Mortality among people with regular or problematic use of amphetamines: a systematic review and meta-analysis

Abstract

Background and aims:

Amphetamines are the second most commonly used class of illicit drugs. We aimed to produce pooled estimates of mortality risks among people with regular or dependent use of amphetamines, with a focus upon all-cause mortality as well as specific causes of death.

Design:

Systematic review and meta-analysis of cohorts of people with problematic use or dependence on amphetamines with data on all-cause or cause-specific mortality.

Setting and participants:

Of 4,240 papers, 30 were eligible, reporting on 25 cohorts that measured all-cause mortality, drug poisoning, suicide, accidental injuries, homicide and cardiovascular mortality. Cohorts (n=35-74,139) were in North America, several Nordic countries, and Asia-Pacific.

Measurement:

Titles/abstracts were independently screened by one reviewer and excluded ones reviewed by a second. Full-text screening was by two reviewers with discrepancies resolved via a third reviewer. We extracted data on crude mortality rates (CMR) per 100-person-years (PY), standardised mortality ratios (SMRs). We imputed SMRs where possible if not reported by study authors. We also calculated mortality relative risks. Data were pooled using random-effects models; potential reasons for heterogeneity were explored using subgroup analyses and meta-regressions.

Findings:

Twenty-three cohorts contributed data for the pooled all-cause CMR: 1.11 per 100PY (95%CI 0.90-1.37). Pooled cause-specific mortality rates were: drug poisoning, 0.14 per 100PY (95%CI: 0.06-0.34); cardiovascular disease, 0.13 per 100PY (95%CI: 0.06-0.55); suicide, 0.20 per 100PY (95%CI: 0.07-0.55); accidental injury, 0.20 per 100PY (95%CI: 0.08-0.47) and homicide, 0.03 per 100PY (95%CI: 0.02-0.06). There was substantial heterogeneity for all pooled CMR estimates except homicide. The pooled all-cause SMR was 6.83 (95%CI: 5.27-8.84). Pooled cause-specific SMRS were: poisoning, 24.70 (95%CI: 16.67, 36.58); homicide, 11.90 (95%CI: 7.82-18.12); suicide, 12.20 (95%CI: 4.89-30.47); cardiovascular disease, 5.12 (95%CI: 3.74-7.00) and accidental injury, 5.12 (95%CI: 2.88-9.08).

Conclusions:

People with regular or dependent amphetamine use are at elevated risk of a range of causes of mortality compared with people without regular or dependent amphetamine use.

Registration:

PROSPERO CRD42018094623

Introduction

Amphetamines are the second most commonly used group of illicit drugs, and amphetamine use and dependence are estimated to have increased over the past two decades¹. In 2016, the United Nations Office on Drugs and Crime estimated 34.2 million people aged 15-64 years used amphetamines (range 13.4-55.2 million).² The Global Burden of Disease (GBD) study estimated the age-standardised rate of amphetamine use disorder in 2017 was 96 per 100,000 population (95%Uncertainty Interval [UI] 70-128; 7.4 million people (95%UI 5.4-9.8 million))³. The United States and Australia are estimated to have among the highest rates of amphetamine use disorders in the world³.

People who use amphetamines may seek varied effects that include euphoria, feelings of increased energy, improved concentration and alertness, and self-confidence. However, a range of adverse physical and psychological consequences may also be experienced, including heart palpitations, sweating, headaches, tremors, paranoia, aggression and symptoms of psychosis^4–6. There are also potentially fatal harms, particularly among people who use amphetamines regularly, those who inject amphetamines, and who are dependent. These include risks of fatal drug toxicity, as well as risk of injuries (both unintentional and intentional, including exposure to violence). Suicide has been noted as elevated among people with problematic amphetamine use^7,8. Cardiovascular toxicity can occur, and includes arrhythmias, acute myocardial infarction and cardiomyopathy^9,10. Cerebrovascular pathology includes stroke, aneurysm, and cerebral haemorrhage¹¹.

An earlier review identified eight prospective or retrospective cohort studies published 1980-2007 on all-cause mortality¹². To our knowledge there has not been a systematic review and quantitative synthesis of the extent to which specific causes of death, including those mentioned above, occur among people who use amphetamines. Hence, although there is concern about these risks, there currently exists no comprehensive quantitative synthesis of the magnitude of risk, and how elevated that risk might be relative to the general population, among people with regular or problematic amphetamine use.

