© 2011 The Royal Statistical Society

drugs commonly used, though patterns of use appear to vary substantially from place to place due to differ-ences in both supply and demand. The use of drugs, notably opiate-related compounds, has grown as opiate availability increases with more prescribing. Synthetic stimulants and hallucinogens are increasingly entering the market around the world, according to the United Nations 2009 World Drug Report. Consumers often do not know exactly what they are taking, and public health and law enforcement are often caught flat-footed by the lack of data on new and potentially dangerous drugs.

Current drug use measurement tools

Current ways of measuring drug use at the population level have many limitations. They include population coverage: how can you reach every drug user? They rely upon self-reporting, there is measurement and selec-tion bias, and substantial time lags greatly reduce the reliability, validity and utility of data2–4. When we think of measuring drug use in a population we generally think of surveys; but survey data have several major weaknesses. They are expensive; the social stigma of drug use brings response bias – few users will admit to a

City-wide drug testing using municipal wastewaterA new tool for drug epidemiology

Background

Drug use is just a big-city problem. Or is it? Measur-ing actual drug use is notoriously difficult. Measuring the effects of drug use is easier – fatal overdoses, drug treatment admissions, emergency department visits, new cases of hepatitis and HIV. But by the time these measurable effects occur and a drug takes hold in a community, any chance to intervene has long since been lost. Similarly our ability to measure the impacts of major drug policy changes or interventions, whether through law enforcement or public health measures, is greatly weakened by our poor ability to measure the true level of drug use for an entire community, in real time or with any resolution into time intervals. Has use in this borough or this suburb gone up or down this year? Does it happen more at weekends or midweek? We generally do not know – and it would be very useful indeed if we could know.

Drug use continues to be a major public health problem throughout the world. In the United States unintentional drug overdoses surpassed traffic-related fatalities in 16 US states in 20061. Marijuana, cocaine, heroin and methamphetamine continue to be the

Drug users do not want to be identified and do not want to reply to surveys. So how do you find out how many of them there are? The answer lies below the city streets. Caleb Banta-Green and Jennifer Field delve deep.

No questionnaires are needed. Data from everyone is

sampled70 june2011

71june2011

surveyor with a clipboard that they have been injecting drugs that morning. Many users are incarcerated or otherwise on the margins of society, which makes it hard to contact them. Other indirect indicators of drug use are often “consequence” data such as emergency department visits, drug treatment admissions, or fatal overdoses. Reviews of limitations of current drug use data are available3, 4. Some of those limitations are summarised in Table 1, using methamphetamine as an example.

A new tool

Caleb Banta-Green, the drug use epidemi-ologist on this interdisciplinary team, was grappling with these severe data limitations for several years as he worked to expand drug epidemiology coverage. He was reporting from the relatively data-rich area of Seattle and wanted to expand coverage to the rest of Washington State. An environmental chemist, Jennifer Field, called him out of the blue in 2006, describing her recent experi-ence testing wastewater in Oregon and her surprise at the amount of methamphetamine detected; she wondered whether this type of information was valuable and what it might mean to drug epidemiologists. Peanut butter meets chocolate. Why was this a match made in heaven?

Municipal wastewater treatment systems – sewerage works in the vernacular – collect waste from residential and commercial customers throughout their catchment areas

and transport it via sewer lines to centralised treatment plants. Plants routinely collect samples of their inflow for their own regula-tory and quality control purposes. These samples are typically collected over a 24-hour period, and automated equipment is used to generate a sample that is representative of the entire day. The equipment collects wastewa-ter subsamples on either an elapsed time (e.g. once every hour) or a flow-proportional (e.g. a sample every set number of litres) basis. Acquiring samples for research purposes is therefore efficient and inexpensive. Most of the work has already been done. The plant operators simply transfer a small volume (15 ml) of the 24-hour composite to a shatter proof vial that is pre-filled with preservative (or they freeze the sample) and ship it rapidly to a laboratory for chemical analysis. For a scientist, collecting samples is often a major cost; here we are able to build off existing infrastructure and efforts and obtain samples at close to no cost, other than materials to collect and ship samples.

