A New Method for Estimating Incidence of First Psychotic Diagnosis in a Medicaid Population

The NYS Medicaid billing and encounter data system (Medicaid) was used to identify patients with new onset of psychosis during the 5-year period between January 1, 2013, and December 31, 2017. Psychosis was defined as having at least one claim or encounter for an outpatient or inpatient service with the following diagnoses: schizophrenia spectrum disorders (ICD-9 and ICD-10 codes 295.0–295.9 and F20.0–F20.9), affective psychosis (296.04, 296.14, 296.24, 296.34, 296.44, 296.54, 296.64, F30.2, F31.2, F31.5, F31.64, F32.3, and F33.3), and other psychotic disorders (297.1, 297.3, 298.8, 298.9, F21–25, F28, and F29). Diagnoses of substance-induced psychosis were not included as qualifying first-episode diagnoses.

The study population referred to as “putative cases” was limited to individuals between the ages of 15 and 35 at the time of diagnosis with no prior Medicaid claims or encounters for service for psychosis. The prior-service lookback period was inclusive of 13 years prior to the study period (2000–2012) where data were available. Finally, the study population was limited to those continuously enrolled in Medicaid for at least 12 months prior to the first identified diagnosis. Individuals with dual eligibility for Medicare and Medicaid were excluded.

Demographic and service-related characteristics of putative cases were extracted from Medicaid. Demographic characteristics (age, gender, and race-ethnicity) were categorized as follows: age at first presentation (15–25 versus 26–35), gender (male or female), and race-ethnicity (non-Hispanic white; racial-ethnic minority, including non-Hispanic black, Hispanic, and other; or unknown). Service-related characteristics were categorized as follows: service setting at first presentation (behavioral health–related emergency or inpatient setting, nonbehavioral health–related emergency or inpatient setting, outpatient behavioral health specialty setting, or outpatient general setting), indication of any antipsychotic medication fill prior to first diagnosis, and Medicaid program type (managed care or fee for service [FFS]).

Annual crude incidence rates (IRs) were calculated for demographic and service-related characteristics as the number of putative cases per year divided by the number of Medicaid recipients continuously enrolled during calendar year 2015, yielding an estimate of annual incidence (putative cases per 100,000 persons per year). Stratified analyses were conducted for age at first diagnosis (15–25 versus 26–35) and for demographic and service-related characteristics. Poisson distribution models were used to estimate IRs of psychosis for the two age groups by demographic and service-related characteristic and compare the results by using rate ratios (RRs) with 95% confidence intervals (CIs) (23, 24).

The OnTrackNY data system was used to confirm putative cases as FEP and to compare date of onset of psychosis with the first service date in Medicaid. The OnTrackNY data system collects participant-level, clinician-reported standardized assessments over the course of care. Medicaid ID and date of onset of psychosis were extracted for Medicaid-enrolled OnTrackNY recipients. The OnTrackNY sample was matched to putative cases for confirmation of identity and to compare date of first presentation with psychosis in Medicaid with the first onset of psychotic symptoms recorded by OnTrackNY.

Medicaid managed care organizations (MCOs) were invited to assist in validation of the algorithm. Three managed care entities (MCEs) agreed to participate, including nine MCOs represented by one behavioral health organization (BHO) and two MCOs. These MCEs represent 65% of the state’s MCE plans (11 of 17 plans) and 53% of the lives covered by Medicaid managed care statewide in 2017. These organizations agreed to participate in a chart review of putative cases identified by the algorithm for calculation of adjusted IRs. MCEs use NYS guidelines to implement FEP-identification protocols that include reviewing individual clinical records (15, 18, 19).

A random sample of 50 cases from each entity was selected for review. Entities completed a tool to confirm putative case identity (Medicaid ID, name, date of birth, and Social Security number) and to indicate confirmation or nonconfirmation of FEP diagnosis and date. Cases were categorized as confirmed if plan records identified the member as having FEP via an algorithm or via clinical assessment. Reasons for nonconfirmation were documented as diagnostic rule-out (records indicate psychotic diagnosis was ruled out rather than confirmed); not continuously enrolled in plan; or other.

Adjusted IRs were calculated by age group and service setting by using the number of putative cases identified and confirmation rates. The 95% CIs for confirmation rates were estimated without continuity correction (25). Initial estimated IRs (based on putative cases) were multiplied by confirmation rates (confirmed cases divided by putative cases) to yield final estimates of adjusted IRs in each stratum.

This study protocol was approved by the Nathan S. Kline Institute Institutional Review Board with a full waiver of informed consent. All statistical analyses were performed with SAS software, version 9.4 (26).

This study adapted a previously published population-based algorithm to identify first presentation of psychosis in Medicaid (21). In this Medicaid population–based study, we estimated the actual IR of first diagnosis of psychosis to be 272 per 100,000 per year. This rate is higher than the rate estimated by the replicated study, in which a majority of the population was privately insured (21). Both this study and the replicated study estimated higher IRs relative to studies using clinical case identification methods (27–29). Higher IRs in a Medicaid population are not surprising, given ample research demonstrating a relationship between lower socioeconomic status and psychosis (30–34).

