Home   About Kurtosis   Courses   Course Calendar   Booking Information   Ideas   Technique   Clients   Contact



Counting Outliers

How should we measure how many hospital inpatients are in the wrong beds?

The word safari means “long journey” in Swahili. A safari ward round is a long journey round a hospital by doctors hunting for their inpatients in a wide variety of—often inappropriate—wards.

Safari ward rounds are a symptom of dysfunction in a general hospital. They happen because inpatients are being accommodated in the “wrong” beds, the “wrong” wards. And the reason why patients end up in the “wrong” beds is because there weren’t enough of the “right” beds when they were needed. Inpatients in the wrong beds are usually known as outliers and it’s important that we are able to measure the extent of this dysfunction. We need to be able to count how many of a hospital’s inpatients are outliers.

The most common way of doing it is to measure the number of outlier bed days. On any given day, how many patients are being accommodated in the wrong beds? Another—less common—way is to measure the number of outlier events. Each outlier is only counted once, regardless of how long they remain in the wrong ward. This measure gives a clearer idea of which days were feeling the most pressure, on which days the most decisions to decant patients took place. The difference between these two measures is a bit like the difference between prevalence and incidence.

So, there are two ways of measuring it, but how do we actually do it in practice?

My experience suggests that in practice a lot of hospitals measure outliers manually. They rely on the bed manager to maintain some kind of paper-based record of how many patients have been moved as outliers. But I want to concentrate here on how to count outliers electronically. Which means that we need to turn to our patient management system (PMS) and get our hands on a data extract that gives us a row for each ward stay. This is not something that is frequently done. It is far more common for PMSs to generate episode-based data extracts or spell-based data extracts; you don’t often come across ward stay-based extracts. But you need one for this, because the definition of an outlier requires it. There has to be a “mismatch” between the specialty and the ward for us to be able to count outliers electronically.

As well as a ward stay data extract, you will also need to have an up-to-date version of the hospital’s specialty bed allocations. You need to know exactly which ward belongs to which specialty. If you’re in luck, you’ll have one already at your fingertips: a bed allocation spreadsheet that’s been lovingly maintained by someone who knows their way around the hospital. If you’re out of luck, you’ll need to compile it yourself.

Thirdly, you need to have a good working relationship with the bed manager. You'll be spending a lot of time in his or her company as you go through this exercise.

Let’s suppose that you’re trying to do this for the most common outlier scenario: medical patients accommodated in non-medical beds. In most general hospitals it’s medical outliers that are the biggest problem. Here’s what you do. First, using your ward stay extract, you need to create a table of data that contains all of the inpatient ward stays that were “live” during a given period. So if—for example—you are looking at the month of August 2011, you will need to get all of the ward stays with a start date on or before 31st August and an end date after 1st August. You’ll also need to include any ward stays that were still “open” at the end of the month.

Second, you need to sort all of these ward stays into strict chronological order within each inpatient admission. It’s probably reasonable to assume that your data extract will contain a unique patient identifier, so if you sort your data by this identifier and then by the ward stay start date, you should achieve this objective. This step is not absolutely strictly necessary to complete the task, but it will be extremely useful when it comes to checking through the records.

Third, for each record in your data extract, you need to match the ward with the specialty to see if the ward stay was an outlier ward stay or not. This is where you need that spreadsheet I mentioned earlier. Essentially, you need to go through every ward in the hospital and decide whether medical inpatients “belong” in that ward or whether they would be classified as “outliers” in such a ward. A lot of the wards will be easy to do. An Acute Medical Unit will be a “belonging” ward, a Surgical ward will be an “outlier” ward. But some wards will be trickier. An ITU, for example, will accommodate inpatients from all kinds of specialties. But the key question to ask is: would a medical inpatient be there because that was the right place for them to be, or because they were there as an outlier? Once you’ve answered all of those questions, you can create a lookup table that you can use to label each ward in your extract. Each ward will be labelled as something like either “Medical” or “Outlier”.

Fourth—and you need to do this at the same time as you do the third thing—you need to group your hospital’s specialties into broad aggregations that match the bed managers’ view of the world. This is the stage where you might, for example, take Acute Medicine, Respiratory Medicine, Gastroenterology, Renal Medicine and Cardiology and group them all under the heading “Medical”. General Surgery, Vascular Surgery and Urology might all get grouped together under “Surgical”. And so on. This will give you another lookup table which you will use to label each specialty in your extract.

Before we go onto the fifth thing that you need to do, we need to mention that one difficulty you may encounter— and this one is potentially a show-stopper—is the complication of multi-specialty wards. If you have a ward—let’s call it Ward 8—that has a mix of Medical and Surgical beds, and there is no “dividing wall”—virtual or otherwise—between the beds, then this presents you with a real problem. You simply have no way of knowing whether a medical patient is in a medical bed on the ward or on a surgical bed in the ward. One solution to this would be that if you know that there are—say—six medical beds, every time there are more than six medical inpatients in Ward 8 then you’ll know that they are outliers. But this is a somewhat ungainly solution to a problem that is probably best dealt with by dividing the ward into two separate wards on the PMS. This problem—by the way—is what will prevent you from working out whether Renal patients are being accommodated in Renal beds or Gastroenterology beds if you only have one ward that is a combination of both and nobody identifies which beds are which.

Sixthly—and finally—you are now in a position to actually count the outliers. If you are doing this in Excel*, you just need to create two pivot tables. One pivot table will have ward as a row label, medical/outlier as a column label, and summed ward length of stay as the data value. This will enable you to count outlier bed days. The second pivot table will be identical apart from the fact that you will count the ward lengths of stay instead of summing them. This will enable you to count outlier events.

Being able to count and report outliers won't of itself make the problem go away. But it will help people get a better grip on the problem and create a greater sense of urgency for those who need to tackle the problem of safari ward rounds.

*If you want to get your hands on a worked example of how to accomplish each of these steps in Excel using dummy, anonymized data, please feel free to send an email to info@kurtosis.co.uk and we’ll reply, sending you an Excel workbook.

[23 September 2011]


Make a comment


Job Title:


Email address:

© Kurtosis 2011. All Rights Reserved.