Since yesterday, therapy advocates from OPIRA, TAHCH, and others, have been meeting with HHSC officials to clarify the amendments that were added to HB 1 and of which were designed to reduce funding for Medicaid therapy reimbursement codes. HHSC replied with several of their answers to questions pertaining to the A&M study from therapy advocates, including, it seems, some of our own.
In our judgment, these answers were inadequate and inaccurate. For example, rather than address the limitation of the 11-state Truven data, HHSC deflected back saying that those 11 states were representative of the US Medicaid population, in general, and were considered along many demographic factors including geography and race. Since Truven cannot release the names of the states involved in such a subgroup, we have no way of scrutinizing this claim, nor does HHSC, as they admitted. The 11-state Truven data-set represents a stratification of only 27.5% of the states offering Medicaid therapy services. Without knowing the 11 states in this data-set, this stratification may be skewed and/or biased and may indeed not statistically represent the US Medicaid therapy population adequately. While effective stratification sampling may be done, we do not know how this stratification sampling was done and with what statistical expertise and to what ends. It turns out that the California rates were extreme low outliers in this study. Nonetheless, the report continued to showcase California’s rate ratios with Texas even though HHSC points out that they used the median indicator to do exactly the opposite.
HHSC pointed out that of the states’ data-sets that were used, the full Medicaid claims populations were collected and hence, no statistical estimation was needed nor appropriate. While this may be true (there still may be gaps in the collected claims data), unfortunately, using 15 out of the 40 states offering subgroups Medicaid therapy services is not a full Medicaid population for the nation and may indeed skew and/or bias the selection if that stratification was not adequately representative of the national Medicaid therapy serviced population. Our full national published rates data bears this out if projected to the actual paid claims data.
HHSC also reiterated that the median was used to better indicate the central tendency (location) of the distribution of paid claims per therapy code among the 11-state and 4-state individual data-sets. Nonetheless, the use of a median does not, in and of itself, guarantee a more robust central tendency of the distribution since we cannot see the distribution itself.
The mean, median, and mode indicators of central tendency depend on the type of metric function that measures the distance between any two points in the distribution. The use of one type of metric over the other is oftentimes not based on practical or even scientific reasons and is mostly chosen based on convenience of calculation. It may also depend on the shape of the distribution itself to be robust. Each central tendency estimator can be expressed in terms of another and so, there is a mathematical precedent to compare why one would use one over the other based on the shape (skewness, kurtosis, and modalities), sparseness, and dynamics of a distribution.
The median is a sledgehammer approach, as we had previously indicated in a post. Both high and low outliers are ignored equally (50% cutoffs on each side) without knowledge of the amount they differ (differential or pseudo-variance) individually from a central tendency location using the appropriate metric function. We had suggested a more surgical approach utilizing alpha-Winsorized or non-parametric approaches.
What is more interesting is that our own complete 40-state published rates data-set median showed a significant difference from the 11-state Truven published rates data-set median. In their own summary, the paid and published rates tendencies were similar. Hence, a 40-state paid claims data-set should show significant differences from the 11-state Truven A&M study results.
In any case, we are trying to gather EOBs to demonstrate real (and bundled) reimbursement patterns among Texas that reimburse more than (managed) Medicaid. We are asking for two things:
(1) EOB copies (with PHI and prov ID removed) for any commercial insurance (variety please) for any services performed with bundled codes (for example 92507 + 92526 used on same date of service) or bundled eval + therapy performed on same date of service.
(2) We also need EOB’s for commercial rates in excess of $70/visit.
Send by these EOBs to: firstname.lastname@example.org. We need these emailed in by end of business day today.
If you have any questions you may call us at (281) 734-1549 or reply to this email. The purpose of this data collection is to demonstrate to HHSC a better indication of how (managed) Medicaid stacks up against commercial rates in Texas and with respect to these particular codes to start things out.
Another approach would beto rank therapy codes based on their utilization in the state (both in frequency billed and collected on), project a truer Medicaid population for 2017-2018, anduse those numbers to weigh each code for a distributed reduction scheme, one in which access to therapy care will not be statistically significantly lowered, while still achieving the state’s targeted saving of $200M.