Guidance, Examples and Tools for Probability Sampling When Designing a Population-Based Biomonitoring Study

GUIDANCE, EXAMPLES AND

TOOLS FOR PROBABILITY

SAMPLING

WHEN DESIGNING

A POPULATION-BASED

BIOMONITORING STUDY

The findings and conclusions in this report are those of the authors and do not necessarily

represent the official position of the Centers for Disease Control and Prevention

Contents

Last mo

dified 05/07/2020

TABLE OF CONTENTS

INTRODUCTION .......................................................................................................................................... 4

Biomonitoring ........................................................................................................................................ 4

Biomonitoring stu

dies........................................................................................................................... 5

Proba

bility sampling designs............................................................................................................... 6

Sources and ty

pes of sampling biases................................................................................................. 7

Sampl

e weights ...................................................................................................................................... 8

STUDY DESIGN GUI

DANCE, EXAMPLES AND TOOLS............................................................................. 9

Flow-cha

rt ............................................................................................................................................. 10

Frequently asked

questions when designing a biomonitoring s tudy..........................................12

Example

s ............................................................................................................................................... 15

Last modified 05/07/2020

PREFACE

This document provides guidance, tools and examples to help select an appropriate probability sampling

design for a population-based biomonitoring study or survey within a state. The intended audiences of

this document include the following:

• State health departments and their biomonitoring study collaborators

• Centers for Disease Control and Prevention (CDC) programs that provide resources to their

recipients

• CDC programs and other programs that present surveillance data, including data from

biomonitoring studies and surveys

The guidance will be useful for developing measures to be displayed on the National Environmental

Public Health Tracking Network (Tracking Network), and can be used when:

• Designing biomonitoring studies, with the aim of making data from these studies useful for

public health surveillance

• Evaluating existing biomonitoring data from Tracking Program state/local recipients

• Evaluating existing biomonitoring data from non-Tracking Program state/local recipients and

other agencies or programs

This document only focuses on biomonitoring studies for population health surveillance and on

examples of population-based or probability sampling. This guidance complements information from

other guidance documents developed by the Association of Public Health Laboratories (APHL) and the

Council of State and Territorial Epidemiologists (CSTE):

• APHL’s Guidance for Laboratory Biomonitoring Programs: Developing Biomonitoring

Capabilities: https://www.aphl.org/aboutAPHL/publications/Documents/EH-Oct2019-

Biomonitoring-Guidance.pdf

• CSTE’s Biomonitoring in Public Health: Epidemiologic Guidance for State, Local, and Tribal Public

Health Agencies: http://www.cste2.org/webpdfs/BioMonISFINAL.pdf

The guidance does not replicate topics already addressed in these documents, such as communicating

biomonitoring results, types of specimens, or laboratory protocols for collecting and storing specimens.

Last modified 05/07/2020

DEFINITIONS

Target population: The complete collection of observations we want to study

Sample:

a sub-set of a population

Sampling frame: a list or map of sampling units in the population from which a sample may be

selected

Sampling unit: a unit that can be selected for a sample. In studying human

populations, observation

units are

often individuals.

Observation unit: an

object on which a

measurement is taken

Public health surveillance: the

ongoing, systematic

collection, analysis, and interpretation

of health-

related data essential to planning, implementation, and evaluation

of public health

practice

Last modified 05/07/2020

INTRODUCTION

Biomonitoring

Biomonitoring assesses human exposure to an environmental chemical by measuring that chemical, its

metabolite(s), or reaction product(s) in human blood, urine, milk, saliva, adipose, or other tissue

[1]. In

the environmental public health field, measurements in individuals are generally taken together to

constitute a population [1]. Biomonitoring helps us determine the environmental chemicals to which

people have been exposed [2]. Biomonitoring can reveal spatial, temporal, and socio-demographic

trends of body burdens of these chemicals in populations for public health surveillance. We can use

these data to identify populations at risk for environmental exposures or to determine the effectiveness

of interventions, laws, and regulations [1].

In exposure studies or surveys, detecting an environmental chemical in human tissue is not enough to

say that the chemical is causing disease

[2]. Other research studies and additional information on health

outcomes are needed to determine whether the chemical level measured is associated with disease [2].

However, results from exposure studies can help inform and prioritize research on health effects from

exposure to environmental chemicals [2].

The Centers for Disease Control and Prevention’s (CDC) National Environmental Public Health Tracking

Program (Tracking Program) maintains the Environmental Public Health Tracking Network (Tracking

Network) (https://ephtracking.cdc.gov/

). The Tracking Network displays standardized environment and

health surveillance data and measures from national, state, and local partners. This information can be

used to show trends over time and make comparisons among states. These data can also be used to

identify common issues among states and support regional responses.

The Tracking Network currently displays the 50

and 95

percentile and geometric mean concentration

levels of analytes from the following chemical groups:

• metals

• perfluoroalkyl and polyfluoroalkyl substances

• personal care and consumer products metabolites

• pesticide metabolites

• polycyclic aromatic hydrocarbon metabolites

• phthalate metabolites

• disinfection by-products

• volatile organic compounds

• tobacco metabolite

The analytes are measured in urine or blood in the U.S. population, collected as part of CDC’s National

Health and Nutrition Examination Survey (NHANES). The results are presented in the National Report on

Human Exposure to Environmental Chemicals (https://www.cdc.gov/exposurereport/index.html

). These

data are national in scope, and a goal of the Tracking Program is to add state-level biomonitoring data to

the Tracking Network.

Last modified 05/07/2020

In 2012, a review of biomonitoring data available to state tracking programs revealed that several of the

same chemicals were measured in state biomonitoring studies [3]. The review also revealed some

disparities in study populations, laboratory methods, and population sampling methods used in studies

that produced these data [3]. These differences, particularly the use of non-probability or convenience

sampling instead of probability population sampling, precluded the development of standard data that

could be representative of a population and displayed on the Tracking Network for public health

surveillance. Unlike probability sampling, convenience sampling does not allow generalization of the

findings to the entire population from which the sample is drawn [4]. However, convenience sampling

can be useful in pilot or exploratory studies, or if limited resources do not allow selecting a

representative sample [4].

Biomonitoring studies

For our review of state biomonitoring data, we broadly classified the biomonitoring studies into the

following categories:

•

Mandatory reporting: Passive collection of data from mandatory reporting of chemical

exposures or body burden levels to a public health agency (e.g., childhood blood lead testing).

•

Population-based study or surveillance: Conducted to detect and measure exposure in a

population (e.g., pregnant women or children). Such studies can be used to study and monitor

spatial, temporal, or demographic differences, or to evaluate the efficacy of public health

actions.

•

Targeted public health investigation: Conducted in response to health concerns in a

community resulting from the discovery of environmental contamination or a cluster of disease

from a possible chemical exposure (e.g., targeting residents using well water, or those near a

hazardous waste site). (Note: some studies target a community or geographic area because of a

health or exposure concern but use a probability sampling design to allow the results to be

representative of that population.)

•

Disease investigation or rapid response: Conducted in response to an exposure event to

evaluate clinical measures in individuals and support diagnosis of poisonings and assessment of

need for medical treatment (e.g., investigating extremely elevated blood lead levels).

•

Research project: Done independently or in collaboration with an existing research effort (e.g.,

Markers of Autism Risk in Babies - Learning Early Signs [MARBLES]).

