Application of an Opioid Use Disorder Cascade of Care in a Large Public Health System

Stage 1: Diagnosed with Opioid Use Disorder (OUD)

Patients were included in Stage 1 if they had at least one inpatient, outpatient, or emergency department encounter at any H+H facility with an ICD-10 diagnosis of OUD documented as the primary or non-primary billing diagnosis in EPIC or Quadramed. ICD-10 codes included are listed in the Appendix. Patients were also included in Stage 1 if they had an OUD diagnosis documented as an active problem on their EPIC Problem List. The EPIC Problem List in each patient’s chart allows providers to document current/past diagnoses. To capture patients with new (index) OUD diagnoses and exclude patients with prior diagnoses who might already be active in treatment, we established a lookback period six months prior to the data observation period. If patients had a documented OUD diagnosis in the past 6 months, they were excluded.

Stage 2: Engaged with addiction services

Patients were included in Stage 2 if they contacted or were reached by addiction specialty services at H+H following their diagnosis. Treatment engagement could consist of one or more completed encounters occurring across four treatment settings throughout the system. We counted engagement whether it occurred with the same encounter as OUD diagnosis, or after the diagnostic encounter. Given heterogeneity across settings, each service setting or intervention had a unique set of inclusion criteria, delineated in Table 2.

Stage 3: Initiated MOUD treatment in an outpatient setting

Patients were included in Stage 3 if they initiated MOUD in tandem with or following their engagement in addiction services at an outpatient setting providing MOUD maintenance. Receipt of MOUD in inpatient or ED settings (Stage 2) did not qualify because this stage aims to capture initiation of maintenance medication therapy, not receipt of medications given to treat withdrawal. Medications included in this stage were: buprenorphine (all formulations except those indicated solely for pain conditions), methadone, and intramuscular extended release (“xr-“) naltrexone. A list of included formulations is provided in the Appendix. Medication data for buprenorphine and xr-naltrexone was accessible through the EHR as documented in medication orders. Methadone administrations are recorded in a software system that is not integrated with the patient’s EHR; therefore, we counted a completed Outpatient Treatment Program (“OTP”) visit as a proxy for methadone administration, thereby excluding use of methadone for pain.

This stage was adapted from the National Quality Forum (NQF) measure for the use of pharmacotherapy for OUD (NQF 3400) (19). Although counseling and behavioral therapies can support patients as part of their treatment, MOUD is considered the gold standard for treating OUD (20). However, fewer than one third of patients in substance use treatment programs for OUD receive medication (21).

Stage 4: MOUD follow up visit within 34 days of initiation

Patients were included in Stage 4 if they had a timely MOUD follow up after their MOUD initiation in an outpatient setting, which was adapted from the Healthcare Effectiveness Data and Information Set (HEDIS) (22, 23) addiction treatment engagement quality measure (24). Patients had to have at least one subsequent medication order for buprenorphine or xr-naltrexone or documented visit in an OTP within 34 days of initiation. Patients with no prescriptions or OTP visits within 30 days from MOUD initiation were excluded and considered discontinued in care.

Due to data limitations, we established rules to determine length of medication coverage to assess MOUD engagement and retention. EPIC data typically contained a start and end date for a prescription, but Quadramed data only contained prescription start dates. Therefore, we created a coverage rule for buprenorphine prescriptions without end dates, where we imputed an end date of 30 days following the start date, consistent with common clinical practice of prescribing no more than a 30 day supply (25). Xr-naltrexone was also considered to cover a patient for 30 days given it is a monthly injection. For OTP visits for which we did not have access to methadone dispensing data or data on take home doses (OTPs manage care through a separate EHR), we established a rule that one visit in an OTP would “cover” a patient for 3 days, in line with federal guidelines around methadone dosing for new patients (25). Patients who had gaps in care of longer than 30 days between prescriptions or OTP visits were considered discontinued.

Stage 5: Retained on MOUD for 6+ months

Patients were included in Stage 5 if they were retained on MOUD for at least six months, consistent with the NQF-endorsed quality measure for continuity of pharmacotherapy of MOUD for a minimum of 180 days (NQF #3175) (26). Patients that had medication orders (buprenorphine or xr-naltrexone), or OTP visits 6 months after their initial MOUD prescription or visit were included in this stage, as long as they did not have a gap between prescriptions or visits longer than 30 days, following the coverage assumption rules defined above. NQF measure #3175 defines the gap at the 7 day mark (26), but our team chose to extend the gap to 30 days to account for patients who were unable to return to the hospital for a follow up prescription appointment within that small of a time window, which may have been an issue during the early days of the COVID-19 pandemic.

