The long‐term productivity impacts of all cause premature mortality in Australia

Objective: To estimate the long‐term productivity impacts of all‐cause premature mortality in Australia by age, sex and cause of death.

P remature mortality, defined here as mortality occurring before the average life expectancy, accounts for about 67,000 deaths and more than half the all-cause mortality in Australia. 1 The value of productivity lost due to premature mortality has been shown to be a significant factor in the assessment of the total costs of illness. 2 Historically, the productivity-related costs of mortality have been estimated by applying average earnings to the period of production lost due to premature death. This is typically referred to as the human capital approach, and has its basis in the cost of illness methodology developed by Rice et al in 1966. 3 Calculation of mortality costs considers earnings over a lifetime rather than a single year since, if an individual had not died, they would have continued to be productive for a number of years according to their life expectancy.
Recent studies have applied this methodology to estimate the productivity related costs of mortality on a disease-specific basis, including for cancer, 4-10 cardiovascular disease (CVD) 11,12 and diabetes. 13,14 Estimates of productivity costs produced by these studies have ranged from $19,000 per CVD death in the European Union, 11 to $288,000 per cancer death in the US. 10 However, there have been no estimates of productivity losses due to premature mortality reported in an Australian context. A literature review identified just one study that quantified the productivity impacts of all-cause mortality across all major illness categories using a consistent and rigorous methodology. 15 However, the use of the friction cost method in this study, where productivity losses were valued only up to the point where an employee could be replaced, 16 meant that the estimates it produced were significantly less than studies that had applied the human capital approach. Given the variation in the methods applied and the types of costs included in previous studies, it was difficult to compare results across countries and diseases. 17 Reliable estimates of the relative productivity impacts of all-cause premature mortality would provide valuable information to decision makers allocating scarce resources amongst competing priorities. Australia, like most developed countries, has an ageing population. The proportion of people aged 65 and over is expected to more than double over the next few decades, resulting in economic consequences that will pose significant policy challenges. Specifically, population ageing is expected to slow Australia's workforce and economic growth, at the very time that burgeoning demands are placed on Australia's health and aged care systems. The Australian Government Productivity Commission's 'Economic Implications of an Ageing Australia' report identified measures to raise labour force participation and productivity as a key strategy to enhance income growth and the capacity to 'pay' for the costs of ageing. 18 Despite this focus, the potential for investment in effective health care interventions to increase the size of the The long-term productivity impacts of all cause premature mortality in Australia productive workforce is often overlooked in both health economic analyses and the broader policy making context. A recent study by Schofield et al presented the results of a microsimulation model developed to estimate the productivity impacts of chronic disease in older Australians. 19 The study reported a substantial productivity impact, with an estimated 112,000 lost productive life years caused by chronic illness in older workers in Australia between 2010 and 2030. The impact of this lost labour force participation on GDP was estimated to be $37.79 billion in 2010, increasing to $63.73 billion in 2030. The use of microsimulation methods in this study allowed for the calculation of productivity costs to be modelled on an individual basis. While the human capital approach was maintained, this technique represented a key advance over previous studies by enabling significantly greater variation and complexity to be captured. However, the model was developed specifically to assess the impact of morbidity and thus did not capture the impacts of mortality. To date, microsimulation methods have not been applied to estimate the productivity impacts of mortality in Australia or internationally.
The aim of this paper was to quantify the productivity related impacts of all-cause premature mortality in Australia by age, sex and cause of death. We developed a new microsimulation model, LifeLossMOD, to estimate the long-term impacts of premature deaths that occurred in 2003 in terms of the projected working years lost and the present value of lifetime income (PVLI) lost out to the year 2030. While previous studies [4][5][6][7][8][9][10][11][12][13][14] have been limited to projecting the productivity impacts of mortality based on assumptions around average wage rates, labour force participation rates and retirement ages, the estimates we report account for a number of individual factors that may influence these variables.

