Stomach Cancer

List of abbreviations
MST	Mean Sojourn Time
PC	Pre-Clinical Cancer
S	Susceptible
C	Clinical cancer

Disease Overview

Stomach/gastric cancer (GC) epidemiology and risk factors Gastric carcinogenesis is a multifactorial, multistep process. Host factors include blood group A, pernicious anemia, prior gastric surgery, family history, hereditary diffuse GC, and genetic syndromes. Smoking, salt, salty and smoked food, red meat, obesity, and low socioeconomic status are environmental factors. Moreover, infection with Helicobacter pylori and Epstein–Barr virus also play a role in gastric carcinogenesis. Information on these risk factors helps characterize individuals at risk of GC during their lifetime. Furthermore, identification of premalignant lesions is important for the purpose of screening and surveillance. Premalignant lesions of GC include atrophic gastritis (AG), intestinal metaplasia (IM), and dysplasia (DYS). It has been estimated that annually 0%–1.8%, 0%–10%, and 0%–73% of the patients with AG, IM, and dysplasia, respectively, progress to GC. The wide variations on the reported progression rates may result from differences in the study design, recruited population, and definitions. The Netherlands cohort study also revealed that premalignant lesions would progress to GC with an annual incidence of 0.2% from AG, 0.25% from IM, 0.6% from mild-to-moderate dysplasia, and 6% from severe dysplasia2.

Endoscopic surveillances in individuals with premalignant lesions may detect GC at an early and curable stage and therefore improve their survival. Prognosis of upper gastrointestinal cancer depends largely on disease stage at diagnosis. The survival rate is less than 10% when diagnosed at an advanced stage but is as high as 85% if detected at an earlier stage. Endoscopic screening can potentially prevent upper gastrointestinal cancers by early diagnosis and early treatment and has been widely adopted in screening programmes.

Stomach cancers tend to develop slowly over many years. Before a true cancer develops, pre-cancerous changes often occur in the inner lining (mucosa) of the stomach. These early changes rarely cause symptoms and therefore often go undetected. Most (about 90% to 95%) cancers of the stomach are adenocarcinomas. A stomach cancer or gastric cancer almost always is an adenocarcinoma. These cancers develop from the cells that form the innermost lining of the stomach (the mucosa). Correa pointed out that the human gastric carcinogenesis is a slow progressive, multistep, and multifactorial pathology process. The multistep process is composed of chronic superficial gastritis, atrophy gastritis, intestinal metaplasia (IM), dysplasia (DYS), and adenocarcinoma (malignant neoplasm).

GBD 2017 Modeling Strategy

The following information was obtained from the GBD 2017 fatal and non-fatal methods appendices [GBD-2017-YLD-Appendix-Stomach-Cancer], [GBD-2017-CoD-Appendix-Stomach-Cancer].

As with the majority of neoplasm causes in GBD, a remission rate for stomach cancer is not explicitly modeled. Prevalence for all cancers is estimated for a maximum of ten years after incidence meaning if a person survives for more than 10 years, he/she is considered ‘cured’. In other words, a surviving person will no longer be a prevalent case 10 years after incidence and does not have excess mortality and morbidity/disability from stomach cancer.

Total yearly prevalence of stomach cancer is split into four sequelae (see page 310-312 of YLD appendix):

1. Diagnosis and primary therapy:

• time from onset of symptoms through to the end of treatment

• assumed 5.2 month duration for stomach cancer

• Disability weight of 0.288 (0.193, 0.399)

1. Controlled phase:

• time between end of primary treatment and earlist of either: cure (defined as recurrence- and progression-free survival after ten years), death from another cause, or progression to the metastatic phase

• duration calculated based on remainder of time after attributing other sequelae.

• Disability weight of 0.049 (0.031, 0.072)

1. Metastatic phase:

• time period of intensive treatment for metastatic disease

• assumed 3.88 month duration for stomach cancer

• Disability weight of 0.451 (0.307, 0.600)

1. Terminal phase:

• 1 month prior to death

• Disability weight of 0.540 (0.377, 0.687)

Note

The disability weights for these sequelae phases are the same across all GBD neoplasms (excluding specific cancers with additional sequelae).

GBD neoplasm models rely on mortality incidence ratios (MIRs), which are estimated in a separate modeling process. According to the GBD modeler, MIRs should be retrieved from the GBD cancer modeler and not calculated from GBD estimates of location-specific incidence and moratlity rates. The fatal estimates are modeled first and then the MIRs are used to model the incidence estimates.

Note

The GBD modeler mentioned that for specific locations, the input data may be primarily cancer incidence registries, although it is possible that the GBD incidence estimates may not align with the incidence input data due to this modeling process.