These categories show that biomonitoring studies are conducted for various uses. The purpose and

objectives of the project will determine the type of biomonitoring study selected, the target population,

sampling frame, and the study design. The study design includes selecting the appropriate population

sampling design. Using the classifications above, this guidance focuses on population-based studies and

surveys or targeted studies within a state. The biomonitoring studies in the categories listed above could

use probability or non-probability sampling methods. However, the focus of this document is on

population-based studies or targeted studies that use a probability sampling design. Such studies

produce findings that can be generalizable to the target population and can be used for population

health surveillance. The study can be a state-level study or community-level within the state.

Last modified 05/07/2020

Probability sampling designs

Overview: In a probability sample, each unit in the population has a known non-zero probability of

selection, and units are randomly selected to be included in the sample. This sample can be used to

make inferences about the larger target population [5]. Probability sampling designs all begin by

identifying a sampling frame, which reflects all the persons in the target population of interest. The

individuals or elements of the sampling frame could be aggregated to become primary sampling units

(PSUs) for multi-stage or cluster sampling. A cluster or PSU consists of a group of elements (e.g., census

tracts, ZIP codes, or households). After selecting the PSUs, you can use various approaches to randomly

select study participants within selected PSUs. Some methods for probability sampling include the

following:

1. Simple random sampling of all elements within the sampling frame [6, 7, 8, 9]

Every eligible person w

ithin the population of interest has the same chance or probability of

being selected for the study. Using random selection, a sample of persons from this population

is selected for the study or survey. Other types of random sampling include systematic, stratified

and cluster sampling.

Pros and cons:

imple random sampling is the simplest form of probability sampling. However,

you need a complete list of all eligible persons before sampling which might not be available,

especially for large populations. Because simple random sampling gives each eligible person an

equal chance of being selected, it may result in samples spread out over a large geographic area.

The logistics of contacting people over a wide geographic area can be challenging [8, 9]. For

these reasons, simple random sampling is rarely used.

2. One-stage sampling [6, 7, 10]

One-stage sampling requires a list of primary clusters. In a cluster sample, observation units in

the population are aggregated into larger sampling units called clusters. When applying this

method, you first identify the primary clusters (e.g., households) for the study area. The

households are randomly selected from the study area, and all persons in each selected cluster

(household) are then included in the study sample. Example 1A

below provides insights on

applications of one-stage sampling.

Pros and cons:

One stage

sampling is simpler

to implement

than multistage

sampling.

complete

list of units

within each sampled primary cluster is needed but

rarely available

for

large populations

and

difficult to assemble.

Because

of this,

studies often use multistage

sampling.

For one stage sampling t

o be beneficial, any given cluster should reflect the variation in the

overall target population. One stage sampling can introduce cluster effects (i.e., concentration

of people based on similar factors) for population estimates when all persons in selected

clusters are included [9].

3. Multistage sampling [6, 7, 11, 12]

a. Two-stage sampling

Last modified 05/07/2020

Two-stage sampling includes a primary and a secondary cluster. The study area

population is divided into groups (primary cluster), then households from these groups

(secondary cluster). Groups are randomly selected, then households within these groups

are randomly selected. All persons within selected households are then included in the

study (See examples 2A-F

below.)

b. Three-stage sampling

Three-stage sampling has 3 levels of clusters. The population in the study area is divided

into primary, secondary, and tertiary sampling units. First, a sample of primary units is

randomly selected. A sample of secondary units is then randomly selected from the

selected primary units. Finally, a sample of tertiary units is randomly selected from the

selected secondary units. (See Example 3A

below.)

Examples of geographic units of primary, secondary and tertiary clusters

One-stage Two-stage Three-stage

•

Primary –

Households

• Primary – Census Blocks

• Secondary – Households

• Primary – Boroughs

• Secondary – Census Blocks

• Tertiary – Households

Pros and cons: Multistage sampling is cost effective and practical, especially for large

populations [9]. For example, it is not possible to list all households in a state for a sample.

However, with two-stage sampling, you could use a county as the primary cluster and census

blocks as the secondary cluster. Then you would only have to identify households in the census

blocks that have been randomly selected for your sample. This makes this method feasible for

large populations. However, multi-stage sampling can produce more sampling errors than other

probabilistic methods [9]. If the members of the different clusters are different from one

another, this reduces the efficiency, i.e., the precision, of the estimates.

4. Using an existing population-based study or survey

Some programs may wish to recruit from existing population-based surveys by asking to add

interest-gauging questions for participation in a biomonitoring study. Information on persons

who agree to participate would then be transferred to the biomonitoring study team, and

separately followed-up on for participation. An example of an existing population-based study

that biomonitoring programs have used is the Behavioral Risk Factor Surveillance System

(BRFSS) survey. (See Examples 4A and 4B

below.)

Pros and cons: This method may be less resource intensive as there is an existing infrastructure

to recruit participants. However, the biomonitoring study would inherit any limitations of the

existing study, for example some existing studies only focus on school-aged children, therefore,

limiting the biomonitoring study to this population.

Sources and types of sampling biases

Under coverage is failing to include all of the target population in the sampling frame [5]. Therefore,

some members of the population are inadequately represented in the sample. This may make it difficult

Last modified 05/07/2020

to provide estimates for hard-to-reach or marginalized populations (e.g., certain race/ethnic minorities)

[13].

Voluntary response bias occurs when the sample consists only of those who volunteered to participate

in the survey. That is, the sample members are self-selected volunteers. The respondents may not be

representative of the whole population. This could lead to biased estimates therefore, affecting the

validity of the estimates.

Non-response occurs when only a fraction of the sampled population responds to a survey such that

they are no longer representative of the whole population. That is, the respondents differ meaningfully

from the non-respondents. Low response rates could lead to underestimating certain populations (e.g.,

younger populations)

[13]. This could lead to biased estimates, thereby affecting the validity of the

estimates.

In general, participation rates in epidemiological studies have been declining over the past 30 years [14].

Some reasons for this include the increasing burden for participants from survey assessments, biologic

sampling, and requests for follow-ups [14]. For biological sampling, biomonitoring studies may use

invasively collected matrices such as blood and adipose tissue or non-invasively collected matrices such

as urine and saliva. Non-invasively collected matrices result in increased participation rates [15].

It has also been increasingly difficult to find and talk to potential study participants. Some reasons for

this include the increasing number of unlisted telephone numbers; use of cell phones; telephone

screening; and unsolicited mail and phone calls [14].

Improving participation rates may require evaluating recruitment methods. Face-to-face recruitment

usually has higher participation rates than telephone or less personal methods [14]. Using mixed

methods for data collection, such as an alternative data collection method for non-respondents might

help increase participation. An example is using telephone interviews for persons who did not respond

to a mail survey that is recruiting study participants [14].

Oversampling methods can also be used to sample larger numbers of subpopulations of interest,

thereby increasing their coverage. These methods can be used to improve the reliability and precision of

estimates from subgroups [16].

(See Examples 5A-C below.)

Studies or surveys using a probability sample design can improve the representation of population

subgroups in their sample by using features such as stratification [17]. This involves dividing the

population into smaller, mutually exclusive groups called strata [17]. The strata are formed on

population members shared attributes or characteristics, e.g., income, race/ethnicity and population

density in geographic area. Each stratum is sampled separately using probability sampling methods, and

then the results are combined across strata to provide estimates for the target population [17]. See

examples 2A, 3A and 5B

below fo

r examples of applying stratification.

