EXECUTIVE SUMMARY

1.0  INTRODUCTION
1.1  Background
1.2  Purpose and Scope
1.3  Methodology
1.3.1 Policy Analysis Framework
1.3.2 Forecasting Impacts
1.3.3 Chronology of Investigation

2.0 AIR QUALITY FORECASTING
2.1 Primary Conclusions
2.2 Current Situation
2.3 Key Issue
2.4 Recommended Strategy
2.5 Summary

3.0 HEALTH EFFECTS
3.1 Primary Conclusions
3.2 Current Situation
3.3 Key Issues
3.3.1 Correlation and Causality
3.3.2 Need for Corroborating Clinical Studies
3.3.3 Role of Each Pollutant
3.3.4 Health Effects Thresholds
3.3.5 Population Risk Factors
3.3.6 Other Confounding Factors
3.4 Recommended Strategy

4.0 HEALTH CARE UTILIZATION AND EXPENDITURES
4.1 Primary Conclusions
4.2 Current Situation
4.3 Key Issues
4.3.1 Guiding Economic Logic
4.3.2 Co-morbidity/Co-mortality
4.3.3 Marginal vs Average Costs
4.3.4 Data Availability
4.3.5 Premature Mortality
4.4 Recommended Strategy
4.5 Summary

5.0 NEXT STEPS AND PRIORITIES
5.1 Detailed Review of Current Version of the AQVM
5.2 Detailed Review of Health Effects Categories
5.3 Preliminary Analysis of Provincial Health Care Data
5.4 Standard Treatment Procedures
5.5 Full Cost Accounting of Fixed Assets
5.6 Health History Profiles
5.7 Premature Mortality Analysis
5.8 Selection of Air Quality Forecasts
5.9 Generate Health Care Expenditure Estimates
5.10 Communication of Results

APPENDIX A - List of Individuals and Agencies Contacted

APPENDIX B - List of Participants in the Expert Review Workshop

APPENDIX C - Summary of the 1996 Air Quality Valuation Model

APPENDIX D - Hypothetical Health History Profile for a Representative Statistical Patient

Background

The OMA has recognized air quality as a significant public health issue. The Association wishes to develop a full understanding of the health care resource utilization and associated expenditures caused by air pollution. This study provides a review of current understanding and recommends a plan of action for the OMA to improve estimates of the health care impacts of air pollution.

Current Situation

Air Quality Forecasting

Numerous, technically sound and current air quality forecasts are available for parts or all of Canada. The number and reliability of these forecasts have increased substantially over the last decade. Additional improvements in air quality forecasts will be forthcoming in the near future, particularly in regards to particulates. Forecasting efforts are underway to distinguish between inhalable (i.e., PM10) and respirable (i.e., PM2.5) particulates. The air quality forecasts currently available are of sufficient rigor and technical credibility to support reasonable estimates of health effects.

Epidemiological Evidence

Much investment in epidemiological research has occurred over the last two decades to determine the relationship between poor air quality and human health effects. This literature is now extensive, technically sound and consistent in the derived quantitative relationships. This epidemiological literature provides a good evidentiary basis to make quantitative predictions of air quality-induced health effects. Nonetheless, some raise criticisms in terms of:
 

1. a failure to demonstrate comparable results in clinical studies,
2. correlation does not demonstrate causality,
3. the failure to distinguish the precise role of each pollutant in the air quality "soup",
4. the potential for health effect thresholds below which no effects occur,
5. the need to account for variations in the risk factors among segments of the population, and
6. other potential confounding factors (e.g., temperature/humidity/seasonality, indoor vs. outdoor air quality, genetic predisposition).

These criticisms are worthy of note but are not cause for inaction. Instead, prudence and a clear declaration of uncertainty are warranted in making any predictions. The current epidemiological evidence has been used extensively in Canada, the U.S. and Europe to make predictions of the health effects of air pollution. These predictions have provided an important foundation for developing responsible environmental policies.
 

Health Care Utilization and Expenditures

Economics are playing an increasingly central role in environmental policy decisions relating to pollution control. Governments are examining environmental issues more and more from a business perspective. Much experience has been gained over the last 15 years using economic valuation techniques to analyze environmental policies, particularly in the U.S. This trend is expected to strengthen in the future, particularly in Canada.

Most primary research in developing economic estimates of health effects from air pollution is U.S.- based. Applying economic measures from one jurisdiction to another, in particular health care expenditure estimates, is unreliable. Comprehensive health care utilization data are available for Canada but these data have not been used extensively up to this point in developing economic estimates of the costs of air pollution.
 

The Air Quality Valuation Model

Health Canada and Environment Canada jointly have developed the Air Quality Valuation Model (AQVM). The AQVM uses the latest air quality, population and epidemiological data to generate forecasts of population-wide health effects related to air pollution. The AQVM produces estimates for 15 types of health effects/treatments, including premature mortality. The AQVM can generate estimates of the probability of health effects and the related economic value.

This model is designed to estimate total economic value. It does not report health care resource utilization or associated expenditures. Nonetheless, many of the key elements required to generate reliable expenditure estimates are provided by the AQVM results.

Outstanding Issues

The focus of this study is on improving estimates of health care resource utilization and associated expenditures caused by air pollution. This study has not examined how to improve estimates of other economic elements related to the health effects induced by poor air quality (e.g., changes in quality of life and premature mortality). As such, estimates of health care resource utilization and associated expenditures provide only part (albeit a significant part) of the total economic benefit which could be realized from air quality improvements.

Focusing only on health care expenditures can potentially create problems, particularly when premature mortality is a significant consideration. Premature mortality can result (at least potentially) in a reduction in health care expenditures. Reducing longevity can reduce health care utilization.

By considering only health care expenditures, results could support the perverse conclusion that air pollution provides a positive economic benefit. Several strategies to deal with this possibility are recommended.
 

1. Reductions in health care expenditures from premature mortality may not actually be observed depending on the specific health care expenditure patterns associated with air pollution. The nature and magnitude of the effect of premature mortality will not be known until reasonable estimates of health care expenditures have been developed. The best strategy to deal with this concern should be reviewed when the full effect of premature mortality is better understood.

 
2. Health care expenditures represent only part of the economic value of improving air quality. The AQVM estimates total economic value. Total economic value includes quality of life estimates. Any reporting of health care expenditures by the OMA should also make reference to the need to consider quality of life issues and the associated total economic value estimates provided by the AQVM.

 
3. Health statistics are faceless and are tied to ambient and episodic air quality. Quality of life issues, including premature mortality, relate to the individual and the nature and duration of their life and the environment to which they are exposed. Health history profiles for representative "statistical patients" need to be constructed from epidemiological data, provincial health care data and discussions with clinical practitioners. Quality of life issues, including premature mortality, become most compelling when considered at the individual level. These profiles are recommended to be used for communication purposes, not for quantitative analysis.

This study sets out a logical sequence of action items for the OMA to pursue. The specific steps recommended are as follows.

a) Provincial health care expenditure data for relevant cardio-respiratory illnesses associated with air pollution need to be compiled, analyzed and interpreted.

b) Unit health care expenditure estimates for each illness/treatment category reported by the AQVM need to be developed based on a full cost accounting framework. The estimates of unit expenditures should include allowance for:

i) labour (e.g., physicians, nurses, hospital administration),

ii) consumable (e.g., drugs), and

iii) fixed assets (e.g., hospitals, doctors' offices, medical equipment).

c) Treatment profiles for each illness/treatment category should be developed in consultation with clinical practitioners. These treatment profiles should be used to ensure that the unit health care expenditure estimates reasonably capture all related costs for treating each type of air quality-induced illness.

d) A comprehensive inventory of forecasts of air quality-induced health effects generated from the AQVM should be compiled. The basis for each forecast should be scrutinized for accuracy and compatibility with the assumptions underlying the unit health care expenditure estimates proposed to be developed. On this basis, the most appropriate forecasts of health effects for estimating health care expenditures should be selected for further analysis.

e) Regional and provincial estimates of the potentially avoided health care expenditures with improved air quality should be prepared and publicized.

f) The effects of premature mortality should be extracted from the estimates of total expenditures and the nature and magnitude of this effect carefully examined. On the basis of this examination, consideration should be given to what additional courses of action are necessary to accurately communicate the economic losses due to premature mortality.

g) Health histories covering the life of representative "statistical patients" expected to suffer from cardio-respiratory illnesses induced by air pollution should be prepared in consultation with clinical practitioners. These health histories should be consistent with the results of leading epidemiological studies and should be used to communicate to the public the acute and chronic nature of air quality-induced health effects.

