背景

烟草流行是中国乃至全球的公共卫生问题，估算吸烟归因和可避免疾病负担具有重要意义。人群归因分数（Population Attributable Fraction, PAF）是估算疾病负担的重要指标。全球疾病负担（Global Burden of Disease, GBD）研究中，吸烟的PAF估算方法经历了吸烟滞后法、Peto-Lopez 法及剂量反应关系法。GBD研究中，前两种方法采用国家或地区的相对危险度（Relative Risk, RR），剂量反应关系法采用全球统一剂量反应RR。在吸烟归因疾病中，脑卒中尤为严重。因此，有必要估算我国吸烟所致脑卒中归因疾病负担，并进一步开展方法学研究。《“健康中国2030”规划纲要》提出了20%的吸烟率目标，因此，估算达标时我国吸烟所致脑卒中的可避免死亡人数，有助于该战略目标的实现。

研究内容和目的

基于中国人群实证数据，采用集成模型的剂量反应关系法估算2000-2020年中国吸烟所致脑卒中的归因死亡人数，为控烟政策的制定提供依据；并与吸烟滞后法和Peto-Lopez 法进行比较，促进疾病负担方法学的发展；估算达到《“健康中国2030”规划纲要》提出的吸烟率目标时中国人群吸烟所致脑卒中可避免死亡人数，预测达标所获得的健康效益，助力战略目标的实现。

方法

数据来源

 吸烟流行率来源于1984-2018年中国烟草流行病学调查；RR值来源于课题组建立的基于7个中英文数据库自建库起至2022年9月10日的亚洲人群吸烟所致疾病的Meta数据库（简称“Meta数据库”）；死亡数据来源于《中国死因监测数据集》中2000-2020年分性别年龄别的脑卒中死亡数据；人口数据来源于国家统计局公布的全国人口普查及统计年鉴，并对年龄别作相应处理。2030年中国分性别年龄别的人口数据来源于联合国人口署发布的中国人口预测结果。

2.   统计分析

（1）2000-2020年中国人群吸烟所致脑卒中的归因死亡人数的估算

获取2000-2020 年的现在吸烟率和戒烟比例数据，拟合相应年份的每日吸烟量（Cigarette Per Day, CPD）以及戒烟年限（Quit Year, QY）分布。基于Meta数据库，采用集成法和指数函数分别拟合CPD-RR及QY-RR曲线。其中集成法是指采用均方根误差（Root Mean Square Error, RMSE）将幂函数、综合暴露反应（Integrated Exposure Response, IER）模型等10个模型集成为一个模型，并采用Lasso回归和Meta回归筛选研究间异质性的六项协变量，将有统计学意义的协变量纳入CPD-RR曲线。结合中国脑卒中死亡人数估算归因死亡人数。

采用GBD研究的Meta回归贝叶斯、正则化、修剪（Meta-Regression—Bayesian, regularised, trimmed, MRBRT）更新法（简称为“MRBRT更新法”）以及传统的剂量反应Meta分析法（简称为“传统法”）进行敏感性分析。并将估算结果与GBD2021研究作比较。

（2）估算2000-2020年中国人群吸烟所致脑卒中归因死亡人数的方法学比较

采用滞后5年和10年的总吸烟率，以及二分类的RR值计算吸烟滞后法PAF，吸烟影响比（Smoking Impact Ratio, SIR）和每SIR单位RR值计算Peto-Lopez 法PAF，进而估算2000-2020年中国人群吸烟所致脑卒中的归因死亡人数，与第一章的结果进行比较。

