The epidemiologic approach: Steps to public health action

Center for Infectious Disease Preparedness UC Berkeley School of Public Health URL: http://www.idready.org Updated June 2006
Created using freely available, open source software: http://www.openoffice.org
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

[INSTRUCTIONS ARE IN CAPITAL LETTERS. Talking points are in normal type.] This lecture reviews the epidemiologic approach used to guide public health and clinical research, and interventions. NOTE: REFERENCES ARE GIVEN BY THE PUBMED IDENTIFICATION NUMBER (PMID). JUST COPY AND PASTE THE PMID INTO http://www.pubmed.org AND GET THE MEDLINE CITATION, ETC. PLEASE SEND US GOOD EXAMPLES OR ARTICLES. THANKS, CIDP FACULTY

Learning objectives: Participants will be able to ...

Define epidemiology; Describe at least five components of the epidemiologic approach to public health action; Describe at least three goals of public health surveillance; List the components of a case definition; Describe the five measures of occurrence; Describe the measures of association Describe at least 3 types of public health actions with examples.

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

These are our learning objectives.

What is epidemiology?

Epidemiology is the study of the distribution and determinants of health-related states or events in specified populations, and the application of this study to the control of health problems.

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is one definition of epidemiology. DON'T SPEND TOO MUCH TIME ON THIS SLIDE. IT'S SELF-EXPLANATORY. By study we mean using quantitative, scientific research methods. By distribution we mean how the amount or burden of risk factors and/or outcomes are distributed in a specified population. By determinants we mean causal factors. The outcome could be having a condition (health-related states) or experiencing a new event (e.g., myocardial infarction). And, of course, the results should be applied to inform and guide public health actions (evidence-based public health).

The epidemiologic approach: Steps to public health action

SURVEILLANCE Detect outbreaks & threats Find cases for intervention Monitor trends Direct interventions Evaluate interventions Generate hypotheses DESCRIPTIVE What (case definition) Who (person) Where (place) When (time) How many (measures) ANALYTIC Why (Causes) How (Causes) MEASURES Count Time Rate Risk/Odds Prevalence STUDY DESIGN Design Implementation Analysis Interpretation Reporting

THREATS TO VALIDITY Chance Bias Confounding INFERENCES Epidemiologic Causal

ACTION Clinical Behavioral Community Environmental

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

This slide outlines this lecture. This slide is important because it summarizes the approaches, tools, and methods used in epidemiology. In fact, we have used this slide by itself to give an overview of epidemiology to a lay audience.

What's is public health surveillance?

Public health surveillance is the ongoing, systematic collection, analysis, interpretation, and dissemination of data about a health-related event for use in public health action to reduce morbidity and mortality and to improve health.

5 PMID: 15129191

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Public health surveillance is the following:

ongoing, systematic collection, analysis, interpretation, and dissemination of data; about a health-related event; for use in public health action

to reduce morbidity and mortality and to improve health

Goals of surveillance system that guide public health actions

Detect outbreaks Detect public health threats Detect infectious cases (case finding) Monitor trends in a target population Monitor exposed individuals for symptoms Monitor treated individuals for complications Direct public health interventions Evaluate public health interventions Generate hypotheses for further evaluation
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

These are the goals of a public health surveillance system that guide public health actions. The keys words are

Detect ... Monitor ... Direct ... Evaluate ... Generate hypotheses ...

The detect category results in a public health action. The monitor category is just thatmonitoring a situation. The remaining categories support evidence-based public health.

Types of disease surveillance

Passive surveillance

Title 17 reportable diseases in California

Enhanced passive surveillance Active surveillance

California Emerging Infections Program (CEIP)

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Passive and active surveillance is defined with reference to the public health authorities that receive (passive) or collect (active) the data. For non-epidemiologists, the most important distinction is knowing the difference between passive and active surveillance. Passive surveillance is health departments receiving California Morbidity Reports (CMRs) from physicians and laboratories for legally reportable conditions. Passive surveillance occurs in the absence of any health department outreach, education, or request. Passive surveillance can be enhanced through outreach and education to encourage reporting. Active surveillance involves health department staff actively collecting the data, usually by field deployment to clinics, hospitals, and clinical laboratories.

