Drug use statistics Library Home Home Universiteit van Amsterdam
Cohen, Peter (1996), Notes on (methods of) drug use prevalence estimation and other drug use research in a city. Presentation for the Joint Seminar about "Addiction Prevalence Estimation: Methods and Research Strategies" 10-14 June 1996, Strassbourg, France. Published as "The relationship between drug use prevalence estimation and policy interests". In: G. Stimson (Ed.)(1997), Estimating the prevalence of problem drug use in Europe, pp. 27-34. Lisbon: European Monitoring Centre for Drugs and Drug Addiction EMCDDA.
© Copyright 1996, 1997 Peter Cohen. All rights reserved.

 

Notes on (methods of) drug use prevalence estimation and other drug use research in a city

Subtitle

Peter Cohen

My comments will focus on the local level of drug use prevalence estimation and other drug use research. It is at the local level that most decisions are made, and where relevance of good data and understanding for drug policy practice is highest. Furthermore, national estimation of drug use or related behaviors and events is difficult without reliable local data. Only if local data collection is good, and when it covers the most important areas of a country, national data have a chance of being relatively decent.

In this introduction I would like to discuss some problems related to the selection of appropriate methods for estimation of drug use or heavy drug use. Without suggesting that I will cover everything, I will look at the kind of problems that have to be solved before one starts or while busy with data. I will also discuss questions about how clarity about a general model of drug policy can improve the relevance and interpretations of local drug use data and estimations.

Decent data are often just partly available or not at all, but , once the importance of collecting reliable data is recognized on the political level this problem can be solved. Then, by means of good sampling, sound survey instruments and by means of good police registration techniques, prevalence and patterns (of different kinds) of drug use can be studied on a regular basis. And once treatment institutions are in place, reliable treatment registration can supply interesting data sets as well.

The inherent problems and error potential of particular methods are not always fully grasped by those who use them. This means that the match between type / quality of data and technique of analysis is sometimes poor. However, I will not discuss here how to create availability of decent sets of local data or the statistical techniques to digest them.

Solutions to 'technical' problems will only come if the perceived political importance, ascribed to data and good estimation rises to a sufficiently high level. Therefore, most important for selection of the most appropriate objects /methods of drug use research in general and prevalence estimation in particular, are problems of drug policy and the way drug policy is made. As long as drug policy is made without a heavy input from systematic and scientific data analysis, the political basis for drug policy is not much more than emotion or orthodoxy.[1]

Every estimate has to be tailor made to its proper use or functions. This implies that estimation problems should not only be described as purely technical, but also as strategic. What exactly is there to estimate, to what broad or narrow policy aspect is the estimate of relevance, what precision of the estimate is wanted, do we need the estimate to be comparable to estimates made in other areas, what are the political implications of estimates? In other words, the sophistication of drug policy itself has a strong influence on the quality of the estimation process.

Drug policy models and what to estimate

Let us take the example of general prevalence estimation of frequent and/or heavy heroin use. In what policy model can such an estimate perform a role? We assume, for brevity's sake, that we have three distinct policy models. One is the predominantly punitive / repressive model, the second the harm reduction model and the third we might define as the cultural integration model.[2]

In the repressive model, the dominant aim is to suppress all drug use. Heavy use estimates will probably work with a definition of 'drug addiction' that includes all types of heroin use. In this model all types of heroin use are considered harmful and abusive. Estimation needs will encompass all forms of heroin use, intravenous or not, regular or not, and problem related or not.The size of the heroin problem is more or less the size of heroin use prevalence.

The function of this estimate repeated over time is to assess the success of suppression. If estimates are reliable and they go down, suppression is deemed successful. If estimates go up, suppression is not successful enough and has to be increased or changed in scope and instrumentation. Prevalence estimates in the punitive model are used to evaluate the short term success of suppression techniques and not its intrinsic merits relative to another model of drug policy. It is also possible that in a particular version of the punitive model the size of the heroin using/abusing population is not so important, but the number of heroin overdose-deaths is. Or the number of emergency room visits of heroin consumers is deemed important. The latter data probably say more about the conditions around drug use than about the seriousness of the problem (in terms of number of users) but the public or politicians often do not perceive this. In a repressive model this kind of data represent deterrence and giving them a high profile is part of the policy. In such a case we will sometimes find that extremely high prevalence estimates may play a role, either in the press or in the speeches of politicians. Such estimates might represent only anecdote (if not worse) but this is easily clouded behind the emotional appeal (which is the real function of such estimates).

