Trajectories of Drug Use and Mortality Outcomes Among Adults Followed Over 18 Years

ABSTRACT

BACKGROUND

For adults in general population community settings, data regarding long-term course and outcomes of illicit drug use are sparse, limiting the formulation of evidence-based recommendations for drug use screening of adults in primary care.

OBJECTIVE

To describe trajectories of three illicit drugs (cocaine, opioids, amphetamines) among adults in community settings, and to assess their relation to all-cause mortality.

DESIGN

Longitudinal cohort, 1987/88 – 2005/06.

SETTING

Community-based recruitment from four cities (Birmingham, Chicago, Oakland, Minneapolis).

PARTICIPANTS

Healthy adults, balanced for race (black and white) and gender were assessed for drug use from 1987/88—2005/06, and for mortality through 12/31/2008 (n = 4301)

MEASUREMENTS

Use of cocaine, amphetamines, and opioids (last 30 days) was queried in the following years: 1987/88, 1990/91, 1992/93, 1995/96, 2000/01, 2005/06. Survey-based assessment of demographics and psychosocial characteristics. Mortality over 18 years.

RESULTS

Trajectory analysis identified four groups: Nonusers (n = 3691, 85.8%), Early Occasional Users (n = 340, 7.9%), Persistent Occasional Users (n = 160, 3.7%), and Early Frequent/Later Occasional Users (n = 110, 2.6%). Trajectories conformed to expected patterns regarding demographics, other substance use, family background and education. Adjusting for demographics, baseline health status, health behaviors (alcohol, tobacco), and psychosocial characteristics, Early Frequent/Later Occasional Users had greater all-cause mortality (Hazard Ratio, HR = 4.94, 95% CI = 1.58–15.51, p = 0.006).

LIMITATIONS

Study is restricted to three common drugs, and trajectory analyses represent statistical approximations rather than identifiable “types”. Causal inferences are tentative.

CONCLUSIONS

Four trajectories describe illicit drug use from young adulthood to middle age. Two trajectories, representing over one third of adult users, continued use into middle age. These persons were more likely to continue harmful risk behaviors such as smoking, and more likely to die.

Electronic supplementary material

The online version of this article (doi:10.1007/s11606-011-1975-3) contains supplementary material, which is available to authorized users.

INTRODUCTION

While illicit drug use is most common among young adults, it can persist into middle age.^1,2 In the US, 20% of 26- to 34-year-olds used an illicit drug in 2009, over half (11.8%) reporting a drug other than marijuana.³ Among adults 50–59, past-year drug use nearly doubled in the past decade from 5.1% to 9.4%.^4,5 Most adults who use illicit drugs do not meet criteria for diagnosis of a drug use disorder.

Identifying whether adult drug use carries long term risks to health or well-being could help clarify whether screening for drug use in primary care is advisable. The US Preventive Services Task Force has judged that current data are inadequate to recommend screening.⁶ Furthermore, epidemiologic studies seeking to correlate drug exposure and health outcomes face important barriers. One barrier is that drug use varies over the lifespan.² Thus long-term outcome analysis after a single exposure assessment is problematic.⁷ Additionally, longitudinal data regarding illicit drug use and outcomes in the general population (as opposed to clinically defined treatment samples) are extremely sparse. Finally, where outcomes such as mortality are considered, important potential confounders such as health status, psychosocial risk and family upbringing must be considered, and yet are often lacking from general population epidemiologic cohorts.

The Coronary Artery Risk Development in Young Adults (CARDIA) Study provides a unique opportunity to observe trajectories of drug use and outcomes over time in a community-based sample, starting during young adulthood (in 1987–88) and following participants into middle age (2005-6).⁸ Using repeated measures of drug use and health from CARDIA and trajectory-based methods,⁹ we sought to describe patterns of exposure over time to three drugs that represent a potentially higher level of deviance and risk than marijuana (cocaine, amphetamines, opioids), often described as “hard drugs”.^10–12 A set of age-based trajectories of drug use was identified and compared to participant characteristics to assess convergence with epidemiologic findings on drug use. Then, as an indicator of whether trajectory groups were likely to be meaningful clinically, we evaluated whether adjusted all-cause mortality was higher among trajectories with higher levels of use.

