Beyond Life Years Lost, Using Different Outcomes for Population Modelling in Tobacco Harm Reduction

The aggregate time lived with a disability across the six smoking associated diseases is modelled. The disease incidence for never smokers is estimated in the base year only, and age and smoking status are assessed at the start of the year. YLD=∑Disease Incidence RateGAS×Disability Weight×Average Duration of Disease $$YLD = \sum {\,Disease\,Incidence\,Rate{\,_{GAS}}\, \times \,Disability\,Weight\, \times \,Average\,Duration\,of\,Disease} $$ Disease Incidence RateGAS= NS Disease Incidence ProbabilityGA×HRGAS × PopulationGAS $$Disease\,Incidence\,Rat{e_{GAS}} = \,NS\,Disease\,Incidence\,Probability{\,_{GA}}\, \times \,HR{\,_{GAS}}\, \times \,Population{\,_{GAS}}$$ NS Disease Incidence ProbabilityGA=Expected Incidence RateGAPopulationGA × ∑s=19 PGAS × HRGAS $$NS\,Disease\,Incidence\,Probability\,_{GA} = {{Expected\,Incidence\,Rate_{GA} } \over {Population\,_{GA} \, \times \,\sum\nolimits_{s = 1}^9 {\,P\,_{GAS} \, \times \,HR\,_{GAS} } }}$$ where: NS

is Never smoker

Disability weights are taken from the Global Burden of Disease (GDB) Study and are factors used to represent the severity of health loss associated with a given health state or disability, where 0 is a full state of health and 1 is death (29). Average duration of disease is approximated using Little’s Law whereby the long-term average number of customers in a stationary system is equal to the long-term average effective arrival rate multiplied by the average time spent in the system. This relationship can be used to approximate disease duration using the current number of cases (disease prevalence) and the number of new cases (disease incidence) as detailed below.

Disease prevalence and incidence data from the GBD study indicate that year on year disease incidence rates have not fluctuated significantly over recent years, thus the RRP2 model can estimate the average time diagnosed with a disease until either successful treatment or death. Disease Duration=Disease PrevalenceDisease Incidence $$Disease\,Duration = {{Disease\,Prevalence} \over {Disease\,Incidence}}$$

The DALY combines into a single indicator the years lived with a disability from the smoking-related diseases model with the life years lost from premature mortality. DALY=YLD+YLL $$DALY = YLD + YLL$$ where: DALY

is Disability-adjusted life years

Using current estimates of disease incidence and the average disease duration, the RRP2 model projects the number of patients diagnosed with the disease. At each time step in the model, the number of diagnosed patients is estimated as the number of diagnosed patients in the previous time step, in addition to any new cases and minus any patients lost due to death or in remission. Diagnosed Patients(t)GA=Diagnosed Patients (t−1)GA + (Disease Incidence RateGA−Remission or Death RateGA) $$Diagnosed\,Patients\,(t)\,_{GA} \, = \,Diagnosed\,Patients\,(t - 1)\,_{GA} \, + \,(Disease\,Incidence\,Rate\,_{GA} - Remission\,or\,Death\,Rate\,_{GA} )$$ Remission or Death Rate(t)GA=Diagnosed Patients(t)GAAverage Disease Duration $$ Remission\,or\,Death\,Rate(t){\,_{GA}} = {{Diagnosed\,Patients\,(t){\,_{GA}}} \over {Average\,Disease\,Duration}}$$ Total Diagnosed Patients (t)=∑Diagnosed PatientsGA $$Total\,Diagnosed\,Patients\,(t) = \sum {Diagnosed\,Patients{\,_{GA}}} $$ Disease Prevalence(t)=Total Diagnosed Patients (t)Total Population(t) $$Disease\,Prevalence(t) = {{Total\,Diagnosed\,Patients\,(t)} \over {Total\,Population(t)}}$$ where: t

is timepoint t (in years)

Nearly all (90%) of smokers in Japan started before the age of 25, with approximately half of men and two-thirds of women first smoking prior to the legal age of 20 years (20, 21).

Observing the prevalence of “Ever” smoking (current or former smoker) in the 20–29 age cohort provides evidence on the trend in smoking initiation rates. An analysis of the recorded Ever smoking prevalence in NHNS survey data estimate that in 2004, 58.2% of males and 23.2% of females in the 20–29 age cohort had ever smoked. The estimated average annual decline in smoking initiation rates across the period was 3.1% for males and 2.6% for females. In our scenarios, all smoking initiation is assumed to occur at age 20.