1.1 Background

The Ontario Medical Association (OMA) has taken a strong stand advocating reductions in air pollution in Ontario. In 1998 the Association released a position paper, "Health Effects of Ground Level Ozone, Acid Aerosols and Particulate Matter"1

in which they laid out their position:

1. More stringent sulphur and nitrogen oxide emission limits should be enacted, including a province-wide SO2 reduction of 75 per cent from current cap levels, and a maximum allowable NOx emission limit of 6,000 tonnes annually from Ontario Hydro.

2. New transportation sector emission limits should include California-level standards for light and heavy-duty vehicles, reductions from off-road engines, an expanded vehicle inspection and maintenance program, and tougher standards for sulphur-in-fuel content.

3. The USEPA Administrator should be petitioned under Section 115 of the U.S. Clean Air Act to require reductions in U.S. emissions of SO2 and NOx which damage the health of Canadian residents and their environment.

4. Physicians should advise patients about the risks of smog exposure, should support more health effects research on air pollution, and should advocate the development of air pollution-related health education materials.

The OMA retained DSS Management Consultants Inc. to provide recommendations on how best to produce reliable estimates of the health care impacts of air pollution. The purpose of the overall project is to present government with a technically sound analysis of the health care resource utilization and associated expenditures which could be avoided with improvements to air quality policies. Expressing these air quality-induced health effects in terms of resource utilization and expenditures is designed to make a compelling "business case" (i.e., clearly demonstrating the direct financial savings) for improving air quality policies immediately.

This study examines the feasibility of generating reliable health care impacts. Specifically, the study included an examination of the current state of knowledge and practice in relation to air quality modelling, forecasting of air quality-induced health effects and the economic assessment of these effects. The study focused on Ontario, however, the potential for expanding the study to other provinces and territories in Canada was a consideration.

1.3 Methodology

This section reviews the steps taken in arriving at the conclusions and recommendations contained in this report.

1.3.1 Policy Analysis Framework

Effectively influencing government air quality policy requires an understanding of how alternative policies are evaluated and decisions on a preferred policy are reached. The process for reaching these decisions differs somewhat from that followed in some other areas of public policy such as health and safety. Environmental policy typically affects a well defined group of "polluters" and often, a much less well defined and identifiable group of "pollutees".

Polluters are typically private enterprises. Although in the case of some air pollutants, private vehicles are a major source of emissions. More stringent environmental policies usually imply a loss in private profit, or at least an encumbrance on normal business practices. As a result, polluters tend to resist policy changes by presenting to government, extensive analyses of the economic consequences for their operations, workers, taxes, industry-based communities and related sectors of the economy. These consequences are based on "hard" economic data (i.e., financial information).

Pollutees are less able to present their case due to:

1. lack of a pre-existing organizational structure comparable to private corporations,
2. limited financial resources (unlike private corporations which stand to receive future revenues to offset the costs of presenting their case), and
3. many of the environmental effects being suffered are not readily quantifiable and are not associated directly to conventional economic measures (e.g., changes in quality of life).

Government decision-makers are faced with balancing the conflicting interests of polluters and pollutees in arriving at environmental policy decisions. An advantage of economics in this context is that monetary values provide a consistent set of widely used, commensurate units of value that can be added and subtracted to arrive at a "bottom line". As well, the fiscal environment faced by governments has become increasingly more severe, demanding a more business-like approach to policy analysis. It is within this policy analysis and decision-making environment that this study has been undertaken.

Underlying the concept of balancing competing interests and making trade-offs is the idea of a range of choices being available; what are referred to as policy alternatives or scenarios. The standard process for analyzing policy scenarios is illustrated in Figure 1.

Each policy scenario consists of a series of actions (e.g., laws, regulations, taxes, rules, guidelines) that are intended to result in a specific change in environmental quality. These changes typically will yield certain benefits (e.g., reduced health effects). As well, there likely will be costs associated with implementing policy changes.

Policy analysis involves developing a comprehensive, technically sound and balanced assessment of these benefits and costs. The assessment of health care impacts is one element in this policy analysis process. The results of any assessment of health care impacts produced by the OMA will be most effective if they are consistent with and readily complement such a policy analysis process. This study has approached this assessment of health care impacts from the overall policy analysis perspective set out in Figure 1.

1.3.2 Forecasting Impacts

The policy analysis process discussed in Section 1.3.1 is based on the need to forecast or predict the consequences (e.g., changes in air quality and associated health effects) of alternative policies. This need is represented by the boxes in Figure 1 entitled "Implications". Forecasts need to be prepared for a range of factors including pollutant sources (e.g., future rates of emissions, effects of pollution control technologies, control costs), air quality patterns (e.g., ambient and episodic concentrations of pollutants, chemical transformations, atmospheric transport), human exposure, health responses, etc.

Figure 2 illustrates a generic forecasting process designed to predict health effects, health care resource utilization and health care expenditures. For the OMA to generate reliable health care impacts, each of the major elements in this process needs to be addressed. This study has examined the current state of knowledge with respect to each of these elements and has made recommendations as to how best the OMA can satisfy these requirements.

1.3.3 Chronology of Investigation

DSS fulfilled these study requirements by following a logical chronology of tasks. Specifically, the order of tasks was as follows:

i) Review of Existing Forecasting Models and Forecasts

ii) Examination of Supporting Scientific Literature

iii) Investigation of Data Requirements and Availability

iv) Expert Review Workshop

Review of Existing Models and Forecasts

Over the last several decades, considerable effort has been spent examining the environmental effects of air pollution. The acid rain controversy first galvanized broad-based public attention on the effects of air pollution and the long range transport of air pollutants. Much effort has since been invested in analyzing air quality policy alternatives. These efforts have given rise to the need for predictive tools (i.e., models) to assess the implications of policy alternatives.

The first task was to review recent air quality policy studies in Canada and the U.S. and to determine the models used to forecast health effects. As part of this task, the air quality and health effects forecasts developed were examined to determine the applicability of the methodology as well as the forecast results.

This task resulted in a good foundation for examining the current scientific literature and for assessing the availability of Ontario-specific data to undertake similar analyses or to expand and refine existing analyses.

Supporting Scientific Literature

The science of atmospheric chemistry and physics is well beyond the expertise and area of concentration of the OMA. Early on in the study, it was decided that a detailed review of the literature in this field would not be undertaken. Instead, the focus was on determining the current situation in terms of available air quality forecasting models and the sources and magnitude of the uncertainty in any forecasts which might be produced. Emphasis was placed on air quality models which had been used most recently for policy analysis purposes in Ontario.

The epidemiological literature relating to air quality-induced health effects has expanded substantially over the last two decades. This study did not attempt to compile a comprehensive review of this primary literature. Instead, numerous reviews of the literature have been conducted and the results incorporated into predictive health effects models. As well, Environment Canada maintains an extensive environmental information database which contains much of the primary literature. This study examined these secondary sources. As well, ongoing discussions among, and reviews by, various experts regarding the application of these studies for policy analysis purposes was considered.

The final area of interest involved the economic valuation of health effects and health care expenditures. Relevant review studies for both the U.S. and Canada were available. Most of this literature, however, deals primarily with the economics of morbidity and mortality in terms of total economic value. Comprehensive quantitative forecasts of health care impacts and expenditures related to air pollution for Ontario or Canada were not found. As a result, this component of the literature review was quite limited. Most of the recommendations in this regard were based on the experience and knowledge of the project team rather than the published literature.

Data Requirements and Availability

Developing Ontario-based estimates of health care impacts requires supporting health care data. This task involved examining the nature of the available provincial data and matching what was available with what was required. This review also extended to other provincial databases, particularly with respect to the data needs for analyzing the special case of premature mortality (see Section 4.3.2 for further explanation) and developing health history profiles for representative statistical patients (see Section 4.4 for further information).

This task did not entail acquiring the data themselves. Instead, the extent of the data and the specific details recorded in each database were assessed.

