A conversation with Eric Seiber, Ph.D., about why defining first-episode psychosis matters

Estimates of how many people in the United States experience first-episode psychosis (FEP) each year vary widely, which makes it harder for payers and health systems to plan and fund early intervention services such as coordinated specialty care, according to Eric Seiber, Ph.D., an economist and professor of health services management and policy at The Ohio State University.

Seiber and his colleagues recently published their research brief, 'Estimating first episode psychosis (FEP) incidence rates using U.S. claims data: a need for standardization,' in Schizophrenia.

They found that much of the variation comes from how researchers define FEP in claims data. When the authors applied the criteria from previously published studies to the same commercial claims dataset, estimated incidence ranged from 20 to 163 cases per 100,000 individuals.

Seiber recently sat down with Managed Healthcare Executive to explain why this matters for research, early treatment and payment models.

This interview has been edited for length and clarity.

What is first-episode psychosis, and why is it important to recognize it early?

First-episode psychosis is broader than schizophrenia. In our analysis, about 15% of cases fell into the schizophrenia category, nearly 50% were mood-disorder-related psychosis, and about 35% were other types of psychosis. Those categories were assigned hierarchically, and cannabis-related psychosis was excluded.

The timing matters because people can go a long time without treatment. The average duration of untreated psychosis in the U.S. is about 76 weeks. Clinical care is most effective when people are connected to treatment early, ideally within the first month or first few months.

Why do estimates of first-episode psychosis vary so widely?

Some of the difference may reflect who is included. Estimates from Medicaid claims have been around 300 to 400 cases per 100,000 people, while estimates from commercial claims have been closer to 60 per 100,000.

But definitions also matter. Claims data are created for payment, not research, and researchers may use different diagnosis codes, age ranges or criteria for deciding whether a case is truly a first episode. When we applied different study definitions to the same population, estimates ranged from about 20 to 163 cases per 100,000.

How did you compare the different definitions?

We reviewed nine studies and identified several common approaches. We then applied those approaches to the same data so that differences in the population would not explain the results.

The choices made a substantial difference. The diagnosis codes used could change the estimate by about threefold. The age range also shifted the estimate, by roughly 50%. And the definition of “first” mattered: a single claim may not reliably indicate a new case.

Requiring either two outpatient claims or one inpatient claim with a psychosis diagnosis helped reduce the chance of counting a one-off claim as a confirmed case. That kind of requirement cut the estimated incidence roughly in half.

Why is it difficult to determine whether a case is truly a first episode using claims data?

People’s insurance coverage changes. If a study requires a long period of continuous coverage before someone’s first recorded diagnosis, it may miss people who had a coverage gap. That can make the estimate look lower, not necessarily because there are fewer new cases, but because some people are no longer observable in the data.

That’s why researchers, clinicians, and payers need to agree on definitions before using the numbers to make decisions. Otherwise, people may think they are discussing the same population when they are not.

Could these issues with definitions affect research beyond psychosis?

I would be surprised if they were limited to psychosis. Claims data are used for billing, and the way a diagnosis is recorded can matter for research, prior authorization and payment. Even when people agree on the broad condition, details in the definitions can affect who is counted and what a payer is expected to cover.

What do you hope readers take away from this research?

Payment can help drive clinical innovation, but we need reliable empirical information to design payment models. Coordinated specialty care teams are an example: fee-for-service payments cover, at best, about half the cost of operating one of these teams.

What are you studying next?

We’re looking more closely at coordinated specialty care: who gets referred, what happens after people enter care and how outcomes and costs change. There are not enough teams to serve everyone who might benefit. Our estimate is that existing capacity reaches less than 10% of cases nationally.

Preliminary analyses suggest that people who are referred may have more severe needs and that emergency, inpatient, and outpatient utilization may drop substantially after they engage with a team. Those findings still need further refinement, but they raise the possibility that coordinated care could improve outcomes while also reducing costs.