（3）2030年达标时中国人群吸烟所致脑卒中可避免死亡人数的估算

设置自然场景和目标场景，自然场景采用自回归差分移动平均（Auto-regressive Integrated Moving Average, ARIMA）模型预测2030年中国总吸烟率和现在吸烟率，其差值为戒烟比例，目标场景以《“健康中国2030”规划纲要》的吸烟率目标为现在吸烟率，采用自然和目标现在吸烟率的比值作为自然和目标总吸烟率的比值，以此计算目标戒烟比例。结合第一章的CPD-RR和QY-RR曲线及相应分布，计算两种场景的PAFs和不可归因PAFs。基于2000-2020年中国人群脑卒中死亡人数和归因死亡人数计算不可归因死亡数，采用ARIMA模型预测2030年的不可归因死亡数，结合两种场景的不可归因PAFs获得脑卒中死亡数，与各自场景的PAF相乘得归因死亡数，两种场景的归因死亡数差值即为可避免死亡人数。

结果

（1）2000-2020年中国人群吸烟所致脑卒中的归因死亡人数

研究期间，合计PAF基本稳定在10.00%-11.00%左右，归因死亡人数在2000年和2020年分别为15.26万和22.30万。其中男性 PAF从2000的19.10%降至2020年的17.35%，归因死亡人数在2000年和2020年分别为14.04万和21.32万；女性PAF值从2000年的2.05%降至2020年的1.02%，归因死亡人数在2000年和2020年分别为1.21万和0.98万。基于MRBRT更新法和传统法拟合CPD-RR曲线估算的归因死亡人数与集成法没有统计学差异。

GBD2021研究的CPD-RR曲线高于本研究的曲线，PAFs基本稳定在15.63%-17.24%，高于本研究同年份5-6个百分点，归因死亡人数从2000年的32.21万上升至2020年的40.28万，高于本研究结果11.00-18.00万。

（2）2000-2020年中国人群吸烟所致脑卒中归因死亡人数的方法学比较

剂量反应关系法估算的归因死亡人数低于吸烟滞后法和Peto-Lopez法，后两类方法的结果接近，各方法均在2000年出现最小值（剂量反应关系法和后两类分别为15.26万和22.00万左右），2005-2020年剂量反应关系法估算结果稳定在20-23万、后两类方法稳定在31.00-37.00万左右。男性占总体归因死亡人数的多数。

（3）2030年达标时中国人群吸烟所致脑卒中的可避免死亡人数

自然情景下，预计到2030年中国人群的现在吸烟率和戒烟比例分别为21.10%和10.22%。目标情景下的戒烟比例为9.69%。男性在自然情景和目标情景的现在吸烟率分别为40.45%和38.34%，远高于女性的1.38%和1.31%，男性在自然情景和目标情景的戒烟比例分别为19.18%和18.18%，女性则分别为1.09%和1.03%。

预计到2030年，估算的自然场景和目标场景的归因死亡人数分别为184,769和175,171人，总体脑卒中可避免死亡人数为9,598人，其中男性可避免死亡人数为9,109人，女性为489人。

结论

（1）本研究基于剂量反应关系法估算的吸烟归因脑卒中死亡人数为15.00-23.00万人左右，提示我国吸烟所致脑卒中的疾病负担形势依然严峻，应进一步加强控烟政策的落实。

（2）相较于吸烟滞后法和Peto-Lopez法侧重累积效应，以CPD为吸烟指标的剂量反应关系法可能更侧重于吸烟所致脑卒中的短期效应，因而剂量反应关系法估算短期效应疾病的结果低于吸烟滞后法和Peto-Lopez法。

（3）2030年，若我国的现在吸烟率达到《“健康中国2030”规划纲要》的战略目标将避免近万人的脑卒中死亡，可见控烟所带来的健康效益，未来应加强控烟工作的落实，以助力战略目标的实现。

Background

Tobacco epidemic is public health issues in China and worldwide. Estimating the attributable and avoidable burden of disease of smoking are of great significance. The Population Attributable Fraction (PAF) is a key indicator for estimating burden of disease. In the Global Burden of Disease (GBD) studies, methods for estimating the PAF of smoking have evolved through the lagged smoking method, the Peto-Lopez method, and the dose response relationship method. In the GBD study, the first two methods adopt region-specific Relative Risk (RR), while the dose response relationship method uses global RR curve. Among smoking-attributable diseases, stroke is particularly serious. It is essential to estimate Chinese smoking-attributable stroke burden and to further conducting methodological research. In the outline for the “Healthy China 2030" initiative, which sets a national goal of reducing smoking prevalence to 20%, estimating the number of stroke deaths attributable to smoking that could be avoided upon achieving the target will help promote the realization of this strategic goal.