Sources of surveillance data

Mortality data

Death registry, medical examiner Legally reportable diseases, including cancer

Morbidity data

Birth registry Hospital discharge diagnoses (utilization data) Special surveys (NHANES, NHIS, CHIS)

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

There are only a few data sources that are population-based, ongoing, and systematic. These include birth and death registries, legally reportable diseases, and hospital discharge diagnoses. OPTIONAL: What the limitations of hospital discharge data? Therefore, because public health surveillance data is very limited, special studies are conducted periodically to fill in the gaps. These include the National Health and Nutrition Examination Survey (NHANES), National Health Interview Survey (NHIS), Behavioral Risk Factor Surveillance System (BRFSS) and, in California, California Health Interview Survey (CHIS), to name just a few. www.cdc.gov/nchs/nhanes.htm www.cdc.gov/nchs/nhis.htm http://www.cdc.gov/BRFSS/ http://ww.chis.ucla.edu/

Sources of surveillance data (continued)

Outbreak reporting Sentinel systems

Influenza-like illness reporting/testing (Kaiser) CDC Acute Hepatitis Sentinel County Study West Nile Virus Surveillance Program

Encephalitis case surveillance Larval and adult mosquito testing Sentinel chicken testing Dead bird surveillance

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Any outbreak is legally reportable. Sentinel surveillance systems are not usually population-based, nor representative of a target population, but they are good enough for government work. For example, during flu season, Kaiser Permanente outpatient clinics provide to the California Dept of Health Service Public Health Labs viral throat culture specimens from patients presenting with an influenza-like illness. These cultures are used to type and subtype influenza strains circulating in the community. While the California Kaiser HMO population is clearly not representative of all Californians, it's good enough. http://www.dhs.ca.gov/dcdc/VRDL/html/FLU/Fluintro.htm The California West Nile virus surveillance system incorporates disease and virus surveillance in humans, animals (sentinel chickens and dead birds), and vectors (mosquitos). http://www.westnile.ca.gov/

The case definition

Inclusion criteria

Clinical criteria (symptoms, signs) Epidemiologic criteria (person, place, and time) Laboratory criteria Suspect Probable Confirmed

Case classification

Exclusion criteria (for suspect and probable) Consider operating characteristics (sensitivity, specificity, predictive value)
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

10

Setting up a surveillance requires creating a case definition. A case definition is based on applying selection criteria: inclusion and exclusion criteria. Subjects that meet the inclusion criteria are consider a case. In general, confirmed cases are either laboratory confirmed or, lacking laboratory confirmation, have convincing inclusion criteria and are linked to a confirmed case. Probable cases have convincing inclusion criteria, but lack laboratory confirmation or a link to a confirmed case. Suspect cases have weak inclusion criteria, no laboratory confirmation, but may have epidemiologic risk factors (e.g., recently returned from a country with active cases). Suspect cases may also be a mild clinical case or early in symptom onset. Sometimes exclusion criteria are used to rule out suspect or probable cases. For example, see the monkeypox case definition: http://www.cdc.gov/ncidod/monkeypox/casedefinition.htm Finally, it's important to consider the sensitivity, specificity, and predictive value of case definition. A loose case definition is more sensitive (lower false negatives) and will pick up mild cases (but also more false positives). In contrast, a tight case definition is more specific (lower false positives), but will miss mild cases. CDC Examples: http://www.cdc.gov/mmwr/PDF/rr/rr4610.pdf

U.S. severe acute respiratory syndrome (SARS) cases reported to CDC, JanuaryJuly 2003.
Displayed in Figure B is the number of unexplained respiratory illness reports received by CDC by week of illness report (N = 1,460). However, only 398 met the U.S. SARS case definition and is displayed by week of illness onset in Figure A.

PMID: 15030681 11

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is a good example of public health surveillance in action during the SARS outbreaks. This is a retrospective analysis of possible SARS cases reported to the CDC. The bottom panel (B) has all the possible cases that were reported (1,460). This epi curve is by date of report. The top panel (A) has the possible cases that met the case definition (398 suspect or probable cases). This epi curve is by date of illness onset which is preferred. Suspect cases met the epi criteria (within 10 days of travel to SARS-affect country or contact to possible SARS case) plus mild/moderate respiratory symptoms, and probable cases met the epi criteria plus evidence of severe respiratory illness (pneumonia or hypoxemia). Of the 398 patients that met the case definition, only 8 were confirmed (black bars in panel A) to have a positive diagnostic test for the SARS coronavirus. This meant that there were up to 390 false positives, suggesting that the CDC used a sensitive case definition. We will study this example in more detail at a later lecture and evaluate the implication of using different case definitions.

