overstates the cumulative incidence of the outcome of interest.
  7. Where exposure measurement is expensive, a nested design should be considered. A nested case-control or case-cohort design can answer the question with exposure measured in a small fraction of the cohort.
  8. Reporting should follow the STROBE cohort checklist. Particular attention should be given to selection, participant numbers at each stage, follow-up distribution, missing data and the direction of residual confounding.

Frequently asked questions

Whether the outcomes had already occurred when the study began. In a prospective cohort the investigator assembles the group, measures exposure, and then waits for outcomes. In a retrospective cohort both exposure and outcome have already happened and are reconstructed from existing records. The inference in both runs from exposure to outcome — the label describes the investigator’s timing, not the logic. Prospective designs give better exposure and confounder measurement because measurement can be designed for the question; retrospective designs are faster, cheaper, and sometimes better for long induction periods, but the investigator is limited to the variables the records captured, which frequently exclude the confounder most needed.

The frequency of the outcome, and therefore everything derived from it. Because a cohort study starts from a defined group and observes who develops the outcome, it can estimate cumulative incidence and incidence rates, and hence risk ratios, rate ratios and absolute risk differences. A case-control study samples on outcome, so it has no denominator and estimates an odds ratio whose interpretation depends on how controls were sampled. The practical consequence is that cohort evidence is easier to use in decisions, since absolute risks and risk differences are what clinicians and payers need, and a risk ratio is more directly interpretable than an odds ratio.

Large enough to accumulate sufficient outcome events, which depends entirely on the incidence rate and the effect size to be detected. The size of the cohort is not the operative quantity: a published nested case-control example drew 74 cases from a parent cohort of 25,802, and 74 cases is what determined its precision. The range in practice is wide and question-driven. A general-practice cohort of 46,000 women was followed for 25 years to estimate mortality associated with oral contraceptive use; at the other extreme, a cohort of 138 patients with oral Kaposi’s sarcoma lesions was sufficient to show median survival in the exposed group at about a third of the comparison group, because the effect was large and the outcome common.

Because it is not random. People who leave a cohort differ systematically from those who stay — more mobile, often less well, sometimes more affected by the exposure — and those differences typically relate to both the exposure and the outcome, which is the definition of a confounded comparison. Differential loss between exposure groups is worse than equal loss, and modest differential loss can bias an estimate more than substantial random loss would blunt it. The appropriate response is not to deny it but to quantify the worst case: attrition should be reported by exposure group, the characteristics of those lost compared with those retained, and a bounding sensitivity analysis run in the direction least favourable to the study’s conclusion.

A nested case-control study is a case-control study conducted inside a defined cohort, with controls sampled from that cohort, and the reason to use one is cost. Where the exposure of interest requires an expensive assay, a stored sample or hours of record abstraction, measuring it in the whole cohort may be unaffordable while measuring it in the cases plus a modest sample of controls is not. In one published example, exposure was measured in 74 cases and 147 controls from a parent cohort of 25,802 — under 1% of the cohort. The costs are that precision is governed by the case count and that the control sampling scheme determines what the resulting odds ratio estimates.

An incidence rate with a person-time denominator should be used whenever follow-up differs between participants, which covers nearly all real cohorts: recruitment spread over months or years, withdrawals, deaths from other causes, and an administrative close-out date shared by everyone regardless of when they entered. Cumulative incidence is appropriate when follow-up is essentially complete and of equal length for everyone, which is uncommon outside short fixed-period studies. A related decision is how to treat deaths from other causes: if they are common, censoring them inflates the apparent cumulative incidence of the outcome of interest, and a pre-specified competing-risks approach is needed rather than an analytic choice made later.

The term is borrowed from the Roman army, where a cohort was a unit of around 300 to 600 men and ten cohorts made a legion. The borrowed sense is precise: a body of people identified together at a defined point and then moving forward together through time. This is what the design requires, and the image is useful in study design because it clarifies why attrition is so damaging. Once the group no longer moves together, because some members have been lost differentially, the group under analysis is no longer the defined cohort, and the denominator no longer means what it meant at entry.