Sample weights

The weighting of sample data permits researchers to produce estimates of the statistics that would have

been obtained if the entire eligible population had been surveyed [18]. Sample weights can be

considered measures of the number of persons in the target population represented by the particular

sampled participant [18]. Weighting takes into account the different probabilities of selection or

Last modified 05/07/2020

inclusion, survey nonresponse, and differences between the distribution of the target population and

the final sampled population [18]. For data analysis, only one final weight per sampled participant is

used which adjusts for all these factors.

Weighting is used for valid statistical inference, to reduce bias, and to keep the weighted sample

distribution close to the distribution of the target population, especially when oversampling is used for

specific subpopulations [19]. Weighting reduces bias, e.g., nonresponse bias, but does not remove all

nonresponse bias.

No one protocol exists for computing weights; rather, computing weights varies among studies

depending on study design, including sample selection and recruitment, and availability of information

on the target population and nonresponse [20]. When computing a weight for each sampled individual,

a combination of the following steps is usually used [20]:

1. Determine the base weight to account for sample selection

2. Adjust for nonresponse

3. Adjust further for undercoverage, i.e., incomplete sampling frame coverage

4. Adjust further to account for distribution in the target population by key characteristics

The final weight for the sampled individual is the product of the value generated in each step above.

This final weight should appear on each respondent data record as a variable to be used in data analysis.

Some datasets could include the individual component weights as well as the final weight.

STUDY DESIGN GUIDANCE, EXAMPLES AND TOOLS

Some considerations and questions to ask when designing a biomonitoring study are listed below. These

also indicate stages at which guidance can be targeted:

• What is the aim of the study?

• What is the target population?

• What probability sampling design will be used?

• Who will be measured (randomly selected persons or everyone in household)?

• What is an appropriate sample size for the study?

• Do I need to oversample certain populations?

• How do I compute weights for study participants?

• What demographic information will be collected and how will it be formatted?

The flowchart, frequently asked questions (FAQ) section with more detailed responses, and examples

that follow provide guidance, examples and tools to help answer these questions. The guidance assumes

that the study aim is population health surveillance, and only focuses on examples of probability

sampling.

Last modified 05/07/2020

Flow-chart

Last modified 05/07/2020

The flow-chart

above guides the selecting of an appropriate probability sampling method to identify participants for a

population-based biomonitoring study within a state.

If the population

of interest is sub-state level, select

from the

following probability sampling designs

and resources:

• One-stage

sampling. See example

1A. Sample size calculation example:

PEATT’s Sampling Strategy and

Data

Analysis Assistance section. For access: email

pfas@cdc.gov.

• Two-stage sampling. See example 2A-F. Sample size calculation resources and examples: CASPER:

https://www.cdc.gov/nceh/hsb/disaster/casper/docs/CASPER-toolkit-3_508.pdf; Emory:

http://web1.sph.emory.edu/users/cdckms/samplesize%20icc%20deff2.html; and WHO:

http://www.who.int/ncds/surveillance/steps/resources/sampling/en/

• Three-stage sampling. See example 3A. Sample size calculation resources and examples: Emory:

http://web1.sph.emory.edu/users/cdckms/samplesize%20icc%20deff2.html; and WHO:

http://www.who.int/ncds/surveillance/steps/resources/sampling/en/

If the population of interest is state-level, select from the following probability sampling designs:

• Recruit from an

existing population-based survey (e.g. BRFSS). See examples 4A

and

4B. Sample size

calculation

resources

and examples: Emory: http://web1.sph.emory.edu/users/cdckms/samplesize%20icc%20deff2.html;

and

WHO: http://www.who.int/ncds/surveillance/steps/resources/sampling/en/

• Three-stage sampling. See example 3A. Sample size calculation resources and

examples:

Emory:

http://web1.sph.emory.edu/users/cdckms/samplesize%20icc%20deff2.html; and

WHO:

http://www.who.int/ncds/surveillance/steps/resources/sampling/en/

For

all probability

sampling designs,

consider

oversampling

to increase

the reliability and precision

of estimates of

certain population subgroups, e.g., under-represented populations. Consider oversampling clusters before the 1

or 2

stage. See section 5.

See other examples 5A-C

For all probability sampling designs use the following resources for computing weights for study participants: NHANES

Estimation Procedures: https://www.cdc.gov/nchs/data/series/sr_02/sr02-184-508.pdf; and GATS Sample Weights

Manual:

https://www.who.int/tobacco/surveillance/en_tfi_gats_sampleweightsmanual_v2_final_15dec2010.pdf

For all probability

sampling designs, consider collecting demographic information in the same format as NHANES

variables

to facilitate comparison:

• Gender: Male; Female

• Race/Ethnicity: Mexican Americans; Non-Hispanic Blacks; Non-Hispanic whites; Non-Hispanic Asians; All

Hispanics

Frequently asked questions when designing a biomonitoring study

What is the aim of the study?

The focus is on population-based studies or surveys, or targeted studies within a state, which produce findings

that can be generalizable to the target population, and that can be used for population surveillance.

What is the target population?

State- or community-level studies

What probability sampling design will be used?

Last modified 05/07/2020

•

You will find guidance and examples of some probability sampling designs in the “Examples

” section below.

These include guidance

using one-stage, two-stage

and

three-stage designs

and existing population-based

surveys or programs.

•

Who will be

measured

(randomly selected persons

everyone in

household)?

The goals of the study

and resources

will determine if the

biomonitoring data

will be

collected on

individuals

the

entire

household.

What is an appropriate sample size for the study?

Two resources offer examples of sample size calculation:

 Per- and Polyfluoroalkyl Substances (PFAS) Exposure Assessment Technical Tools (PEATT)

This resource provides information on how to conduct a one-stage cluster sampling approach and

estimate sample size at the household level. For a copy of this guidance, email the Centers for Disease

Control and Prevention at [email protected].

 Community Assessment for Public Health Emergency Response (CASPER)

This resource provides information on two-stage cluster sampling in which 30 clusters are selected, and

then seven households in each cluster are interviewed. It also provides methods for how to modify this

30 × 7 design. The sampling methodology is available from the Centers for Disease Control and

Prevention at: https://www.cdc.gov/nceh/casper/sampling-methodology.htm.

Two publicly available resources can help you calculate an appropriate sample size for any study:

 Sample Size Calculations for a Proportion for Cluster Surveys

Kevin Sullivan in the Department of Epidemiology at Emory University, Atlanta GA, developed this online

program. The program, available at

http://web1.sph.emory.edu/users/cdckms/samplesize%20icc%20deff2.html

, gives investigators various

options for calculating the number of observations per cluster and the number of clusters. It also

provides definitions for terminology, and formulas where necessary.

 STEPwise approach to surveillance (STEPS) Sample Size Calculator and Sampling Spreadsheet

This resource from the World Health Organization, available at

http://www.who.int/ncds/surveillance/steps/resources/sampling/en/, provides a spreadsheet with a

sample size calculator. It also provides a sampling spreadsheet with worksheets for:

Probability proportional to size (PPS) sampling

Simple random sampling, and

Last modified 05/07/2020

o Weighting your data

•

Do I

need to oversample certain populations?

We oversample to increase the reliability and precision of estimates of certain population subgroups.