Syndromic surveillance rationale for early detection

Centers for Disease Control and Prevention. Syndromic Surveillance: Reports from a National Conference, 2003. MMWR 2004:53(Suppl).

12 PMID: 15714620

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Traditional public health surveillance relies on diagnoses, usually laboratory confirmed. However, with an large intentional release of a bioterror agent, such as Bacillus anthracis spores, traditional surveillance systems will only detect advanced disease, meaning that most of those exposed are destined to die from inhalational anthrax, even with antibiotic therapy. To reduce morbidity and mortality, we would need to detect exposed persons with early, pre-diagnostic symptoms to initiate life-saving therapy. Unfortunately, early symptoms are nonspecific and might only be detectable by evaluating nontraditional systems (e.g., pharmacy sales). These types of systems have been called syndromic surveillance. The fundamental objective of syndromic surveillance is to identify illness clusters early, before diagnoses are confirmed and reported to public health agencies, and to mobilize a rapid response, thereby reducing morbidity and mortality. ...syndromic surveillance might help determine the size, spread, and tempo of an outbreak after it is detected, or provide reassurance that a large-scale outbreak is not occurring, particularly in times of enhanced surveillance (e.g., during a highprofile event).

Inferences in epidemiology
Descriptive epidemiology
Who, what, where, when, & how many? Rule out: Chance Bias Confounding Descriptive study: Design Implementation Analysis Interpretation Observation

Analytic epidemiology
Why & how?

Make comparison

Hypotheses

Control for: Chance Bias Confounding Analytic study: Design Implementation Analysis Interpretation

Epidemiologic inference

13

Causal inference

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In general, epidemiologic studies are either descriptive or analytic. Descriptive studies measure the burden and distribution of health-related states; i.e., describing the what (case definition), who (person), where (place), when (time), and how many (measure of occurrence). Study findings inform and guide public health services. When we are interested in determining the underlying causes (the why and how) of health-related event states, we design and conduct analytic studies (analytic epidemiology). Analytic studies involve generating and testing causal hypotheses. These hypotheses are often generated from descriptive studies. We can think of descriptive and analytic epidemiology as two sides of the same coin. In both types, we are always interested in minimizing threats to making valid conclusions (inferences). We'll come back to threats to making valid inferences, for now, let's look at the process of descriptive epidemiology more carefully. NEXT SLIDE

Descriptive epidemiology

Study of the distribution of health-related states or events

For a specific outcome, make comparisons and note differences across one or more dimensions (e.g. time series) Seek known explanations to account for observed differences (rule out chance, bias, confounding as explanations) Draw conclusions from descriptive study (epidemiologic inference #1)
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

14

It is human nature that when we observe occurrence of healthrelated states, we make comparisons across some dimension; for example, by neighborhood, by sex, by environmental exposures, etc. We might notice that they are not randomly distributed, and we imagine reasons to explain our observations (hypothesis generation). A layperson may prematurely conclude that their causal explanation is correct; in contrast, an epidemiologist first seeks alternative explanations (chance, bias, confounding) for the observations, and, if none are found, may then entertain new causal hypotheses that may need to be tested in an analytic study. Therefore, epidemiologists and non-epidemiologists share similar thinking processes when observing/describing data. However, an epidemiologist will first rule out alternative explanations (chance, bias, and confounding) before considering new causal hypotheses. A layperson is more likely to infer causality without considering threats to validity and without hypothesis-testing studies. Sometimes, these different approaches is a source of tension and misunderstanding between public health epidemiologists and community advocates who disagree on what conclusions can be drawn from descriptive studies.

