All-cause and cause-specific mortality among people with regular or problematic cocaine use: A systematic review and meta-analysis

Abstract

Aims:

To estimate pooled all-cause and cause-specific mortality risk for people with regular or problematic cocaine use.

Methods:

Systematic review and meta-analysis of prospective or retrospective cohort studies (n≥30) of people with regular or problematic cocaine use with data on all-cause or cause-specific mortality. Of 2808 papers, 28 were eligible and reported on 21 cohorts with a total 170,019 individuals. Cohorts identified based on acute care for drug poisoning or other severe health presentation were excluded. Title/abstract screening was conducted by one reviewer; a second reviewer independently checked 10% of excluded studies. Two reviewers conducted full-text screening. Data were extracted by one reviewer and checked by a second. A customised review-specific study reporting quality/risk of bias tool was used. Data on crude mortality rates (CMR) and standardised mortality ratios were extracted for both all-cause and cause-specific mortality. Standardised mortality ratios were imputed where not provided by the author using extracted data and information from the Global Burden of Disease Study 2017. Data were pooled using a random-effects model.

Results:

The pooled all-cause crude mortality rate was 1.24 per 100 person-years (95% CI: 0.86, 1.78; n=16 cohorts), but with considerable heterogeneity (I²=98.8%). The pooled all-cause standardised mortality ratio was 6.13 (95%CI: 4.15, 9.05; n=16 cohorts). Suicide (SMR 6.26, 95%CI 2.84, 9.68), accidental injury (SMR 6.36, 95%CI 4.18, 9.68), homicide (9.38, 95%CI 9.38, 3.45, 25.48), and AIDS-related mortality (SMR 23.12, 95%CI 11.30, 47.31) were all elevated compared with age and sex peers in the general population.

Conclusions:

There are elevated rates of mortality among people with regular or problematic cocaine use for traumatic deaths and deaths attributable to infectious disease.

Review registration:

PROSPERO CRD421018094623

INTRODUCTION

Cocaine manufacture is estimated to be at the highest level recorded and trafficking routes continue to expand globally (1). Greater availability has been accompanied by increased use, with an estimated 0.37% (CI 0.31–0.42) of adults aged 15–64 years reporting use globally in 2017, equivalent to 18.0 (CI 15.5–21.0) million people (1). Confluence of increasing availability and demand, coupled with indicators of increasing purity in some regions (1, 2), creates a higher risk environment for increased problematic use and health harms.

The most common forms of cocaine are hydrochloride salt (typically a fine white powder) and base (‘crack’ cocaine), the latter appearing crystal- or rock-like and typically being higher purity (3). Cocaine produces stimulant effects, increasing heart rate and blood pressure, enhancing alertness and producing feelings of euphoria, with a half-life of 30–60 minutes (4). Regular use of cocaine is associated with adverse health effects, predominantly cardiovascular (e.g., arrhythmia, myocardial infarction, stroke) and psychiatric (e.g., psychotic episodes, suicidal ideation) in nature (5, 6). There are additional and greater severity health risks depending on route of administration, including respiratory problems with smoking (6), nasal ulceration with snorting (7) and bloodborne virus transmission with injection, as well as via form used (e.g., higher risk of dependence with crack versus powder cocaine; 6).

A systematic review of studies published until 2008 suggested that people with regular cocaine use had a four to eight times higher mortality risk than their age and sex peers in the general population(8). Only seven cohorts were located, so a lack of data precluded quantifying risks of cause-specific mortality or examination of reasons for variation in mortality rates. In the intervening period, there has been significant research exploring specific harms associated with cocaine use from which cause-specific mortality estimates could be computed. Increased cocaine supply and demand globally and increasing poisoning deaths in a number of countries (e.g., United States; 9), including the highest rate of deaths in England and Wales observed in 2017 (10), makes it imperative that the magnitude of mortality risk by cause be quantified for this population. As such, the aims of this systematic review were to:

1. Update the previous systematic review and perform meta-analysis to calculate pooled estimates of all-cause crude mortality rates (CMRs) and standardised mortality ratios (SMRs) for people with regular or problematic cocaine use globally (regardless of treatment status); and

2. Compute pooled estimates of cause-specific CMRs and SMRs, with a focus on those causes of death that could be causally related to regular or problematic cocaine use.

METHODS

We report the systematic review methodology in accordance with the PRISMA guidelines (11) (Appendix A). This review protocol was registered with PROSPERO (CRD42018094623; Appendix B).

Search strategy and study screening

Medline, Embase and PsycINFO peer-reviewed literature databases were searched using the OVID™ interface/platform for articles published between 2009 and 22 February 2018. Relevant articles published between 1980 and 2008 were identified through a previous review conducted by the research team (8). In line with the previous review, search strings incorporating keywords and Medical Subject Headings (MeSH terms) related to cocaine/crack cocaine and mortality epidemiology were used to identify relevant articles (see Appendix C and D). Searches were limited to human literature. There were no restrictions on publication type or language; papers published in languages other than English were read via Google Translate or by a team member fluent in the language.

Citations were imported into an Endnote™ library where duplicate citations were removed, and imported into Covidence, a web-based screening tool (12). Titles and abstracts were reviewed by one team member (LTT or research assistants); 10% of excluded studies were checked by a second person (AP) to monitor accuracy. Full-text articles were reviewed by two reviewers (LTT and AP or LD); discrepancies were resolved by a third reviewer (AP or LD). Reference lists for relevant systematic reviews identified in the peer-review literature search were hand searched for additional papers not already identified.

