The sources of air pollution in Estonian capital Tallinn (~390 000 inhabitants) are quite complex with an important role played by local heating. Thus, the health impacts of the air pollution are best characterized using PM_2.5 using relative risk assumptions from studies of a mix sources. The negative effects of chronic exposure to fine particles, even at low concentrations, has previously been shown in various epidemiological studies 1819.

The aim of this study was to estimate the added local health impacts due to emissions in Tallinn. HIAs in such towns as Tallinn help to quantify the health effects of air pollution in less polluted average-size cities and in less studied regions where economic transitions have been very rapid. The current study improves the explanatory power of HIA methods by incorporating modeling and sectioning approaches for cities where a air quality measuring network is rare or absent.

Population data for Tallinn is based on the Population Register (02.02.2007) according to address and registration in the following age groups: 0–6, 7–17, 18–27, 28–37, 38–47, 48–57, 58–67, 68+ years. The citizens' residences were divided into sections according to neighborhoods (regions with similar geographical, socio-economic, etc., patterns), forming small administrative units (smaller than city districts) used in city planning and management. Altogether 84 sections (Fig 1, 3 and 4) were formed in order to identify site-specific exposure to air pollution and identify the areas with greatest risk. A 'neighborhood' is considered to be a small and homogeneous section, where air pollution as a risk factor is assumed to be similar. The age-structure of the population in each section was identified and used for calculation of YLL with life-tables methodology. Each section also belongs to one of the 8 city districts (Table 1).

The total regional baseline mortality was retrieved from statistics on Estonia (International Classification of Diseases – ICD-10, A00-Y98). The morbidity calculations were carried out using hospitalization data from the Estonian Health Insurance Fund (EHIF), which covers the whole population and is the sole purchaser of health care services in the country.

Hospitalizations due to two main disease groups were included in the calculations: cardiovascular (I00-I99) and respiratory causes (J00-J99). Cardiac admissions (I20-I25) and cerebrovascular admissions (I60-I69) were also used for the exposure-response work on cardiovascular hospitalizations. The short-term effects of high pollution levels on mortality were not calculated separately as according to several authors 2916 these are already included in exposure-response function of long-term mortality.

The annual levels of locally-emitted PM_2.5, as well as PM₁₀ for model validation were estimated using model AirViro 20, based on emission data for traffic, industry, local and central heating along with meteorological parameters with grid resolution 200 × 200 meters. A database of local heating emissions was developed during the current study, using a previous questionnaire study on fuel consumption results 21 and the European Environmental Agency's emission factors for small combustion devices based fuel consumption. The traffic flows have been measured in Tallinn in 2005 and 2006. The emission factors were taken from CORINAIR 22 for traffic and from a database of pollution licenses for industry, central heating etc. The temporal resolution for traffic was one hour, a working day for industrial sources and one month for boiler houses and residential heating.

For model validation the PM_2.5 and PM₁₀ modeled levels were compared with air quality monitoring data from Rahu, Õismäe and Liivalaia receptor points (Fig. 1) for three meteorological years, 2004–2006. The Liivalaia monitoring station is located in the city centre and it represents a typical city hotspot. The Õismäe monitoring station is an urban background station located in a typical city region with apartment houses where a large number of the Tallinn population reside. The Rahu monitoring station is located in a region of residential houses together with a nearby railway with a number of the diesel trains passing by on a daily basis. In each monitoring station, the concentration of PM₁₀ is routinely measured by beta-attenuation analyzers (Thermo Andersen FH-62). In the Õismäe station, the PM₁₀ levels are measured by the reference method (Digitel DHA-80) as well. The PM_2.5 level is monitored only at the Õismäe station, and this is done by beta-attenuation analyzer (Thermo Andersen FH-62).

The annual levels of PM_2.5 were calculated for all 84 Tallinn sections using the average concentration of modeled grid cells in a section. The average concentration for each section was then assigned to all residents of that neighborhood. Only individuals of age 28+ were included in analyses, as the US cohort study 23.

