Epidemiological and model structure characteristics of the simulation models
| Model | Model type Macrosimulation or microsimulation | Time component | Model steps through diseases/directly on mortality | Time trends in BAU epidemiological parameters: | Lag time (cessation to disease and mortality reduction) | Source of RR associating tobacco with disease or mortality | Heterogeneity in RR | Uncertainty | Validation and calibration | |||
| Time steps explicitly in model | Default time horizon (scenarios) | All-cause mortality rate/incidence rate/case fatality rate | Tobacco use | Probabilistic about input parameters | Discrete scenario and sensitivity analyses (examples) | |||||||
| Abridged SimSmoke | (Discrete first-order Markov process). Macrosimulation (I) | No | No (but do estimate future impacts; table 3) | Directly on mortality | No | No in default. (Future impacts do allow for trends in prevalence.) | No | Doll et al 28 | By country-level economic development (from tobacco use to mortality) | No | Bounds on effect size parameters of the different policies | Validation by examining projected prevalence from full SimSmoke and with national survey data for some countries |
| BENESCO | Markov model. Macrosimulation | Yes (annual) | Life-time (<100 years) | On mortality through four diseases (COPD, CHD, stroke and lung cancer) | No (I) | No | No | Howard et al 23 | Age, gender and smoking status for incidence RRs | Yes (1000 iterations)22 | No | No |
| BODE3 | Markov, proportional multistate life table model. Macrosimulation | Yes (annual) | Life-time (<110 years). Some papers also shorter (eg, <20 years) | On mortality through CHD, stroke, COPD, lower respiratory tract infection and multiple cancers: lung, oesophageal, stomach, liver, head and neck, pancreas, cervical, bladder, kidney, endometrial, melanoma and thyroid (with smoking protecting against the latter three cancers) | Trends in incidence and case fatality rates for each disease and in mortality rate for other than included disease | Yes, through rates of uptake and cessation | Ex-smoker disease risks decline with time since quitting | NZ Census-Mortality Study,68 American Cancer Society CPS-II | Age, sex and ethnicity | Intervention effect size, intervention costs and baseline parameters | Multiple scenario analysis | Validation by verification of cause-specific mortality rates |
| EQUIPTMOD | Markov state transition cohort model. Macrosimulation | Yes (annual) | Lifetime, and 2-year, 5-year and 10-year time horizons (I) | On mortality through four diseases: lung cancer, CHD, COPD and stroke | No (I) | No (I) | No (I) | Smoking to all-cause mortality: Doll, et al 28; smoking to health outcomes69 | Not reported | Probability distribution on natural history parameters, RRs and ORs, costs and utilities | None | None (I) |
| CA Higashi | Markov chain technique: smoking model and multistate life table model. Macrosimulation | Yes: annual cycle | Lifetime | On mortality through IHD, cerebrovascular accident, six cancers, COPD and lower respiratory tract infections | No for diseases and all-cause mortality (I) | Yes: the uptake and cessation rates were fitted to the observed prevalence at three time-points | No (I) | Mortality RR: CPS-II. Smoking to disease69–71 | Not reported | Yes, to provide the UI of incremental cost-effectiveness ratios (ICERs), 2000 iterations | Different scenarios pertaining to cost of non-smoking signboards were analysed, increase in tobacco excise taxes | Face validity and formal (verification) validity* |
| CA. Hoogenveen | Dynamic multistate model. Macrosimulation | Yes: annual | Lifetime (I) | On mortality through diseases: lung cancer, CHD, stroke, COPD, congestive heart failure, diabetes and cancer of the lung, stomach, larynx, oral cavity, oesophagus, pancreas, bladder and kidney | No (I) | Yes72 | Yes: for each disease the RR decreases according to a negative-exponential curve starting from the value for current smokers and converging to the value 1 | Appendix of van Baal et al 73 | Gender and age | No | No | Not reported |