Study eligibility

Studies were included if they were cohort or case-control studies or clinical trials (n≥30 people) where: i) at least 90% of the sample reported regular or problematic use of cocaine and ii) data on all-cause or cause-specific CMRs or SMRs were available (see Appendix E for detailed inclusion and exclusion criteria). This could include cohorts that were identified based on criteria other than cocaine use but reported mortality data for a sub-group of people reporting regular or problematic cocaine use. Cohorts were defined as comprising people with regular or problematic use if individuals reported cocaine as their primary drug used, cocaine injection, cocaine dependence, treatment for cocaine dependence or other healthcare presentation related to the effects of cocaine. Cohorts defined by any cocaine use (with no other indicators of cocaine-related problems) were excluded. Cohorts defined by cocaine overdose/poisoning or other serious adverse health presentations with high mortality risk were excluded. This included cohorts defined by cardiovascular presentations related to cocaine exposure or HIV positive status; estimates from these studies are available in Appendix F.

Data extraction

The data extraction worksheet was developed in Microsoft Excel based on the previous review. extraction was standardised through the use of a manual detailing data entry rules (8). Data were independently extracted by one reviewer (LTT) and checked by a second reviewer (TS).

Variables extracted included study information (e.g., country of sample, length of follow-up, recruitment setting) and sample information (e.g., age, sex, percentage of sample injecting, percentage engaged in treatment, form of cocaine used). With respect to the outcomes of interest (all-cause and cause-specific mortality), we extracted sample size, number of observed deaths (all-cause and cause-specific), person-years of observation, CMRs and SMRs. Data were extracted for specific causes of death which might be causally related to regular or problematic cocaine use through direct acute or chronic effects of the drug, or indirectly via other risk pathways (e.g., use of non-sterile injection equipment, high risk sexual behaviours). These causes of death comprised: drug-related, accidental injury, suicide, homicide, cardiovascular disease, respiratory disease, cancer, AIDS-related and digestive disease (the latter including liver diseases)(6). We also extracted International Classification of Diseases (ICD) codes or other information used to define cause-specific mortality. Where data for a study were incomplete, other published papers using the same cohort were sourced to locate missing data. If supplementary information could not be located or did not detail the data needed, authors were contacted by email for additional information.

Quality of reporting and risk of bias

There are no standard tools available for assessing risk of bias in descriptive studies reporting estimates of CMRs or SMRs. However, risk of bias tools for observational studies assessing the effects of exposures are under development and some domains of these tools are clearly relevant (13). We assessed risk of bias for the two domains that we considered most relevant for this type of study design: sample representativeness (i.e., the number of site locations and sample types) and outcome measurement (i.e., ascertainment of death by death registry/certificate versus indirect sources). Studies were rated as being at higher or lower risk of bias on these domains. The tool also measured three components of study reporting quality, comprising cohort description (i.e., age and/or sex data for the cohort), data completeness (i.e., numerator and denominator for outcome variable) and definition of cause-specific deaths (i.e., ICD codes or other information to define cause of death). Studies were assessed as having higher or lower risk or bias or quality reporting on each of these domains (see Appendix G for full details of tool).

Data analysis

Calculation of CMR, SMR, and relative risks

If not reported by study authors, CMRs were calculated as per 100 person-years (100PY). Where person-years were not reported nor made available by the authors, an imputed CMR follow up calculation was undertaken using other data reported by the authors. This calculation was based on the assumption that deaths occurred halfway through the follow-up period so that each person who died contributed half the person-year follow-up of survivors (see Appendix H for formula and example computation).

Standardised mortality ratios (SMRs) represent the CMR ratio between those exposed to the risk and the general population (including those exposed to the risk). If not reported by study authors, SMRs were calculated by dividing the sample CMR by the CMR for the for the respective age, sex, location, and years from the Global Burden of Disease (GBD) 2017 study (14). The GBD 2017 study used vital registration, verbal autopsy, registry, survey, policy, and surveillance data to model mortality estimates for 282 causes of death in 195 countries and territories. Due to the nature of GBD data classification, a match between the GBD database and study classification for respiratory and digestive disease mortality was unable to be obtained resulting in CMRs being unable to be converted into SMRs. Standard errors of log-transformed CMRs and SMRs were estimated using Rothman Greenland method (15).

Relative risks (RRs) comprise the ratio of mortality risk between those exposed to the risk (i.e., regular or problematic cocaine use) and those not exposed to the risk. With a low prevalence exposure such as regular and/or problematic cocaine use, SMRs and RRs should be similar, but RRs provide an estimate that is useful for further estimation of burden of disease. RRs were estimated from SMRs using the method described by Jones and Swerdlow (16) by adjusting the SMR by the proportion of the general population that experiences cocaine dependence (see Appendix H for the formulae).

Pooling all-cause and cause-specific mortality CMR and SMR

DerSimonian and Laird random-effects meta-analysis (17) was conducted in STATA version 14.2(18) to pool all-cause and cause-specific CMR and SMR estimates. This allows for heterogeneity between and within studies (noting that each study could only contribute one estimate to each meta-analysis). Heterogeneity was quantified using the I² statistic and described as low (≤30%), moderate (>30% and ≤50%), substantial (>50% and ≤90%) or considerable (>75% and 100%)(19).

Understanding variation in CMR and SMR

Sources of heterogeneity in all-cause CMR and SMR were investigated through univariate meta-regressions, including aspects of the study design (including the percentage of the cohort that was female, type of cocaine used, and geographical region) and the characteristics of the sample (including year of final follow-up, and recruitment setting). Variables were only included in meta-regressions if 5 or more data points were available per variable.

Sensitivity analyses

We undertook several addition analyses to examine potential impacts of our imputed metrics (CMRs, person years, SMRs) for studies that did not report all of these metrics upon the estimates generated in our meta-analyses. First, we generated estimated person years and CMRs for studies using our methods described above for studies that had reported all of these and compared the resulting estimates with the author-reported estimates; we found reasonable consistency for four out of seven studies (Appendix H). Second, we compared pooled estimated CMRs that included studies with imputed CMRs with the pooled estimates that only included author-reported CMRs. Finally, we contrasted the pooled estimated SMRs including studies where we used derived SMRs with pooled estimates that only used author-reported SMRs.