Covariates used in the fatal stomach cancer model for GBD 2017 included page 189 in YLL/CoD appendix):

Level 1: diet high in sodium +, cumulative cigarettes + (5, 10, 15, and 20 years), smoking prevalence +, tobacco + (cigarettes per capita), log-transformed SEV scalar: Stomach C +, SEV unsafe water +, SEV unsafe sanitation +

Level 2: vegetables adjusted (g) -, fruits adjusted (g) -, mean BMI +, sanitation (proportion with access) -, improved water source (proportion with access) -, healthcare access and quality index -

Level 3: Education (years per capita) -, LDI ($ per capita) 0, socio-demographic index 0

Stomach Cancer ICD Codes used for GBD 2017

ICD 10	ICD 9
C16-C16.9, D00.2, D13.1, D37.1	151-151.9, 211.1, 230.2

Cause Hierarchy

Restrictions

The following table describes any restrictions in GBD 2017 on the effects of this cause (such as being only fatal or only nonfatal), as well as restrictions on the ages and sexes to which the cause applies.

GBD 2017 Cause Restrictions

Restriction Type	Value	Notes
Male only	False
Female only	False
YLL only	False
YLD only	False
YLL age group start	age_group_id = 8	15-19 years
YLL age group end	age_group_id = 235	95+ years
YLD age group start	age_group_id = 8	15-19 years
YLD age group end	age_group_id = 235	95+ years

Vivarium Modeling Strategy

Scope

This Vivarium modeling strategy is intended to simulate stomach cancer incidence/morbidity as well as mortality so that it reflects the estimates and assumptions of GBD. Additionally, this cause model intends to allow for the differentiation of preclinical screen-detectable (asymptomatic) phase of stomach cancer and the clinical (symptomatic) phase of stomach cancer.

Assumptions and Limitations

1. This model will assume the existence of a “recovered” cause model state in an attempt to be consistent with the GBD assumption that no morbidity due to stomach cancer occurs more than ten years past incidence of the clinical phase of stomach cancer. The assumption also asserts that there is no recurrance of stomach cancer.

2. This model assumes that the GBD incidence rate corresponds to the incidence of all pre-clinical asymptomatic stomach cancer (PC state) rather than symptomatic clinical stomach cancer arising from symptomatic presentation at the doctor’s office. This assumption has a few notable downstream limitations, including:

• simulation incidence of clinical stomach cancer will lag slightly behind forecasted incidence of stomach cancer due to the mean sojourn time period delay

• this should not cause too much trouble for stomach cancer as we assume a short mean sojourn time (<1 year)

Todo

think more about these assumptions in relation to the sojourn time

3. For stomach cancer, we are assuming there is a 5% H. pylori screening coverage in the insured population (double check if we want to bake this into the general population or we create an insured population for our baseline scenario?). For now, we assume that the insured population is the general GBD population. There is no endoscopy screening in theinsured population/general GBD population as there is no reliable data for this, hence all the cancers in the general population would have been detected as symptomatic clinical presentations. This means there are no PC states in the general population. Therefore prevalence_S, general population = 1 - prevalence_c414

4. In our simulation, we will model th pre-clinical state because we will introduce endoscopy screening which will detect pre-clinical canceers and move pay-out forward (earlier). The prevalence of pre-clinical (the pre-clincial cancer is screen-detectable for stomach cancer) is assumed to be equal to incidence of pre-clinical cancer (per assumption #2, this would be i_c414 which is the GBD incidence rate among the susceptable population) x duration in the pre-clinical state which is the mean sojourn time (MST).

5. In our simulation, we will intialize the model with nobody in the C state in line with the insured population: people cannot get insurance if they already know they have cancer.

Cause Model Diagram

This causal diagram reflects the simulation population which is different from the general population (GBD). The simulation population reflects as close as possible the insured population. The simulation population has addition PC state and will be initlized with no one in the C state (the general population refers to the GBD population). We assume there is a 5% H. pylori coverage in the general population, and is the same in the simulation population.

State and Transition Data Tables

State Definitions

State	State Name	Definition
S	Susceptible	Without cancer condition (may have pre-cancer states)
PC	Pre-clinical asymptomatic cancer, endoscopy detectable	With asymptomatic condition, detectable through endoscopy screening, will progress to clinical symptomatic phase
C	Clinical stomach cancer	With symptomatic condition
R	Recovered	Without condition; not susceptible

States Data

State	Measure	Value	Notes
S	prevalence	1 - prev_PC - prev_C - prev_R	Note: we assume no initial prevalence in C or R state (prev_C and prev_R =0)
S	birth prevalence	0
S	excess mortality rate	0
S	disabilty weights	0
PC	prevalence	prev_PC = \(\frac{\text{i\_pc}}{\text{(1 - prev\_c414)}}\) x MST	we scale prev_PC by 1-prev_c414 to account for 0 prevalence in the C state at initialization so as to preserve the ratio of people without stomach cancers to people with pre-clinical cancers.
PC	birth prevalence	0
PC	excess mortality rate	0
PC	disability weights	0
C	prevalence	0	No clinical cancers at initialization because those with clinical cancers cannot purchase insurance and are therefore not included in our sim population. See assumption #5.
C	birth prevalence	0
C	excess mortality rate	csmr_c414 / prev_c414
C	disabilty weights	\(\displaystyle{\sum_{s\in\text{s\_c414}}}\scriptstyle{\text{disability\_weight}_s\,\times\,\frac{\text{prev}_s}{\text{prev\_c414}}}\)	Total stomach cancer disability weight over all sequelae with IDs s248, s249, s250, s251
R	prevalence	0	No initialization into recovered state
R	birth prevalence	0
R	excess mortality rate	0	No excess mortality in recovered state assumed
R	disabilty weights	0	No long term disability in recovered state assumed

