Smoking Risk Effects

Risk Overview

Todo

Provide a brief description of the risk, including potential opportunities for confounding (factors that may cause or be associated with the risk exposure), effect modification/generalizability, etc. by any relevant variables. Note that literature reviews and speaking with the GBD risk modeler will be good resources for this.

GBD 2017 Modeling Strategy

[GBD-2017-Risk-Factors-Appendix-Smoking-Effects]

Note

This section will describe the GBD modeling strategy for risk effects. For a description of GBD modeling strategy for risk exposure, see the risk exposure page.

The smoking risk factor affects several outcomes in GBD, including tuberculosis, lower respiratory tract infections, oesophageal cancer, stomach cancer, bladder cancer, liver cancer, laryngeal cancer, lung cancer, breast cancer, cervical cancer, colorectal cancer, lip and oral cancer, nasopharyngeal cancer, other pharyngeal cancer, pancreatic cancer, kidney cancer, leukaemia, ischaemic heart disease, ischaemic stroke, haemorrhagic stroke, subarachnoid haemorrhage, atrial fibrillation and flutter, aortic aneurysm, peripheral arterial disease, chronic obstructive pulmonary disease, other chronic respiratory diseases, asthma, peptic ulcer disease, gallbladder and biliary tract diseases, Alzheimer disease and other dementias, Parkinson disease (protective), multiple sclerosis, type‐II diabetes, rheumatoid arthritis, low back pain, cataracts, macular degeneration, and fracture.

Notably, the relative risks for the smoking risk factor in GBD are defined separately for current smokers (intensity of smoking) and former smokers (time since quitting). The TMREL for the smoking risk factor is never smokers.

The relative risks for the smoking risk factor cannot be pulled using standard tools. Rather, filepaths to the relative risk data (both for current and former smokers) are available in this excel document.

The relative risk data for the smoking risk factor are defined as a continous risk curve. This curve is modeled according to “mesh points” that are documented in the above excel file and the risk curve is assumed to connect linearly between each mesh point.

Smoking-Affected Cause IDs

Outcome	Outcome type	Outcome ID	Affected measure	Note
Tuberculosis	Cause	297
Lower respiratory tract infections	Cause	322
Oesophageal cancer	Cause	411
Stomach cancer	Cause	414
Bladder cancer	Cause	474
Liver cancer	Cause	417		Not most detailed cause
Laryngeal cancer	Cause	423
Lung cancer	Cause	426
Breast cancer	Cause	429
Cervical cancer	Cause	432		Female only cause
Colorectal cancer	Cause	441
Lip and oral cancer	Cause	444
Nasopharyngeal cancer	Cause	447
Other pharyngeal cancer	Cause	450
Pancreatic cancer	Cause	456
Kidney cancer	Cause	471
Leukaemia	Cause	487		Not most detailed cause
Ischaemic heart disease	Cause	493
Ischaemic stroke	Cause	495
Haemorrhagic stroke (Intracerebral hemorrhage)	Cause	496
Subarachnoid haemorrhage	Cause	497
Atrial fibrillation and flutter	Cause	500
Aortic aneurysm	Cause	501
Peripheral arterial disease	Cause	502
Chronic obstructive pulmonary disease	Cause	509
Other chronic respiratory diseases	Cause	520
Asthma	Cause	515
Peptic ulcer disease	Cause	527
Gallbladder and biliary tract diseases	Cause	534
Alzheimer disease and other dementias	Cause	543
Parkinson disease	Cause	544		Smoking exposure is protective for this cause
Multiple sclerosis	Cause	546
Type‐II diabetes	Cause	976
Rheumatoid arthritis	Cause	627
Low back pain	Cause	630		YLD only cause
Cataracts	Cause	671		YLD only cause
Macular degeneration	Cause	672		YLD only cause
Fracture	Cause			YLD only cause

Vivarium Modeling Strategy

Note

This section will describe the Vivarium modeling strategy for risk effects. For a description of Vivarium modeling strategy for risk exposure, see the risk exposure page.

