Predicting Medicare Underpayments Using an LMS algorithmPowerPoint PresentationSlide 3Slide 4Slide 5Predicting Medicare Underpayments Using an LMS algorithmTed ShultzDecember, 2001University of WisconsinProblem descriptionBill:Band-Aid $0.12Aspirin $1.04New Hip $1,000.00Gauss $12.00Gloves $3.75-------------------TOTAL $1016.91MEDICAREPayment:(no explanation)-------------Total $412.63Medicare pays Vanderbilt a different amountVanderbilt Bills Medicare for one amountProblem Explanation?Why?Comparison Comparison of Methods:Never been done before with Medicare!Simultaneous equations methods:Inverse matrix methodMuch to large a matrix to inverted on a convention computerOrthogonal-triangular decomposition (Matlab backslash operator )Unable to sort though possible answer to determine optimal solution based on input parametersModified LMS methodSlow, but able to bracket answer LARGE data file (443,964 purchases)Techniques Techniques used to handle large data file•Do all file manipulations in a data base program•Significant time savings•Bracket weights after each weight recalculation•Know Medicare will pay between 0-100%•Automatically resize •Start larger, but shrink for accuracy•Auto save and resume capabilities are required•CAE tethered server crashes every few days•Requires two days to load and format Matrix! (400Mhz)•Two weeks of calculations (by project definition)Results-Conclusions Only about 1 week of way into calculationsFull reimbursementFixed percentNo paymentStill moving ornegotiated rateBilled amountGuess pay amountPotential to have HUGE impactAbout $32M in charges, $5.5 M in
View Full Document