Writing this review my original intent was to provoke new action via radical thinking about new candidate phenotypes for research at the intersection of molecular genetics and epidemiological research on drug dependence (hereinafter, ‘genetic epidemiology of drug dependence’), as might be put to work in future genome wide association studies (GWAS), with large and rigorously drawn epidemiological samples of community residents in multiple countries. By ‘radical,’ I mean only that we will work back to the original roots of the ‘phenotype’ concept. In the process, some new phenotypes can be considered, which might prove to be useful in the genetic epidemiology of drug dependence, particularly as we try to gain an initial multi-national and then a global perspective on this topic.
In some respects, the review has the quality of an editorial of a type written about ten years ago to advocate a major reorientation of scientific work on the genetics of bipolar disease and schizophrenia (1
). What I am trying to accomplish is to reorient research on genetic epidemiology of drug dependence, and in the process I will make a very practical recommendation about how this work can be accelerated via a re-conceptualization of the phenotypes under study. I provide a practical example and describe experiences in a recent pilot study that was intended to set in motion a future elaboration of large-sample epidemiological field studies that now are not being made informative with respect to the genetic epidemiology of drug dependence.
One facet of what I shall recommend involves a return to practices of the past. Another facet involves new points of departure. In specific, the acceleration of progress I have in mind may require somewhat of a departure from the drug dependence or ‘addiction’ phenotypes as they presently are framed in contemporary American psychiatry, and which may not be as useful outside of America as they are in the American context.
What I will share may come as a surprise to readers who know my work as an epidemiologist who mainly has tried to extend trail-blazing paths laid by the late Professor Lee Nelkens Robins, together with Professor John E. Helzer, with respect to a focus on the drug dependence concept derived from contemporary diagnostic criteria and case definitions as originally framed in the American Psychiatric Association's Diagnostic and Statistical Manual, Third Edition (DSM, DSM-III), and in the later DSM-IV. In the end, I shall recommend a deliberate departure from that research approach in order to accelerate progress in drug dependence epidemiology, and to promote creation of parallel tracks of research. That is, I judge that for a time we can accelerate the progress of these scientific fields by working along non-intersecting tracks that run alongside the still- evolving DSM diagnostic traditions and guide the clinical practice of psychiatry.
My proposal is that the greatest acceleration might be achieved via parallel track research over the next 5-10 years, with an occasional ‘crosswalk’ across a bridge first built orthogonally, until there is a sufficient accumulation of evidence about non-parallelism of the tracks. I will provide a conceptual model, illustrated with a descriptive figure, in order to make clear that we must begin with non-parallel tracks because in virtually all observational research of our time the field has ignored a reciprocity (sometimes erroneously termed ‘reverse causation’) when prior crosswalks have been created.
Before presentation of the figure, a note about Professor Lee Nelkens Robins may be in order in that she passed away no more than a few months ago, and her work may not be well-known in some sectors of this field. Lee was a trailblazer in 20th century psychiatric epidemiology generally and in drug dependence epidemiology specifically. Her PhD sociology studies with Talcott Parsons at Harvard University occurred when Parsons was attempting to move structural functionalism in the direction of action theory and the AGIL framework of institutional necessities for successful society formation and development. [AGIL is an acronym for adaptation (A) to environmental conditions, consensus setting with respect to goal attainment (G), harmonization and integration (I) of the society as in adoption of common values, norms, and language, and successful development of latency (L) mechanisms such as bonds to family and school through which values, norms, and language are passed from generation to generation.] For Lee's collaborators who were aware of Parsons' AGIL framework, it was obvious to see her attempts to harness that framework in the creation of the small social groups through which successful scientific progress might be made.
