The Epidemiology and Burden of Skin Disease

The Epidemiology and Burden of Skin Disease: Introduction

Scientists in health-related fields focus on phenomena at different levels. For laboratory scientists, the focus is at the molecular, cellular, or organ system level; for clinical scientists, the focus is on the patient; and for public health practitioners, the focus is on the population. Epidemiology is the basic science of public health.

Epidemiology has many subdivisions and offshoots. Often the epidemiology of a disease in a clinical review refers primarily to its frequency and distribution in the population and estimates of its morbidity and mortality. These data are derived by descriptive epidemiology. Case-control, cohort, and cross-sectional studies may seek to identify risk factors and causes of disease and form the core of analytical epidemiology. Evaluations of public health interventions (experimental epidemiology) constitute the third major branch of classic epidemiology. The basic principles of epidemiology have found broad application in many areas, including understanding the public health implications of naturally occurring and synthetic compounds (molecular epidemiology), the complex interactions of genetic and environmental factors in disease (genetic epidemiology), the formulation of better diagnostic and treatment strategies for patients based on available evidence (clinical epidemiology), and the structuring of health care delivery for better outcomes and greater efficiency (health services research). The reader is referred to other sources for a more detailed discussion of various topics in dermatoepidemiology.13

Types of Epidemiologic Studies

Three of the many types of epidemiologic studies are mentioned here because of their prominence in epidemiologic research. The randomized, controlled trial is a particularly rigorous type of study appropriate to the evaluation of public health interventions. In general, the intervention is performed on a random sample of the study population, and the entire study population is then observed for the occurrence of the outcome in question. The random assignment of intervention allows the more rigorous application of many statistical techniques and reduces the potential for bias. Elimination of biases permits these studies to evaluate the efficacy and impact of an intervention more accurately than trials that do not assign the intervention randomly. Standards for reporting have been published4 (, accessed Jul 7, 2010) and adopted by leading dermatology journals to improve assessment of their validity and their use in subsequent systematic reviews5 (see Chapter 2).

When evaluating risk factors for disease, it is frequently impossible to assign the risk factor randomly. Hence, inference is based on observational studies. In classical cohort studies, a group with exposure to the risk factor and a group without are chosen and observed over time. Occurrences of the study outcome are counted and compared between groups. Although more vulnerable to bias than randomized trials, cohort studies, in which exposure to the risk factor is known well before the study outcome is knowable, avoid some potentially serious biases. In a cohort study, the incidence of the study outcome can be measured directly in each group, and the relative risk can be measured directly as the ratio of the incidence between the two groups.

Cohort studies often are quite expensive to conduct because they require following a large population over time and may be impossible if the outcome being studied is uncommon. Hence, observational studies often use the case-control approach, in which cases with the outcome being studied and appropriate controls are investigated to determine their past exposure to the risk factor. Relative risks can be estimated by this approach, although incidence of the disorder cannot. Readers are referred to standard texts for more detail regarding epidemiologic study designs.6 Case-control and cohort study methods in dermatology also have been reviewed.79

Bias and Confounding

The problem with inference from observational studies is that one may be led to draw erroneous conclusions. In particular, an association that is found between an exposure and a disease may be an artifact due to one or more of the many forms of bias or confounding. Proper inference regarding cause and effect requires understanding these possible artifacts and their potential impacts.10

Selection bias occurs when factors that lead to selection of the study population affect the likelihood of the outcomes or exposures evaluated. For example, a case-control study of cutaneous lymphoma may recruit its cases from sources that typically include a high proportion of referred patients. If controls are recruited from a local clinic population, their socioeconomic status and location of residence may be substantially different from those of the cases simply due to the method of recruitment. Under these circumstances, an association of cutaneous lymphoma with occupation may be noted. It then becomes important to note that the observed association may be due not to a carcinogenic chemical in the workplace but rather to the method by which cases and controls were selected. Similarly, if one were conducting a cohort study of the effect of breast-feeding on the risk of atopic dermatitis, it would be important to select breast-fed and bottle-fed infants from similar environments.

Information bias occurs when the assessment of exposure or outcome may differ between the groups being compared. People who were exposed to a publicized environmental toxin may be more likely to seek care for minor symptoms or signs (and hence be more likely to be diagnosed and treated) than those who were not so exposed, even if the exposure had no biologic effect. Similarly, people who are diagnosed with a disease may be more likely to recall past exposures than healthy controls.