Data from two waves (2019, 2020) of a cross-sectional survey of nicotine product use in Japan were used to estimate THP initiation (22, 23, 24). As the survey studied only adults, any THP initiation from never tobacco users prior to age 20 is not captured. To indirectly estimate THP initiation prior to age 20 we used the ever THP prevalence in the 20–24 age cohort and adjusted for initiation rates whilst in that cohort.

In our scenarios, all THP initiation is assumed to occur at age 20.

Quit probabilities were estimated by calibration of the model estimates to the observed smoking prevalence from the NHNS (4) by gender and age cohort over the period 2004–2019. The calibration method applied was to minimize the Mean Absolute Error over Mean (MAEoM) statistic. The calibrated quit probabilities are shown in Table 5.

These are the probabilities that a smoker reaches 12 months abstinence from smoking, though they may still subsequently resume smoking beyond one year.

Data from the first two waves (2019, 2020) of a cross-sectional survey to assess tobacco and nicotine product use since the introduction of THPs in Japan (22, 23, 24) were combined to generate estimates of the annual transition probabilities between tobacco products or to no tobacco product. The two waves consisted of separate cohorts of participants and combining the two provided a total of 1,937 current tobacco product users, 1,423 males and 514 females. This sample size was not sufficient to determine transition rates by age groups in addition to gender, so we utilize age-adjusted transition probabilities.

For both products, cigarettes and THP, individuals are defined as either never, current, or former users based on the following definitions.

• Never: Never tried the product or used less than 100 times in lifetime

• Current: Used the product 100+ times and used within last year

• Former: Used the product 100+ times but not used within last year

Users were asked about their current tobacco use status and that at 12 months ago. The compiled data were used to generate the transition matrix in Table 6 and Table 7. Data from the cross sectional surveys were weighted to account for oversampling in the 20–29 year-old age groups. The survey also provided prevalence estimates for the nine tobacco use statuses (Table 8).

The overall adult current smoking prevalence was 15.3% (males 24.3%, females 7.0%) and current THP use prevalence was 6.0% of adults (males 9.6%, females 2.6%). Lifetime smoking prevalence was 34.9% (male 55.6%, female 15.6%) and lifetime THP use was 6.6% of adults (male 10.7%, female 2.9%).

A similar population-based survey of tobacco use in Japan conducted in 2018 (25) estimated the age-adjusted THP prevalence as 5.0% (males 8.3%, females 1.95%). For lifetime THP use, the survey estimated 8.7% of adults (males 14.1%, females 3.7%).

The alternative population survey estimates of THP prevalence, both current use and lifetime use, were slightly higher than the survey data we utilised. One reason could be the definition of current user, where the alternative scenario included anyone who had used in the past 30 days, even if only once.

The model enhancement to project non-mortality indicators of population health impacts has required extra input data for the RRP2 model for the incidence, prevalence, and smokers’ excess risk of six tobacco-associated diseases (Table 9).

The first four diseases are those most traditionally associated with smoking. The last two diseases were chosen as representatives for diseases less customarily linked to smoking. These data have been drawn from the Global Burden of Disease study via the data portal hosted by Institute for Health Metrics and Evaluation (IHME) (26) at the University of Washington. For the RRP2 model, the data are used under a commercial licence, but the same data are freely available under a Creative Commons licence. The incidence, prevalence and smoking risk factor data used by the RRP2 model for these diseases are provided in Supplementary File 2.

Prior to comparing scenarios, the RRP2 model’s projection results of Japan’s demographics were assessed for realism by comparing them to independent projections. The data in Table 10 provides the RRP2 model demographic projections and the graphs in Figure 2, Figure 3 and Figure 4 compare these projections against those published by United Nations projections for Japan (28). The RRP2 demographic projections were within the 95% prediction interval of those from the United Nations.

Figure 2.

Projected population.

Figure 3.

Projected population by age group.

Figure 4.

Projected annual mortality rate.

