John Macleod, Rachel Oakes,AlexCopello, liana Crome, Matthias Egger, Mathew Hickman, Thomas Oppenkowski, Helen Stokes-Lampard, George Davey Smith
Departments of Primary Care and Gen= eral Practice (J Macleod PhD, R Oakes BSc, T Oppenkowski Msocsc4 H Stokes-Lam= pard ',sou') and Psychology (A I Copello PhD), University of Birmingham, Birmingham, UK; Academic Psychiatry Unit, Keele University, Stoke on Trent, UK (Prof I Crome Mo); Department of Social Medicine, University = of Bristol, Bristol, UK (Prof M Egger MD, Prof G Davey Smith osc); Department of Social and Preventive Medicine, University of Beme, Berne, Switzerland (Prof M Egger); and Centre for Research on Drugs and Health Behaviour, Social Science and Medicine, imperial College, London, UK (M Hickman PhD)
Correspondence to: Dr John Macleod, Department of Primary Care and General Practice, Primary Care Clinical Scie= nces and Learning Centre building. University of Birmingham, Edgbaston, Birmingham B15 2TT, UK
Background Use of illicit drugs, particularly cannabis, by young people is widespread and is associated with several types of psychological and social harm. These relations might not be causal. Causal relations would suggest that recreational drug use is a substantial public health problem. Non-causal relations would suggest that harm-reduction policy based on prevention of drug use is unlikely to produce improvements in public health. Cross-sectional evidence cannot clarify questions of causality; longitudinal or interventional evidence is needed. = Past reviews have generally been non-systematic, have often included cross-secti= onal data, and have underappreciated the extent of methodological problems associated with interpretation.
Methods We did a systematic review of general population longitudinal studies reporting associations between illicit drug use by young people and psychosocial harm= .
Findings We identified 48 relevant studies, of which 16 were of higher quality and prov= ided the most robust evidence. Fairly consistent associations were noted between cannabis use and both lower educational attainment and increased reported u= se of other illicit drugs. Less consistent associations were noted between cannabis use and both psychological health problems and problematic behaviour. All these associations seemed to be explic= able in terms of non-causal mechanisms.
Interpretation Available evidence d= oes not strongly support an important causal relation between cannabis use by young people and psychosocial harm, but cannot exclude the possibility that such a relation exists. The lack of evidence of robust causal relations prevents t= he attribution of public health detriments to illicit drug use. In view of the extent of illicit drug use, better evidence is needed.
The use of illicit drugs amongst yo= ung people seems to be widespread and may be increasing.' Cannabis is the most widely used illicit substance, although use of psychos= timulants also appears quite common; use of opiates seems less common. Most of these = drug users do not access drug treatment services and the consequences of their d= rug use are unclear. Physical health problems aside, there are concerns that illicit drug use, particularly cannabis use, could cause psychological and social problems.' Cannabis use has been shown to be associated with psychol= ogical health problems, use of other illegal drugs, reduced educational attainment, and antisocial behaviour.' The causal basis of = these associations has not been established. If associations are non-causal, harm-reduction policies based on the prevention of drug use are likely to be ineffective. Conversely, a causal association could mean that "recreational" illicit drug use, in view of its apparent extent, represents an important, and substantially hidden, public health problem.
Causal explanations for associations between drug use and psychosocial harm compete with three altemative explanations: reverse causation, where drug use is a consequence, rather th= an a cause, of psychosocial problems; bias, where the association is an artifact= of study methodology; and confounding, when drug use is associated with other factors that predispose to psychosocial problems.
A causal relation between drug use = and psychosocial harm could plausibly be mediated by two principal mechanisms: directly, through neurophysiological pathways, = or indirectly, through involvement in the criminal culture and commerce associ= ated with use of an illegal substance.'' Past reviews of the relevant evidence h= ave often been non-systematic and have used restricted search strategies. Much = evidence is cross-sectional and derives from highly selected samples. Such evidence = is limited as a basis for inferring true causal relations and their possible relevance to public health. We therefore undertook a systematic review of general population, longitudinal studies relating illicit drug use by young people to subsequent psychological and social harm.
Search strategy and selection crite= ria
We searched the general electronic databases MEDLBVE, EMBASE, CINAHL, PsycLIT, and= Web of Science, and the specialist databases of the Littde= smith Center, Drugscope, US National Institute on Drug Abuse and Substance Abuse and Mental Health Services Administration, and Addiction Abstracts, with an agreed battery of search terms (available from= the authors) in July, 2000. This search was updated in July, 2001, and again in June, 2003. Addiction Abstracts was hand-searched for the period not covere= d by the electronic database. Key individuals in the specialty of addictions
(details= available on request) were asked to identify evidence unlikely to be found through the other sources. Both published and unpublished evidence, along with that not published in English (which was translated), was judged.
We included all prospective studies= based in the general population that measured use of any illicit drug by individu= als aged 25 years or younger at the time of use and related these data to any measure of psychological or social harm assessed subsequently.