Confounding occurs when an observed association (or lack thereof) between exposure and disease is due to the influence of a third factor on both the exposure and the disease. For example, people who use sunscreens may have more intense sun exposure than those who do not, and intense sun exposure is one cause of melanoma. Hence, observational studies may mistakenly conclude that sunscreen use is a cause of melanoma when the observed association is due to sunscreen use serving as an indicator of a lifestyle involving intense sun exposure.

Causal Inference

Key issues in the public health arena often must rely on observational data for inferring cause and effect; in these situations, the validity and generalizability of the individual studies and of the totality of the evidence must be carefully examined. The following criteria generally are applied for causal inference when an association is found. Although they are described for inferring causality between an exposure and a disease, they are more generally applicable to epidemiologic causal inference.

Time Sequence

The exposure must precede the disease. This concept is simple and obvious in the abstract but sometimes difficult to establish in practice because the onset of disease may precede the diagnosis of disease by years, and the timing of exposure is often not well defined.

Consistency on Replication

Replication of the observed association is key and provides the strongest evidence if the replications are many and diverse and with consistent results. The diversity of the replications refers to varied contexts as well as to study designs with different potential weaknesses and strengths.

Strength of Association

True causal relationships may be strong (i.e., high relative risk) or weak, but artifactual associations are unlikely to have a high relative risk. If the association between factors x and y is due to the association of both with confounding variable z, the magnitude of the association between x and y always will be less than the magnitude of the association of either with z.

Graded Association

Also described as biologic gradient, this criterion refers to an association of the degree of exposure with occurrence of disease, in addition to an overall association of presence of exposure with disease. This dose-response relation may take many forms, as degree of exposure may, for example, refer to intensity, duration, frequency, or latency of exposure.


Coherence refers to plausibility based on evidence other than the existence of an association between this exposure and this disease in epidemiologic studies. Coherence with existing epidemiologic knowledge of the disease in question (e.g., other risk factors for the disease and population trends in its occurrence) and other disorders (including but not limited to related disorders) supports inference. Coherence with existing knowledge from other fields, particularly those relevant to pathogenesis, is critically important when those fields are well developed. It may involve direct links, which are preferred, or analogy. Just as observations in the laboratory assume greater significance when their relevance is supported by epidemiologic data, the reverse is equally true.


Experimental support is critical when feasible. As noted in Section “Types of Epidemiologic Studies,” the strongest inferences derive from results of randomized trials, although other experimental designs and quasi-experimental designs may contribute useful evidence.

More detailed discussions of these issues are available.11,12

Investigation of Disease Outbreaks

Although outbreaks of disease vary tremendously, use of a standard framework for investigation is important to address the public health issues efficiently (see Chapter 4). The Centers for Disease Control and Prevention has outlined this framework as a series of ten steps, which are described in more detail at

  1. Preparation. Before initiating fieldwork, background information on the disease must be gathered, and appropriate interinstitutional and interpersonal contacts should be made.

  2. Confirm the outbreak. Publicity, population changes, or other circumstances may lead to an inaccurate perception that more cases than expected have occurred. Hence, local or regional data should be sought to confirm the existence of an increased frequency of disease.

  3. Confirm the diagnosis. Symptoms and signs of persons affected should be determined and laboratory findings confirmed, perhaps with the assistance of reference laboratories.

  4. Establish a case definition, and find cases. Careful epidemiologic investigation will involve precise and simple case definitions that can be applied in the field. Efforts to find and count additional cases beyond those reported initially are key to defining the scope of the outbreak.

  5. Establish the descriptive epidemiology. The cases can now be characterized in terms of time, including development of an epidemic curve that describes the changes in magnitude of the outbreak; place, including mapping the distribution of cases; and person, the demographic and potential exposure characteristics of cases.

  6. Develop hypotheses. On the basis of the data gathered in steps 1 through 5 and the input of other individuals, plausible hypotheses about causality can be developed for further evaluation.

  7. Conduct analytical epidemiologic investigations. If the data gathered do not yet clearly prove a hypothesis, cohort and case-control investigations can be conducted to verify or disprove the hypotheses.

  8. Revise hypotheses and obtain additional evidence as needed. Steps 6 and 7 are repeated, each building on prior iterations, to establish the causal chain of events.