METHODS

Study Design and Sample

We utilized 18 years of data from the Coronary Artery Risk Development in Young Adults (CARDIA) study. CARDIA is an institutional review board-approved longitudinal cohort of 5,115 black and white women and men recruited from four U.S. cities, aged 18–30 years and healthy at the time of enrollment in 1985.¹³ Participants were solicited by random digit dialing in three urban communities: Birmingham, Alabama; Chicago, Illinois; and Minneapolis, Minnesota; supplemental door-to-door recruitment was implemented in areas with low telephone subscription rates. Participants from Oakland, California, were drawn randomly from membership files of Kaiser Permanente Medical Care Program, which had broad market penetration. Consenting participants underwent a baseline examination and follow-up examinations at years 2, 5, 7, 10, 15, and 20.

Our analysis focuses on participants who completed: (a) the second in-person examination (1987/88), when the target illicit drugs (cocaine, opioids, amphetamines) were first queried; and (b) at least one additional in-person examinations in 1990/91, 1992/93, 1995/96, 2000/01 or 2005/06.

Drug use Measures

The target variable for analysis was a self-reported indicator combining recent use (past 30 days) of cocaine, opioids and amphetamines. Drug misuse questions were drawn from the 1985 National Household Survey on Drug Abuse.¹⁴ Participants were asked: “Have you ever used (substance)?” and “During the last 30 days, on how many days did you use (substance)?” Accordingly, a value of 0 was assigned to persons with no use in the prior 30 days. Substances queried included cocaine, amphetamines (not specifically methamphetamine) and opioids, including heroin as well as prescription pain relievers for nonmedical reasons. The number of days for the three substance categories was summed to produce a score from 0 to 90 (“drug days score”). Marijuana use was similarly queried.

Other Measures

Most other measures were obtained in 1987/88, unless collected only on another exam (typically 1985/86, the CARDIA Study baseline exam). In addition to race, gender and age, socio-demographics included education (<high school, high school only, >high school), and marital status. Ability to afford all basic necessities (yes/no) served as a more plausible indicator of economic status than income among college-age adults. Psychosocial measures included self-report of a psychiatric diagnosis (1985/86), a social support measure indicating instrumental and emotional support as well as social network adequacy,^15,16 and anxiety (1990/91).¹⁷ A score from a validated questionnaire assessed how much the childhood family environment was unsupportive, cold, neglectful, or characterized by substance use (“risky family” score).^18,19

Smoking was classed as Current (last 30 days), Past or Never.²⁰ Alcohol use was summed from the typical number of weekly drinks.²¹ To compare drinking at levels that connote significant risk, we used World Health Organization thresholds of >21 drinks/week for men, and >14 drinks/week for women,²² higher than US thresholds.²³

Covariates for the mortality analysis included body mass index (<25, 25-29, >29) in 1987/88, and general self-reported health (Excellent, Good, Fair, Poor) collected in 1985/86, as well as depressive symptoms according to the Center for Epidemiologic Studies Depression Scale (CES-D),²⁴ applying the threshold ≥16 for significant depressive symptoms.

Information on deaths among cohort members was obtained through direct reporting to staff at protocol-specified contacts with the participant or family (every 6 months) and by record linkage with the National Death Index. Follow-up was censored for analysis 12/31/2008.²⁵