Expert Review Workshop

The final task in the study was to review the preliminary results of our investigations with experts in i) the fields of air quality modelling, ii) forecasting air quality-induced health effects and iii) health economics. The workshop provided an opportunity for the participants to modify and expand on the study results. The feedback provided through this workshop, shaped the final recommendations presented in this report.

2.0 AIR QUALITY FORECASTING

The health effects of interest in this study are those induced by air pollution. Figure 1  illustrates the need to consider a range of policy alternatives in arriving at a preferred strategy to improve air quality. Figure 2  illustrates how air quality forecasts play an important role in estimating health effects. This section examines the current situation in terms of air quality forecasting and the implications for generating reliable estimates of health care resource utilization.

Air quality modelling requires consideration of:

a) the geographic location of pollutant sources

b) pollutant types

c) emission rates and patterns

d) atmospheric diffusion and transport

e) chemical transformations and combinations

f) deposition and/or degradation

Government air quality policy options are targeted primarily at reducing emission rates. However, to evaluate the potential benefits of policy alternatives, reduced emission rates must be interpreted in relation to changes in exposure to air pollutants by the population and the natural environment. Air quality modelling is the science used to estimate the effects of changes in emission rates on ambient air quality.

2.1 Primary Conclusions

The following primary conclusions arise from the assessment, presented in this section, of air quality modelling in Canada generally, and Ontario specifically.

1. Developing reliable air quality forecasts is a demanding, technical undertaking far outside the expertise of the OMA.

 
2. Many reliable forecasts for a variety of air pollutants are available, primarily from government-sponsored studies and initiatives.

 
3. Additional improvements in air quality forecasts will be forthcoming in the near future, specifically in regards to particulates.

 
4. The degree of uncertainty in making air quality forecasts varies significantly among pollutants. The lowest uncertainty has been realized with SOx and NOx. Higher uncertainty is involved in forecasting ozone, particulates and aerosol ammonia.

Modelling of air quality is a demanding exercise for a variety reasons, including:

1. a three-dimensional medium

 
2. complex circulation patterns exhibiting highly variable, short-term, stochastic behaviour

multiple (and often variable) emission sources, rates, mixtures of pollutants and potential emission control efficiencies and costs
 
3. long residency time and transport distances of some pollutants

complex atmospheric chemical reactions driven by multiple factors, including the presence and concentration of other pollutants, sunlight, humidity, temperature, etc.

Much progress has been made in accurately measuring air pollutants and modelling air quality. Nonetheless, the complexity of the problem remains. Generating reliable air quality forecasts is a highly demanding task requiring technical expertise in mathematical modelling, atmospheric chemistry and related skills.

In the 1990's, multiple forecasts of air quality involving ground-level ozone, particulate matter (PM) and acid aerosols have been produced for Ontario. These forecasts are the foundation for various estimates of health effects and costs, as well as, for recommendations for changes to government air quality policy, regulations and standards. Table 1 provides a list of studies and reports for the more comprehensive forecasts which have been produced in the 1990's which deal exclusively or partially with Ontario. As well, much effort is currently underway to improve the resolution and reliability of these forecasts, with new forecasts being generated on a regular basis. In particular, much effort is being expended to develop improved estimates of inhalable (i.e., PM10) and respirable (i.e., PM2.5) ambient air concentrations.

Table 1 - Listing of Primary Air Quality Forecasts for Ontario in the 1990's
 

Study Name	Date	Types of Pollutants	Geographic Scope
Environmental and Health Benefits of Cleaner Vehicles and Fuels	1995	Ozone, PM10, Toxics	Canada, by province
Human Health Benefits of Sulfate Reductions under Title IV of the 1990 Clean Air Act Amendments	1995	SO2	Eastern Canada, US
Towards a Smog Plan for Ontario	1996	VOC, NOx, IP	Ontario
Benefits and Costs of the Clean Air Act, 1970 to 1990	1996	TP (total particulates), NOx, VOC, SOx, CO, toxics, lead	Eastern Canada, US
Hamilton Wentworth Air Quality Initiative	1997	Ozone, toxics, acid aerosols, SOx, NOx, PM	Hamilton Wentworth
Canada-wide Standard for Particulates	Ongoing	PM10 PM2.5	Canada, by province
Compendium on Inhalable Particulates and Respirable Particulates	Early 1999	PM10 PM2.5	Ontario

2.3 Key Issue

The OMA is not expert in modelling air quality. Most likely the OMA will rely on existing air quality forecasts developed by other organizations. Nonetheless, the OMA must understand the limitations of existing forecasts and ongoing efforts to overcome these limitations and must choose a strategy as to how best to use existing air quality forecasts.

Uncertainty in air quality forecasts has a cascading influence. All aspects of the health care resource utilization estimation procedure are affected. Uncertainty in terms of the air pollution exposure levels (and how they might change with changes in emissions sources and rates) leads to uncertainty as to the potential benefits of more stringent environmental policies.

The types and importance of uncertainty in air quality forecasts were discussed at the expert review workshop with air quality modelling experts. The conclusions arising from these discussions are as follow:

1) The uncertainty in air quality modelling is not constant but varies significantly depending on the chemical being modeled.

2) The acid aerosol (i.e., NO_x, SO_x) portion is relatively well understood and can be accurately modeled.

3) The ability to model ground-level ozone is good and is expected to improve significantly in the future.

4) The ability to model particulate matter depends greatly on the fraction of interest. Concerns with respirable particulates (i.e., PM_2.5) have recently emerged. Prior to this time, air quality monitoring data did not distinguish on the basis of this smaller fraction. As a result, limited data are available to model this fraction.

5) Other pollutants, like ammonia, are poorly understood and predictive capabilities are limited.

Areas targeted for further study include:

1) More monitoring data, especially in rural areas - Expanding the rural ambient air quality monitoring network is needed to provide data against which to compare urban air quality. Better estimates for pollutants both locally and regionally could be achieved with the availability of improved rural air quality monitoring data. Better rural monitoring would also improve understanding of the role of "fugitive" (i.e., mobile, inconsistent, and/or non-point) sources of air pollution.

2) Improved emissions inventories - Current inventories are limited for some pollutants, especially those originating from fugitive sources (e.g., particulate matter). Improved understanding will involve identifying the sources of these pollutants and the relative contribution of each to ambient air quality conditions.

3) Increased research into atmospheric chemical reactions - Synergistic reactions, the generation of secondary aerosols, and certain basic atmospheric chemical reactions are poorly understood. Much gain in understanding has been achieved over the last several decades. Considerable potential to improve this understanding remains.

Efforts to improve air quality modelling are currently underway world-wide through government and industry-sponsored research at university, government and private sector institutions. Environment Canada and the National Research Council are cooperating with the U.S. EPA. This joint initiative is developing improved models for application to Canada.²

These investments will improve the ability to forecast changes in ambient air quality as a result of potential policy initiatives. As is the case with research of this nature, progress is slow, particularly for the more poorly understood pollutants. Sustained support for this research is important over the long term.

In the meantime, ambient air quality forecasts based on the best available knowledge are available for Ontario for the major pollutants of concern. These forecasts provide the best available basis for estimating health effects.

Particular attention needs to be given to ensuring that the uncertainty in these forecasts is reflected in any related forecasts of health care impacts.

2.5 Summary

1) Air pollution modelling involves predicting the behaviour of a complex natural system exhibiting stochastic characteristics.

2) As well, our understanding of the physical and chemical behaviour of the atmosphere is limited.

3) This level of knowledge varies significantly among pollutants.

4) Improvements in the level of knowledge will be achieved in the future, given current investments in research being made in Canada and worldwide.

5) Nonetheless, given the nature of the atmosphere, significant uncertainty will continue to be a reality in forecasting ambient air quality for the foreseeable future.

6) Accordingly, estimates of health care impacts relying on air quality forecasts should reflect this uncertainty.

7) Forecasts of ambient air quality are available for various pollutants from reputable sources, primarily government.

8) These forecasts reflect the current state of the science and are a suitable basis for making forecasts of health care effects.

9) Forecasts of health care impacts need to be periodically updated as significant improvements in air quality forecasting abilities are achieved.

3.0 HEALTH EFFECTS

This study focuses on cardio-respiratory illnesses induced by air pollution. The primary contributing air pollutants are ground-level ozone, particulates and acid aerosols (derived from SO_x and NO_x). This section examines the scientific evidence supporting a direct linkage between air quality and health effects.