When fully deploying the Cascade, we are considering testing dynamic definitions of follow-up engagement (Stage 4) and retention (Stage 5) via sensitivity analyses using different cutoff periods to determine treatment discontinuation. Although we operationalized our allowable treatment gap at 30 days, the existing literature on medication gaps ranges between 5 to 90 days (27). In a study measuring the duration of buprenorphine treatment episodes, Dong et al. (27) demonstrated that changing the allowable gap between MOUD episodes (their team analyzed gaps at 7, 14, 30, and 60 days) impacted the researchers’ understanding of patients’ continuity of care. This is particularly relevant as federal regulations for OTP visits and methadone take-home doses have changed in response to COVID-19. Prior to the pandemic, patients starting treatment up until their first 90 days in care were required to visit an OTP daily and patients in the second 90 days of treatment could visit every other day and were allowed up to two weekly take-home doses (25). In March 2020, the Substance Abuse and Mental Health Services Administration (SAMHSA) relaxed regulations, issuing a waiver permitting up to 28 days of take homes for stable patients and up to 14 days of take homes for less stable patients (28). We anticipate that these changes will impact the rules we establish regarding medication coverage.

Summary statistics are presented in Table 3 showing the total number of individuals identified with OUD in the H+H system (n= 33,616) and their progression through sequential stages of the Cascade. Consistent with prior studies (14), we found that attrition was most steep in earlier stages of the Cascade.

Data Access and Linkages

The experience of H+H shows the feasibility of building an OUD Cascade of Care in a large health system even with shifts in data management software. We show that committing to stage definitions with fidelity is only possible after understanding a system’s data constraints and responding to challenges accordingly.

Although our team had access to a wide variety of data sources, the complexity of H+H’s data systems limited our ability to access certain records and understand linkages over time. Because we relied on EPIC and Quadramed datasets, we were unable to rely on a numerical identifier attached to the patient in both EHR systems. To work around this, we developed our own unique patient identifier that concatenated the first 5 digits of a patient’s last name, the first 2 digits of a patient’s first name, an 8 digit birthdate, and an F or M initial for the patient’s sex. This process was undertaken using SQL queries and required several rounds of manual review and validation checks. Patient name misspellings, changes in sex, or other data inconsistences may have contributed to one patient having more than one identifier, despite our best efforts to develop a unique ID for each person.

While compiling and synthesizing data sources across a large hospital system allows us to capture patients who may have received treatment in multiple H+H facilities, our analysis is limited by our inability to access data outside of the system. There may be patients with OUD included in our sample who sought treatment or initiated/discontinued MOUD at another facility outside of the H+H system but could not be identified. Additionally, because our research team did not have access to insurance claims data, part of the criteria for medication initiation and subsequent stages relied on prescription order records, which could result in the assumption that a patient adhered to their medication when they did not in fact fill their prescription or take the medication. Thus, analyses that can track multi-system service utilization using claims data are also needed to understand true patient trajectories.

Data disaggregation using patient demographics is also a critical step in identifying population level disparities in access to care. Given the significant racial and ethnic disparities in receipt of OUD services (29), the Cascade can be an important tool in understanding where a healthcare system is best engaging certain patient groups, and where additional organizational level strategies are needed to deliver culturally competent care. The encounter-level datasets compiled for the Cascade contain fields that indicate patient race, ethnicity, and sex, therefore it is possible to investigate potential discrepancies in access and retention in care. However, due to issues including limited race and ethnicity category options within the EHR and patients’ unwillingness to report, some data may be missing or a race/ethnicity category may be unspecified. The ability to link datasets by patient identifier allows for the reduction of some missing data. If demographic data is missing from one patient record, but a patient has returned for an encounter within the system and demographic data is entered, the records can be linked to substitute for the missing data. Steps will be taken in future analyses to reduce missing and unspecified variables.

Defining Measures

The accurate measurement of the overall patient population may be impacted by the use of diagnosis codes as the main indicator to identify patients with OUD. Stigma may prevent patients who use opioids from disclosing use to healthcare providers or seeking treatment (30, 31). This may contribute to an undercount of patients analyzed using the model. Other models may consider adding an “at risk” stage to analyze patients who may not meet the clinical criteria for OUD but take opioids for pain management or score within a certain range on validated substance use screeners including the DAST-10 or CAGE-AID (10, 32, 33, 34).

Another ongoing challenge is establishing rules for defining gaps in care and patient drop offs between stages. We established a rule for MOUD engagement and retention that if a patient had a coverage gap of longer than 30 days, they would not achieve the subsequent retention stage. Creating rules around patient drop off due to a coverage gap is complex given the nature of OUD as patients may start and stop MOUD multiple times (35, 36). While establishing an exclusion rule may mean that fewer patients can achieve subsequent stages, we chose to initially design the Cascade based on established standards (e.g. HEDIS or NQF-endorsed measures) as a quality improvement tool for the H+H system.