Analysis

Trajectory Group Formation

This analysis classed participants into age-based trajectories using the summed drug days score as measured at each examination. Semiparametric group-based trajectory models treat the sample as a mix of unobserved but implicit groups, each summarized by a finite set of polynomial functions of age or time.²⁶ The method identifies groups with similar trajectories based on maximum likelihood estimates, with each person assigned a probability of membership in each group. The approach seeks to explain as much variation as possible while maintaining parsimony and interpretability. Solution selection was guided by a statistical indicator of explanatory power that penalizes overfitting (Bayesian Inference Criterion, BIC) and by the investigators’ assessment of the interpretability of the groups, as recommended.²⁶ The average posterior probability of membership for any trajectory exceeded 0.90, with 0.70 being the recommended minimum. Because the most common pattern is no recent drug use, a zero-inflated Poisson distribution was assumed. Time was based on age at data collection (e.g. age 20, 21) rather than study wave. All models were conducted with SAS 9.0 and its PROC TRAJ module. (http://www.andrew.cmu.edu/user/bjones/download.htm).

Testing Reasonableness of the Groups

Reasonableness of the trajectory groups was assessed based on their agreement with population-based epidemiologic patterns with regard to (a) gender, (b) psychosocial predictors and (c) tobacco/alcohol. Unadjusted comparisons used chi-square tests and analyses of variance. Multivariable analyses used multinomial models where drug use trajectory was the dependent variable. All models included race, sex and age (>25 years in 1987/88 versus ≤25), with age dichotomized to reflect the distinction between emerging adulthood (18-25) and thereafter. Race (black, white) and gender (male, female) were treated as a 3-degree of freedom variable based on the balanced sample design and prior analyses of CARDIA. The psychosocial predictor model included self-reported psychiatric diagnosis (at baseline), risky family score, social support, and economic status. The tobacco/alcohol model included tobacco smoking and alcohol consumption in 1987/88. To adjust for potential differences in cohort retention, all models included an indicator of whether participants attended all six CARDIA examinations after baseline (1987/88, 1990/91, 1992/93, 1995/96, 2000/01, 2005/06).

All-Cause Mortality

All-cause mortality was compared with Kaplan–Meier curves and the log-rank test. A proportional hazards model adjusted for potential confounders in 1987/88, including age, sex and race, baseline general health status, body mass index, education, economic status, depressive symptoms and location. Because prior analyses had shown that smoking and alcohol use changed over time, with differing implications for health outcomes, this model used measures from both 1990/91 and 1995/96. General self-reported health was included, but it was assessed only in 1985/86 (Excellent, Good, Fair, Poor).

Sensitivity to Differences Based on Birth Year

The conceptual framework for trajectories assumes behaviors are influenced by historic context.²⁷ Epidemiologically, this may be construed as a combination of period and cohort effects.²⁸ Because most trajectory analyses draw on persons of similar age during a single inception year, such effects are not typically explored. However, we tested the degree to which the trajectory shape varied for persons who differed by age at study entry (using the TCOV “age at study entry” statement in SAS PROC TRAJ^26,29). We tested the interactions between study year and participant age at study entry (including Time*Age, Time*Age², and Time*Age³). To illustrate interactions, we plotted model-derived trajectory curves for a person of minimum (20 years), average²⁶ and maximum³² age in 1987/88. Finally, we tested whether younger members of the sample dominated trajectories with heavier use by comparing age (at inception) across trajectory groups.³⁰

RESULTS

Study Sample

Among 5115 participants, 4301 (84%) fulfilled criteria of completing the drug questionnaire in 1987/88, and at least one additional drug assessment. Starting with 1987/88, the maximum number of exams a person could attend by 2005/06 was six. The modal response pattern (62%) was to attend 6 exams, with 78.4% attending ≥5. Attendance in the final exam year (2005/06) was not different among the identified trajectory groups (72% to 77% across groups, p = 0.27), and the mean number of exams attended (5.25, SD = 1.1), did not differ among the groups, (df = 3, p = 0.26).

Compared to excluded participants, analyzed participants were more likely to be white (51.6% versus 31.8%, p < 0.001) and female (55.1% versus 51.1%, p = 0.04), a pattern seen previously in this cohort¹² and others.³⁰ Analyzed and excluded participants differed in baseline smoking (28.7% versus 39.4%, respectively, p < 0.001), and marijuana use (ever used, 71% versus 66%, p < 0.001), but not in weekly alcohol (M = 4.8 versus 5.3 standard drinks, p = 0.21).