Quality assessment was undertaken a= fter initial searches in July, 2000. Two reviewers assessed methodological quality of studies independently against set criteria (sample size and representativeness, age at recruitment, duration and completeness of follow-up, apparent validi= ty and reliability of exposure and outcome measures, and degree of adjustment = for potential confounding factors). Formal quantitative quality scoring was not used, since it can be misleading and give a false sense of objectivity.'
Reviewers made an independent overa= ll assessment of study quality based on the above criteria, and assigned studi= es to categories of higher quality, uncertain quality, or lower quality. Studi= es were judged to be of higher quality if the probability of selection bias se= emed low, exposure to drugs was assessed with a validated instrument, follow-up = was over several years, and analyses were adjusted for important confounding factors. Validity and relevance of psychosocial outcome measurement was also considered. Initial agreement between reviewers was high (weighted a>0.9= ). Reviewers then discussed, and agreed, which studies of higher or uncertain quality warranted more detailed consideration. Corresponding authors on pap= ers deriving from these studies were contacted and asked to supply any relevant= unpublished data.
We assessed the potential for quant= itative synthesis of study results against criteria for combinability. Results were also summarised descriptively.
Role of the funding source
The sponsors of the study had no ro= le in study design, data collection, data synthesis, data interpretation, or writ= ing of the report
We located more than 200 publicatio= ns deriving from 48 longitudinal studies reporting associations between drug u= se by young people and psychological or social outcomes. Five studies were not published in English. All studies were observational. All had published res= ults in peer-reviewed joumals; however, some additio= nal publications in books and unpublished papers were identified through person= al contact. Many studies used composite measures of illicit drug use, making it impossible to infer effects of specific drugs. Most drug-specific results related to use of cannabis. Many studies reported substantial losses to follow-up and made either no, or little, attempt to adjust estimates for possible confounding factors. 16 studies were classified as of higher methodological quality (table 1). The remaining 32 studies are summarised, in terms of their ostensible findings and= with a brief methodological critique, in table 2. All studies were judged, but appraisal was focused on evidence from the 16 in table 1.=
Recruitment strategies, and thus the precise relation of the study population to the general population, varied substantially (tables 1 and 2). In all studies, exposure to illicit drugs w= as measured through uncorroborated self-report. Although some measures were similar across studies, no two studies measured either illicit drug exposur= e or psychosocial outcome in the same way. Additionally, potential confounding factors were inconsistently assessed across studies. Because of these considerations, we felt that quantitative synthesis (meta-analysis) was lik= ely to be misleading and did not attempt to do this.'°
We report our principal findings on relations between cannabis use and educational attainment, use of other dru= gs, psychological health, antisocial behaviour, and= other social problems. Illustrative crude and adjusted effect estimates in relati= on to these outcomes are described in table 3. Findings on relations between u= se of other illicit drugs and psychosocial problems are also summarised. Key publications are cited; a full list of publications is available on request.
Cannabis use was consistently assoc= iated with reduced educational attainment. Most relevant studies indexed this out= come through objective and apparently valid measures. The strength and magnitude= of the association varied. Adjustment of estimates for potential confounding factors generally led to their attenuation, which was often substantial.
Cannabis use was consistently assoc= iated with use of other drugs. In all but one relevant study, other drug use was indexed by uncorroborated self-report (in one study, use of injected drugs = was corroborated by inspection of injection sites)." The strength and magnitude of these associations varied, although in one study, both were substantial." In this study, as with most studies, the outcome reported was other drug use, rather than drug problems. Adjustment of estimates for potential confounding factors generally led to their attenuation.
Cannabis use was inconsistently ass= ociated with psychological problems. Some studies found no association, although ot= hers reported associations between increased use and increased problems. Within these latter studies, patterns of association with specific psychological problems were inconsistent. In most studies, psychological problems were indexed through self-report of symptoms, some assessed according to standard diagnostic criteria. The outcome was clinical mental illness (schizophrenia= ) in only one study." This report also mentioned a crude association between cannabis use and mortality from suicide, but did not report actual estimates." A crude association with all-cause mortality disappeared on adjustment for confounding factors. Adjustment of other estimates of increa= sed psychological problems for potential confounding factors generally led to t= heir attenuation, which was often substantial.
Cannabis use was inconsistently ass= ociated with antisocial or otherwise problematic behaviour. In most studies these outcomes were indexed through uncorroborated self-rep= ort. In some studies corroboration was sought from other sources. In studies that did report associations between greater use and behavi= oural problems, adjustment of estimates for potential confounding factors general= ly led to their attenuation, often substantially so. `
Evidence of effect modification acc= ording to sex and ethnic origin (where these were reported separately) was inconsistent across studies. Cannabis use at a younger age was consistently associated with greater subsequent problems.
Two studies reported associations b= etween use of cocaine and opiates and subsequent psychological symptoms; results w= ere mixed."-'°'N Amphetamines and ecstasy (3,4-methylenedioxymethamphetamine, MDMA) seem to be widely used illicit drugs.' We identified no studies meeti= ng our selection criteria that reported effects of either amphetamine or ecsta= sy use.