Descriptive epidemiology in outbreak investigation

Describe cases

What (case definition)

Outcome (Case status)

How many? (measures of occurrence) When? (time) Where? (place) Who? (person)

Rule out chance, bias, confounding Generate causal hypotheses

Exposures (Causes) Known Suspected

15

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In outbreak investigations, the process of descriptive epidemiology occurs when we describe identified cases. Immediately, we start to consider alternative explanations (chance, bias, and confounding). If the outbreak is real, we generate causal hypotheses base on our observations and on current fund of knowledge. In the ideal scenario, we have sufficient information to implement control measures. Sometimes we don't have enough information to implicate a cause, or control measures are failing. In this case, we might conduct an analytic study to test causal hypotheses. We'll cover this approach in detail in another lecture.

Analytic epidemiology

Study of the determinants of health-related states or events

Generate hypotheses from descriptive studies Design and conduct studies to test hypotheses (control for chance, bias, confounding) Draw conclusions from analytic study (epidemiologic inference #2)

16

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Analytic epidemiology is the complementary process of descriptive epidemiology. Whereas in descriptive epidemiology we measure and describe differences across strata, in analytic epidemiology we measure and test differences across strata (exposed vs non-exposed, for example) in order to test a causal hypothesis. In descriptive epidemiology, epidemiologists rule out chance, bias, and confounding as explanations of observed differences. In analytic epidemiology, epidemiologist control for chance, bias, and confounding in study design and data analysis (more on these later). The conclusions we draw from descriptive and analytic studies is called epidemiologic inference. Chance, bias, and confounding are always threats to making valid inferences in both types of studies. Descriptive epidemiology is backbone of studies (including public health surveillance) conducted at health departments. Analytic studies are more commonly conducted at academic institutions. However, both require require scientific rigor.

John Snow's cholera map, 1854

17

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here's a classic example of analytic epidemiology. John Snow hypothesized that cholera was waterborne (presumably from the process of descriptive epidemiology). Two companies provided water to households. The Lambeth Company recently switched to a cleaner river source. He recognized this as a natural experiment to test his causal hypothesis. He predicted in the next cholera outbreak more cholera deaths would occur in households that received water from the Southwark and Vauxhall Company that used river water contaminated with human sewage. In contrast, households that received water from the Lambeth Company would have lower cholera death rates. The cholera outbreak occurred in 1854. In the 40,046 households supplied by the S & V Co, there were 1,263 deaths (315 deaths per 10,000 households); and in the 26,107 households supplied by the Lambeth Co, there were 98 deaths (37 deaths per 10,000 households). He made the epidemiologic inference that the cholera death rates differed. He also made the causal inference that cholera was waterborne and recommended the removal of the Broad St. pump. http://www.ph.ucla.edu/epi/snow/socbulletin34(2)3_7_2001.pdf

Describing disease occurrence

Measures of occurrence

Count, time, rate, risk, prevalence Person (Age, ethnicity, sex/gender, risk factors) Place (Country, state, county, city, census tract) Time (Calendar time, seasonality)

Describing disease occurrence by

18

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In order to conduct descriptive epidemiology, we need to measure the occurrence of disease cases. We now turn our focus to understanding epidemiologic measures of occurrence.

Epidemiologic measures: Overview

Types of measures How we combine numbers

Measures of occurrence Measures of association Measures of attribution (not today)

19

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

THIS IS AN OVERVIEW SLIDE We will discuss types of epidemiologic measures. We will discuss how we combine numbers into new measures.

Types of measures

Quantitative

Continuous numbers

Time, height, weight, etc. Integers (..., -4, -3, -2, -1, 0, 1, 2, 3, 4, ...) Counting (natural) numbers (0, 1, 2, 3, 4, ...)

Discrete numbers

Qualitative

Nominal categorical (non-ordered)

Male vs. female; ethnicity (white, Latino, Asian, etc.) Outcomes are usually binary (event vs. no event) Low, medium, high
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Ordinal categorical (ordered)

20

THIS SLIDE DESCRIBES THE TYPES OF MEASURES

Epidemiologic Measures: How we combine numbers

Rate

Proportion

x t a p= a b r=
P=

Change in x per change in t

Numerator is part of the denominator

x y

a 100 a b

Percent

Ratio
21

R=

x and y are different and are compared

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In epiodemiology we often combine numbers to construct a rate, proportion, or ratio. A rate is the change in one measure (usually counts) per change in another measure (usually person-time). In a proportion, the numerator (usually a count) is part of the denominator. We use proportions to estimate risk (2 out 10 exposed subject got ill) or prevalence (8 out of 10 injection drug users are anti-hepatitis C antibody positive). We use ratios to compare two measures. For example, a ratio of rates is called a rate ratio, and a ratio of risks is called a risk ratio.

