Describing Epidemiologic Data

As a field epidemiologist, you will collect and assess data from field investigations, surveillance systems, vital statistics, or other sources. This task, called descriptive epidemiology, answers the following questions about disease, injury, or environmental hazard occurrence:

• What?
• How much?
• When?
• Where?
• Among whom?

The first question is answered with a description of the disease or health condition. “How much?” is expressed as counts or rates. The last three questions are assessed as patterns of these data in terms of time, place, and person. After the data are organized and displayed, descriptive epidemiology then involves interpreting these patterns, often through comparison with expected (e.g., historical counts, increased surveillance, or output from prevention and control programs) patterns or norms. Through this process of organization, inspection, and interpretation of data, descriptive epidemiology serves multiple purposes (Box 6.1).

Organizing descriptive data into tables, graphs, diagrams, maps, or charts provides a rapid, objective, and coherent grasp of the data. Whether the tables or graphs help the investigator understand the data or explain the data in a report or to an audience, their organization should quickly reveal the principal patterns and the exceptions to those patterns. Tables, graphs, maps, and charts all have four elements in common: a title, data, footnotes, and text (Box 6.2). In this chapter, additional guidelines for preparing these data displays will appear where the specific data display type is first applied.

Tables are commonly used for characterizing disease cases or other health events and are ideal for displaying numeric values. In addition to the previously mentioned elements in common to all data displays (Box 6.2), tables have column and row headings that identify the data type and any units of measurement that apply to all data in that column or row. A well-structured analytical table that is organized to focus on comparisons will help you understand the data and explain the data to others. In arranging analytical tables, you should begin with the arrangement of the data space by following a simple set of guidelines (Box 6.3) (1).

Cases are customarily organized in a table called a line-listing (Table 6.1) (2). This arrangement facilitates sorting to reorganize cases by relevant characteristics. The line-listing in Table 6.1 has been sorted by days between vaccination and onset to reveal the pattern of this important time–event association. Commonly in descriptive epidemiology, you organize cases by frequency of clinical findings (Table 6.2) (3). If the disease cause is unknown, this arrangement can assist the epidemiologist in developing hypotheses regarding possible exposures. For example, initial respiratory symptoms might indicate exposure through the upper airways, as in Table 6.2.

Counts

A first and simple step in determining how much is to count the cases in the population of interest. Always check whether data sources are providing incident (new events among the population) or prevalent (an existing event at a specific point in time) cases. For incident cases, specify the period during which the cases occurred. This count of incident cases over time in a population is called incidence. Never mix incident with prevalent cases in epidemiologic analyses.

The counts of incident or prevalent cases can be compared with their historical norm or another expected or target value. These case counts are valid for epidemiologic comparisons only when they come from a population of the same or approximately the same size.

Rates, Ratios, and Alternative Denominators

Rates correct counts for differences among population sizes or study periods. Thus, incidence divided by an appropriate estimation of the population yields several versions of incidence rates. Similarly, prevalent case counts divided by the population from which they arose produce a proportion (termed prevalence). Strictly speaking, in computing rates, the disease or health event you have counted should have been derived from the specific population used as the denominator. However, sometimes the population is unknown, costly to determine, or even inappropriate. For example, a maternal mortality ratio and infant mortality rate use births in a calendar year as a denominator for deaths in the same calendar year, yet the deaths might be related to births in the previous calendar year. To assess adverse effects from a vaccine or pharmaceutical, consider using total doses distributed as the denominator. Another example is injuries from snowmobile use, which have been calculated both as ratios per registered vehicle and as per crash incident (4). Returning now to counts, you can calculate expected case counts for a population by multiplying an expected (e.g., historical counts, increased surveillance, or output from prevention and control programs) or a target rate by the population total. This expected or target case count is now corrected for the population and can be compared with the actual observed case counts.

Measurements on a Continuous Scale

Disease or unhealthy conditions also can be measured on a continuous scale rather than counted directly (e.g., body mass index [BMI], blood lead level, blood hemoglobin, blood sugar, or blood pressure). You can use empirical cutoff points (e.g., BMI ≥26 for overweight). These can then be counted and the rates calculated. However, a person’s measurements can fluctuate above or below these cutoff values. To calculate incidence, special care therefore is needed to avoid counting the same person every time a fluctuation occurs above or below the cutoff point. For prevalence, this fluctuation amplifies the statistical error. A more precise approach involves computing the average and dispersion of the individual measurements. These can then be compared among groups, against expected values, or against target values. The averages and dispersions can be displayed in a table or visualized in a box-and-whisker plot that indicates the median, mean, interquartile range, and outliers (Figure 6.1) (5).