In terms of its ability to identify a younger population prior to accumulation of significant disability related to psychosis, the importance of this algorithm should not be underestimated. Compared with individuals with a serious mental illness in the Medicaid system (35), the individuals identified as putative cases by this algorithm were less disabled. In addition, this analysis should alert the Medicaid program to examine the need for FEP services in the populations that were carved out of Medicaid managed care but that remain covered by FFS Medicaid (36). These populations may include individuals who are enrolled in or who have a history of being in the child welfare system, who have intellectual disabilities, or who have other chronic health needs. In this study, the group covered by FFS Medicaid had a higher crude IR of FEP compared with the group covered by Medicaid managed care. This high crude IR for the population covered by FFS Medicaid may point to disparities in need for FEP services for populations excluded or excepted from Medicaid managed care (37).

This study also supports continued work to develop CSC models that are more broadly inclusive of the population in need, including those experiencing affective psychosis or those with a comorbid substance use disorder. It also points to a potential demand based on estimated adjusted IRs that exceeds capacity (38, 39). That said, we utilized broad inclusion criteria to support planning related to early intervention programs. Case definitions in both this study and the previous study included patients with new claims for any of a broad set of psychosis diagnoses (21). It is possible that individuals with a co-occurring substance use disorder or co-occurring mood disorder were included as cases in the sample and that these individuals may later be clinically determined to have substance or mood disorders rather than schizophrenia spectrum disorders. Clinically it is known that initial diagnostic classification among schizophrenia disorders changes over time; however, more work needs to be done to understand the stability of psychosis diagnosis in insurance claims (40).

In addition, this study indicates that programs focusing on outpatient mental health and acute settings would identify many of the incident cases. Programs like NYC START may have important potential to identify incident cases of FEP in acute settings, a critical point in the presentation of early psychosis (41). However, FEP case identification methods likely need to be enhanced in outpatient mental health settings as well.

Translating this algorithm from research to practice is an important next step. This will require additional epidemiologic research on the algorithm and operational work to pilot outreach, case confirmation, and treatment planning for identified individuals. Preliminary conclusions using OnTrackNY case confirmation indicate that the algorithm has high sensitivity, provided continuous eligibility (CE) requirements are relaxed. However, given the findings of high crude and adjusted IRs, it is likely that improving the specificity and reducing false positives identified by the algorithm may also be required. As mentioned previously, having Medicaid MCO-reported data on cases of FEP will allow us to conduct sensitivity analyses on the algorithm. The critical areas to examine include the window of CE prior to first diagnosis and the definition of a qualifying diagnostic event or set of events. Currently the qualifying event is one diagnosis of psychosis following a clean period with CE of 12 months. Criteria that can be modified are the number and type of events, time windows between events, and windows of CE prior to the defined events. In this study, individuals in the majority of identified cases had an antipsychotic prescription fill prior to the first psychosis diagnosis. Research to examine the weighting of events by type could lead to a more robust case definition and allow inclusion of less specific information, such as psychotropic medication fill. Next steps in terms of piloting an operational approach to outreach to identified individuals will be planned with the state Medicaid and state mental health authorities, Medicaid MCOs, and service providers. Care will be needed to address privacy issues for these individuals.

We should acknowledge some important study limitations. First, the algorithm identified putative cases by only one claim for psychosis in primary or secondary positions in Medicaid and 1 year of CE in Medicaid prior to this claim. Imposing a CE criterion on a sample derived from insurance is a method used commonly to provide a sample with an equal opportunity to be included in a measure (42). In this study, a 1-year window of CE was used for purposes of comparing results with the previous study (21). It is likely that this study missed individuals who qualified for Medicaid for a shorter duration prior to a first-presentation diagnosis. Second, it is possible that the single identifying claim was a rule-out diagnosis rather than a clinical assessment of psychosis. These limitations could both over- and underestimate the true incidence and as such will be important factors to examine in future work to refine the algorithm.

Third, individuals who did not present for treatment are not captured in the algorithm estimate. That would be expected to underestimate true incidence in the population. However, such individuals may be identified in the algorithm as symptoms escalate and care is required. Fifth, the adjusted IR calculations were based on record review by plans that agreed to participate. The record review was a time burden on MCEs, so the state could not randomly assign the task. However, the participating plans represent a majority of lives covered by the state’s Medicaid managed care plans, and these plans are expected to follow guidance provided by the state Medicaid authority for the definition of FEP (18–20). As such, the adjusted IR calculations in this study may miss individuals who are covered by other MCOs or who are in FFS Medicaid. In 2019, MCOs are required to submit all first-episode cases to the state for review and oversight. Future work will include these data for calculating adjusted incidence rates.

This study has important implications for the public mental health system. This algorithm presents a mechanism to identify a high-risk population before individuals accumulate significant disability. A comprehensive system for outreach, assessment, and treatment for FEP can be appropriately resourced by using these estimates. Additional research is needed to fine-tune this administrative data algorithm for use as a basis for communication and active outreach by early intervention programs.