RESULTS

As evident in the PRISMA study flow diagram in Figure 1, the search identified 2,808 papers, of which 209 were screened in full. Of these 28 papers were deemed eligible, including 21 cohorts reported in primary publications and 7 secondary publications providing additional data for those cohorts reported in the primary publication (see Appendix I for excluded studies and Table 1 for details of included studies).

Figure 1.

PRISMA flow diagram of studies reporting on mortality among people with regular or problematic cocaine use

Table 1:

Characteristics of included studies

Study	Country	Years conducteda	Sample	N people	N person yearsb
1. Accurso, 2015(39)	United States	1990–2010	People “abusing” cocaine and presenting for detoxification at Chemical Dependence Unit at John Hopkins Bayview Medical Center in Baltimore between 1990–1991	315	5780*
2. Arendt, 2011(40)	Denmark	1996–2006	People receiving publicly funded treatment for illicit substance use disorder and primarily using cocaine across Denmark between 1996–2006, identified from the Danish Substance Abuse Treatment Register	838	2571*
3. Barrio, 2013(41)	Spain	2004–2010	People who reported regular cocaine use (≥ 52 days/last year) recruited from drug scenes and non-treatment settings in Madrid, Barcelona, and Seville, Spain, between 2004–2006	714	3922
4. Bohnert, 2017(23)	United States	2006–2011	People receiving Veteran Health Administration (VHA) care in the 2005 financial year diagnosed with cocaine use disorder and still alive in 2006 as identified using VHA National Patient Care Database	83808	468560*
5. Callaghan, 2013(22) (Callaghan, 2012(42))	United States	1990–2005	People hospitalised with a cocaine use disorder diagnosis in California between 1990–2005 from the Patient Discharge Database	48949	395738
6. de la Fuente, 2016(43) (Brugal, 2016(44); Colell, 2018(45); Molist, 2018(21))	Spain	1997–2008	People starting drug treatment for cocaine use disorder in a publicly owned or funded treatment centre in Barcelona and Madrid between 1997–2007	11905	65849*
7. Dias, 2011(46)c	Brazil	1992–2006	People who were consecutively admitted patients to Taipas General Hospital’s inpatient treatment for crack/cocaine dependence between 1992–1994	131	1182*
8. Gossop, 2002(47)c	United Kingdom	1995–1999	People who self-reported cocaine misuse recruited to treatment programs throughout England in 1995 as part of National Treatment Outcome Research Study (NTORS) cohort	227	926*
9. Hayashi, 2016(48) (Tyndall, 2001(49))	Canada	1996–2011	People who injected cocaine in the 6 months prior to enrolling in the open cohorts of Vancouver Injection Drug Users Study (VIDUS) and AIDS Care Cohort to Evaluate access to Survival Sciences (ACCESS) in Vancouver between 1996–2011	1719	11749
10. Hser, 2012(50)	United States	2002–2010	Mothers enrolled in a drug treatment program with cocaine as the primary drug of concern between 2000–2002 across California as identified through California Treatment Outcome Project (CalTOP)	511	5471*
11. Lopez, 2004 (OFDT)(51)c	France	1992–2001	People arrested in 1992, 1993, 1996 and 1997 for cocaine/crack use/dealing as identified through database of police questioning files for use of narcotics	2212	11496
12. Markota, 2016(52)	United States	1999–2011	People aged 13–18 years old attending drug and alcohol treatment and with a positive cocaine urinary toxicology screen administered at clinical sites within Mayo Health Care System between 1999–2011	63	308*
13. Martell, 2009^(53)	United States	2003–2005	People with cocaine and opioid dependence enrolled in a randomised clinical trial for cocaine vaccine from greater New Haven between 2003–2005	114	58*
14. Nielsen, 2011(54)	Denmark	1999–2009	People diagnosed with cocaine abuse and at least one contact with a homeless shelter in Denmark between 1999–2009 as identified by the Danish Homeless Register	525	5362*
15. O’Driscoll, 2001(55)	United States	1994–1997	People who inject drugs within Seattle and King County and reported cocaine as their primary drug recruited between 1994–1996	340	931*
16. Pavarin, 2017(56) (Pavarin, 2008(57); Pavarin, 2013(58))	Italy	1989–2013	Individuals admitted to a public drug treatment for problems caused by primary use of cocaine in Bologna (North Italy) between 1989–2013	678	4753*
17. Ryb, 2009(59)	United States	1983–1997	People discharged from R Adams Cowley Shock Trauma Center with a positive cocaine urinary toxicology screen at admission between 1983–1995	2451	15932*
18. Sanvisens, 2014(60)	Spain	1985–2008	Patients admitted to hospital detoxification for primary cocaine abuse at one of three tertiary care facilities in Barcelona and the surrounding metropolitan area between 1985–2006	945	7155
19. van Haastrecht, 1996(61)c	Netherlands	1985–1993	People who are HIV+ and HIV- who self-reported injecting cocaine and were recruited in Amsterdam between 1985–1992 through “low-threshold” methadone programs and clinic workers for people who use drugs and engage in sex work.	632d	194
20. Vlahov, 2008(62)	United States	1997–2002	People who inject drugs with self-reported injecting of cocaine everyday recruited from five U.S. cities between 1997–1999 through community-based outreach methods and enrolled in the second Collaborative Injection Drug Users Study (CIDUS-II)	102	486
21. Wang, 2005(63)c	United States	1988–2005	People who inject drugs who self-reported cocaine use in the previous 6 months, recruited through community in Baltimore between 1988–1989 and 1994–1998 and enrolled in AIDS Linked to Intravenous Experience study (ALIVE)	518#	3727

Note. Italics denotes associated secondary paper for the cohort;

^*Person years were not reported by study but calculated using formula within Appendix H;

^{^}Denotes that the study was a randomised controlled trial (RCT);

^#Denotes that the information was provided by the authors on request;

^aPeriod covers the start of recruitment until the end of follow-up;

^bPerson-years were rounded to nearest whole number, though exact person-years reported were used for analysis;

^cStudy was included in the previous review and information differs due to additional information being provided;

^dStudy does not specify proportion of cohort that uses cocaine, but provides number of deaths within those who use cocaine.