Time has special importance in interpreting epidemiologic data in that the initial exposure to a causative agent must precede disease. Often, this will follow a biologically determined interval. The disease or health condition onset time is the preferred statistic for studying time patterns. Onset might not always be available. In surveillance systems, you might have only the report date or another onset surrogate. Moreover, with slowly developing health conditions, a discernable onset might not exist. On the opposite end of the scale, injuries and acute poisonings have instantaneous and obvious onsets.

Similarly, times of suspected exposures vary in their precision. With acute infections, poisonings, and injuries, you will often have precise exposure times to different suspected agents. Contrast this with chronic diseases that can have exposures lasting for decades before development of overt disease. Other relevant events supplementing a chronologic framework of a health problem include underlying environmental conditions, changes in health policy, and application of control and prevention measures.

Relating disease with these events in time can support calculation of key characteristics of the disease or health event. If you know both time of onset and time of the presumed exposure, you can estimate the incubation or latency period. When the agent is unknown, the time interval between presumed exposures and onset of symptoms helps in hypothesizing the etiology. For example, the consistent time interval between rotavirus vaccination and onset of intussusception (Table 6.1) helped build the hypothesis that the vaccine precipitated the disease (1). Similarly, when the incubation period is known, you can estimate a time window of exposure and identify exposures to potential causative agents during that window.

Depicting Data by Time: Graphs

Graphs are most frequently used for displaying time associations and patterns in epidemiologic data. These graphs can include line graphs, histograms (epidemic curves), and scatter diagrams (see Box 6.4 for general guidelines in construction of epidemiologic graphs).

Contact Diagrams

Contact diagrams are versatile tools for revealing relationships between individual cases in time. In contact diagrams (Figure 6.2, panel A) (5), which are commonly used for visualizing person-to-person transmission, different markers are used to indicate the different groups exposed or at risk.
Epidemic Curves

Epidemic curves (Box 6.5) are histograms of frequency distributions of incident cases of disease or other health events displayed by time intervals. Epidemic curves often have patterns that reveal likely transmission modes. The following sections describe certain kinds of epidemic situations that can be diagnosed by plotting cases on epidemic curves.

Point Source

An epidemic curve with a tight clustering of cases in time (≤1.5 times the range of the incubation period, if the agent is known) and with a sharp upslope and a trailing downslope is consistent with a point source (Figure 6.3) (6). Variations in slopes (e.g., bimodal or a broader than expected peak) might indicate different ideas about the appearance, persistence, and disappearance of exposure to the source. Of note, administration of antimicrobials, immunoglobulins, antitoxins, or other quickly acting drugs can lead to a shorter than expected outbreak with a curtailed downslope.

To approximate the time of exposure, count backward to the average incubation period before the peak, the minimum incubation period from the initial cases, and the maximum incubation period from the last cases. These three points should bracket the exposure period. If a rapidly acting intervention was taken early enough to prevent cases, discount the contribution of the last cases to this estimation.

Point source outbreaks result in infected persons who might have transmitted the agent directly or through a vehicle to others. These secondary cases might appear as a prominent wave after a point source by one incubation period, as observed after a point source hepatitis E outbreak that resulted from repairs on a broken water main (Figure 6.4) (7). With diseases of shorter incubation and lower rates of secondary spread, the secondary wave might appear only as a more prolonged downslope.

Continuing Common Source

Outbreaks can arise from common sources that continue over time. The continuing common source epidemic curve will increase sharply, similar to a point source. Rather than increase to a peak, however, this type of epidemic curve has a plateau. The downslope can be precipitous if the common source is removed or gradual if it exhausts itself. The rapid increase, plateau, and precipitous downslope all appeared with a salmonellosis outbreak from cheese distributed to multiple restaurants and then recalled (Figure 6.5).

Propagated

A propagated pattern arises with agents that are communicable between persons, usually directly but sometimes through an intermediate vehicle. This propagated pattern has four principal characteristics (Box 6.6).

The epidemic curve accompanying the severe acute respiratory syndrome (SARS) contact diagram (Figure 6.2, panel B) illustrates these features, including waves with an approximate 1-week periodicity. Certain behaviors (e.g., drug addiction or mass sociogenic illness) might propagate from person to person, but the epidemic curve will not necessarily reflect generation times. Epidemic curves for large geographic areas might not reveal the early periodicity or the characteristic increase and decrease of a propagated outbreak. For these larger areas, stratifying the epidemic curves by smaller subunits can reveal the underlying periodicity.