| DYNAMO-HIA | Markov-based lifetable model. Macrosimulation | Yes | 25–50 years | Onto mortality and summary health measure through diseases: diabetes, IHD, stroke, lung cancer, oral cancer, oesophageal cancer, colorectal cancer, breast cancer and COPD | No (stated in study limitations) | No (stated in study limitations) | Yes: exponential decay from current to never smoker risk using Hoogenveen equations3 43 | CPS-II age-specific relative risks, 1982–1988 | By age and sex in all-cause mortality and by sex in IHD, stroke, diabetes mellitus, COPD, lung cancer, colorectal cancer, oral cancer, breast cancer and oesophageal cancer | No | No | No (I) |
| PMI | Markov chain state-transition model. Smoking prevalence is determined by microsimulation of individuals, and the health impacts are calculated at an aggregate level of sex by age group (and presumably future year), which is essentially a macrosimulation of health impacts | Yes: annual | 20 years | On mortality through lung cancer, IHD and stroke | No | Yes | Yes: for former cigarette smokers, the RR estimates by time since quitting are derived from the corresponding estimates for current smokers, assuming that the decrease in excess risk after quitting follows a negative exponential function | Meta-analyses of epidemiological data | Country, disease, sex, age | No | No | No (I) |
| SimSmoke | Discrete first-order Markov process. Macrosimulation | Yes | 50 | Directly on mortality74 | No (I) | Yes | Yes: for ex-smokers, RRs were assumed to decline with years since quitting at the rate observed in the USA51 52 | CPS-II | Age and gender | No | No | Yes: validation of prevalence with national survey data |
| Tobacco Policy | Dynamic computer simulation model. Macrosimulation (I) | Yes: annual | 21–75 years | Directly on mortality accounting for fertility, smoking and net migration | Yes (all cause mortality) | Yes56 | No (I) | Not given | Not given | No | Assumptions regarding change in smoking and mortality levels into future | Calibration against external estimates of smoking prevalence, population size and life expectancy |
| CA Vos | Deterministic Markov model, multistate life tables. Macrosimulation | Yes | Lifetime, 100 years | Onto mortality through diseases: lung, upper aerodigestive, pancreas, cervix, bladder and kidney cancers, COPD, other respiratory diseases, CHD, stroke, peripheral arterial disease, other cardiovascular disease and other tobacco-related causes | Yes, but unclear on which parameters | Yes | Yes: risk reversal based on years since quit among former smokers | CPS-II | Sex for all diseases and age and sex for IHD and stroke and all-cause mortality | Yes | No | No |
| CA Warner & Mendez | Dynamic simulation model. Macrosimulation | Yes (annual) | Lifetime (110 years) | Directly on mortality | Yes (death rates) | Yes62 | No | Estimated by authors, and CPS-II63 | Age, sex and smoking status63 | Yes63 | Based on assumptions on vaping induced initiation and cessation rates60 | Against national survey prevalence data64 |
| WHO-CHOICE | PopMod: Markov-based lifetable model. Macrosimulation | Yes (annual) | Lifetime (101 years) | Onto mortality through diseases: CHD, COPD, lung cancer, stroke | No (I) | No | No (I) | Murray and Lopez75 | Age and sex | Yes | Effect of change in prices and without age, weight and discounting | No (I) |
*Formal validity concerns how well an argument conforms to the rules of logic to arrive at a conclusion that must be true, assuming that the premises are true.76
†While some papers clearly state the SimSmoke model works directly onto all-cause mortality, we note that a SimSmoke Technical Report74 states: ‘Besides total smoking attributable deaths, SimSmoke distinguishes smoking attributable deaths due to lung cancer, heart disease, stroke and COPD’. (pp. 18) Therefore, we are unclear on the precise SimSmoke approach.
BAU, business as usual; BENESCO, Benefits of Smoking Cessation on Outcomes; BODE³, Burden of Disease Epidemiology, Equity & Cost-Effectiveness Programme; CA, Common author; CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; CPS, Cancer Prevention Study; DYNAMO-HIA, Dynamic Model of Health Impact Assessment; (I), inferred (not explicitly stated in papers); IHD, ischaemic heart disease; PMI, Philip Morris International; RR, relative risk; WHO-CHOICE, WHO-CHOosing Interventions that are Cost Effective.