All cohorts (or subsample of cohorts) included here wholly comprised people who reported regular or problematic cocaine use (see Appendix J). Cohort size ranged from 63 to 83,808 individuals (total 170,019 individuals) and follow up of the cohort ranged between 58 to 468560 PY (total 1,012,147 PY). Cohorts were recruited from nine countries: eight of which were classified as high-income and one (Brazil) as middle-high income. Over two-fifths of the cohorts (43%, n=9) were recruited from the United States, three cohorts from Spain and two from Denmark. Relative to the previous review (9 cohorts)(8), 13 new cohorts were identified and three cohorts had additional published follow-up data. One study(20) from the original review was excluded as, on review for the current study, the population was not identified as reporting regular or problematic cocaine use.

Risk of bias and study reporting quality

Two-fifths (42.9%) of the cohorts were at risk of bias from poor representativeness; that is, the sample were recruited from a single site location or from a single sample type (e.g., people who were in treatment; see Table 2). Only three cohorts (14.9%) did not use an official death registry to identify deaths within the cohort, suggesting low risk of bias in outcome ascertainment.

Table 2:

Summary of risk of bias and quality of study reporting

	Study risk of bias		Study reporting quality
	Entire cohort		Entire cohorta			Cocaine sub-cohorta
	Representativeness	Outcome measurement	Cohort description (age/sex)	Completeness of data (numerator/denominator)	Cause of death definitionb	Cohort description (age/sex)	Completeness of data (numerator/denominator)
Accurso, 2015(39)	↑	↓	↑	↓	NA	↓	↓
Arendt, 2011(40)	↓	↓	↑	↑	NA	↓	↓
Barrio, 2013(41)	↓	↓	↑	↑	↓	↑	↑
Bohnert, 2017(23)	↑	↓	↑	↓	↑	↓	↓
Callaghan, 2013(22)	↓	↓	↑	↑	↑	↑	↑
Callaghan, 2012(42)	↓	↓	↑	↓	NA	↑	↓
de la Fuente, 2016(43)	↓	↓	↑	↑	NA	↑	↓
Brugal, 2016(44)	↓	↓	↑	↑	↑	↑	↑
Colell, 2018(45)	↑	↓	↑	↑	NA	↑	↑
Molist, 2018(21)	↓	↓	↑	↑	↑	↑	↑
Dias, 2011(46)	↑	↑	↑	↓	↓	↑	↓
Gossop, 2002(47)	↑	↓	↓	↓	↑	↓	↓
Hayashi, 2016(48)	↓	↓	↑	↑	↑	↓	↓
Tyndall, 2001(49)	↓	↓	↑	↓	↑	↓	↓
Hser, 2012(50)	↓	↓	↑	↓	↓	↓	↓
Lopez, 2004 (OFDT)(51)	↑	↓	↑	↓	↓	↓	↓
Markota, 2016(52)	↓	↑	↓	↓	↓	↓	↓
Martell, 2009^(53)	↑	↑	↑	↓	NA	↑	↓
Nielsen, 2011(54)	↓	↓	↑	↓	↑	↓	↓
O’Driscoll, 2001(55)	↓	↓	↑	↑	↓	↓	↑
Pavarin, 2017(56)	↑	↓	↑	↓	↓	↑	↓
Pavarin, 2008(57)	↑	↓	↑	↑	↑	↑	↑
Pavarin, 2013(58)	↑	↓	↑	↓	↓	↑	↓
Ryb, 2009(59)	↑	↓	↑	↓	↓	↑	↓
Sanvisens, 2014(60)	↑	↓	↑	↑	↑	↑	↑
van Haastrecht, 1996(61)	↓	↓	↓	↑	↓	↓	↑
Vlahov, 2008(62)	↓	↓	↑	↑	NA	↓	↓
Wang, 2005(63)	↓	↓	↑	↑	↓	↓	↓
Higher risk of bias/ Lower quality of reportingc	9/21 (42.9%)	3/21 (14.9%)	3/21 (14.9%)	11/21 (52.4%)	10/17 (58.8%)	13/21 (61.9%)	16/21 (76.2%)

Note. Full details of the risk of bias and quality of study reporting assessment are available in Appendix H, noting that studies with insufficient information to evaluate each criteria were deemed ‘higher risk of bias’/’lower quality of reporting’; ↓ denotes lower risk of bias or lower quality of reporting; ↑ denotes higher risk of bias or higher quality of reporting

^a‘Entire cohort’ refers to reporting for sample of interest in the original study which may include people who do not consume cocaine; ‘cocaine sub-cohort’ refers to reporting where mortality outcomes were reported for the subsample of interest for this paper (i.e., people with regular/problematic cocaine use) but whom may not be the primary focus of the paper.

^bIncludes 17 cohorts with cause-specific mortality data; ‘NA’ denotes cohorts where cause-specific mortality data were not collected;

^cTotals only include the primary publication if there are multiple publications reporting on the same cohort.

Outcomes for quality of reporting were mixed. Both age and sex characteristics of the cohort were reported for all except three cohorts (14.9%). However, half (52.4%) of the cohorts had incomplete mortality data, where the numerator or denominator for the main analyses of mortality were not reported. Of the 17 cohorts for whom cause-specific mortality was reported, 10 (58.8%) did not provide definition(s) of the cause of death (see Appendix J for ICD codes used).

