Estimation of the impacts of substance use on workplace productivity: a hybrid human capital and prevalence-based approach applied to Canada

Abstract

Objective

Policy makers require evidence-based estimates of the economic costs of substance use-attributable lost productivity to set strategies aimed at reducing substance use-related harms. Building on a study by Rehm et al. (2006), we provide estimates of workplace costs using updated methods and data sources.

Methods

We estimated substance use-attributable productivity losses due to premature mortality, long-term disability, and presenteeism/absenteeism in Canada between 2007 and 2014. Lost productivity was estimated using a hybrid prevalence and incidence approach. Substance use prevalence data were drawn from three national self-report surveys. Premature mortality data were from the Canadian Vital Statistics Death Database, and long-term disability and workplace interference data were from the Canadian Community Health Survey.

Results

In 2014, the total cost of lost productivity due to substance use was $15.7 billion, or approximately $440 per Canadian, an increase of 8% from 2007. Substances responsible for the greatest economic costs were alcohol (38% of per capita costs), tobacco (37%), opioids (12%), other central nervous system (CNS) depressants (4%), other CNS stimulants (3%), cannabis (2%), cocaine (2%), and finally other psychoactive substances (2%).

Conclusion

In 2014, alcohol and tobacco represent three quarters of substance use-related lost productivity costs in Canada, followed by opioids. These costs provide a valuable baseline that can be used to assess the impact of future substance use policy, practice, and other interventions, especially important given Canada’s opioid crisis and recent cannabis legalization.

Electronic supplementary material

The online version of this article (10.17269/s41997-019-00271-8) contains supplementary material, which is available to authorized users.

Introduction

Policy makers require empirical estimates of the societal impacts of substance use (SU) to prioritize allocation of public resources and inform policy in efforts to promote population health. Here, we define “substances” as those whose use can be harmful, whether due to the acute psychoactive effects of their use (i.e., accidental poisoning attributable to non-medical opioid use) or as a result of consumption (i.e., chronic obstructive pulmonary disease resulting from tobacco use or infection transmission from injecting substances). In the context of this research, “harms” refers to premature death or disablement attributable to the consumption of substances. We focus on the most commonly used harmful substances: alcohol, tobacco, cannabis, opioids (including heroin and prescribed pain relievers), other central nervous system (CNS) depressants (including benzodiazepines and barbiturates), cocaine, other CNS stimulants (including amphetamine and methamphetamine), and other substances (including hallucinogens and inhalants).

Harms may be measured in terms of burden of disease and economic costs arising from providing healthcare and criminal justice services, as well as effects on workplace productivity resulting from SU (Degenhardt et al. n.d.; Single et al. 2003; Larg and Moss 2011; Single et al. 1998; Rehm et al. 2006). In June 2017, the Canadian Substance Use Costs and Harms (CSUCH) Scientific Working Group released a full assessment of the societal costs and harms associated with substance use in Canada within these domains spanning 2007 to 2014 (Canadian Substance Use Costs and Harms Scientific Working Group 2018).

Previous studies indicate that lost productivity is the leading source of economic costs from SU, with larger contributions than from healthcare or crime (Single et al. 1998; Rehm et al. 2006). For example, in a study of the cost associated with alcohol dependence in 14 high-income countries, the indirect costs of productivity losses were estimated to account for more than 70% of the total cost (Mohapatra et al. 2010). We attribute lost productivity to SU when a substance-using individual displays reduced productivity on the job or is removed from the workforce as a result of premature mortality or disablement.

This paper discusses the methods and results from the lost productivity portion of the CSUCH assessment (Canadian Substance Use Costs and Harms Scientific Working Group 2018). Details on the other domains of costs arising from SU, such as healthcare and criminal justice, are forthcoming. Building upon methods previously employed by Rehm et al. (2006) and Single et al. (1998), updated data sources and methods are described and the costs of SU-attributable lost output due to premature mortality,1 long-term disability (LTD), and reduced productivity at work in Canada from 2007 to 2014 are estimated.

Methods

This study is a comprehensive account of societal costs using an ecological population-level framework. We do not account for private costs (i.e., the cost of purchasing substances) or intangible costs (nonmonetary outcomes such as pain and suffering) associated with SU (Single et al. 2003; Larg and Moss 2011).