References

  1. Schauberger G, Klug SJ, Berger M. Random forests for the analysis of matched case-control studies. BMC Bioinformatics. 2024;25(1):253. doi:10.1186/s12859-024-05877-5 read 9 October 2026. doi.org/10.1186/s12859-024-05877-5
  2. Zeng L, Brignardello-Petersen R, Hultcrantz M, Mustafa RA, Murad MH, Iorio A, et al. GRADE Guidance 34: update on rating imprecision using a minimally contextualized approach. Journal of Clinical Epidemiology. 2022;150:216-224. doi:10.1016/j.jclinepi.2022.07.014 read 9 October 2026. doi.org/10.1016/j.jclinepi.2022.07.014
  3. Schunemann HJ, Neumann I, Hultcrantz M, Brignardello-Petersen R, Zeng L, Murad MH, et al. GRADE guidance 35: update on rating imprecision for assessing contextualized certainty of evidence and making decisions. Journal of Clinical Epidemiology. 2022;150:225-242. doi:10.1016/j.jclinepi.2022.07.015 read 9 October 2026. doi.org/10.1016/j.jclinepi.2022.07.015
  4. Wan F, Colditz GA, Sutcliffe S. Matched Versus Unmatched Analysis of Matched Case-Control Studies. American Journal of Epidemiology. 2021;190(9):1859-1866. doi:10.1093/aje/kwab056 read 9 October 2026. doi.org/10.1093/aje/kwab056
  5. Guyatt GH, Oxman AD, Kunz R, Brozek J, Alonso-Coello P, Rind D, et al. GRADE guidelines 6. Rating the quality of evidence - imprecision. Journal of Clinical Epidemiology. 2011;64(12):1283-1293. doi:10.1016/j.jclinepi.2011.01.012 read 9 October 2026. doi.org/10.1016/j.jclinepi.2011.01.012
  6. Guyatt G, Oxman AD, Akl EA, Kunz R, Vist G, Brozek J, et al. GRADE guidelines: 1. Introduction - GRADE evidence profiles and summary of findings tables. Journal of Clinical Epidemiology. 2011;64(4):383-394. doi:10.1016/j.jclinepi.2010.04.026 read 9 October 2026. doi.org/10.1016/j.jclinepi.2010.04.026
  7. Balshem H, Helfand M, Schunemann HJ, Oxman AD, Kunz R, Brozek J, et al. GRADE guidelines: 3. Rating the quality of evidence. Journal of Clinical Epidemiology. 2011;64(4):401-406. doi:10.1016/j.jclinepi.2010.07.015 read 9 October 2026. doi.org/10.1016/j.jclinepi.2010.07.015
  8. Schulz KF, Altman DG, Moher D. CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials. Trials. 2010;11(1):32. doi:10.1186/1745-6215-11-32 read 9 October 2026. doi.org/10.1186/1745-6215-11-32
  9. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP; STROBE Initiative. Strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. BMJ. 2007;335(7624):806-808. doi:10.1136/bmj.39335.541782.AD read 9 October 2026. doi.org/10.1136/bmj.39335.541782.AD
  10. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Epidemiology. 2007;18(6):800-804. doi:10.1097/EDE.0b013e3181577654 read 9 October 2026. doi.org/10.1097/EDE.0b013e3181577654
  11. Ridker PM, Cook NR, Lee I-M, Gordon D, Walter JM, Buring JE, et al. A randomized trial of low-dose aspirin in the primary prevention of cardiovascular disease in women. New England Journal of Medicine. 2005;352(13):1293-1304. doi:10.1056/NEJMoa050613 read 9 October 2026. doi.org/10.1056/NEJMoa050613
  12. Grimes DA, Schulz KF. Compared to what? Finding controls for case-control studies. The Lancet. 2005;365(9468):1429-1433. doi:10.1016/S0140-6736(05)66379-9 read 9 October 2026. doi.org/10.1016/S0140-6736(05)66379-9
  13. Ioannidis JP. Contradicted and initially stronger effects in highly cited clinical research. JAMA. 2005;294(2):218-228. doi:10.1001/jama.294.2.218 read 9 October 2026. doi.org/10.1001/jama.294.2.218
  14. Lawlor DA, Davey Smith G, Ebrahim S. Commentary: the hormone replacement-coronary heart disease conundrum: is this the death of observational epidemiology? International Journal of Epidemiology. 2004;33(3):464-467. doi:10.1093/ije/dyh124 read 9 October 2026. doi.org/10.1093/ije/dyh124
