In the last essay, I explored the data architecture issues associated with constructing the Golden Egg, a model that would align employee data and patient data to allow one to analyze movement in one space based upon movement in the other space. On top of these concerns, there are issues with the broader logical construction that makes the analysis difficult to execute. In this essay, we will explore the logical issues that add to the struggle for a good model intersecting patient and employee attitudes.
Again, before I start, I will add my caveat that I have not seen this work done well. If there are those out there who have cracked the problems (or are attempting to crack them), please reach out to me. I would love to hear how you have addressed the data and logical issues and give you your roses.
The Bucket Issue
In the last essay, I suggested that the best way to solve the data architecture issue is to bucket your data by discharge unit. It is the smallest unit of analysis that will allow the connection of data. The problem is that the discharge unit is not a clear or clean division of data.
From patient’s perspective, they are likely to have multiple encounters in many spaces, and their discharge unit is just the tip of the iceberg. A patient might be admitted through the emergency department. They may be transferred across multiple units. Saying that they were discharged from 5West, a med/surg unit, does not say very much about their actual journey though the hospital. You know, unless every patient follows the same journey.
From the employee’s perspective, there are a large number of staff who are not tagged to a specific unit. On the clinical side, folks in lab, rehab, or imaging for example, will treat patients from any unit in the hospital, sometimes by coming to the patient and sometimes by having the patient brought to them. In some hospitals, depending on the workflow, they may have inpatients, emergency department patients, and outpatients all mixed into their daily worksheet. Plus, to the extent that it matters, they may not even know what sort of patient they are working with.
The non-clinical crew is even more scrambled. Housekeeping, facilities management, food service are all delivering their services wherever they are needed and are not locked into one unit, or even a subset of units. Some hospitals may have assigned spaces for housekeeping or food service, but often hospitals use a more flexible next-up model, where when a housekeeper is finished in one place, they go to the next need. If a room needs to be turned over after a discharge, that is often a all-hands-on-deck moment because of the urgency to admit another patient. Likely the most extreme example of this is patient transport, who will often have an ad hoc approach to their workday.
Perhaps the only employee of the hospital who is tagged to a specific unit is the nursing staff. They usually are assigned to a specific unit based upon skillset, license and preference. Still, even here, there is variation, with float nurses going where needed. At some critical access hospitals, even staff nurses may be moved from, say, OB to med/surg, based upon need and whether there are any new moms and babies in the hospital.
What all this means is that while we can accept and perhaps even account for some noise based on how a patient moves through the hospial, we struggle to create clean lines of demarcation in how to assign the employees. Some might choose to forego including non-clinical staff, or even clinical staff who cannot be appropriately designated. This may be cleaner, but it also is removing many employees who interact with patients every day and have a role in forming that patient’s perceptions.
Visible versus Invisible
This allocation problem becomes more problematic when you consider the broader collection of staff at a hospital. While I am fond of pointing out that all staff are patient-facing, because they all deal with patients directly or indirectly, they are not all patient-facing in the same way or to the same degree.
So, while in theory, no employee is invisible, in practical terms, some are decidedly less visible than others. For example, the FNS (food and nutrition services) person who cooks all day in the kitchen and rarely sees a patient is less visible than the nurse performing hourly rounds on a patient. Further, there are employees, like scheduling, registration, and billing, who primarily bookend a patient encounter rather than being in the middle of the care arc. And then, there are the administrative leaders, who may not interact with patients directly, but certainly set the tone and priorities for their staff.
Then there are the staff that may only interact with a subset of patients and in very specific circumstances. While their jobs are far more complex, in a patient’s eyes, they may only see someone from security or facilities management when there is a problem that needs to be addressed, which colors the importance and the directionality of the experience. In other words, if I only need to see facilities management because my TV is broken, that repairman’s disposition may matter less to my hospital evaluation than the perception that the hospital is falling apart. After all, if they cannot keep the TV working, how can I be sure the MRI is working?
Even in situations where we can be confident that the staff is visible, it doesn’t really mean very much. Afterall, while a unit may have forty nurses on staff, the patient may only interact with four or five of them. The same goes for hospitalists or imaging techs. The patient is not getting a perception of ALL of them, but only the ones that they interact with.
Model building
Now, you may roll your eyes at me, imagining that I am trying to make this seem more complicated than it needs to be. After all, we don’t need to examine every grain of sand to know we are standing on a beach. Heck, I just acknowledged in the previous section that while everything matters, it all doesn’t matter to the same extent.
The moment, though, you make the case that there are degrees of importance in the ranking of employees, you are advocating for the creation of a model where some elements are weighted heavier than others. So, for example, a patient who comes through the emergency department gets those employees added to the equation, but those who walked in by the front door do not, though they may get registration staff employee data included instead.
As you piece apart and reassemble the data, you can quickly realize that this can be a complicated exercise. As you add elements, like an ED admission, or experiences on multiple units, having a team of doctors, while getting lab work, imaging and food service across several days, you can create a very long and detailed equation. If you further add weights to allow for nurses to be more important than transport people, this can grow to monstrous proportions. As any statistician can tell you, having a more complicated and nuanced model does not make it more useful in explaining variance nor easier to turn into action-plans. This is often called “over-fitting.” On the other hand, having NO model and putting everything into one big stewpot may not be subtle enough to detect any relationship at all.