In this review, we found little evi= dence from longitudinal studies in the general population about the outcomes of exposure to any illicit drugs other than cannabis. We confirmed the existen= ce of evidence of associations between cannabis use and psychosocial harm; however, the extent and strength of this evidence seemed less than is perha= ps sometimes assumed. Furthermore, the causal nature of these associations is = far from clear. Some seem to fulfil at least some o= f the traditional criteria for establishing causality." They are fairly consistent; cause seems to precede effect, and a plausible mechanism can be advanced. The criterion of specificity of association was less consistently fulfilled. In several studies (tables 1 and 2) tobacco and alcohol showed similar associations as cannabis with psychosocial outcomes. This finding d= oes not suggest a causal mechanism mediated through drug-specific neurophysiological effects or involvement in criminalised commerce, since tobacco and alcohol have distinct neurophysiological effects, and they a= re not illegal. Existence of a dose-response relation, in which magnitude of the outcome varies with magnitude of the exposure is another criterion often invoked. In many studies, existence of such a relation was impossible to as= sess since only binary exposure categories were examined. Where effects of more = than two exposure categories were reported, a graded association with outcome fr= om higher to lower exposure was sometimes noted. Interpretation of these gradi= ents was complicated by the fact that in almost all studies, frequency of drug u= se, rather than dose, was assessed. Quantity used was probably closely related = to frequency, and frequency measures allowed inference of extent of drug involvement, which is of relevance to social mechanisms of causation.
However, empirical evidence has sho= wn that associations can fulfil these criteria, and sti= ll be unlikely to be causal."'" Alternative explanations of reverse causation, bias, and confounding are discussed.
Psychosocial problems might be more= a cause than a consequence of cannabis use, especially with regard to associations between use and mental illness. Some studies adjusted for psychological symptoms reported at baseline or excluded incident problems occurring during early follow-up. Nevertheless, unreported or subclinic= al psychological problems might have preceded and precipitated cannabis use. Individuals with a pre-existing tendency to experience psychological difficulties might have a greater inclination to develop problematic pattems of drug use (for example, depressed individua= ls are more likely to start smoking tobacco and less likely to stop than those who= are not depressed)." Cannabis use might also exacerbate existing predispositions to psychological problems.
Exposure to cannabis use and experi= ence of psychosocial problems might have been associated with both study recruitment and retention leading to selection bias that could affect the apparent association between cannabis use and harm. Measurement bias is another possibility. Some empirical evidence suggests reasonable validity of self-r= eported drug use, although other evidence shows that in some situations, especially general population studies in which the drug-use status of participants has= not been previously recorded, this method can be unreliable."'" Random misclassification of drug-use status will simply lead to dilution of appare= nt effects, but systematic misclassification, especially when it affects both exposure and outcome measurement, can generate spurious effects. For exampl= e, an individual may have a general tendency to value either conformist or non-conformist, behaviour, and this tendency may influence their reporting. In this situation one would expect artefactual associations between greater reported use= of cannabis and greater reported use of other drugs or other non-conformist behaviours. Since most associations of cannabis use w= ith use of other drugs, and with antisocial behaviour, are based exclusively on self-reported measures, the effect of this type of bias must be considered. In other contexts, reporting bias has been shown t= o be capable of generating strong and substantial associations between measures that, individually, seem to have high validity."
Discounting confounding is probably= the most serious interpretational challenge in observational epidemiology." Both cannabis use and adverse psychosocial outcomes seem to share common antecedents related to various forms of childhood adversity, and factors relating to peer-group and family.""•"' The relation between cannabis use and harm might simply reflect these associations; cann= abis use could be a marker, rather than a cause, of a life trajectory more likel= y to involve adverse outcomes.
There are no completely reliable me= ans to identify confounded associations within observational data, and instances w= here apparently robust observational evidence has later been shown to be serious= ly misleading are common." The importance of this issue to the epidemiolo= gy of drug use might have been underestimated. In particular, the extent to wh= ich confounding can be overcome through statistical adjustment seems to have be= en overestimated. Adjustment is useful, but its power to abolish the confounded component of an association depends on the completeness and precision of measurement of the confounders." Only three studi= es'°"s" included in our analysis had any prospectively measured indices of the early life factors that may covary with both cannabis= use and harm. It seems unlikely that even these measures were complete or preci= se.
Unmeasured, as well as measured, po= tential confounders can be taken into account through techniques such as fixed effe= cts regression and latent variable modelling."= '' These approaches allow more sophisticated adjustment. The main value of adjustment is to allow the comparison of adjusted with unadjusted estimates= , but few studies provided both of these estimates. The most informative examples= of those that did are summarised in table 3. Atten= uation of estimates towards the null value, on adjustment, suggests confounding by= the adjustment factor. In this situation, residual confounding can be assumed t= o be present. Unchanged or strengthened estimates suggest that confounding by the factor adjusted for is unlikely—confounding by another factor is still possible. In table 3, almost all adjusted estimates are substantially attenuated towards the null value. With attenuation of this relative magnit= ude even small degrees of measurement imprecision in the confounders could acco= unt for the residual effects.