Methods
The productivity impacts of premature mortality were estimated using the human capital approach, where the impacts were assumed to be equal to the expected lifetime outcomes that would have been realised had the death been avoided. This is the traditional method for valuing production losses 3 and it remains the most widely adopted approach in the recent cost of illness literature. 4,5,7,8 We developed a new microsimulation model, LifeLossMOD, to apply a counterfactual life trajectory to each individual that died prematurely in 2003. These alternate lifespans provided a set of outcomes that were assumed to occur under a hypothetical scenario whereby an individual's death in 2003 was prevented. The process by which LifeLossMOD was developed is described in detail in Carter et al. 20 and is summarised below.  22 we excluded all deaths occurring in individuals aged 80 and above from the analysis. This is consistent with the Australian Institute of Health and Welfare definition of premature mortality as deaths occurring before a selected age cut-off, which may be determined or informed by average life expectancy. 23 A sensitivity analysis using 75 years as the threshold for premature death was conducted to assess the impact of uncertainty around this cut-off.

Building the microsimulation model: LifeLossMOD
Each mortality record contained variables describing the underlying cause of death as well as the individual's age, sex and the socioeconomic status of the suburb they resided in (as measured by the ABS's socioeconomic index for areas (SEIFA) quintiles). 24 However, the mortality dataset did not have information on the key economic outcomes of interest, including labour force participation and income. It was therefore necessary to assign estimates of additional variables onto the original mortality records, which was achieved using data from APPSIM.
APPSIM is a dynamic microsimulation model of the Australian population developed by the National Centre for Social and Economic Modelling to evaluate the impact of future fiscal and social policies. 25 The model uses a one percent sample of the 2001 Australian Census (188,000 records) as its base population. Future lifetime outcomes for this population are then projected using data from large surveys including the Household, Income and Labour Dynamics in Australia (HILDA), 26 the Longitudinal Study of Immigrants to Australia, 27 as well as official demographic data and projections. 28 APPSIM is also able to simulate changes to the population over time, including births, deaths and migration, as well as couple formation and separation and children leaving home. 29 In order to estimate the potential productivity gains forgone due to deaths occurring in 2003, each individual in the mortality dataset was matched with a similar individual existing in year 2003 of APPSIM. The resulting set of outcomes projected by APPSIM to the year 2030 were assumed to represent a series of counterfactual life trajectories for each individual that died in 2003. This process consisted of two steps. First, records in both the 2003 mortality dataset and the 2003 APPSIM population were grouped into homogenous cells, or 'bins' , based on their combination of age, sex and SEIFA quintile. In the second step, each mortality record within a particular bin was matched at random with an individual from the APPSIM dataset that appeared in the same bin.
To allow for the effects of uncertainty in the pairing of records, the matching process was replicated 100 times to create 100 uniquely matched datasets. These 100 simulated datasets comprise LifeLossMOD. The results contained throughout this paper report the mean of the 100 datasets. Where present, 95% confidence intervals have been calculated using the percentile method.

