The randomness in episode costs has always troubled the Singletrack Analytics team. Even for high volume episodes, randomness makes it difficult to gain a good understanding of financial performance. After CMS released the mock reconciliations, we started wondering how much variation would be present in the quarterly average costs that would be reconciled throughout the performance period. This triggered the analysis below.
How randomness affects episode cost
In this analysis, the hospital participated in three episode families – cardiac valves, spinal fusion and major joint replacements. Since episode cost must be computed on a DRG basis (rather than as an episode family) we looked at the average variation across quarters for individual DRGs. For this analysis we arbitrarily eliminated DRGs having fewer than an average of 10 episodes per quarter, since those DRGs had widely varying episode costs. For the remaining DRGs we computed the standard deviation of the quarterly average episode cost, and from that computed the coefficient of variation (CV), which is the standard deviation divided by the average. The coefficient of variation shows the range within which the average cost would lie approximately 2/3 of the time, assuming the data is normally distributed.
The results are shown below. For DRG 216, having average of 17 episodes per quarter, the cost per episode is likely to vary ± 11% from quarter to quarter. Even for DRG 470, having 150 episodes per quarter, the quarterly variation in episode cost could be ± 5%. This may mean that quarterly reconciliations may require a constant stream of payments back and forth between the participant and CMS, even for episodes in which performance is generally improving.
Note that this variation is lower in the Medicare BPCI program than it would be in many commercial bundled payment programs for two reasons. The first is that Medicare hospital payments are based on case rates that don’t vary with length of stay (absent outliers). Many commercial payer contracts are based on per diems, which means that the episode cost variation would be much greater than for the same episode in which hospitals are paid on DRGs. The second reason is that episode counts for many commercial payers may be lower than they are for Medicare, which would increase the variation due to random patient selection.
Effect of variation on reconciliation
The effect of this variation on quarterly reconciliations will be significant. The table below shows an example of the quarterly variations in reconciliation amounts from the above data, using the Grand Total average episode cost as a surrogate for the target payment amount. As can be seen, the reconciliation payments can vary between quarters, which may make predicting cash flow from reconciliations difficult.
For the participants to understand their actual performance, a longer-term view will be necessary. The graph below shows the average episode costs for the above DRGs plotted alongside exponential trend lines. Exponential trends are used because they indicate the percentage change over time. (If you want to prove this to yourself, create an Excel table that starts with 100 and increases by 10% down a column; then graph it and add an exponential trendline. You’ll find that the trendline exactly fits the data.) The exponent of e in the equation for the line shows the average percentage increase or decrease over the time period. Thus, the average rate of decrease in episode costs for DRG 219 is 2% per quarter, determined from the portion of the equation e-0.02x. DRG 460’s cost is increasing at an average rate of about .3% per quarter, where DRG 470’s cost is decreasing by about .7% each quarter.
This amount should be compared to the trend factor that CMS will be applying to historical costs to compute the quarterly targets. If the trends continue, the participants cost should ultimately end up lower than the target if their average rate of increase or decrease is lower than the corresponding CMS trend factor.
Drilling down in the data using trends is also helpful. In this case the hospital had implemented a strategy of migrating post-acute care from an inpatient rehabilitation setting to a SNF setting. As the IP rehab costs dropped the SNF costs increased, although not commensurately. Additional quarters of data will be necessary to determine the new cost trends under this care management arrangement.
Measuring trends will be increasingly important as care management strategies are implemented. They’ll also be important for financial reporting, since estimates of surplus or deficit must be accrued for financial statements and the variations will need to be “smoothed” to provide an annual estimate of the results. Similar smoothing will be necessary to determin amounts eligible for gainsharing payments. Hopefully costs will trend lower as these strategies begin to work, and thoughtful analysts will measure the “before” and “after” trend rates on an ongoing basis to evaluate their effectiveness.