The focus of the discussion is on the epidemiological evidence and how that evidence should be applied in a policy analysis context. In particular, the key question to be answered is:

Is the epidemiological evidence adequate to develop quantitative concentration-response functions for the primary health effects induced by air pollution?

The following primary conclusions arise from the assessment of the epidemiological evidence relating to air pollution and human health effects presented in this section.

1) Populations are exposed to a complex mixture (i.e., "soup") of air pollutants that in combination, cause highly predictable human health effects.

2) The statistical strength of these concentration-response functions varies among pollutants and among health effects.

3) These concentration-response functions have been used repeatedly in Canada, the U.S., Europe and elsewhere as a primary basis for making environmental policy decisions.

4) Some interests (primarily representatives of industries potentially affected by more stringent air quality regulations) continue to note uncertainties in the epidemiological evidence. These uncertainties relate to:

i) correlation does not ensure causality,

ii) failure to demonstrate comparable effects in clinical studies for some types of air pollutants at observed concentrations,

iii) failure to distinguish the precise contributory role of each pollutant in the air quality "soup" to each health effect,

iv) potential for health effect thresholds below which no effects occur (or at least cannot be practically detected),

v) failure to account for variations in the risk factors among populations, and

vi) other potential confounding factors (e.g., temperature/humidity/seasonality interactions, indoor vs outdoor air quality, genetic predisposition).

5) These uncertainties are not cause for inaction and have been addressed to a large extent through the use of well-designed statistical analyses.

6) Given the inherent nature of environmental problems, a high level of uncertainty is inescapable. These uncertainties are cause for prudence and a clear declaration of uncertainty in making predictions of air quality-induced health effects.

7) The Air Quality Valuation Model (AQVM), developed jointly by Health Canada and Environment Canada, is the most comprehensive model of its kind in Canada and is a reasonable means to forecast air quality-induced human health effects.

Air pollution consists of a complex mixture (what is commonly referred to as a "soup") of pollutants. These pollutants vary in concentration and physical and chemical characteristics. As well, they undergo chemical, and even physical, transformations in the atmosphere. These physical and chemical features are constantly changing, sometimes rapidly over a large range in variation.

As a result of these factors, the outdoor air quality environment to which individuals are exposed is difficult to characterize, not to mention to replicate in a controlled laboratory environment. This complexity creates an important barrier for the application of the conventional scientific method (i.e., testing cause-effect hypotheses using controlled experiments) in researching the health effects of air pollution. If the conditions to be tested cannot be readily characterized (let alone replicated), setting up conventional experiments involving treatments and controls is not possible. However, the complexity of outdoor air quality is only one of several complicating factors in understanding these relationships.

The susceptibility of individuals in a population to many health effects is known to vary considerably. Some individuals with high risk factors will react more severely to lower concentrations of pollutants than others will to higher concentrations. Likewise, these risk factors differ among various types of health effects. These risk factors include personal health habits (e.g., eating, smoking, drinking), individual health history, family health history, occupation and personal activity patterns. The role of factors like personal activity patterns and occupation, often must be considered over the course of many years where chronic illness or cumulative exposure are at issue.

Environmental policies are designed to affect the air quality enjoyed by the entire population. To develop reasonable estimates of the benefits of policy alternatives, the health effects for the entire exposed population must be considered. In other words, approaching the problem using the conventional scientific method, one would need to stratify the "treatments" to ensure a reasonably representative cross-section of the personal habits and history of similar groups in the population. The results of these controlled experiments would then need to be extrapolated to the entire population to produce the information necessary for analyzing alternative environmental policies. Doing so is prohibitively complex and costly.

A final complexity is the relationship between outdoor and indoor air quality. Canadians spend the majority of their life indoors. In this environment, they are exposed often to a quite different "soup" of air pollutants than is present outdoors. Additionally, indoor air quality differs substantially from building to building. Lastly, no long-term, broadly representative monitoring data for indoor air quality (comparable to that available for outdoor air quality) are available. Setting up rigorous experiments controlling for outdoor and indoor air quality is essentially impossible.

These complexities are common to real world environmental and public health problems. Several conclusions can be reached given this reality.

i) The conventional scientific method of research is not adequate (at least exclusively) to develop practical relationships (i.e., dose/response functions) between air quality and human health effects.

ii) The range of effects will be significant within a large population given the range of individual differences.

iii) Reaching definitive conclusions as to the precise cause of illness for individual patients is difficult when environmental factors are contributory causes. Indeed, the contributory role of these environmental factors may only be detectable at the population level.

iv) A high degree of uncertainty is inescapable, particularly when higher levels of resolution are attempted (e.g., in terms of individual air pollutants, individual locations and individual groups of people).

Despite the underlying complexity and variation among the populations and air quality environments studied, a consistent trend in the results is evident. Repeatedly, it has been concluded that morbidity and mortality are being caused by air pollutants at the levels regularly encountered in Canada. While variations in the exact value of the dose-response coefficients are evident, these studies consistently demonstrate a positive correlation between air quality and the incidence of health effects.

These positive relationships are statistically stronger for certain air pollutants and certain human health effects. The strength of the relationships depends on:

1. the quality and length of the supporting air quality monitoring record, and

 
2. the nature of the related health effect (e.g., ease of diagnosis, severity, required treatment, time lag between exposure and response, significance of co-morbidity/co-mortality factors).

The effects of particulate matter are not so clear. The chemistry of particulates, as well as their physical dimensions, likely play an important role. As a result, correlating measurements of total particulate matter with health effects may not capture all of the subtlies of the underlying dose-response relationships of each particulate fraction. Considerable work is currently underway to differentiate among the various constituents of particulate matter and their relative contribution to observed health effects. Nonetheless, a positive relationship between particulate matter concentration and health effects, especially in urban environments where the particulate matter is primarily from anthropogenic sources, has been repeatedly observed. These observations have given rise to an initiative to develop a Canada- wide standard for air-borne particulate matter.

The epidemiological literature has advanced to the point that quantitative dose-response coefficients for human health effects have been developed for a range of air pollutants and for a range of human health effects. These dose-response relationships have been used repeatedly to evaluate the implications of air quality policy alternatives. In particular, Health Canada and Environment Canada funded the development of an Air Quality Valuation Model (AQVM) which relies on the best available epidemiological evidence to forecast air pollution-induced health effects. Appendix C provides a review of the structure and contents of the AQVM. The AQVM is now the standard tool for developing health effects forecasts for air quality policy alternatives in Canada. This model has been used at federal and provincial levels. Similar models have been developed and applied for policy analysis in the U.S. and Europe.

3.3 Key Issues

This section examines important uncertainties associated with predicting air quality-induced health effects.

3.3.1 Correlation and Causality

The refrain, correlation does not mean causality, is drummed into students from the first day they are exposed to the principles and theory of statistical correlation techniques. The reason for this refrain is that the actual underlying causal factor(s) may exhibit a collinear pattern of behaviour with the dependent variables being analyzed. If this is the case, an observed correlation may appear to be causal but in fact, is an artifact of collinearity.

Epidemiologists face this challenge daily. While causality can never be conclusively proven, the potential for confounding factors (i.e., collinearity artifacts) can be minimized by well designed statistical strategies. These involve searching and testing for confounding factors. Indeed, for epidemiological studies to be used in the AQVM, one of the requirements is a well structured statistical design that controls for known or suspected confounding factors.

Another strategy is to replicate the statistical analysis using multiple independent data sets for independent populations. In other words, as the number of independent studies reporting similar results increases, confidence increases that observed relationships are in fact causal, and not simply coincidental correlations. The current epidemiological literature consists of numerous independent studies demonstrating similar responses by separate populations exposed to air pollution.

Causality can never be proven beyond a shadow of doubt in any field of science. However, the epidemiological literature has been repeatedly examined by panels of experts, independent researchers and others. Attention is commonly drawn to underlying uncertainties, but time and again, the overriding conclusion reached has been that air pollution is causing morbidity and mortality. There is no comparable scientific literature supporting a contrary conclusion.

3.3.2 Need for Corroborating Clinical Studies

The inherent complexities in analysing the health effects of air pollution on a large population are discussed in Section 3.3. Despite the compelling epidemiological evidence that hundreds of Canadians are suffering premature mortality due to air pollution, not one death certificate has been issued listing air pollution as the cause of death. Why not?

