4.4 Model structure
4.4.3 Intervention arm and risk reduction estimates
Building from Chapter 3
As explained in Chapter 3, to evaluate an intervention such as air pollu- tion reduction, the model requires: a “baseline” arm populated with baseline TP to be compared against (ii) an “intervention” arm for which, similarly to a health impact function, baseline TP are combined with risk reduction estimates (RRE). The latter are epidemiological risk estimates scaled to the intervention-related exposure decrement. Since TP are non-linear function of time, the multiplication of TP with RRE is carried out on the transition rate scale. Adjusted transition rates are then converted back to probabilities to parameterise the intervention arm.
Chapter 3 presented two approaches to the application of Markov modelling methodology to quantitative risk assessment, which differ in the scope of the morbidity-mortality interactions encompassed and in the way life expectancy impacts are modelled. Like the “full-modelling approach”, the present model aims at encompassing thoroughly the chronic morbidity effects associated with long-term PM exposure and their impacts on life-expectancy, including their influence on individuals’ susceptibility to dying prematurely due to PM expo- sure. However, since air pollution has been associated with a positive excess risk of all-cause mortality, a lower risk of death under pollution decrement was also applied to those “healthy” individuals who did not enter any disease path- way. Importantly, as advocated in Chapter 3 when describing the “focused” modelling approach, to avoid double-counting of life expectancy effects, the change in mortality risk for those “healthy” individuals exclusively pertains to all the other causes of death than the ones modelled (see parameter RREg in
Table 4.1).
Finally, as the present objective is to assess the total QALY and health care resource impacts of reducing particulate air pollution, as opposed to disentan- gling impacts by causal pathways and population subgroups as was done in
Chapter 3, only one intervention arm fitted with all relevant and appropriately scaled risk estimates was built.
Risk estimates: sources and application
Although the model is to be applied to evaluate an intervention of air pol- lution control in the UK, risk estimates were taken from studies performed in various developed countries, mainly in North America and Europe, with only one risk estimate derived from a UK study (Tonne & Wilkinson, 2013). However, this is unlikely to be a major limitation since developed countries are characterised with a similar range of P M2.5 concentrations - ranging from
about 5 to 35 µg/m3 - across which linearity in health impacts and absence
of threshold to effects has repeatedly been found (Lepeule et al., 2012; Crouse et al., 2012; Krewski et al., 2009). These characteristics of the dose-response function have two implications. Within the above concentrations range: (i) estimates of health effects for a different level of pollution reduction may be obtained by simple proportional scaling of results obtained per one unit decre- ment in ambient P M2.5 concentrations; (ii) RRE can be derived from risk
estimates, that are typically expressed for 4P M 2.5= +10µg/m3, using simple
logarithmic multiplicative scaling (see Appendix C of Chapter 3).
In an effort to encompass all existing relevant evidence and to decrease pa- rameter uncertainty, risk estimates were preferably sourced from meta-analyses. In light of the number of studies pertaining to all-cause mortality and lung cancer development or mortality published in recent years, it was decided to carry out a systematic review of such studies and two meta-analyses. This was performed separately in Chapter 6.
To limit extrapolation of epidemiological findings, application of risk esti- mates in the model sought, to the extent that it was feasible, to be in line with study subjects’ main characteristics in terms of age and/or health condition. For instance, the only available piece of evidence on the excess risk of death associated with PM exposure in COPD patients is based on individuals aged above 65 years old, identified using hospital discharge data (Zanobetti et al.,
2008). As the risk of hospital admission for COPD greatly increases with dis- ease severity, the study’s risk estimate was applied only to those individuals aged 65 and above if they were in GOLD 3 or 4 states. In other words, it was conservatively assumed that individuals with COPD in stages 1 and 2 or in stages 3 and 4 but aged below 65 faced the same PM-related excess risk of mortality as the general population. Similarly, the PM-related excess risk of mortality in individuals with CHD was informed by a study from Tonne & Wilkinson (2013), based on patients above 25 years of age admitted to hospital following acute coronary syndrome (ACS). ACS reflects a more severe health condition than CHD as a whole. Since the risk of ACS linearly increases with age (Simms et al., 2012), Tonne & Wilkinson (2013)’s risk estimate was only applied to individuals suffering from CHD if they were aged 75 or above. In- dividuals with CHD aged below 75 were therefore conservatively assumed to face the same PM-related excess risk of mortality as the general population.
Table 4.1 links baseline transition probabilities with relevant risk reduction estimates expressed for a 1 µg/m3 decrement in P M
2.5 concentrations. As
lung cancer is very deadly, the impact of P M2.5 exposure on the lung cancer
pathway was restricted to disease development, i.e. no further PM-related excess risk of death applied to individuals suffering from lung cancer.
Figure 4.2 represents the model’s intervention arm, with RRE associated with various transition paths. Dotted arrows represent RRE-adjusted transi- tions, i.e. transitions for which the underlying risk of event is reduced under pollution reduction, whereas full arrows represent transitions for which the underlying risk of event is assumed to be unchanged under the intervention.
Parameter Transition Pop. Risk Reduction Estimates (RRE)
Name Probability age Risk Estimates Mean (95%CI) PX,Y Definition Source 4P M 2.5= −1µg/m3
RREa PH,COP Di All ORDev.COP D(a) Schikowski et al. (2014) 0.988 (0.918-1.065) i=GOLD1,..., 3
RREb PH,CHD All HRDev.CHD Cesaroni et al. (2014) 0.976 (0.949-1.004)
RREc PH,LC All HRDev.LC see Chapter 6(b) 0.985 (0.980-0.991)
RREd PCOP Di,D All HRDeathAC see Chapter 6 0.993 (0.991-0.995)
i=GOLD1,...,2
RREe PCOP Di,D < 65 HRDeathAC see Chapter 6 0.993 (0.991-0.995)
i=GOLD3,...,4 ≥ 65 HRDeathAC|COP D Zanobetti et al. (2008)(c) 0.980 (0.976-0.984)
RREf PCHD,D < 75 HRDeathAC see Chapter 6 0.993 (0.991-0.995)
≥ 75 HRDeathAC|CHD Tonne & Wilkinson (2013) 0.982 (0.968-0.996)
RREg PH,D All HRDeathAOC|H Pope III et al. (2002) 0.999 (0.994-1.005)
Table 4.1: Risk reduction estimates for intervention arm.
Abbreviations: PX,Y : age and gender-specific annual probability of developing disease/ experiencing event “Y”, conditional on being in health state “X”; Dev. = developing; COPD = chronic obstructive
pulmonary disease; CHD = coronary heart disease; LC = lung cancer; H=healthy; D= dead; HR=hazard ratio; HRY |X: hazard ratio of event “Y” in population with health condition “X”; OR= odd ratio; AC= all causes; AOC = all other causes.
(a)When events are rare, i.e. with a probability of event occurrence in the unexposed group less than 10%,
which is the case of COPD, OR can be considered equivalent to RR (Sistrom & Garvan, 2004).
(b)In order to be in line with the most recent published work of Hamra et al. (2014), the pooled estimate
of the excess risk of lung cancer associated with P M2.5exposure used to parameterise the present model was taken from sensitivity analysis run 3 in Chapter 6, as presented in Figure 6.7.
Figure 4.2: Diagram of the model’s intervention arm.
Abbreviations: COPD = chronic obstructive pulmonary disease; CHD = coronary heart disease; LC = lung cancer: Yr = year. Risk reduction estimates RREa, ...g are defined in Table 1 and apply to