For example, the National Health and Nutrition Examination Survey (NHANES) used oversampling to sample

larger numbers of subgroups of interest, such as minorities, adolescents, and older adults [16]. This increases

reliability and precision of estimates in these population subgroups [16]. Rural populations might be another

subgroup of interest when considering oversampling to help ensure that you achieve a large enough N value for

appropriate statistical power.

For sub-populations that are at least 10% of the total population, a general sample will usually produce reliable

estimates [21]. For subpopulations between 1% and 10% of the total population, the oversampling methods

described in the Examples 5A-C

below are needed [21].

By considering oversampling clusters before the first or second stage, you have a better opportunity to net a

desired sample size.

 Possible oversampling methods include increasing the number of units (e.g., census blocks) in your first

stage [22]. You could also increase the number of units (e.g. households) in the second stage [22].

 An example of oversampling to increase cluster selection is a modified application of the CASPER study

design whereby the 30 × 7 design is modified to a 35 × 7 therefore increasing the N value by 16.6% [23].

The NHANES study design, as another example, draws its sample in the following stratified, four stages [24]:

• Stage 1: PSUs are first stratified according to population size, and then PSUs are selected from each

stratum. These are mostly single counties or, in a few cases, groups of contiguous counties with

probability proportional to a measure of size (PPS).

• Stage 2: The PSUs are divided up into segments (generally city blocks or their equivalent). As with each

PSU, sample segments are selected with PPS.

• Stage 3: Households within each segment are listed, and a sample is randomly drawn. In geographic

areas where the proportion of age, ethnic, or income groups selected for oversampling is high, the

probability of selection for those groups is greater than in other areas.

• Stage 4: Individuals are chosen to participate in NHANES from a list of all persons in selected

households. Individuals are drawn at random within designated age-sex-race/ethnicity screening

subdomains. On average, 1.6 persons are selected per household.

stage 3,

households from each segment are randomly drawn. For geographic

areas where

the proportion of

age, ethnic, or income groups selected for oversampling is high, the probability

of selection for those groups is

greater than in other areas

[16]. This can be replicated in other biomonitoring

studies that apply probability

sampling to

oversample populations

of interest.

• How do I compute weights for study participants?

Two examples of study protocols that can serve as resources for computing weights are listed below:

• NHANES provides information on the health and nutritional status of the noninstitutionalized civilian

resident population of the United States. The sample for NHANES is selected using a complex, four-

stage sample design. NHANES carries out sample weighting in three steps. The first step computes base

Last modified 05/07/2020

weights to compensate for unequal probabilities of selection. The second step adjusts for nonresponse

to reduce bias. In the third step, sample weights are post stratified to the reference (target) population.

The procedures for computing these weights are detailed in the National Health and Nutrition

Examination Survey, 2015 – 2018: Sample Design and Estimation Procedures document

(https://www.cdc.gov/nchs/data/series/sr_02/sr02-184-508.pdf

• The Global Adult Tobacco Survey (GATS) is a nationally representative household survey of adults 15

years of age or older, and its used to enhance the capacity of countries to design, implement and

evaluate tobacco control interventions. GATS uses a geographically clustered multistage sampling

methodology to identify households to be included in the study. GATS recommends a three-step

approach in computing sample weights: compute a base weight for each sample respondent, adjust the

base weights for nonresponse, calibrate the adjusted weights to known population totals. Their Sample

Weights Manual

(

https://www.who.int/tobacco/surveillance/en_tfi_gats_sampleweightsmanual_v2_final_15dec2010.p

df) details requirements and recommendations on how to approach computing sample weights for a

country GATS. The step-by-step

manual can be modified to country-specific sample design and analysis

needs.

•

What demographic information will be collected and how will

be formatted?

Consider collecting demographic information in the same format as NHANES so that you can compare local

estimates with national estimates.

Examples

1. One-stage sampling

Identify the primary cluster unit (e.g., households). Households are randomly selected from the study area. All individuals in each selected cluster/household

are included in the sample.

Example 1A: Per- and Polyfluoroalkyl Substances (PFAS) Exposure Assessment Technical Tools (PEATT) — One-

stage cluster sample enrolling entire households within a PFAS-affected community

Last modified 05/07/2020

Program: Centers for Disease Control and Prevention (CDC)

Access: To request PEATT documentation, email CDC at pfas@cdc.gov

Methods:

• The sampling frame is a list of households in a community of interest

• A complete listing of all households within the samp

ling frame is obtained

• A sample size for the study is determined

• Househo

lds from the list are randomly selected to fit th

e selected sample size

• All household members are recruited to participate in the study

• Statistically based

. Information learned

can be applied to others in the tar

geted

population (those not tested).

• Presents formulas to be used to

determine the sample size with step-

by-step example of how to apply it.

• Mentions Census.gov as backup

resource if municipal water/well

sampling frame information are

unavailable.

• Gives two options for calculating

sample size based on whether loc

Cons

• Information l

earn

can only be applied to others

within the sampling frame.

• Need

to have some level of statistical experience

understand the formulas. Most significantly,

applying the formulas requires a sampling frame

which requires a complete list of all households in

the area and the total population of the

geographic area of interest. For water

contamination issues, a list of those on public

water or on private well water is needed,

depending on the exposure; the latter may be

hard to come by.

Recommendations for

use and m

odifications to

consider

• Can modify to select

sampling fram

e based on

geographic area (i.e., not an exposed community).

• Can only use this strategy if you know the number

of household

s within the geographic area being

assessed (or census data), total population of the

area, and analyte data (either from preliminary

testing in that area or from NHANES).

Last modified 05/07/2020

preliminary biomonitoring results from

the exposed community are av

ailable

and if not, then states to use NHANES

data for the analyte of interest.

• Gives formula for adjustment of

usehold

sample size based on an

estimate of non-response.

•

Census.gov

information on housing units

is only

available at

the state

and

county level and for

towns and

cities with populations greater than

5,000. That means

one

can't use

the calculations

in the PEATT for communities lacking addresses of

households

on municipal water

supply or

private

wells

(if exposure applies

to analyte[s]

of interest)

or those

with

<5,000 people.

•

The formulas can't be used if neither preliminary

nor NHANES data are available (e.g.,

for emerging

contaminants).

•

Need to have an idea

of non-response rate to use

the formula.

2. Two-stage sam

pling

The state

populatio

n is divided into groups (primary cluster). Groups are randomly selected, then persons within these groups are randomly selected and

included in the study.

Example 2A: Survey of the Health of Wisconsin — Two-stage cluster sampling

Reference: Nieto FJ, Peppard PE, Engleman CD, McElroy JA, Galvao LW, Friedman EM, et al. The Survey of the Health of Wisconsin (SHOW), a novel

infrastructure for population health research: rational and methods. BMC Public Health 2010; 10:785. doi: 10.1186/1471-2458-10-785

Link: https://www.med.wisc.edu/show/

Methods:

• The target population at initial selection is Wisconsin non-institutionalized/non-active duty adult residents aged 21 – 74 years

• The sampling frame is a list of census block groups or clusters of census block groups in Wisconsin state

• The initial sampling frame is constructed using Census 2000 data to generate 4,388 census blocks groups (CBG) or clusters of CBGs as the primary

sampling units (PSUs)

• CBGs with <40 households are merged with a neighboring CBG to form a cluster

• CBGs that fall in sovereign native American nation territories with no authorization to be included in survey are excluded

Last modified 05/07/2020

• PSUs are stratified according to 1) congressional district and 2) percentage of population living below 100% poverty level

• CBGs are randomly selected within each stratum using the Sampford explicit probabilities proportional to size without replacement method (36-72

CBGs usually are selected)

• A list of households addresses by CBG is generated using U.S. Postal Service sequence delivery files

• From the household sampling frame, 12-38 addresses are randomly selected using simple random sampling

Pros

• Provides a mechanism to

collect subjective (self-

reported) and objective

(physical exam and

biospecimen) individual

data.