Estimating the economic impacts of premature mortality in 2003
The expected labour force participation lost due to premature mortality was estimated by accumulating the number of hours worked per week between 2003 and 2030 for each individual in LifeLossMOD. The impacts of mortality on labour force participation across individuals working both full-time and part-time were generalised using a full-time equivalent working year metric. This was derived by dividing the accumulated number of hours worked by the number of hours in a standard Australian working year (1,976, or 38 hours per week for 52 weeks per year: Fair Work Act (Cth) Section 62).
In order to quantify the productivity loss due to premature mortality, an estimate of the Present Value of Lifetime Income (PVLI) was derived. The PVLI represents the potential private income forgone due to premature The labour force analysis revealed that a total of 284,000 working years were lost due to premature deaths occurring in 2003. Male deaths were responsible for approximately 218,000 working years lost between 2003 and 2030, more than three times the number of working years lost due to female deaths ( Figure 1). The 45 to 54 years age bracket accounted for the greatest number of working years lost for both men and women.
The PVLI lost due to premature deaths in 2003 was estimated to be $13.8 billion between 2003 and 2030, with a 95% confidence interval of $13.7 billion to $13.9 billion ( Table  2). Male deaths accounted for 81% of the total PVLI lost, which was a function of the higher number of premature deaths among men (15,836 more men than women died prematurely in the year 2003), their higher labour force participation ( Figure 1) and their higher average incomes. 32 Male deaths between the ages of 25 and 54 accounted for more than half the total PVLI impact.
The PVLI lost due to premature mortality by cause of death was estimated (Table 3). Cancer was responsible for the greatest loss in PVLI (30%), followed by cardiovascular disease (19%), deaths from unintentional injuries (predominately transport accidents and falls) (15%) and deaths from intentional injuries (predominately suicide, self-harm and assault) (13%).
The most costly cause of death, per death, was intentional injuries, representing an average PVLI loss of $765,000 per death. This was followed by mental disorders ($654,000 per death) and unintentional injuries ($595,000 per death). The majority of premature deaths classified as having an underlying cause of mental disorder related to drug and alcohol dependence, but also included schizophrenia, anxiety, depression, bipolar disorder, personality disorders and eating disorders.
The diseases associated with the highest total PVLI loss for both men and women were assessed across 10 year age categories ( Figure 2). For men aged 15-34, unintentional injuries resulted in the greatest total PVLI lost ($781 million), followed by intentional Working years forgone the total PVLI lost. This reflected the relatively large number of deaths from these diseases as a proportion of total deaths (66%).
In addition to considering the total PVLI lost, we reported the average PVLI lost per death by cause of death. Deaths from injuries, both unintentional and intentional, and mental disorders were the most costly in this regard, each accounting for a loss of more than $595,000 per death. This is reflected in our analysis of the costs of disease by age category and sex which revealed that injuries accounted for the greatest PVLI loss in men aged 15-34, a group which have the largest potential earning capacity to lose.
It should be noted that we have based our estimates on the human capital approach, which is heavily influenced by the earnings potential of individuals. This method therefore gives greater weight to conditions affecting working age men compared with women, the young, the elderly, the indigenous and other ethnic minority groups. This approach is consistent with that adopted by most recent studies in this field, 4,5,8,9,33 but is an important consideration when interpreting our results. We have also excluded the value of care giving, household work and earnings from the informal economy in our analysis.
Our reference to 'productivity' costs throughout this manuscript has been adopted to reflect the common use of the term in the literature. However, we suggest that 'production' costs may be the more precise term for the types of losses we describe. By definition, production is the process of combining inputs to produce output, while productivity is a measure of the efficiency of each input in producing output. It is therefore conceivable that premature mortality could lead to a scenario where overall production goes down (because fewer people are working) but productivity (the average amount produced per person) remains the same, or even improves.
There are relatively few studies evaluating the productivity impacts of premature mortality available for comparative purposes. The 'Economic Burden of Illness In Canada 2005-2008' (EBIC) report 15 provided estimates of the value of lost production due to premature mortality by age, sex and cause of death. While previous EBIC reports have applied the human capital approach to measure the value of lost production, the most recent report applied the friction method. 16 Rather than measuring the present value of an injuries ($700 million). Cancer was the most common and costly cause of death for men aged 35-80 ($2.9 billion). The PVLI associated with premature mortality from cardiovascular disease became more significant in men aged 45 and above. Diseases of the digestive system were the fifth most costly category overall.
In women, the PVLI lost due to cancer dominated the cost of all other causes of death in those aged 25-80 ($1.0 billion). The next most costly categories in terms of PVLI were cardiovascular disease, unintentional injuries, intentional injuries and nervous system and sense organ disorders.
A sensitivity analysis was performed whereby the younger age cut-off of 75 years was used to classify deaths as premature. This resulted in the exclusion of 4,932 deaths relative to our main analysis. The impact on total productivity costs was insignificant, with 0.2% fewer working years lost and a decrease of 0.1% in the total PVLI loss reported.