Sensitivity analyses are another me= ans to explore the possibility of confounding. A recent application of this princi= ple to North American data showed that confounding by a factor termed "propensity for drug use" could explain associations between cann= abis use and use of other drugs." Both environmental and genetic factors co= uld underlie such a propensity."
Further evidence against a simple c= ausal explanation for associations between cannabis use and psychosocial harm rel= ates to population patterns of the outcomes in question. For example, incidence = of schizophrenia seems to be strongly associated with cannabis exposure over a fairly short period (four-fold to five-fold relative risks over follow-up of 10-30 years). Cannabis use appears to have increased substantially amongst young people over the past 30 years, from around 10% reporting ever use in 1969-70, to around 50% reporting ever use in 2001, in Britain a= nd Sweden.'" If the relation between use and schizophrenia were truly cau= sal and if the relative risk was around five-fold then the incidence of schizophrenia should have more than doubled since 1970. However population trends in schizophrenia incidence suggest that incidence has either been st= able or slightly decreased over the relevant time period''"
The above considerations suggest th= at a non-causal explanation is possible for most associations between cannabis exposure and both psychological and social harm. It is important to clarify these questions, and evidence meeting this requirement could come from seve= ral sources. Birth cohorts provide the ideal prospective design within which to investigate the role of early life factors." They are expensive and ti= me consuming, and ensuring complete follow-up is challenging. However two of t= he studies we identified successfully adopted this design."'" Other = ongoing birth cohorts whose participants are now entering adolescence exist." These studies could provide valuable information, especially if they incorporated approaches to measurement other than those completely reliant = on uncorroborated self-report.
The principle of "Mendelianrandomisation" is proving useful in cardiovascular and cancer epidemiology." If level of exposure to a putative environmental cause is substantially affected by a particular gene= tic polymorphism, then analysis of effect by genotype is unlikely to be confoun= ded by environmental factors. Study of polymorphisms affecting neuroreceptor affinity for the psychoactive components of cannabis may have potential in = this regard." The statistical power is generally low in such studies, howev= er, and sample sizes need to be large.90° Finally, experimental studies are the traditional approach to overcoming problems of selection bias and confounding. If experimental reduction in cannabis expos= ure were associated with reductions in psychosocial harm, this would be stronger evidence for a true causal relation. Currently, this approach is limited by= the absence of interventions that substantially or reliably reduce exposure to cannabis.'° Concerns have been expressed about the public health effect= s of ecstasy use;10' the same principles should guide research to provide evidence relating to this drug. Evidence on public heal= th effects of opiate use seems likely to be most feasibly obtained through follow-up of population-based cohorts of opiate users."
In this review we did not consider = physical health outcomes. Clearly, some types of illicit drug use lead to serious physical harm, but the extent of this problem outside known treatment populations is unclear. It is probable that cannabis use is associated with some physical harm, since most users apparently smoke the drug with tobacco. Intermittent use confined to adolescence or early adulthood might have small effects, but data confirming that this pattern of use predominates, or measuring the prevalence of other usage pattems= , are limited. Little reassurance is available from the evidence we identified. O= nly one study reported mortality up to middle adulthood and found no increase w= ith cannabis use, however the same study showed no mortality increase associated with tobacco use.'°
Drug policy is sometimes justified = on the basis of a causal relation between drug use and psychosocial harm. We have shown that evidence for this relation is not strong. However it would be na= ive to assume that scientific evidence is generally an important determinant of policy, especially in this area.10''°^
No search strategy can ensure identification of all relevant evidence. Our search was the most comprehens= ive of any we are aware of in this field and was recently updated. However, it = is probable that we missed some potentially relevant evidence. Given the gener= al issues of interpretation we have discussed, it seems unlikely that such omissions would have substantially altered our conclusions: Our quality assessment was inevitably subjective; however, we undertook it as a guide to readers and to make the task of the review more manageable. We contacted on= ly authors of higher-quality studies to identify further evidence, although ag= ain it seems unlikely that this procedure introduced substantial bias.
Despite widespread concern, we have= found no strong evidence that use of cannabis in itself has important consequences for psychological or social health. This finding is not equivalent to the conclusion that use of cannabis is harmless in psychosocial terms; problems with the available evidence render it equally unable to support this proposition. Better evidence is needed in relation to cannabis, which is wi= dely used, and in relation to other drugs that, although less widely used, might have important effects.
