- Supply Chain Solutions
- Elite™ Enterprise
- Elite™ Healthcare
- Omni™ Retail
- Who We Serve
- About Us
Back to List View
After having painstakingly gone through all the legal, procedural, and data engineering pain points that I described earlier, I now have in my possession a data set from a hospital system that I can eagerly sink my exploratory analytics teeth into. The data, which is now ingested, formatted, prepared, cleansed, combined, scaled, and loaded in my data science environment, is ready to lead me on my Exploratory Data Analysis and Discovery journey.
What Was the Business Problem Again?
Remember, if you do not have a guiding business problem, you are wandering around aimlessly in what could be characterized as a data wasteland. However, in this case it happens to be preference card optimization. With many surgeons, each having their own procedure-specific preference cards, performing a variety of procedures many times over, and using a large number of disparate products, the logistics can quickly get out of hand—or so I am told. However, what does ‘getting out of hand’ mean? Let me dig in to find what it could potentially be.
The Process Behind the Data
Now that I know my business problem, sort of, I also have the data. So that comes next? Knowing more about the process behind all that data will put me in a better position to start making sense of it all.
Although this may come as second nature to perioperative staff, I have now learned that in preparation for any type of surgical intervention, products need to be planned, allocated, picked, prepared, and delivered to the OR according to individual preference cards. During the intervention and depending on how things go, more products are potentially requested, fetched, used, put aside, or discarded, and so on. After the intervention is over, products that are consumed during the procedure are tallied up for accounting/invoicing purposes. And those that are not are collected, sorted, sterilized (if they are reusable tools), and returned for storage until the next time, when the cycle repeats.
Also, during the surgical procedure, I have learned that the surgeon may request additional products (which may or may not be on their preference cards). These are fetched right on the spot by the clinical (as opposed to logistics) staff, who in most cases rush to get the items to have the least impact on the surgery, at least in terms of preventing a delay. Armed with this additional subject matter expertise in process insight, I feel that I am better equipped to dive deeper into the data and start making more sense of it.\
How Is All This Represented?
My curiosity and analytical perspective will naturally gravitate toward what I certainly understand and of course relate to, which in this case happens to simply be product usage information.
I discover that for each product necessary for a procedure, it is indicated as requested (presumably through preference cards), pre-operatively allocated and picked and post-operatively indicated as either used or returned. If used, items are tallied up for accounting purposes. If not, they are collected and returned to storage. Keep in mind that some of the products are reusable instruments and would require sterilization before storage. Simple enough. If you look at it in tabulated form, the data might be represented like this:
Note that all this data is for illustration purposes and real values have been completely replaced. I also do not differentiate between non-reusable and reusable products in this example.
Data! Tell Me “What Happened?”
I discovered that within a few short months (the period that the data in my possession covered), hundreds of surgeons performed thousands of procedures and used hundreds of thousands of products. While by themselves these numbers do not mean much, other than the fact that they represent a large and busy hospital system servicing a large population, the scale seemed staggering.
I discover that there are a multitude of situations, albeit in a smaller percentage, where multiple surgeons performed multiple procedures on a single patient. However, to simplify matters initially, I will intentionally overlook these cases and direct my focus to single-surgeon/single-procedure cases.
The data indicates that there have been hundreds of different types of procedures that have taken place at various frequencies. Also, it looks like total knee replacements (knee arthroplasty) have occurred at a much higher rate than other procedures, which in turn have been performed by a relatively large number of (seemingly orthopedic) surgeons working at the hospital. No great insight there but suffice it to say that the data is coming from a large orthopedic-specialty hospital, where numerous total knee replacement surgeries are performed.
Is a 30% Product Return Rate Too High?
As I dive slightly deeper, I discover that the total number of products which have been requested and returned ‘unused,’ is almost 30%, roughly half of which are reusable instruments and the other half are not. From an operational inefficiency perspective, that percentage of return seems to be rather high. All things considered, one hospital system estimates that fetching and returning a product costs around $17. Based on what I found with respect to the absolute number of products returned, which is in the hundreds of thousands, the labor cost not directly contributing to the well-being of the patient seems sobering.
After some further exploration, the data tells me that there have been a significant number of surgical cases where a considerable number of products is being returned post-surgery, unused. However, the puzzling part of this observation was that the products that were returned were not been requested in the first place. This seems odd! Is it a process failure or lack of understanding on my behalf? Naturally, the time seems right for me to start gravitating toward the ‘why did it happen?’ However, let me sort through the data further.
