By Sarianne Gruber
How effective is diagnosis in healthcare? Constantine Gatsonis PhD, Professor and Chair of the Department of Biostatistics at Brown University, has been asking this question for over twenty years with diagnostic imaging studies on a variety cancers, primarily within the context of clinical trials. His interdisciplinary research has delivered answers on MRI and CT screenings for cervical cancer detection, and the landmark DMIST study in 2007 that settled the question of digital over plain film mammography for breast cancer screening. One of his largest research projects has been with the National Cancer Institute Lung Screening Trial comparing CAT scans and x-rays of smokers. Dr. Gatsonis shared with me the goals and challenges of his most recent study, the Imaging Dementia Evidence Amyloid Scanning –the IDEAS Study. He was the Plenary Speaker at the International Conference on Health Policy Statistics held earlier this month in Providence, Rhode Island. Here is a gently edited version of our interview at the ICHPS conference.
What is the IDEAS Study? And how is this study unique from other disease screening clinical trials?
Right now I am involved in the IDEAS study- Imaging Dementia Evidence Amyloid Scanning study, evaluating the impact of amyloid scan tests for Alzheimer’s disease. The PET scans look for amyloid plaques in the brain, and these plaques have been shown to be predictive of Alzheimer’s disease. We are trying to understand the impact of doing this type study today in an environment that is kind of complicated because there is no real treatment for Alzheimer’s, whereas with cancer when you find something, for example, in someone’s breast you take it out. There is a lot of demand for this test so the Centers of Medicare and Medicaid Service (CMS) decided to fund a “coverage with evidence” initiative. CMS will cover the cost of imaging for studies that would generate new information in order for Medicare to approve this coverage. The IDEAS study will have about 20,000 people in a registry setting. These patients must be currently seeing a neurologist because they have certain symptoms and fall into a category called “appropriate use” criteria for amyloid imaging. We will see whether the test on these patients has an impact on their care because these persons are already receiving some type of care for dementia. They could be getting various work ups, and an amyloid plague testing could shed light on their current situation.
How will you evaluate utilization for patients with and without amyloid plaque testing scans?
With the IDEAS study, we will be looking to see if patients’ medical care changes after getting the test. Doctors could change medications or make recommendations on counseling with this knowledge. We want to see their healthcare utilization one year after the test, and if healthcare utilization is better than if they did not have the test. We will have to select a group of controls, people who have similar symptoms but did not have the test and then compare utilization. For example, were they hospitalized in the last 12 months or did they go to the emergency room in the next 12 months. These are things that we will be able to address. Sometimes undiagnosed people end up in hospital because they forget to take their medications – like diabetes. We will be able to see if these tests lower admissions to hospitals because the patient’s doctors knew what is going on and family members could be more prepared and act differently. Also, these tests will eventually become important when drugs for Alzheimer come about. If there are drugs that slow down the disease, amyloid testing will be linked to downstream therapies or prevention therapies.
In our study, patients will be referred by their doctors. Doctors will provide their current care and what they are planning. And then 90 days after the test, the doctors need to tell us whether the test results changed their mind on the course of care. There have been other studies for PET and cancer that my group has been involved through the National Oncology PET Registry. Those studies showed that in aggregate 36% of all cancer patients that get a PET, there is a change in their course of treatment. With 20,000 patients, CMS could spend up to $80M just paying for the tests, at least $4,000 for CMS prices. If this becomes a prevalent type of testing, naturally CMS will be very careful as to what type of evidence they need to approve the cost of this test. Especially, in the context of a disease where there is no therapies.
How will the control sample be created to compare patients to those without amyloid plaque testing scans?
We plan to enroll about 20,000 people in the group that will get the scan. Then we will identify controls from Medicare records. Mapping out the approach to indentify controls will be very difficult. We have to look through lots and lots of Medicare records to identify people with longitudinal profiles that look similar to the profiles of people who actually enlisted in the registry. This study is a hybrid study with an observational part that is the registry. The claims data is the secondary data. Selecting controls in that context is complicated and a challenge. Statistical methods will include propensity score matching, traditional matching by categories and by covariates. We do need to match by geographic location like hospital areas, since this is a national study. We were hoping to match by provider, the neurologist, but it won’t be realistic. We will have to look through millions of records, longitudinal records of CMS that combines data from all of their files. Because we also want to see the profile not just in terms of the symptoms, but hospital utilization in terms of the drugs they received and existing co-morbidities. It is going to be a real challenge. Just as a test bed, we are getting a 20% sample of all CMS records for three consecutive years. We want see what these profiles look like, and eventually we are going to have to tap into the 100% Medicare data. Medicare records take a while to become complete due to an information lag. You will not get immediate answers when you have to rely on secondary databases. We can handle the big data, but it is assembling it that is a challenge. Medicare has a new VRDC system – the CMS Virtual Research Data Center, where “you buy a seat” to access to their databases and computing. One can only do analyses behind all of their firewalls and built-in protections. It is relatively new, just in the last few years. In this way, you can access more recent Medicare data.
Will this type of study design become a gold standard for disease management and outcomes?
Our study will be in the regular fee for service Medicare. It is a kind of a big data that is easier to work. We have a much more controlled situation, which makes this not a “completely exploring project”. There are set questions that we are seeking answers for by looking at people trajectories over a 2-3 year period of claims data. We plan to work closely with neurologists and imagers. There is a full scientific board, a Medical PI and advisory members who are also subject matter experts. We will build an image archive where we will collect the scans for all the 20,000 people. This will be an enormous resource. The image archive will be linked to the clinical information that we will have in the study. These are the actual scans. Researchers will be able to run data for interpreting and predicting. They will also know what is happening from the claims data. You will know what is happening to patients for a very long time. Starting from their initial meeting with their physician to their point of entry, and when things happen to them down the line you can trace them. This will be a major type of resource. This will be a study that you will see more and more, because it has a prospective data collection, and it is also combined with other data sources from the same people or on control. It is an important study for Alzheimer’s. And it is methodologically important.