Development of Trajectory Groups

We designated four groups with respect to drug use from ages 20 to 50 (Fig. 1, Table 1). The 4-group model offered parsimony in terms of fit and interpretability. A 3-group model had inferior fit (based on BIC = -6961) and did not distinguish a subgroup with low levels of use that appeared to escalate slightly during middle age (noted as “Persistent Occasional Users”; see below). The Bayesian Information Criterion (BIC) was slightly improved with a 5-group solution (BIC = -6572), but produced two groups that were not distinguishable visually, and thus we retained the 4-group solution (see Fig. 1).

Figure 1.

Trajectories of drug use, combining last 30 days’ use of cocaine, amphetamines and opioids as reported over 18 years follow-up from 1987/88 to 2005/06, with individuals self-reporting at up to six in-person examinations (mean 5.25, SD 1.1). Drug use days were summed to generate a score ranging from 0 to 90, and group-based trajectory models included a covariate reflecting age at cohort entry (see Methods).

Table 1.

Characteristics of Participants in Four Drug Use Trajectory Groups (Years 1987/88-2005/06; Ages 20-50) Demographic and Psychosocial Characteristics

Characteristics	Non-user	Early Occasional User	Persistent Occasional User	Early Frequent/Later Occasional User	p†
	N = 3691	N = 340	N = 160	N = 110
Demographic
Age in 1987/88
Mean, (SD)	27.0 (3.6)	27.1(3.3)	26.2(3.5)	26.8 (3.5)	0.04
Sex/Race
Black Male (%)	18.3	30.6	46.3	31.8
Black Female (%)	28.8	21.5	19.4	26.4
White Male (%)	24.2	25.9	22.5	21.8
White Female (%)	28.8	22.1	11.9	20.0
Educational attainment as of 1987/88‡
< High School (%)	6.0	10.6	17.0	26.4	<0.001
High School only (%)	23.8	37.9	42.8	31.8
> than High School (%)	70.1	51.5	40.3	41.8
Marital Status in 1987/88
Married (%)	32.8	24.7	13.8	20.9	<0.001
Partnered, Non-married (%)	12.2	18.5	21.4	17.3
Single, Divorced, Other (%)	55.0	56.8	64.8	61.8
Difficulty in Paying Basic Necessities 1987/88 (%)	27.4	37.7	44.0	41.8	<0.001
Psychosocial
Self-reported mental disorder diagnosis, 1987/88 (%)	6.8	6.2	10.0	10.0	0.22
Risky family upbringing (z-transformed) ‡	-0.04	0.1	0.45	0.16	<0.001
Social Support Score (z-transformed), 1987/88	0.03	-0.04	-0.22	-0.10	0.005
Trait-Anxiety Score (z-transformed) ‡ 1990/91	-0.05	0.10	0.46	0.44	<0.001
Ever arrested, 1985/86-1987/88(%)	3.1	12.4	22.0	26.9	<0.001
Ever convicted of a crime: 1985/86-1987/88(%)	1.2	5.3	9.4	11.1	<0.001
Ever jailed, 1985/86-1987/88(%)	2.6	8.8	18.4	21.3	<0.001
Alcohol, smoking and marijuana use
Risky Alcohol Use (above weekly threshold) (%)
1987/88‡	3.1	12.1	17.0	23.2	<0.001
2005/06‡	3.2	11.7	12.7	12.8	<0.001
Smoking (last 30 days)(%)
1987/88‡	24.4	50.4	64.2	60.8	<0.001
2005/06‡	14.1	43.8	57.8	47.4	<0.001
Marijuana Use (last 30 days)(%)
1987/88‡	17.1	70.0	66.3	60.0	<0.001
2005/06‡	7.9	34.2	38.2	28.2	<0.001
Marijuana, mean number of days in last 30 (SD)
1987/88	1.3 (4.4)	7.1 (9.0)	7.4(8.5)	6.9 (9.3)	<0.001
2005/06	0.8 (4.1)	3.5 (7.3)	3.9 (7.6)	4.6 (9.2)	<0.001

