6.4.1 Main findings
A total of 18 studies, published from 1950 until April 2014, have been identified as relevant for informing the association between chronic exposure to P M2.5 and respectively all-cause mortality and lung cancer incidence or
mortality. Mean effect estimates and 95% confidence intervals obtained by pooling study-specific results using the random-effect model are respectively: 1.07 (1.05 - 1.10) for all-cause mortality and 1.13 (1.07 -1.20) for lung cancer incidence or mortality.
Moderate heterogeneity was found between studies, which justifies the use of the random effect model. Several sources of between-studies heterogeneity were identified including: (i) varying approaches to spatio-temporal assessment and assignment of exposure ; (ii) differences in study-population age, gender and socio-economic status mix and (ii) expected differences in particulates’ chemi- cal composition and toxicity. A meta-regression of the pooled estimate against study characteristics was however, not performed for the following two reasons. First, performing these analyses requires a sufficient number of studies, with a suggested minimum of ten estimates per characteristic modelled (Higgins & Green, 2011), which were not available from the present pool of studies. Sec- ond, since studies are not randomised across potential effect-modifiers, their findings are problematic to interpret (Deeks et al., 2008).
6.4.2 Comparison with work published most recently
The present results, which take into account the most recent evidence pub- lished until end of April 2014 are consistent with those previously obtained by Chen et al. (2008) for the period 1950-2007. The authors reported random
effect-pooled estimates of respectively 1.06 (1.03 - 1.10) for all-cause mortality and 1.15 (1.06 - 1.24) for lung cancer incidence or mortality.
Since the start and completion of this piece of work, two meta-analyses on the association between P M2.5 exposure and respectively all-cause mortality
(Hoek et al., 2013) and lung cancer incidence or mortality (Hamra et al., 2014) were published (in May 2013 and September 2014 respectively).
Hoek et al. (2013) reported a random effect-pooled estimate of 1.06 (1.04 - 1.08) for all cause mortality, which is very similar to the present findings. The authors however, only included studies published until January 2013 and as a consequence, did not encompass results from three cohort studies: ESCAPE, English cohort and ACS California sub-cohort (see Table 6.2 ). In addition, the authors encompassed in their scope studies based on elderly populations (Zeger et al., 2008; Enstrom, 2005), which may further explain the slight difference with the present result.
Hamra et al. (2014) reported a random effect-pooled estimate of 1.09 (1.04 - 1.14) for lung cancer incidence or mortality, which is smaller than the present pooled estimate. After careful analysis, it appeared the difference in estimates was driven by two factors. First, Hamra et al. (2014) used the risk estimate from the American Cancer Society (ACS) national full cohort based on PM exposure for the years 1979-1982. By contrast, the present analysis relied on results based on the 1999-2000 exposure period, which included 42% more study participants (see section 6.2.3). Unfortunately, Hamra et al. (2014) did not justify their choice of exposure period. Sensitivity analysis using the risk estimate based on the 1979-1982 period of PM exposure yielded a slightly lower pooled estimate (1.11 - 95% CI: 1.06 - 1.17, see section 6.3.4), which is more in line with Hamra et al. (2014)’s findings.
Second, Hamra et al. (2014) included a result from a cohort study in China (Cao et al., 2011). This study was excluded from the pool of relevant studies in the present analysis as it did not provide a risk estimate for particulate air pollution exposure but only for total suspended particle (TSP). While Hamra et al. (2014) had similar inclusion criteria and clearly stated that they excluded studies which did not provide quantitative estimates for particulate matter, they apparently made an exception for this study - without justification - and
converted the risk estimate for TSP to a risk estimate of P M2.5 applying a 3:1
ratio. This led to an estimate of 1.07 (95% CI: 1.0 - 1.07). Due to a relatively small standard error, the weight attributed to the Chinese study in Hamra et al. (2014)’s meta-analysis was high (21%).
In order to assess the impact of these two factors on the pooled risk estimate for lung cancer mortality or incidence, a third sensitivity scenario was run by (i) using the risk estimate from the ACS national full cohort based on PM exposure for the years 1979-1982 and (ii) adding results from Cao et al. (2011)’s study. Results are presented in Figure 6.7. In this scenario, the pooled estimate and its 95% CI exactly match with Hamra et al. (2014)’s findings.
Figure 6.7: Sensitivity-analysis for meta-analysis results for lung-cancer using Hamra et al. (2014)’s study scope.
6.5
Conclusion
A systematic review and two meta-analyses of the association between long- term exposure to fine particulate matter (P M2.5) and respectively all-cause
mortality and lung cancer incidence or mortality were performed. These quan- titative analyses update past work done by Chen et al. (2008), by including all the relevant evidence published over the last seven years. These results are important for public health practitioners and policy-makers who need to assess air pollution control interventions based on all existing evidence.
Present results are consistent with two meta-analyses published after com- pletion of this work: Hoek et al. (2013) and Hamra et al. (2014). For lung cancer incidence or mortality, the difference between the presently obtained pooled estimate and Hamra et al. (2014)’s results appears to be driven by two unjustified choices made by the authors and does not put into question the quality of the present work. Nevertheless, since Hamra et al. (2014) results were published, they will be used to parameterise the model built in Chapter 4. By contrast, the presently obtained pooled estimate for all-cause mortality, which closely matches with Hoek et al. (2013)’ estimate, but includes most up to date evidence and excludes studies on elderly subjects, will be used to parameterise the model of the health effects of pollution exposure.
Chapter 7
QALY gain, health care
resources impact and
cost-effectiveness of air pollution
control in England and London
7.1
Introduction
Chapter 4 developed a Markov model of the health effects of air pollution exposure, in order to fully capture air pollution’s joint effect on quality and length of life as well as to assess the total health care budget impact of a reduction in air pollution. The model required a number of parameters, a subset of which were estimated in Chapters 5 and 6.
This chapter presents the results from the application of the developed model to the UK case study detailed in Chapter 4. The intervention under- pinning the case study, hereafter referred to as “the intervention”, consists of an immediate and sustained 1µg/m3 decrease in population-weighted mean
P M2.5 concentrations in England and Wales and London. This is expected
trations. Section 2 provides the total expected health gain and health care cost impact associated with the intervention, as well as the distribution of outcomes by age and gender. Section 3 focuses on uncertainty surrounding outcomes and also evaluates results’ sensitivity to the choice of discount rate and to dynamics in risk reduction. Section 4 compares the developed Markov model with the simple life-table approach currently used in health impact as- sessment (HIA) and contrasts estimates of un-discounted life expectancy gain with results from past HIAs. Based on case study results, section 5 evaluates the cost-effectiveness of reducing air pollution in London, whether such an intervention would be funded by the NHS or through general taxation.