As such, the primary aim of this systematic review was to produce updated pooled estimates of all-cause mortality risk, and new estimates of cause-specific mortality risk, the latter focused on those conditions w4hich might be casually related to amphetamine use. We sought to produce gender and age-specific estimates of these measures where possible. We also aimed to study potential reasons for variation in mortality across cohorts, including methodological aspects as well as features of study populations.

Methods

We undertook this review using methods consistent with PRISMA and GATHER guidelines (see checklists in Appendix). This review was also registered with PROSPERO (CRD42018094623).

Search strategy and study screening

We used the OVID^™ platform to search the electronic databases Medline, Embase and PsycINFO (see summary of these databases in Appendix A) for articles published from 01 January 2008 to 22 February 2018 containing data describing all-cause and cause-specific crude mortality rates (CMR) and/or standardised mortality ratios (SMR) among people who use amphetamines.

Sets of search strings incorporating both keywords and Medical Subject Headings (MeSH terms) reflecting drug type and mortality epidemiology were used and are provided in full in Appendix B. A previous review conducted by the research team was used to identify studies published from 1980 to 2007.¹² Searches were limited to human literature, with no restrictions placed on language or publication type. Citations for papers in languages other than English were read via Google Translate.

Citations were imported into an Endnote^™ library, where duplicate citations were removed, and imported into the web-based screening tool, Covidence.¹³ Titles and abstracts were screened by one reviewer (TT) and all papers marked as excluded were reviewed by a second person (ES) to ensure accuracy in first-pass screening. Full text articles were screened by two independent reviewers (ES and LD) with discrepancies resolved via consultation with a third reviewer (AP or SL) when needed.

Reference lists for relevant systematic reviews identified in the peer-review literature search were hand searched for additional papers not already identified, and we additionally searched table of contents of relevant journals. A final list of included studies was distributed to experts to check if any relevant studies were missed.

Inclusion and exclusion criteria

The population of interest was people with problematic use of or dependence on amphetamines. The exact definition of this varied across cohorts and we summarise the way in which the cohorts were defined and amphetamine use operationalised in Table 1. Inclusion criteria for the study population comprised:

• Cohort studies of people using amphetamine (where ≥90% of the sample reported use of the substance of interest; people may have reported using multiple drugs);

• Cohort studies that were based on a sub-group of amphetamine users e.g. people who use amphetamines who are incarcerated.

• Cohort studies that were not based on drug use (e.g. a cohort of homeless people), but an amphetamine-using sub-group was identified, and results were presented separately for this sub-group;

• Case-control studies where cases were defined by amphetamine use and mortality was reported for cases and controls separately;

• Clinical trials of interventions to treat amphetamine use disorder (randomised controlled trials; non-randomised trials) where mortality may be reported at follow-up. Interventions may include pharmacotherapies or non-pharmacological treatments;

• Secondary publications of an included cohort where the outcomes were not mortality (these were used to assess methodology or to extract other baseline/demographic data not reported in the primary publication).

Table 1.