Research contents and objectives

Based on the empirical data from Chinese population, this study employed dose response relationship method with ensemble model to estimate the smoking attributable stroke deaths in China from 2000 to 2020, providing scientific evidence to formulate tobacco control policies; and compare it with the lagged smoking method and the Peto-Lopez method, to promote the development of disease burden methodology; then  estimated the smoking avoidable stroke deaths if China achieves the smoking rate targets proposed in the outline for the “Healthy China 2030" initiative, providing health benefit information for helping achieving this strategic goal.

Methods

1.  Data sourse

The smoking exposure data came from epidemiological surveys of smoking in the Chinese population conducted from 1984 to 2018. The RR values were derived from a Meta database of smoking-related diseases in Asian populations established by the research group, based on seven Chinese and English databases from the inception to September 10, 2022 (referred to as the "Meta database"). Mortality data were taken from stroke mortality data by sex and age from 2000-2020 in the "China Cause of Death Monitoring Dataset." Population data were obtained from National Population Censuses published on the National Bureau of Statistics website, as well as Statistical Yearbooks, and took corresponding treatments for age groups. The sex- and age-specific population data for China in 2030 came from Chinese population projection results released by the United Nations Population Division.

Statistical analysises

Estimation of Smoking Attributable Stroke Deaths among the Chinese Population from 2000 to 2020

Current smoking rates and quit ratios from 2000 to 2020 were obtained. Cigarettes Per Day (CPD) and Quit Years (QY) distributions for the corresponding years were fitted. Base on the Meta database, we applied an ensemble method and exponential functions to fit the dose-response relationship curves for CPD-RR and QY-RR, respectively. The ensemble method involved integrating 10 models, including power functions and Integrated Exposure Response (IER) models and so on, into a single model using Root Mean Square Error (RMSE). Statistically significant indicators from six covariates that might influence heterogeneity between studies were screened using Lasso regression and Meta regression, to ultimately fit the CPD-RR curve. We estimated the Chinese smoking attributable stroke deaths from 2000 to 2020, combined with stroke mortality data for the Chinese population.

Sensitivity analysis was conducted using the Meta-Regression-Bayesian, regularized, trimmed (MRBRT) update method (referred to as the "MRBRT update method") employed by the GBD research, as well as the traditional dose-response Meta analysis method (referred to as the "traditional method"). Simultaneously, the estimation results were further compared with the findings from the GBD2021 study.

Comparison of Methods for Estimating Smoking Attributable Stroke Deaths in the Chinese Population from 2000 to 2020

Using 5 years lagged and 10 years lagged smoking prevalence, as well as dichotomous RR values, to calculate the lagged smoking method PAF; and the Smoking Impact Ratio (SIR) and the RR value per SIR unit were used to calculate Peto-Lopez method PAF, then the smoking attributable stroke deaths in the Chinese population from 2000 to 2020 were further estimated, and compared the results with those estimated by the dose-response relationship method in Chapter One.

Estimating the Smoking Avoidable Stroke Deaths in the Chinese Population when Reaching Targets in 2030

The natural scenario and target scenario were set up. In the natural scenario, the Auto-regressive Integrated Moving Average (ARIMA) model was used to predict the total smoking rate and current smoking rate of the Chinese population in 2030. The difference between these two rates represented the quit ratio. For the target scenario, the smoking rate target (20.0%) from the outline for the “Healthy China 2030” initiative was used as the current smoking rate. The ratio between the natural and target current smoking rates was applied to calculate the ratio between the natural and target total smoking rates, which was then used to calculate the target quit ratio. Combined with the CPD-RR and QY-RR curves fitted in Chapter One, as well as correspending distributions, the PAFs and non-attributable PAFs for both scenarios were calculated. Based on the total number of stroke deaths and the smoking attributable stroke deaths among the Chinese population from 2000 to 2020, calculated the number of non-attributable deaths for corresponding year. The ARIMA model was used to predict the non-attributable stroke deaths of Chinese population in 2030. By incorporating the non-attributable PAFs from both scenarios, stroke deaths were obtained. These were then multiplied by the respective PAFs to derive attributable deaths. The difference in attributable deaths between the two scenarios represented the number of avoidable deaths.