Cons

• Results are limited to those from the same

communities.

• Requires many resources.

• Household canvassing requires significant staff

numbers to accomplish in timely manner.

Recommendations for use and modifications to consider

• SHOW infrastructure can be used to do specific biomonitoring

studies.

• To increase participation, a public relations campaign is

launched in communities 6-8 weeks before recruitment in

that location.

• The selected households are mailed an information package

1-3 weeks before the team arrives in the selected block

group. The package includes a description of the project, how

households were selected, and benefits of participation.

Example 2B: Minnesota East Metro PFAS Biomonitoring Projects —Two-stage cluster sample enrolling randomly-

selected adults likely to be exposed to PFAS and from a PFAS-affected community

Program: Minnesota Biomonitoring: Chemicals in People, Minnesota Department of Health.

Reference: Landsteiner A, Huset C, Johnson J, Williams A. Biomonitoring for perfluorochemicals in a Minnesota community with known drinking water

contamination. J Environ Health. 2014 Dec;77(5):14-9

Link: https://www.health.state.mn.us/communities/environment/biomonitoring/projects/pfas.html

Methods:

• The study location was a community with known PFAS exposures

• The sampling frame is a list of households from two groups in the community:

o People using community water— billing addresses of all households receiving municipal water service from city of Oakdale before January 1,

2005 (before the remediation)

o People using private wells— private well sampling results were used to identify all households with PFOS or PFOA above trace levels

Last modified 05/07/2020

•

A household survey was se

nt to a random sample of people on water billing records, and to all homes with contaminated private wells

• The households were as

ked to enumerate all eligible adults who lived there. Eligibility was restricted to those who lived in Oakdale before January 1,

2005

• A sample size, deter

mined by state legislature, was identified for the study

• Eligible adults wer

e randomly selected from the list to fit the selected sample size. For anyone who declined, replacements were selected

Pros

• Popu

lation-based sample.

• Results can be applied to others from

the same communities (those not

tested).

• Using an existing list as sampling frame

was an ec

onomically efficient means of

recruitment.

•

With extensive prior agency outreach

and co

mmunity interest, participation

rates can be high.

• These sampling frames work especially

well when

water is the source of

exposure.

• Able to assess some differences

betw

een participants and refusals.

Cons

• Res

ults are limited to thos

e from the same

communities and not the general

population.

• Renters and non

-homeowners are

underrepresented on water billing lists, so

may be missing important, possibly

vulnerable sub-populations. (This was

addressed in the third study by including a

group of renters, but it was difficult to find

a sampling frame and participation rates

were lower.)

Recomm

endations for use and modifications to consider

• Community outre

ach key to make residents familiar with

issue/study and likely to participate.

• Important to consider health equity issues and alternate

ways to include renters and non-homeowners.

• Don’t h

ave to restrict residence length.

Example 2C: Washington Environmental Biomonitoring Survey (WEBS) — Two-stage sampling to measure

exposure in the general population of WA state

Program: Washington State Department of Health

Links: https://www.doh.wa.gov/Portals/1/Documents/1500/WEBSOverview2013.pdf

https://www.doh.wa.gov/Portals/1/Documents/1500/WEBSFactSheet.pdf

Methods:

• The study area is Washington State

Last modified 05/07/2020

• The sampling frame is a list of census bl

ock groups in Washington state

•

70 census block groups were randomly selected from the state

• 27 housing units were ran

domly selected from each census block group

• All household residents aged 6 years and older were invited to participate in the survey

• 1422 participants from 666 households were included in the survey

Pros

• Populat

ion-based sample.

• Results can be applied to others from the

same communities (those not tested).

• Measured analytes also measured by

NHANES: speciated arsenic, 12 metals

(antimony, barium, beryllium, cadmium,

cesium, cobalt, lead, molybdenum,

platinum, thallium, tungsten and

uranium), pesticide metabolites

(chlorpyrifos, and four pyrethroids),

bifenthrin metabolites, BPA and

phthalates.

Cons

• Children <6 years were

not included

in the sample.

Recommenda

tions for use and modifications to consider

• Online publication of the re

sults by population

characteristics, which the local health departments and

community can access to understand the chemicals

opulations are being exposed to.

Example 2D: Community Assessment for Public Health Emergency Response (CASPER) — Two-stage cluster

sampling design in which 30 clusters are selected and then seven interviews are completed in each of the 30

clusters

Program: Centers for Disease Control and Prevention (CDC)

Link: https://www.cdc.gov/nceh/hsb/disaster/casper/overview.htm

Methods:

• A sampling frame boundary is defined and includes a list of census blocks in that boundary

• A sample of 30 census blocks, weighted by household numbers (using information from the U.S census), is selected from within the sampling frame.

Use sampling with replacement

• Census blocks are visited by on-the ground teams over 3 days to gather seven interviews per census block

Last modified 05/07/2020

• Each interview asks about everyone within the entire household

• Residential units are selected systematically by dividing total units (using census information) by 7 and visiting each nth residence. Units are replaced

by adjacent units after three failed attempts at contact

• Results are weighted based on census block size and completion rate (if unable to achieve seven interviews in each block)

Pros

• Some states/counties/localities

already have experience in

using CASPER for post-disaster

health surveillance and disaster

preparation.

• A national disaster

epidemiology coalition is

available for support.

Cons

• A list of all clusters within the sampling frame is

required.

• Household canvassing requires significant staff

numbers to accomplish in timely manner.

Recommendations for use and modifications to consider

• CASPER sampling methodology can be used to identify

potential study participants. However, other appropriate and

more feasible methods must be used to contact those

persons and to collect their samples for the study (e.g., via

mail).

• In rural areas, clusters might have fewer than seven

households which might require modifying the traditional 30

(clusters) x 7 (households) design of CASPER. Rural area

clusters (census blocks) with fewer than seven households

might make it hard for sampling teams to interview the

needed number from that cluster.

Example 2E: Optimizing cluster survey design for planning schistosomiasis preventive chemotherapy —School-

based two-stage cluster sampling implementation

Reference: Knowles SCL, Sturrock HJW, Turner H, Whitton JM, Gower CM, Jemu S, et al. Optimising cluster survey design for planning schistosomiasis

preventive chemotherapy. PLoS Negl Trop Dis 2017; 11(5): e0005599.

Link: https://doi.org/10.1371/journal.pntd.0005599

Methods:

• The sampling frame was a list of primary schools in a health or educational school district

• 15-20 primary schools were randomly selected, and then 30 children were randomly selected per school

• Children aged 10 - 14 years were eligible for participation

• Eligible children were line up and a sampling interval was used to select the required 15 participants of each sex

• The sample size was determined based on precision-based sample size calculations that took into account spatial heterogeneity in prevalence of

schistosomiasis using methods from Lohr, S. Sampling: design and analysis. Boston , MA: Cengage Learning; 2010

Last modified 05/07/2020

Pros

• Could use schools as

distribution points for

recruitment material and

collection points for

samples.