Characteristics of included studies

Study (Author, ref)	Country	Years conducted	Quality score (max 16)	Sample	N	Person years FU	% women	Age at baseline	% PWID
1. Ahman, 201834	Sweden	1987-2013	11	People who self-report injecting amphetamine recruited from the Malmo Needle Exchange Program between 1987-2011	2019	27698	23	33	100
2. Arendt, 201135	Denmark	1996-2006	14	Persons receiving treatment in all specialised institutions for illicit substance use disorders between 1996-2006 in Denmark	1553	6111.11*
3. Bartu, 200417	Australia	1985-1998	13	People who use amphetamine admitted to Perth metropolitan hospitals or psychiatric institutions between 1985-1998	4280	19913
4. Bohnert, 201744	USA	2006-2011	13	People accessing the US Veterans Health Administration healthcare system with an amphetamine or other psychostimulant use disorder who received services in financial year 2005	12033	67368.42*	5.9
5. Callaghan, 201336	USA	1990-2005	13	Inpatients discharged from a California licensed hospital with a substance-related diagnosis identified via Patient Discharge Database between 1990-2005	74170	455598	55	35.2 (mean)
6. Chen, 201037	Taiwan	1998-2005	11	People with drug offences who served at least 1 day in a correctional facility and involved in Schedule II drugs in Taiwan	21149	62551.8
7. Davstad, 201148	Sweden	1969-2004	11	Swedish male conscripts born in 1949-1951 and conscripted between 1969-1970 with self-reported experience with stimulants	221	7153.99	0
8. Ericsson, 201419	Sweden	2000-2008	12	People reporting primary amphetamine use interviewed with the addiction severity index (ASI) in the Swedish criminal justice system during 2000-2006	1396	5978	15	37.4 (mean)
9. Forsyth, 201820	Australia	2008-2013	8	Adults administered an interview prior to prison release identified as high-risk methamphetamine use before prison entry, recruited between 2008-2010	162	534.6*
10. Fridell, 200621	Sweden	1988-2004	7	People who use drugs consecutively admitted to psychiatric detoxification and short-term rehabilitation unit at Sankt Lars Hospital in Lund, Sweden between 1988-1989	48	592.11*
11. Fugelstad, 201438	Sweden	1985-2007	7	Patients admitted to drug addiction department at Sabbatsberg hospital in Stockholm between 1981-1988	578	12140	30.8
12. Hayashi, 201622	Canada	1996-2011	7	People who inject crystal methamphetamine identified via the Vancouver Injection Drug Users Study (VIDUS) and AIDS Care Cohort to Evaluate Exposure to Survival Services (ACCESS) open cohorts in Vancouver recruited between 1996-2011	205	730.04
13. Herbeck, 201523	USA	1999-2012	6	People who use methamphetamine who received or did not receive substance use disorder treatment recruited in 1999-2001	563	5123.3*	39.8		43.2
14. Jones, 201539	Canada	2008-2014	11	Socially disadvantaged residents in the Downtown east-side neighbourhood recruited from 2008-2012	298	1035.214*
15. Karjalainen, 201040	Finland	1993-2006	10	All apprehended individuals by police on suspicion of driving under the influence of amphetamines between 1993-2006	658	2646*	17.9	29.86 (mean)
16. Kittirattanapaiboon, 201041	Thailand	2000-2007	8	Patients with methamphetamine psychosis initially hospitalised in Suan Prung psychiatric hospital in 2000-2001	1116	3224*	9.	33.3 (mean)
17. Kuo, 201142	Taiwan	1990-2007	10	People with methamphetamine dependence admitted to Taipei City Psychiatric Center for detoxification between 1990-2007	1254	10032.8	19.5	28.4 (mean)
18. Lejckova, 200743	Czech Republic	1997-2002	16	Patients hospitalised for amphetamine-related disorders between 1997-2002 identified using data from hospitalisation register	3039	9748.4	36.72%
19. Nyhlen, 201124	Sweden	1970-2006	9	Inpatients admitted to the detoxification and short-term rehabilitation unit at St. Lars psychiatric hospital in Lund, Scania County between 1970-1978	236	6395.6*
20. Onyeka, 201625	Finland	1997-2008	8	Consecutive clients seeking treatment for drug use at Helsinki Deaconess Institute between 1997-2008	1334	11472.4*
21. Quan, 200726	Thailand	1999-2002	9	Patients admitted for detoxification of amphetamine dependence at Northern Drug Treatment Center in Chiang Mai province, Thailand	320	954			16.3
22. Turpeinen, 200127	Finland	1971-1992	8	Schoolchildren living in Helsinki who were interrogated by narcotics police during 1971-1972 and reported intravenous use of amphetamine	35	735*	51.4		100
23. van Haastrecht, 199628	Netherlands	1985-1993	9	Individuals recruited through “low threshold” methadone programs and sexually transmitted disease clinic for sex workers during 1985-1992	632	672	39.6		77
24. Vlahov, 200829	United States	1997-1999	7	People who inject drugs enrolled in the second Collaborative Injection Drug Users Study (CIDUS-II) in 5 U.S. cities between 1997-1999	584	2669			100
25. Weber, 201530	Switzerland	2007-2013	8	Swiss HIV Cohort Study (SHCS) participants who reported any amphetamine use and registered prior to 2007	261	31215			0

^*Denotes where PYFU was calculated based on available data (see Appendix F for formulae used).

% PWID – percentage of the sample who were noted as people who inject drugs.

Inclusion criteria for the study outcome comprised:

• Reporting mortality data (all-cause or cause-specific) or studies where such data could be obtained via contacting study authors.