I Macleod, A C= opello, I Crone, M Egger, M Hickman, and
G Davey= Smith devised the search strategy. Electronic searches, expert contact, and retri= eval of references were undertaken by R Oakes and T Oppenko= wski. Hand searches were undertaken by J Macleod and
I Crone.= Quality assessment was undertaken by J Macleod, M Egger, and
H Stokes-Lampard. Data synthe= sis and interpretation was discussed by
J Macleod, A Copello, I Crone, M Egger, M Hickman,
H Stokes-Lampard, and G Davey Smith. J Macleod wrote the first draft of this report, all authors contribu= ted to the final draft.
Conflict of interest statement
This review was funded through the = UK Department of Health, Drug Misuse Research Initiative (project grant 121720= 6). JM and MH are supported by Primary Care and Public Health Career Scientist Fellowships from the Department of Health. All views expressed are those of= the authors and not necessarily those of the Department of Health.
1 Aust = R, Sharp C, Goulden C. Prevalence of drug use: key findi= ngs from the 2001/2002 British Crime Survey, findings 182. London: Home Office Research, Developme= nt and Statistics Directorate,
2 Hall ,=Solowij N. Adverse effects of cannabis. Lancet 1998; 325: 1611-16.
3 Solowij N. Cannabis and cognitive functioning. Cambridge: Cambridge University Press, 1= 998.
4 Lenton S. Cannabis as a gateway drug: comments on Fergusson and = Horwood. Addiction 2001; 96: 511-15.
5 Juni = P, Witachi A, Bloch R, Egger M. The = hazards of scoring the quality of clinical trials for meta-analysis. JAMA 1999; 282: 1054-60.
6 Resnick MD, Bearman PS, Blum RW, et al. Prote= cting adolescents from harm: findings from the National Longitudinal Study on Adolescent Health. JAMA 1997; 278: 823-32.
7 Dornbusch SM, Lin I-C, Munroe PT, Bianchi AJ. Adolescent polydrug use and viole= nce in the United Staes. Intl Adolesc Med Health 1999; Il: 197-219.
8 Guy SM, Smith GM, Bender PM. Adol= escent socialization and use of lick and illicit substances—impact on adult = health.Psycho[ Health 1993; 8: 463-87.
9 Stein JA, Smi= th GM, Guy SM, Bender PM. Consequences of adolescent drug use on young adult job behav= ior and job satisfaction. JApplPsycho(1993; 78: 463-74.
10 Brook JS, Cohen P, Brook DW. Longitudinal study of co-occurring psychiatric disorders and substance use. JAm Aced Child Adolesc Psychiatry 1998; 37: 322-30.
11 Brook JS, Whiteman M, Finch Si, Cohen P. Young adult drug use and delinquency: childhood antecedents and adolescent mediators. JAm Acad Child Adolesc Psychi= atry 1996; 35: 1584-92.
12 Brook JS, Richter L, Whiteman M,= Cohen P. Consequences of adolescent marijuana use: incompatibility with the assumption of adult roles. Genet Soc Gen Psycho/Moeagr 1999; 125: 193-207.
13 Brunswick AF, Messed P. Drugs, lifestyle, and health: a longitudinal study of urban black youth. Am l Publ= ic Health 1986; 76: 5257.
14 Brunswick AF, Messed PA. Life stage, substance use and health decline in a community cohort of urban African Americans. JAddict Die 1999; 18: 53-71.
15 Fergusson DM= , Horwood Q. Does cann= abis use encourage other forms of illicit drug use? Addiction 2000; 95: 505-20.
16 Fergusson DM= ,Lynskey MT, Horwood Q. The short-term consequences of early onset cannabis use. JAbnonn Child Psycho/ 1996=;24:499-512.
17 Fergusson DM= , Horwood LJ, Swain-Campbell N. Cannabis use and psycho= social adjustment in adolescence and young adulthood. Addiction 2002; 97: 1123-35.
18 Fergusson DM, Horwood LJ. Early onset cannabis use and psychosocial adjustmen= t in young adults. Addiction 1997; 92: 279-96.=
19 Fergusson DM= , Horwood Q, Swain-Campbell NR. Cannabis dependence and psychotic symptom= s in young people. Psycho/ Med 2003; 33: 15-21.
20 Silva PA, Stanton WR. From child to adult: the Dunedin multi disciplinary health and development study. Auck= land: Oxford University Press, 1996.
21 McGee R, Williams S, Poulton R, Moffitt T. A longitudinal study of cannabis use and = mental health from adolescence to early adulthood. Addiction 2000; 95: 491-= 503.
22 Arseneault L, Cannon M, Poulton R, Murray R, Caspi A, Moffit TE. Cannabis = use in adolescence and risk for adult psychosis: longitudinal prospective study. B= MJ 2002; 325: 1212-13.
23 Brook JS, B= alka EB, Whiteman M. The risks for late adolescence of early adolescent marijuana use. Am JPablic Health 1999; 89: 1549-54.
24 Newcomb MD, Scheier LM, Bender PM. Effects of adolescent drug use on adult mental health: a prospective study of a community sample. Exp Clin Psychophannacol 1993; 1: 241.