Estimating substance use-attributable fractions

While some conditions are wholly attributable to SU (e.g., alcoholic gastritis), others are considered partially attributable (e.g., fibrosis and cirrhosis of the liver). For these latter events, an attributable fraction (AF) is used to determine the portion of events that can be attributed to SU. The AF is defined as the proportion of all cases of an event that is associated with a particular cause (Levin 1953). An AF is developed by (1) identifying the health conditions attributable to SU; (2) determining the risk estimates of developing a health condition associated with different consumption levels based on our best knowledge from literature reviews (Sherk et al. 2017; National Center for Chronic Disease Prevention and Health Promotion (US) Office of Smoking and Health 2014; Degenhardt et al. 2016; Degenhardt et al. 2013); and (3) these estimates are then applied to the prevalence of SU in the population of interest. An AF of one is used for wholly attributable conditions.

Detailed methods used to estimate attributable fractions were based on methods published by Sherk et al. (2017), Kehoe et al. (2012), Rehm et al. (2010), and Greenland (2008) and are available online (www.csuch.ca). For the current study, a list of 83 SU-attributable health conditions and SU-related risk estimates were identified using the World Health Organization’s Global Status Report on Alcohol and Health 2014 (World Health Organization 2014), US Surgeon General’s 2014 report (National Center for Chronic Disease Prevention and Health Promotion (US) Office of Smoking and Health 2014), a previous Canadian cost study (Rehm et al. 2006), preparatory work from the Global Burden of Disease study (Degenhardt et al. 2016), and the International Model of Alcohol Harms and Policies (Sherk et al. 2017). Health conditions were defined using the International Statistical Classification of Diseases and Related Harms codes (World Health Organization 2016). Lists of these conditions are presented as supplementary material in Appendix A.

Estimating substance use prevalence

Detailed methods for estimation of SU prevalence are described elsewhere and in a forthcoming publication, and are beyond the scope of the current analysis (Zhao et al. 2018). Briefly, estimates were drawn from the Canadian Tobacco, Alcohol and Drugs Survey; the Canadian Alcohol and Drug Use Monitoring Survey; and the Canadian Community Health Survey (CCHS) and Northwest Territories Substance Use Survey (Statistics Canada 2016a; Statistics Canada 2014; Statistics Canada 2017a; Northwest Territories Health and Social Services 2012). Missing and marginally reliable data were predicted using empirical best linear unbiased prediction methods, supplemented with year- and region-specific census data on alcohol and tobacco sales as covariates (Statistics Canada 2017b; Statistics Canada 2017c).

Premature mortality and potential years of productive life lost

Mortality data from Statistics Canada’s Vital Statistics Death Database were used to determine the number of incident cases of premature mortality and average age at death by age groups (i.e., 0–14 years, 15–34 years, 35–64 years), due to SU-related conditions (Statistics Canada 2017d). Assuming a retirement age of 65, potential years of productive life lost (PYPLL) were then calculated as the average age at death subtracted from 65, by age group. For the deceased in age group 0–14 years, we assumed 50 productive years lost (i.e., these individuals would not contribute any production).

Lost productivity costs due to substance use-attributable premature mortality

To assess the costs associated with premature mortality, we employed the human capital approach (Larg and Moss 2011; Rehm et al. 2006; Dobrescu et al. 2012). This approach involves estimating the impact of premature mortality on current and future productivity. It is especially suited to estimating costs of incident data and age subgroups as it is sensitive to PYPLLs (Pearce et al. 2015).

The human capital cost of lost productivity was estimated for unemployment-adjusted premature mortality using the current value of annuity formula and assuming a discount rate of 2% per annum using Eq. 1: (Brealey et al. 2016)

$${\text{Cost of lost production}}_{\text{premature mortality}}=\left(\text{Total mortality cases}\ast \mathrm{AF}\right)\ast \left(\frac{\text{Yearly wage}}{0.02}\right)\ast \left[1-\frac{1}{{\left(1+0.02\right)}^{\text{PYPLL}}}\right]$$ 1

where the terms in this equation are defined as follows:

Total mortality cases is the number of deaths for the year and region of interest for those aged 0-64 years, by age group, with adjustment for workforce participation and health conditions causally associated with SU. When cause of death was known to be substance use, but the substance could not be uniquely determined based on the ICD10 codes present on the Vital Statistics Database, deaths were apportioned among possible diagnoses based on corresponding inpatient hospitalizations proportions that were attributable to SU by substance-type, age, sex, region and year that were calculated for the larger CSUCH project (Canadian Substance Use Costs and Harms Scientific Working Group 2018),

AF is the SU-category and health condition-specific attributable fraction,

Yearly wage is the median reported all-type weekly work wage by year and province/territory from the Survey of Employment, Payrolls and Hours converted to yearly wages (Statistics Canada 2017e), and

PYPLL are the potential years of productive life lost.