It is important to note that quality of reporting outcomes was assessed for the total cohort. In many instances, people with regular or problematic cocaine use were a subsample of a larger cohort of people who use drugs. Age/sex characteristics and both numerator and denominator for mortality specifically for the sample who report regular or problematic cocaine use were thus only available for 8 (38%) and 5 (24%) cohorts, respectively (Table 2).

All-cause mortality

Data from 16 cohorts yielded an all-cause CMR of 1.24 per 100 PY (95% CI: 0.86, 1.78) with substantial heterogeneity (I²=98.8%) (Table 3; Figure 2). Meta-regression analyses to explore sample and study characteristics that could explain high heterogeneity showed that the percentage of the cohort reporting injecting drug use was positively associated with higher CMRs (Table 4; Appendix L). Similarly, recruitment of the cohort from a single city (versus subnational/national recruitment) was associated with higher CMRs. There was little or no statistical evidence of an association of CMR with other variables explored (Table 4).

Table 3:

Pooled all-cause and cause-specific CMRs and SMRs for people with regular/problematic cocaine use, stratified by sex, age, drug use characteristics and region

	Crude mortality rate					Standardised mortality ratio
	N cohorts	N people	Pooled crude mortality rate per 100PY (95%CI)	I2	References	N cohorts (N author-reported SMR)e	N people	Pooled standardised mortality ratio (95%CI)	I2	References
All-cause mortality
Overall	16	69,954a	1.24 (0.86–1.78)	98.8%	(22, 39–41, 43, 46, 48, 51, 52, 54–56, 60–63)	16 (6)	69,932a	6.13 (4.15–9.05)	99.0%	(39–43, 46, 48, 51, 52, 54–56, 60–63)
Sex
Women	6	25,217b	0.66 (0.54–0.81)	55.6%	(22, 40, 43, 52, 54, 58)	6 (3)	25,202b,c	4.59 (2.68–7.87)	94.2%	(40, 42, 43, 52, 54, 57)
Men	6	37,041b	0.89 (0.50–1.56)	98.3%	(22, 40, 43, 52, 54, 58)	6 (4)	37,056b	3.42 (2.86–4.10)	77.5%	(40, 42, 43, 52, 54, 57)
Age
< 30	2	3,677	0.67 (0.16–2.89)	97.1%	(21, 46)	2 (1)	3,677	7.75 (5.67–10.58)	0.0%	(21, 46)
≥ 30	3	8,358	1.01 (0.49–2.09)	81.5%	(21, 46, 57)	3 (1)	8,358	3.09 (1.77–5.37)	69.4%	(21, 46, 57)
Lifetime cocaine injection	5	2,679a	3.60 (3.32–3.90)	88.4%	(48, 55, 61–63)	5 (0)	2,679a	13.74 (12.67–14.90)	91.2%	(48, 55, 61–63)
Cocaine dependence/use disorder	8	64,286	1.09 (1.07–1.13)	98.4%	(22, 39, 40, 43, 46, 54, 56, 60)	8 (5)	64,286	3.24 (3.15–3.33)	97.6%	(22, 39, 40, 43, 46, 54, 56, 60)
GBD region*
High-Income North America	7	52,006	1.56 (0.83–2.95)	99.2%	(22, 39, 48, 52, 55, 62, 63)	7 (1)	51,984	5.13 (2.34–11.25)	99.5%	(39, 42, 48, 52, 55, 62, 63)
Western Europe	8	17,817a	0.93 (0.49–1.78)	98.6%	(40, 41, 43, 51, 54, 56, 60, 61)	8 (4)	17,817a	6.01 (4.16–8.68)	94.6%	(40, 41, 43, 51, 54, 56, 60, 61)
Tropical Latin America	1	131	2.28 (1.57–3.33)	-	(46)	1 (1)	131	14.75 (9.92–21.17)	-	(46)
Cause-specific mortality
Drug-relatedc	8	16,857	0.34 (0.10–1.15)	98.6%	(41, 44, 46, 47, 49, 55, 56, 63)	8 (2)	c	c	c	c
Suicide	8	100,854	0.07 (0.04–0.10)	72.6%	(21, 23, 41, 49, 52, 56, 59, 63)	8 (2)	100,854	6.26 (2.84–13.80)	94.4%	(21, 22, 46, 52, 56, 59)
Accidental injury	6	64,177	0.09 (0.04–0.22)	95.4%	(21, 22, 46, 52, 56, 59)	6 (3)	64,177	6.36 (4.18–9.68)	71.9%	(21, 42, 46, 52, 56, 59)
Cardiovascular disease	4	14,085	0.13 (0.07–0.24)	77.3%	(21, 49, 60, 63)	4 (1)	14,085	1.83 (0.39–8.57)	96.7%	(21, 49, 60, 63)
Homicide	3	14,487	0.09 (0.01–0.54)	98.4%	(21, 46, 59)	3 (1)	14,487	9.38 (3.45–25.48)	90.1%	(21, 46, 59)
AIDS-related	7	7,293	0.28 (0.12–0.63)	95.1%	(41, 45, 46, 49, 56, 60, 63)	6 (1)	6,576	23.12 (11.30–47.31)g	90.1%	(41, 45, 46, 56, 60, 63)
Cancer	5	14,763	0.11 (0.05–0.25)	87.1%	(21, 49, 56, 60, 63)	5 (2)	14,763	1.49 (0.70–3.16)	85.5%	(21, 49, 56, 60, 63)
Respiratory disease	5	6,217	0.09 (0.04–0.17)	63.2%	(41, 45, 49, 56, 63)	1 (1)f	852	24.12 (6.03–96.43)	-	(56)
Digestive diseases	3	13,140	0.14 (0.03–0.73)	96.1%	(21, 49, 63)	1 (1)f	11,905	1.90 (1.20–2.89)	-	(21)
Liver-relatedd	3	4,666	0.06 (0.01–0.55)	99.9%	(45, 46, 60)	3 (0)	4,666	3.36 (0.51–22.10)	92.5%	(45, 46, 60)

Note.

