Simulation models can be used to project the long-term outcomes associated with implementing a public health intervention or program at the population level. These models can help to estimate the effectiveness of the specific program in improving health outcomes as compared to one or more control scenarios.
The Colorectal Cancer Control Program (CRCCP) aims to improve colorectal cancer (CRC) screening among age-eligible adults, with the goal of reducing CRC incidence and mortality, by funding state and tribal organizations to provide and promote screening.
Method and model description
The main goal of simulation modeling, in this case, was to estimate the long-term CRC outcomes associated with receipt of a first CRC screening through the CRCCP program among low-income and underserved patients in the U.S.
Using digital twin simulation, the research team created simulated individuals who match those in our CRCCP patient population to simulate and evaluate their future CRC outcomes which cannot yet be observed. Also, they recalibrated a simulation model previously calibrated based on a real-world mix of insurance and demographic factors.
The model included 2 groups of people:
- Those who underwent a screening colonoscopy (“Screening Colonoscopy Cohort”).
- Those who completed a stool test and, after testing positive, underwent the recommended follow-up diagnostic colonoscopy.
For each cohort, researchers compared the CRCCP patients’ outcomes with and without this single CRC test.
Individuals were simulated from birth until death from either CRC or other non-CRC natural causes. During their lifespan, individuals can develop polyps, which may progress from non-cancerous to cancerous as the individuals' transition within and across health states.
Due to the large differences between the actual and simulated results, it was necessary to calibrate the model to the colonoscopy findings for our CRCCP screening colonoscopy cohort. The best combination of multipliers to match the actual CRCCP data were identified for three parameters:
- Individual risk factor
- Incidence rate
- Transition rate from polyp to cancer.
Researchers used Latin hypercube sampling (LHS) to achieve our calibration objective. LHS is a sampling method, in which a grid of all possible variables is created and samples are selected from non-overlapping intervals across each dimension. To account for randomness in the simulation model, each LHS sample was run 30 times and average values were obtained.
Constructing of simulation equivalents included several steps:
- Combinations of demographic characteristics were identified for each cohort with a focus on the race, gender, and age (Step 1 and Step 2).
- After creating the reduced cohort of T combinations of individuals (i.e., individual types), their lifetimes with and without receipt of the CRCCP screening colonoscopy were simulated numerous times (Step 3).
- Then, simulated equivalents were chosen – simulated individuals that matched the colonoscopy result as their real counterparts. (Step 4).
- To estimate this cohort’s long-term outcomes, individuals in the CRCCP screening colonoscopy cohort were matched to their simulated equivalents and then sampled the required number of replications (Step 5).
- The number of replications for the CRCCP screening colonoscopy cohort was established to produce results that were within 1% of the total number of cancers in the population. To estimate the impact of the CRCCP screening colonoscopy intervention compared to no intervention, the corresponding simulated equivalents with the same replication number in the set of no CRCCP intervention replications were identified (Step 6).
Process of creating simulated equivalents and matching CRCCP patients to their equivalent for our screening colonoscopy cohort
Results
By assigning simulated equivalents to members of both of our CRCCP modality-based cohorts, we were able to replicate the CRC-related health status of each individual. We were also able to assess the impact of the screening colonoscopy intervention and the FIT/FOBT testing followed by a diagnostic colonoscopy compared to no intervention for each respective cohort. Both intervention types proved to be effective in averting cancers and reducing both the number of deaths and life years lost by more than half over the individuals’ lifetimes.
In general, the simulated equivalent method proved to be effective and can be extended to include past screening history and/or other aspects relevant to simulating a cohort. Planned future work includes estimating the quality-adjusted life years (QALYs) gained and the cost-effectiveness of the CRCCP intervention for both cohorts compared to no intervention.