Staff turnover has been an area of concern and vigorous debate during the Pandemic. The media has published numerous articles bludgeoning the nursing home industry for pretty much anything and everything, some of it warranted, most of it not. Politicians, as they often do, have assigned blame to others (the nursing home industry, staffing agencies, private equity firms, etc.) liberally, whilst at the same time ignoring their own culpability in creating a situation where the price of labor has significantly increased, conditions in nursing homes, already challenging, are made more so by an incessant wave of novel and ever-changing regulations, laws, and in a sick twist of policy, reimbursement diminution. The Germans are really good at naming unfortunate things or situations. They have a word, backpfeifengesicht, which means a “face that needs to be slapped”. Say you trip, and some guy has a smirk on his face as you get up and you feel like smacking that smile off his face? That’s backpfeifengesicht. For some reason, the politicians feel backpfeifengesicht towards SNFs.
My diatribe above aside, one area where the government or CMS, in particular, has been very good has been the release of timely, topical, and elucidating SNF data. One interesting dataset for the analysis of staff churn is the PBJ Employee Detail Nursing Home Staffing Dataset. This dataset provides granular daily staffing for Q2-2020 through Q3 2021 (18 months). This allows for some interesting insights into the staffing situation in SNFs during the heart of the Pandemic.
One area of staff turnover that has, in my opinion, been underreported is that of the turnover of SNF nursing leadership. In large multi-faceted organizations like SNFs, leadership provides institutional direction, consistency, and memory. This is particularly important during and after times of crisis. The data within the PBJ Employee Detail Nursing Home Staffing Dataset paints a dour picture of leadership churn.
Our initial analysis is that of Directors of Nursing a/k/a DONs. Let’s start by taking a look at the total number of Directors of Nursing per SNF by state from April 2020 through September 2021.
The chart shows that most SNFs had between one and ten DONs during this time period. In itself, this is concerning. However, many SNFs had far more DONs – more than 25 and up to 64 DONs. The skeptic in me suspects that these extreme outliers contain reporting errors, especially in the far tail(s).
It is obvious that the distribution of the count of DONs per SNF does not follow a normal distribution, which means that an arithmetical mean would not be a good way to characterize the average number of DONs per SNF. Fortunately, we can lean on Bayesian estimation. Our model estimates the average number of DONs per SNF as follows:
We can see that Florida, North Carolina, Maryland, Virginia, and Ohio had the highest turnover, in terms of average DONs during the observed time period. These SNFs in these states had approximately 3 DONs, during this 18-month period. Or put in another way, they churned 2 DONs. Remember, this was the average outcome …. Even on the other side of the scale, say in South Dakota or Utah, the average facility still churned a DON. It is highly unlikely that a single DON remained in place for the entirety of the time period, even in the best of circumstances.
Now, one might be tempted to take the time period and divide it by the average number of DONs to arrive at the average tenure or length of employment. This would not only be naive but also wrong. This owes to what is called “censoring”. The data we have has a window of April 1, 2020, and September 30, 2021. Most of the people we see in April 2020 will have been DONs at the SNF before that date. We just cannot see their real first date at the SNF. This is called “left censoring”. We also have people who start late in the window, say in September 2021 whose end date we do not actually see. In the data, it shows up as the last day(s) of September, but it is more likely that they were actually employed past September 2021. This is called “right censoring”. Another case of “right censoring” are people who start before September 2021 and end in September 2021. For a better estimate of tenure, one has to attempt to account for censoring, lest estimates fall artificially low. Our model sets left and right censoring parameters, and the model will estimate the censored values after it has fit the non-censored data. The results are below:
We can see that the average DON tenure in North Carolina is between 95 and 102 days; or just over 3 full months. On the other side of the scale in NY, which is the best performing, it is between 125 and 133 days, or just over 4 full months. This is suggestive to me of some sequence of multiple shorter tenures, and a singular longer tenure, probably at the start of the pandemic. However, this would require further analysis to conclude.
SNF staff churn is going to be an ongoing issue. The real question is for how long?