^*Regions are defined as per the Global Burden of Disease (GBD) project. No studies were found for the following GBD regions: Central Asia, Central Europe, Eastern Europe, Australasia, High-Income Asia Pacific, Southern Latin America, Andean Latin America, Caribbean, Central Latin America, North Africa & Middle East, South Asia, East Asia, Oceania, Southeast Asia, Central Sub-Saharan Africa, Eastern Sub-Saharan Africa, Southern Sub-Saharan Africa or Western Sub-Saharan Africa.

^aExcept for van Haastrecht,1996(61), all studies reported the N of people who reported regular/problematic cocaine use.

^bExcept for Arendt, 2011(40), all studies reported the N of people who reported regular/problematic cocaine use.

^cThough study estimates were available for SMR, these were deemed unstable and therefore not included. See Appendix L for the pooled estimate using author reported SMRs. It should be noted that drug poisoning deaths, for some studies, could include poisoning due to any drug and mental and behavioural disorders due to psychoactive substance use and other causes deemed by the authors to be drug-related (see Appendix J for ICD codes where reported).

^dThere is partial overlap with digestive diseases deaths, but liver-related deaths includes those that were specified identified as liver related whereas digestive diseases deaths included any deaths within a broader definition encompassing the digestive system (i.e., Chapter 10 of ICD-10 codes).

^eThe number in brackets denotes the number of cohorts where SMRs were reported by the authors, noting we imputed SMRs for those cohorts where these data were not reported (see Appendix L for pooled estimates using only author-reported SMRs.

^fDue to the nature of GBD classifications, expected number of deaths was unable to be estimated resulting in no imputed SMRs to be calculated.

^gTyndall, 2001(49) excluded as estimate was not logical.

Figure 2.

Pooled estimates of all-cause crude mortality rates (CMR) per 100 person-years (left) and standardised mortality ratio (SMR) (right) among people with regular or problematic cocaine use, overall and by region

Table 4:

Meta-regression of potential sources of heterogeneity in the pooled all-cause crude mortality rate (CMR) and standardised mortality ratio (SMR)

	Crude mortality rate				Standardised mortality ratio
	N studies	Coefficient (SE)	Adj. R2	P	N studies	Coefficient (SE)	Adj. R2	P
Sample characteristics at baseline
% Women	9	0.194 (0.543)	−9.76%	0.576	9	0.038 (0.051)	39.96%	0.045
% Injecting	8	5.871 (3.272)	68.62%	0.019	8	3.091 (2.472)	17.35%	0.208
Type of cocaine use			−1.54%				3.19%
Cocaine/Cocaine and crack cocaine	16	ref			16	ref
Speedball/Cocaine and Heroin	6	1.383 (0.546)		0.422	2	2.113 (1.400)		0.276
Geographic region			1.20%				−7.77%
% Western Europe	8	ref			8	ref
% High-Income North America	7	1.689 (0.755)		0.262	7	−0.207 (0.494)		0.682
% Tropical Latin America	1	2.447 (2.199)		0.337	1	0.844 (0.995)		0.412
Study characteristics
Year of final follow-up	16	0.939 (0.034)	12.54%	0.103	16	0.925 (0.035)	16.88%	0.056
Sample size	15	1.000 (< 0.001)	−7.82%	0.797	15	1.000 (< 0.001)	0.29%	0.345
Person years	16	1.014 (0.038)	−6.45%	0.713	16	0.973 (0.038)	−4.64%	0.502
Recruitment setting			−14.62%				20.51%
Treatment clinics and other health services	6	ref			6	ref
Hospital	3	0.750 (0.492)		0.664	3	0.247 (0.146)		0.034
Other	7	0.961 (0.489)		0.939	7	0.587 (0.268)		0.264
Sampling frame			34.92%				24.65%
National/sub-national	9	ref			9	ref
City	7	2.776 (0.985)		0.012	7	2.579 (1.042)		0.034

Note. Regions are defined as per the Global Burden of Disease (GBD) project. No studies were found for the following GBD regions: Central Asia, Central Europe, Eastern Europe, Australasia, High-Income Asia Pacific, Southern Latin America, Andean Latin America, Caribbean, Central Latin America, North Africa & Middle East, South Asia, East Asia, Oceania, Southeast Asia, Central Sub-Saharan Africa, Eastern Sub-Saharan Africa, Southern Sub-Saharan Africa or Western Sub-Saharan Africa. Ref: reference category; SE: standard error

Authors of studies for seven cohorts reported all-cause SMRs; we imputed all-cause SMRs for a further 9 cohorts. The pooled all-cause SMR across the 16 cohorts was 6.13 (95%CI: 4.15, 9.05), with substantial heterogeneity (I² = 99.0%) (Table 3, Figure 2). This estimate was similar to that observed from the seven cohorts where all-cause SMRs were reported by the authors (5.58, 95%CI: 3,90, 7.99; I² = 96.6%) (Appendix L). Excess mortality was particularly elevated among cohorts reporting lifetime injecting (13.74, 95% CI: 12.67, 14.90; I²=91.2%; n=5 cohorts).

The percentage of the cohort that was female was negatively associated with excess mortality (Table 4; Appendix L). Study characteristics associated with excess mortality comprised recruitment setting (with higher SMRs for cohorts recruited from hospital versus treatment services and other settings) and sampling frame (with higher SMRs for cohorts recruited from a single city versus subnational/national recruitment).