Exclusion criteria comprised:

• Case reports/case series of drug-related deaths;

• Works that did not present original data (e.g. letters to the editor, editorials, commentaries);

• Works that did not present mortality data and where such data could not be obtained from study authors;

• Systematic reviews (we hand searched reviews for relevant studies, which were assessed independently against the inclusion/exclusion criteria);

• Studies where use of amphetamines and amphetamine-related mortality were not reported for the same sample of people;

• Studies where it was unclear that everyone in the cohort reported intentional use of amphetamines;

• Clinical trials of interventions with people using amphetamines for non-substance use disorders e.g. a clinical trial of HCV medicines;

• People prescribed amphetamines, where they were recruited at the point of which they started being prescribed (e.g. for Attention-Deficit Hyperactivity Disorder or Parkinson’s Disease);

• People prescribed stimulants using them as prescribed (i.e., no extra-medical use);

• Cohort studies consisting of 30 or fewer participants.

Data extraction

The data extraction form was developed in Microsoft Excel 2016 based on the previously conducted review,¹² and recommendations outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)¹⁴ statement. Data were independently extracted by one member of the research team (TT) and checked by a second (TS). Bibliographic information was extracted in addition to study specific information. Data entry was standardised by use of a manual, which contained data entry rules.

Variables extracted included study information and sample information (drug treatment status, HIV status, gender, percentage of sample injecting). Crude mortality rates (CMRs) and standardised mortality ratios (SMRs) were extracted as mortality measures.

Cause of death information was extracted and where possible, deaths were grouped to be consistent with guidelines developed by expert groups (summarised in¹⁵). We focused on specific causes of mortality, namely drug overdose/poisoning, accidental injuries, suicide, homicide and cardiovascular disease. Small numbers of studies reported on several other causes of mortality; these are summarised in Appendix I for the interested reader.

In instances where data reporting in the publications was incomplete, supplementary information and documents were sourced to locate missing data. If supplementary information could not be located or did not provide the necessary data needed, study authors were contacted by email for additional information.

A quality index used in the previous review¹² was adopted for consistency. The quality index assesses each study on nine individual criteria, each with a rating scale where higher scores indicate higher quality: case ascertainment [0-3], measurement [0-2], diagnosis [0-2], estimate [0-1], numerator and denominator [0-1], data catchment [0-1], completeness [0-2], representativeness [0-2] and age/sex variables [0-2] (see Appendix D). Individual scores for each criterion were tallied to provide an overall quality score, where the greater the quality score, the higher the methodological quality of the study (maximum score: 16). Study information necessary for quality assessment was extracted to the Excel template by one reviewer (LT) and double checked by a second (TS). Discrepancies were resolved via consultation with a third reviewer (ES, AP or SL).

Calculation of crude mortality rates, standardised mortality ratios and risk ratios

If not reported by study authors, crude mortality rates (CMRs) were calculated as per 100 person-years (100PY). Where person-years were not reported nor made available by the authors, approximate person-years of follow-up were calculated, with the assumption that deaths occurred halfway through the follow-up period, so that each case contributed half the person-year of follow-up of survivors.

If not reported by study authors, standardised mortality ratios (SMRs) were calculated by dividing the sample CMR by the CMR for the respective age, sex, location, and years from the GBD 2017 study. The GBD 2017 incorporates vital registration, verbal autopsy, registry, survey, policy, and surveillance data to model mortality estimates for 282 causes of death in 195 countries and territories¹⁶.

SMRs represent the CMR ratio between those exposed to the risk and the entire general population (including those exposed to the risk). This process was conducted for 13 cohorts^17–30.

Relative risks (RRs) illustrate the CMR ratio between those exposed to the risk versus those not exposed to the risk. RRs were estimated from SMRs using the method described by Jones and Swerdlow³¹ that adjusts the SMR by the proportion of the general population that is exposed to the risk. See Appendix E for a full overview of the formulae and calculations used in the review.