Short-term effects of air pollution were calculated using mean daily average concentrations of PM₁₀ from the 3 monitoring stations in 2006 and morbidity data for all age-groups. For each neighborhood the most representative station was used based on the location.

To describe the long-term effect of air pollution on mortality, the broadly employed US ACS study relative risk RR = 1.06 (95% CI 1.02–1.11) per 10 μg/m³ increase of PM_2.5 was used as the exposure-response relationship 23. For calculations of respiratory hospitalizations due to short-term air pollution episodes, RR = 1.0114 (95% CI 1.0062–1.0167) per 10 μg/m³ increase of PM₁₀ was used 24. For cardiovascular hospitalizations we used a weighted average RR = 1.0073 (95% CI 1.0047–1.0093) per 10 μg/m³ increase of PM₁₀ based on the effect on cardiac and cerebrovascular admissions from a COMEAP meta-analysis 25.

The cases (mortality and morbidity) were calculated in absolute and relative numbers for all sections in Tallinn. The following equation was used:

ΔY = (Y₀ × pop) × (e^{β × X} - 1)

where Y₀ is the baseline rate; pop the number of exposed persons; β the exposure-response function (relative risk) and X the estimated excess exposure.

The number of YLL was calculated using "life-tables" methodology, where the hypothetical life expectancy is compared with the life expectancy affected by air pollution. The calculation of YLL and changes in life expectancy were facilitated by a WHO Centre for Environment and Health developed program AirQ 2.2.3 (Air Quality Health Impact Assessment Tool) 26. For calculation of hospitalizations, the short-term effects module of AirQ was used. The number of hospitalization cases was determined at different exposure intervals (10–19.9; 20–29.9; ...); no effect was assumed below 10 μg/m³.

Air pollution affects economic and social well-being through mortality and morbidity. Morbidity, in turn, affects the health and productivity of the labor force. In this study, the direct costs related to morbidity were calculated using costs of hospitalization, salary compensation during sick leave and loss of labor input (based on GDP per capita). The data for hospitalization cost calculations were provided by EHIF, where the average costs of a hospitalization case due to respiratory disease or general internal disease in 2005 was €1239 and €778 respectively 27. The same source was used to identify the time spent in hospital (6.9 days) and the value of the average compensation of a workday (~€10).

For the country as a whole and its development prospects, the long-term outcomes and costs of air pollution effects are even more important than the direct costs. This means that in a case of premature death, people can lose decades of life-years but direct costs appear only in the actual year of death. The concept of Statistical Value of Life (SVL) and Value Of Life Year (VOLY) are used to express the cost of lost lives and life-years. These concepts stem from people's contribution to GDP, typical work time, salary and sometimes health care (compensation and decreased productivity) costs 2829. As there are no comprehensive statistical life valuation studies in Estonia, the conversion coefficient between GDP and the statistical value of life was derived from international meta-analyses (statistical value of life being on average equal to 120 times GDP per capita in a country) 3028. Value of a life year was calculated from the statistical value of life following the formula:

V O L Y = S V L A T − A MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemOvayLaem4ta8KaemitaWKaemywaKLaeyypa0tcfa4aaSaaaeaacqWGtbWucqWGwbGvcqWGmbatdaWgaaqaaiabdgeabbqabaaabaGaemivaqLaeyOeI0Iaemyqaeeaaaaa@3A59@

where VOLY is statistical value of life year; SVL statistical value of life; A age when the case happened; T life expectancy; T - A loss of life.

A sensitivity analysis was performed using minimal and maximal economic statistical values of life and life-year to describe the range of potential errors.

Altogether, 388 964 registered residents of Tallinn were identified in 84 sections of the city. Population-wise the biggest sections had more than 15 000 residents while some of the smallest had less than 100. The population density varied a great deal as well. In the majority of sections, the number of residents ranged from 3 000 to 16 000.