Detailed tabular results are provided in Supplementary File 3. The number of cigarette smokers is projected to fall in all three scenarios (Figure 5). By 2030, there are still 12 million (12.0%) adult smokers in the Smoking Only scenario compared to 6.2 million (6.2%) in the Harm Reduction scenario. The number of adult smokers does not fall below one million until 2092 in the Smoking Only scenario, some 32 years after that in the Harm Reduction scenario.

Figure 5.

Adult current smokers.

By the end of the projection period there are a negligible number of smokers in any of the three scenarios.

Comparing use of any tobacco product (Figure 6) reveals a different picture. The total number of adult tobacco product users in the Harm Reduction scenario exceeds that in the Smoking Only scenario in all years after the date the THP was introduced. By 2100 there would still be 1.1 million adults using tobacco products, compared to just 733,000 in the Smoking Only scenario.

Figure 6.

Adults using tobacco products.

The number of adults that have ever smoked was similar in both the Smoking Only and Harm Reduction scenarios (Figure 7), with the number slightly higher in the Harm Reduction scenario.

Figure 7.

Adult ever smokers.

As the number of adult ever smokers is similar, but the number of current adult smokers is lower in the Harm Reduction scenario compared to Smoking Only, it indicates a smoking habit is sustained for a shorter period. This is supported by the comparison of the average age at which adults quit smoking (Figure 8).

Figure 8.

Average age at quitting smoking.

From 2030 onwards, the average age at quitting smoking was at least 7 years earlier for the Harm Reduction scenario. The reduction in the number of smokers is reflected in the consumption of cigarettes, where across 2004–2100 nearly 2.5 trillion fewer cigarettes were consumed in the Harm Reduction scenario (Figure 9).

Figure 9.

Cumulative cigarette consumption

The projected number of THP users will peak around 2030 at just over 8.3 million, around 8.2% of the adult population. After 2030, THP users gradually decline to one million by 2100. Figure 10 shows that most of the early adopters of THPs were cigarette smokers. After many of the smokers had switched completely to THPs, former smokers became the dominant smoking status of THP users. By 2100, 45.5% of THP users had never smoked, 43.1% used to smoke 11.4% still smoked.

Figure 10.

Smoking status of THP users.

Within each smoking status, the projected use of THP behaviour is shown in Figure 11.

Figure 11.

THP use within smoking status.

By 2100, half of any remaining cigarette smokers also use a THP. THP use within former smokers peaks at 20% around 2028, then declines to 10% from 2060 onwards. The prevalence of THP use within never smokers does not ever exceed 1%. Dual use of tobacco products peaks at 3.1 million adults (3.0%) around 2020.

Comparing premature mortality (Figure 12) even in the Complete Abstinence scenario, deaths associated with tobacco use will still occur right up to 2100. The decline in smoking attributable mortality is steepest in the Complete Abstinence scenario as would be expected, falling from nearly 100,000 deaths per years in 2014 to near zero by 2100.

Figure 12.

Tobacco-related deaths.

The decline in smoking-related deaths is faster in the Harm Reduction scenario, but by 2100 the decline has plateaued and is beginning to converge with the Smoking Only scenario mortality rate. Over the period 2004–2100 however, there are 850,000 fewer premature deaths than in the Smoking Only scenario. For comparison, over the same period in the Complete Abstinence scenario, there were 2.3 million fewer premature deaths.

The difference between attributable mortality amongst the three scenarios is indicated by the years of life lost statistic (YLL) (Figure 13).

Figure 13.

Years of life lost.

Compared to the Smoking Only scenario, 13.3 million fewer life years would be lost in the Harm Reduction scenario and 34.5 million saved if everyone were to cease tobacco product use in 2014.

Figure 14 contrasts the number of years lived with a disability and lower quality of life from the six tobacco associated diseases modelled.

Figure 14.

Years lost due to a disability.

Individuals diagnosed with each disease are allocated a disability weight which reflects the loss of quality of life with each disease. The profile is similar across the three scenarios from 2080 onwards. Prior to then a greater number of years lived with a disability was observed in the Smoking Only scenario, followed by the Harm Reduction scenario and the best of all three was the Complete Abstinence scenario. Accumulating the years lived with a disability over 2004 to 2100 produced a total of 25.2 million, 24.3 million and 22.9 million from the Smoking Only, Harm Reduction and Complete Abstinence scenarios, respectively.

The projected disability-adjusted life year (DALY) statistic (Figure 15), which is a combined measure of tobacco disease burden, merging the number of years lost due to disability or premature death displays a similar picture as that for YLD.