Transition Data

Transition	Source	Sink	Value	Notes
i_pc	S	PC	\(\frac{\text{i\_c414*}}{\text{(1 - prev\_c414)}}\)	*draw at age ‘current age + MST’
i_c	PC	C	1/MST per person-year	See MST definition in table below
i_c414			GBD incidence
r	C	R	0.1 per person-year for each sex and age group	To be consistent with 10 year GBD assumption

Note

• we need to draw from i_c414 at current age + MST because otherwise we are making people get clinical cancer a period of +MST older than they would have otherwise by giving them the pre-clinical cancer first with i_c414 and then waiting MST time to get clinical cancer. To keep clinical cancer incidence consistent with the right age groups, we can draw the incidence rates for preclinical cancer from the future- age group MST-time older than the stimulants current age. This depends on what duration of MST we end up using- if its shorter than the time incidence rates increase (1 year?), then we might not need to add this period.

Data Sources

Measure	Sources	Description	Notes
prevalence_c414	414_ets_prevalence_scaled_logit_phi_89_minmax_3_1000_gbd19.csv	CSU stomach cancer prevalence forecasts	2020-2040; defined as proportion of population with condition
csmr_c414	414_ets_deaths_scaled_logit_phi_89_minmax_3_1000_gbd19.csv	CSU stomach cancer cause specific mortality rate forecast	2020-2040; defined as deaths per person-year in general population
incidence_rate_c414	414_ets_incidence_scaled_logit_phi_89_minmax_3_1000_gbd19.csv	CSU stomach cancer cause-specific mortality rate forecast	2020-2040; defined as incidence cases per person-year in general population
disability_weight_s{248, 249, 250, 251}	YLD appendix	Sequela disability weights	0.288 (0.193-0.145), 0.049 (0.031-0.072), 0.451 (0.307-0.6), 0.54 (0.377-0.687)
prevalence_s{248, 249, 250, 251}	GBD 2019, COMO, decomp_step=’step4’	stomach cancer sequelae prevalence	Not forecasted
MST	2.37 years (95%CI: 1.78 to 2.96); distribution of uncertainty at draw level	Mean sojourn time; duration of time between onset of the asymptomtic stomach cancer to the clinical phase	See below for instructions on how to sample and research background.

Mean Sojourn Time

The MST that Bae 2014 estimated for population of Korean men is 2.37 years (95%CI: 1.92 to 2.96). We will use a lower bound of 1.78 instead to make the distribution symmetrical. This means that the MST is 1.92-1.78 = 0.14 years shorter ~ 1.6 months for approximately 5% of the population.

Parameter for Use in Model:

This parameter is be sampled at the draw level from the distribution detailed below and should be applied universally to all simulants within that draw.

from scipy.stats import norm

# mean and 0.975-quantile of normal distribution for mean difference (MD)
mean = 2.37
q_975 = 2.96

# 0.975-quantile of standard normal distribution (=1.96, approximately)
q_975_stdnorm = norm().ppf(0.975)

std = (q_975 - mean) / q_975_stdnorm # std dev of normal distribution

# Frozen normal distribution for MST, representing uncertainty in the parameter
mst_distribution = norm(mean, std)

Reference:

• Bae et al. Mean Sojourn Time of Preclinical Gastric Cancer in Korean Men: A Retrospective Observational Study J Prev Med Public Health 2014;47:201-205

Validation Criteria

The incidence and prevalence of clinical stomach cancers in the general population should approximately validate to the GBD incidence and prevalence of stomach cancers. The mortality rates (CSMR and EMR) of stomach cancer should validate to those of GBD.

References

[GBD-2017-YLD-Appendix-Stomach-Cancer]

Pages 310-317 in Supplementary appendix 1 to the GBD 2017 YLD Capstone:

(GBD 2017 YLD Capstone) GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2018; 392: 1789–858. DOI: https://doi.org/10.1016/S0140-6736(18)32279-7

[GBD-2017-CoD-Appendix-Stomach-Cancer]

Pages 190-198 in Supplementary appendix 1 to the GBD 2017 CoD Capstone:

(GBD 2017 CoD Capstone) GBD 2017 Causes of Death Collaborators. Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2018; 392: 1736–88. DOI: http://dx.doi.org/10.1016/S0140-6736(18)32203-7