Which epidemiologic study design allows us to evaluate temporal relationship?

Tonne, C*; Beevers, S*; Kelly, F*; Jarup, L†; Wilkinson, P‡; Armstrong, B‡

*King's College London, London, United Kingdom; †Imperial College London, London, United Kingdom; and ‡London School of Hygiene and Tropical Medicine, London, United Kingdom.

Abstracts published in Epidemiology have been reviewed by the organizations of Epidemiology. Affliate Societies at whose meetings the abstracts have been accepted for presentation. These abstracts have not undergone review by the Editorial Board of Epidemiology.

Epidemiology 19(6):p S232, November 2008. | DOI: 10.1097/01.ede.0000340190.04996.43

• Free

ISEE-1199

Background:

There is increasing interest in estimating the effect of long-term changes in air pollution on changes in health outcomes across areas. Such study designs are less vulnerable to confounding than those evaluating only spatial contrasts in exposure. We developed a study design to estimate the association between changes in traffic-related air pollution and emergency hospitalizations over time across small areas within Greater London. This design was applied to data from a two-year period surrounding the introduction of the London Congestion Charge Scheme (CCS).

Methods:

Annual average NOX concentration was estimated for the year before (2002) and after (2003) the introduction of the CCS using an emission-dispersion model. Emergency hospitalizations for ischemic heart disease were aggregated at the ward level. Change in NOx was defined as the difference in ward average NOX for the post-pre CCS period. We used a binomial distribution to model the number of hospitalizations occurring in the post-CCS period out of total hospitalizations during the study years in each ward.

Results:

Across London, annual average NOX was 32 ppb in 2002 and 36 ppb in 2003. Differences in NOX (2003–2002) varied by an IQR of 1.8 ppb across wards. The IQR of the percentage of hospitalizations in 2003 was 11%. Deprivation, an important confounder in spatial studies, was predictive of change in NOX (P = 0.003), but not of ward level changes in hospitalizations (P = 0.24). On average, an increase in NOX was associated with a decrease in hospitalizations across wards: OR 0.93 95%CI (0.91, 0.95) per IQR difference in NOX.

Discussion:

This design is unlikely to be confounded by factors that vary only across areas (e.g. deprivation) or time (e.g. weather). However, it may be confounded by differential trends in ward level socio-demographics. The design is being applied to longer time periods as well as additional cardio-respiratory diagnostic categories.

Cross-sectional designs are mainly used to investigate the distribution of epigenetic variation across populations with different characteristics. A substantial number of epigenetic studies are cross-sectional. In the cross-sectional design, subjects' characteristics and epigenetic measures are collected at one time point or within a short time interval.

Results from the cross-sectional design are merely associations, and they do not imply causation. However, these caveats do not necessarily dismiss the utility of cross-sectional studies. For example, several CpGs have been found to be cross-sectionally associated with smoking status, i.e., being differently methylated between smokers and nonsmokers. If such associations are true (e.g., not being due to confounders), DNA methylation at these CpGs could potentially be used as biomarkers of smoking even if the causation of the association is unknown.

Cross-Sectional Study

Cross-sectional studies measure the cause (exposure) and the effect (disease) at the same point in time. They compare the rates of diseases or symptoms of an exposed group with an unexposed group. Strictly speaking, the exposure information is ascertained simultaneously with the disease information. In practice, such studies are usually more meaningful from an etiological or causal point of view if the exposure assessment reflects past exposures. Current information is often all that is available but may still be meaningful because of the correlation between current exposure and relevant past exposure.

Cross-sectional studies are widely used to study the health of groups of workers who are exposed to possible hazards but do not undergo regular surveillance. They are particularly suited to the study of subclinical parameters, such as blood biochemistry and hematological values. Cross-sectional studies are also relatively straightforward to conduct in comparison with prospective cohort studies and are generally simpler to interpret.

3.4.2.4 The Cross-Sectional Study

The cross-sectional study is an observational study that assesses exposure and the outcome at one specific point in time in a sample population. There is no prospective or retrospective follow-up. Both the RR and the OR can be calculated to describe the association between the exposure and the outcome. The incidence and prevalence, important descriptive measures, can also be calculated from a cross-sectional study.

The cross-sectional study cannot be used to infer causality because a temporal sequence cannot be established. Nevertheless, this type of study is used to generate descriptive statistics regarding the disease/outcome burden in a population, or to determine background exposure rates. All of which can be very useful, especially during the premarket phase of a product's life cycle. Fig. 3.3 below shows a schematic representation of a cross-sectional study.

Cross-Sectional Studies and Cluster Samples

Cross-sectional studies survey a sample of the population at one point in time and estimate the prevalence of a condition, an infection or a disease. The survey tool may be a questionnaire; a physical assessment (e.g. of weight, height or blood pressure); a blood test (e.g. for malaria parasites or HIV infection); or a diagnostic procedure (e.g. chest X-ray). Unlike prospective surveillance studies, cross-sectional surveys do not provide information on incidence, i.e. the number of new cases per population during a specific time period. However, they may provide other important public health information. For example, for diseases that induce life-long antibodies (e.g. HIV, hepatitis A and B), sero-epidemiological studies may show the age groups most affected and when done at different time periods and geographic locations, may indicate the effectiveness of prevention and control strategies. HIV seroprevalence in pregnant women is often used as an indicator of the burden of HIV/AIDS in a community. A major challenge of cross-sectional studies is ensuring that the sample selected and included in the survey is representative of the population of interest.

Cross-sectional studies

Cross-sectional studies are useful for characterizing the prevalence of a condition or risk factor in a particular population. Measurements are made at a specified time in a population, one patient at a time. There is no longitudinal component of the investigation. For example, all children in the emergency room during a certain month may have their blood drawn once to determine the incidence of antibodies indicating prior exposure to a particular virus. These studies can be inexpensive and easy to conduct.

However, cross-sectional studies may not detect certain data depending on the timing and method of data collection. A good example is the hidden mortality of CDH. In Norway mortality from CDH was thought to be 30%, as reported from hospital records. However, on a more comprehensive survey of neonatal deaths it was found that many infants died soon after birth from CDH and never presented to a major referral center. The true incidence of CDH was at least 1 in every 5000 live births and the actual mortality from CDH was closer to 66%.

Cross-sectional studies have other limitations. Although they are useful for characterizing the prevalence of a condition or a risk factor in a study population, their inability to demonstrate a temporal relationship limits the ability to infer causation. However, these studies do provide preliminary data to justify further epidemiologic investigation.

Which study design is most effective at establishing a temporal sequence of events?

Which study design is best for a study about prognosis?

What is the best design to study the etiology of a disease?

Which study design uses historical data to determine exposure level at a specific baseline in the past?