Page 119

TEAMKY Demonstration

Approval Period: January 1, 2025 through December 31, 2029

Page 118 of 155  total expenditures;  SUD and non-SUD expenditures (with SUD expenditures disaggregated into IMD and non- IMD expenditures);  expenditures disaggregated by source of treatment—namely, inpatient expenditures, emergency department (ED) expenditures, non-ED outpatient expenditures, pharmacy expenditures, and long-term care expenditures.

Because of the demonstration's focus on SUD care, the sample population for which expenditures will be calculated will consist of Medicaid beneficiaries with an SUD diagnosis or who have used SUD treatment services during the period of interest. In particular, following the protocol specified in Attachment A of the SUD Evaluation Guidance Technical Assistance document, beneficiaries will be included in monthly expenditure calculations if they have received an SUD diagnosis or have used SUD treatment services that month or in the previous 11 months. If there is no SUD diagnosis or SUD treatment service utilization after these 12 months, beneficiaries will be excluded from subsequent expenditure calculations. Monthly expenditures will thus be based on pooled cross-sectional samples rather than a specific cohort of beneficiaries. To identify beneficiaries with an SUD diagnosis or who have used SUD treatment services, we will use codes in the value sets specified in Appendix A of the SUD Evaluation Guidance Technical Assistance document.

As with quantitative analyses of utilization, we will report summary statistics of expenditures for the two pre-demonstration years, as well as each successive year of the evaluation. We will also estimate the following model:

Yi,m = β0 + β1 timem + β2 I[post]m + β3 timem × I[post] m + β' controlsi,m + i,m

where Y denotes expenditures; time is a linear time trend; I[post] is a binary indicator of demonstration implementation (1 if yes, 0 otherwise); controls are a vector of covariates (e.g., beneficiary characteristics); ε is the disturbance term; m indexes the month; and i indexes the individual beneficiary. The outcome measure of interest for the cost analyses is average monthly expenditure per (SUD) beneficiary.

For the expenditure analyses, we are interested in β2, which reflects the shift in spending in the post-demonstration period, and β3, which reflects the expenditure effect of the demonstration. We hypothesize that expenditures for outpatient visits will initially increase, while spending for more costly services such as inpatient care and ED visits will decrease, generating net cost-savings over time.

We are aware that the validity of the cost analysis is dependent on the quality and completeness of the financial measures in the MCO encounter data. The Penn team's preliminary analysis of the data suggests a relatively high-quality dataset with plausible beneficiary and case counts, few missing values, and plausible paid amount values and distributions. For the evaluation, we will conduct a more thorough graphical and statistical analysis of the expenditure measures, checking for missing and implausible extreme values, anomalous distributions, and signs of selection bias (based on beneficiary characteristics). Prior to formal statistical analyses, we will take care to clean the data, correcting errors as necessary.