  9. Implement control measures. As soon as the causal chain of events is understood, prevention and control measures are initiated.

  10. Communicate results. An outbreak investigation is not complete until the results have been appropriately communicated to the relevant communities.

Descriptions of Disease in Populations: Measures of Disease Burden

No single number can completely describe the burden of skin disease because that burden has many dimensions and because the term skin disease itself is rather ambiguous. Many disorders with substantial morbidity or mortality, such as melanoma or lupus erythematosus, affect multiple organ systems. The degree of skin involvement may vary widely from patient to patient and within the same patient from time to time. Diseases not typically treated by dermatologists, such as thermal burns, often are excluded from estimates of the burden of skin disease even though they primarily involve the skin. In addition, some diseases treated most often by dermatologists may be classified in a different category by funding agencies or others [e.g., melanoma is classified as an oncologic disorder as opposed to a disease of the skin by the National Institutes of Health and by the International Classification of Diseases, (, accessed Jul 7, 2010) even though it almost always arises in the skin]. Organ systems are interrelated, and the overlap is sufficiently great that any definition of skin disease is necessarily arbitrary, and any global estimate of the public health burden of these diseases is therefore open to challenge. Typical measures of disease burden are discussed in the following sections.


Mortality is a critical measure of disease impact. Death certification is universal in the United States, and the International Classification of Diseases code of the underlying cause of each death is recorded. For the year 2006, there were 16,163 deaths reported as due to “skin disease” in the United States, of which most were due to melanoma (Table 1-1). Additional major causes included other skin cancers (primarily keratinocyte carcinomas), infections of the skin, and skin ulcers (primarily decubitus ulcers). Bullous disorders represented less than 2% of these deaths. The total number of skin disease deaths, of course, depends critically on the definition of skin disease, as noted in Section “Descriptions of Disease in Populations: Measures of Disease Burden.”

Table 1-1 Skin Disease Deaths, United States, 2006 

In addition to the total number of deaths, mortality typically is expressed as an age-adjusted rate to facilitate comparisons among populations with different age distributions. Statements of age-adjusted rates of mortality (or other results standardized by age) should be accompanied by an indication of the standard used in the adjustment to avoid potentially misleading inferences. For example, when 1998 melanoma mortality rates are estimated using the 2000 US population standard, the result is 50% higher than when the 1940 US standard population is used (1.8 vs. 1.2 per 100,000 per year for women and 4.1 vs. 2.7 per 100,000 per year for men). Similarly, when years of potential life lost are reported, the reader must be wary of different definitions that may be applied. In one analysis, a decline in years lost from melanoma was noted by one definition that was not observed with another.13

Careful analyses of mortality include assessment of the validity of the data. Melanoma mortality statistics appear to be reasonably accurate.14,15 However, deaths from keratinocyte carcinomas are overestimated by a factor of 2 (mostly due to the erroneous inclusion of mucosal squamous cell carcinomas of the head and neck region),16,17 and conventional estimates of deaths from cutaneous lymphoma miss about half of the actual deaths.18


Incidence refers to the number of new cases of a disorder. Mortality is low for most skin diseases; hence, incidence may be a more useful measure for the assessment of burden of skin disease. However, many features of skin diseases make their incidence difficult to measure. For example, for many skin disorders, there are no diagnostic laboratory tests, and, in fact, some disorders may evade physician diagnosis (e.g., allergic reactions). Incidence for reportable communicable diseases in the United States is published periodically based on reports to health departments, although underreporting of skin diseases due to failure to present for medical care or to misdiagnosis is a concern (Table 1-2). Incidences of melanoma and cutaneous lymphoma have been published based on data from a system of nationwide cancer registries, yet underreporting remains a potential concern with these data.19,20 Special surveys have been conducted and administrative datasets analyzed to estimate incidence of other disorders, such as keratinocyte carcinomas, although a system of sentinel registries would improve nationwide assessment.21,22 For some diseases unlikely to evade medical detection due to their severity, such as toxic epidermal necrolysis, efforts to estimate incidence have met with considerable success.23,24 Specific contexts that permit more accurate incidence estimates include the workplace; for example, where occupational skin disease is a prevalent problem.25

Table 1-2 New Cases of Selected Reportable Diseases in the United States