American Journal of Ophthalmology
Volume 150, Issue 1 , Pages 3-5, July 2010

Cohort Studies: Design and Pitfalls

  • Shu E. Soh
    • ,
  • Seang Mei Saw

      Affiliations

    • Corresponding Author InformationInquiries to Seang Mei Saw, Department of Epidemiology and Public Health, Yong Loo Lin School of Medicine, National University of Singapore, 16 Medical Drive (MD 3), Singapore 117597, Republic of Singapore

Department of Epidemiology and Public Health, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Republic of Singapore

Accepted 16 March 2010.

Article Outline

 

Back to Article Outline

History 

Cohort studies often are viewed as the gold standard in observational epidemiology, and the first large cohort study was conducted as early as 1913. Wilhelm Weinberg reported in the book Die Kinder der Tuberkuloesen (Children of the Tuberculous) on a 20-year follow-up of 18 212 exposed children whose parent died of tuberculosis and 7574 unexposed children to compare mortality and fertility rate.1 This early cohort study was ambidirectional and comprised both retrospective data collection of the parents' death records to determine exposure and nonexposure of the child and prospective follow-up to determine the outcome.

This important early epidemiologic work set the stage for different types of cohort study designs. In a prospective cohort study, the investigator first determines the exposure in the population and follows up the defined population through time to determine the outcome. The design of the retrospective cohort study is similar, but historical data are used to ascertain exposure, follow-up, and outcome.2

Back to Article Outline

Measure of Disease Burden and Association 

During the course of a cohort study, new cases arise during that given period in the specified group of people. This incidence of outcome frequency can be measured as cumulative incidence or incidence rate. The Beaver Dam Eye Study, conducted since 1988, provided United States population-based estimates of a high 25% cumulative incidence of loss of vision in people aged 75 years or older over a 15-year period, raising the need to address this substantial public health concern.3 To address loss to follow-up in cohort studies, incidence rate is based on person-time, taking into account the varying follow-up period of each individual. The Blue Mountains Eye Study, comprising 3654 white adults 49 years of age and older, identified the person-specific 10-year incidence rate of early and late age-related maculopathy as 2.8% and 10.8%, respectively, after age standardization.4 The incidence rate in cohort studies derived from the exposed and nonexposed group can lead to the derivation of the relative risk directly. This direct measure of risk cannot be calculated from case-control studies, which do not include incidence data. In case-control studies, the odds ratio, which is related to probability, approximates relative risk when the disease is rare, but not when the disease is common.

Back to Article Outline

Temporality 

The temporal sequence between exposure and disease can be elucidated clearly and is an important feature of cohort studies to establish causal associations. Lead time needs to be considered for an exposure such as dietary factors to have an effect on the outcome because of the long incubation period (lag phase) with the prolonged exposure. Outcomes in the first few years of the cohort may not be regarded as such because lag time needs to be considered. Misleading results also can occur in cohort studies that evaluate screening or treatment strategies. An adjustment is needed for the advancement of diagnosis by screening to ensure a common baseline for the time-to-event analysis. This takes into account the time between diagnosis by screening and diagnosis, which normally would have occurred through the usual course of medical care. In the Beaver Dam Eye Study, the 5-, 10-, and 15-year follow-up examinations determined that cigarette smoking, alcohol consumption, and extended sunlight exposure contribute to the incidence of age-related macular degeneration (AMD), cataract, and visual impairment. Because the exposures status could change over time, multiple time point measurement on the exposure with semiquantitative exposure variables incorporating exposed time information were used. Smoking status defined by total pack-years smoked found that current smokers had an increased risk of early AMD (odds ratio, 1.47; 95% confidence interval [CI], 1.08 to 1.99; P = .01) and progression (odds ratio, 1.43; 95% CI, 1.05 to 1.94; P = .02) in the recent 15-year follow-up, after adjusting for age, sex, and baseline AMD severity.5 In heavy alcohol drinkers who consumed 4 servings or more of alcoholic beverages daily, late AMD was 6.94 times more likely to develop.6 Extended exposure to sunlight was found to be a risk factor for the development of both AMD and cataract.7, 8 In the Singapore Cohort study Of Risk factors for Myopia (SCORM) study, the risk factors related to myopia in a 10-year cohort of children with rapidly evolving refractive error were evaluated. Children with 2 myopic parents were found to be 1.55 times (P = .002) more likely to experience myopia compared to those with no myopic parents, whereas lower intelligence quotient score in the third tertile was found to have a higher relative risk of 1.50 (95% CI, 1.19 to 1.89) of myopia development than first tertile intelligence quotient score.9 In contrast, the temporality of the associations cannot be documented in cross-sectional studies because it is unclear whether the exposure to the factor preceded outcome.