From the mid-1970s forward, she taught many of us a conceptual model for drug dependence that I have characterized as a stage-transition model, and have put to good use in the study of tobacco dependence and related processes, often harnessing latent class and transition approaches (e.g., 2
). Lee taught that the first stage in the drug dependence process was having a chance or opportunity to try a drug. Borrowing from Wade Hampton Frost's concept of ‘exposure opportunity,’ first expressed some 80 years ago in a theory of infectious disease epidemiology, my own research group has worked up various measurements of that first chance to try a drug. We now have evidence of a fairly general proposition that a male excess in occurrence of drug dependence often has its origins in a male excess in the timing of the first drug exposure opportunities. In most countries in which we have conducted our studies, age by age, males are more likely to have had a chance to try illegal drugs, as compared to females in the same birth cohorts. Once the chance to try occurs, the next stage-transition is actual use of the drug. Here, the male-female variation is much attenuated in general; females seem to be just as likely as males to use the drug, once the chance to try has occurred. Then, once drug use occurs, some drug users develop a drug dependence syndrome while others do not. Sometimes there is a male-associated excess conditional risk of becoming dependent once use starts; sometimes not. For example, we have found male-female differences in the risk of becoming cannabis dependent soon after onset of cannabis smoking, less pronounced male-female differences for alcohol, and even smaller male-female differences for cocaine (2
). As a side note, one might contemplate epidemiological research on the later stage-transitions (e.g., from dependence into recovery), but our research group always has been concerned about attrition biases in the large-sample epidemiological research context once the drug dependence process starts – given excess mortality risk attributable to drug dependence, as well as potential under-representation of heavy drug users in community sample surveys (11
The first formal epidemiological work on the stage-transition model from drug use to dependence was completed by John Helzer during a followup study of male Vietnam era veterans, for which Lee Robins was Principal Investigator (12
). No complex latent transition modeling was involved in that research. Relatively simple contingency table analyses proved to be enough to convey that (a) most veterans who had used opioid drugs during Vietnam era service had not developed sustained opioid ‘addiction’ problems that complicated their lives back at home, and (b) some characteristics were predictors of the post-service ‘addiction’ outcome while other characteristics were not predictive at all.
Seeking examples from the published literature for this review article, I came across some interesting elaborations of the stage-transition model that didn't make much sense to me. To illustrate, consider a simple plot of drug dependence prevalence estimates on the y-axis of an x-y graph, with the count of the number of days or occasions of drug use on the x-axis. This plot can be envisioned, and with appropriate estimates and data, it can be plotted. But it is valid to draw this plot only if the relationships linking the counts of drug experiences and the occurrence of the drug dependence process are in conformity with the acyclic assumption of standard dose-response modeling.
I believe that behavioral pharmacologist pioneer Joseph Brady was the first to recognize a violation of the acyclic assumption in relation to the process of starting to use a drug and then becoming dependent upon it, but his regrettably neglected article on the topic is buried within a NIDA research monograph that has not been cited frequently (13
). In , I have re-drawn Brady's earlier and more artistic torus-like structure of feedback loops that link repetitions of drug self-administration experiences with the formation of clinical features of drug dependence. In this rendition, I have tried to give a very clear depiction of a violation of the acyclicity assumption, with the drug dependence features feeding back and influencing the repetition of drug self-administration experiences. Readers familiar with the work of Koob and LeMoal (14
), among others, will be familiar with this type of feedback loop and re-setting of set points in the processes of neuroadaptation and repetitive drug self-administration.
Figure 1 Violation of the acyclic assumption. Often, there is an insidious onset of the drug dependence syndrome, once use starts. Plus, as the dependence develops, feedback loops appear, such that the choice to quit becomes more difficult. The count of drug experiences (more ...)
shows the onset of an individual's drug use at 1. At Step 2, there is a piling up of occasions or days of drug use after first use. Thereafter, at Step 3, tangible clinical features of drug dependence begin to show up (e.g., subjectively felt tolerance, craving). At some point before or during the formation or coalescence of these individual clinical features into a drug dependence syndrome, there is a feedback loop such that the syndrome formation process begins to drive up the count of drug-taking occasions and the associated rate of drug use per unit time, just as the accumulating count of drug-taking occasions drives forward the drug dependence process.
With a reciprocity or feedback loop of this type, the acyclicity assumption of standard dose-response modeling is violated. It becomes difficult to make sense of a plot that expresses the prevalence of drug dependence as a function of the count of drug-taking experiences as if this were some standard type of acyclic ‘dose-response’ relationship. Here, the response is allowed to influence and drive up the size of the ‘dose’ in a clear violation of assumptions for standard statistical modeling of dose-response relationships.
Advanced models exist for use when the acyclicity assumption is violated, but the data for these models necessarily must be longitudinal in character. In the context of the multi-national or global GWAS investigations mentioned in the first paragraph of this article, to gather such longitudinal data on the scale of epidemiological sampling for a GWAS investigation would drive up the cost of any multi-regional global drug dependence epidemiology research program, not to mention a need for future replications with equally sized samples and other refinements to satisfy the GWAS critics.