In a harm reduction model the aim is still to reduce drug use in general, but the emphasis is not placed on punitive action but on prevention of potential harmful effect of drug use behavior and drug distribution. One allows for the insight that drug use exists and that it will continue. In spite of the goal to reduce its prevalence, reducing individual and social risks of the existing types of drug use is seen as a major concern. In this model the need for data could be very different than in a suppression model of drug policy. Policy makers that work within a harm reduction model are mainly interested in the number of heroin users that are clients and potential clients for harm reduction intervention, be it medical or non medical. Such clients could belong to quite different subgroups. One subgroup may consist of i.v. drug using prostitutes working on the street under marginalised conditions, another one might be prostitutes working in a completely different site of the sex markets, like clubs or expensive brothels. Other subgroups might consist of homeless users, inmates, ethnic groups or students and employees. Providing assistance to each of these subgroups -where needed- is deemed important. However, attention for these different groups will probably have different policy priorities, different budget consequences and different kinds of political usefulness.

Estimates in this model serve for assessment of treatment/assistance needs of particular groups based on knowledge about drug use and risk patterns in these groups. The estimates may also be needed to find out if some groups remain without adequate assistance. For instance, one may have data that generate estimates of high i.v drug use in certain ethnic groups, but in assistance institutions these groups are rarely seen. In that case, the treatment system will have to be able to relocate or devise options that are offered. Likewise, in the Dutch city of Rotterdam a general increase in the use of base cocaine by heavy drug users is reported (Grund, 1993, Visser, 1996, NRC 1996) The adequacy of the present methadone based treatment system has to be investigated in the light of these developments.

Another function of estimations in a harm reducing policy model is related to evaluating the level of harm reduction that is characteristic for the context of particular kinds drug use. For instance, if drug overdose deaths are reliably estimated (by excluding double counting, by including corrections for false cause of death registration and by maintaining unambiguous definitions) and these estimates rise, something may be wrong with the style of police activity in a city. Crude suppression may result in high variability of drug purity, unavailability of quiet drug use locations, etc. This in its turn could result in rising drug overdose deaths numbers. Since drug policy in a harm reduction model aims at reducing risks, interventions into the style and intensity of police suppression may even be more important for risk reduction than the development of accessible assistance institutions.

So, if one investigates the prevalence of i.v. drug use in a particular city and finds this is much higher than elsewhere, one might start to investigate the possible reasons for this. In Amsterdam, of heavy heroin users only a minority inject and a large majority (of about 70%) are current smokers of heroin (Korf, 1995, p. 323, Grapendaal et al, 1995, p. 144). Those who inject have a higher mortality than the non injectors in Amsterdam (Van Haastrecht et al, 1996). Although this difference might not apply in other settings, it is very interesting from a local mortality prevention point of view. Would it be possible to decrease heroin related mortality in cities by having heavy users change route of ingestion? Could we find out under what conditions (heavy) heroin users abandon injection as route of ingestion or start smoking heroin straight away? Might cheap heroin base, of stable and relatively high purity be such a condition? And is such a condition attainable if law enforcement against individual heroin use and sales is given a low priority, as is the case in Amsterdam?

In this example, the availability of the estimate of low prevalence of i.v drug use among heavy heroin users in Amsterdam is the starting point for broader drug policy questions.

In the third policy model we distinguished, the cultural integration model the main aim is no longer to suppress drug use per se. Overall aim is to bring drug use under the normal regulatory mechanisms societies have for controlling accepted behavior and attributed meanings of behavior. Drug use in this model is as 'normal' as are its problems. In the Netherlands this approach is called 'normalization of drug use'[3] and it explicitly implies that possible problems are no longer treated as extreme or principally different from other problems society has, like traffic, broken marriage or stress in the office. In the drug policy model of cultural integration the basic aims are exactly opposite to those in a suppression model: in the latter, drug use is maximally ostracized and marginalized. In the former it is integrated into normal social process.