Most researchers are interested in what flows out of a wastewater treatment plant. They look at it after it has been treated, and want it as clean as possible because of its potential downstream impacts on the aquatic environment, and, eventually, on the hu-man environment. However, examining the inflow to a plant is a way to look upstream to (potentially) describe unfiltered, literally and figuratively, human behaviour. Wastewater testing has identified pharmaceuticals that

have passed through the human body5, and applying wastewater testing to abused drugs was first proposed in 20016 and implemented in 2005 in Italy7.

Detecting drugs and other substances of interest

Detecting extremely low concentrations (nanograms per litre or parts per billion) of substances has become increasingly easier in recent years. Dr Field’s laboratory uses state-of-the-art tandem mass spectrometry equipment and analytical procedures to iden-tify specific substances, their metabolites, and human urinary indicators. Her laboratory has advanced the state of the science in this area for the detection of licit and illicit drugs in wastewater, including the development of more efficient and cost-effective analytical procedures8. Because of the extreme sensitiv-ity of this equipment, the amount of sample that needs to be collected and analysed is very small, which makes collection, shipping, stor-age and analysis even more efficient.

Does it work?

We have demonstrated the validity of data collected from 96 cities in Oregon for discern-ing rural and urban patterns of drug use, using the geographic and spatial data that is available in wastewater9. The total popula-tion served by the 96 participating treatment plants was 2 478 168, approximately 65% of the state’s population. The smallest munici-pality served by a single treatment plant had a population of 170 people, the largest 562 690, with a median of 5595, so this was the range of the population groups that we could analyse. We used the Rural Urban Commut-ing Area (RUCA) classification scheme10 to characterise each community: we classified 36 municipalities as urban, 26 as large rural cities or towns and 34 as small rural towns. We analysed for methamphetamine and for cocaine – the latter is metabolised in the human body into benzoylecgonine (BZE), which is what we measure to indicate cocaine use. For each of them we could draw maps of high, medium or low usage at each treatment plant’s area. Comparing them, we found their patterns of usage were very different.

Figure 1 shows the results. The spatial data for BZE, reported as index loads (in milligrams per person per day, adjusting for stated population and wastewater flow), were Aerial view of a modern water treatment plant. © iStockphoto.com/Dan Barnes

72 june2011

Tabl

e 1.

