Association of Project ECHO Training With Buprenorphine Prescribing by Primary Care Clinicians in Minnesota for Treating Opioid Use Disorder

This matched-cohort study examines the association between the training received by primary care clinicians in Minnesota under Project ECHO (Extension for Community Healthcare Outcomes) and the number of clinicians obtaining a waiver to prescribe buprenorphine for treating opioid use disorder.


Initial screen
After collecting baseline characteristics of ECHO-trained clinicians and potential comparison clinicians, we screened out comparison clinicians whose baseline age or practice characteristics fell well outside the range observed for ECHO clinicians of the same clinician type. The screen was applied to ensure no comparison clinicians who were completely outside the values of the ECHO-trained clinicians. Age was bounded exactly (e.g., comparison physicians aged less than 29 or greater than 79 were excluded). Comparison clinicians were also excluded if their maximum or minimum value in any quarter of certain characteristics was outside of a buffered range of corresponding values among ECHO-trained clinicians of the same clinician type. The characteristics were: the count of unique patients per month, unique patients with prior substance use disorder diagnosis per month, unique patients with opioid use disorder (OUD) diagnosis per month, buprenorphine prescriptions per OUD patient per month, and opioid analgesic prescribing as morphine milligram equivalent (MME) per month per unique patient. The buffer was set as the floor or ceiling of the second significant digit of the minimum or maximum value for ECHOtrained clinicians. As an example, the maximum value of unique MHCP patients per month among ECHOtrained nurse practitioners (NPs) was 252; we screened out comparison NPs who treated more than 260 unique patients per month in any one quarter.

Propensity score matching
We fitted a propensity score model as a logistic regression including both time invariant covariates and time-varying patient panel and clinical practice variables. The time-varying covariates included measures of the outcomes of interest, included to reduce bias from pre-exposure deviations from parallel trends.

COVID-19 time period
The COVID-19 pandemic spanned the final quarter of our follow-up period (March -June 2020), which represented 10 percent of the full study period. During this time, there were substantial shifts in health care, such as most appointments moving from in-person to virtual, that could potentially change the patterns of ECHO attendance and the outcomes. To examine this possibility, we conducted a set of sensitivity checks.
ECHO sessions were offered with the usual frequency in March -June 2020. We examined the index dates for ECHO clinicians across quarters and found no difference between the number of clinicians whose first ECHO session date was between March and June 2020 and the number of clinicians whose first ECHO session date was during earlier quarters. This suggests that clinicians were starting ECHO at a similar rate before and during COVID.
We conducted a sensitivity analysis, presented in the legislative report (Appendix F), in which we removed data that was collected after March 1, 2020, and re-estimated the primary outcome models. Essentially, this tested what would have happened if the study period had ended on March 1, 2020 (before COVID-19) instead of June 11, 2020. Then, we visually compared plots of the estimated means over time and their 95% confidence intervals, for models with and without the COVID time period. None of the outcomes showed different patterns depending on whether the COVID time period was included or excluded.
Based on these results, we concluded that it was unlikely that including the one quarter of data during the beginning of the COVID-19 pandemic unduly influenced our findings.

Alternative specification of DATA-waiver status
Due to gaps in the availability of Controlled Substances Act Registry data, we tested an alternative specification for DATA-waiver. In this sensitivity analysis we defined DATA-waiver as either (1) confirmation in the Controlled Substances Act Registry or (2) having prescribed buprenorphine to any patient with OUD up to and including the quarter in question. The main analysis, which likely undercounts DATA-waivers, and the sensitivity analysis, which likely over-counts DATA-waivers, can be seen as bounding the true association between ECHO training and DATA-waiver attainment.