Epidemiologic Measures: Measures of occurrence

Count Time Rate Risk (or Odds) Prevalence

L D
S L I R D = = = = =

Susceptible Latent infection Infective Recovered Dead

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

22

There are ONLY five types of epidemiologic measures of occurrence. That's it! No more, no less. The multi-state model depicts a susceptible person becoming infected but not infectious (latent state), becoming infectious (infective state), then recovering (recovery state). At any time the subject can die. We can count the number of events (new infections) during some time interval, or we can count the number of subjects in a health state (e.g., susceptibles) at a point in time. We can measure the waiting time until a subject becomes newly infected. This is also call time-to-event. We can calculate the average (per capita) rate of new infections as the count of new infections divided by person-time at risk. We can calculate the risk for new infection (for a specific time interval) as the count of new infections divided by the number of subjects infection-free at the start of the time interval. Odds is just the numerical transformation of risk: (more on this later). Prevalence is the proportion that are in a specific state (e.g., susceptible) at a point in time. For epidemics, prevalence measures can change significantly over time.

Community outbreak of shigellosis, San Francisco, 2000

During JuneDecember 2000, 230 cases of culture-confirmed* S. sonnei infection were reported to the San Francisco Department of Public Health; an average of 21 cases (range: 1329 cases) occurred during the same period from 1996 to 1999. [MMWR 2004;50(42):922] Center for Infectious Disease Preparedness

23

*Residents of San Francisco County aged >=15 years.

UC Berkeley School of Public Health www.idready.org

This epidemic curve depicts a community outbreak of shigellosis in San Francisco. Epidemic curves most commonly use counts of newly identified cases (also call incident cases or incidence). We avoid the term incidence because there are other terms that use the word incidence but have very different meanings:

incidence (count of new cases) incidence rate (a rate) incidence density (a rate) cumulative incidence (a risk)

Unfortunately, epidemiology terminology can sloppy. So our best recommendation is to use precise terms and to read methods sections carefully to understand precisely the actual measures used in papers.

Epidemiologic Measures: Measures of occurrence - Rate

rate = Number of events Person-time at risk D 1989 D 1990 D1991 K 1989 K 1990 K 1991

Period rate1989 1991 =

D x =Events in year x K x = Midyear population estimate in year x

24
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

The average or per-capita rate is calculated by dividing the number of events divided by the person-time at risk. Rates can be calculated for an open (dynamic) cohort or a closed (fixed) cohort. Average rates are most useful when the actual rate is approximately constant during the observation period, or the observation period is sufficiently short so that the rate is approximately constant during that time interval. For geographic regions, where it is not practical to measure individual person times, person-time at risk is approximated using population estimates. For example, if a population size is approximately constant during a 2-year period, then the personyears at risk is approximated by the mid-interval population size times 2 years. If population estimates are available every year, use all the data. Rates calculated using this method are called period rates because we are using data that is defined by calendar periods instead of individual observation times.

Measures of occurrence Cancer rates in an open cohort study, Person-time data

rate =

#cases person-time i

7 cases 85.7 person-years

=0.08168028 =8.2 cases per 100 py

25
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is an example of calculating an average rate from 13 subjects. This is an open (dynamic) cohort because subjects are moving in and out of the study. Subjects that leave the study before developing cancer are censored at the last time their status was known. For the numerator we count the number of cancer cases. For the denominator we sum the individual person-times at risk.

Measures of occurrence Period rates: Female breast cancer deaths, San Francisco
Annual period rates
CATEGORIES YEAR 1989 Breast cancer deaths 125 Female population 361,975 Rate per 100,000 per year 34.5 1990 130 361,401 36.0 1991 131 366,613 35.7

Period rate
r 1989 1991=

125 130 131 361,975 361,401366,613

=35.4 per 100,00 per year

26
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is an example using period data to calculate either (1) annual period rates for 3 years; or (2) a period rate combining 3 years of data. The midyear population estimates are used to estimate the personyears at risk per year.