The answer is simple in a sense. For the same reasons that the conventional scientific method cannot deal with the complexity of many environmental problems, so to is it impossible to isolate at the individual patient level the contributory role of air pollution. Population-wide environmental health effects are rarely attributable to only one causal agent. Multiple causal agents are commonly the case. This reality presents problems when demands are made for corroborating clinical evidence.

With some air pollutants like ozone, a consistent physiological response can been induced through clinical tests. These responses can be tied to one causal agent, are easily measured, occur more or less immediately, are reversible and are non-life threatening at low dosages. However, when health effects do not exhibit all of these features, the utility of clinical tests becomes limited.

Certainly clinical evidence at the individual patient level will continue to provide important insights into some types of environmental-related health effects. However, if corroborating clinical evidence is a prerequisite for improved environmental policies, few initiatives could be defended on scientific grounds. For the foreseeable future, epidemiology and the analysis of population-wide health responses will continue to be the leading investigative technique for detecting and quantifying environment-related health effects.

3.3.3 Role of Each Pollutant

Individuals are exposed daily to a complex "soup" of air pollutants. This "soup" changes as one moves from location to location. Epidemiological studies typically choose one air pollutant as the independent variable against which to correlate observed health effects. Revealed correlations are attributed to the air pollutant being examined. But what role did other constituents of the "soup" play? Do the effects of different combinations of pollutants and concentrations result in significantly different responses?

These are the types of questions that epidemiologists regularly face in developing rigorous study designs. Appropriate data sets and statistical analyses must be used to reduce the potential for confounding factors and to reveal the complexities of the interactions among multiple independent variables. The full range of potential synergistic effects of complex mixtures cannot be practical investigated. Nonetheless, well designed epidemiological studies can minimise the role of confounding factors and can extract the highest level of information possible from a given data set.

Two considerations flow from these observations.

i) Every reasonable effort should be made to use only high quality, epidemiological studies that meet minimum criteria in terms of study design. (Appendix C describes the criteria which are use with the AQVM to screen candidate epidemiological studies.)

ii) Efforts should continue to be made to improve the environmental and health effects data on which these epidemiological studies rely. Ultimately, the resolution and reliability of any statistical analysis rests on the quality of the supporting data.

3.3.4 Health Effects Thresholds

The concept of thresholds is based on the notion that humans have a certain level of resilience to certain pollutants. Indeed in the case of substances like vitamins, a certain minimum supply is vital.. On the other hand, an excess can be fatal in some cases. These response levels represent a type of threshold.

Pollutant thresholds are important from an environmental policy perspective. If a threshold does in fact exist, an investments to improve environmental quality below the threshold level will yield no benefit in terms of reducing any related health effects. Often efforts have been directed to detecting specific threshold levels as a basis for setting ambient environmental standards.

The existing epidemiological evidence does not indicate the presence of a threshold for the air pollutants being examined in this study. As well, the question of thresholds has been reviewed in the literature and by expert panels. The majority view is that no threshold should be assumed. This view is consistent with the precautionary principle which is widely applied in environmental issues. Nonetheless, the relative importance of potential thresholds for key findings like health care impacts should be explored. Having knowledge of the potential magnitude of this effect would be helpful in making policy decisions.

3.3.5 Population Risk Factors

Risk factors for environment-related health effects are known to vary substantially within a population. Epidemiological studies often stratify sample populations into risk factor groups. The practical level of stratification depends on the resolution available in the population database. As well, if the results are to be applied to another population sufficient resolution to characterise the frequency of these strata must be available. Stratification is an effective statistical strategy to improve the resolution and to reduce the uncertainty of forecasts when adequate data are available.

Stratification will only reduce uncertainty where significant differences are expected or observed among populations of interest. The epidemiological evidence does not suggest this to be the case. Sample populations have been analysed from diverse locations, yet, similar trends in air quality induced health effects have been observed. Certainly the resolution of health effects forecasts will be improved in the future with greater stratification for risk factor groups. When such results are available, they should be used for policy analysis. In the mean time, variations in risk factor groups with populations do not appear to be a major factor.

3.3.6 Other Confounding Factors

The possibility of unsuspected and unknown confounding factors can never be totally eliminated in epidemiological studies. The probability, however, can be minimised by well designed statistical analyses. Considerable care needs to be exercised in relying only on the best available epidemiological evidence. As well, improvements in the evidentiary record will be made over time. Forecasts of health impacts need to be periodically updated to reflect these developments. However, the unsuspected and the unknown should not be cause for inaction but cause for prudence, care in knowing all underlying assumptions and limitations, and a full exploration of the implications of uncertainty in terms of the forecasts being made.

3.4 Recommended Strategy

The results of epidemiological studies will continue to be the primary basis to understand and predict environmental related health effects for the foreseeable future. The results of these studies have inherent weaknesses that can be minimized but never eliminated. Accordingly prudent use of these epidemiological results is warranted.

Care must be taken to screen epidemiological studies to ensure only the most reliable and relevant results are used to develop formats of health effects. A well conceived set of screening criteria have been developed and applied in selecting the epidemiological foundations for the AQVM. For this reason among others, the AQVM is considered a useful and reliable tool to develop fundamentals of air quality, induced health effects.

Epidemiological research is ongoing. New results are frequently being published. For this reason, forecasts of health effects need to be revisited periodically and updated accordingly. The nature and rate of production of new epidemiological evidence should dictate the frequency for re-analysis.

Clinical tests provide valuable corroboration of epidemiological results. When the results of appropriate tests are available, they should be used to refine concentration-response coefficients. The availability of supporting clinical evidence, however, should be a prerequisite for making health effects forecasts

Uncertainty is a fact of life in developing environmental policies. This underlying uncertainty needs to be reflected in forecasts of air quality-induced health effects. The AQVM includes a rigorous, quantitative procedure to estimate the effects of uncertainty on health effects. This procedure is technically sound and useful for analyzing alternative policies. The results of this uncertainty analysis in any forecasts of the health care impacts will need to be incorporated.

4.0 HEALTH CARE UTILIZATION AND EXPENDITURES

The OMA is examining the impacts of air pollution on public health and the health care system. Impacts on the health care system involve increased resource utilization and associated expenditures which are related to treating air quality-induced illnesses.

This section assesses the current information available on health care system impacts and develops an economic framework appropriate for examining health care resource utilization impacts. The data requirements and their availability are also examined.

4.1 Primary Conclusions

The following primary conclusions arise from this assessment of the required economic foundation for generating reasonable health care expenditure estimates. The details of this assessment are presented in the following subsections.

1. The majority of the economic data being used in current analyses comes from foreign-based studies, largely from the U.S.

 
2. Health care expenditure data from one jurisdiction (e.g., the U.S.) are unlikely to represent accurately expenditures in another jurisdiction (e.g., Ontario), particularly when the economic foundations for the health care systems differ substantially.

 
3. Quality-of-life and value-of-life economic measures have accounted for much of the focus of recent economic analyses of air pollution in Canada.

 
4. Little use has been made of comprehensive provincial health care, data to develop estimates of the economic value of air quality-induced health effects.

 
5. The Air Quality Valuation Model (AQVM) is a useful tool to provide estimates of air quality- induced health effects. The AQVM results are a reliable basis on which to develop reasonable estimates of health care resource utilization and associated expenditures.

 
6. Developing reliable expenditure estimates will require assigning appropriate clinical treatment procedures to each category of predicted air quality-induced illness. The health care expenditures related to each treatment procedure need to be developed from the existing health care databases.

 
7. The potential impact of premature mortality requires special consideration. Premature mortality could, under certain circumstances, lead to a reduction in health care expenditures.

 
8. On the other hand, premature mortality is a major factor in the total economic value estimates produced by the AQVM. Care needs to be taken to coordinate the reporting of health care impacts with impacts on total economic value.

4.2 Current Situation

Illnesses caused by air pollution can cause morbidity and/or mortality. The nature of the induced illness will determine the medical care requirements of the patient and the associated costs, including both direct health care expenditures and losses due to reduced quality of life and/or length of life. Using economic measures to value human health impacts has become a widely accepted practice. Considerable expertise and experience in Canada and elsewhere has developed in applying economic techniques to assist in environmental policy decisions. Most assessments of air quality initiatives now include estimates of the economic value of health effects.

