Mental health continues to be a national crisis that is only becoming more prevalent, especially since the pandemic. According to the National Alliance on Mental Illness (NAMI), 22.8% of U.S. adults experienced mental illness in 2021, which represents about 1 in 5 adults or 57.8 million people. Furthermore, Trilliant Health found that behavioral health volumes reached 18.1% above pre-pandemic levels by Q2 2022. And Mental Health America (MHA) found that between 2019 and 2020, 54.7% of adults with a mental illness did not receive treatment, totaling over 28 million individuals.
For Mental Health Awareness Month, PM360 spoke with Dr. Carl Marci, Chief Psychiatrist and Managing Director of Mental Health and Neuroscience at OM1, about how the company is using its Mental Health and Neuroscience Real-World Data Network to help improve personalized treatment for mental health patients and increase awareness about issues related to mental health disorders. Currently, OM1 can harness real-world data from clinical notes obtained from more than 9,000 clinicians working in 2,500 clinics across all 50 states, and the company recently added data from more than 148,000 patients with schizophrenia and bipolar this past March.
PM360: What are the benefits of using real-world data to gain insights into mental health disorders?
Dr. Carl Marci: I like the term “pragmatic” trial, because we’re looking at real people as they get prescribed in the real world, which can be a little easier for clinicians to relate to. This is different from clinical trials which are really designed to look at efficacy and safety in a very controlled environment, so controlled trials are very good at answering those two questions. But clinical trials are not as good at generalizing to the real world and what happens when these medications hit the market and clinicians are using them in slightly different ways. Through real-world data studies, we have longer horizons to look at prescribing patterns and to see what’s happening with patients over time.
For example, we co-authored a study with Alkermes examining weight gain in bipolar disorder I patients treated with second generation antipsychotics. This has been a challenge with second-generation antipsychotics really from the beginning. They’re wonderful medications, blockbuster sales, but real side effects. Another side effect that tends to be related to weight gain is the metabolic syndrome, which increases your risk for cardiac events and that increases your risk for sudden death and heart attacks.
The second-generation antipsychotics have now been out for a couple of decades, so we have a lot of data that we can learn from. The idea was to put some numbers to the amount of weight gain in different populations using second-generation antipsychotics as a class, and then evaluate the risk for cardiometabolic disorders in this category in a real-world population that all clinicians can relate to. We found, consistent with prior research, that the people who were most at risk for weight gain tended to be younger and not obese, so people who weren’t already in an overweight status. Then, a large proportion were at risk for cardiometabolic events. The other thing that stood out was a finding that 90% of patients stopped taking the medication within a year either due to side effects or weight gain, so that tells you something about how poorly tolerated these medications are in the real world.
How can you use that data to then help these patients stay on their medication?
I think the best way is to come up with alternatives that don’t have these side effects, because then the problem is solved. But with the existing class of medications, which are now in generic form and very affordable, there is a lot of education that goes into it. Even myself, looking at a study like this, if I get a relatively young person who is not overweight, I’m going to have a very different conversation with him or her starting this medication than I am someone who is in their 50s, is already overweight, and in a different stage of life.
But, as you can imagine, clinically, talking to anybody about weight gain is a sensitive topic. As soon as you say there is a potential for weight gain, everyone is like, “I’m not taking that.” So it helps to have real statistics that show it’s only one or two pounds on average, and if you commit to a good diet and exercise then most people can avoid gaining any weight. That’s a benefit of having recent data that can be provided as talking points for doctors and their patients rather than information from a clinical trial that’s over a decade old.
That study is an example of a retrospective observational study, which can sometimes deliver biased results from biased selection of data, measurement errors, confounding factors, and methodical errors. How do you approach studies like this to avoid those biases?
Yeah, these studies can have their limitations, but let’s first emphasize the benefits. Typically, our studies are done on tens of thousands, hundreds of thousands, sometimes over a million patients, so size allows us to look at things that are more subtle in terms of different effects. Another advantage is speed, because we’re mining big data that’s already digitized, we can do things in weeks or a few months that sometimes take years in a randomized controlled trial. It’s also a lot cheaper for a group of data scientists, biostatisticians, and epidemiologists to access data and essentially do virtual comparator trials than it is to recruit real patients.