  15. Doll R, Peto R, Boreham J, Sutherland I. Mortality in relation to smoking: 50 years' observations on male British doctors. BMJ. 2004;328(7455):1519. doi:10.1136/bmj.38142.554479.AE read 9 October 2026. doi.org/10.1136/bmj.38142.554479.AE
  16. Grimes DA, Schulz KF. Bias and causal associations in observational research. The Lancet. 2002;359:248-252. read 9 October 2026. qmplus.qmul.ac.uk
  17. Schulz KF, Grimes DA. Case-control studies: research in reverse. The Lancet. 2002;359(9304):431-434. doi:10.1016/S0140-6736(02)07605-5 read 9 October 2026. doi.org/10.1016/S0140-6736(02)07605-5
  18. Grimes DA, Schulz KF. Cohort studies: marching towards outcomes. The Lancet. 2002;359:341-345. doi:10.1016/S0140-6736(02)07500-1 read 9 October 2026. doi.org/10.1016/S0140-6736(02)07500-1
  19. King G, Zeng L. Estimating risk and rate levels, ratios and differences in case-control studies. Statistics in Medicine. 2002;21(10):1409-1427. doi:10.1002/sim.1032 read 9 October 2026. doi.org/10.1002/sim.1032
  20. Zondervan KT, Cardon LR, Kennedy SH. What makes a good case-control study? Design issues for complex traits such as endometriosis. Human Reproduction. 2002;17:1415-1423. read 9 October 2026. ebm-netzwerk.de
  21. Beral V, Hermon C, Kay C, Hannaford P, Darby S, Reeves G. Mortality associated with oral contraceptive use: 25 year follow up of cohort of 46 000 women from Royal College of General Practitioners' oral contraception study. BMJ. 1999;318:96-100. read 9 October 2026. qmplus.qmul.ac.uk
  22. Rothman N, Cantor KP, Blair A, Bush D, Brock JW, Helzlsouer K, et al. A nested case-control study of non-Hodgkin lymphoma and serum organochlorine residues. The Lancet. 1997;350:240-244. read 9 October 2026. qmplus.qmul.ac.uk
  23. Bland JM, Altman DG. Regression towards the mean. BMJ. 1994;308:1499. read 9 October 2026. qmplus.qmul.ac.uk
  24. Wacholder S, McLaughlin JK, Silverman DT, Mandel JS. Selection of controls in case-control studies, I: principles. American Journal of Epidemiology. 1992;135(9):1019-1028. read 9 October 2026. med.libretexts.org
  25. Wacholder S, Silverman DT, McLaughlin JK, Mandel JS. Selection of controls in case-control studies, II: types of controls. American Journal of Epidemiology. 1992;135(9):1029-1041. read 9 October 2026. med.libretexts.org
  26. Rodrigues L, Kirkwood BR. Case-control designs in the study of common diseases: updates on the demise of the rare disease assumption and the choice of sampling scheme for controls. International Journal of Epidemiology. 1990;19(1):205-213. doi:10.1093/ije/19.1.205 read 9 October 2026. doi.org/10.1093/ije/19.1.205
  27. Sackett DL. Bias in analytic research. Journal of Chronic Diseases. 1979;32:51-63. read 9 October 2026. qmplus.qmul.ac.uk
  28. Prentice RL, Pyke R. Logistic disease incidence models and case-control studies. Biometrika. 1979;66:403-411. read 9 October 2026. kiwix.lema.org
  29. Miettinen O. Estimability and estimation in case-referent studies. American Journal of Epidemiology. 1976;103(2):226-235. doi:10.1093/oxfordjournals.aje.a112220 read 9 October 2026. doi.org/10.1093/oxfordjournals.aje.a112220
  30. Doll R, Hill AB. Smoking and carcinoma of the lung; preliminary report. British Medical Journal. 1950;2(4682):739-748. doi:10.1136/bmj.2.4682.739 read 9 October 2026. doi.org/10.1136/bmj.2.4682.739

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Cohort Studies: Definition, Design Types and Measures of Effect. Pubrica. https://pubrica.com/academy/concepts-definitions/what-is-cohort-studies/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Cohort study design and reporting stub

For use when designing a cohort study or checking one before submission. The attrition block is the section reviewers read first and the one most papers leave incomplete.