Based on mortality data, the mortality rates in different age groups were found (average 1 136 cases per 100 000 citizens per year) and the numbers were calculated in all 84 sections for the reference year 2006. The baseline hospitalization rates were determined separately for cardiovascular and respiratory admissions per 100 000 people using the same principles. The analysis showed 3 945 and 1 266 annual admission cases respectively per 100 000 people.

The city centre and nearby regions with local heating can be clearly differentiated as areas with higher exposure to fine particles (Fig. 1). The concentrations are also higher in other regions, especially those adjacent to the city centre. High concentrations also appear in smaller residential areas, particularly near busy streets (Fig. 1). The lowest concentrations were in the Tallinn fringe area with small but densely populated residential neighborhoods.

The exposure to fine particles was calculated by sections, using the modeled annual PM_2.5 levels in 200 × 200 m grids in Tallinn. Modeling was based only on local sources as no data about background fine PM in Estonia are available. The background could be illustrated by the difference between modeled values (based on local sources) and measured values (local sources and regional background). Since we found that Tallinn itself contributes approx 0.4 μg/m³ to the unknown regional background, it was subtracted from the modeled exposure values in order to calculate health impacts only associated with the levels above those outside the city.

The difference between modeled and measured mean PM_2.5 values in Õismäe station was 21% in 2006 (Fig. 2). The 2–3 μg/m³ variation in PM_2.5 annual values from the model indicates a somewhat lower background than expected. The average difference for all three monitoring stations above modeled PM₁₀ levels over three years measurements was 18.8%. The biggest difference was in Rahu monitoring station, which is close to a railway with diesel locomotives and less known emissions, where measured and modeled PM₁₀ values differed by 37%. At both Liivalaia and Õismäe monitoring stations the measured and modeled concentrations of PM₁₀ differed by only 11% (Fig. 2). As agreements between the measured and modeled PM₁₀ levels for the monitoring stations were fairly good, we have assumed that the model could also represent real particle levels fairly well at other receptor points in the city.

The concentration of particulate matter (PM₁₀) at the 3 monitoring stations differed quite much. Generally, the concentration was the highest in the town centre and the lowest in the residential areas. This is expected, as the former has busy traffic and the latter is an apartment house area. It is noteworthy that in spring for some time the concentration at Õismäe station was surprising higher than in the centre town. As PM_2.5 levels at that time were not very high, these high pollution episodes were presumably driven by coarse particles (PM_2.5–10).

As some neighborhoods ('sections') had very few deaths, the estimated number of premature deaths attributed to the additional (local) particle pollution is presented at the level of city district (Table 1), whereas YLL is given at the level of neighborhood section (Fig. 3).

Our analysis shows that locally emitted air pollutants could be estimated to cause 296 (95% CI 76–528) premature deaths per year in Tallinn. According to the AirQ calculations using life tables these deaths correspond to 3 859 (95% CI 1 023–6 636) YLL, which is 988 YLL per 100 000 citizens. As a total number, the greatest loss (235–650 YLL) was in neighborhoods with a large number of citizens (25 000–50 000), e.g., Mustamäe, Lilleküla, Väike-Õismäe and Laagna (Fig. 3). The relative largest losses appeared in the city centre neighborhoods Kompassi, Südalinn, Tõnismäe.

Air pollution in Tallinn would then reduce the life-expectancy of the residents by on average 0.64 (95% CI 0.17–1.10) years, what corresponds to 7.7 months. The reduction is much greater in the city centre, e.g., in the Kompassi neighborhood where it reaches up to 1.17 years, whereas in the least polluted neighborhoods the decrease of life-expectancy remains between 0.1–0.3 years (Fig. 4). The average number of YLL per premature death is approx. 13 years.

Many of the negative health influences emerge in risk groups (people with respiratory and cardiovascular disease, elderly, etc.) 19. Nevertheless, healthy people may be affected as well. Synergistic interactions with air pollution can appear to occur with other diseases.