The downside is there is always the potential for bias. The way we offset that is by using large networks and then having our epidemiologists check to see if the data matches the population. For example, in schizophrenia we know roughly 1% to 3% of the population in the U.S. has the disorder, so we want to make sure the data we are using also represents 1% to 3% of the population in our world. We’ll always check our cohorts from an epidemiologic perspective.
The other thing we do is what are called attrition tables, in which we start with a very large population of patients, say, with schizophrenia that we’re looking at weight gain, and then we narrow it down. For example, we constrain the dataset by maybe age as well as the populations we actually have measured weights for. Then, again, we make sure that the patients with outcome measures are representative of the population.
It really leverages the best practices from epidemiology and public health. The difference is OM1’s data is all digitized, so we’re not going through charts by hand, which is how a lot of historic retrospective observational studies were done. Now that everything is digitized, we can access it with the press of a button.
Since you are working off of digitized data, are you mostly reviewing structured data or can you also incorporate unstructured data in your studies?
We have a data science team that is using machine learning and artificial intelligence approaches to mine the unstructured information we have access to from electronic health records (EHRs) for deeper insights. For example, we used a natural language processing-based approach to extract depression symptom severity and suicidal ideation from clinical notes. With something like suicidal ideation, sometimes that can be found in a structured field within an EHR. Other times it’s just in the clinical narrative, so we do a combination of pulling structured and unstructured mentions of anything related to suicide. Then, we have computer programs that will help determine whether it’s an affirmation or a negation. Additionally, we have a very rigorous quality control (QC) process where we’ll then have humans double check a portion of the charts to confirm that the computer is right.
In that case, we were looking at the prevalence of suicidal ideation among a real-world population of depressed patients. I think everyone would agree we need to do a better job screening and documenting suicidal thoughts in a world where there’s a suicide every few minutes in this country. Our hope is to create, essentially, machine monitoring tools that will take the population of patients who have expressed positive endorsement of suicidal ideation and then learn what these patients are like, so we can predict who may go on to have that kind of risk. Then, once we have those models, deploy those at scale at the bedside so that in case someone forgets to ask, the system can begin to flag patients who are at risk.
How else can you use the data that you have access to? What other types of studies or programs are you working on?
In our mental health and neuroscience division, we’ve just launched a product around reasons for discontinuation. We are able to generate a report that shows stacked ranks of the different reasons for discontinuation for antidepressants and antipsychotic medications in depression, bipolar, and schizophrenia.
Another thing we haven’t talked about, but should, is comparator outcome studies. For example, in major depressive disorder, we’re doing a number of studies looking at different types of interventions and their impact on treatment-resistant depression and how people respond to different types of approaches. In the bipolar world, we’re looking at the impact of adding second-generation antipsychotics to antidepressants as a strategy for managing mood.
We’re also exploring studies in schizophrenia that look at the different antipsychotics being used to treat the “positive” symptoms of schizophrenia (such as hearing voices, seeing visual hallucinations, having active paranoia or delusions, etc.) vs. the newer antipsychotics coming out to treat the “negative” symptoms (such as social withdraw, anhedonia, flat affect, low motivation, etc.). Having a dataset where you have both documented over time the patients who are treated with an existing class of medications as well as with the newer medications as they come out, we’ll be able to look at treatment comparisons that go beyond randomized control trials. Do they have any side effects associated with them? What kinds of patients are being prescribed new medications? What kinds of clinicians are prescribing them? Do they tend to be rural or in academic settings?
Those are good examples of the types of studies we do.
Beyond that, are there any areas within the mental health space that you would like to see or believe deserve more study?
It’s exciting to see companies working on new medications with novel mechanisms in the brain, including zuranolone from Biogen and Sage, which has a novel neuromodulation and fast onset mechanism for treating depression; COMPASS Pathways’ COMP360 psilocybin, which is a synthetic, pharmaceutical-grade psychotropic magic mushroom; and Karuna Therapeutics, which recently announced a novel treatment that looks at muscarinic-type receptors as opposed to dopamine and serotonergic receptors for schizophrenia.
But big insurance companies are going to be hesitant to write checks for fancy new medications without proof they work, or proof that they have fewer side effects, and they lead to either higher productivity or lower utilization over time. So at some point, we’re going to have to figure out what we are going to pay for and how are we going to reimburse for new treatments relative to their outcomes. This is why I think real-world health economics studies are also important and will be crucial to getting these new treatments to patients.