†Categorical data are compared with Chi-square/Fisher exact tests. Continuous variables were compared with Analysis of Variance

‡Number of persons not providing data for specific variables are as follows: Education in 1987/88 (n = 4). Trait-Anxiety Score in 2000/01 (n = 322), History of risky family upbringing in 2000/01 (n = 863), Risky Alcohol Use in 1987/88 (n = 14) and in 2005/06 (n = 1075), Smoking in1987/88 (n = 24) and in 2005/06 (n = 1033), Marijuana Use (in last 30 days) in 1987/88 (n = 3) and in 2005/06 (n = 1126)

Unadjusted Comparisons

Non-users (n = 3691, 85.8%) reported no current cocaine, amphetamines or opioids at any CARDIA exam. The three other trajectories differed in time-course of use. They included persons with infrequent use that extinguished (Early Occasional Users, n = 340, 7.9%); persons with infrequent use that persisted or increased in middle age (Persistent Occasional Users, n = 160, 3.7%), and persons with high use in young adulthood persisting at lower levels thereafter (Early Frequent/Later Occasional Users, n = 110, 2.6%). Among these three groups, over 1/3 fell into trajectories with continuing use into middle age. For these groups, the mean days of use in the past 30 was less than 7 for any one substance, with cocaine being most common (Table 2).

Table 2.

Use of Cocaine, Amphetamines and Opioids in Four Drug Use Trajectory Groups. (CARDIA, Years 1987/88-2005/06; Ages 20-50)

Characteristic	Nonmarijuana Drug Use Trajectories (N = 4301)
	Nonuser	Early Occasional User	Persistent Occasional User	Early Frequent/ Later Occasional User	p†
	N = 3691	N = 340	N = 160	N = 110
At Baseline (1987/88):
Any use in past 30 days (%,) of:
Cocaine	0	60.9	51.3	51.8	<0.001
Amphetamines	0	7.4	7.6	12.7	<0.001
Opioids	0	0.3	2.5	10.9	<0.001
Two or More Drugs	0	1.8	5.6	10.0	<0.001
Current use, mean (SD) days in last 30 of:
Cocaine	0	1.1(1.2)	1.9 (2.9)	6.5 (8.2)	<0.001
Amphetamines	0	0.2 (0.6)	0.4 (1.7)	1.5 (4.6)	<0.001
Opioids	0	0.003 (0.05)	0.1 (0.6)	1.2 (4.2)	<0.001
At last follow-up (2005/6):
Any use in past 30 days (%,) of:
Cocaine‡	0	9.7	28.1	11.8	<0.001
Amphetamines‡	0	2.5	3.6	6.3	<0.001
Opioids‡	0	2.4	12.4	6.4	<0.001
Two or More Drugs	0	0	7.9	5.1	<0.001
Current use, mean (SD) days in last 30 of:
Cocaine	0	0.3(0.8)	3.0(5.2)	2.9 (7.7)	<0.001
Amphetamines	0	0.07 (0.5)	0.2 (1.2)	1.1 (4.9)	<0.001
Opioids	0	0.1 (0.5)	0.8 (2.4)	1.1 (4.8)	<0.001

†Categorical data are compared with Chi-square/Fisher exact tests.

‡Missing data for Any Drug Use in past 30 days in 2005/06: Cocaine (n = 841), Amphetamines (n = 1073), Opioids (n = 1042)

The unadjusted analyses showed that the three drug-using trajectories were less likely to have graduated high school (1987/88) or college (2005/2006), less likely to be married, more likely to report economic difficulty, and had a more adverse upbringing. Black men were over-represented in these groups, especially Persistent Occasional Users (46% were black men, compared with 23% in the sample overall). Psychiatric diagnosis was common among Persistent Occasional Users and Early High/Later Occasional Users, while social support was lower, and anxiety higher for these two groups (Table 1).