Discussion
This is the first study to provide estimates of the productivity impacts of all-cause premature mortality in Australia. When projected to 2030, premature deaths occurring in 2003 accounted for a total of 295,000 working years lost and over $13.8 billion in PVLI lost. Deaths from cancer and cardiovascular diseases were the most costly overall, together accounting for almost half LifeLossMOD allows individuals to continue earning income beyond the traditional retirement age, and accounts for the projected increase in the age at which people will retire in the future. The higher LifeLossMOD figure may also be explained by the broader scope of the PVLI estimate and its inclusion of income sources outside of wages.
The relatively large variation in the estimates of the economic burden of mortality across different studies and countries demonstrates how geographical differences in the pattern of deaths and labour force dynamics may influence the nature and size of the estimates produced. This appears to be further compounded by differences in the type of data used and the methodological approach taken. A key advance of our study is the use of a single, complete mortality dataset and a consistent methodology to generate results across deaths from all causes while maintaining the capacity to report on impacts by cause of death. This allows for highly reliable order of magnitude comparisons of the impacts of mortality across various disease groups. Because each death in our mortality dataset was assigned only to the primary cause of death, we have avoided overstating the impacts by 'double-counting' a death in more than one disease category.
Our results further extend on the previous research by applying a microsimulation modelling technique that is able to account for significantly greater variation and complexity in generating projections. This approach builds on recent microsimulation methods developed by Schofield et al. that were applied to determine the economic impacts of illness in older Australians, 34 and has several advantages. A key strength of this approach was its ability to impute key variables of interest onto individual level mortality records, using reliable data sources and statistically robust techniques. It was therefore possible to significantly enrich the data available for analysis. The advantage of this approach over the use of aggregate datasets is that, because individual-level data is being used, the variation and complexity in the analysis could be significantly increased. It was therefore possible to account for differences in income across individual combinations of age, sex and socio-economic status. In addition, the estimates are able to account for projected trends in wage growth, labour force participation and retirement age. A further significant advantage of the microsimulation approach is its capacity to account for a broader scope of impacts than lost wages alone, including the impacts on labour force participation and total income which includes income generated from business profits and investments.
The model is limited by the predictive ability of the covariates used to link the mortality records with similar individuals in the APPSIM microsimulation model, those being age, sex and SEIFA index. We attempted to address this uncertainty by bootstrapping the process used to assign these counterfactuals to create 100 unique simulations. The results presented here report the mean of the 100 simulations along with confidence intervals generated by the bootstrapping process. The relative narrowness of the confidence intervals indicate that our results are robust to the effects of uncertainty in the record matching process.
Our analysis of productivity impacts was based on the officially registered underlying cause of death. This is defined in accordance with the International Classification of Diseases as the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury. 35 While this approach was necessary to avoid the double-counting of productivity impacts at an all-cause mortality level, the contribution of certain diseases to overall productivity costs may be underestimated. This is most likely evident for common comorbidities including chronic and unspecified kidney failure, diabetes, asthma, COPD, and dementia and Alzheimer disease. In interpreting our results, attention must be given to the broader cause of death categories we have applied; these are detailed in the Burden of Disease and Injury in Australia 2003 study. 21 Australia, like most developed economies, has an ageing population. As more people move into older age groups, overall workforce participation rates are projected to drop from around 63.5% in 2003-04 to 56.3% by 2044-45, and hours worked per capita will be about 10% lower than without ageing. 18 In this context, maximising workforce participation is increasingly being recognised as a key policy focus required to sustain economic growth. This paper highlights the significant labour force impacts associated with premature mortality, and thus the potential for investment in interventions that prevent mortality to have positive impacts on the size of the workforce.
There is a wealth of existing literature assessing the cost effectiveness of health care interventions and highlighting interventions which provide good value for money. For example, the 2013 World Health Organisation Global Action Plan for the Prevention and Control of Noncommunicable Diseases 36 identified a number of interventions that are affordable and give a good return on investment, generating one year of healthy life for a cost that falls below the gross domestic product (GDP) per person. Similarly, the recent ACE-Prevention study assessed the relative cost-effectiveness of a comprehensive set of preventive interventions for noncommunicable disease in Australia, together with a selected set of comparator care/cure intervention. 37 The study then recommended a 'menu' of cost effective interventions from which policy makers can select within available budgets. However, previous studies such as these have typically taken a health care perspective, with outcomes valued in terms of the likely health benefits gained but overlooking the potential economic benefits that would also arise. Our results provide decision makers with valuable information on the nature and scope of this additional set of productivity benefits. When interpreted alongside information on the cost effectiveness of health care interventions, this allows for an alternative, but complementary perspective on the long-term returns of health care investment which may be used to inform priority setting.

Conclusions
The cost of premature mortality to the Australian economy is substantial, with the long-term impacts of deaths in 2003 accounting for over 284,000 full time equivalent working years lost, and $13.8 billion in the PVLI lost, to 2030. The results from this study provide an assessment of the relative productivity impacts associated with premature mortality across the major cause of death categories, as well as a comprehensive overview of the distribution of these impacts by age and sex. This information can be used by decision makers in allocating scarce resources between competing priorities and may provide valuable information to governments seeking to improve not only the health but also the productivity of a nation.