Maj

or s

ourc

es o

f ep

idem

iolo

gica

l dat

a on

met

ham

phet

amin

e, a

nd t

he m

ajor

lim

itat

ions

of

thei

r us

e

Data

nam

ePo

pula

tion

Num

ber

of

even

ts/s

ize

of

popu

lation

Data

in

terv

al

Tim

e la

g fo

r da

ta

avai

labi

lity

Geog

raph

yTe

rmin

olog

y M

ajor

str

engt

hsM

ajor

lim

itat

ions

Drug

Abu

se

War

ning

Net

wor

k,

surv

eilla

nce

syst

em

Emer

genc

y de

part

men

t pa

tien

ts

Hun

dred

s–th

ousa

nds

Annu

al12

mon

ths

3 Co

unty

Met

ro

Area

Spec

ific

drug

nam

e m

etha

mph

etam

ine

Popu

lati

on-b

ased

es

tim

ates

An

nual

tre

nd d

ata

Self-

repo

rtin

g bi

ases

To

xico

logy

oft

en n

ot d

one

Nat

iona

l tel

epho

ne

surv

eys

Adul

tsH

undr

eds–

tens

of

tho

usan

dsAn

nual

1–2

year

sRe

gion

/sta

teSp

ecifi

c dr

ug n

ame

met

ham

phet

amin

ePo

pula

tion

-bas

ed

esti

mat

es

Annu

al t

rend

dat

a

Self-

repo

rtin

g bi

ases

H

ighe

st d

rug

usin

g po

pula

tion

s of

ten

not

in

sam

ple

Publ

ic s

choo

l su

rvey

sSt

uden

tsH

undr

eds–

thou

sand

sBi

-ann

ual

12 m

onth

sCi

tySp

ecifi

c dr

ug n

ames

m

etha

mph

etam

ine

Anon

ymou

s, s

elf-

repo

rt

surv

ey,

larg

e sa

mpl

eOu

t-of

-sch

ool y

outh

mis

sing

So

cial

des

irabi

lity

repo

rtin

g bi

as

Drug

tre

atm

ent

adm

issi

ons

Publ

icly

fu

nded

tr

eatm

ent

Hun

dred

s–th

ousa

nds

Ongo

ing

4 m

onth

s5-

digi

t zi

p co

des

Spec

ific

drug

nam

es

met

ham

phet

amin

eIn

dica

tion

of

prob

lem

atic

use

of

drug

s

Larg

e nu

mbe

rs

Trea

tmen

t ty

pica

lly la

gs d

rug

use

onse

t by

yea

rs

Mid

dle

and

uppe

r in

com

e pe

rson

s no

t in

clud

ed

Fata

l dru

g ov

erdo

ses

–

med

ical

exa

min

er

and

toxi

colo

gy la

b

All s

udde

n,

unex

pect

ed

and

unna

tura

l de

aths

Tens

–hun

dred

sOn

goin

g4

mon

ths

5-di

git

zip

code

sPr

ecis

e ch

emic

al

nam

es.

Quan

tita

tive

che

mis

try

Popu

lati

on-b

ased

, an

nual

tre

nd d

ata

Diffi

cult

to

assi

gn c

ausa

tion

to

spe

cific

dru

g in

mul

ti-d

rug

case

s

Com

mun

ity-

base

d su

rvey

sVa

ries

– ev

ent

or lo

cati

on

base

d

Tens

–hun

dred

sOn

e ti

me

Shor

t/m

onth

sVa

ries

Deta

iled

nam

es a

nd

slan

g te

rms,

e.g

. cr

ysta

l, cr

ank,

met

h

Patt

erns

of

use,

co

nseq

uenc

esCo

nven

ienc

e sa

mpl

e

One-

tim

e su

rvey

So

cial

des

irabi

lity

repo

rtin

g bi

as

Was

tew

ater

tr

eatm

ent

plan

tsTo

tal

popu

lati

on

on s

ewer

Hun

dred

s–te

ns

of t

hous

ands

Varia

ble/

flexi

ble

Wee

ksM

unic

ipal

ity/

fle

xibl

ePr

ecis

e ch

emic

al

nam

esPo

pula

tion

-bas

ed

Dire

ct m

easu

re

Aggr

egat

ed d

ata

Prec

isio

n to

be

dete

rmin

ed

Accu

racy

to

be d

eter

min

ed

73june2011

significantly different across different types of community. Larger loads were significantly more likely to occur in more urban areas. In contrast, methamphetamine was present at quantifiable concentrations in raw influent from every treatment plant location, urban or rural. The distribution of index loads for methamphetamine was statistically equivalent across community types (RUCA classifica-tions) and there was no trend in drug index load by RUCA code. These quantitative data were analysed as ordered categorical data, because some samples yielded results that were below the level of quantification or were below detection and because all of the elements of the proper error term had yet to be assembled.

This novel research demonstrates the feasibility of simultaneous collection of a rela-tively large number of unbiased samples from large and small communities for testing for multiple substances. Most of the participating municipalities were too small to be captured by conventional measures of drug use. The observed geographic distributions were con-sistent with generally expected urban–rural patterns based on conventional drug use indi-cators, including drug treatment admissions, morbidity and mortality data11–13, which is an indication that our technique is valid.

Scientists on at least three continents, Europe, North America and Australia, are actively working on developing and refining the method. Recent research has examined levels of drug excretion between cities and differential patterns of excretion for various compounds by day of the week14,  15. We are currently determining sampling approaches

to accurately and efficiently determine annual estimates of drug excretion.

Wastewater-derived data indicate variability in index loads and geographic distributions within and between drugs. The computed index loads represent a quantifiable measure of community drug excretion that is not a threat to individual privacy and that is not impacted by self-report bias. These two alone are a great advance. Wastewater treatment plant data provide fine geographic detail. It also gives substantial coverage of the population, given that the majority of the US population is connected to sewers16 as is much of the industrialised world.

Strengths and weaknesses of wastewater testing for drugs

While the analytical methodology is very sensitive, there are some compounds that are difficult to detect. Some compounds are consumed at such low levels that they cannot be detected – benzylpiperazine is one – and others are metabolised into non-specific com-pounds. For example, heroin metabolises into morphine, which can also be present due to legitimate pharmaceutical usage. Wastewater analysis cannot distinguish between legiti-mate and illegitimate use.

