Screening

Screening Modality Preferences

Each person in the model is randomly assigned one of two CRC screening tests – Fecal Immunochemical Test (FIT) or colonoscopy – as the routine screening test. While there are other modalities available to screen for CRC, we focus on FIT and colonoscopy because they are the two most commonly used modalities. In a prior claims analysis, we found that other types of modalities comprised less than 4% of all CRC tests.

The model assumes that an individual uses the same type of CRC screening modality each time he or she completes routine screening, as long as his or her insurance status remains the same and he or she does not have a polyp removed. However, when insurance status changes, the individual’s screening test modality may change, as it is affected by provider and system-level factors as well that are likely to change along with insurance. In addition, if the individual is initially assigned to FIT and has at least one polyp removed during a diagnostic colonoscopy, the modality will be changed to colonoscopy for the remainder of his or her lifetime for both screening and surveillance.

The probabilities used to randomly assign tests are obtained by one of two methods: user-specified and model-based. Using either method, the actual proportion of the population assigned to each test should match the probabilities in expectation only. That is, for any given run of the model, the observed proportions may differ from the expected proportions, but over a large number of runs the proportions should be approximately equal.

User-Specified Probabilities

When using this method, the user manually enters the proportion of the population to be assigned to each test, and the model randomly assigns a test to each person with probabilities equal to the specified proportions. If the specified proportions sum to less than one, the remaining proportion of the population will not be assigned a routine test and will therefore not be eligible for routine testing.

Model-Based Probabilities

When using this method, a statistical model is used to assign a preferred test to each person. Each person is assigned a probability and then a preferred test is randomly assigned to each person using the model-based probability. The statistical model uses individual characteristics such as gender, race, income, insurance status, enrollment in the state health insurance plan, and marital status. County-level characteristics are used as well. These county-specific characteristics are: distance to endoscopy facility, procedure volume, the count of generalist physicians, poverty level, educational attainment, uninsurance rate, and the proportion of non-whites. The statistical model was developed only for the two tests used in this model; thus, individuals may only screen by FIT or colonoscopy.

Routine Screening

Routine screening is defined as testing that is performed for the purpose of early detection even when a person is showing no signs of disease. The age at which a person begins routine screening, the time interval between tests, and the age at which the individual ends routine screening are all taken from the parameters associated with the person’s assigned routine screening test. For example, suppose the model assigns colonoscopy to a person for his or her routine screening test, and suppose the user sets the colonoscopy start age to 50, the end age to 80, and the frequency to 10 years. This person is then eligible to take a colonoscopy routine screening test at ages 50, 60, 70, and 80, but not eligible to take the test at any other age. At each eligible age, he or she may choose to take the test or choose not to take the test.

The routine screening test yields either a positive or negative result. The test result (positive or negative) depends on whether the person has lesions, as well as the lesion stage-specific sensitivity and specificity of the test.

If the person has lesions:

The model checks each lesion individually to see whether the test detects the lesion(s). The probability that the test detects a lesion is controlled by the test’s sensitivity parameters. The appropriate sensitivity parameter for each lesion is selected based on the size of the lesion. For example, if the lesion is a small-sized polyp, the model uses the test’s small polyp sensitivity parameter. After checking for detection of every lesion, the model produces the overall test result. A positive test result occurs when the test detects at least one of the lesions, and a negative test result occurs when the test fails to detect every lesion present.

If the person does not have lesions:

The test will correctly produce a negative result with probability equal to its specificity parameter. The test will produce a positive result (a false positive) with probability equal to (1 – specificity).

Positive screening leads to a recommended diagnostic test

After a positive routine screening test, the person becomes eligible for a diagnostic follow-up test for confirmation of the positive result. Diagnostic testing is performed by colonoscopy. For example, if an individual whose preferred routine test is FOBT has a positive (i.e. abnormal) result, the diagnostic test is a colonoscopy. In cases where a person’s preferred routine test is a colonoscopy, the model skips the routine screening step and immediately executes the logic outlined in the diagnostic test section.

Screening may result in immediate death or non-fatal perforation of the colon

Perforation and death are the only types of complications that may occur as a result of screening in this model. Both adverse effects are possible during either a positive or negative test result. Death from a screening test is counted in the model as a non-CRC death. A perforated colon is handled in the model through the additional cost for treating the perforation, but it has no other effect on the person’s life cycle. There is an average risk of perforation per colonoscopy, rather than a differential risk of perforation based on the size and number of polyps that an individual may have.