EACMo (CUIMC) Columbia University Irving Medical Center

Model Overview

Developed by the CISNET modeling group at the Massachusetts General Hospital (MGH) and Columbia University Irving Medical Center (CUIMC), the Esophageal AdenoCarcinoma Model (EACMo) is a population-level Markov model that depicts the natural history of Barrett’s Esophagus (BE) and progression to esophageal adenocarcinoma (EAC). Simulated populations enter the model in the Normal Population state and may progress to BE, either directly or via the symptomatic gastroesophageal reflux disease (GERD) state. Further progression may then occur from BE to Low or High Grade Dysplasia, Undetected Cancer, Detected Cancer, and finally Death. Since EAC incidence has risen dramatically, it is imperative that our modeling incorporate mechanisms that reflect the potential causes of this increase. To accomplish this, we generalize the traditional age-period-cohort (APC) formalism (1-3) by applying age, period, and cohort trends to transition rates between health states within the natural history model. A simplified schematic of the natural history model is shown in Figure 1.

Figure 1. EACMo (CUIMC) model structure.

We inserted into the population-level natural history model (EACMo) a microsimulation module that allows clinically realistic modeling of screening, surveillance, and treatment at the individual patient level. Following is a brief description of this module and its use in a recent analysis of surveillance and treatment strategies for BE. A simulation was first run of the entire US population within the population-level natural history model. When patients reached the designated age (60 in the base case) for the beginning of surveillance, the subpopulation with BE was identified and removed from simulation. The characteristics of this subpopulation were used to automatically initialize the individual-level microsimulation, which continued to simulate the progression of the disease during endoscopic surveillance and radiofrequency ablation (RFA) treatment that attempts to remove the BE and dysplastic tissue. The individual and population-level results were aggregated to produce a single set of outputs. Before RFA treatment, patients within the microsimulation could progress in each cycle according to the transition probabilities of the natural history model. Endoscopy was performed at scheduled intervals based on diagnosed health state (shown as boxes in Figure 1). Patients received RFA treatment depending on the treatment strategy being analyzed and the patient’s histologic status as detected by endoscopic biopsy. Endoscopic surveillance schedules after ablation were determined by the outcome of treatment and the pre-ablation health state of the patient.

If RFA treatment failed, patients remained in their pre-ablation health state, undergoing endoscopic surveillance according to the same schedule as before ablation with no further attempts at RFA treatment. Progression to EAC was then simulated using the natural history transition probabilities. Patients who received successful treatment (complete eradication of dysplasia and intestinal metaplasia) or partially successful treatment (complete eradication of dysplasia) did not progress according to the natural history transition probabilities. Instead, there was a constant probability in each cycle that a patient would undergo a recurrence event (e.g. the detection of recurring BE). When a recurrence event occurred, a second random draw would be performed to determine the post-recurrence health state. Once in a post-recurrence state, the patient could again progress according to the natural history. If recurrent BE or dysplasia was detected by endoscopic surveillance before EAC developed, the patient could receive touch-up RFA treatment, with a maximum of three touch-ups after the end of the initial two-year treatment period.

References

1. Clayton D, Schifflers E. Models for temporal variation in cancer rates. II: Age-period-cohort models. Statistics in Medicine. 1987;6(4):469-81.
2. Holford TR. The estimation of age, period and cohort effects for vital rates. Biometrics. 1983;39(2):311-24.
3. Luebeck EG, Moolgavkar SH. Multistage carcinogenesis and the incidence of colorectal cancer. Proceedings of the National Academy of Sciences of the United States of America. 2002;99(23):15095-100.