These costs were then summed across health condition and province/territory to arrive at a total lost production cost due to SU-attributable premature mortality by year and substance category.

Long-term disability

Yearly point prevalence of LTD was used as the basis for cost estimates following the methods described by Schroeder (2012). First, estimates of the prevalence of individuals permanently removed from the workforce due to LTD were acquired from the 2007–2014 cycles of the CCHS (Statistics Canada 2016b), a multi-year national survey. The survey asks respondents, “Last week, did you work at a job or a business?” One possible response, “Permanently unable to work,” was assumed to estimate the frequency of respondents unable to work due to permanent disability. To validate this assumption, we compared this with the frequency responding to a similar question from a different survey—the CCHS Mental Health (MH) survey (Statistics Canada, 2017a). Specifically, they responded to the question, “What is the main reason that you have not worked at a job or business in the past three months?” with the following response option, “Respondent’s own chronic physical or mental health condition diagnosed by a health professional.” The percentages were very similar (2.6% CCHS vs. 2.2% CCHS-MH). Responses were limited to those of productive age, or aged 15–64 years.

The total costs of lost productivity due to LTD were then estimated by multiplying these cases by the median yearly wage (Statistics Canada 2017e). To estimate the cost associated with SU-related LTD, we applied the corresponding inpatient hospitalizations to the total counts and costs of lost productivity due to LTD.

Absenteeism and presenteeism

Absenteeism refers to an employed individual being absent from work, whereas presenteeism refers to decreased productivity while on the job. Presenteeism can happen when an employee’s output is decreased due to impairment or “hangover” effects resulting from use of psychoactive substances.

In the CCHS, a subsample of working respondents who report past-year SU were asked, “In the past 12 months, how much did your [drinking or all-type drug use] interfere with your ability to work at a job?” Responses may range from zero (no interference) to 10 (very severe interference). The frequency of workplace SU interference was assumed as the proportion of non-zero responses. These responses were used as a composite indicator of absenteeism and presenteeism. Average frequency of respondents who reported interference and interference levels (as a proportion of 10) were calculated for alcohol and other psychoactive substances. For example, an interference response level of three would correspond to a 30% reduction in productivity. Costs were then estimated as follows in Eq. 2:

$$\text{Cost of}\mathrm{SU}\text{interference}=\text{Population}\ast \left(1-\text{Unemployed rate}\right)\ast \text{Proportion reporting interference}\ast \left(\text{Interferences level}\ast \text{Yearly wage}\right)$$ 2

where the terms in this equation are defined as follows:

Population is the Canadian provincial/territorial population, restricted to working ages 15-64, taken from Statistics Canada’s Annual Demographic Estimates: Canada, Provinces and Territories (Statistics Canada 2017b), and

Unemployment rate is the Canadian unemployment rates expressed as a proportion of working individuals by year, sex, and age group from the Labour Force Survey (Statistics Canada 2017f), and

Proportion reporting interference as described above, and

Interference level as a proportion, described above.

To distribute other substances’ costs by substance type, we applied the SU-attributable proportion estimates.

Absenteeism as a result of tobacco use

In the CCHS, respondents were not asked to estimate workplace interference resulting from tobacco use and so we were not able to apply the above methodology. It is well known that tobacco use is a risk factor for a number of chronic and communicable diseases that may lead to absenteeism. Using data from the CCHS, Bounajm et al. (2013) demonstrate that smokers take about 10 sick days per year, vs. eight taken by non-smokers, after controlling for age, sex, education, and occupation (National Center for Chronic Disease Prevention and Health Promotion (US) Office of Smoking and Health 2014; Bounajm et al. 2013; Tsai et al. 2005). The cost of tobacco-attributable absenteeism was estimated using Eq. 3 below:

