The QBM predicts that the benefits of quitting smoking are substantial in the subsequent 10-year period. For every 1000 quitters, randomly selected from the Australian smoker population, savings of almost A$400,000 in health care costs, and prevention of 40 cases of the four most common smoking-associated diseases and 18 deaths can be expected. Sensitivity analyses indicated that these predictions are robust to plausible variations in the model parameters.
The validity of a model such as the QBM is dependent on, first, how well the model structure represents the outcomes of smoking and quitting in real life, and, second, on the accuracy of the model parameters. In developing the QBM, when assumptions about model structure or choices about data sources were needed, we took the most conservative course, i.e. we chose to under-estimate the adverse health and health care cost consequences of smoking and to under-estimate the benefits of quitting. In relation to disease incidence, health care costs and quality of life, the QBM considers only four of the hundreds of smoking-related diseases, [42
] making the model conservative with respect to cost savings and QALYs, although these four diseases do account for over 80% of morbidity (and mortality) associated with smoking. Furthermore we only modelled the incidences and costs associated with the first of these four smoking-related diseases which occurred in an individual [12
]. This is conservative, as people who have had an AMI, for example, might also subsequently have a stroke, COPD or lung cancer. However the mortality rates following a diagnosis of one of the four diseases did include deaths from any other causes.
The QBM only considers the impact of quitting on the smoker. This is also conservative, because smoking adversely affects others, in particular by increasing the risk of coronary heart disease in those exposed to environmental tobacco smoke [43
] and increasing the risk of a pregnant woman who smokes having a low-birth-weight infant. Quitting smoking can reduce these risks, [44
] but this was not incorporated in the QBM.
The most important parameters in the QBM are the exponential models describing the decline after quitting in the risk of the four smoking-associated diseases and other causes
mortality. These models were all based on relative risks from studies involving very large numbers of subjects. For AMI and stroke, the models were previously developed by one of us (SH) [4
] and were based on relative risks from all available published studies that met specified criteria. Relative risks for AMI were sourced from four case-control studies with more than 8000 cases and more than 16,000 controls, [46
] and one cohort study involving almost 1000 cases and over 120,000 subjects [50
]. For stroke, relative risks came from two cohort studies with over 600 cases and more than 120,000 subjects[51
]. For lung cancer and COPD, the model was based on relative risks from a lung cancer case-control study with more than 600 cases and more than 2000 controls [26
]. The mortality model was based on the ACS CPS II study of 1.2 million people [29
The model structure and data sources in relation to mortality are also important aspects of the QBM. Mortality rates for the four smoking-associated diseases, were based on case fatality data from a variety of sources. To estimate the other causes mortality rates, cause-specific Australian population mortality rates for these four conditions were subtracted from the Australian all causes mortality rates. To estimate the other causes mortality rates for quitters we assumed that the decline in risk for quitters relative to smokers could be estimated by the function estimating the decline in risk for all causes mortality. If this decline in risk has been overestimated, we would have underestimated the number of quitters who would die of other causes and hence overestimated the benefit of quitting with respect to other causes mortality. However we would have correspondingly underestimated the benefit of quitting with respect to the incidence of the four smoking-related diseases and their health care costs, as the quitters would be at greater risk of these diseases through their reduced risk of death from other causes. This highlights the "competing risk" aspect of the model.
The QBM parameter estimates for disease probability, case fatality, disease cost and disease utility parameter estimates were based on the best available data. With the exception of COPD, the disease probability and case fatality data came either from Australian population-based studies or routine data collections, and the parameter estimates should therefore be reasonably accurate. The COPD incidence and fatality data were derived from Australian data using modelling and are therefore inherently less reliable. However, better estimates are unlikely to be available without a large scale prospective study because of the typically insidious onset of COPD.
Of the health care cost estimates, the stroke data came from the most reliable source, a comprehensive, prospective, population-based study[32
]. The AMI and lung cancer health care cost estimates were both based on Australian hospitalisation data, and are therefore under-estimates. There were no Australian data on the costs of managing patients with COPD, so data from the Canadian arm of an international study of COPD were used,[35
] as Canada has a similar health care system to Australia. The disease utility estimates for lung cancer, stroke and AMI came from an international database and the stroke utility estimate came from a meta analysis.
So, in summary, the QBM under-estimates the benefits of quitting, as the model only considers the impact of quitting on a sub-set of the adverse health effects of smoking and the assumptions underpinning the parameter estimates were mostly conservative.
In this paper we have used the QBM to produce summaries of the benefits of quitting for an individual and for a randomly selected cohort of 1000 quitters. The model is also a tool that can be easily used to evaluate the consequences and cost-effectiveness of quitting programs. To estimate the impact of a quitting program on a particular outcome, such as QALYs, the number of quitters for each age-group and sex category is simply multiplied by the predicted difference between smokers and quitters for that outcome, and the products are summed over all age-sex categories. The QBM can also easily be adapted to incorporate different epidemiologic data. Key input tables for the QBM can be stored in a set of linked EXCEL spreadsheets, and imported into the TreeAge model to form a package. These input tables can be readily updated from the base year of 2001, for example, when more recent data becomes available. Similarly, a set of incidence, mortality and cost data for a different population could be input to obtain predictions of the benefits of quitting for that population.
The QBM also has a variety of other potential applications. For example, one of the major current challenges for tobacco control is the differential in smoking rates associated with socio-economic status. In Australia, for example, people in the occupation category regarded as having the lowest socioeconomic status are almost three time more likely to smoke than those in the highest socioeconomic occupation category[1
]. The QBM could be used to compare the cost-effectiveness of tobacco control programs targeted at smokers of low socio-economic status with population-based programs.