Given the size of the data, I decide to narrow the scope and the scale of my initial exploratory analysis by focusing on the total knee replacement procedure, which has been performed the largest number of times. Well, it looks like there are dozens of orthopedic surgeons who are performing total knee replacement procedures.
For this type of intervention, it came as no surprise that these orthopedic surgeons are requesting roughly the same number of different types of products to perform the same procedure. However, the surprising part (at least to me) was finding that the collective pool of unique products used by the surgeons for knee replacements was almost an order of magnitude more than what each used. Could this mean the more surgeons performing the same procedure, the more products to manage? Is this unique for total knee replacement?
Why Would a Requested Product Not Be Used?
Given my earlier assessment of 30% overall return, it comes as no great surprise that many of the orthopedic surgeons have requested numerous products that they never use. It seems that many of the surgeons are consistently requesting products (including re-usable instruments) but are returning them unused.
Is this because their preference cards require modification and updates, or is it because the products need to be there just in case? The ‘why did it happen?’ diagnostic analytics part seems to make its presence known. The data as it stands will not reveal that information, as there are no ‘non-negotiable’ or ‘just-in-case’ provisions in the data when requesting products. Maybe that is something to address in the future.
Being analytically driven, obviously I look at the numbers first and foremost. And from what they are telling me, I see an opportunity for improvement, at least from a labor overhead cost saving perspective, especially taking into consideration the estimates I cited earlier.
However, it seems that there is more than meets the eye. I am bewildered to find out that labor overhead—although a factor that can have benefits if addressed—is not at the heart of this healthcare business problem.
If Not Overhead, What Else Is Important?
Much to my surprise discovery, it seems that the labor overhead challenge, although beneficial to solve, does not seem to be the hospital’s top-of-mind concern. So, what is?
Rather than increasing labor efficiency by reducing product returns, it turns out that the primary business driver is patient health. Specifically, minimizing the risk of subjecting patients to post-surgical infection. Puzzled, I continue my inquiry. How is the high rate of product returns related to patient health?
After consulting with Tecsys point of use subject matter experts, it turns out that picking and returning products unused many times over, with trips back and forth from the OR, will unnecessarily start affecting their packaging. This often-undiscernible compromise in products’ sterile packaging likely increases the risk to patients in developing post-surgical infection.
Why could I not think of that? Simply because I was only looking at the numbers, purely from an analytical perspective, sort of in a subject matter expert vacuum. This is a classic example of how the data scientist, with individual skillsets, and the perioperative subject matter experts on their own could only do so much. However, when they synergistically come together and focus their efforts and skillsets on what matters the most and the reasons behind it—which neither group could do on their own—the whole becomes much bigger than the sum of the parts.
In order to develop a better appreciation for the problem and to see a couple of examples firsthand, I am curious to know more about the contents of a particular procedure’s preference card for various surgeons. Much to my dismay, I find out that preference card information for a surgeon/procedure pairing resides in another system not directly accessible, and thus not available. Leaving that hurdle aside for the time being, I let the data tell me what happened?
Continuing my explorations, I realize that there are more than a dozen orthopedic surgeons. Among them, one seemingly mysterious surgeon (designated as D-14), has been the most prolific. I shall hereafter refer to him as Dr. Smith, who has performed the lion’s share of knee replacement procedures, again during the time that the data covers. Furthermore, Dr. Smith has been consistently using a considerably larger number of unique products than the rest of his peers. In fact, Dr. Smith uses as much as 77% more than the average of all his peers, which is between 60-70 products per case.
From a bird’s eye view of Dr. Smith’s product usage for knee replacement procedures, I get a glimpse at his preference card make-up. If the same set of products is requested for each procedure, then chances are, they are on Dr. Smith’s preference card for total knee replacement. I do not need the list after all from this third-party system; I know exactly what it is. No great revelation there. Had I had access to the actual preference card, I would be in a position to validate my finding. But I don’t.
October 31, 2018
Perioperative Analytics: Exploratory Data Analysis in the OR (Part 1 of 4)
May 14, 2019
Perioperative Analytics: Exploratory Data Analysis in the OR (Part 4 of 4)
April 2, 2019