A partial cross-sectional study solution might be derived by yoking an appropriate model for count or rate responses with an appropriate model for multivariate response profile of clinical features, allowing the interdependency of these count and multivariate response constructs to be estimated in a correlation that is agnostic with respect to direction of influence (e.g., drawn in a Directed Acyclic Graph with a double-headed curved arrow). The cross-sectional sample to model these relationships within the first month after onset of drug use would have to be quite large in order to include a sufficient number of earlier-onset users of recent onset vintage. Estimation of the parameters of separate models, month by month, for each month of elapsed time after onset of drug use could foster development of a cross-sectional mosaic of the time course of the drug dependence process, expressed as a function of genes and environmental conditions and processes interacting. The result would be a useful development, much needed to derive starting estimates for planning of prospective and longitudinal research for more definitive evidence on these interrelationships.
Returning to 's steps 1 and 2, it is possible to see that the drug dependence process actually begins with a count process that predates the formation of the drug dependence syndrome. This count process might be measured as the number of self-administrations per interval of time under study, or the number of occasions of use or number of days of use during that interval.
Recognition of these developments that antedate the clinical features of the drug dependence process prompted our research group to turn its attention to count process models that might be used in research on the very early stages of drug dependence. It is fortunate that in the past 20 years there have been remarkable advances in the statistical methods for addressing questions about count process parameters of considerable interest in the study of the process of becoming drug dependent. In a later section of the article, a count process model is described in more detail, with an illustration that forecasts what might be done in large multi-national epidemiological research to complement current approaches based on the stage-transition tradition pioneered by the Robins research group, including our own research in genetic epidemiology with community and other samples (e.g., see 15
). As compared with the genetic epidemiological studies of dependence on internationally regulated drugs such as cannabis and cocaine, there has been a much greater diversity of phenotypes under study in genetics research on the dependence syndromes involving tobacco and alcohol. To illustrate, one might consider the broad range of phenotypes outlined in a recent National Cancer Institute (NCI) tobacco control monograph that can be downloaded from a web site (16
). Of course, there is an advantage to the Robins tradition of orienting the stage-transitions studied in epidemiology to the same diagnostic constructs for drug dependence that are being used in the clinical practice of psychiatry and allied professions responsible for treating patients with these conditions. The resulting epidemiological estimates have clinical implications that are not too difficult to communicate to clinicians who are treating or doing research with these patients. Epidemiologists' use of the DSM-type diagnostic constructs also seems to serve well in the introductions and background sections of pre-clinical research reports, helping to substantiate the public health significance of the drug dependence syndromes.
In the rest of this article, I will review two lines of research that may help motivate a departure from this time-honored stage-transition tradition. One of the lines of research is based upon the work of the World Mental Health (WMH) Surveys Consortium, for which I have some responsibility as an affiliated Principal Investigator. The other line of research is based upon the work of the Office of Applied Studies of the Substance Abuse and Mental Health Services Administration (SAMHSA), which directs the National Surveys on Drug Use and Health (NSDUH) and then makes NSDUH public use datasets available to non-governmental research groups such as the one for which I am responsible.
The WMH Surveys Consortium is important in this context because it is taking the Robins tradition of standardized diagnostic assessment of drug dependence and related conditions into multiple countries and all regions of the world, with a forward plan for research on the genetic epidemiology of drug dependence as well. The NSDUH is important because its sample size produces enough newly incident cases of drug dependence and related phenotypes to make feasible future genome wide association studies (GWAS), with multiple replications of the type required to distinguish falsely positive leads from candidate genes that deserve more attention and work-up in efforts to understand pathophysiological pathways and molecular mechanisms with the most public health importance.
Nevertheless, in these two lines of research, we see reason and opportunity to broaden the ranges of candidate phenotypes under study, including a return to the original concept of a candidate phenotype as a population characteristic and not just an individual-level characteristic of each organism. Due to the emphasis upon phenotype as an individual-level characteristic, many genetic epidemiologists working in the field of drug dependence have forgotten or never learned that Wilhelm Johanssen originally coined and defined the term ‘phenotype’ with a population perspective that complements a perspective on phenotype as a characteristic of an individual. This population perspective motivates thinking about the ‘phenotype’ in terms of what we might call the moments of the statistical distribution from a population. The foundation of individual-level research approaches in modern genetic epidemiology have grown out of post-Johanssen advances in cytology and work at the intersection of molecular biochemistry and genetics (17
). It now may be useful to think of some population-level candidate phenotypes for new work at the intersection of molecular genetics and epidemiological research on drug dependence. This thinking is prompted in part by some recent observations from the World Mental Health Surveys, which will be discussed next. The thinking also is prompted by a consideration of potential alternatives as might be applied in future genetic epidemiology studies that can exploit the large sample replications of the NSDUH with some minor elaborations, which will be reviewed after coverage of the WMH Surveys Consortium observations.