Cons

• Designed to sample school age children. Would

need to send information home with children to

families to recruit for sampling. Would be biased

toward people with school-age children.

Therefore, this structure might not be suitable for

other biomonitoring studies.

• Home-schooled children are left out of sampling

frame.

• Study was based on detection of schistosomiasis

endemicity. The number needed to sample will

likely not translate exactly for biomonitoring.

Recommendations for use and modifications to consider

• Could be modified at the levels of sampling such that schools

are randomly selected from within a county or a primary care

area.

• Could also do probability-based sampling for the first stage

instead of simple random sampling to oversample more

populated areas or certain populations of interest.

Example 2F: Two-stage cluster sampling based on census block and household units

Reference: Smith DA, Akira A, Hudson K, Hudson A, Hudson M, Mitchell M, Crook E. The effects of health insurance coverage and the doctor-patient

relationship of health care utilization in high poverty neighborhoods. Prev Med Rep 2017; 7:158–61.

Link: https://doi.org/10.1016/j.pmedr.2017.06.002

Methods:

• The target population is defined based on census tract

• The primary sampling frame is a list of census blocks in a defined census tract, and the primary sampling unit is randomly selected census blocks from

within the tract, where the number of census blocks selected is proportional to the number of blocks in the tract

• The secondary sampling unit is five randomly selected housing units from within each sampled census block

• The selected houses are contacted either by letter or a home visit

Pros Cons Recommendations for use and modifications to consider

Last modified 05/07/2020

• By randomly selecting

houses from within a census

block, eliminates the labor-

intensive need to go door-

to-door and could recruit by

mailer.

• Relatively simple strategy.

• Recruitment of five selected households might be

difficult if by mail (generally low response rate).

• Five randomly selected houses might not be

enough.

• Could be modified to where sampling frame is county, primary

sampling unit is census tract, and secondary unit is census

block. Entire blocks could be contacted by overlaying blocks

with a GIS layer with address and zip, and send mailers

targeting census block.

3. Three-stage cluster sampling

The population is divided into primary, secondary and tertiary sampling units (e.g., counties, cities, households). First, a sample of primary units is randomly

selected. A sample of secondary units is then randomly selected from the selected primary units, then a sample of tertiary units is randomly selected from

the selected secondary units.

a. Three-stage cluster sampling at the city-level

Example 3A: New York City Health and Nutrition Examination Survey (NYC-HANES)

Last modified 05/07/2020

Links: http://nychanes.org/wp-content/uploads/sites/6/2015/11/NYC-HANES-Training-Slides_part-1_08222016.pdf

https://www1.nyc.gov/assets/doh/downloads/pdf/hanes/hanes-manual.pdf

https://www.ncbi.nlm.nih.gov/pubmed/26844121

https://www.ncbi.nlm.nih.gov/pubmed/16776895

Methods:

•

New York City i

s divided into about 20,000 segments, then 144 segments are randomly selected; stratified by borough

• The primary sampling unit is segments which are an aggregation of census block groups (probability proportional to size)

• The secondary sampling unit is households within the primary sampling unit (simple random sample)

• The tertiary sampling unit is participants within households (simple random sample)

Pros

• Instruments and laboratory

methods benchmarked

against the national HANES.

Cons

• Only city-wide estimates available.

• Small sample sizes for some demographic

subgroups.

•

Selection/non-response bias (partially corrected

through non-response weighting adjustment).

Recommendations for use and modifications to consider

• Can be modified to where the sampling frame is the state.

4. Recruit from an existing population-based survey or program

Last modified 05/07/2020

Some programs may wish to recruit from existing population-based surveys by asking to add interest-gauging questions for participation in a biomonitoring

study. Information on persons who agree to participate would be transferred to the biomonitoring study team, and separately followed-up on for

participation.

a) The Behavioral Risk Factor Surveillance System (BRFSS) is an example of an existing population-based survey that some states have worked with. BRFSS

uses a disproportionate stratified sample design. Landline users are divided into two groups (high-density and medium-density), which are sampled

separately. To provide adequate sample sizes for smaller geographically defined populations of interest, many states sample disproportionately from

strata that correspond to sub-state regions. Since 2011, BRFSS has been conducting surveys by landline and by cellular telephone. In conducting the

landline telephone survey, interviewers collect data from a randomly selected adult in a household. In conducting the cellular telephone survey,

interviewers collect data from those adults who answer the cellular telephone call and lives in a private residence or college housing.

Example 4A: Behavioral Risk Factor Surveillance System (BRFSS) – Disproportionate stratified sample design.

Program: Centers for Disease Control and Prevention (CDC)

Methods:

• In conducting the landline telephone survey, interviewers collect data from a randomly selected adult in a household

•

Landlines are divided into tw

o groups (high-density and medium-density), which are sampled separately

• In conducting the cellular telephone survey, interviewers collect data from adults answering the cellular telephones residing in a p

rivate residence or

college housing

• State health departments collaborate during survey development and co

nduct the interviews themselves or use contractors

Pros

• D

oes not require many

res

ources for sampling.

• Can add questions.

Cons

• N

on-response bias could be large.

• E

ven though BRFSS participants are randomly

elected, the biomonitoring sampling frame

includes only those who participate in BRFSS, and

then say yes to participating in the biomonitoring

program. The program then follows up with these

people.

Rec

ommendations for use and modifications to consider

• Can be costly to contact all interested participants to collect

info

rmation on them:

o Because the number identified from the B

RFSS survey

could be large (>2,000), consider using a call center to

contact all interested participants.

o Also consider having a website for i

nterested

participants, where they can provide more

information about themselves (e.g., demographic

Last modified 05/07/2020

• Can be costly for some states to add

biomonito

ring questions to the BRFSS

questionnaire.

• Can be a significant time delay between

when

people are surveyed by BRFSS and when they are

re-contacted by the

biomonitoring staff.

information)

to reduce costs of contacting

each

participant

to collect

this information.

o The website can

provide

inter

ested participants

with

information

what samples are needed and where

to drop

them

off (e.g., at a hospital). This requires

partnership with health care providers

or local health

departments

collect and store samples.

Example 4B: Minnesota Healthy Rural and Urban Kids Project — One-stage cluster sample enrolling preschool-

aged children from selected communities in the state.

Program: Minnesota Biomonitoring: Chemicals in People, Minnesota Department of Health.

Link: https://www.health.state.mn.us/communities/environment/biomonitoring/projects/ruralurbankids.html

Methods:

• The study location is two geographically distinct communities in the state with different community exposure concerns: 1) group of rural counties in

North-Central Minnesota, and 2) two ZIP codes in urban North Minneapolis.

•

The sampling frame is prescho

ol-age children who live in these areas and came in for their required pre-kindergarten early childhood screening visits

during June – September 2018. Early childhood screening programs are administered by school districts (Minneapolis) or, in some cases, local public

health agencies (rural counties in the study).

• Early childhood screening staff recruited children and conduct

ed data collection:

o Staff contacted parents with appointments scheduled and informed t

hem about possibility to participate.

o On the day of visit,

staff introduced the study again to families, conducted informed consent, administered the survey while child was being

screened, and collected the urine sample.

• Eligibility was restricted to childre

n whose parents reported that their child could provide a urine sample (i.e., potty-trained and no health issues that

would interfere), and to one child per family.

