Cohort study reporting under STROBE has to show how long each exposure group was followed, who was lost and how often the outcome occurred. State incidence as a risk over a named period or a rate per person-years, align the start of follow-up with exposure assignment to avoid immortal time bias, and report losses, with reasons, by exposure group.
The STROBE checklist for cohort studies has 22 items, and starred items 8, 13, 14 and 15 are reported separately for exposed and unexposed groups. STROBE (2007) remains the current version; the EQUATOR Network record, updated in September 2026, lists no newer one. It does list an extension for prevalence and incidence studies, STROBE EPIC (2026); check it when estimating incidence is the main aim.
Related guides cover the rest: building Table 1 of baseline characteristics (items 14(a) and 14(b)), reporting confounding and bias in observational studies (items 16(a), 19 and 20) and reporting Kaplan-Meier survival analysis (censoring and survival curves).
Report incidence as a risk over a stated period or a rate per person-years
Incidence comes in two forms; name yours. The incidence proportion, or cumulative incidence, is the share of people at risk who develop the outcome within a stated period, such as 15% over three years. It is a risk and needs its period. The incidence rate is new events divided by person-time at risk, such as 6.0 per 100 person-years. A rate is not a percentage; always show its time unit.
A closed cohort, with membership fixed at the start, can estimate both. An open cohort, which people join and leave, estimates rates, as Box 1 of the STROBE explanation and elaboration paper sets out. Proportions give a risk ratio, rates a rate ratio and survival models a hazard ratio, so label each. When death or another competing event prevents the outcome, 1 minus the Kaplan-Meier estimate overstates risk; use a cumulative incidence function. For item 16(c), absolute risk over a meaningful period, see reporting effect sizes with confidence intervals.
Count person-time from entry to the first exit, not as people times years
Person-time is the total time participants spent at risk and under observation. Each clock starts at entry and stops at the first of four exits: the outcome, loss to follow-up, death, or the end of the study or observation window. The example below is invented to illustrate the structure; its numbers are not real data. Five adults are followed for incisional hernia for up to three years after open abdominal surgery.
| Participant | How follow-up ended | Months | Person-years |
|---|---|---|---|
| A | Hernia diagnosed | 18 | 1.50 |
| B | Lost to follow-up | 12 | 1.00 |
| C | Died | 9 | 0.75 |
| D | Study closed | 30 | 2.50 |
| E | Completed three years | 36 | 3.00 |
| Total | 105 | 8.75 |
One event in 8.75 person-years is 11.4 per 100 person-years; five people times three years would give a falsely low 6.7. Item 14(c) asks for follow-up time, such as the average and total, as the Kaplan-Meier survival guide also notes; the mean times the number of participants gives total person-years.
Give events, person-years and a rate for each exposure group
Item 15 asks for outcome events or summary measures over time, by exposure group. The explanation paper suggests a rate per person-year, or events in intervals if risk changes during follow-up. The invented cohort below compares 160 smokers with 240 non-smokers; confidence intervals (CIs) are 95%.
| Row | Smokers | Non-smokers |
|---|---|---|
| Enrolled | 160 | 240 |
| Hernia (person-years before diagnosis) | 24 (24) | 18 (27) |
| Lost to follow-up (person-years before loss) | 16 (16) | 12 (18) |
| Completed three years (person-years) | 120 (360) | 210 (630) |
| Total person-years | 400 | 675 |
| Mean follow-up per participant, years | 2.50 | 2.81 |
| Rate per 100 person-years (exact Poisson CI) | 6.0 (3.8 to 8.9) | 2.7 (1.6 to 4.2) |
The rate ratio is (24/400) ÷ (18/675) = 2.25. By the log method, the standard error of its logarithm is √(1/24 + 1/18) = 0.312, giving a CI of 1.22 to 4.15 and p = 0.009. The rate difference is 3.3 per 100 person-years (Wald CI 0.6 to 6.0). Name each interval method; others give slightly different limits.
Align time zero with exposure to avoid immortal time bias
Immortal time is "a span of cohort follow-up during which, because of exposure definition, the outcome under study could not occur", in Suissa's 2008 definition. A patient classed as a drug user from discharge had to survive until the first prescription. If that waiting time is credited to the exposed group, the drug looks more protective than it is.
Hernán and colleagues (2016) trace the bias to a failure to align eligibility, exposure assignment and the start of follow-up. State time zero for each group and how later-starting exposure was classified. One established remedy treats exposure as time-dependent: a patient prescribed the drug at month 8 who has the event at month 20 contributes 8 unexposed and 12 exposed months. STROBE does not name immortal time; address it under item 9, bias.
The example below is invented to illustrate the structure; its numbers are not real data. Fifty patients start a drug with no true effect six months after discharge, and 150 never do.
| Analysis | Exposed: deaths / person-years | Unexposed: deaths / person-years | Rate ratio |
|---|---|---|---|
| Time-dependent: users' first six months counted as unexposed | 5 / 100 | 20 / 400 | 1.00 |
| Immortal time counted as exposed | 5 / 125 | 20 / 375 | 0.75 |
| Immortal time dropped | 5 / 100 | 20 / 375 | 0.94 |
Counting the 25 immortal person-years as exposed invents a 25% lower death rate, and dropping them still distorts the ratio.