As expected, trajectories of illicit drug use tracked with tobacco and alcohol (Table 1). Persistent Occasional Users were especially likely to report smoking at 20 year follow-up.

Multivariable Adjusted Analyses

Multinomial logistic regressions are shown in Tables 3 and 4. One model tests for association with age, race, and psychosocial variables (Table 3); the other incorporates age, race and alcohol/tobacco use (Table 4). In both, black men had significantly greater odds of representation in the Persistent Occasional User and the Early Frequent/Later Occasional User groups, and black women had lower odds of being among the three drug-using groups, although the estimate was not always statistically significant. Among social factors, a more risky family background was associated with membership in the Persistent Occasional User group, while lower economic status in 1987/88 was modestly associated with being an Early Occasional or Persistent Occasional User. As expected, at-risk drinking and smoking were associated with illicit drug use. In these models, age at study entry was usually not associated with being in a particular trajectory group. Nor was attendance at a greater number of CARDIA’s in-person examinations.

Table 3.

Predictors of Membership in Drug Use Trajectories*. Adjusted for Demographics, Psychological and Social Contextual Factors

Drug Use Trajectories (N = 3432) In relation to “ Nonusers”
Predictor	Early Occasional User		Persistent Occasional User		Early Frequent/ Later Occasional User
	OR	(95% CI)	OR	(95% CI)	OR	(95% CI)
Demographics
Age greater than 25 in 1987/88	1.04	(0.91,1.18)	0.82	(0.68,0.99)	1.04	(0.83,1.32)
Sex/Race
White Female	REF	REF	REF	REF	REF	REF
Black Male	1.54	(1.24,1.92)	2.70	(2.00,3.65)	1.41	(0.94,2.11)
White Male	1.22	(0.99,1.51)	1.21	(0.86,1.71)	1.1	(0.73,1.65)
Black Female	0.65	(0.52,0.82)	0.67	(0.47,0.96)	0.87	(0.59,1.29)
Educational attainment (1987/88)
Did not complete high school	1.17	(0.85,1.60)	1.71	(1.21,2.41)	3.05	(2.09,4.46)
Graduated high school	REF	REF	REF	REF	REF	REF
Some Higher Education	0.64	(0.52,0.79)	0.48	(0.36,0.63)	0.42	(0.30,0.59)
Attended 5 CARDIA Examinations (since 1987/88)	0.94	(0.81,1.09)	1.17	(0.94,1.47)	1.05	(0.80,1.38)
Psychological
Self-reported mental disorder diagnosis (1987/88)	0.87	(0.66,1.13)	1.15	(0.84,1.58)	1.28	(0.88,1.84)
Social Context
Risky family upbringing (impact of 1-standard deviation increase)	1.10	(0.97,1.25)	1.44	(1.21,1.72)	1.10	(0.88,1.38)
Social Support Score (impact of 1-standard deviation increase), 1987/88	1.01	(0.89,1.15)	0.91	(0.75,1.09)	1.11	(0.88,1.40)
Difficulty Paying for Basic Necessities in 1987/88	1.23	(1.07,1.40)	1.33	(1.10,1.62)	1.04	(0.80,1.33)

*Interpretation of Odds Ratios: Each odds ratio expresses how the modeled predictor is associated with the odds of an individual being in a specified drug use trajectory (e.g. Early Occasional User) as opposed to the status of Nonuser. For example, the odds ratio of 1.71 reflects that person who had not completed high school had 71% increased odds of being an “Early Occasional User” compared to persons who had graduated high school and were otherwise similar with respect to variables included in the model

Table 4.