• A sample size for study was identifie

d, and monthly target recruitment goals were set.

Pros

• Children are re

quired to complete Early

Childhood Screening before entering

kindergarten, so the sampling frame is based

Cons

• Results are limited to children from

the selected communities and not to

the state-wide population.

Recommendations for use and modifications to consider

• Because program happens across the state, can tailor

geography to meet study/community needs.

Last modified 05/07/2020

on an existing, universal program for

preschool-aged children.

• Contracting with local staff from existing

programs was economically efficient and

helped achieve high participation rates

–

people

knew and trusted them.

• Being able to complete all components

the study when

families came in for their

pre-scheduled screening visit increased

retention, survey completion and sample

collection.

• Study embedded in

existing public health

programs that

focus on

children’s health and

have strong interest in results.

• Collecting a blood sample would be

more complicated and resource-

intensive with this design.

• Lack of comparison data from NHANES

and other sources for this age group.

Note: NHANES samples children from

ages 3 years and older for some

analytes (not all).

5. Over sampling populations

Oversampling is done to increase the reliability and precision of estimates of certain population subgroups.

For example, studies such as the National Health and Nutrition Examination Survey (NHANES) used oversampling to sample larger numbers of

subgroups of interest such as minorities, adolescents, and older adults [16].

This increases the reliability and precision of estimates in these population

subgroups [16]. Rural populations may be another subgroup of interest when considering oversampling to help ensure that you achieve a large enough

N value for appropriate statistical power.

nsideration of oversampling clusters before the first or second stage provides a benefit for netting a desired sample size:

 Possible oversampling methods could include increasing the number of units (e.g. census blocks) in your first stage [22].

In the second stage the

number of units (e.g. households) also can be increased [22].

 An

example of oversampling to increase cluster selection is a modified application of the CASPER study design whereby the 30 × 7 design is

modified to a 35 × 7 therefore increasing the N value by 16.6% [23].

The NHANES study design, as another example, draws its sample in the following stratified, four stages [24]:

• Stage 1: PSUs are first stratified according to population size, and then PSUs are selected from each stratum. These are mostly single counties

or, in a few cases, groups of contiguous counties with probability proportional to a measure of size (PPS).

Last modified 05/07/2020

• Stage 2: The PSUs are divided up into segments (generally city blocks or their equivalent). As with each PSU, sample segments are selected with

PPS.

• Stage 3: Households within each segment are listed, and a sample is randomly drawn. In geographic areas where the proportion of age, ethnic,

or income groups selected for oversampling is high, the probability of selection for those groups is greater than in other areas.

• Stage 4: Individuals are chosen to participate in NHANES from a list of all persons residing in selected households. Individuals are drawn at

random within designated age-sex-race/ethnicity screening subdomains. On average, 1.6 persons are selected per household.

In stage 3, where households from each segment are randomly drawn, for geographic areas where the proportion of age, ethnic, or income groups

selected for oversampling is high, the probability of selection for those groups is greater than in other areas [16]. This can be replicated in other

iomonitoring studies that apply probability sampling to oversample populations of interest.

For su

b-populations that are at least 10% of the total population, a general sample will usually produce reliable estimates [21]. F

or, subpopulations

between 1% and 10% of the total population, the oversampling methods described in the tables below are needed [21].

Example 5A: Screening—Used when sampling frame does not have the subpopulation domain identifiers.

Reference: Kalton G. Methods for oversampling rare subpopulations in social surveys. Survey Methodol. 2009;35(2):125–41

Methods:

•

Used if a sampled person’s membership in a rare population can be determined inexpensively, e.g., from responses to a few questions

• A large first phase sample size is identified from wh

ich to select members of subpopulation of interest

• This is the minimum sample size th

at will produce the required (or larger) sample sizes of subpopulations of interest

• All the members of the subpopulations o

f interest are then included in the second phase sample

Pros

•

Uses probability sampling

methods.

Cons

• Large sample size required.

• Requires expensive procedures.

• With screening, the sample size for a rare

pulation is a random variable and could be

more or less than expected.

• Noncoverage can be a significant problem when

large-scale screening is used to identify rare

populations.

Rec

ommendations for use and modifications to consider

• A large screening sample size is required to get an adequate

sample size for the subpopulation(s) of interest,

o Use an inexpensive mo

de of data collection (e.g.,

telephone surveys or mail questionnaire).

o Share costs across surveys – e.g

., partner with

existing population-based surveys

• When possible and useful, pe

rmit the collection of screening

data from persons other than those sampled. For example,

Last modified 05/07/2020

o Even though a frame has good ov

erall

coverage, its coverage of a rare domain

might be inadequate

other household members may be able to accurately report

the rare population status of the sampled member.

•

Use a sampling fraction to be

st ensure the sample size will

meet the minimum needed for a rare population. Sample

fractions can be in context of subdomain such as age, sex or

income groups.

• Consider non-response rates.

• Allow for under-rep

resentation by taking into account issues

such as noncoverage, non-response or misclassification of

domain membership at the design stage to produce the

required sample size.

Example 5B: Disproportionate stratification—Identifying strata based on subgroups of interest

Reference: Kalton G. Methods for oversampling rare subpopulations in social surveys. Survey Methodol. 2009;35(2):125–41.

Methods:

• Used if the rare population is concentrated in certain identifiable parts of the sampling frame

• Ma

ke sure subpopulation is more prevalent in the strata

• Make sure the strata has a high proportion of subpopulation

• The cost of collecting data from the subpopulation should not be high

Pros

• Disproportionate

stratification, with higher

sampling fractions in the

strata where th e prevalence

of the rare population is

higher, can reduce the

amount of screening

needed.

Cons

• Assumes the prevalence of the rare population in

each stratum is known.

• Assumes the identification of rar

e populations is

relatively easy.

• Errors in the prevalence es

timates will reduce the

precision gains achieved with disproportionate

stratification and could even result in a loss of

precision for the survey estimates.

Rec

ommendations for use and modifications to consider

• This method will yield substantial gains in efficiency if the

following conditions hold:

o The rare population mu

st be much more prevalent in

the oversampled strata

o The over-sampled strata must contain a high

roportion of the rare population

o The cost of the

main data collection per sampled unit

must not be high

• If a list with names is accessible (e.g., telephone numbers

erged with U.S. Postal Service delivery sequence file

addresses), the names can be used to build strata of likely

Last modified 05/07/2020

members of some racial/ethnic groups. Nam

es can be

effective for identifying certain ethnic groups.

•

If area sampling is used, data from the

last census and other

sources can be used to allocate the area clusters to strata

based on their prevalence estimates for the rare population.

Example 5C: Two phase sampling—Used when accurate identification of subpopulation members is expensive

Reference: Kalton G. Methods for oversampling rare subpopulations in social surveys. Survey Methodol. 2009;35(2):125–41.

Methods:

• Used if a sampled person’s membership in a rare population can only be determined using expensive methods, e.g., medical examinations

• Start with an imperfect screening classification in the firs

t phase e.g. screening questionnaires with high sensitivity to detect subpopulation of interest

• Follow with accurate identification for a disproportionate stratified subsample in sec

ond phase (e.g., assessment by medical personnel)

Pros

• Ben

eficial when first phase

accu

rate identification is not

known or expensive.