In order to generate a per capita estimate of excreted drugs, a population estimate is necessary. To date, most researchers have used a fixed population estimate for a municipality, typically one based upon census data. Anyone who has lived in a place for any period of time knows that the number of people can shift

substantially from midweek to weekend, at holiday times, as well as seasonally. Ideally, population can be estimated from the same wastewater sample used for drug analysis. If every person produces on average the same amount of a compound, measuring the concentration of the compound will let us cal-culate the population. Which compounds are most chemically stable and best approximate to total populations has yet to be determined.

Missing data

A major issue is missing data, or censoring. As with other environmental samples, such as air pollutants, working with the resulting data can be tricky. Often no compound may be detected, or the level at which it is detected is so low that assigning an accurate quantitative value is not possible. These data are consid-ered to be censored; there is a cut point below which data appear to be missing and there are different reasons for this “missingness”. Pre-liminary data from a study we are conducting in Oregon and Washington yielded two cities with the following average per capita loads of MDMA (commonly called “ecstasy”): 0.007 mg/person/day for City 1 and 0.008 mg/person/day for City 2. At first, these appear to be similar averages. The problem is that they are inappropriate as summaries of drug use as they do not incorporate the fact that the pro-portion of weekly samples with no detectable MDMA was 16% for City 1 and 58% in City 2 in 2009. Information is contained in the pattern of censoring and statistical methods exist for censored data. While there is debate

Figure 1. Drugs detected in Oregon municipal wastewater from a single day’s sample in 2008

74 june2011

about which statistical approach is best, what is clear is that it is critical that the distribution of data are fully and accurately reported and that decisions about appropriate statistical summaries and tests are made that incorpo-rate the nature of the data and the intended uses. An excellent description of the types of data problems and possible approaches to addressing them is provided by Helsel17.

Another important statistical issue is that of determining and incorporating error bounds. Without error bounds, or confidence intervals, it is not possible to state whether the drug levels are higher on Saturdays than Tuesdays or in the city than in the suburbs. Error bounds need to involve several com-ponents including analytical error associated with chemical analysis, flow estimate error, population estimate error, and sampling error.

Developing an error associated with a 24-hour sample is needed and involves assumptions about the number of pulses of a drug that are expected to move through the sewer system. Many factors will affect this. They include pharmacokinetics – the speed at which the human body processes the drug; the frequency and timings of drug use within the day; and how long it takes to transit the sewer system. The size and frequency (and mode, e.g. flow-proportional or timed) with which samples are taken at the treatment plant must be optimised for obtaining representative samples and for minimising error. The challenges in developing error bounds, measures of uncertainty, and reliable and valid 24-hour daily estimates are well described by Ort et al.18, but have yet to be applied in practice.

What lies ahead?

We have a sensitive analytical method for de-tecting a range of drugs and other compounds for entire municipalities. Wastewater drug testing is a potent new tool that supports drug epidemiology with a direct measure of drug use for a potentially far broader and more

representative proportion of the population than has been possible up to now. However, it is not a measurement that should be used in isolation. It must be used in context and interpreted with other measures of drug use and consequences. Given standardised approaches to within-day sampling, between-day sampling, analytical chemistry and statis-tical treatments of resulting data, a reliable and valid comparison of drug use can be made across diverse places across the earth. Quality drug epidemiology derived from wastewater treatment plant drug testing can only help to inform our debates about drug policy as well as law enforcement and public health interventions.

References1. Warner, M., Chen, L. H. and Makuc,

D. M. (2009) Increase in Fatal Poisonings Involving Opioid Analgesics in the United States, 1999–2006. NCHS data brief no 22. Hyattsville, MD: Na-tional Center for Health Statistics.

2. Al-Tayyib, A. A., Rogers, S. M., Gribble, J. G., Villarroel, M. and Turner, C. F. (2002) Effect of low medical literacy on health survey measure-ments. American Journal of Public Health, 92, 5–8.

3. National Institute on Drug Abuse (2006) Assessing Drug Abuse Within and Across Communi-ties: Community Epidemiology Surveillance Networks on Drug Abuse (2nd edition). NIH Publication No. 06-3614. Bethesda, MD: National Institute on Drug Abuse.