25 Newcomb MD, Bender PM. Consequen= ces of adolescent drug use: impact on the lives of young adults. Newbury Park, CA= , USA: Sage Publications Inc, 1= 988.
26 Newcomb MD. Psychosocial predictors and consequences of drug use: a developmental perspective within a prospective study. )= Addict D is 1997; 16: 51-89.
27 Newcomb MD, = Vargas-Cremona J, Galaif ER. Dreg problems and Psychological Distress among a community sa= mple of adults: predictors, consequences, or confound? J Comm Psycho) 1999; 27: 405-429.
28 Kandel D. Stages in adolescent involvement in drug use. Science 1= 975; 190: 912-14.
29 Kandel D, Chen K, Gill A. The impact of drug-use on earnings - A life-= span perspective. Social Forces 1995; 74: 243-70.
30 Friedman AS, Kramer S, Kreisher C, Granick S. The relationships of substance abuse to illegal and violent behavior, in a comm= unity sample of young adult African American men and women (gender differences). =J Subst Abuse 1996; 8: 379-402.
31 Kaesmer R. New estimates of th= e effect of marijuana and cocaine use on wages. Indasrr Labor Relar Rev 1994; 47: 454-70.
32 Windle M. Mate similarity, heavy substance use and family history of problem drink= ing among young adult women. if Stud Alcohol 1997; 58: 573-80.
33 White HR, L= oeber R, Stouthamer-Loeber M, Farrington DP. Developmental associations between substance use and violence= . Develop Psycho-pasha/ 1999; 11: 785-803.
34 Ellickson P, Bui K, Bell R, McGuigan KA. Does early drug use increase the risk of dropping out of high school? if Drug Issues 1998; 28: 357-80.
35 Ellickson PL, McGuigan KA. Early predictors of adolescent violence. Am 7Pub Health 2000; 90: 566-72.
36 Bray JW, Za= rkin GA, Ringwatt C, Qi = J. The relationship between marijuana initiation and dropping ou= t of high school. Health Ecmt 2000; 9: 9-18.
37 Stenbacka M, Allebeck P, Brandt L, Romelsjo A. Intravenous drug abuse in young men: risk factors assessed in a longitudinal perspective. Scand7 Soc Med 1992; 20: 94-101.
38 Zammit S, Allebeck P, Andreasson S, Lundberg I, Lewis G. Self reported cannabis use as a risk factor for schizophrenia in Swedish conscripts of 1969: historical cohort study. BMJ 2= 002; 325: 1199-1201.
39 Andreasson S, Allebeck P. Cannabis and mortality among young men: a longitudinal study of Swedish conscripts. Scand J Soc Med 1990; 18: 9-15.
40 Bates ME, L= abouvie EW. Adolescent risk factors and the prediction of persi= stent alcohol and drug use into adulthood. Alcohol Clin Exp Res 1997; 21: 94= 4-50.
41 Juon HS, Ensminger ME. Childhood, adolescent, and young adult predictors of suicidal behaviors: a prospective study of African Americans. J Child Psycho/ Psychiae Allied Disciplines 1997; 38: 553-63.
42 Sadava SW. Patterns of college student drug use: a longitudinal social learning study. Psycho/ Rep 1973; 33: 75-86.
43 Annie HM, Watson C. Drug use and= school dropout: a longitudinal study. Canadian Counsellor 1975; 9: 155-62.
44 Benson G. Dr= ug-related medical and social conditions in military conscripts. ActaPsychiatrica Seandinavica 1984; 70: 550-58.
45 Benson G. Course and outcome of = drug abuse in military conscripts. Acre Psychiatrica Scandinavian 1985; 71: 38-47.
46 Friedman AS, Utada AT, Glickman NW, Morrissey MR. Psychopathology = as an antecedent to, and as a "consequence" of, substance use, in adolescence. if Drug Educ 1987; 17: 233-44.
47 Choquet M, Ledoux S, Menke H. Drug i= ntake, health and demands for help in adolescents: a longitudinal approach.=
P sychiatne et Prychobialogie 1988; 3:= 227-44.
48 Farrell AD, = Danish SJ. Peer drug associations and emotional restraint - causes or consequences of adolescents drug-use. 7 Consuls Clin Psycho(1993; 61: 327-34.
49 Huizinga D, Locher R, Thomberry TP. Longitudinal study of delinquency, drug use, sexual activity and pregnancy amongst children and youth in three cities. Pub Health Rep 1993; 108 (suppl 1):90-96.
50 Sanford M, = Offord D, McLeod K, Boyle M, Byrne C, Hall B. Pathways into the work force: antecedents of school and work force status. JAm Aced Child Ado/ere Psychiatry 1994; 33: 1036-46.
51Scholenberg J , = Bachman JG , O'Malley PM ,= Johnston LD. High school educati= onal success and subsequent substance use: a panel analysis following adolescents into young adulthood.
J Health Soc Behan 1994; 35: 45-62.
52 Anthony JC, =Petronis KR. Early-onset drug use and risk of later d= ing problems. Drug Alcohol Depend 1995; 40: 9-= 15.