Rates of acute hepatitis A, San Francisco, 1986-1999

Rates of Acute Hepatitis A by Year San Francisco, 1986-1999 Reported Cases per 100,000 Population 10 20 30 40 50 60 70 80 90 0 1986 1987

1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 Year
Source: Com m unity Health Epidem iology and Disease Control

1998 1999

27

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is a graphical display of annual period rates (time series) for hepatitis A cases reported in San Francisco.

Epidemiologic Measures: Measures of occurrence - Risk

Risk is a probability Probability models are used to estimate risk Here is useful and common approach*:

Consider N subjects as disease free at time 0 From time 0 to t, x subjects develop disease Proportion x/N estimates the risk of disease

*Binomial probability model

28
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In epidemiology we commonly estimate risks. Risks are probabilities. To estimate probabilities we use probability models. The simplest, familiar model we use is the binomial model. ADVANCED:Other probabilities models for estimating risk:

Binomial model (our focus) Hazard-based models (not today)

Constant hazard (constant rate) model Non-constant hazard model Exponential formula method (fixed time intervals) Kaplan-Meier method (time-to-event data)

Measures of occurrence Risk estimation with binomial data

R 0, t = #events in 0, t 18 = =0.1622 Population at risk at time 0 111

29PMID: 14681502

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

In March, 2003, a single passenger with SARS boarded an airplane that flew 3 hours from Hong Kong to Beijing. Out of the 111 passengers that were disease-free at the start of the flight, 18 developed SARS. The risk of infection was 18/111 or 16.2%. QUESTION: What conclusions can we draw from this data? ANSWER: SARS can be transmitted during an airplane fight.

Epidemiologic Measures: Measures of occurrence - Odds

Odds is a transformed risk (R) estimate
Odds 0, t = R 0, t 1 R 0, t

Example
If R=1 / 4 Then ,Odds 0, t = 1 /4 1 = 3/ 4 3

30

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

An odds is just the numerical transformation of a risk estimate. Gamblers think of risk in terms of odds. Epidemiologists use odds for mathematical convenience. For now, we only need to understand the definition of odds.

Measures of occurrence Prevalence

Point prevalence
P= Number of existing cases , at a point in time Number in total population

31

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Prevalence is proportion of a target population with an existing condition at a point in time.

Prevalence of hepatitis C virus (HCV) infection by age and race/ethnicity United States, 19881994

32

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is a graphical display of the prevalence of hepatitis C infection (indicated by anti-HCV antibodies) by age group and ethnicity. Note: Age group is a ordinal categorical variable. Ethnicity is a nominal categorical variable. Prevalence of anti-HCV positivity is a proportion.

Measures of occurrence: Prevalence, rate, and duration

new cases
Prevalence

removed by recovery, migration, or death

Under steady state conditions and small P,

Prevalence Rate Duration
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

33

It is useful to know that the prevalence of a condition is a function of rate of new cases and the duration of illness.

Measures of association

Measures of occurrence Rate Risk Odds

Measures of association Rate ratio Risk ratio Odds ratio

34

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Measures of association are used to compare a measure of occurrence in an exposed group vs. a non-exposed group. For example, if R1 is the rate of lung cancer among smokers, and R0 is the rate of cancer among non-smokers, the RR = r1/r2 is the rate ratio comparing smokers to non-smokers. Interpretation: RR > 1 means smoking is associated with lung cancer RR < 1 means not smoking is associated with lung cancer (that is, smoking is protective) RR 1 means no association

Outbreak of nodding off at CIDP lectures

The Centers for Disease Control (CDC) is responding to an outbreak of nodding off at training conferences provided by a west coast CDC Center for Public Health Preparedness. The UC Berkeley Center for Public Health Preparedness trainees are nodding off during lectures at an alarming rate. Local public health authorities have requested assistance from the CDC to determine the cause of this problem.
35
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Here is an example of a fictitious outbreak.