$$\mathrm{C}\mathrm{o}\mathrm{s}\mathrm{t}\mathrm{of}\mathrm{t}\mathrm{o}\mathrm{b}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{o}\text{-}\mathrm{a}\mathrm{t}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{b}\mathrm{u}\mathrm{t}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}\mathrm{a}\mathrm{b}\mathrm{s}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{e}\mathrm{i}\mathrm{s}\mathrm{m}=\mathrm{P}\mathrm{o}\mathrm{p}\mathrm{u}\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\ast (1-\mathrm{U}\mathrm{n}\mathrm{e}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{o}\mathrm{y}\mathrm{e}\mathrm{d}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{e})\ast \mathrm{S}\mathrm{m}\mathrm{o}\mathrm{k}\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{v}\mathrm{a}\mathrm{l}\mathrm{e}\mathrm{n}\mathrm{c}\mathrm{e}\ast \left(\frac{\mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{l}\mathrm{y}\mathrm{w}\mathrm{a}\mathrm{g}\mathrm{e}}{\mathrm{W}\mathrm{o}\mathrm{r}\mathrm{k}\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}}\right)\ast 2\text{sick}\mathrm{d}\mathrm{a}\mathrm{y}\mathrm{s}$$ 3

where the terms in this equation are defined as follows:

Population as described above,

Unemployment rate as described above,

Smoking prevalence is the frequency of current smokers estimated as above, and

Working days is assumed to be 226 days per year (365 days per year, minus 104 weekend days, minus 11 holidays, minus 15 vacation days, minus nine sick days).

In order to make our findings comparable across years, cost results have been converted to 2014 real Canadian dollars using the National Consumer Price Index (Statistics Canada 2017g). All analyses were conducted by year, sex, province/territory, and age group for each substance category. However, only national totals by substance are presented in this report. All analyses were carried out using R version 3.4.3.

This study was approved by the Research Ethics Board of the University of Victoria (17-068).

Results

While the national results presented focus on 2014, the most recent year of analysis, we provide figures presenting results across our time series, 2007–2014. Detailed aggregate cost and harms estimates across years, substances, and regions are provided as supplementary material in Appendix B and in an online data visualization tool created for this project (www.csuch.ca/explore-the-data/).

Premature mortality

In 2014, a total of 20,713 productive-aged individuals were removed from the workforce due to SU-attributable premature mortality, representing 284,324 PYPLLs. Included in Table 1 are counts, average age at death, and PYPLLs for Canadians removed from the workforce due to premature death as a result of alcohol, tobacco, and other SU.

Table 1.

SU-related premature mortality counts, average age at death and PYPLLs in Canada by substance among those aged 0–64, 2014

Substance	Mortality count	Average age at death	PYPLL
Alcohol	6845	49	110,148
Tobacco	9480	57	74,622
Cannabis	565	53	6892
Opioids	2108	41	49,533
Other central nervous system (CNS) depressants	708	43	15,582
Cocaine	275	35	8219
Other CNS stimulants	457	36	13,068
Other substances	277	42	6359
Total	20,713	51	284,324

Tobacco was responsible for the highest burden of SU-attributable premature mortality among productive-aged Canadians, with 9480 cases in 2014. During this year, alcohol use was responsible for the second highest number of deaths at 6845 cases with 110,048 PYPLLs.

While tobacco use led to a greater number of productive-aged deaths in 2014 when compared with alcohol, its use is associated with a higher average age of death compared with alcohol (57 vs. 49 years, respectively). As such, death from tobacco-related conditions was associated with fewer overall PYPLLs than alcohol (see Table 1).

Long-term disability

In 2014, 67,193 Canadians were unable to work due to SU-related LTD, contributing to a corresponding number of PYPLLs. Tobacco, alcohol, and opioids were the three substances responsible for the largest burden. Results are presented in Table 2.

Table 2.

Estimated potential years of productive life lost (PYPLLs) due to SU-attributable long-term disability (LTD) in Canada for persons aged 15–64 years, 2014

Substance	PYPLLa
Alcohol	28,932
Tobacco	30,247
Cannabis	1544
Opioids	2528
Other CNS depressants	1861
Cocaine	608
Other CNS stimulants	875
Other substances	598
Total	67,193

^aUnder our prevalence-based LTD cost methodology, case counts correspond to PYPLLs

Age at disability was not available for LTD; however, the average age among LTD cases was 50 years in 2014.

Absenteeism and presenteeism (interference)

National SU workplace interference frequency and levels of interference estimates across our time series are presented in Table 3. Self-reported workplace interference due to alcohol use was most prevalent. However, overall reported SU interference was infrequent, with no substance identified as interfering with work performance for more than a half a percent of the working population. The level of workplace interference was greater for substances other than alcohol.

Table 3.