Drug-related deaths

Eight cohorts reported data on drug-related deaths. The definition of drug-related deaths was not provided for six cohorts; for the remaining two cohorts, ‘drug-related deaths’ comprised poisoning deaths, deaths attributed to mental and behavioural disorders due to psychoactive substance use, and deaths from other causes that were thought to be drug-related by the authors following consultation with forensic and toxicological services (e.g., ICD-10 code J81 ‘pulmonary oedema’)(21). The pooled drug-related CMR was 0.34 per 100PY (95%CI: 0.10, 1.15), again with considerable heterogeneity (I² = 98.6%) (Table 3, Figure 3). SMRs were only reported for two cohorts, with a very high pooled estimate observed (44.37, 95% CI: 37.28, 52.81; I²=63.7%) (Appendix L). We have not reported the pooled estimate for all cohorts - including those for whom we imputed SMR as the estimate was deemed unstable.

Figure 3:

Pooled estimates of cause-specific crude mortality rates (CMR) per 100 person-years (left) and standardised mortality ratio (SMR) (right) among people with regular or problematic cocaine use

Traumatic deaths: suicide, accidental injury, and homicide

The pooled suicide CMR was 0.07 per 100PY (95%CI: 0.04, 0.10; I²=72.6%; n=8 cohorts), with both pooled accidental injury (n=6 cohorts) and homicide (n=3 cohorts) CMR estimated at 0.09 per 100PY (95%CI: 0.04, 0.22; I²=95.4% and 95%CI: 0.01, 1.54; I²=98.4% respectively) (Table 3, Figure 3). The pooled SMR based on author-reported and imputed estimates was 6.26 (95% CI: 2.84, 13.80; I²=94.4%) for suicide, 6.36 (95% CI: 4.18, 9.68; I²=71.9%) for accidental injury and 9.38 (95% CI: 3.45, 25.48; I²=90.1%) for homicide (Table 3, Figure 4). Pooled estimates of the subsample of author-reported SMRs were higher for suicide and lower for accidental injury and homicide (Appendix L) but fell within the confidence intervals of the former estimates. There was substantial heterogeneity in these pooled CMR and SMR estimates.

Figure 4:

Pooled estimates of cause-specific standardised mortality ratio (SMR) (right) among people with regular or problematic cocaine use

* Imputed SMR was calculated for the study

AIDS-related deaths

The pooled CMR for AIDS-related deaths based on 6 cohorts was 0.28 per 100PY (0.12–0.63; I² = 95.1%) (Table 3, Figure 3). Only one study reported a SMR for AIDS-related deaths, with an excess mortality rate five times that expected in the general population (4.98, 95% CI 0.70, 35.34; Appendix L). The pooled SMR based on author-reported and imputed estimates was higher (23.12, 95%CI 11.30–47.31; I² = 90.1%; n=6 cohorts) (Table 3, Figure 4) but fell within the confidence interval for the former estimate.

Digestive disease deaths

The pooled CMR for digestive diseases based on three cohorts was 0.14 per 100PY (95%CI: 0.03, 0.73; I² = 96.1%). The SMR was only reported by authors for one cohort (1.90, 95%CI 1.20, 2.89). Three different cohorts provided CMRs and SMRs for liver-specific deaths, with these deaths occurring more than three times the expected rate (pooled SMR: 3.36, 95%CI 0.51, 22.10; I² = 92.5%; n=3 cohorts) (Table 3, Figure 4).

Other causes of death

Pooled CMRs were derived from four cohorts for cardiovascular disease (0.13 per 100PY; 95%CI: 0.07, 0.24; I² = 77.3%), five cohorts for cancer (0.11, 95% CI: 0.50–0.25; I² = 87.1%), and five cohorts for respiratory diseases (0.09, 95%CI 0.04, 0.17; I² = 63.2%) (Table 3; Figure 3). These deaths were elevated relative to the expected rate, particularly respiratory diseases (pooled SMR: 24.12, 95% CI 6.03, 96.43) (Table 3, Figure 4), although this estimate should be treated with caution being based on one cohort.

Mortality relative risks

Estimated study-level mortality RRs for all-cause and cause-specific mortality are presented in Appendix K and are similar to SMRs.

We conducted several additional analyses to examine our approach to imputation of data. First, we generated estimated CMRs and person years for studies that had already reported all these metrics and found reasonable consistency across these (Appendix H). Second, we compared pooled estimates that did not include imputed CMRs with those that did, and again found reasonable consistency (Appendix L). Finally, as noted earlier, we also contrasted pooled SMRs that only included author-reported SMRs with those that included our imputed SMRs (Appendix L); in some cases there were very few or only one cohort that had author-reported SMRs, so there are few data, but the all-cause estimates were remarkably similar. Finally, we examined the potential impact of the two very large cohorts on pooled estimates by generating pooled estimates without those two cohorts (22, 23). Pooled all-cause CMR was 1.23 per 100PY (95%CI 0.75, 2.02; I²=98.7%; n=15 cohorts) compared to 1.24 per 100PY (95% CI: 0.86, 1.78; I²=98.8%; n=16 cohorts) including those cohorts; pooled accidental injury CMR was 0.13 per 100PY (95%CI 0.06, 0.25; I² =82.5%; n=5 cohorts) compared to 0.09 per 100PY (95%CI: 0.04, 0.22; I²=95.4%; n=6 cohorts) including those cohorts; and pooled suicide CMR was 0.08 per 100PY (95%CI 0.04, 0.15; I²=73.0%’ n=7 cohorts) compared to 0.07 per 100PY (95%CI: 0.04, 0.10; I²=72.6%; n=8 cohorts) including those cohorts.

DISCUSSION

This review suggests that people with regular or problematic cocaine use have, on average, an excess mortality risk six times (95%CI four to nine times) the expected rate for their age-matched counterparts in the general population. The most highly elevated causes of deaths were drug-related or arising from traumatic causes (suicide, accidental injury or homicide) – all of which are preventable. Mortality was also elevated for communicable diseases (e.g., AIDS-related mortality) and other natural causes of death, including cardiovascular disease. These findings are consistent with the known effects of cocaine (6) and other risk pathways associated with these health outcomes (6). The potential for a growing population of people with regular or problematic cocaine use – coupled with the current findings of elevated mortality risk – reinforces the need for expansion of evidence-based prevention and intervention efforts to reduce health harms.