Risk Outcome Relationships for Vivarium

Outcome	Outcome type	Outcome ID	Affected measure	Note
Lung cancer	Cause	426	Incidence
Ischemic heart disease	493	Cause-specific mortality rate	For the smoking related mortality model
Chronic obstructive pulmonary disease	509	Cause-specific mortality rate	For the smoking related mortality model

Lung Cancer Incidence

See the relevant documentation for the lung cancer cause model and the forecasted smoking risk exposure model.

Incidence Relative Risk Data

The lung cancer relative risks cannot be pulled using get_draws or other standard tools.

The mesh points for the relative risk curves for current smokers can be found here: /home/j/WORK/05_risk/risks/TEAM/sub_risks/tobacco/raw_data/metadata/rr/systematic_review_extraction_sheets/draws_for_PAF/426_lung_cancer/draws_pack.csv

The mesh points for the relative risk curves for former smokers can be found here: /home/j/WORK/05_risk/risks/TEAM/sub_risks/tobacco/raw_data/metadata/rr/systematic_review_extraction_sheets/draws_for_PAF/426_lung_cancer/draws_quit.csv

The following code demonstrates how to assign relative risk values to individual simulants based on their exposure values.

from scipy.interpolate import interp1d

"""
rr_i =: simulant's individual relative risk
smoking_status_i =: simulant's smoking status exposure
draw_x =: selected draw for a given model run
sex_i =: simulant's sex
age_group_i =: simulant's age group
pack_year_exposure_i =: simulant's pack year exposure value, if applicable
years_since_quitting_exposure_i =: simulant's years since quitting exposure value, if applicable
"""

if smoking_status_i.isin(['never','unassigned]):
  rr_i = 1

elif smoking_status_i == 'current':

  rr_current = pd.read_csv('/home/j/WORK/05_risk/risks/TEAM/sub_risks/tobacco/raw_data/metadata/rr/systematic_review_extraction_sheets/draws_for_PAF/426_lung_cancer/draws_pack.csv')
  rr_current_i = rr_current.loc[rr_current.draw=draw_X].loc[rr_current.sex_id==sex_i].loc[rr_current.age_group_id==age_group_i]
  x = rr_current_i.exposure.values
  y = rr_current_i.rr.values
  current_rr_function_i = interp1d(x, y)

  rr_i = current_rr_function_i(pack_year_exposure_i)

elif smoking_status_i == 'former':

  rr_former = pd.read_csv('/home/j/WORK/05_risk/risks/TEAM/sub_risks/tobacco/raw_data/metadata/rr/systematic_review_extraction_sheets/draws_for_PAF/426_lung_cancer/draws_quit.csv')
  rr_former_i = rr_former.loc[rr_former.draw=draw_X].loc[rr_former.sex_id==sex_i].loc[rr_former.age_group_id==age_group_i]
  x = rr_former_i.exposure.values
  y = rr_former_i.rr.values
  former_rr_function_i = interp1d(x, y)

  rr_i = former_rr_function_i(years_since_quitting_exposure_i)

Incidence PAF Calculation

The lung cancer PAF specific to an age, sex, location, and year demographic group for smoking should be calculated according to the following equation:

\[PAF_\text{a,s,l,y} = \frac{\overline{rr_\text{a,s,l,y}} - 1}{\overline{rr_\text{a,s,l,y}}}\]

Where, \(\overline{rr_\text{a,s,l,y}}\) is the mean value of relative risks for all simulants in a given age, sex, location, and year demographic group.

Application of Risk Factor

The smoking risk factor should affect the incidence rates of the preclinical and indolent lung cancer cause model states, \(incidence_\text{PC}\) and \(incidence_I\), respectively, using the same relative risk values such that:

\[incidence_\text{PC_i} = incidence_\text{PC} * (1 - PAF_\text{a,s,l,y}) * rr_i\]

And,

\[incidence_\text{I_i} = incidence_\text{I} * (1 - PAF_\text{a,s,l,y}) * rr_i\]

