Harvard-GC (BCH) Boston Children’s Hospital

Summary

The Harvard-GC model is a population-based microsimulation model designed to simulate gastric cancer incidence and mortality across diverse U.S. populations. It operates on an annual cycle and simulates individuals from birth, tracking their progression through precancerous and cancerous stages. The model simulates cancers by site (cardia vs non-cardia) and histology group. The model simulates adenocarcinomas (intestinal and diffuse) via the Correa cascade, and also models MALT lymphoma, non-MALT lymphoma, GIST, NETs, and other rare subtypes. This broad scope enables the model to project tumor-type-specific and overall gastric cancer outcomes while evaluating screening, prevention, and treatment strategies.

At the start of the simulation, individuals are assigned a birth cohort-specific risk factor profile and categorized by sex, race/ethnicity, and nativity (U.S.-born vs. foreign-born). The model includes detailed subgroup analyses for Non-Hispanic White, Non-Hispanic Black, Hispanic, Non-Hispanic American Indian/Alaska Native, Non-Hispanic Asian/Pacific Islander, and Non-Hispanic individuals of two or more races. Each year, individuals transition through precancerous and cancerous states, with progression risks influenced by H. pylori infection, smoking, and residual risk factors. The model assumes that H. pylori infection increases the risk of transition at all points in the precancerous process, while smoking increases the risk of developing intestinal metaplasia, dysplasia, and cancer for adenocarcinomas, as well as pre-cancerous lesions and cancer in non-adenocarcinoma subtypes.

H. pylori infection is simulated in each year for individuals still at risk, based on the demographic and HP generator which accounts for secular trends by sex, race/ethnicity, and nativity. Smoking histories, including initiation and cessation ages, are also derived from the demographic generator that incorporates age-, sex-, race/ethnicity-, and nativity-specific trends. The Harvard-GC model also accounts for trends in residual risk factors that may further influence gastritis and precancerous lesion development.

Individuals who develop clinical cancer face stage-specific mortality rates (relative survival) based on SEER estimates (SEER 1975-2019 for cancer incidence; SEER 2004-2015 and 2018-2019 for stage distribution and survival). The model does not allow for regression ("point of no return"), meaning that once an individual progresses to a more advanced stage, they cannot revert to an earlier state. Competing mortality is modeled using U.S. birth cohort-specific life tables, adjusted for sex, race/ethnicity, and smoking status.

The Harvard-GC model is calibrated to age-, sex-, and race/ethnicity-specific prevalence data on precancerous lesions (from systematic reviews and expert input) and SEER-based cancer incidence and stage distribution. The model’s calibration process uses simulated annealing to identify many “good-fitting” parameter sets, and the top 100 parameter sets are used to account for parameter uncertainty for all model estimates.

Structure of Harvard-GC Model

Figure 1: Because of its flexible framework, Harvard-GC is readily adaptable for subgroup-specific analyses within the U.S. and can be extended to non-U.S. settings, making it a versatile tool for gastric cancer prevention and screening policy evaluation.