Estimates of economic damages are often dominated by the high value assigned to mortality (e.g., up to 90% of the total economic value). Economists differ on the best data and best technique(s) to estimate the value of life. A large literature exists which examines these issues in detail. The primary source of controversy with respect to estimates of economic damages is the appropriate economic measures for quality of life and premature mortality. Given the chosen focus of the OMA on health care system impacts, the economics of measuring quality of life and value of life are not an issue. Accordingly, the economic arguments relating to these non-market³

losses are not discussed further in this report. Nonetheless, from a policy analysis perspective, the need to consider total economic value is clear. Estimates of health care impacts capture a significant portion of the financial (i.e., market) losses caused by air pollution, but are not a substitute for total economic value. Health care impact assessments are complementary to total economic value assessments and provide greater insight into this component of total economic value.

The AQVM has been the primary analytical tool used recently to generate estimates of the economic value of air quality-induced health effects in Ontario. The objective of the AQVM is to generate estimates of total economic value. Total economic value includes direct expenditures plus the value of damages not reflected directly in financial transactions. However, the economic coefficients in the AQVM do not separate out the direct expenditures component. As a result, it is not possible to generate directly estimates of the health care impacts using the AQVM.

As well, the monetary values in the AQVM come largely from U.S.-based studies. While standard conversions from U.S. currency to Canadian currency have been performed, U.S.-based medical costs do not accurately reflect Canadian medical costs. These differences are due to differences in prescribed treatments for certain illnesses, different accounting principles and the significant difference in the role of government in providing health care services.

Some attention has been given recently to improving estimates of Canadian health care expenditures. Nonetheless, the AQVM does not make full use of the extensive health care data available for Canada. In its current form, the AQVM does not forecast health care impacts in terms of resource utilization categories or associated health care expenditures.

A major objective of this study is to examine whether (and if so, how) complementary health care data could be used in combination with the AQVM.

4.3 Key Issues

This section examines the key issues which need to be addressed in developing reliable health care expenditure estimates.

4.3.1 Guiding Economic Logic

The objective, from a policy analysis perspective, is to develop reliable estimates of the health care resource utilization attributable to air pollution. The attributable portion consists of those resources which are consumed above and beyond the "base case" (i.e., the level of resource utilization which would be expected to occur at a lower level of air pollution). This concept is illustrated in Figure 3. The attributable expenditures are the difference between the two cases (i.e., the area lying between the two expenditure lines in Figure 3).

Ideally, forecasts of health care resource utilization would be based on the treatment history of individual patients in the population over the entire course of their lives. This approach would allow inclusion of all health care utilization impacts to be included (e.g., diagnostic, treatment and post- treatment expenditures for acute and chronic illnesses). The total impact would be estimated by aggregating these results for individual patients across the entire population exposed to air pollution.

This economic framework is difficult to implement due to limits in scientific knowledge relating to i) the effects of different air pollutants and combinations of pollutants on individual patients, ii) personal exposure to air pollutants and iii) variations in health responses among individuals to different air pollutants. As well, the cost of undertaking such a detailed analysis is prohibitive given the data and analytical requirements. Nonetheless, the framework is a valuable reference point for judging compromises in the analytical methodology practically necessary to produce reasonable health care expenditure impacts.
 

4.3.2 Co-morbidity/Co-mortality

The epidemiologica l literature reports the proportion of morbidity and mortality attributable to a specific effect (in this case, air pollution). Statistical techniques are used to compare the responses of sample populations exposed to varying levels of air pollution. Once confounding factors are netted out, the remaining difference in morbidity and mortality is attributed to the independent variable (i.e., a certain air pollutant or combinations thereof).

As a result, the epidemiological literature provides estimates of health effects consistent with the economic framework discussed in the preceding section. In theory at least, estimating health care resource utilization and expenditures simply requires developing appropriate expenditure coefficients for each predicted illness category and patient category. In practice, several important underlying assumptions need to be considered.

The "attributable" segment of the population utilizing health care resources cannot be identified at an individual-patient level. For example, one cannot examine two chronic bronchitis sufferers and conclude that one case is attributable to air pollution and another is not. Indeed, the etiology of this disease and the known physiological effects of air pollutants lead to the conclusion that air pollution will be a contributing factor to the condition of most, if not all, chronic bronchitis sufferers. While those sufferers already stressed by other factors (e.g., smoking) may be less affected, multiple environmental factors (including air pollution) likely play a role, to one extent or another, in the health response pattern of individual patients.

Why is this a concern, if the epidemiological studies have identified the attributable sufferers of the illness? The significance relates to estimating appropriate expenditure factors for each air quality- induced illness category. For example, considerable range in health care resource utilization is evident among chronic bronchitis sufferers. More serious cases typically demand greater utilization of health care resources. This knowledge leads to the question, does air pollution contribute primarily to causing more serious cases, or less serious cases, or average severity cases, for each illness category?

This issue would not arise if the ideal economic framework could be implemented. The severity of the illness would be tied directly to a specific patient, their risk factors and health history. But this is not practical. Instead, what is known about the effects of air pollution must be used to arrive at the best practical estimates.

Accepting that air pollution acts as a co-morbidity/co-mortality factor in the majority (if not all) of the illness categories attributable to air pollution, then patients suffering all ranges of severity of an illness are affected. In other words, air pollution is a contributing factor in most, if not all, cases. The epidemiological literature describes the increase in total cases over a particular interval of time. But these attributable cases are not unique or distinguishable from the "normal" range of cases of a particular illness at an individual patient level. Certainly, these cases are not distinguishable below the level of resolution used in the medical records on which the epidemiological studies are based.

Given these observations, using the average representative resource utilization rates for each illness category is reasonable. This level of expenditure assumes that the range of severity of attributable cases of a particular illness is similar to the range for the "normal" cases. If future epidemiological studies provide finer resolution among illness categories in terms of severity of illness, appropriate adjustments to the expenditure estimates can be made. In the mean time, given the co-morbidity/co- mortality behaviour of air pollution, using average representative costs for each illness category is defensible.

4.3.3 Marginal vs Average Costs

Economic theory demands that policy decisions designed to improve efficiency should be based on an analysis of marginal values (i.e., for both benefits and costs). This principle is important in deciding on what health care resource utilization components should be attributed to a particular illness category. For example, say, air pollution is improved such that one less emergency room visit per year in Ontario is expected. Is it reasonable to assign a portion of the capital costs of building hospitals to this marginal reduction in resource utilization? The answer depends on two considerations, namely,

i)the combined magnitude of the resource utilization under consideration, and

ii)the time horizon for the analysis.

As the magnitude of the impact increases, a greater potential exists for there to be a strong influence on future decisions affecting the capacity of the health care system. In other words, if significant reductions in the demand for health care services are realized by improving air quality, this could lead eventually to downsizing of the health care infrastructure. While it is highly unlikely that, for example, some hospitals will be closed if air quality in improved (at least over the short term), the health effects attributable to air pollution are substantial and are not a negligible factor in examining health care resource utilization.

Regarding the influence of time horizon, if the results of this study were intended only to guide short- term policy decisions (e.g., in the extreme case, whether to improve air quality over a single month on a one-time basis), capital investments in hospitals, medical equipment, research, etc. would largely be viewed as sunk costs⁴.

In other words, the policy decision would not affect the levels of these investments. Therefore, including allowance for a potential capital cost savings as a benefit of the policy decision would be incorrect. Only the variable costs (e.g., consumed drugs) directly attributable to the change in air quality should be included. This hypothetical policy change, given its brief planning horizon, would not affect the level of investment in the health care system infrastructure.

On the other hand, if the types of policy decisions being analyzed have long-term consequences (i.e., if improvements in air quality are likely to be put into effect on an essentially permanent basis), all investments in the health care system are relevant and should be included. For example, if improved air quality is expected to reduce for the foreseeable future emergency room visits, over the long-term this reduction will be reflected in the emergency room capacity built in the future. Accordingly, a portion of these fixed capital costs is reasonably assigned as a health care expenditure attributable to air pollution-induced illnesses.

In summary, the types of impacts attributable to air pollution are of significant magnitude to affect health care expenditures at a local and provincial level. As well, the types of policy decisions under consideration involve essentially permanent, long-term changes. Both of these considerations argue for inclusion of variable and fixed expenditures in the resource utilization factors for each illness category. These allowances should be based on the long-term average costs⁵ of these health care system components.

