In my August 4th Star article on Healthcare and Moore’s Law I said: “…if you talk off-the-record to surgeons about a given procedure such as hip replacement, they will tell you they can now do two to four a day instead of one to two. The length of stay in hospital after such an operation is half what it was a decade ago. It is likely that the costs are actually declining even if the reported numbers appear flat.”

How could this be the case?  How can Ontario Case Costing Initiative (OCCI) based on CIHI’s MIS procedures be showing an upward trend when all of the “common sense” data and what we know about the procedures would suggest that the cost should be going down?

Well fortunately, I had a talented group of students at Rotman and a number of other mentors who taught me a lot about OCCI, case costing and about CMG’s 352 (simple hips) and 354 (simple knees) in particular.  Let me group the comments into: 1) Overhead allocations, 2) Outlier exclusions, and 3) Black box modelling

  1. Overhead allocations for OCCI for acute care hospitals are 2-3 times direct variable costs and sometimes more.  For every dollar of costs that are directly tied to a case, there are 2-3 “allocated fixed costs”. The costing model is very sensitive to getting variable cost allocations correct because each dollar results in 3-4 dollars of total cost.  It also means that total cost can go up even if the direct costs for a procedure are going down when fixed (overhead) costs are going up for the hospital.  

In general, these overhead costs are considered unaddressable and yet they are where the savings are.  Not because the hospital has “too much overhead” or “too many administrators”.  That is just silly talk from politicians.  The problem is that general hospitals are too complex an institution to be doing simpler stuff.  Or paraphrasing from Clay Christensen in the Innovator’s Prescription, once we have a procedure well specified and understood we need to take it out of the high overhead site and/or manage it as a value added process.  So for example, to apply this thinking to hips and knees: If we have $10,000 of total cost split $2500 variable and $7500 fixed, rather than getting the $2500 variable down to $1500.  Where we can save the big money is to move the surgery to a lower complexity center and cut the fixed costs from $7500 to $3000.

There is an alternative explanation raised by a person I trust who reviewed this paper.  The hospitals may just not be doing the work to allocate costs to individual procedures.  With the demise of the JPPC in Ontario, there is no discipline to the allocation process.  This explanation, if true, is really concerning as it would mean that the data we are looking at is largely junk.  And worse, that there will be lots of room to “reallocate” (aka game) when dollars flow on the basis of prior period costs.

  2. Outlier exclusions have a huge impact on the averages reported for all right-skewed distributions.  If a variable like cost per case has the following distribution 5*$6000, 5*$7000, 5*$8000,  and 1*$40000.  A reasonable outlier exclusion like “2x the normal case cost (or LOS)” will take the “average from $9500 down to $7000.  This is exactly what appears to have happened with CMG’s 352 and 354 in the time series from 2002/3 to 2008/9.  So for 2002/3 the maximum LOS in the sample for hips (352) is 12 days or less across all reporting institutions.  In 2008/9 the maximum LOS are (11, 14, 16, 22, 22, 24, 24, 28, 34, 44, and 49).  And even with the much higher exclusion numbers the reported “average” LOS  decline over the time period.  But the word “average” doesn’t mean the same thing when you change your outlier exclusion. 

There are a lot of good arguments as to whether it is better to have an exclusion or not.  For my part, I would say yes, as we want to pay for “tough cases” using a different bucket of funds to reward academic excellence.  But one thing is very clear: you cannot present a time series with an outlier exclusion change and not adjust the data. 

  3. Black box modelling is a derisive term for any model which is so complex that the policy maker who is the user no longer truly understands the workings of the model.  An ADM or Director needs to be able to understand conceptually why a model makes sense and how it is working.  When someone says that a particular procedure has increased in cost while the LOS and OR time are both down 50%, we need to have the courage to say “why?”  Why is that cost model showing an increase?  Run through the detailed math and explain to me how that could be so counter-intuitive.  Yes, there are legitimate reasons why new and better technologies may add some costs, or a new prosthetic might add costs while reducing OR time or LOS.  In general, policymakers need to stop using models that they fail to understand.  Particularly when the person or institution being paid through does understand it and is collecting the data that feeds it.  Otherwise, the information asymmetry will constantly be used to the detriment of the policy maker.

In general, prior period cost-based models have serious limits and are grossly overused and under-comprehended.  Normative rather than empirical modelling is probably a better way to go to set prices.  It is to that topic that we will turn in Part Two of this essay next week.

About the Author

Will Falk (@willfalk) is an Executive Fellow in Residence, Mowat Centre for Policy Innovation, School of Public Policy & Governance and Adjunct Professor, Rotman School of Management, University of Toronto.

For recent articles and speeches click here.