Figure 15.

Disability-adjusted life years (DALY).

Cumulatively across the period, the Smoking Only scenario totalled 99.3 million years, Harm Reduction scenario 85.1 million years and the Complete Abstinence scenario 62.6 million years.

Figure 16 shows the cumulative cases for each of the six specific diseases, expressed as a percentage reduction compared to the projected Smoking Only scenario.

Figure 16.

Disease cases.

The greatest fall was seen in lung cancer cases, the major disease most linked to smoking, where complete abstinence would achieve a 22.6% reduction and the Harm Reduction scenario a 7.9% reduction.

The number of people diagnosed with a smoking-related disease is lower across each of the diseases modelled for the Harm Reduction scenario, but still greater than in the Complete Abstinence scenario, with the greatest reductions seen in lung cancer and COPD (Figures 17–19).

Figure 17.

Tracheal bronchus and lung cancer patients.

Figure 18.

Chronic obstructive pulmonary disease patients.

Figure 19.

Heart disease patients.

Full details and outputs for sensitivity and scenario variants are contained in Supplementary File 4. Product transition sensitivity outputs are available for several endpoints and multiple THP relative risk scenarios have been investigated, thus results are too substantial to report in the main paper.

Model outputs were compared to observed data or alternative source projections for each of the key output indicators to check the RRP2 projections were consistent. In the majority of cases the projected data follows a similar trend to and are in line with verification data. For the full comparison, see Supplementary File 5.

With the introduction of THPs, by 2050 the average age of current smokers had fallen to 46 years, compared to 51.2 years in the Smoking Only scenario. This was reflected in the average number of years smoked in current smokers, which by 2050 had fallen by five years from 32.7 years to 27.5 years.

Although the average length of time spent smoking was shorter in the Harm Reduction scenario, a greater number of people had smoked at some point during their lives. The peak in this extra number of Ever-Smokers occurred around 2060, where 900,000 more people had smoked compared to the Smoking Only scenario. As we assumed only 50% of those that initiate THP would have smoked had the THP not been available, this would mean the overall rate of tobacco product initiation would be greater in the Harm Reduction scenario. As a proportion of those who initiate a THP as their first tobacco product would later transfer to smoking, a greater number of Ever-Smokers would be expected.

The RRP2 model projects that the number of THP users will peak around 2030 at 8.3 million and then start to fall year on year. By 2100, there would only be one million THP users. In 2030, 2.1% of THP users had never smoked, compared to 76.4% which had previously smoked. The transition from smoking to THP would result in a peak of THP use, as fewer smokers remain to transition to THP. By 2100, of the one million THP users, 45.5% had never smoked and 43.1% used to smoke and 11.4% were still smoking. Dual use prevalence reached 3% of the adult population in 2020 but fell in all subsequent years. Dual use appears to be a transitionary phase, as smokers adapt to complete switching to THPs. Data from the first two waves (2019, 2020) of a cross-sectional survey estimated that around a third of dual users (33% males, 39% females) transitioned to solus THP use each year, whilst around 7% (7% males, 6.6% females) transitioned to solus smoking (22, 23, 24). Incorporating the transitioning to no tobacco use would indicate that only 50% of current dual users would still be a dual user in the following year.

Over the period 2004–2100, the cumulative number of person-years lived with a disability from one of the six smoking-related diseases modelled (Table 8) fell from 25.2 million in the Smoking Only scenario to 24.3 million in the Harm Reduction scenario, a fall of 3.5%. The health risks from smoking do not disappear the day someone stops smoking. The duration of smoking, cumulative number of cigarettes smoked, and the elapsed time since quitting will influence the residual excess risk levels. In the Complete Abstinence scenario an 9.0% drop was projected in years lived with a disability.

In each of the six smoking associated diseases modelled, the annual disease incidence rate was lower in the ‘Harm Reduction’ scenario than the Smoking Only scenario, but higher than in the Complete Abstinence scenario. The cumulative reduction in incidence over the 2004–2100 period varied considerably by disease. Lung cancer showed the biggest fall at 7.9%, followed by COPD at 6.7% and esophageal cancer at 5.2%. The fall in disease cases for ischemic heart disease, stroke and lower extremity peripheral arterial disease were 1.0%, 0.9% and 2.6%, respectively.