SU work interference prevalence and level/reduction in productivity, converted to percentage (%) among employed respondents reporting SU, years 2007–2014

Substance	Prevalence of working population reporting SU interference (mean %, (range))	SU interference level/reduction in productivity (mean %, (range))
Alcohol	0.39 (0.26–0.52)	23 (21–25)
Any-type other psychoactive substancesa	0.15 (0.08–0.20)	31 (28–36)

^aIncludes cannabis, opioids, other CNS depressants, cocaine, other CNS stimulants, and other substances

Our analysis for tobacco absenteeism allowed us to calculate the years of lost productivity, which accounted for 42,435 PYPLLs in 2014 with little variation across our study period.

Total costs of substance use-attributable lost productivity

In 2014, alcohol use was responsible for the greatest productivity-related costs, accounting for 38% of the total costs. This was followed closely by tobacco use, which accounted for 37% of lost productivity costs in Canada, 2014 (see Table 4). Opioid use contributed 12% of costs. Total productivity loss due to SU in 2014 amounted to a total of $15.7 billion, an increase of 17% from $13.4 billion in 2007.

Table 4.

SU-related premature mortality and long-term disability lost productivity costs in Canada, 2014 (2014 CND dollars × 1,000,000)

Substance	Premature mortality	LTD	Interference	Total (premature mortality + LTD + interference)
Alcohol	3874	1367	675	5916
Tobacco	3007	1414	1422	5844
Cannabis	249	72	46	368
Opioids	1622	118	92	1832
Other CNS depressants	522	87	76	685
Cocaine	248	29	17	294
Other CNS stimulants	395	42	22	459
Other substances	211	28	23	261
Total	10,129	3158	2372	15,659

National per capita costs for 2014 are presented below. The total cost of $440.66 per person increased 8% from 2007. Again, we find that alcohol and tobacco contributed the largest lost productivity per capita costs, accounting for a combined 75% of total costs (Table 5).

Table 5.

Per capita SU-related lost productivity costs in Canada, 2014 ($ CND)

Substance	Per capita total (premature mortality + LTD + interference)	% total
Alcohol	166.49	38
Tobacco	164.45	37
Cannabis	10.36	2
Opioids	51.54	12
Other CNS depressants	19.27	4
Cocaine	8.28	2
Other CNS stimulants	12.91	3
Other substances	7.36	2
Total	440.66	100

2007–2014 costs of substance use-attributable lost productivity trends

Across our time series, productivity losses due to alcohol and tobacco use were similar and generally increasing. Opioid use contributed the third largest costs, with an increasing trend noted. The remaining substances were associated with variable costs and trends. Total costs over our study period are presented in Figs. 1 and 2.

Fig. 1.

Total substance use-related lost productivity costs in Canada (2014 CND dollars × 1,000,000), other psychoactive substances shown in aggregate, 2007–2014

Fig. 2.

Substance use-related lost productivity costs in Canada, excluding alcohol and tobacco (2014 CND dollars × 1,000,000), other psychoactive substances separated by substance type, 2007–2014

Discussion

We present the lost productivity costs of SU in Canada due to premature mortality, LTD, and presenteeism/absenteeism for the years 2007–2014. We find substances legal during this time, tobacco and alcohol, responsible for the largest proportion of costs: 75% of productivity losses were due to these substances in 2014. Further, costs have increased steadily over the years 2007–2014. These increasing costs may be partially explained by a number of simultaneously occurring factors. Despite adjusting all costs to 2014 dollars, real wages have increased across our time series. Furthermore, a growing and ageing population is an intuitive contributor.

Opioids as a category deserve special attention because it is the only examined substance that saw a decrease in use while the related productivity losses increased substantially between 2008 and 2014. This is largely due to the high mortality rate associated with opioids resulting from the appearance of fentanyl and other novel synthetic opioids into the Canadian illicit opioid market (Canadian Community Epidemiology Network on Drug Use 2014; Canadian Community Epidemiology Network on Drug Use 2015a; Canadian Community Epidemiology Network on Drug Use 2015b; Canadian Community Epidemiology Network on Drug Use 2016). Deaths from opioid overdose have risen substantially during 2014 to 2017 and hence economic costs will also have risen since then (Fischer et al. 2018; Public Health Agency of Canada 2019). This is in contrast to cocaine productivity costs, which have steadily decreased with declining prevalence of use.