Data analysis

STATA version 15.1 metan and random commands were used to conduct DerSimonian and Laird Mantel-Haenszel random-effects meta-analysis to pool all-cause and cause-specific CMR and SMR estimates. The random effects model allows for heterogeneity between and within studies; the random-effects model was used in all meta-analyses as we anticipated high levels of heterogeneity between cohorts. Heterogeneity was quantified using the I² statistic, and described as low (≤25%), moderate (>25% and ≤50%) or substantial (≥75%).³²

We explored the impact of cohort and study characteristics on CMR and SMR values via subgroup, sensitivity and meta-regression analyses. Variables included: proportion of the sample that was female, geographic region (as defined by GBD), country income (low, middle, high), sampling frame (national, sub-national, city) and year of final follow-up. There were too few studies to examine whether mortality was associated with participants characteristics such as mean age, HIV status at baseline, and percentage of participants injecting amphetamines. Study quality variables included case ascertainment, completeness and study representativeness. Formal comparisons between subgroups were performed using meta-regression, via the metareg command.³³ Bubble plots were generated where significant differences were identified between subgroups. The adjusted R² index was employed to quantify goodness-of-fit for each model. Statistical significance for all analyses was set at p<0.05.

Results

The search produced 4,240 unique papers, of which 101 were screened in full, and 30 were included in the review (Figure 1), reporting on 25 cohorts. Appendix G provides the list of excluded studies and Appendix H further details on the included cohorts. Appendix I provides details of supplementary analyses including calculation of relative risks, subgroup analyses and sensitivity analyses.

Figure 1.

PRISMA flowchart displaying selection of studies in the review

Cohorts were primarily undertaken in North America and several Nordic countries (Sweden, Finland and Denmark); there were several in the Asia Pacific region (Taiwan, Thailand and Australia; Table 1). Cohorts varied widely in size (n=35 to n=74,139). Reporting of cohort demographics was incomplete: for example, fewer than half reported on the proportion who were women. Seven studies reported the percentage who had injected amphetamines; fewer studies reported other characteristics such as HIV and HCV status, mental health and exposure to environmental risks such as homelessness/unstable housing and incarceration (data not shown). We estimated the within-study relative risks for women vs. men in studies that included both genders: the pooled relative risk was 2.10 (95%CI: 1.77-2.49; see Figure I3 in Appendix I).

Crude mortality rates per 100 person-years (PY)

Twenty-three cohorts^{17,19–30,34–43} (n=115,223 participants) provided data to inform the pooled all-cause CMR. This was 1.11 per 100PY (95%CI 0.90-1.37; Figure 2, Table 2), with substantial heterogeneity (I²=96.9%). All-cause CMRs were higher in Southeast Asia^26,41 (CMR: 2.02 per 100PY, 95%:CI: 0.23-17.98, n=2 studies) and Western Europe^{19,21,24,25,27,28,30,34,35,38,40} (CMR: 1.09 per 100PY, 95%CI: 0.84-1.40, n=11 studies). They were lower in the single study⁴³ from central Europe (CMR: 0.49 per 100PY, 95%CI: 0.37-0.65).

Figure 2. Forest plots of pooled estimates of all-cause crude mortality rates (CMR) per 100 person-years (left) and standardised mortality ratio (SMR) (right) among people who use amphetamines, overall and by region.

CMR indicates how many deaths are expected to occur within 100 years of following an individual who uses amphetamines. SMR indicates the excess mortality risk that an individual who uses amphetamine has compared to an age- and sex- equivalent peer.

Table 2:

Pooled estimates of crude mortality rates (CMRs) and standardised mortality ratios (SMRs) among people who have problematic or dependent use of amphetamines

