Chapter 4 Design Strategies and Statistical Methods Used in Descriptive Epidemiology

What Is a Study Design? The program that directs the researcher along the path of systematically collecting, analyzing, and interpreting data It allows for descriptive assessment of events, and for statistical inference concerning relationships between exposure and disease, and defines the domain for generalizing the results

Descriptive Epidemiology Means of organizing, summarizing, and describing epidemiologic data by person, place, and time Descriptive statistics can take on various forms, including tables, graphs, and numerical summary measures Application of statistical methods makes it possible to effectively describe the public health problem

Why Is Descriptive Epidemiology Helpful? Provides information about a disease or condition Provides clues to identify a new disease or adverse health effect Identifies the extent of the public health problem Obtains a description of the public health problem that can be easily communicated Identifies the population at greatest risk Assists in planning and resource allocation Identifies avenues for future research

Objective: Describe uses, strengths, and limitations of descriptive study designs Four types of descriptive studies: 1. Ecologic studies 2. Case reports 3. Case series 4. Cross-sectional surveys

Ecologic Study Involves aggregated data on the population level Ecologic fallacy

Example

Case Reports and Case Series A case report involves a profile of a single individual A case series involves a small group of patients with a similar diagnosis Provide evidence for larger scale studies (hypothesis generating)

Cross-sectional survey (sometimes called prevalence survey) Conducted over a short period of time (usually a few days or weeks) and the unit of analysis is the individual There is no follow-up period

Cross-Sectional Survey Strengths Can be used to study several associations at once Can be conducted over a short period of time Produces prevalence data Biases due to observation (recall and interviewer bias) and loss-to-follow-up do not exist Can provide evidence of the need for analytic epidemiologic study

Cross-Sectional Study Weaknesses Unable to establish sequence of events Infeasible for studying rare conditions Potentially influenced by response bias

Serial Surveys Cross-sectional surveys that are routinely conducted United States Census Behavior Risk Factor Surveillance System National Health Interview Survey National Hospital Discharge Survey

Example

China: 1.32 billion (19.84%) India: 1.13 billion (16.96%) United States: million (4.56%) Indonesia: million (3.47%) Brazil: million (2.8%) Pakistan: 163 million (2.44%) Bangladesh: million (2.38%) Sample Data Approximately 4.3 billion people live in these 15 countries, representing roughly two-thirds of the world’s population. Nigeria: 148 million (2.22%) Russia: 142 million (2.13%) Japan: million (1.92%) Mexico: million (1.6%) Philippines: 88.7 million (1.33%) Vietnam: 87.4 million (1.31%) Germany: 82.2 million (1.23%) Ethiopia: 77.1 million (1.16%)

Objective: Define the four general types of data Nominal (dichotomous or binary) Ordinal Discrete Continuous

Types of Data DescriptionExamples NominalCategorical – Unordered categories Two levels – Dichotomous More than two levels – Multichotomous Sex, disease (yes, no) Race, marital status, education status OrdinalCategorical – Ordering informative Preference rating (e.g., agree, neutral, disagree) DiscreteQuantitative – IntegersNumber of cases ContinuousQuantitative – Values on a continuum Dose of ionizing radiation

Objective: Define ratio, proportion, and rate

Ratio In a ratio, the values of x and y are independent such that the values of x are not contained in y The rate base for a ratio is typically 1

Proportion In a proportion, x is contained in y A proportion is typically expressed as a percentage, such that the rate base is 100

Rate A rate may be thought of as a proportion, with the addition that it represents the number of health- related states or events in a population over a specified period

Rate Equations

Cumulative Incidence Rate (Attack Rate) Diseases or events that affect a larger proportion of the population than the conventional incidence rate

Objective: Distinguish between crude and age-adjusted rates The crude rate of an outcome is calculated without any restrictions, such as by age or sex, on who is counted in the numerator or denominator These rates are limited if we try to compare them between subgroups of the population or over time because of potential confounding influences, such as differences in the age distribution between groups

Example of the importance of age-adjustment In 2002, the crude mortality rate in Florida was 1,096 per 100,000 compared with 579 per 100,000 in Utah The crude mortality rate ratio is 1.9, meaning the rates in Florida are 1.9 times (or 90%) higher than in Utah However, the age distribution differs considerably between Florida and Utah In Florida, 6.3% of the population is under five years of age and 16.7% of the population is 65 years and older Corresponding percentages in Utah are 9.8% and 8.5%