Predictors of Membership in Drug Use Trajectories. Adjusted for Demographics and Substance Use

Drug Use Trajectories (N = 4301) In relation to “Nonuser”
Predictor	Early Occasional User		Persistent Occasional User		Early Frequent/ Later Occasional User
	OR	(95% CI)	OR	(95% CI)	OR	(95% CI)
Demographics
Age greater than 25 in 1987/88	1.04	(0.92,1.16)	0.85	(0.72,1.01)	1.00	(0.82,1.23)
Sex/Race
White Female	REF	REF	REF	REF	REF	REF
Black Male	1.52	(1.25,1.84)	2.53	(1.95,3.27)	1.46	(1.05,2.03)
White Male	1.07	(0.88,1.31)	1.08	(0.79,1.46)	0.94	(0.66,1.34)
Black Female	0.76	(0.61,0.93)	0.73	(0.53,1.00)	0.94	(0.66,1.32)
Attended 5 CARDIA Examinations (since 1987/88)	1.05	(0.93,1.18)	1.16	(0.98,1.38)	0.96	(0.78,1.18)
Substance Use
Drinking above Weekly Threshold in 1987/88	1.69	(1.39,2.06)	1.84	(1.45,2.34)	2.42	(1.87,3.14)
Current (last 30 days) Smoking in 1987/88	1.7	(1.51,1.91)	2.19	(1.85,2.61)	1.94	(1.57,2.38)

*Interpretation of Odds Ratios: Each odds ratio expresses how the modeled predictor is associated with the odds of an individual being in a specified drug use trajectory (e.g. Early Occasional User) as opposed to the status of Nonuser. For example, the odds ratio of 1.70 indicates that current (1987/88) smokers had 70% increased odds of being “Early Occasional Users” compared to persons who were not current smokers at that time but were otherwise similar with respect to variables included in the model

All-Cause Mortality

By 12/31/2008, 4.6% of the analysis cohort had died, and the percentage who died was higher among the Persistent Occasional (8.1%) and Early Frequent/Later Occasional (6.4%) Users compared to Early Occasional (5.0%) and Nonusers (3.1%) (p = 0.003, df = 3). Survival curves differed (Log Rank Test, p = 0.003, df = 3). Compared to Nonusers, the adjusted hazard ratio for death was elevated for Early Frequent/Later Occasional Users (Hazard Ratio, HR = 4.94, 95% CI = 1.58-15.51, p = 0.006), and nonsignificantly elevated for Persistent Occasional Users (HR = 3.28, 95% CI = 0.95-11.31, p = 0.06) (Table 5).

Table 5.

Drug Use Trajectory and All-Cause Mortality at 18 Years’ Follow-up

	Drug Use Trajectories (N = 4301)
	Nonusers (n = 3691)	Early Occasional User (n = 340)	Persistent Occasional User (n = 160)	Early Frequent/ Later Occasional User (n = 110)	p
Mortality as of 12/31/2008					0.003
No. dead	122	17	13	7
%	3.3%	5%	8.1%	6.4%
Adjusted Hazard Ratio for Death (95% CI)	Referent	2.17 (0.87,5.44)	2.52 (0.76,8.35)	4.94 (1.58,15.51)	0.024

Mortality is for all causes as ascertained by 12/31/2008. Hazard ratios computed with Cox proportional models adjusting for age, sex, race, education, economic status, smoking, at-risk drinking, self-reported general health and depressive symptoms, body mass index, and original recruitment location (Birmingham, Minneapolis, Oakland, Chicago)

Effects Related to Age of Study Enrollment

Persistent Occasional and Early Frequent/Later Occasional Users were slightly younger than the other groups (Table 2). In adjusted models, persons older than 25 in 1987/88 had marginally lower odds of being Persistent Occasional Users, an association that reached significance in one of two models (Table 3, with OR = 0.82, 95% CI = 0.68-0.99, p = 0.02). On the whole, the trajectory groups do not appear dominated by subgroups entering CARDIA at different ages.