4.3.4 Data Availability

The AQVM reports air quality-induced illnesses by 15 illness categories (see Appendix C). Health care resource utilization factors are required for each of these categories. In some cases, the illness categories are specific (e.g., chronic bronchitis, asthma). In others, the categories are more broad (e.g., respiratory hospital admissions, emergency room visits). These illness categories do not correspond directly in all cases with the standard disease codes used in provincial health care databases.

Several matters need to be resolved in this regard. The first is that the broad illness categories in the AQVM need to be tied to corresponding ICD9 codes if provincial health care data are to be used effectively. In most cases, the illness categories will correspond to multiple ICD9 codes. Judgement will need to be applied in arriving at the best means to integrate the data for multiple ICD9 codes such that a reasonable overall estimate of health care utilization requirements can be developed for each illness category.

A second matter relates to tying all of the health care utilization data together into a logical and representative treatment procedure for each illness category. Of particular importance is to ensure that non-specific diagnostic expenditures which often precede identification of a specific illness are captured in the expenditure estimates for each illness category. Developing standard sequences for treatment procedures for each illness category will demand the experience and knowledge of clinical practitioners. This is particularly so, if allowances for non-illness-specific diagnostic procedures are to be included.

Another consideration is the health care resource utilization categories to be used to organize the compilation and integration of expenditure data. These expenditures consist of three basic components:

1)Labour (e.g., doctors, nurses, support staff, administration),

2)Consumable (e.g., drugs, laboratory services), and

3)Fixed costs (e.g., hospitals, doctors' offices, medical equipment).

A final consideration relates to developing reasonable estimates of the full cost of these health care resources. A common problem in large organizations, particularly publicly funded organizations, is ensuring all costs are reasonably captured. For this reason, a full cost accounting framework needs to be adopted in developing these expenditure estimates. Adequate allowance for overhead and support services needs to be included. These support services should include education and training of clinical staff as well as medical research. These expenditures are in addition to conventional overhead allowances for operation and maintenance of infrastructure.

The data required to develop these expenditure estimates are available but not from one source, nor in a directly applicable form. Considerable effort will be required to compile and synthesize the data in a form suitable for the intended purpose.

4.3.5 Premature Mortality

Epidemiological studies have repeatedly demonstrated significant increases in mortality in association with air pollution. Premature mortality can complicate estimating the portion of health care expenditures attributable to air pollution. Figure 4 illustrates the effect of premature mortality. Two significant effects need to be considered.

The first is that the pattern of health care utilization is contracted in the case of premature mortality. The highest rate of utilization typically occurs immediately before death. Premature mortality caused by air pollution can cause these expenditures to occur sooner. Economic theory includes the time value of money.⁶

The time value of money is reflected by the discount rate.⁷

Applying this theory will result in the end-of-life health care expenditures for premature mortality being of higher present value⁸
than the corresponding expenditures for the base case (i.e., the premature mortality expenditures will be incurred sooner). The magnitude of the difference in present value will depend on the discount rate and on the difference in the length of time between the premature mortality occurrence and the base-case mortality occurrence. Considering this factor alone, premature mortality will result in an increase in the present value of these end-of-life expenditures.

Counteracting this effect is the potential for an overall reduction in the total health care utilization between the two cases. A shorter life often implies a lower total utilization of health care resources. This tendency could be a concern. If one considers only health care expenditures, premature mortality could appear to be a benefit (i.e., lower expenditures). In general, the earlier one dies, the less health care utilization.

Obviously, this conclusion ignores the value of life. It is for this reason that any reporting of changes in health care utilization should be accompanied by reference to total economic value and the need to consider quality of life and value of life in addition to direct health care expenditures.

The net effect of premature mortality on overall health care expenditures has not been quantitatively analyzed with respect to air pollution. As a result, whether or not an overall perverse relationship will be evident is unknown. An important aspect of any future study of health care resource utilization should be to isolate the impact of premature mortality and to determine the magnitude of this effect.

4.4 Recommended Strategy

Impacts of air pollution on health care resource utilization have not been prepared for Ontario or Canada. This recommended strategy is designed to produce suitable estimates of these impacts. A related aspect of the strategy involves communicating effectively to the public the underlying economic principles of this strategy and its limitations.

The OMA should start with the generic categories of health effects forecast by the AQVM. For each health effect category, standard treatment procedures need to be developed. These procedures should include allowances for diagnostic, treatment and post-treatment phases. The resources required by each procedure need to be itemized according to three broad accounting categories (i.e., labour, consumables and fixed assets). Costing fixed assets and support services need to be done using a full cost accounting framework.

Developing practical treatment procedures will require input from experience clinical practitioners. It is recommended that for each health effect category, experienced practitioners develop a range of treatment procedures depending on the severity of the individual illness. These procedures will be required for each of the specific illnesses expected to fall within the broad health effect categories used in the AQVM.

The epidemiological studies on which the AQVM relies should be reviewed to gain greater resolution as to the specific illnesses included in each health effect category and any related information on the severity of the subject cases. The results of these more detailed investigations should be combined with the standard treatment procedures and the unit health care resource requirements for each procedure to arrive at an aggregate estimate of the average health care impact of an incident for each health effect category. These aggregate estimates can be applied to any forecast of health effects based on the corresponding health effect categories.

The issues of premature mortality should be given special attention. The quantitative impact of premature mortality on health care resource utilization and associated expenditures should be examined separately from the effects of morbidity. To capture accurately the effects of premature mortality, health history profiles for patients experiencing premature and normal mortality will need to be constructed. Epidemiological results should be used as a primary basis for constructing these profiles. As well, input from clinical practitioners and provincial databases (which allow tracking of the treatment history of individual patients) should be used where practical.

Communicating the results of future analyses of health care resource utilization and associated expenditures should be done carefully for several reasons.

1. Health care resource utilization does not capture all of the economic value of air quality-induced illnesses. Future estimates will be important for understanding impacts on the health care system. Total economic value should also be a central measure on which to judge policy alternatives. The distinction between, and purpose of, these two economic measures should be consistently explained in any public statements on this subject by the OMA.

 
2. The potentially perverse impacts of premature mortality can easily be misinterpreted and discredit supporting economic analyses of mortality, as well as those for morbidity. Premature mortality should be presented as a special case and the effects on the health care system explained in this context.

Analyses of health care statistics and economics are anonymous and often do not accurately communicate the human suffering associated with air pollution. This is particularly the case with subtle, long-term and broad-based illness factors like air pollution. Even experienced medical practitioners cannot, in most cases, conclude definitively that a particular illness is due exclusively to air pollution. Multiple co-morbidity/co-mortality factors are the norm.

Accordingly, the attributable portion of an illness caused by air pollution is essentially invisible at the individual patient level. Life-long health history profiles are recommended to be developed for representative "statistical patients". These representative patients would include, among others, those individuals of the population who have high risk factors for suffering air pollution-induced illnesses. The expected health history of each patient would be tracked in a poor and good air quality environment to demonstrate the types and incident rates of health effects that could be expected under both air quality circumstances. These health history profiles are recommended for qualitative purposes, not for quantitative analyses (although as discussed in Section 4.1, if these profiles could be developed at an individual patient level, the greatest precision in estimating health care impacts would be achieved). Appendix D provides an example of hypothetical health history profile for a representative statistical patient.

4.5 Summary
 

1. Estimates of the health care impacts of air pollution are not available nor can they be produced directly from current forecasts of air quality-induced illnesses and the associated total economic value.

 
2. The current analytical procedures available for estimating air quality-induced health effects do provide some of the essential information for developing reasonable forecasts of health care resource utilization impacts.

 
3. Provincial health care data need to be combined with the knowledge and experience of clinical practitioners to develop standard treatment procedures and related health resource use rates for each of the health effect categories associated with air pollution.

 
4. Premature mortality represents a special case which needs to be given particular attention.

 
5. Communicating the results of any analyses of health care impacts needs to be done carefully. Confusion regarding the underlying economic framework and its relevance in a policy analysis context needs to be avoided.

 
6. Economic analyses are essential to make a "business case" for improved air quality standards but do not capture the human suffering element of air pollution. Developing health history profiles for representative statistical patients could be an effective means to communicate the human dimension of quantitative economic measures.