Example of the importance of age-adjustment (cont’d) Using the direct method of age adjustment based on the 2000 U.S. standard population, yielded rates of 762 in Florida and 782 in Utah per 100,000 After adjusting for differences in the age distribution, the rate in Florida is 0.97 times that in Utah

Example

Two methods for calculating age-adjusted rates Direct Indirect

Direct Method Suppose that we want to know the rate for females, assuming they had the same age distribution as males Multiply the age-specific female cancer rates by the age- specific population values for males to get expected number of cases for females for each age group, assuming they had the same age distribution as males These expected counts are then summed and divided by the total male population

Direct Method The resulting malignant cancer rate for females age-adjusted to the male population is: The crude rate is 1.09 times (or 9%) higher for males than females The age-adjusted rate ratio for males to females is now 1.28 This means that if females had the same age distribution as males, malignant cancer incidence would be 28% higher for males than females, as opposed to 9% higher found using crude rates

Example 2 Population A Age (years)Population# Deaths Attack Rate , ÷ 1000 = , ÷ 4000 = , ÷ 6000 = Total11, ÷ = Population B Age (years)Population# Deaths Attack Rate , ÷ 5000 = , ÷ 2000 = ÷ 500 = Total 7, ÷ 7500 =.0187

Example 2 (cont’d) Population A Age (years)PopulationAttack Rate Pop. BExpected ,000 ×.024 = ,000 ×.005 = ,000 ×.020 = 120 Total 11, Age-adjusted rate: 164/11000=.0149 Crude rate ratio:.0146 ÷.0187=.7822% lower in population A Adjusted rate ratio:.0146 ÷.0149=.98 2% lower in population A

Objective: Define the standardized morbidity (or mortality) ratio In situations where age-specific rates are unstable because of small numbers or some are simply missing, age adjustment is still possible using the indirect method

Standardized Morbidity (or Mortality) Ratio (SMR)

Interpretation of the SMR SMR = 1 Health-related states or events observed were the same as expected from the age-specific rates in the standard population SMR > 1 More health-related states or events were observed than expected from the age-specific rates in the standard population SMR < 1 Fewer health-related states or events were observed than expected from the age-specific rates in the standard population

Example of SMR Suppose that some or all of the female age-specific counts are unavailable, but that the total count is available Further suppose that the age-specific rates for males can be calculated Now multiply the age-specific rates in the male (standard) population by the age-specific female population values to obtain the expected number of all malignant cancer cases per age-specific group (see following table)

Example of SMR

Sum the expected counts to obtain the total number of expected malignant cancers in the comparison population This ratio indicates that fewer malignant cancer cases (about 25%) were observed in females than expected from the age-specific rates of males

Example 2 – Indirect Method Population A Age (years)Population# Deaths Attack Rate , ÷ 1000 = , ÷ 2000 = , ÷ 3000 = Total 6, Population B Age (years)Population# Deaths , Not Available Not Available Total 5,000 95

Example 2 – Indirect Method (cont’d) Age (years)Population BAttack Rate AExpected Deaths , Total 5, SMR = Observed ÷ Expected = 95 ÷ 73 = 1.3 The ratio indicates 30% more deaths than expected, based on the age-specific rates of population A (standard population)

Objective: Be familiar with tables, graphs, and numerical methods for describing epidemiologic data Tables Line listing Frequency distribution Graphs Bar chart, pie chart Histogram Epidemic curve Box plot Two-way (or bivariate) scatter plot Spot map Area map Line graph

What cancer is more common? Breast cancer in women? Prostate cancer in men? For U.S. Whites, Female breast cancer rate = per 100,000 Male prostate cancer rate = per 100,000

Example

Numerical Methods Measures of central tendency Mean Median Mode Measures of dispersion Range Inter-quartile range Variance Standard deviation Coefficient of variation Empirical rule Chebychev’s inequality

Objective: Be familiar with measures for evaluating the strength of the association between variables For discrete and continuous variables Correlation coefficient (denoted by r) Coefficient of determination (denoted by r 2 ) Spearman’s rank correlation coefficient Slope coefficient based on regression analysis Slope coefficient based on multiple regression analysis

Objective: Be familiar with measures for evaluating the strength of the association between variables (cont’d) For nominal and ordinal variables Spearman’s rank correlation coefficient Slope coefficient based on logistic regression analysis Slope coefficient based on multiple logistic regression analysis

Other Measures of Association Under analytic epidemiologic studies, the risk ratio (also called relative risk) and odds ratio are commonly used to measure association