Report the flow after inclusion separately for each exposure group
Item 13(a) asks how many people remained at every step: possibly qualifying, screened, meeting the criteria, enrolled, followed to the end and analysed. Item 13(b) asks for reasons at each stage; give both by exposure group, as item 13 is starred. Attrition before inclusion is covered in writing the Methods for a retrospective chart review.
The explanation paper notes that follow-up rates have no universally agreed definition, so say how yours was calculated. STROBE item 13(c) says to consider a flow diagram; the explanation paper sets no format and says it may include the number of outcome events. Neither STROBE nor its explanation paper sets an acceptable share lost.
Compare participants lost with those retained, in each exposure group
Item 12(d) asks how loss to follow-up was addressed. Beyond the censoring and losses by group covered in the Kaplan-Meier survival guide, three checks apply.
- Compare baseline characteristics. Report key prognostic factors for those lost and retained, by exposure group. The explanation paper's item 12(d) example compares baseline CD4 cell counts in patients lost and followed up.
- Compare the proportions lost. In the invented cohort, 16 of 160 smokers (10.0%) and 12 of 240 non-smokers (5.0%) were lost. Loss that differs by exposure and tracks outcome risk can cause selection bias.
- Show how far loss could move the estimate. Item 12(e) asks for sensitivity analyses. An extreme-case analysis gives the event to every lost participant in one group and none in the other.
| Three-year risk of hernia | Smokers | Non-smokers | Risk ratio |
|---|---|---|---|
| Complete cases only | 24/144 (16.7%) | 18/228 (7.9%) | 2.11 |
| Lost counted as hernia-free | 24/160 (15.0%) | 18/240 (7.5%) | 2.00 |
| Every lost smoker had a hernia | 40/160 (25.0%) | 18/240 (7.5%) | 3.33 |
| Every lost non-smoker had a hernia | 24/160 (15.0%) | 30/240 (12.5%) | 1.20 |
Every scenario points the same way, but the risk ratio ranges from 1.20 to 3.33, and none equals the rate ratio of 2.25. State the likely direction of bias from loss, as item 19 asks.
Follow exposed and unexposed participants with the same methods
Item 6(a) asks for follow-up methods, and the explanation paper adds whether they were the same for everyone. Detection bias, also called medical surveillance bias, is finding more outcomes in exposed people because they are examined more often. Item 8 asks whether assessment was comparable between groups. Say how outcomes were found, such as clinic visits or linkage to a disease registry, and whether assessors knew exposure status. Ascertainment and follow-up schedules in general are covered in the Clavien-Dindo surgical complications guide. For a matched cohort, item 6(b) asks for matching criteria and numbers exposed and unexposed.
Cohort study reporting mapped to selected STROBE items, with template sentences
| STROBE item | What the cohort report states | Section |
|---|---|---|
| 5 | Dates of recruitment, exposure, follow-up and data collection | Methods |
| 6(a), 6(b) | Eligibility, sources and follow-up methods for each group; matching criteria and numbers, if matched | Methods |
| 8 | How outcomes were assessed; comparability between groups | Methods |
| 9 | Efforts to address bias, including time zero for each group and how later-starting exposure was classified | Methods |
| 12(d), 12(e), 17 | Handling of loss to follow-up and sensitivity analyses, such as extreme cases (12); their results (17) | Methods and Results |
| 13(a) to 13(c) | Numbers at each stage and reasons for non-participation, by group; consider a flow diagram | Results |
| 14(a), 14(b) | Characteristics and missing values by group (Table 1 guide) | Results |
| 14(c) | Average and total follow-up time, by group | Results |
| 15 | Events, person-years and rate, or risk over a stated period, for each group | Results |
| 16(a), 16(c) | Unadjusted and adjusted estimates (confounding guide); absolute risk over a meaningful period | Results |
| 19 | Likely direction and size of bias from loss to follow-up and from unequal outcome detection | Discussion |
Methods. Follow-up started at [time zero] in both groups and ended at the first of [outcome], loss to follow-up, death or [end date]. Exposure that began after time zero was classified as [method]. Outcomes were ascertained by [method] in both groups, by assessors [aware or unaware] of exposure status. Lost participants contributed person-time until last contact; extreme-case sensitivity analyses assigned [outcome] to every lost participant in one group and none in the other, then the reverse.
Results. Follow-up was completed by [n/N] ([%]) exposed and [n/N] ([%]) unexposed participants; the [n] exposed and [n] unexposed losses were due to [reasons]. Mean follow-up was [x] years in exposed and [y] years in unexposed participants. [Outcome] occurred in [n] exposed participants during [n] person-years ([rate] per 100 person-years; 95% CI [lower] to [upper]) and [n] unexposed participants during [n] person-years ([rate]; [lower] to [upper]): rate ratio [x] (95% CI [lower] to [upper], [method]). Those lost [resembled or differed from] those retained in [characteristics]. Under extreme-case assumptions, the risk ratio ranged from [x] to [y].
At Directive Publications, the author guidelines ask authors to match the manuscript to the relevant EQUATOR checklist: STROBE for an observational cohort. Review is double-blind, with at least two independent expert reviewers sought per research manuscript, and our reviewer guidelines list reporting completeness among the points a complete review usually considers. Our policy pages set no minimum follow-up, maximum loss or cohort flow-diagram requirement. To flag an error in this guide, report the problem to us.