Back to Article Outline

Multiple Outcomes 

The potential to study multiple outcomes in relation to given exposure is unique in cohort studies. The Blue Mountains Eye Study of 3654 white adults aged 49 years and older assessed the incidence and associated risk factors such as cigarette smoking and visual impairment, cataract, AMD, glaucoma, and diabetic and other vascular retinopathy over a 15-year period since 1992. Nuclear cataract was more likely to develop in those who had ever smoked (relative risk, 1.41; 95% CI, 1.09 to 1.83) compared with those who had never smoked. Nuclear cataract further developed in current smokers at a slightly younger mean age of 65.2 years than in nonsmokers at 67.5 years of age (P = .049).10 In addition, these current smokers were at a 4 times higher long-term risk of incident late AMD than never smokers.11 Higher mean intraocular pressure also was observed in current smokers (16.34 mm Hg) than in nonsmokers (16.04 mm Hg).12

Back to Article Outline

Biases 

Compared with case-control studies, there are fewer biases in cohort studies. Recall bias in the strict sense is less of a concern in a cohort study because the disease status is not known when exposures are recalled and thus there is no issue of differential recall leading to misclassification. Similarly, interviewer bias is minimized because the interviewer asks the subjects equally about exposure among those who are diseased and nondiseased because the outcome is still unknown.

Inconsistent quality of information can result in disparity in the assessment of outcome and may be different among the exposed and nonexposed, leading to observer bias. Interobserver variability between ophthalmologists at different levels of experience can lead to uninterpretable outcomes. Changes such as transition of staffs and investigators in addition to alteration of study protocols and methodology also will have an impact on the consistency of the study over time. Furthermore, the person who grades the cataract should be blinded to the exposure status to prevent differential misclassification of outcome. For example, in subjects who drink more alcohol, the assessor may overdiagnose cataracts, and for subjects who do not drink alcohol, the assessor may underdiagnose the disease. The Beaver Dam Eye Study,13 the Blue Mountains Eye Study,14 the Barbados Eye Study,15 the Melbourne Visual Impairment Project,16 and the Longitudinal Study of Cataract,17 all of which reported varying cataract incidence data in similar populations, highlighted the importance of objective, measurable, and comparable diagnoses.

One major concern in cohort studies is loss to follow-up when the subjects do not follow through the full length of the study, especially if the cohort is conducted over a long period. This will bias the results because subjects who are lost to follow-up often are different from those who choose to continue with the cohort with regard to the exposure–disease relationship. Association of dietary antioxidant intake and 5-year incidence of early AMD could not be established in the Blue Mountains Eye Study, because more than 35% of the baseline cohort was not followed up as a result of death, migration, and declining to participate.18 In this study, Flood et al analyzed that antioxidant intakes at baseline did not differ between those who were lost to follow-up and those who returned for 5-year examinations. However, data were available only from 69% of the cohort. Sample size calculation at the study design stage should allow for loss to follow-up due to death in a cohort study of the elderly, because the mortality rate ascends when the study progresses over time. Differential losses among the exposed and nonexposed group may severely impact the reliability and validity of the study. Furthermore, a prospective cohort design is relatively costly when the follow-up is long compared with case-control or cross-sectional studies for rare diseases or for diseases with long latency periods. Although the Blue Mountains Eye Study did not find an antioxidant–AMD association in the 5-year follow-up data, this association was documented in the 10-year follow-up.19 This example highlights the effect of lead time, and a very long follow-up is needed because of the lag phase to detect the effect of dietary factors.

Back to Article Outline

Statistical Analyses 

Survival analysis models the time to outcome and can be used to account for censoring as a result of incomplete follow-up. The Kaplan-Meier method with a log-rank test compares survival curves in 2 groups or more, whereas Cox proportional hazards regression modeling allows the calculation of the hazard ratio and adjustment for the effects of other covariates. A time-dependent risk factor may change over time and may weaken or strengthen the association with outcome measure. Serial measurements of this risk factor in relation to subsequent mortality are collected for time-to-death information. Generalized estimating equations approach and multivariate generalized linear mixed model allow the use of both eyes and account for correlations between the two eyes.

Back to Article Outline

Conclusions 

Because of the many strengths and weaknesses of the cohort study, the nested case-control study stems from the framework of a cohort study combining the advantages of both the cohort and the case-control study. A case-control study nested in a cohort study eliminates recall bias because the exposure data are accrued before the disease develops. If the outcome is rare with low incidence of the disease within the cohort, the nested case-control study design is more economical by performing the laboratory test on only the chosen cases and on controls after the disease has been established. The association between abnormalities of the retinal microvasculature structure and risk of death from ischemic heart disease and stroke was established in a nested case-control study conducted within the Beaver Dam Eye Study.20

The available evidence generated from cohort studies for eye diseases has been promising with the capacity to evaluate multiple outcomes with a measure of absolute risk in addition to a measure of association. The effort and expense to conduct cohort studies can be justified with these high levels of evidence. Continual follow-up of the current cohort studies should be encouraged and new studies should be put together to close any existing gaps in eye diseases. Investigating multiple exposures may be crucial in revealing the associations with multifactorial complex diseases. The design of future cohort studies should take into consideration the possible interactions between multiple exposures, which add another dimension to the cohort study design, and therefore a much larger sample size is needed to determine the associations.