Data Mush and Interaction Effects
So, you decide to save the heartache and just jam all the data together to maximize n-size and avoid having to play reindeer games with the math. But this assumes that all staff are equally happy or unhappy, or that in combining the data, it all averages out. But that averaging out is part of the problem. Imagine three hospitals have the same employee satisfaction score for their med/surg unit. At one hospital, no one is thrilled or miserable, everyone is just average. At another hospital, the clinical staff is thrilled, and the non-clinical staff is miserable. At the third hospital, this is reversed, the non-clinical staff loves their jobs, but the clinical staff hates working at the hospital. All scores are the same, but are all staff the same? Would we expect that the patient experience scores are the same at all three? I would argue that, NO, we would not expect that they would be the same. We could grab a beer and argue over which hospital produces the best or worse experience, but I think we would agree that the experience is not the same.
It gets more complex when you consider that much of the impact that, say, food service has on a patient’s experience is filtered through the nursing staff. If a patient is hungry, does the nurse say, “There is a card on your end table, but at this time of day, don’t hold your breath” or do they say, “We are coming up on lunch time, so let me help you get an order in, so it will come up as fast as possible.” I will always say that if you don’t set a patient’s expectations, they will set them for themselves. While the friendliness of the food service staff is critical, so also is the solicitousness of the floor staff.
I worked with a four-hospital system that had the food for all four come out of one kitchen.1 That one kitchen would prepare and package all the food and it would be driven to each hospital where it was microwaved and served. You might think that the food service scores would be great at the one hospital that had the kitchen and horrible at the hospitals with only microwave ovens. But it was the reverse. This was because the staff at the satellite hospitals knew there was a delay in getting food to patients, so they assisted patients in getting their lunch orders early, so there would be no delay when lunchtime came. By helping patients be strategic, they could deliver on the experience, and the patient would never know that the food they ate came from across town. Meanwhile, the staff at the hospital with the kitchen didn’t have to think strategically about food service, so they didn’t. Perhaps they figured that the kitchen was in their building, so they got priority, or perhaps they didn’t even know that all the system’s food was coming out of one kitchen. Either way, they didn’t manage the perception and that cost them.
All of this means that creating a model that captures the reality of a patient’s experience is challenging. It may also be unique, given that there is variation in how hospitals staff or schedule. So, the model that might explain variance at one hospital is useless at another, based upon whether they have hospitalists or not, run 8- or 12-hour shifts, employ or use third-party vendors for housekeeping, food service or anything else.
Key Variables and Key Drivers
One other issue with linking the patient and employee data is that one often forgets that the datasets were collected for very different reasons.
- Patient data is collected to satisfy CMS requirements and is focused on doing the things that maximize the likelihood that a patient will have a positive outcome. Outside of this, there is a desire to have patients continue to be patients and not leave for a different health system. If possible, it would also be nice if the patients said nice things about the hospital at family get-togethers or block-parties.
- Employee data is collected to improve retention. Outside of keeping staff from looking at other options in a tight labor market, they can try and determine what things make the organization an appealing destination for prospective employees. Beyond that, having employees tell their friends and family that this hospital is a great place to work is a nice cherry on top of the sundae.
Our intentions for surveying are similar, perhaps, but not the same. And while making the broad platitude that “happy staff makes happy patients” feels truthful, I am not convinced that it is that simple or that linear. Just as I am not convinced that “unhappy staff makes unhappy patients.” The reality is that it depends on how one defines “happiness.”
What drives an employee’s happiness or unhappiness may or may not affect patient experiences. One thing that is often important for staff satisfaction is Senior Leader Communication. But what communication is it? I will argue that staff who are confused about appropriate policy regarding transferring patients within the system, or what drives an ED going on diversion or limited diversion MIGHT impact patient care, because these are elements directly related to what happens to a patient. If there is confusion, patients are in limbo and that breeds anxiety. If the staff is confused about the changes in the health plan or the 401K, this may lead to dissatisfaction with the hospital as a place to work, but I am not convinced that this frustration will bleed into interactions with patients.
Yes, in both cases, if an employee complains to a patient about the hospital while delivering care, it will impact how a patient perceives the hospital. But setting aside that direct bitching and moaning, I am not convinced that general dissatisfaction with senior leaders will necessarily lead to impacts on patients. If a patient is boarded in the emergency department because the transfer policy won’t prioritize them, that can lead to bad scores. I am not convinced that dropping the employer match from 7% to 5% though will affect how patients see the care. Put very simply, if staff treat patients differently because they did not get the raise that they were expecting, you have much bigger problems with staff than as seen in patient perceptions.
In the end, while I do think that ecstatic staff is better at delivering care than morose staff is, I also think that for the 70% of organizations within one standard deviation of the mean, expecting slight shifts in one number to be reflected in another is demanding too much from the model. Combining this with issues in data construction and underlying logic, there is not enough juice here to justify the squeeze. Of course, given the market potential of marrying these two concepts, none of this will stop people from trying anyway. Good luck and Godspeed to you, noble warrior. Let me know how it works out.
1If this seems strange, consider how much work at a hospital (like laundry, purchasing, etc.) is farmed out or centralized in hospital systems.
PX Matters with Joseph Snipp 
Leave a comment