Genetic epidemiology of multiple sclerosis.

Epidemiological studies of multiple sclerosis have been performed on almost an industrial scale over the past 90 years. Morbidity statistics have been used to generate aetiological hypotheses, to assess local needs for the provision of services and the allocation of resources, and to define the natural history of the disease. Methodological factors have limited the extent to which the many surveys have yielded definitive conclusions in any one of these contexts. Most vulnerable have been the comparisons of prevalence between regions and the serial studies of single places. In planning an epidemiological study in multiple sclerosis, the usual practice is to retrieve cases from lists of those already known to be affected. Because in most parts of the world the diagnosis is coordinated through hospital clinics, scrutiny of departmental and office notes provides the best source of information. In some situations, a case can be made for retrospectively adjusting statistics to include those people who would have featured in a population based survey if their whereabouts or clinical status had been known at the time (onset adjusted prevalence'); their exclusion can then be regarded as an error of administration, recognition, disease expression, or any one of the quirks which makes one person seek medical advice in advance of another. Rigid application of criteria for inclusion, and the decision to omit suspected cases, will vary depending on the purposes of the study. For surveys examining biological features, the error should be towards inclusion of those who probably have the disease process even if this is not yet clinically declared. In other contexts, it is essential to restrict the register to those who definitely have the disease. Cases of different racial origin should not be grouped because they may differ for important characteristics. Sociohistorical factors are known to create significant differences in risk status even across quite small regions; conversely, some questions relating to the epidemiology of multiple sclerosis, which involve cohort studies, can only be answered by comparing specifically different locations. It makes little sense to plan a study requiring the recruitment of significant numbers of patients with a rare manifestation of multiple sclerosis, such as twinning or familial disease, in a community which has a low overall prevalence of the disease. Similarly, a ubiquitous but biologically important feature may not differ significantly between groups in places where multiple sclerosis is relatively common. It follows that there are usually better opportunites for identifying risk factors which make a significant contribution to the disease but are frequent in the at risk population by working in areas of low prevalence; conversely, risk factors for multiple sclerosis which are not overrepresented in the normal population will be identified more easily in high prevalence regions. The extent to which complete ascertainment is achieved depends much on the structure, size, and distribution of the denominator, and whether the population has previously been surveyed. The few patients with multiple sclerosis in a medium to low prevalence island community with a demographically stable population of around 20 000 and a normal age structure, can easily be ascertained but when surveying the disease in one at risk group living within a large metropolitan but ethnically mixed community, it may prove impossible to ascertain with confidence either the numerator or denominator, especially if recent population censuses are not available. Improved provision of facilities for the disabled inflates both prevalence and mortality; and the arrival of an investigator with a special interest in the disease abruptly increases morbidity estimates although these will plateau once ascertainment is saturated. Underestimating the absolute number of cases may not affect the definition of geographical gradients in the distribution of multiple sclerosis but it does matter in serial studies of a single region where a rise in prevalence resulting from reduced mortality and altered diagnostic criteria has to be distinguished from a real increase in incidence. It follows that investigator vigilance is a major confounding factor in comparative epidemiology.