	Crude mortality rate					Standardised mortality ratio
	No. studies	No. people included	Pooled crude mortality rate per 100PY (95%CI)	I2	References	No. studies	No. people	Pooled standardised mortality ratio (95%CI)	I2	References
All-cause mortality	23a	115,223	1.11 (0.90-1.37)	96.9%	17–30,34,35,37–39,41–44	23	115,223	6.83 (5.27-8.84)	98.0%	17–30,34,35,37–39,41–44
Women	7	44,580	0.56 (0.40-0.77)	87.8%	17,18,26,34,35,42,43	7	44,580	5.91 (3.74-9.35)	94.3%	17,18,26,34,35,42,43
Men	10	42,248	0.97 (0.76-1.24)	95.2%	17–19,26,34,35,42,43	10	42,248	4.68 (3.43-6.40)	97.2%	17–19,26,34,35,42,43
GBD region
East Asia	2	22,403	0.83 (0.35-1.97)	98.6%	37,42	2	22,403	4.48 (2.53-7.93)	96.8%	37,42
Southeast Asia	2	769	2.02 (0.23-17.98)	94.8%	26,41	2	769	7.30 (0.82-65.30)	94.8%	26,41
High-income North America	5	75,820	1.19 (0.84-1.68)	86.2%	18,22,23,39	5	75,820	5.09 (3.55-7.30)	87.3%	18,22,23,39
Central Europe	1	3,039	0.49 (0.37-0.65)	-	43	1	3,039	6.22 (4.64-8.34)	-	43
Western Europe	11	8,750	1.09 (0.84-1.40)	91.4%	19,21,24,25,28,30,34,35,38	11	8,750	7.81 (5.03-12.12)	87.3%	19,21,24,25,28,30,34,35,38
Australasia	2	4,442	1.03 (0.51-2.07)	49.5%	17,20	2	4,442	10.52 (9.30-11.89)	0.0%	17,20
Cause-specific mortality
Drug poisoning	5	28,166	0.14 (0.06, 0.34)	95.5%	19,23,34,37,43	5	28,166	24.70 (16.67-36.58)	75.3%	19,23,34,37,43
Accidental injury	7	101,000	0.20 (0.08, 0.47)	98.4%	18,19,23,34,37,41,42	7	101,000	5.12 (2.88-9.08)	95.5%	18,19,23,34,37,41,42
Suicide	6	40,085	0.20 (0.07, 0.55)	98.2%	19,34,37,41,43,44	6	40,085	12.20 (4.89-30.47)	97.7%	19,34,37,41,43,44
Cardiovascular disease	4	24,985	0.13 (0.06, 0.29)	94.2%	19,23,34,37	4	24,985	5.12 (3.74-7.00)	54.9%	19,23,34,37
Homicide	4	25,127	0.03 (0.02, 0.06)	45.2%	19,23,34,37	4	25,127	11.90 (7.82-18.12)	13.6%	19,23,34,37

Note: For countries included in GBD regions please see www.healthdata.org/gbd/FAQ#What countries are in each region.

^aStudies included those that had both men and women in their cohort and reported all-cause mortality.

Cause-specific CMRs are summarised in Table 2 and presented in Figure 3. There was substantial heterogeneity for all pooled CMR (Table 2, Figure 3).

Figure 3:

Forest plots of pooled estimates of cause-specific mortality crude mortality rates (CMR) per 100 person-years (LHS) and standardised mortality ratio (SMR) (RHS) among people who use amphetamines

Small numbers of studies reported on several causes of mortality other than those presented here. These are summarised in Appendix I for the interested reader.

Standardised mortality ratios (SMRs)

Authors of studies of eight cohorts^{34,35,37–39,41–43} reported all-cause SMRs; we imputed all-cause SMRs for a further 13 cohorts^17–30. The pooled all-cause SMR across these 23 studies was 6.83 (95%CI: 5.27-8.84; Table 2, Figure 3). There were 9 studies^{17–19,26,34,35,40,42,43} (n=86,942 participants) that provided all-cause SMRs separately for men (SMR: 4.68, 95%CI: 3.43-6.40) and women (SMR: 5.91, 95%CI: 3.74-9.35; see Table 2 and Figure I2 in Appendix I). SMRs did not differ significantly according to GBD region (Table 2, Figure 2, and Appendix I).

Pooled cause-specific SMRs are summarised in Table 2 and Figure 3. Unsurprisingly, drug poisoning was the cause most highly elevated compared to the general population. The pooled homicide SMR was 11.90 (95%CI: 7.82-18.12) and the suicide SMR 12.20 (95%CI: 4.89-30.47). Cardiovascular disease and accidental injury SMRs were also elevated.

Mortality relative risks (RRs)

Appendix H presents estimated study-level mortality relative risks (RRs) for all-cause and cause-specific RRs, and Appendix H presents pooled estimates of all-cause and cause-specific mortality RRs. The pattern of findings was similar to those observed for SMRs (see above).

Factors associated with CMR and SMR

We ran a series of meta-regressions to explore potential variables that may have accounted for the very high heterogeneity observed for all-cause CMRs and SMRs (Table 3). There was no association of percentage women, years of study conduct, region or country income with either CMRs or SMRs. Methodological features of studies were also not significantly associated with CMRs or SMRs. CMR estimates were significantly lower for studies using national-level data than studies using subnational or city-level recruitment (see Table 3 and supplementary Figure I4 in Appendix I); by contrast, SMR estimates were not (Table 3, Figure I4 in Appendix I).