Human–Vector–Human

Vectorborne diseases propagate between an arthropod vector and a vertebrate host. Six biologic differences in human–vector–human propagation affect the size and the shape of the epidemic curve (Box 6.7). The last two factors listed in the box will lead to irregular peaks during the progression of the outbreak and precipitous decreases.

An outbreak of dengue arising from a single imported case in a South China town reveals several of these features (Figure 6.6) (8). After the initial case, 15 days elapsed until the peak of the first generation of new cases. Control measures targeting the larva and adults of the mosquito vectors Aedes aegypti and A. albopictus began late in the first generation. The line indicates the rapid decrease in Aedes-infested houses (house index). A rapid decrease in dengue cases follows this decrease in vector density.

Zoonotic

The epidemic curve for a zoonotic disease among humans typically mirrors the variations in prevalence among the reservoir animal population. This will be modified by the variability of contact between humans and the reservoir animal and, for vectorborne zoonoses, contact with the arthropod vector.

Environmental

Epidemic curves from environmentally spread diseases reflect complex interactions between the agent and the environment and the factors that lead to exposure of humans to the environmental source. Outbreaks that arise from environmental sources usually encompass multiple generations or incubation periods for the agent. You should include on the epidemic curve a representation of the suspected environmental factor (e.g., rainfall connected with leptospirosis in Figure 6.7 [9]). In this example, nearly every peak of rainfall precedes a peak in leptospirosis, supporting the hypothesis regarding the importance of water and mud in transmission.

Relative Time

As an alternative to plotting onset by calendar time, plotting the time between suspected exposures and onset can help you understand the epidemiologic situation. For example, a plot of the days between contact with a SARS patient and onset of SARS in the person having contact indicates an approximation of the incubation period (Figure 6.8) (5).

Multiple Strata Display

To reveal distinctive internal patterns (e.g., by exposure, method of case detection, place, or personal characteristics) in time distributions, epidemic curves should be stratified (Figure 6.9). This puts each stratum on a flat baseline, enabling undistorted comparisons. Stacking different strata atop one another (as in Figure 6.7, which is not recommended) defeats attempts to compare the time patterns by group.

Examining Rates by Time

Temporal disease rates are usually illustrated by using a line graph (Box 6.4). The x-axis represents a period of interest. The y-axis represents the rate of the health event. For most conditions, when the rates vary over one or two orders of magnitude, an arithmetic scale is recommended. For rates that vary more widely, a logarithmic scale for the y-axis is recommended for epidemiologic purposes (Figure 6.10) (10). You should also use a logarithmic scale for comparing two or more population groups. Equal rates of change in time (e.g., a 10% decrease/year) will yield misleading, divergent lines on an arithmetic plot; a logarithmic scale will yield parallel lines.

Secular Trend

For most conditions, a time characteristic of interest is the secular trend—the rate of disease over multiple years or decades. Secular trends of invasive cervical cancer (Figure 6.11) reveal steady decreases over 37 years (11). New health policies in 1970 and 1995 that broadened coverage of Papanicolaou smear screenings for women were initially followed by steeper decreases and subsequent leveling off of the downward trend. This demonstrates how review of secular trends can bring attention to key events, improvements in control, changes in policy, sociologic phenomena, or other factors that have modified the epidemiology of a disease.

Seasonal and Cyclical Patterns

For certain conditions, a description by season, month, day of the week, or even time of day can be revealing. Seasonal patterns might be summarized in a seasonal curve (Box 6.8). Stratifying seasonal curves can further expose key differences by place, person, or other features (Figure 6.12) (12).

When creating graphics and interpreting distributions of disease by place, keep in mind Waldo Tobler’s first law of geography: “Everything is related to everything else, but near things are more related than distant things” (13). These distance associations of cases or rates are best understood on maps. In addition, maps display a wealth of underlying detail to compare against disease distributions. In creating epidemiologic maps, you should follow certain basic guidelines (Box 6.9).

Information about place of affected persons might include residence, workplace, school, recreation site, other relevant locales, or movement between fixed geographic points. Distinguish between place of onset, place of known or suspected exposure, and place of case identification. They are often different and have distinct epidemiologic implications. Information about place can range in precision from the geographic coordinates of a residence or bed in a hospital to simply the state of residence. Because population estimates or censuses follow standard geographic areas (e.g., city, census tract, county, state, or country), determination of rates is also restricted to these same areas.