COHORT STUDY DESIGN AND REPORTING STUB

1. THE COHORT
   Question, in one sentence: __________________________________
   Source population: _________________________________________
   Eligibility criteria: ______________________________________
   Entry date definition: _____________________________________
   [ ] Cohort confirmed FREE OF THE OUTCOME at entry
       (if not, this is a prognostic study - say so)
   Direction: [ ] prospective  [ ] retrospective
              [ ] ambidirectional

2. EXPOSURE
   Definition: ________________________________________________
   How measured: ______________________________________________
   Measured BEFORE outcome occurred?  [ ] yes
   Exposure categories or scale: ______________________________
   Expected misclassification, and its likely direction: ______

3. OUTCOME
   Definition: ________________________________________________
   Ascertainment method: ______________________________________
   [ ] Ascertainment is identical for exposed and unexposed
       If not, what differs and what bias that creates: _______
   Adjudicated?  [ ] yes - by whom, blinded? ___________________

4. FOLLOW-UP
   Start of follow-up for each participant: ___________________
   Exit criteria: [ ] outcome  [ ] death  [ ] withdrawal
                  [ ] administrative close-out on ____________
   Expected follow-up duration and spread: ____________________
   Denominator: [ ] person-time  [ ] fixed-period proportion
     (person-time unless follow-up is genuinely equal)

5. CONFOUNDING
   Confounders identified: ____________________________________
   Measured?  all [ ]  some [ ] - which are missing: __________
   Strategy: [ ] adjustment [ ] stratification [ ] restriction
             [ ] matching  [ ] other: ___________________
   Likely DIRECTION of residual confounding: __________________

6. COMPETING EVENTS
   Competing events expected: _________________________________
   Expected frequency: ______%
   Approach, pre-specified: [ ] cause-specific
                            [ ] subdistribution  [ ] censoring
   Justification: _____________________________________________

7. PRECISION
   Expected number of OUTCOME EVENTS: ______
   (not cohort size - this is what sets precision)
   Detectable effect at that event count: ______
   If events will be too few, consider:
     [ ] longer follow-up  [ ] nested case-control
     [ ] case-cohort  [ ] different outcome definition

8. ATTRITION BLOCK  (complete at analysis; reviewers read this first)
   Enrolled: ______   Contributing follow-up: ______
   Person-time: ______
   Median follow-up and IQR: ______
   Lost to follow-up:
     Exposed:   ______ of ______ = ______%
     Unexposed: ______ of ______ = ______%
   [ ] Characteristics of lost vs retained tabulated
   How the lost differed: _____________________________________
   [ ] Bounding sensitivity analysis run, least favourable to
       our own conclusion.  Result: __________________________

9. REPORTING
   [ ] STROBE cohort checklist mapped
   [ ] Participant numbers given at every stage
   [ ] Follow-up distribution reported, not only a mean
   [ ] Missing data handling stated
   [ ] Residual confounding discussed WITH A DIRECTION

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica designs cohort studies around the event count rather than the headcount, which usually changes the follow-up plan and sometimes changes the design to a nested one. Pubrica writes the attrition analysis in full — loss by exposure group, a comparison of those lost with those retained, and a bounding sensitivity analysis in the direction least favourable to the authors — because that is the paragraph reviewers test hardest. Pubrica also maps the manuscript against the STROBE cohort items and flags where residual confounding is acknowledged without a stated direction.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Pubrica reviews cohort protocols and drafts and assesses whether the follow-up plan is likely to produce enough events to answer the research question.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.