Table 3:

Associations between study-level variables and all-cause CMRs and SMRs derived from meta-regression analyses

	Crude mortality rate				Standardised mortality ratio
	N studies	Coefficient (SE)	Adj. R2	P	N studies	Coefficient (SE)	Adj. R2	P
% women	10	0.189 (0.261)	6.15%	0.263	10	0.646 (0.586)	−9.83%	0.643
Geographic region			1.18%				−21.35%
Western Europe	11	ref			11	ref
Australasia	2	0.836 (0.530)		0.781	2	1.147 (0.804)		0.847
High-Income North America	5	1.077 (0.448)		0.860	5	0.668 (0.311)		0.399
Central Europe	1	0.436 (0.343)		0.306	1	0.803 (0.713)		0.808
East Asia	2	0.744 (0.430)		0.615	2	0.580 (0.376)		0.412
Southeast Asia	2	2.151 (1.344)		0.237	2	0.592 (0.410)		0.460
Country incomea	23	1.822 (1.659)	−1.87%	0.517	23	4.511 (4.068)	8.31%	0.110
Year of final follow-up	23	0.991 (0.030)	−6.74%	0.766	23	0.977 (0.030)	−4.30%	0.468
Sampling frame			61.25%				14.66%
National	6	ref			6	ref
Sub-national	5	1.500 (0.495)		0.233	5	0.981 (0.481)		0.970
City	13	3.249 (0.813)		0.000	13	1.967 (0.735)		0.085
Case ascertainment	23	0.748 (0.076)	34.90%	0.010	23	0.872 (0.112)	1.32%	0.298
Study representativeness	23	0.736 (0.219)	1.39%	0.313	23	0.975 (0.317)	−6.36%	0.938
Completenessb	23	N/A	N/A		23	N/A	N/A

^aExcept for Quan, 2007²⁶ (upper middle income), all studies were conducted in a high income country.

^bAll studies had the same completeness score, resulting in a meta-regression being unable to be run.

Ref – reference category

Sensitivity analyses

We further examined whether there were differences between pooled estimates generated using only author-reported SMRs and compared to those that included our imputed SMRs. Details of these comparisons are provided in Appendix I Table 3. The pooled all-cause SMRs were very similar: 5.71 (95%CI: 4.44-7.35) for author-reported SMRs (n=8 studies^{34,35,37–39,41–43}), compared to our pooled all-cause SMR of 6.08 (95%CI 4.41-8.37; n=23 studies).

There were fewer cause-specific SMRs reported by study authors to permit robust sensitivity analyses, limiting our ability to compare findings. Nonetheless pooled cause-specific SMRs were similar for author-reported SMRs as for our pooled SMRs. These are summarised in Appendix I Table 3.

We also examined whether the removal of the single HIV-positive cohort³⁰ impacted on the overall all-cause CMR and SMR estimates. The pooled all-cause CMR and SMR were comparable to the original estimates with a CMR of 1.09 (95% CI: 0.97-1.22) and SMR of 6.47 (95%CI: 4.96-8.43). These forest plots can be found in Figure I5 in Appendix I.

Finally, we examined evidence for potential publication bias in two ways: through funnel plots and by examining whether there was an association between cohort size and either CMR or SMR. These are reported in Appendix J. The results did not suggest any evidence to support that a significant bias existed within this review.

Discussion

To our knowledge, this is the most detailed and comprehensive systematic review of mortality among people with regular or dependent amphetamine use conducted to date. We identified 25 eligible cohorts, who were followed over more than 750,000 person-years. Overall, it was estimated that people with regular or dependent amphetamine use had a 6.3-fold elevated rate of mortality compared to their age peers. In studies where both women and men were included, men had 2.1 times the risk of mortality compared to women. Data on age-specific mortality risk were rarely reported, so it is difficult to make any conclusions about the levels and patterns of mortality across the lifespan among this population.

The most common causes of death were drug poisoning, accidental injury, suicide and cardiovascular disease. Compared to the general population, the most highly elevated causes of death among people with regular or problematic amphetamine use were drug poisoning, homicide and suicide. Considerable attention has been placed on mortality among people who use opioids, particularly given the risk of fatal overdose. This review suggests that the elevations among people who use amphetamines (around 6.2-fold relative to the general population) are somewhat less marked than among people with dependent opioid use (around 14-fold elevated relative to the general population)⁴⁵.