53 Farrington D= P. The 12th jack-Tizard-Memorial-Lec= ture—the development of offending and antisocial-behavior from childhood: key findin= gs from the Cambridge= study in delinquent development. J Child Peychol Psychiatry 1995; 36: 929-64.
54 Krohn MD, Lizorte AJ, Perez CM. The interrelationship between substance use and precocious transitions to adult statu= ses.,) Health Soc Behan 1997; 38: 87-103.
55 Luther SS, Cushing G. Substance = use and personal adjustment among disadvantaged teenagers: A sin-month prospective study. J Youth Adolesc 1997= ; 26: 353-72.
56 Stanton B, Fang X, Li X, Feigelman S, Galbraith J, Ricardo I. Evolution of risk behaviors over 2 years among a cohort of urban African American adolescents. Arch PediaerAdolesc Med 1997; 151: 398-406.
57 Rao = U, Daley SE, Hammen C. Relationship between depression a= nd substance use disorders in adolescent women during the transition to adulth= ood.JAm Aced Child Adolesc Prychiany 2000; 39: 215-22.
58 Weiser M, Knobler HY, Noy S, Kaplan= Z. Clinical characteristics of adolescents later hospitalized for schizophreni= a. Am )Med Genes 2002; 114: 949-55.
59 Epstein I, = Temir A. Health-related behavior of adolescents: change over mné.JAdolesc Health Care 1984; 5: 91-95.
60 Kaplan HB, Martin SS, Johnson RJ, Robbins CA. Escalation of marijuana use: application of a general theory of deviant behavior. J Health Soc Beh= an 1986; 27: 44-61.
61 Tubman JG, Vicary JR, von Eye A, Lerner JV. = Longitudinal substance use and adult adjustment. if Subse Abuse 1990; 2: 317-34.
62 Scheier LM, Newcomb MD. Differentiation of early adolescent predictors of drug use vers= us abuse: a developmental risk-factor model..7 Subs Abuse = 1991; 3: 277-99.
63 Hammer T, Vaglum P. Further course of mental health and use of alcohol and tranquilizers after cessation or persistence of cannabis use in young adulthood: a longitudinal study. Stand J Soc Med 1992; 20: 143-50.
64 Degonda M, Angst J. The Zurich study. XX. Social phobia and agoraphobia. Eur= Arch Psychiatry Clin Neumsci 1993; 243: 95-102.
65 Romero E, Luengo MA, Otero-Lopez JM. The relationship between self-esteem and drug use in adolescents: a longitudinal analysis. Revises de = Psicologia Social 1995; 10; 149-59.
66 Andrews JA, Duncan SC. Examining= the reciprocal relation between academic motivation and substance use: effects = of family
relationships, self-esteem, and general deviance. j Behavior Med 1997; 20:523-49.
67 Patton GC, Harris R, Carlin JB, = et al. Adolescent suicidal behaviours: a population-ba= sed study of risk. Psycho/ Med 1997; 27: 715-24.
68 Hansell S, White HR. Adolescent drug use, psychological distress, and physical symptom= s..?Health Sac Behan 1991; 32: = 288-301.
69 White HR. Hansell S. Acute and long-term effects of drug use on aggression from adolescence into adulthood. .7 Drug Issues 1998; 28: 837-58.
70 Costello EJ, Erkarili A, Federman E, Asgold A. Development of psychiatric comorbidity with sub= stance abuse in adolescents: effects of timing and sex. J Clin Child Psycho/ 1999; 28: 298-311.
71 Duncan SC, Strycke= r LA, Duncan TE. Explori= ng associations in developmental trends of adolescent substance use and risky sexual behavior in a high-risk population. if Behavior Med 1999; 22: 21-34.
72 Perkonigg A, Lieb R, Hoffer M, Schuste= r P, Sontag H,
Winchen= H-U. Patterns of cannabis use, abuse and dependence over time: incidence, progression and stability in a sample of 1228 adolescents. Addiction 1999; 94: 1663-78.
73 Huertas Zareo I, Pereira Berenguer I, Chover Larea J, et al. Sc= hool failure in a cohort of adolescents. Atencion Pdmana 1999; 23: 289-95.
74 Braun BL, Murray DM, Sidney S. L= ifetime cocaine use and cardiovascular characteristics among young adults: the CARD= IA study. Am J Public Health 1997; 87: 629-34.
75 Brook DW, Brook JS, Pahl T, Montoya I. The longitudinal relationship between dru= g use and risky sexual behaviors among Colombian adolescents. Arch Pediatr Ado/ere Med 2002; 156: 1101-07.
76 Egger M, Sch= neider M, Davey Smith G. Spurious precision? meta-analysis of observational stu= dies. BMJ 1998; 316: 140-44.
77 Lynskey MT, Coffey C, Degenhardt L, Carlin JB, Paton G. A longitudinal study of = the effects of adolescent cannabis use on high school completion. Addict= ion 2003; 98: 685-92.