Results

Estimating the Smoking Attributable Stroke Deaths in Chinese Populations from 2000 to 2020

The overall PAF remained relatively stable at around 10.00% to 11.00%, with attributable deaths recorded at 152.6 thousand in 2000 and 223.0 thousand in 2020. Among males, the PAF decreased from 19.10% in 2000 to 17.35% in 2020, with the attributable deaths being 140.4 thousand in 2000 and 213.2 thousand in 2020. For females, the PAF value dropped from 2.05% in 2000 to 1.02% in 2020, with attributable deaths of 12.1 thousand in 2000 and 9.8 thousand in 2020. The attributable death estimates based on CPD‑RR curves fitted using the MRBRT updated method and traditional method showed no statistical difference from the ensemble method.

The CPD-RR curve in the GBD2021 study is significantly higher than the curve in this study. The PAFs of the GBD study remained relatively stable at 15.63% to 17.24%, which is 5 to 6 percentage points higher than those of our study in the same years; the number of attributable deaths increased from 322.1 thousand in 2000 to 402.8 thousand in 2020, with results for each year being 110.0 to 180.0 thousand higher than those of our study.

Comparison of Different Methods for Estimating Smoking Attributable Stroke Deaths in the Chinese Population from 2000 to 2020

The estimated number of attributable deaths using the dose-response relationship method is lower than that from the lagged smoking method and the Peto-Lopez method, with the results from the latter two methods being similar. All methods reached their minimum values in 2000 (approximately 152.6 thousand for the dose-response relationship method and around 220.0 thousand for the other two methods). From 2005 to 2020, the estimates from the dose-response relationship method remained stable at around 200.0 thousand to 230.0 thousand, while the other two methods stabilized at 310.0 thousand to 370.0 thousand. Males account for the majority of the overall attributable deaths.

Estimating the Smoking Avoidable Stroke Deaths in the Chinese Population when Reaching Targets in 2030

In the natural scenario, the current smoking rate and smoking cessation rate in Chinese population were projected to be 21.10% and 10.22% respectively by 2030. In the target scenario, the smoking cessation rate was 9.69%. For males, the current smoking rates in the natural and target scenarios were 40.45% and 38.34%, respectively, which were much higher than females' rates of 1.38% and 1.31%. The smoking cessation rates for males in the natural and target scenarios were 19.18% and 18.18% respectively, while for females were 1.09% and 1.03% respectively.

It was estimated that by 2030, the attributable deaths in the natural scenario and target scenario will be 184,769 and 175,171 respectively, with a total of 9,598 avoidable stroke deaths. Among these, 9,109 avoidable deaths were male and 489 were female.

Conclusions

(1) Based on the dose-response relationship method, this study estimated that approximately 150.0 thousand to 230.0 thousand stroke deaths were attributable to smoking, highlighting the persistently severe burden of disease of smoking attributable stroke in China. The implementation of tobacco control policies should be further strengthened.

(2) Compared to the lag smoking method and the Peto-Lopez method, which emphasized cumulative effects, the dose-response relationship method using CPD as a smoking indicator might have focused more on the short-term effects of smoking attributable stroke. Therefore, the results estimated by the dose-response relationship method for short-term effect diseases were lower than those from the lag smoking method and the Peto-Lopez method. 

(3) By 2030, if the Chinese smoking rate reach to the strategic target set in the outline for the “Healthy China 2030" initiative, nearly 10.0 thousand stroke deaths could be avoided. This demonstrated the health benefits of tobacco control. Further efforts should be strengthened the implementation of tobacco control measures, to help achieve the strategic goal.