Advertisement

Methodological issues in the study of rare exposures and rare outcomes in perinatal epidemiology

      Abstract

      Epidemiologic study of rare exposures and rare outcomes is complicated. For perinatal epidemiology, a prospective cohort study design is preferred for reproductive outcomes with incidences of at least 5%. Case-control studies are generally used to investigate rare outcomes with incidences from 5 per 1000 to less than 1 per 1000. However, obtaining data on large numbers of rare cases can be difficult. The use of nested case-control designs or nested prospective studies may be particularly useful in perinatal epidemiology. (J Allergy Clin Immunol 1999;103:S354-5.)
      The study of rare diseases poses difficult methodological problems in epidemiology. The investigative process becomes more complicated when exposures that are possibly related to a health outcome are rare.
      • Bracken MB
      Musings on the edge of epidemiology.
      The epidemiologic method is most severely challenged in situations in which both disease and exposure occur at low prevalences—a condition often found in studies of maternal drug exposure and reproductive outcomes.
      • Bracken MB
      Methodologic issues in the epidemiologic investigation of drug-induced congenital malformations.
      Ideally, an epidemiologist chooses to investigate a problem prospectively. This approach avoids many potential biases (eg, patient recall, interviewer bias, or selection bias) and permits the precise calculation of relative risk.
      • Breslow NE
      • Day NE
      Statistical methods in cancer research. II. The design and analysis of cohort studies.
      The prospective design allows for a more accurate measure of confounding factors and may enable the researcher to perform repeated measures of exposure throughout a pregnancy to investigate “windows” when exposure may be especially problematic.
      • Bracken MB
      • Leaderer B
      • Belanger K.
      Role of biologic markers in epidemiologic studies of prenatal drug exposures: issues in study design.
      Additionally, biomarkers may be used to quantify maternal and fetal drug exposure. Use of the prospective design is ideally suited to investigate maternal drug exposure and reproductive outcome because the effect of drug use during pregnancy can be studied over a period of 9 months. Perinatal epidemiologists typically use the prospective cohort study design when the reproductive outcome has an incidence of about 5% or more. Examples of such outcomes include spontaneous abortion, fetal growth retardation, and low birth weight. When exposures are reasonably prevalent, occurring in approximately 15% or more of the cohort, feasibly sized studies of about 3000 women are possible.
      A case-control design is preferred for studies investigating rare outcomes, such as specific birth defects, which range in incidence from about 5 per 1000 to less than 1 per 1000. Even when maternal exposure is uncommon (eg, 5%), the case-control study design would allow for detection of a 50% increase in risk (relative risk = 1.5) with 1300 case patients and 2600 control subjects (α = .05, 1-β = .80). However, obtaining data on 1300 cases of a rare condition requires considerable resources. If the prevalence of maternal exposure is 2%, as it appears to be for inhaled bronchodilator use during pregnancy,

      Belanger K, Bracken MB, Beckett WS, et al. Prevalence of asthma medication use during pregnancy: surveillance results from 13,000 consecutive pregnancies. Submitted for publication.

      3100 cases of the rare condition would be required for a study. If one type of inhaler was of interest and the prevalence of exposure was 1%, 6100 cases are required.
      To obtain data on such large numbers of rare cases, nested case-control designs may be used. In this design, case-patients may be obtained from large cohorts, such as state-wide birth defect registries (eg, those found in the Nordic countries) and other large networks. Control subject selection is not straightforward in these designs, although population-based control subjects from the same source population that produced the case-patients (ie, the preferred method) can be derived in some circumstances.
      • Breslow NE
      • Day NE
      Statistical methods in cancer research. I. The analysis of case-control studies.
      In order to investigate problems in which both the maternal exposure and reproductive outcome are relatively rare, we have designed nested prospective studies.
      • Bracken MB
      • Leaderer B
      • Belanger K.
      Role of biologic markers in epidemiologic studies of prenatal drug exposures: issues in study design.
      • Doucette JT
      • Bracken MB
      Possible role of asthma in the risk of preterm labor and delivery.
      This design retains advantages of a prospective study while allowing the investigator to minimize the size of the cohort through selection of appropriate subpopulations. For example, a traditional cohort study of the effect of inhaled bronchodilator use (prevalence, 2%) designed to detect a 50% increase in intrauterine growth retardation (prevalence, 5%) would require a cohort of almost 40,000 mothers, because the great majority of subjects would yield no information. That is, 98% of the mothers would not have exposure to inhaled bronchodilators and 95% would not have newborns with intrauterine growth retardation. Use of the nested prospective study design substantially reduces the size of the cohort necessary to detect the same level of increased risk. In the nested prospective study, large numbers of women are screened by using a relatively simple approach. From this cohort, subpopulations of exposed and unexposed subjects are selected for more detailed study. This design is particularly useful when intensive interventions, such as monitoring protocols of the collection of biomarkers, are required. A cohort of 1300 inhaler users and 2600 unexposed subjects would detect a 50% increase in the incidence of intrauterine growth retardation. To construct this cohort, 65,000 women would need to be screened initially. Large cohorts for screening may be obtained from large health maintenance organizations, Medicaid, or other special databases.

      References

        • Bracken MB
        Musings on the edge of epidemiology.
        Epidemiology. 1997; 8: 337-339
        • Bracken MB
        Methodologic issues in the epidemiologic investigation of drug-induced congenital malformations.
        in: Perinatal epidemiology. Oxford University Press, New York1984: 423-449
        • Breslow NE
        • Day NE
        Statistical methods in cancer research. II. The design and analysis of cohort studies.
        International Agency for Research on Cancer, Lyon, France1987
        • Bracken MB
        • Leaderer B
        • Belanger K.
        Role of biologic markers in epidemiologic studies of prenatal drug exposures: issues in study design.
        NIDA Res Monogr. 1992; 117: 41-60
      1. Belanger K, Bracken MB, Beckett WS, et al. Prevalence of asthma medication use during pregnancy: surveillance results from 13,000 consecutive pregnancies. Submitted for publication.

        • Breslow NE
        • Day NE
        Statistical methods in cancer research. I. The analysis of case-control studies.
        International Agency for research on Cancer, Lyon, France1987
        • Doucette JT
        • Bracken MB
        Possible role of asthma in the risk of preterm labor and delivery.
        Epidemiology. 1993; 4: 143-150