The preceding sections have systematically reviewed each of the primary elements in the air quality/health care expenditure model illustrated in Figure 2 . This section organizes the recommendations arising from these reviews into a series of steps. In essence, this section sets out a study plan to develop estimates of health care resource utilization associated with air pollution. These steps are presented in a logical progression based on the interdependencies among each.

5.1 Detailed Review of Current Version of the AQVM

The AQVM is recommended as a primary basis to forecast air quality-induced health effects. Accordingly, the AQVM will have a direct and major impact on any derived estimates of health care resource utilization. Of particular importance will be the assumptions underlying the health effects categories used to predict impacts.

Technical documentation for the latest version of the AQVM was not available for this study. The latest available documentation was for the 1996 version. Apparently, various refinements to the AQVM have been made since 1996.

Documentation for the latest version of the AQVM needs to be obtained. As well, a chronology of the refinements made since 1996 is required, particularly if forecasts of health effects from intermediate versions of the AQVM are proposed to be used as a basis for analyses of health care resource utilization impacts.

Health Canada and Environment Canada are in the midst of negotiations to offer the AQVM as a commercial software product. Any such product will require extensive documentation. Accordingly, the OMA can be reasonably confident that appropriate documentation exists and that it will be available at some point. Efforts need to be made to confirm when appropriate technical documentation will be available.

5.2 Detailed Review of Health Effects Categories

The AQVM reports air quality-induced health effects according to 15 categories (see Appendix C). Each category includes one or more specific types of illness and varying levels of severity for each. As a result, health care resource utilization can vary substantially within a given health effects category.

These health effects categories originate from the epidemiological research on which the AQVM is based. The illnesses included in each health effects category needed to be examined carefully. The objective is to ensure that any links between the AQVM health effects categories and provincial health care data are accurate, technically sound and consistent with the underlying epidemiological research.

As well, developing standard treatment procedures for each health effects category will involve characterizing the specific illnesses and severities included in each. These characterizations should be consistent with the profile of illnesses included in the epidemiological studies. In some cases, the authors of the epidemiological studies may need to be contacted directly to obtain the details required.

This step should result in a profile of specific illnesses included in each health effects category. These illnesses should be assigned ICD 9 codes so that appropriate linkages with the provincial data can be made. As well, the approximate frequency of each illness within a category and the severity should be included, if possible.

5.3 Preliminary Analysis of Provincial Health Care Data

The provincial health care data will be used for several purposes. One is to derive estimates of health care resource utilization coefficients for specific illnesses. These coefficients will reflect the treatment procedures prescribed for each illness. An important result of this task will be to report these treatment procedures in a style readily interpretable by clinical practitioners.

Other economic data need to be compiled on fixed assets and overhead as well as research, education and related costs. Some of these data may be included in conventional provincial health care data but past experience has shown some cost components are poorly represented and need to be refined considerably.

A major challenge of this task will be to extract the most appropriate and useful data as efficiently as possible. A clear understanding of the contents of each database and the underlying assumptions and limitations is essential to ensure that any economic analyses relying on these data appropriately reflect the underlying assumptions and limitations. A purpose of this step will be to document these features of the data and confirm their compatibility with forecast health effects categories.

5.4 Standard Treatment Procedures

Aggregate standard treatment procedures for each health effects category are required. Developing these standard procedures will involve analyses of provincial health care data bases. As well, the input of clinical practitioners experienced in prescribing and administering these procedures will be valuable.

Provincial data will indicate, on average, the health care resources consumed by patients afflicted with various illnesses. By combining this information in the form of a clinical treatment procedure, standard treatments for each illness can be derived. These standard procedures are likely to be more reliable for data from large hospitals than those derived from records submitted from individual doctors. However, some illnesses are treated more commonly through doctors' offices. As a result, treatment procedures for various levels of severity and for various locations/ institutions will likely be required.

These initial derivations of standard treatment procedures will need to be reviewed by clinical practitioners directly familiar with the treatment procedures for various cardio-respiratory illnesses. This review should refine these procedures and identify health care resources likely to be consumed but not included in the initial deviations.

5.5 Full Cost Accounting of Fixed Assets

A comprehensive full cost accounting framework is rarely used in public institutions, including the health care system. Developing reasonable allocations of the costs of capital assets and other fixed costs to individual illnesses is particularly difficult, partly as a result of how these costs are tracked and partly due to the diversity of health care services, patients and illnesses.

Considerable judgement will need to be exercised in this task. This task, alone, is a study on to itself. The relative significance of these associated health care expenditures will determine the appropriate investment of effort in developing these estimates. This task will result in an expenditure allocation for fixed costs for each health effects category.

5.6 Health History Profiles

Health history profiles are required for two purposes. The first is to examine the net effects of premature mortality on total health care expenditures. Health history profiles are required for various air pollutants (and ideally, combinations thereof) over a range of ambient and episodic concentrations. Developing these profiles will require analysis of the epidemiological literature, provincial health care data (particularly using provincial databases that allow tracking of the health histories of individual patients) and the experience and knowledge of clinical practitioners. Given the acute nature of premature mortality, temporal tracking (i.e., over a period of years) of individual patients should be possible. As well, the economic effects of premature mortality are evident primarily in the later years of life. Accordingly, to analyze the effects of premature mortality, a complete health history profile over the entire life of a patient is not an essential prerequisite. Likely these profiles will need to cover only the last 5 to 10 years of life to capture the major impact on health care expenditures.

The second purpose of the health history profiles is for public communication of the "human side" of air pollution. The use of these profiles for this purpose is discussed in Section 4.4.

5.7 Premature Mortality Analysis

The effects of premature mortality on health care resource utilization need to be assessed separately from the effects of increased morbidity. This step will involve analyzing for different representative health care profiles, the effects of premature mortality on health care expenditures during the later years of life, as well as on the time pattern of these expenditures. This analysis will require converting the stream of expenditures in the years leading up to death, into present value terms. Doing so will allow direct comparisons between the total expenditures with premature and normal mortality.

Of particular importance will be a careful examination of the epidemiological literature. This review will focus on gaining an understanding of the average reduction in longevity caused by air pollution and any exceptional medical complications commonly related with these premature deaths. Extracting this information will be difficult given the multiple contributing factors to premature mortality and the subtle effects of air pollution. Some crude approximations may be necessary given the complexity of these relationships.

Once these details are obtained, the remainder of this task is largely a matter of accounting. These economic analyses will rely on the results of earlier steps. This step involves essentially compiling the supporting data in a suitable form and using conventional economic techniques to calculate the present value of the expected stream of expenditures for patients experiencing premature and normal mortality.

Once these results are produced for a representative group of patients expected to suffer premature mortality, their significance needs to be assessed. It may be necessary to refine the proposed strategy to deal with the special case of premature mortality depending on the results produced.

5.8 Selection of Air Quality Forecasts

The ultimate objective of this initiative by the OMA is to influence government air quality policy. Air quality policy ultimately is expressed as a change in the ambient and/or episodic nature of air pollution. To implement the policy analysis framework outlined in Figure 1, alternative policies need to be analyzed. This raises the question as to which changes in air quality policy should be analyzed.

This step will involve examining existing forecasts of air quality for various pollutants given various policy alternatives. The policy alternatives to be analyzed in terms of health care resource utilization impacts will need to be chosen carefully. Part of this selection process will involve assessing whether the air quality and selected health effects of each policy alternative are available and whether they are based on reasonably current versions of the best air quality and health effects models available. Full supporting technical documentation should be obtained for any forecasts that are selected for further analysis.

5.9 Generate Health Care Expenditure Estimates

Health care expenditure estimates for each policy alternative need to be developed. Estimates of total expenditures, as well as a detailed breakdown by health care resource components and by morbidity and mortality, should be produced. These estimates will rely on the expenditure coefficients developed in the preceding steps for each health effects category.

5.10 Communication of Results

The final step is to organize the results of the economic analysis in a form easily understood by the general public as well as by government officials. As part of this communication package, health history profiles for representative statistical patients that correlate to the policy alternatives under consideration should be prepared. As well, the basis for developing the expenditure estimates needs to be clearly set out along with limitations in the data and methodology. Any published health care resource utilization estimates should be accompanied by a supporting technical report.