Regarding morbidity, short-term exposure to PM₁₀ is estimated to cause 71 (95% CI 43–104) respiratory hospitalizations per year in Tallinn. Even though RR is lower, the baseline frequency of cardiovascular hospital admissions is higher than for respiratory and the attributed absolute number is greater; 204 (95% CI 131–260) pollution related hospitalizations per year.

There were 275 short-term air pollution related hospitalizations in Tallinn. Drawing from the average treatment cost data, the total direct cost to treat these air pollution related hospitalizations would be ~€245 000. For the days spent in the hospital ~€2 600 was paid to compensate temporary loss of income. The national economy lost input from hospitalized individuals to the value of €44 000 annually.

The statistical value of a life in Estonia in 2005 was estimated around 1 million € and the statistical value of a life-year ~€39 000. Based on these values the total indirect loss from air pollution caused premature deaths (296) adds up to €150 (95% CI 40–260) million.

In summary, the major external economic costs related to exposure to outdoor air pollution add up to €150.3 (95% CI 40.2–260.4) million. The majority stems from loss of life-years from premature deaths. This represents ~2.9% of Tallinn's GDP (€5.2 billion in 2005).

While the methodology for HIA in this study follows generally accepted principles 31, major differences have been found in the exposure assessments. In the case of Tallinn, air pollution has been measured in only 3 monitoring sites. Thus, it was necessary to use dispersion modeling to gain an adequate level of detail for exposure assessment by city sections. The model validation showed fairly good agreement with the monitored levels, although the model generally underestimated the particle concentrations. The reasons for this may be the lack of a background concentration and an incomplete emission database; for example the high levels of PM measured at springtime because of road dust.

In our study, the sectioning was determined by a variety of natural and social factors, but the final results were found to follow to a large degree the municipal districts in Tallinn. Detailed population data was easily available even for small neighborhoods. In this case, it was feasible to employ this data in order to increase the accuracy of the analysis. Nevertheless, such detailed data is not always available. Population exposure is usually calculated as the average of monitored particulate matter in a large area subsequently adjusted for population density. We assume that in smaller cities with sparse monitoring networks, it would be more relevant to use GIS technologies for calculating average concentration by neighborhood and explicit demonstrating any variations. There might be contradictions with larger studies such as the one undertaken here, as the latter exposure-response coefficients are determined using different exposure assessment methods, but the differences would not be major.

The place of residence was used as the exposure position presuming that the greatest portion of the day is spent there. This is similar to other epidemiological studies from which exposure-response coefficients were taken. Furthermore, site of dwelling was the only data available from the population register. The amount of time a person spends in the residence area and outside of it (work, studies, etc.) affects individual exposure levels, however current methodology does not permit consideration of these variations. Neither could they be considered in the studies providing our exposure-response functions. When doing analysis with such accuracy (which is possible with modeling), individual factors such as a home's exact distance from the street, other pollution sources, individual sensibility to pollutants, etc. could play an important role.

As expected, the exposure was highest in the city centre and close to busy streets. People who live or work there, rather than the people who drive through, are more exposed to pollution. The number of people living in Tallinn is possibly greater than that indicated by the data used because of optional registration. In the sectioning process we also lost ~3% of persons, due to mismatching between Population and Land Register datasets.

Firstly, the baseline population and health data are vital determinants affecting HIA results. Similar statements were made by Tainio et al. 32 in their statistical modeling study. In Tallinn, when looking at the absolute numbers, the highest number of casualties occurred in Lasnamäe, as the largest number of people lives in this neighborhood. If we look at relative values (e.g. YLL per 100 000), the greatest numbers are in the city centre and Kristine neighborhood. Thus, in Mustamäe and Lasnamäe the absolute number of casualties is similar as air pollution exposure is much higher in Mustamäe (Table 1). The age structure of the different populations also plays a role, but is quite minor.