In the cultural integration model of drug control the need for data is different again. One might want to know how many heavy drug users are treated in the normal care system (general practitioners, hospitals) and how many are still referred to special institutions. Taking care of heavy drug users in special institutions belongs more to suppressive and harm reduction models. In these models heavy drug use is so deviant and its problems are so special that those concerned need special institutions. But, as we can see in the field of alcohol use (which is culturally integrated), much of the treatment takes place in the normal system. In the Netherlands this is mostly general practitioners, prescribing vitamins, monitoring health status and referring to other health care where needed. Only very few of the heavy alcohol users are referred to specialized institutions. In suppressive and harm reduction systems heavy drug users are still seen as so deviant, that most do not have a chance to be treated in regular psychiatric institutions, altho a certain percentage of them would need such treatment. (In spite of 'normalization' attempts in Amsterdam, the Municipal Health Service has had to set up a small psychiatric care unit within the Municipal Health Care system for heavy drug users). Other data that could be needed in a cultural integration model would focus on what social process is ongoing that hinders cultural integration. A good example is the shift from a suppressive policy model around homosexuality in the Netherlands toward a cultural integration model. In the latter it is relevant to ask what the prevalence is of teachers that are kept outside the schools because of their homosexuality, or what career problems homosexual officers in the military might have, and how to redress them.

This short overview illustrates that "estimates of prevalence of heroin addiction" or related estimates can only have meaning and a function within the local models of policy. If such functions are not well ascribed, estimation will probably not only never reach an acceptable level of accuracy, it will also have very little impact on drug policy. Fuller accounts of types of estimates and their possible uses are given by Reuter (1993) and by Anglin, Caulkins and Hser (1993).

On a higher level of abstraction one could say, that almost in all countries some sort of local construction of the drug problem is made. This construction appears in

  1. the kind of problems that are considered relevant,
  2. the causal attributions that are made to 'explain' these problems,
  3. the policy interventions that are produced to 'do something about it'
  4. the type of expertise (e.g. medical, social, legal) that is considered dominantly relevant in the situation.

It is usually a change in the latter factor (dominance of particular expertise) that shows the introduction of a change in the type of paradigm that governs the problem construction.[4]

What precision is needed for estimates?

Functional considerations for precision

Creating estimates for size of small albeit well defined groups is an intricate procedure. For some groups it may entail case finding and nomination techniques or costly survey and or field work methods to provide data that, in combination with other registries might yield reasonable estimates. Highest possible precision of the estimates is always important from a scientific point of view. But since policy makers will have to provide the funds for estimation making, the elevated costs for higher levels of precision will, unfortunately not always be considered legitimate. Precision here is defined as a narrow band of confidence intervals.

Required precision will often be dictated by drug policy aims, by budget, by simple technical skills, by the quality of the raw data that are available but most of all by the political agenda's of those who will fund and use estimations.

Estimating prevalence of rare or hidden behaviors or events is a matter of combining different outcomes of different methods for estimating. If estimates vary considerably among methods, something is wrong and precision probably low. Field work and databases must be improved. But if really independent estimates are not too far apart per method, or if variations are clearly understood as artifacts of data collection[5] one could be relatively satisfied with the results. In such cases precision is sufficient.

If one wants to evaluate the potential of a serious infection threat, for instance hepatitis C, one should know accurately how many drug users are at risk and where they are. Carefully set up field surveys, test sites, and use of a maximum of estimations by different methods, and improvement of data bases might after some time result into an adequate level of precision.

Political considerations for precision

An important aspect to be discussed when speaking about the desired level of estimate precision is the political use that is made of estimates. Since political differences about the right kind of drug policy occur quite often, an estimate might be of use to a particular type of drug political rhetoric, and not to another. Imagine a situation in which the size of the group of 'addicts' is an important political issue. If someone responsible for drug policy can say that this group has diminished during her administration, this can be used as support for the particular policy. So, estimations can become weapons in a political struggle.[6] This gives the researchers behind these estimates a particular responsibility. It requires the making of estimates of the highest possible quality and precision, with a clear minimum level as bottomline. It is hard to give a standard 'bottom line' but this topic can not be ignored.

Sometimes strong pressure is applied by politicians to drug policy functionaries to supply estimates of drug use or addiction. "Just give me some figure that looks like something".

But, if decent estimates cannot be produced, this should be made very clear, for instance by stating how uncertain the estimate is, and why. One could supply different estimates and not just one, by making use of different assumptions and data sets. If under conditions of high uncertainty more than one estimate is systematically provided, this prevents the development of quasi knowledge. The absence of 'clear figures' because of insufficient data should be used to legitimate funding the research and registration improvements needed to be able to make good and firm estimations after all. Specialists in the area of epidemiology should not be forced or let themselves be forced to supply unwarranted estimates; they are not magicians. They should stick to the adagium that if the data do not exist, estimates can not be provided.