It is important to note that it is not the case that elevations in mortality necessarily reflect direct causal effects of cocaine use. In some instances elevated mortality may reflect other lifestyle factors and exposure to risk environments. But there are some causes of death for which there is good evidence of a direct causal impact. Cocaine carries a clear risk for cardiovascular events (e.g., heart attack, arrhythmia or stroke) (24). Risk of mortality can only be mitigated through reduced use. Yet, a major hurdle in reducing mortality associated with cocaine use is the lack of treatment options for cocaine dependence. There is no strong clinical evidence to support pharmacotherapies for cocaine dependence despite trials of various medicines(25–30). Psychosocial treatment (primarily contingency management) may decrease frequency of use and increase length of abstinence; longer-term impacts post-treatment are less clear (31). However, accessibility and affordability of psychosocial treatment are major issues in many countries, particularly for consumers of crack cocaine who are typically more socially marginalised relative to consumers of powder cocaine (32).

There are a range of secondary interventions that can address mortality risk pathways. Prevalence of HIV, HCV and other infectious diseases is often higher among people who use cocaine (and crack cocaine specifically) relative to the general population (33), driving AIDS-related mortality and likely contributing to excess mortality from liver disease. Regular HIV and HCV testing coupled with access to HIV antiretroviral therapies and HCV direct-acting antiviral agents could significantly reduce HIV and HCV related mortality. Access to these services is suboptimal in almost all countries (34). Strategies to prevent transmission include provision of sterile needles and smoking pipes, free condoms, and pre-exposure prophylaxis for HIV and sexually transmitted infections (25). Rigorous evaluation of some of these interventions is yet to be undertaken (particularly with respect to their impacts on mortality) however efforts to directly engage people who smoke crack cocaine in particular are laudable as existing services targeted at people who inject drugs may not meet the needs of this group.

Whilst preventable, reducing excess mortality due to traumatic causes (e.g., suicide, homicide, and accidental injury) is challenging. People with regular or problematic cocaine use are often from socio-economically disadvantaged areas and are more likely to be exposed to environmental risk factors (e.g., violence, crime) than people who use other drugs (35). There is also a significant gap in evidence regarding interventions to reduce agitation related to stimulant intoxication and to manage violence risk more broadly amongst this group (36). Treatment for cocaine dependence thus needs to considerate of possible means for reducing the risk of injury and violent behaviour against a backdrop of broader environmental risk factors. CBT can also reduce suicide risk in substance-using populations (37).

Strengths and limitations

There has been a significant growth in the literature on mortality among people who report regular or problematic cocaine use, with 13 new cohorts identified since the previous review in 2008(8). Cohorts were recruited from regions with the highest levels of cocaine use (i.e., North America, Western Europe and Latin America; 1), with Australasia and Central Europe being notable exceptions. There are limited data on cocaine use in many of the remaining regions (e.g., Asian and African regions; 1), highlighting the challenge of capturing health harms associated with regular or problematic cocaine use in these regions.

Limitations of existing studies included uncertain representativeness of samples and poor reporting of methodological detail. Many of the cohorts included here were derived from a single sample type: typically, those individuals engaged with health services. Linkage of administrative data from healthcare services could be used in future studies. Secondly, data were often missing on how causes of death were defined and, where available, varied between studies. Although there are likely many contributors to heterogeneity in cause-specific mortality estimates, use of standardised definitions of cause-specific mortality (38) would reduce this potential source of variability. Nonetheless, it is important to acknowledge that varying definitions of causes likely contributed to heterogeneity across studies. Ideally, an approach where people used standardised definitions of causes of mortality might occur in future studies (e.g., 38), permitting some examination of whether this explains some of the variation observed.

We were thorough in searching for relevant studies, however, we may have missed eligible cohorts or made errors. Where necessary, we sought additional data from authors and we generated estimates of CMRs, SMRs and RRs where possible. There are potential biases introduced by our computation of person years where this was not reported, including impaired capacity to account for censoring, however sensitivity analyses showed relative consistency between imputed and study-reported person-years when we tested our approach on those studies which did provide the latter information. Computing SMRs and RRs was achieved by using estimates from the GDB study, the most comprehensive effort globally to estimate prevalence of cocaine dependence and of population-level cause-specific mortality (14). Pooled SMRs including estimates imputed from GBD typically fell within the confidence interval of the pooled SMRs from author-reported estimates only. We acknowledge that the GBD data may have some limitations and there was minor variability between pooled estimates of author-reported SMRs versus author-reported and imputed SMRs. Reporting of all-cause and cause-specific SMRs in future work would improve confidence in existing estimates. Further, RRs should be treated with caution, being based on prevalence of cocaine dependence extracted from the GDB study, and likely of lower prevalence than regular and/or problematic cocaine use.

Finally, some variability in cohort definition and study design should also be noted. This was explored via meta-regression and stratified meta-analyses; although cohort definition (i.e. meeting criteria for dependence vs not) was not associated with heterogeneity in CMR/SMR, other aspects of study design (e.g., recruitment setting, sampling frame) were associated with heterogeneity in estimates.

Conclusions

There has been increased study of cohorts of people reporting regular or problematic cocaine use. For this reason, we could quantify excess mortality by cause of death in the current review. Synthesis of this evidence suggests people with regular or problematic cocaine use have, on average, six times (95%CI four to nine times) higher mortality risk than their age and sex peers in the general population. Excess mortality risk is particularly evident for traumatic causes of deaths and causes likely attributable to infectious diseases. These deaths are mostly preventable. A lack of treatment options for cocaine dependence mean current efforts rely heavily on other prevention and intervention strategies to address risk pathways to mortality.