78 Patton GC, Coffey C, Carlin JB, =Degenhardt L, Lynskey M, = Hall W. Cannabis use and mental health in young people: cohort study. BM' 2002; 325: 1195-98.
79 Hill AB. The environment and disease: association or causation. Prot R= Soc Med 1965; 58: 295-300.
80 Davey Smith G, Phillips AN, Neaten JD. Smoking as "independent" risk factor for suicide: illustration of an artifact from observational epidemiology? Lancet 1992; 340: 709-712.
81 Macleod J, = Davey Smith G, Heslop P, Metcalfe C, Carroll D, Hart = C. Psychological stress and cardiovascular disease: empirical demonstration of bias in a prospective observational study of Scottish men. BMJ 2002; 324: 1247-51.
82 Lipkus IM, Barefoot JC, Williams RB, Siegler IC. Personality measures as predictors of smoking initiation and cessation in the UNC Alumni Heart Study. Health Psycho! 1994; 13: 149-55.
83 Colon HM, Ro= bles RR, Sahai H. The validit= y of drug use responses in a household survey in Puerto R= ico: comparison of survey responses of cocaine and heroin use with hair tests. IntJEpidemiol 2001; 30: 104249.
84 Anthony JC, Neumark YD, Van Essen ML. Do I do what I say? a perspect= ive on self-report methods in drug dependence epidemiology. In: Stone AA, Turkkan JS,Bachrach CA,Jobe JB, Kurtzman H= S, Cain VS, eds. The Science of Self-report: implication= s for Research and Practice. Mahwah: Lawrence Erlbaum, 2000: 175-198.
85 Davey Smith G, Ebrahim S. Data dredging, bias or confounding. BMJ 2002; 325: 1437-38.
86 Robins LN. Sturdy childhood predictors of adult antisocial behavi= our: replications from longitudinal studies. Psycho Med 1978= ; 8: 611-22.
87 Maughan B, McCarthy G. Childhood adversities and psychosocial disorders. Br Med Bull 1997; 53: 156-69.
88 Phillips AN,=Davey Smith G. How independent are independent effects? relative ri= sk estimation when correlated exposures are measured imprecisely. Clin Epidemic( 1991; 44: 1= 223-31.
89 Berkane M. Latent variable modelling and applications to causality. New York: Springer Verlag, 1997.
90 Morral AR, McCaffrey DF, Paddock SM. Reassessing the marijuana gateway effect. Addiction 2002; 97: 1493-1503.
91 van den Brc= e MB, Johnson EO, Neale MC, Pickens RW. Genetic and environmental influences on drug use and abuse/dependence in male and female twins. Drug Akohol Depend 1998; 52: 231-241.
92 Jablensky A. Epidemiology of schizophrenia: the global burden of disease and disability.=Eur Arch Psychiatry Olin Neumsci= 2000; 250: 274-85.
93 Degenhardt L, Hall W, Lynskey M. Testing hypotheses about = the relationship between cannabis use and psychosis. Drug Akohol Depend2003; 71: 37-48.
94 Susses E, Terry MB, Mane T. The birth cohorts grow up: new opportunities for epidemiology. Paediatr Penna. Epidermal 2000; 1= 4: 98-100.
95 Golding J, Pembrey M, Jones R, for the ALSPAC Study Team. The Av= on Longitudinal Study of Parents and Children, I study methodology. Paediarr Perinat Epidemia= l2001; 15: 74-87.
96 Davey Smith G,Ebrahim S. "Mendelian= randomisation": can genetic epidemiology contrib= ute to understanding environmental determinants of disease? I= ncEpiderniol 2003; 32: 1-22.
97 Gadzicki D, Muller-Vahl K, Stuhrman= n M. A frequent polymorp= hism in the coding exon of the human cannabinoid receptor (CNRI) gene. Mal Cell Probes 1999; 13: = 321-3.
98 Davey Smith G,Harbord R, Ebrahint= S. Fibrinogen, C-reactive protein and coronary heart disease: does Mendelian randomization suggest the associations are non-causal? Q7Med2004; 97: 163-66.
99 Davey Smith G,Ebrahim S. Mendelian randomisation: prospects, potentials and limitations.=Int 3 Epidemic,! 2004; 33: 30-42.
100 White D, Pitts M. Educating you= ng people about drugs: a systematic review. Addiction 1998= ; 93: 1475-87.
101 Ricaune GA, McCarm UD. Assessing long-term ef= fects of MDMA (ecstasy). lancet 2001; 358: 1831-32= .
102 Copeland L,= Budd J, Robertson JR, Elton R. Changing parterre in causes of death in a cohort of injecting drug users 1980-2001. Arch Inter= n Med (in press).
103 Davey Smith G, Ebrahim S, Frankel S. How policy informs the evidence. BM,7 2001; 322: 184-85.
104Kendell RE. Cannabis condemned: the proscription of Indian hemp. Addiction 2003; 98: 143-151.