Another problem attached to the outcomes of estimations is the realistic threat such outcomes may pose to particular groups of professionals within the field of drug policy and treatment. One does not need much imagination to see that particular groups of professionals may have an interest in these estimations to be high or going up. Such groups could be specialised forces in the police, customs, treatment personnel etc. By providing their own estimates they can influence the political debate in the way they desire. If a scientifically sound , independent and regular estimation of the most important drug use phenomena is in place, such interest related estimates pose less of a danger to serious discussion about drug policy.

Standardized estimates for purposes of comparison

This introduction has focused up till this point on the thesis that well described local policy needs should be at the basis for all local prevalence estimation work. I will conclude with some remarks about estimates that can be used as general indicators. It is possible to create particular prevalence estimates that are relevant across different models of drug policy. If it could be accomplished that the same phenomenon is estimated reliably in different places (which is not necessarily in the same methodological way) thinking about the impact of drug policy in relation to the drug use phenomenon can be improved.

In spite of the fact that an estimate of all heroin use is seen as not very important in a drug policy model that focuses on harm reduction for heavy users, an overall estimate of all heroin use in the general population is very useful as a general descriptive indicator. Such an indicator can be seen as a baseline indicator. This means for instance that if one would find relatively high prevalence of heroin use among homeless adults, one would want to know if heroin use prevalence among the household population is relatively high as well. The meaning of estimates or measurements relating to subgroups in a city can only be established

  1. if one knows the size of the same phenomenon in the city population as a whole
  2. if one knows the relative size of the same phenomenon in more cities.
  3. if one knows the sub group data in these different cities

In order to make such measurements and comparisons thereof one needs to standardize the definitions of a phenomenon, and the methods of how one looks for it. Some of the epidemiological work of the Council of Europe-Pompidou group of experts in epidemiology- is an example of this drive for comparability between cities.

To be able to evaluate the total level of heroin or other drug use in a population, one would need estimates valid for

  1. the general household population ( from 12 years on)
  2. the most important population subgroups that can not be measured via household surveys (e.g. prison inmates, the homeless in some cities, etc.).

Type of drug use experience to be measured could be recency/frequency: lifetime experience with a drug, last 12 month experience, last 30 days, and last 24 hours experience. Within each of the time spans one could measure the frequency of drug use. This way it becomes possible to define many different patterns of drug use, among which long term frequent drug use. The advantage of this simple way of measuring recency and frequency of drug use is that an overview is created over all drug use. One will often see that long term frequent drug use is rare. This is true in different drug control systems and slowly we start to see that drug policy may have less impact on drug use level than politicians assume.

Total heroin use would be near the sum of household and sub group use, and this summation should be done in different cities according to standardized methods. This could yield massive information on the effects (or of the lack of effect) of drug policy, the effect of economic or cultural factors on levels of use, the prevalence of particular patterns of use or the prevalence of particular problems (like overdose, first treatment, crime).

Conclusion

Estimation of drug use and the interpretations of such data is, like all other drug use data collection, very much a function of the drug political situation in a city or country. Technical matters, like statistical techniques, are only a small part of the problems that epidemiologists and other researchers of drug use have to struggle with. This presentation has tried to give some insight in these other than technical problems and to show that these other problems have a much larger impact on data collection and interpretation than normally acknowledged. It further stressed the need for high quality comparitive work in this area.