Where,

Parameter Definitions

Parameter	Definition	Note
\(incidence_\text{PC_i}\)	Individual simulant’s preclinical lung cancer incidence probability
\(incidence_\text{PC}\)	Population level incidence rate of preclinical lung cancer	As defined in the lung cancer cause model document
\(incidence_\text{I_i}\)	Individual simulant’s incidence lung cancer incidence probability
\(incidence_\text{I}\)	Population level incidence rate of indolent lung cancer	As defined in the lung cancer cause model document
\(PAF_\text{a,s,l,y}\)	Lung cancer PAF for smoking for simulant’s demographic group	As calculated in the Incidence PAF Calculation section
\(rr_i\)	Individual simulant’s relative risk value	Assigned as described in the Incidence Relative Risk Data section

Validation and Verification Criteria

While validating the exact application of the relative risks will be difficult to do with Vivarium simulation outputs, results should be stratified by smoking status and then, the incidence rates for lung cancer should be lowest for never smokers, higher for former smokers, and highest for current smokers.

Assumptions and Limitations

This model is limited in that it relies on the GBD relative risk structure that does not differentiate relative risks of former smokers by pack-year history so that current smokers with low pack-year histories who become former smokers may increase their risk of lung cancer by doing so, according to the GBD relative risk curves.

This model is limited in that it assumes the relative risk for smoking and lung cancer applies to preclinical and indolent lung cancer incidence rates equally. However, there is data that suggests that while indolent lung cancers occur at higher rates among smokers than nonsmokers, lung cancers are more likely to be indolent among non-smokers than among smokers, as tumor growth rates tend to be higher in smokers than never smokers [Mackintosh-et-al-2014].

As described in the forecasted smoking risk exposure model, some age groups will have unassigned smoking status exposures which we are assuming have a relative risk value of one. For groups with unassigned smoking status exposures, the smoking risk factor will not have an affect on lung cancer incidence.

Bias in the Population Attributable Fraction

As noted in the Population Attributable Fraction section of the Modeling Risk Factors document, using a relative risk adjusted for confounding to compute a population attributable fraction at the population level will introduce bias.

Todo

Outline the potential direction and magnitude of the potential PAF bias in GBD based on what is understood about the relationship of confounding between the risk and outcome pair using the framework discussed in the Population Attributable Fraction section of the Modeling Risk Factors document.

Smoking-Related Mortality

See the relevant documentation for the smoking related mortality model and the forecasted smoking risk exposure model.

Mortality Relative Risk Data

Relative risks for the causes included in the smoking related mortality model model are located at the filepath /home/j/WORK/05_risk/risks/TEAM/sub_risks/tobacco/raw_data/metadata/rr/systematic_review_extraction_sheets/draws_for_PAF/ (unless otherwise specified, as for COPD) and the subfolder specified in the table below.

File paths for relative risk data

Cause	Pack-years	Years since quitting
Stomach cancer (414)	414_stomach_cancer/draws_pack.csv	414_stomach_cancer/draws_quit.csv
Colorectal cancer (441)	441_colon_and_rectum_cancer/draws_pack.csv	441_colon_and_rectum_cancer/draws_quit.csv
COPD (509)	/share/gbd/WORK/05_risk/TEAM/sub_risks/tobacco/smoking_direct_prev/rr/modeling/outputs/decomp3/copd/draws_for_PAF/509_copd/draws_pack.csv	509_copd/draws_quit.csv
IHD (493)	493_ihd/draws_cig.csv	493_ihd/draws_quit.csv
Stroke (494)	494_stroke/draws_cig.csv	494_stroke/draws_quit.csv

The following code demonstrates how to assign relative risk values for a single cause to individual simulants based on their exposure values.

from scipy.interpolate import interp1d

"""
rr_i =: simulant's individual relative risk for a given cause
smoking_status_i =: simulant's smoking status exposure
draw_x =: selected draw for a given model run
sex_i =: simulant's sex
age_group_i =: simulant's age group
pack_year_exposure_i =: simulant's pack year exposure value, if applicable
years_since_quitting_exposure_i =: simulant's years since quitting exposure value, if applicable
"""

if smoking_status_i.isin(['never','unassigned']):
  rr_i = 1

elif smoking_status_i == 'current':

  rr_current = pd.read_csv({filepath})
  rr_current_i = rr_current.loc[rr_current.draw=draw_X].loc[rr_current.sex_id==sex_i].loc[rr_current.age_group_id==age_group_i]
  x = rr_current_i.exposure.values
  y = rr_current_i.rr.values
  current_rr_function_i = interp1d(x, y)

  rr_i = current_rr_function_i(pack_year_exposure_i)

elif smoking_status_i == 'former':

  rr_former = pd.read_csv({filepath})
  rr_former_i = rr_former.loc[rr_former.draw=draw_X].loc[rr_former.sex_id==sex_i].loc[rr_former.age_group_id==age_group_i]
  x = rr_former_i.exposure.values
  y = rr_former_i.rr.values
  former_rr_function_i = interp1d(x, y)

  rr_i = former_rr_function_i(years_since_quitting_exposure_i)