Outbreak of nodding off at CIDP lectures

Based on initial interviews, the CDC field investigators suspect that the lectures given by Dr. Toms Aragn, Director of the Center, is putting his students to sleep. At the next 3-day conference, investigators set up a surveillance system to measure students nodding off and who was lecturing when the nodding off occurred.
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

36

Outbreak of nodding off at CIDP lectures

Investigators set up hidden cameras in the lecture hall with unobstructed views of all the students and the lecturers. A case of nodding off was classified as probable if the student appeared to be sleeping with their eyes closed for 2 or more minutes; a case was classified as confirmed if the student snored audibly for 15 or more seconds. All student attendance times, and who was lecturing and for how long was measured.
37
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Outbreak of nodding off at CIDP lectures

Investigators hypothesize that students exposed to Dr. Aragn were at higher risk for nodding off. The 3-day conference had 100 attendees. There were 6 hours of lecture per day. Dr. Aragn lectured 6 of 18 hours. Risk of nodding off was estimated as the proportion that nodded off at each lecture. Rates were also calculated.
38
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Outbreak of nodding off at CIDP lectures: Risk ratio & Odds ratio
Exposure Toms Other
Individuals nodding off Cases of any nodding off Number started conference 19 21 100 12 13 100
Why is this analytic approach not recommended?

R 1=

19 = 0.19 100 R1 =1.58 R0

R0=

12 =0.12 100

RR =
39

OR =

R1 / 1 R1 =1.72 R0 / 1 R0
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

p value =0.24

Outbreak of nodding off at CIDP lectures: Risk ratio & Odds ratio
Exposure Toms Other
Cases of nodding off No. started lecture 21 559 13 1151

R 1=

21 = 0.0376 559 R1 =3.33 R0

R 0=

13 =0.0113 1151

RR =

OR =

R1 / 1 R1 =3.42 R0 / 1 R0
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

p value =0.0006
40

Outbreak of nodding off at CIDP lectures: Rate ratio

Exposure Toms Other
Cases of nodding off Person-hours at lecture 21 522.7 13 1047.3

r1 =

21 522.7

r1 =

13 1047.3

rr =

r1 =3.27 r0

p value =0.0008
41
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Epidemic curve of nodding off

For 3 days, Dr. Aragn lectured the hour before and the hour after lunch. What else could explain the nodding off? What is this called?

42

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Epidemiologic jargon!!!

Ambiguous terminology

Attack rate Case fatality rate, Case fatality ratio Survival rate Prevalence rate Incidence Incidence rate Incidence density Cumulative incidence
Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

The many lives of incidence

43

Avoid sloppy terminology. If you call a measure a rate, be sure it is actually a rate. Be aware that the word incidence is used epidemiologic terms with very different meanings. Incidence means new case counts. Incidence rate and incidence density mean rate. Cumulative incidence means risk. Instead just precise, unambiguous terminology:

Count: e.g., number of new cases Count: e.g., number of prevalent cases Time: e.g., median survival time Rate Risk (or odds) Prevalence

Epidemiologic study designs

Study designs

Experimental (Randomized control trial) Observational (Cohort, Case-control) Design (study protocol) Implementation (operations manual) Analysis Interpretation

Steps

44

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Threats to validity

Chance: Random error

Confidence interval (precision) P value (Pr{X x}, under null hypothesis) Selection bias Measurement bias (information bias)

Bias: Systematic error

Confounding

45

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

notes pending

Inferences in epidemiology
Descriptive epidemiology
Who, what, where, when, & how many? Rule out: Chance Bias Confounding Descriptive study: Design Implementation Analysis Interpretation Observation

Analytic epidemiology
Why & how?

Make comparison

Hypotheses

Control for: Chance Bias Confounding Analytic study: Design Implementation Analysis Interpretation

Epidemiologic inference

46

Causal inference

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

We have already discussed the process of descriptive and analytic epidemiology. In both approaches we need to address threats to making valid epidemiologic inferences (chance, bias, and confounding) from our studies. Causal inference is concluding that an exposure is causal. This involves incorporating current investigation findings with existing evidence from other studies and scientific disciplines. In public health, when findings are inclusive, we generally err on the side of protecting the public health.

Public health action

Clinical Behavioral Community Environmental

Adapted from Haddix AC, et al. Prevention Effectiveness: A Guide to Decision Analysis and Economic Evaluation. Oxford University Press 2003, 2nd Edition

47

Center for Infectious Disease Preparedness UC Berkeley School of Public Health www.idready.org

Public health interventions occur at four levels:

In the clinic with patients. Changing individual behaviors Changing community norms. Changing the environment so that it is safer.