Notes

  1. Cf Alain Ehrenberg in Liberation, 15 feb 1996: "Contre les théologies antidrogue" (Against antidrug religions).
  2. In reality these models are mixed and may move from one direction to another.
  3. E.L. Engelsman: "The drug problem is serious, but in principle not more serious than other health and social problems." "Drug users will function like other civilians in a social system that is characterized by mutuality and responsibility." In: M.S. Groenhuijsen en A.M. van Kalmthout (eds), Nederlands Drugsbeleid in Westeuropees Perspectief. cf E.L. Engelsman: Het Nederlandse drugbeleid in Westeuropees perspectief, page 137-144. Arnhem, 1989. Gouda Quint.(translation P. Cohen).
  4. In the Netherlands this has been very clear in the late sixties, when drug policy commissions were created that included many criminological and social scientific experts, and not only medical experts. This made a more complete view of the drug use phenomenon possible, with dramatic consequences for the construction of the 'drug problem'. See also P. Cohen. "The case of the two Dutch drug policy Commissions.An exercise in Harm Reduction 1968-1972." Forthcoming in Patricia Erickson et al (Eds) "New Public Health Policies and Programs for the Reduction of Drug Related Harm" University of Toronto Press, 1997.
  5. Lately, in the USA for the first time a military general has been made the director of drug policy ("Drug Czar") after this function had been in the hands of a mere police officer. This might show a new direction of policy development and problem construction.
  6. For instance, if one knows that a data base of drug related arrests is biased for certain ethnic groups because of unequal arrest probability between ethnic groups, the error resulting from this will be reflected in estimates using capture -recapture in which arrest data bases are used. This can be dealt with to a certain degree simply by allowing a certain correction for such an error. But, if such errors are not well quantifiable, one might have to decide to set up the estimates from different assumptions on arrest probability, to set up capture recapture using other data, or to move to different methods altogether.
  7. In the recent exchange of opinions on drug policy between France and the Netherlands, the Dutch prime Minister used French estimates of addiction prevalence to show that addiction was less prevalent in the Netherlands than in France. For a discussion of the quality of French addiction estimates see T. Boekhout van Solinge: Heroine, cocaine en crack in Frankrijk, 1996 CEDRO, University of Amsterdam. The quality of the Dutch addiction estimates, even if marginally better than the French, is not high enough to play a role in a serious comparison of addiction prevalence between the two countries. This leaves untouched the question of the relevance of such a comparison for an evaluation of drug policy; addiction prevalence is probably more determined by broad economic and cultural conditions, immigration and quality of social security than to drug policy. (cf Quensel et al Zur Cannabis-Situation in der Bundesrepublik Deutschland and Harrison, Backenheimer and Inciardi: Cannabis use in the United States. Implications for policy. In Cohen P. and Sas A. (eds): Cannabisbeleid in Duitsland, Frankrijk en de Verenigde Staten, CEDRO Universiteit van Amsterdam, 1996 (Cannabis policy in Germany, France and the United States of America. -Text in the language of the evaluated country. The text of the report about France is printed in Dutch as well, as is the introduction by Peter Cohen-)

References

Anglin, Douglas, Caulkins, Jonathan and Yih-Ing Hser (1993, Spring), Prevalence Estimation: Policy Needs, Current Capacity, and Future Potential. In Rachin, Richard L., Editor, Prevalence Estimation. Techniques for Drug-Using Populations. Journal of Drug Issues, 1993 Spring.

Boekhout van Solinge, Tim (1996), Heroine, cocaine en crack in Frankrijk. Handel, gebruik en beleid. CEDRO, University of Amsterdam. (Heroin, cocaine and crack in France. Trade, consumption and policy).

Cohen, Peter and Arjan Sas (eds) (1996), Cannabisbeleid in Duitsland, Frankrijk en de Verenigde Staten. CEDRO, Universiteit van Amsterdam.

Cohen, Peter (1997), The case of the two Dutch drug policy Commissions. An exercise in Harm Reduction 1968-1972. In: Patricia Erickson et al (Eds); New Public Health Policies and Programs for the Reduction of Drug Related Harm. University of Toronto Press.

Engelsman, Ed (1989), Het Nederlandse drugbeleid in Westeuropees perspectief. page 137-144. In: M.S. Groenhuijsen en A.M. van Kalmthout (eds), Nederlands Drugsbeleid in Westeuropees Perspectief. Arnhem, Gouda Quint.

Grapendaal, Martin, Ed Leuw, and Hans Nelen (1995), A world of opportunities. Lifestyle and economic behavior of heroin addicts in Amsterdam. State University of New York Press.

Grund, J.P.C. (1993), Drug Use as a Social Ritual: Functionality, Symbolism and Determinants of Self-Regulation. Rotterdam, Instituut voor Verslavingsonderzoek (IVO).

Haastrecht, H.J.A. van, et al (1996), Predictors of mortality in the Amsterdam Cohort of Human Immunodeficiency Virus (HIV)-positive and HIV-negative Drug Users. American Journal of Epidemiology, 1996, Vol 143, no 4. pp. 380-391.

Korf, D.J. (1995), Dutch Treat. Formal control and illicit drug use in the Netherlands. Amsterdam, Thesis Publishers.

Reuter, P. (1993, Spring), Prevalence estimation and Policy Formulation. In: Rachin, Richard L., Editor, Prevalence Estimation. Techniques for Drug-Using Populations. Journal of Drug Issues, 1993 Spring.

Visser, Hans (1996), Perron Nul. Opgang en ondergang. Meinema Uitgeverij.

Swol, Coen van (1996), NRC Handelsblad, 1 juni 1996.

Last update: May 25, 2016