Back to Article Outline

 

The authors indicate no financial support or financial conflict of interest. Both authors were involved in Design and conduct of study; Collection and management of the data; Analysis and interpretation of data; and Preparation of the manuscript.

Back to Article Outline

References 

  1. Morabia A, Guthold R. Wilhelm Weinberg's 1913 large retrospective cohort study: a rediscovery. Am J Epidemiol. 2007;165(7):727–733
  2. Gordis L. Epidemiology. 4th ed.. Philadelphia: Elsevier Saunders; 2009;
  3. Klein R, Klein BE, Lee KE, Cruickshanks KJ, Gangnon RE. Changes in visual acuity in a population over a 15-year period: the Beaver Dam Eye Study. Am J Ophthalmol. 2006;142(4):539–549
  4. Wang JJ, Rochtchina E, Lee AJ, et al. Ten-year incidence and progression of age-related maculopathy: the Blue Mountains Eye Study. Ophthalmology. 2007;114(1):92–98
  5. Klein R, Knudtson MD, Cruickshanks KJ, Klein BE. Further observations on the association between smoking and the long-term incidence and progression of age-related macular degeneration: the Beaver Dam Eye Study. Arch Ophthalmol. 2008;126(1):115–121
  6. Klein R, Klein BE, Tomany SC, Moss SE. Ten-year incidence of age-related maculopathy and smoking and drinking: the Beaver Dam Eye Study. Am J Epidemiol. 2002;156(7):589–598
  7. Cruickshanks KJ, Klein BE, Klein R. Ultraviolet light exposure and lens opacities: the Beaver Dam Eye Study. Am J Public Health. 1992;82(12):1658–1662
  8. Tomany SC, Cruickshanks KJ, Klein R, Klein BE, Knudtson MD. Sunlight and the 10-year incidence of age-related maculopathy: the Beaver Dam Eye Study. Arch Ophthalmol. 2004;122(5):750–757
  9. Saw SM, Shankar A, Tan SB, et al. A cohort study of incident myopia in Singaporean children. Invest Ophthalmol Vis Sci. 2006;47(5):1839–1844
  10. Tan JS, Wang JJ, Younan C, Cumming RG, Rochtchina E, Mitchell P. Smoking and the long-term incidence of cataract: the Blue Mountains Eye Study. Ophthalmic Epidemiol. 2008;15(3):155–161
  11. Tan JS, Mitchell P, Kifley A, Flood V, Smith W, Wang JJ. Smoking and the long-term incidence of age-related macular degeneration: the Blue Mountains Eye Study. Arch Ophthalmol. 2007;125(8):1089–1095
  12. Lee AJ, Rochtchina E, Wang JJ, Healey PR, Mitchell P. Does smoking affect intraocular pressure? (Findings from the Blue Mountains Eye Study). J Glaucoma. 2003;12(3):209–212
  13. Klein BE, Klein R, Lee KE, Gangnon RE. Incidence of age-related cataract over a 15-year interval the Beaver Dam Eye Study. Ophthalmology. 2008;115(3):477–482
  14. Kanthan GL, Wang JJ, Rochtchina E, et al. Ten-year incidence of age-related cataract and cataract surgery in an older Australian population (The Blue Mountains Eye Study). Ophthalmology. 2008;115(5):808–814 e1
  15. Leske MC, Wu SY, Nemesure B, Li X, Hennis A, Connell AM. Incidence and progression of lens opacities in the Barbados Eye Studies. Ophthalmology. 2000;107(7):1267–1273
  16. McCarty CA, Mukesh BN, Dimitrov PN, Taylor HR. Incidence and progression of cataract in the Melbourne Visual Impairment Project. Am J Ophthalmol. 2003;136(1):10–17
  17. Leske MC, Chylack LT, Wu SY, et al. Incidence and progression of nuclear opacities in the Longitudinal Study of Cataract. Ophthalmology. 1996;103(5):705–712
  18. Flood V, Smith W, Wang JJ, Manzi F, Webb K, Mitchell P. Dietary antioxidant intake and incidence of early age-related maculopathy: the Blue Mountains Eye Study. Ophthalmology. 2002;109(12):2272–2278
  19. Tan JS, Wang JJ, Flood V, Rochtchina E, Smith W, Mitchell P. Dietary antioxidants and the long-term incidence of age-related macular degeneration: the Blue Mountains Eye Study. Ophthalmology. 2008;115(2):334–341
  20. Witt N, Wong TY, Hughes AD, et al. Abnormalities of retinal microvascular structure and risk of mortality from ischemic heart disease and stroke. Hypertension. 2006;47(5):975–981

PII: S0002-9394(10)00222-9

doi:10.1016/j.ajo.2010.03.008

American Journal of Ophthalmology
Volume 150, Issue 1 , Pages 3-5, July 2010

Article Tools