We found significant linear, quadratic and cubic interactions between age at study entry and age itself; these indicate that specific trajectory curves would differ slightly depending on age at cohort entry. The trajectory results (Fig. 1) show trajectories for persons of average age at time of cohort entry. When curves were iterated for persons of different ages at cohort entry, modest differences were apparent, but the descriptive categories (i.e. Nonusers, Early Frequent/Later Occasional, Persistent Occasional, Early Occasional) remained broadly applicable (Figure 2A, B, C, available online).

DISCUSSION

Using longitudinal data from a large four-community sample of initially healthy adults followed from 1987/88 to 2005/06, four trajectories of drug use were identified. The trajectories conformed to epidemiologic expectations regarding gender, alcohol and tobacco, family upbringing and education. Black men, but not black women, were at higher risk of being in a high drug use trajectory, as seen elsewhere.^4,31

The longitudinal measurement of drug use provided by the CARDIA sample, when subject to trajectory analysis, clarifies patterns that have not been easily appreciated from prior epidemiologic data. Longitudinal studies have shown drug use typically starts in late adolescence or young adulthood, and is less common thereafter, with frank abuse or dependence present only among 1.4% and 0.4% of adults, respectively, in the past 12 months.³² Among CARDIA participants, 6.3% had continuing use of opioids, cocaine, or amphetamines in middle age. These trajectory analyses show that such use originates typically in young adulthood. Several characteristics predict who will continue drug use, including demographic characteristics (i.e. black men), and social characteristics, including social support and whether the family of origin was characterized by abuse/neglect and parental substance use.

Limitations

This large four-city cohort, while broader than single-community samples,³³ was not nationally representative. However, there has been no trajectory analysis of these drugs in a national sample. The fact that our sample had slightly higher drug use prevalence than in contemporaneous US data suggests that CARDIA did not recruit a sample biased toward nonusers.³⁴ It is likely that persons with severe and persistent drug use disorders would be underrepresented in a cohort study like CARDIA. However, the mean number of self-reported days of cocaine among this sample’s Early Frequent/Later Occasional users (6.5 days) is similar to the mean number among persons with illicit drug abuse or dependence in the 2009 National Survey on Drug Use and Health (6.7 days).³⁵

We note that the studied drugs reflect those common when the CARDIA young adult cohort began, in 1985/86. Ongoing examinations within this cohort now specifically query methamphetamine, and distinguish use of pain medication for nonmedical reasons from other opioids.

There are statistical challenges to trajectory analysis that typically go unacknowledged. Trajectories are less precise at either end of the age spectrum, where observations are fewer. And, while the modeling approach considered relevance of birth year as a marker for cohort and/or period effects, we do not believe these effects are fully separable within our data. Others suggest conceptual barriers to fully disentangling these effects.³⁶ On the whole, the utility of statistical clustering should not be confused with a formal taxonomy of human biology or behavior and one statistical originator cautions against “reifying” the groups.³⁷

The positive association found between drug use and all-cause mortality is not likely one of direct cause and effect. We speculate that continuing adult drug use signifies complex patterns of vulnerability or behavior that contribute to mortality in ways not captured by our models. Nevertheless, the mortality association remained significant after adjusting for several plausible confounders, including tobacco and alcohol, general health status, body mass index, educational and economic status, and conditions of family upbringing. Moreover, CARDIA offers what is likely to be for some time the best descriptive data on this question from a large, diverse general population cohort. CARDIA is unique in providing repeated drug use measures across adulthood, health outcomes, and a range of biomedical and psychosocial measurements not available from other long-term studies in the general population.

Implications

The identification of drug trajectories assists in indicating who is likely to continue their use past young adulthood. Groups characterized by continuing drug use experienced premature mortality in ways not directly attributable to typical confounders. Some clinicians may prudently incorporate these findings into the education they offer to patients who use illicit drugs. However, definitive proof favoring screening and brief intervention for drug use awaits randomized controlled trials to assess whether the interventions reduce drug use and improve health outcomes at reasonable cost.³⁸ In the meantime, screening tools are available,^39,40 and clinicians may wish to use these results to help them target their screening where it is likely to be positive.