Mortality PAF Calculation

The PAF specific to a cause for an age, sex, location, and year demographic group for smoking should be calculated according to the following equation:

\[PAF_\text{c,a,s,l,y} = \frac{\overline{rr_\text{c,a,s,l,y}} - 1}{\overline{rr_\text{c,a,s,l,y}}}\]

Where, \(\overline{rr_\text{c,a,s,l,y}}\) is the mean value of relative risks for a given cause among all simulants in a given age, sex, location, and year demographic group.

Application of Risk Factor

The smoking risk factor should affect the mortality rate of the individual modeled causes, as defined in the smoking related mortality model documentation, such that:

\[mr_i = ACMR - CSMR_\text{c} + CSMR_\text{c} * (1 - PAF_\text{c,a,s,l,y}) * rr(c)_i\]

Where,

Parameter Definitions

Parameter	Definition	Note
ACMR	All-cause mortality rate
\(CSMR_\text{c}\)	Cause-specific mortality rate for cause c	Should use forecasted rates from 2020-2040 as documented on the smoking related mortality model document
\(PAF_\text{c,a,s,l,y}\)	PAF for smoking and cause c for simulant’s demographic group	As calculated in the Mortality PAF Calculation section
\(rr(c)_i\)	Individual simulant’s relative risk value for cause c	Assigned as described in the Mortality Relative Risk Data section

Validation and Verification Criteria

While validating the exact application of the relative risks will be difficult to do with Vivarium simulation outputs, results should be stratified by smoking status and then, the all cause mortality rates should be lowest for never smokers, higher for former smokers, and highest for current smokers.

Assumptions and Limitations

This model is limited in that it relies on the GBD relative risk structure that does not differentiate relative risks of former smokers by pack-year history so that current smokers with low pack-year histories who become former smokers may increase their risk of lung cancer by doing so, according to the GBD relative risk curves.

As described in the forecasted smoking risk exposure model, some age groups will have unassigned smoking status exposures which we are assuming have a relative risk value of one. For groups with unassigned smoking status exposures, the smoking risk factor will not have an affect on smoking-related mortality due to the causes covered in this document.

Bias in the Population Attributable Fraction

As noted in the Population Attributable Fraction section of the Modeling Risk Factors document, using a relative risk adjusted for confounding to compute a population attributable fraction at the population level will introduce bias.

Todo

Outline the potential direction and magnitude of the potential PAF bias in GBD based on what is understood about the relationship of confounding between the risk and outcome pair using the framework discussed in the Population Attributable Fraction section of the Modeling Risk Factors document.

References

Mackintosh-et-al-2014

Mackintosh JA, Marshall HM, Yang IA, Bowman RV, Fong KM. A retrospective study of volume doubling time in surgically resected non-small cell lung cancer. Respirology. 2014 Jul;19(5):755-62. doi: 10.1111/resp.12311. Epub 2014 May 6. PMID: 24797504. Available here.

Todo

Update the GBD 2017 Risk Factor Methods appendix citation to be unique to your risk effects page (replace ‘Risk-Effects-Model-Template’ with ‘{Risk Name}-Effects’)

Update the appropriate page numbers in the GBD risk factors methods appendix below

Add additional references as necessary

GBD-2017-Risk-Factors-Appendix-Smoking-Effects

Pages ???-??? in Supplementary appendix 1 to the GBD 2017 Risk Factors Capstone:

(GBD 2017 Risk Factors Capstone) GBD 2017 Risk Factor Collaborators. Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2018; 392: 1923-1994. DOI: https://doi.org/10.1016/S0140-6736(18)32225-6