This review, with the much larger number of studies, countries and participants compared to a previous systematic review¹², permits greater confidence in the synthesised results. In this review, we located 25 cohorts of which 23 could be used to estimate all-cause mortality (compared to the eight cohorts in the previous review¹²). It also meant that in contrast to the previous review, we were able to consider specific causes of death. Those that were most common were consistent with the known effects of amphetamines, which have been studied in clinical samples and studies using coronial and autopsy data (e.g.^7,9).

This review clearly found that people with regular or problematic amphetamine use are at elevated risk of several important causes of death, including suicide, homicide and injuries, as well as cardiovascular mortality. Interventions that effectively address amphetamine use disorders may therefore have the attendant benefits of reducing not only direct harms of acute drug overdose, but also of other causes of death that appear to be causally related to use. Although there has been substantial investment in research examining effective interventions for amphetamine use, there is no approved medication yet identified for use in withdrawal from and maintenance of abstinence from amphetamines. Cognitive behavioural therapy has shown some evidence of effectiveness, as has contingency management⁴⁶, yet it is often the case that people with problematic stimulant use do not seek treatment, or remain in treatment for their substance use; it is likely that only a minority of people with problematic or dependent amphetamine use receive treatment each year⁴⁷.

Limitations

It is important to acknowledge the limitations of this review and of the existing studies identified in this review. Although there was a substantial increase in the number of cohorts since our previous review a decade ago, nonetheless, gaps remain in terms of coverage of countries where problematic amphetamine use is known to occur. There were also limits in the design and reporting of cohort studies when they were undertaken. Although we endeavoured to obtain additional data from authors, and generated estimates of mortality rates and ratios where reporting was incomplete, but we had capacity to produce estimates ourselves, it is important that authors of studies using cohort designs improve the quality of reporting of study details, even if in supplementary material.

Reporting of cause-specific mortality was varied, and in many cases, authors were not clear on the exact manner in which cause of death had been ascertained (e.g. from coronial files, or if according to ICD, and if the latter, which codes had been used for specific causes of death). Greater use of definitions for specific causes of death generated by specialists in those fields might assist future endeavours to be clearer and more consistent¹⁵. Further research is needed to clarify many of what are categorised as drug related toxicity deaths in order to better understand this category in the context of amphetamines as contrasted with the clearer understanding of opioid related toxicity.

There was variability in the way that cohorts had been assembled and in how regular or dependent amphetamine use had been defined or operationalised. This may have explained some of the substantial heterogeneity in some estimates, rather than reflecting true variation in mortality risk among amphetamine users across those studies. We explored this via subgroup, sensitivity and meta-regression analyses. Variables reflecting study quality were not significantly associated with mortality observed. One exception was the sampling frame of studies: studies located within a single city had much higher mortality than those with a national sampling frame, which no doubt reflects the fact that single city samples may have been drawn from areas where problems had been identified (which may have been the reason for selection of the site). Another potential source of variability (which we could not directly examine) was that older cohorts (e.g. those before 2001) may well have differed from more recently conducted cohort studies in that the type of amphetamine being consumed differed, with likely much higher methamphetamine compared to amphetamine (including the more potent crystalline form) in more recent cohorts. Finally, mortality rates were not reported such that we could examine differing patterns of polydrug use and comorbid mental health problems, which may have affected the magnitude and nature of mortality risk across cohorts.

Although we endeavoured to be as thorough as possible in searching for studies examining mortality among people using amphetamines, we may have missed studies or made errors. We endeavoured to minimise this through independent checking of screening and data extraction, and then further stages of checking during the analysis and write-up phase by those not involved in extraction and generation of estimates. Given the previously mentioned gaps in reporting in the original studies, we used previously developed methods to backfill gaps where sufficient data had been presented to allow generation of estimates. The Global Burden of Disease estimates of expected age-specific mortality rates that we used to generate SMRs may have limitations, which we acknowledge. However, the GBD study is the most comprehensive effort directed at estimating precisely such data for every country in the world. Using these data allowed us to include substantially more datapoints compared to any previous review, and the robustness of these calculations was tested in sensitivity analyses.

Conclusions

There has been a substantial increase in studies examining mortality among people with problematic or dependent amphetamine use over the past decade. This has permitted more robust estimates of the extent and elevations in mortality among this population, as well as the first synthesised estimates of mortality for specific causes of death. People with regular or dependent amphetamine use are at elevated risk of a range of causes of mortality. Interventions to reduce the risks of overdose, suicide and traumatic deaths in this population are clearly warranted.