4. Banta-Green, C., Goldbaum, G., King-ston, S., Golden, M., Harruff, R. and Logan, B. K. (2005) Epidemiology of MDMA and associated club drugs in the Seattle area. Substance Use and Misuse, 40, 1295–1315.

5. Ternes, T. A. (1998) Occurrence of drugs in German sewage treatment plants and rivers. Water Research, 32, 3245–3260.

6. Daughton, C. G. (2001) Illicit drugs in municipal sewage: proposed new non-intrusive tool to heighten public awareness of societal use of illicit/abused drugs and their potential for ecological consequences. In C. G. Daughton and T. Jones-Lepp (eds), Pharmaceuticals and Personal Care Products in the Environment: Scientific and Regulatory Issues, Symposium Series 791, pp. 348–364. Washington, DC: American Chemical Society.

7. Zuccato, E., Chiabrando, C., Castiglioni, S., Calamari, D., Bagnati, R., Schiarea, S. and Fanelli, R. (2005) Cocaine in surface waters: a new evidence-based tool to monitor community drug abuse. Environmental Health, 4, 14.

8. Chiaia, A. C., Banta-Green, C. and Field, J. (2008) Eliminating solid phase extraction with large-volume injection LC/MS/MS: analysis of illicit and legal drugs and human urine indicators

in US wastewaters. Environmental Science and Technology, 42, 8841–8848.

9. Banta-Green, C. J., Field, J. A., Chiaia, A. C., Sudakin, D. L., Power, L. and de Montigny, L. (2009) The spatial epidemiology of cocaine, meth-amphetamine and MDMA use: a demonstration using a population measure of community drug load derived from municipal wastewater. Addiction, 104, 1874–1880.

10. Hart, G. L., Larson, E. H. and Lishner, D. M. (2005) Rural definitions for health policy and research. American Journal of Public Health, 95, 1149–1155.

11. Substance Abuse and Mental Health Ser-vices Administration (2006) Methamphetamine/ Amphetamine Treatment Admissions in Urban and Rural Areas: 2004. The Drug and Alcohol Services Information System Report No. 27. Rockville, MD: Office of Applied Studies.

12. National Institute on Drug Abuse (2008) Epidemiologic Trends in Drug Abuse. Proceedings of the Community Epidemiology Work Group, Volume 1, June 2007, NIH Publication No. 08-6200A. Bethesda, MD: National Institute on Drug Abuse.

13. Substance Abuse and Mental Health Services Administration (2005) Substance Abuse or Dependence in Metropolitan and Nonmetropolitan Areas, 2004 Update. The NSDUH Report. Rock-ville, MD: Office of Applied Studies. http://www.oas.samhsa.gov/2k5/metro/metro.htm.

14. Van Nuijs, A. L. N., Pecceu, B., Theunis, L., Dubois, N., Charlier, C., Jorens, P. G., Bervoets, L., Blust, R., Neels, H. and Covaci, A. (2009) Co-caine and metabolites in waste and surface water across Belgium. Environmental Pollution, 157, 123–129.

15. Metcalfe, C., Tindale, K., Li, H., Ro-dayan, A. and Yargeau, V. (2010) Illicit drugs in Canadian municipal wastewater and estimates of community drug use. Environmental Pollution, 158, 3179–3185.

16. Environmental Protection Agency (2008) Clean Watersheds Needs Survey 2004: Report to Congress, January 2008. http://www.epa.gov/cwns/2004rtc/cwns2004rtc.pdf

17. Helsel, D. (2005) Nondetects and Data Analysis: Statistics for Censored Environmental Data. Hoboken, NJ: Wiley-Interscience.

18. Ort, C., Lawrence, M. G., Rieckermann, J. and Joss, A. (2010) Sampling for pharmaceuticals and personal care products (PPCPs) and illicit drugs in wastewater systems: are your conclusions valid? A critical review. Environmental Science and Technology, 44, 6024–6035.

Caleb Banta-Green is Research Scientist at the Alcohol and Drug Abuse Institute, University of Washington, Seattle. Jennifer Field is Professor of Environmental and Molecular Toxicology at Oregon State University.

Without error bounds it is not possible to say whether

drug levels are higher in cities, or at weekends