Spot Maps

Use spot maps to reveal spatial associations between cases and between cases and geographic features. Cases can be plotted on a base map (Figure 6.13 [14]), a satellite view of the area, a floor plan, or other accurately scaled diagram to create a spot map. Dots, onset times, case identification numbers for indexing with a line listing, or other symbols might represent disease cases (Box 6.10). The example spot map of a dengue outbreak uses larger dots to represent cases clustered in time and space and numbers these clusters to reference to a table (not shown). It reveals the location of the first case in the business district and the large initial cluster surrounding it (Figure 6.13) (14). Cases not included in clusters are marked with smaller dots. These are widely dispersed, indicating that they did not acquire their infection from their local environs.

You might also use spot maps to represent affected villages, towns, or other smaller population units. If the denominator of the population unit is known, spots of different size or shading (Box 6.10) can represent rates or ratios.

Spot maps that plot cases have a general weakness. The observed pattern might represent variability in the distribution of the underlying population. When interpreting spot maps, keep in mind the population distribution with particular attention to unpopulated (e.g., parks, vacant lots, or abandoned warehouses) or densely populated areas.

Area Maps and Rates

Rates are normally displayed on area maps (e.g., patch or choropleth). The map is divided into population enumeration areas for which rates or ratios can be computed. The areas are then ranked into strata by the rates, and the strata are shaded (Box 6.10) according to the magnitude of the rate.

Compute and plot rates for the smallest area possible. For example, the map of spotted fever rickettsioses in the United States effectively displays multiple levels of risk for human infection (Figure 6.14) (15). Avoid using area maps to display case counts. Plotting only numerators loses the advantage of both the spot map (indicating exact location and detailed background features) and the area map (indicating rates).

Scatter Plots

Scatter plots are versatile instruments for exploring and communicating data. They indicate the association between two numerically scaled variables (Figure 6.15) (16). Each spot in the plotting area represents the joint magnitude of the two variables. As a convention in plotting epidemiologic or geographic association, the explanatory variable (exposure, environmental, or geographic) is plotted on the x-axis, and the outcome (rate or individual health measurement [e.g., BMI]) is plotted on the y-axis.

When the pattern of the spots forms a compact, linear pattern, suspect a strong association between the two variables. In Figure 6.15, a distinctive pattern of rapidly increasing cholera death rates is apparent as the altitude approaches the level of the River Thames. This reveals that factor and that an environmental exposure also related to low altitude (e.g., poor drainage of sewage) might have contributed to cholera incidence.

Recognizing disease patterns by personal attributes (e.g., age, sex, education, income, or immunization status) constitutes the fifth element in descriptive epidemiology. Two important qualifications apply to person data assessments. First, determining rates is more often necessary than for time and place. Second, age is a strong independent determinant for many causes of morbidity and mortality.

Social Groupings and Personal Contact

Social groupings might be as compact as a household or as diffuse as a social network linked by a common interest. The underlying epidemiologic process might produce disease distributions within and among social groupings that range from strong aggregation to randomness or uniformity. Clustered distributions might result from common exposures of group members, an agent that is transmissible through personal contact, an environmental exposure in the living or meeting areas, or localization of houses near or within an environmental area of high risk. Random or uniform distributions indicate that the exposure lies outside the group.

For diseases or behaviors spread through personal contact or association, contact diagrams reveal the pattern of spread plus such key details as index cases and outliers. In the example diagram, closeness and quality of relationships, timing between onsets, and places of contact are all displayed through different symbols and shading (Figure 6.2) (5). To support person-to-person transmission, you should also see that the timing between onsets of cases approximates the known incubation periods for the disease (Figure 6.8) (5).

Age

In most descriptive analyses, the epidemiologist will determine disease rates by age. This can be as simple as finding that a health event is affecting only a limited age group or as complicated as comparing age-specific rates among multiple groups. Age represents three different categories of determinants of disease risk (Box 6.11). Because age is a pervasive determinant of disease and because population groups often differ in their age structures, age adjustment (standardization) is a useful tool for comparing rates between population groups (17). Age-adjusted rates can be used for comparing populations from different areas, from the same area at different times, and among other characteristics (e.g., ethnicity or socioeconomic status).

An analysis of BMI by age from Ajloun and Jerash Governorates, Jordan, draws attention to increasing BMI and accumulating overweight prevalence for persons aged 18–75 years (Table 6.3) (Ajloun Non-Communicable Disease Project, Jordan, unpublished data, 2017). As an alternative to using tables, charts (Box 6.12) (e.g., dot charts) (Figure 6.16, panel A) or horizontal cluster bar charts (Figure 6.16, panel B) improve perception of the patterns in the data, compared with a table. Cluster bar charts with more than two bars per cluster (e.g., Figure 6.16, panel B) are not recommended.