Cons

• A fairly

common practice with two-pha

se designs

is to take no second-

(or third-) phase sample from

the stratum of

those classified as nonmembers

the rare domain, based on

their responses at

the

previous stage. If the prevalence

of the rare

domain

more than

minimal in this stratum, a

substantial proportion of the domain

may go

unrepresented.

Recomm

endations for use and modifications to consider

• The imperfect screening methods in

the

first phase should be

less expensive than

the accurate identification in

the

second

phase. I

f area sampling is used, data from the last census and

other

sources can be used

to allocate the area clusters to

strata based on their prevalence estimates for the rare

population.

Last modified 05/07/2020

REFERENCES

1. Needham L, Calafat AM, Barr D. Uses and issues of biomonitoring. Int J Hyg Environ Health 2007;

210(3–4):229–38.

2. Centers for Disease Control and Prevention. The fourth national report on human exposure to

environmental chemicals. Atlanta, GA: Centers for Disease Control and Prevention; 2009.

Available from: https://www.cdc.gov/exposurereport/pdf/fourthreport.pdf

3. Namulanda G. Biomonitoring and Environmental Public Health Tracking. Journal of

Environmental Health 2015; 77(9).

4. Council of State and Territorial Epidemiologists. Biomonitoring in public health: epidemiologic

guidance for state, local, and tribal public health agencies. Atlanta, GA: Council of State and

Territorial Epidemiologists; 2012. Available from:

http://www.cste2.org/webpdfs/BioMonISFINAL.pdf

5. Lohr SL. Sampling: Design and Analysis. 2

Edition. Brooks/Cole Cenage Learning

6. Sullivan KM. Sampling for epidemiologists. Atlanta, GA: Rollins School of Public Health, Emory

University; 2010.

7. Kalton G. Introduction to survey sampling. Quantitative applications in the social sciences. Vol

35. Thousand Oaks, CA: Sage Publications; 1983.

8. Levy PS and Lemeshow S. Sampling of Populations; Methods and Applications. John Wiley and

Sons Inc

9. Gaganpreet S. Pros and cons of different sampling techniques. Int J Appl Res. 2017;3(7):749–52.

10. Centers for Disease Control and Prevention; Agency for Toxic Substances and Disease Registry.

Per- and Polyfluoroalkyl Substances (PFAS) Exposure Assessment Technical Tools. Atlanta, GA:

Centers for Disease Control and Prevention; Agency for Toxic Substances and Disease Registry;

2018. Available from:

https://www.health.pa.gov/topics/Documents/Environmental%20Health/PFAS%20Exposure%2

0Assessment%20Technical%20Tools.pdf.

11. Centers for Disease Control and Prevention. Community Assessment for Public Health

Emergency Response (CASPER) toolkit. Third edition. Atlanta, GA: Centers for Disease Control

and Prevention; 2019. Available from:

https://www.cdc.gov/nceh/hsb/disaster/casper/docs/CASPER-toolkit-3_508.pdf

12. Nafiu LA, Oshungade IO, Adewara. Alternative estimation method for a three-stage cluster

sampling in finite population. Am J Math Stat. 2012;2(6):199–205.

13. Schneider KL, Clark MA, Rakowski W, Lapane WL. Evaluating the impact of non-response bias in

the Behavioral Risk Factor Surveillance System (BRFSS). J Epidemiol Comm Health

2012;66(4):290–5.

14. Galea S,

Tracy M.

Participation rates in epidemiologic studies. Ann Epidemiol. 2007;17(9):643–

53.

15.

Association of Public Health Laboratories. Guidance for Laboratory Biomonitoring Programs:

Developing Biomonitoring Capabilities. 2019. Available from:

https://www.aphl.org/aboutAPHL/publications/Documents/EH-Oct2019-Biomonitoring-

Guidance.pdf

16. National Center for Health Statistics. National Health and Nutrition Examination Survey

(NHANES) Tutorials. Module 2: Sample Design. Hyattsville, MD. Available from:

https://wwwn.cdc.gov/nchs/nhanes/tutorials/Module2.aspx

17. Groves R et al. Survey Methodology. 2

Edition. 2009. John Wiley and Sons

Last modified 05/07/2020

18. Chen TC, Clark J, Riddles MK, Mohadjer LK, Fakhouri THI. National Health and Nutrition

Examination Survey, 2015-2018: Sample design and estimation procedures. National Center for

Health Statistics. Vital Health Stat 2(184). 2020. Available from:

https://www.cdc.gov/nchs/data/series/sr_02/sr02-184-508.pdf

19. ICF International. Demographic and Health Survey Sampling and Household Listing Manual.

MEASURE DHS, Calverton, Maryland, U.S.A; 2012.

20. Global Adult Tobacco Survey Collaborative Group. Global Adult Tobacco Survey (GATS): Sample

Weights Manual, Version 2.0. Atlanta, GA: Centers for Disease Control and Prevention, 2010.

21. Kalton G. Methods for oversampling rare subpopulations in social surveys. Survey Methodol.

2009;35(2):125–41.

22. Repp KK, Hawes E, Rees K, Vorderstrasse B, Mohnkern S. Lessons learned from an

epidemiologist-led countywide community assessment for public health emergency response

(CASPER) in Oregon. J Public Health Manag Pract. 2019;25(5):472–8.

23. Centers for Disease Control and Prevention. Sampling methodology. Community Assessment for

Public Health Emergency Response (CASPER). Atlanta, GA. Available from:

https://www.cdc.gov/nceh/casper/sampling-methodology.htm

24. Johnson CL,

Dohrmann SM, Burt VL, Mohadjer LK. National Health and Nutrition Examination

Survey: Sample design, 2011–2014. National Center for Health Statistics. Vital Health Stat

2(162). 2014

.. Available from: https://www.cdc.gov/nchs/data/series/sr_02/sr02_162.pdf.

_____________________________________________________________________________________

Last modified 05/07/2020

Suggested citation: Environmental Public Health Tracking Program, Centers for Disease Control and

Prevention. Guidance, examples and tools for probability sampling when designing a population-based

biomonitoring study. 2020.

CONTRIBUTORS

Biomonitoring Study Design Team, Environmental Public Health Tracking Program, Centers for Disease

Control and Prevention.

Name Affiliation

Gonza Namulanda CDC–Division of Environmental Health Science and Practice

Kristin Dortch CDC–Division of Laboratory Sciences

Wellington Onyenwe CDC–Division of Laboratory Sciences

Karen Wilson NY Tracking Program

Monica Nordstrom NY Tracking Program

Matt Roach AZ Tracking Program

Hsin Lin Cox AZ Tracking Program

Niki Lajevardi-Khosh AZ Tracking Program

Melissa Kretchmer AZ Tracking Program

Amanda Cosser NH Biomonitoring program

Jessie Sagona NH Tracking Program

Cliff Mitchell MD Tracking Program

We acknowledge and thank all reviewers from the Centers for Disease Control and Prevention’s National

Center for Health Statistics, Division of Laboratory Sciences, and Environmental Public Health Tracking

Program (Tracking Program); the Agency for Toxic Substances and Disease Registry (ATSDR); and the

Tracking Program State/local recipients for their feedback on improving this document. We also

acknowledge and thank Nerissa Wu, Kathleen Attfield and Jennifer Mann from the California

Department of Public Health for their participation on the Biomonitoring Study Design Team, and for

their input in developing this document.

ENVIRONMENTAL PUBLIC HEALTH TRACKING