Original article
The design of prospective epidemiological studies: More subjects or better measurements?

https://doi.org/10.1016/0895-4356(93)90120-PGet rights and content

Abstract

Prospective epidemiological studies which seek to relate potential risk factors to the risk of disease are subject to appreciable biases which are often unrecognized. The inability to precisely measure subjects' true values of the risk factors under consideration tends to result in bias towards unity in the univariate relative risks associated with them—the more imprecisely a risk factor is measured, the greater the bias. When correlated risk factors are measured with different degrees of imprecision the adjusted relative risk associated with them can be biased towards or away from unity. When designing a new prospective study cost considerations usually limit the total number of subject-evaluations that are available. The usual design approach is to maximize the study size and evaluate each subject on one occasion only. An alternative approach involves recruitment of a smaller number of subjects so that each can be evaluated on more than one occasion, thus resulting in a more precise measure of subjects' risk factor values and hence less bias in the relative risk estimates. In this paper we use a simulation approach to show that under conditions that prevail for most major prospective epidemiological studies the latter approach is actually more likely to produce accurate relative risk estimates. This emphasizes the importance of bias due to exposure measurement imprecision and suggests that attempts to anticipate and control it be given at least as high a priority as that given to sample size assessment in the design of epidemiological studies.

References (18)

There are more references available in the full text version of this article.

Cited by (58)

  • Alcohol, microbiome, life style influence alcohol and non-alcoholic organ damage

    2017, Experimental and Molecular Pathology
    Citation Excerpt :

    Despite best efforts to improve design and analysis of observational studies, some correctly stated that “Proof is impossible in epidemiology” (Conner, 2016). This is primarily due to the limited number of confounders measured in epidemiological studies, and the inevitable measurement errors in assessing both the exposure and the potential confounders (Phillips and Smith, 1991, 1993). The problem is more accentuated in alcohol epidemiological studies because all these studies rely on self-reporting to determine the amount and type of alcoholic beverage consumed, which inevitably introduces recall bias (Klatsky et al., 2014).

  • A review of five cardiology journals found that observer variability of measured variables was infrequently reported

    2008, Journal of Clinical Epidemiology
    Citation Excerpt :

    The benefits of improving reliability of measurements have been described by Perkins et al., who showed that improving the reliability coefficient leads to a significant decrease in sample size requirements, and is both time and cost-effective [39]. Exposure measurement imprecision leads to biased estimates of relative risk in prospective epidemiologic studies, and it was recommended that controlling measurement error was at least as high a priority as sample size assessment [40]. Our study has limitations.

  • Change in incidence and aetiology of intracerebral haemorrhage in Oxfordshire, UK, between 1981 and 2006: a population-based study

    2007, Lancet Neurology
    Citation Excerpt :

    Clearly chance effects cannot be excluded in all of these trends, given the small number of cases in each study period. Nevertheless, such detailed analysis is really only possible in smaller studies, and we believe that this study's design is preferable to that of larger studies with routinely collected data.36 Our results have implications for health service provision when predicting the future burden of intracerebral haemorrhage in the UK, as well as the demand for new acute treatments.2

  • Treating Individuals 4: Can meta-analysis help target interventions at individuals most likely to benefit?

    2005, Lancet
    Citation Excerpt :

    When adjusting for baseline troponin concentration, the sex difference was no longer evident.15 Although such findings are observational in nature, and can be subject to residual confounding and measurement error,40 in this case troponin might be the more important moderator of the effect of PGIs than sex per se. However, troponin data were only available for 35% of the entire population, and when restricting the analysis to this subgroup the unadjusted sex difference was no longer evident.

  • Systematic Reviews of Epidemiological Studies of Etiology and Prevalence

    2022, Systematic Reviews in Health Research: Meta-Analysis in Context: Third Edition
  • Correlation without a cause: An epidemiological odyssey

    2020, International Journal of Epidemiology
View all citing articles on Scopus
View full text