Other Personal Attributes

Analysis by other personal attributes in descriptive epidemiology involves comparing rates or other numeric data by different classes of the attribute. For example, overweight prevalence in the Ajloun data can be compared by using different education levels. A more precise approach to estimating how much for measurements on a continuous scale, discussed earlier in this chapter, might be to compute the average and dispersion of the individual BMI measurements, as shown on a box-and-whisker plot (Figure 6.1).

The tables, graphs, and charts presented in this chapter have been determined experimentally to perform best in conveying information and data patterns to you and others. Accordingly, less efficient and inaccurate displays, although common, were avoided or noted as not recommended.

1. Ehrenberg AC. The problem of numeracy. Am Stat. 1981;35:67–71.
2. Centers for Disease Control and Prevention. Intussusception among recipients of rotavirus vaccine—United States, 1998–1999. MMWR. 1999;48:577–81.
3. Hawley B, Casey ML, Cox-Ganser JM, Edwards N, Fedan KB, Cummings KJ. Notes from the field: respiratory symptoms and skin irritation among hospital workers using a new disinfection product—Pennsylvania, 2015. MMWR. 2016;65:400–1.
4. Centers for Disease Control and Prevention. Injuries and deaths associated with use of snowmobiles—Maine, 1991–1996. MMWR. 1997;46:1–4.
5. Xie S, Zeng G, Lei J, et al. A highly efficient transmission of SARS among extended family and hospital staff in Beijing, China, April 2003. Presented at the 2nd Southeast Asian and Western Pacific Bi-Regional TEPHINET Scientific Conference, November 24–28, 2003, Borocay, Philippines.
6. Tauxe RV, Tormey MP, Mascola L, Hargrett-Bean NT, Blake PA. Salmonellosis outbreak on transatlantic flights; foodborne illness on aircraft: 1947–1984. Am J Epidemiol. 1987;125:150–7.
7. Abu-Sbeih A, Fontaine RE. Secondary water-borne hepatitis E outbreak from water storage in a Jordanian town [abstract LB01]. Presented at the TEPHINET 2000 First International Conference, April 17–21, 2000, Ottawa, Canada.
8. Dou F, Sun H, Wang ZJ. Dengue outbreak at a fishing port: Guangdong Province, China, 2007. Presented at the International Conference on Emerging Infectious Diseases, March 16–19, 2008, Atlanta, Georgia.
9. Ko AI, Galvao Reis M, Ribeiro Dourado CM, Johnson WD, Jr., Riley LW. Urban epidemic of severe leptospirosis in Brazil. Salvador Leptospirosis Study Group. Lancet. 1999;354:820–5.
10. Centers for Disease Control and Prevention, National Center for Health Statistics. National Vital Statistics System. Death rates for selected causes by 10-year age groups, race, and sex: death registration states, 1900–32, and United States, 1933–98. https://www.cdc.gov/nchs/nvss/mortality/hist290.htm
11. National Institutes of Health, National Cancer Institute, Surveillance Epidemiology, and End Results Program. SEER*Explorer. http://seer.cancer.gov/explorer/
12. Fontaine RE, van Severin M, Houng A. The stratification of malaria in El Salvador using available malaria surveillance data [Abstract 184]. Presented at the XI International Congress for Tropical Medicine and Malaria, September 16–22, 1984, Calgary, Alberta, Canada.
13. Tobler W. A computer movie simulating urban growth in the Detroit region. Econ Geogr. 1970;46(Suppl):234–40.
14. Vazquez-Prokopec GM, Kitron U, Montgomery B, Horne P, Ritchie SA. Quantifying the spatial dimension of dengue virus epidemic spread within a tropical urban environment. PLoS Negl Trop Dis. 2010;4:e920.
15. Biggs HM, Behravesh CB, Bradley KK, et al. Diagnosis and management of tickborne rickettsial diseases: Rocky Mountain spotted fever and other spotted fever group rickettsioses, ehrlichioses, and anaplasmosis—United States. MMWR Recomm Rep. 2016;65(No. RR-2):1–44.
16. Registrar-General. Report on the mortality of cholera in England 1848–49. London, UK: Her Majesty’s Stationery Office; 1852.
17. Fleiss JC. Statistical methods for rates and proportions. New York: John Wiley & Sons; 1981.