Due to differing methods and data sources, our estimates are not comparable with the previous Canadian cost of SU study carried out by Rehm et al. (2006). Perhaps the most significant departure involved our not applying the human capital approach to assess lost productivity due to LTD. As with other cost of SU studies, the prevalence of the population absent from the workplace due to LTD was available (Single et al. 1998; Rehm et al. 2006; Dobrescu et al. 2012). As discussed by Schroeder (2012), estimates of impacts on future productivity should only be made if incidence rather than prevalence data are available. To explain, a cohort of prevalent cases of individuals unable to work due to LTD will predominantly remain classified as such in the subsequent year of analysis. Restricting costs to the year of analysis will prevent year-on-year double counting (Schroeder 2012). Additionally, advancements in the literature have allowed us to estimate the harms among former drinkers. We have also utilized updated estimates of the risk of various cancers and ischaemic heart disease associated with alcohol use, another departure from the previous study (Zhao et al. 2017).

While our study provides valuable insight into the magnitude of SU issues in Canada by showing how they affect labour productivity, our estimates are also subject to other limitations. Mainly, the data in the CSUCH study rely on secondary analysis of other existing data and are subject to measurement error. Additionally, due to a limitation in the data available, we were unable to separate costs related to absenteeism and presenteeism, and therefore provide these in aggregate. Future studies should seek to identify disaggregated indicators of this, or in the absence of such, valid methods to disaggregate the composite indicator used for this analysis.

In addition, there are some limitations inherent to similar cost of SU studies to be acknowledged. Annual LTD data may include people who were ill or injured for only part of the year, and their inclusion in analysis may be a source of overestimation. While it was not possible to separate incident and prevalent LTD cases as a limitation of the data, our assumption that LTD cases are comprised predominantly of disabilities caused by SU in earlier years seems reasonable. Examining the rates of Canadian Pension Plan (CPP) Disability beneficiaries between 2011 and 2014 shows a slightly decreasing trend (Statistics Canada 2017b; Employment and Social Development Canada 2019). This suggests that a proportion of beneficiaries similar to the proportion of incident cases effectively “age out” of CPP Disability eligibility based on their birthday. Because our unit of time is one year, those 65 years of age during any year will not be captured by our methodology, therefore counteracting inclusion of incident partial year cases.

Throughout our analysis, we define lost productivity as the value of lost marketed output, and exclude non-marketed output in the form of home production, volunteer work, and non-marketed care for family members, etc. While these are important sectors, our objective of estimating reductions in the availability of tangible resources precludes these inclusions (Larg and Moss 2011).

Our choice of applying the median wage to derive cost estimates deserves attention. Some literature exists suggesting individuals using these substances may discount future health heavily, and in turn, may not place much weight on future benefits. This may then put the individual on a lower lifetime trajectory in regard to education/training and correlate to lower earning potential (Evans and Montgomery 1994). However, we feel the evidence base of methods to adjust for this insufficient to include in our analysis. Furthermore, our objective to provide estimates for the whole of society, across all levels of consumption, accounts for the breadth of users across socio-economic strata. As such, we believe the choice of median wage is appropriate.

Further, our estimates of the costs arising from premature disability are sensitive to our applied discount rate. Convention is to apply a discount rate approximately equal to the real rate of return on long-term government bonds (Larg and Moss 2011). Across our study period, this rate averaged 1.3% (Statistics Canada 2019a). Allowing for market variation, and in consultation with health economists on our advisory committee, we have chosen 2% as an appropriate and conservative discount rate.

The Vital Statistics Death Database used for this analysis was limited to one ICD10 code representing the underlying cause of death; secondary diagnoses were not available. This led to limitations regarding substance use poisoning deaths, as the underlying cause of death is listed as poisoning, but the involved substance(s) are not listed. This is a data limitation and results should be interpreted with caution accordingly.

Overall, the current study supports the conclusions of previous studies in Canada, which show that substances that have been legal for some time in Canada (i.e., alcohol and tobacco) are responsible for the largest proportion of the disease burden and productivity loss (Single et al. 1998; Rehm et al. 2006). A steady increase in SU costs was found from 2007 to 2014 for almost all substances, and these increases highlight the need for new and ongoing policy efforts to prevent such losses in Canada. Within the context of an ongoing opioid crisis and recent cannabis legalization, our findings provide a consistent baseline dataset for the period preceding these events (Government of Canada 2018; Statistics Canada 2019b). With plans to provide updates as data are available, we will be able to examine the consequences of these events.

Conclusion

We demonstrate a hybrid approach to cost of SU studies when prevalence and incidence data are available. Our research provides a comprehensive estimate of the annual costs of lost productivity due to SU in Canada from 2007 to 2014, with legal substances again accounting for the vast majority of costs. Our findings are also unique in being the first to estimate greater costs for alcohol than for tobacco.

Compliance with ethical standards

This study was approved by the Research Ethics Board of the University of Victoria (17-068).