Questions may arise about the possibility of over (under)estimation of the health impacts. The main basis for overestimation is the high baseline mortality rate (driven by external causes) in Estonia. This magnifies the relative exposure impact. If the health of the population is generally weak, the residents could likely be more sensitive to air pollution. In some cases, as in the Reshetin & Kazazyan study, where air pollution was said to cause 15–17% of mortality in Russia 33, the health impacts are probably overestimated because of very high base-line mortality (to a large extent related to alcohol consumption). Of course, we should not be too conservative in our estimations. In Helsinki, where the air pollution influence of busses was assessed, the results could be underestimated because only exhaust particles, which are seen as more toxic, were taken into account 34.

Secondly, the choice of a dose-response relationship represents an even more critical determinant of HIA results than any consideration of baseline data. For long-term mortality assessment, we applied the relative risk (Pope et al. 23) from the ACS study that is broadly used in HIA studies. Since most of the stations in that survey were in urban surroundings, the combustion particles (prevailing in the city centre) could cause relative risk up to 1.17 per 10 μg/m³ increase of PM_2.5, as Jerret et al. found in the California cohort used in the ACS study 35. Thus, we could have underestimated the effects on mortality. There are many other exposure-response coefficients available, but there is a great deal of differences of opinion, and additional studies are needed in order to determine which coefficient is the best.

A third problematic aspect lies in the choice of pollutant as the pollution indicator. We assumed that most of the health effects of air pollution could be quantified with PM_2.5, especially due to the impact of local heating. In the HIA by Tonne et al. (2007), where effects of PM_2.5 and NO₂ from traffic were compared, there was a slightly greater influence on health recognized when NO₂ was the indicator 36. This means that we may have underestimated the pollution impact in the city centre. One possibility would be to conduct HIAs with several pollutants, so that alternative exposure rate scenarios could be designed. However, a simple addition of the effects of different pollutants would lead to overestimation and would be methodologically wrong.

The fourth critical issue is the exposure data in combination with any assumed threshold level of health effects. Studies have shown that fine particulate matter can cause negative effects on concentrations below current limit values 37. In principle, we have assumed that the local contribution has the main impact. The background concentration is often used as the reference concentration. However, as the background is here presumed (as there is no study) to be quite low in Tallinn (2–3 μg/m³), we might have a slight over-estimation in our results, if a threshold level exists above this background. Our modeled PM levels seem to correspond to measured levels quite well except for one measuring site. For the short-term effects on hospital admissions, we calculated effects only above daily means of 10 μg/m³. If we had built this calculation on yearly mean values instead, and calculated the attributable cases above an estimated regional background (likely 4–6 μg/m³), we would have estimated a larger number of excess cases.

Even though air pollution exposure in Tallinn is relatively low, the number of premature deaths and hospital admissions is rather high. As baseline cardiovascular hospitalization is known to be much higher in Estonia compared to respiratory hospitalization, the big differences between them was predictable. The negative effects on morbidity could also be found in family physician and other doctors' outpatient records. However, due to lack of relevant data, these morbidity cases were not taken into account.

The average loss of life expectancy (at birth) estimated here is slightly less (7.7 months) than the average of all EU citizens (8.6 months) 13. The rate of premature deaths (76/100 000) is almost the same as the EC study which showed 75/100 000 among EU-25 residents 12.

The total external costs of air pollution estimated here at €150.3 million make up 2.9% of the Tallinn GDP (in 2005). This is somewhat smaller compared to findings from Russia 2.6–6.5% 14 and Beijing 6.55% 38, which are of course much more polluted. But compared to the 1.5% for Europe (WHO assessment), it is slightly higher 13. The main reason for that might be quite high decrease of life expectancy per premature death case.

The use of HIA as an assessment tool is based on assumptions that local situations and social factors are similar to the reference conditions used in the epidemiological studies from which the exposure-response coefficients are derived. The design of epidemiological studies for long-term effects of air pollution is complicated, especially in small or average size towns where air quality and social patterns vary. Thus, the design used in the current study would be most applicable where resources are limited.