A hospital focused on improving its clinical performance will spend some time on an analysis of its patient data including length of stay, potential excess days and other clinical quality measures as compared to benchmarks. When the data is organized on a month by month basis, it can be difficult to look at meaningful trends by month if there are a few abnormal or atypical discharges that distort the data.
Management may want to exclude some of these atypical or unusual patient accounts to obtain a clearer picture from the data. These unusual patients can be considered to be outliers if they have a length of stay that is atypical for the hospital, such as a length of stay of 50-100 days or more. An outlier in terms of length of stay can have a significant influence on the mean and standard deviation of length of stay for a month. But how do we know when excluding outliers is the right thing to do and will provide us with a more meaningful and useful analysis? Just because a patient stays longer than average doesn’t make them an outlier. In fact, extreme data points can contain useful information about the hospital and clinical outcomes, particularly if they occur with some regular frequency. A data point isn’t an outlier if it happens every month.
Excluding the outlier isn’t the only option. Instead of removing the patient from the data, it may make sense to truncate the outlier(s) at a more reasonable length of stay, such as 30 days or something that occurs fairly regularly in the data. Then the analysis still contains some partial recognition of each of these observations. A simple rule of thumb based on standard deviations above the mean may be a good place to start in terms of recognizing what can be considered an outlier. Sometimes a visual inspection of the data may also provide some clues. For example, see the graph below for a visual example of an outlier. Also, no data point should be removed before the reason for the long length of stay has been investigated for the individual patient (occasionally there may be erroneous data).
Its also important to ensure that all time periods, such as a baseline and actual time period, are treated consistently with respect to outliers. In other words, comparisons must be done on an apples-to-apples basis. As an example, suppose a hospital has implemented a process which will result in several long term patients being moved from the hospital to skilled nursing facilities. As those patients are discharged in the current months, length of stay may appear to temporarily increase due to the effect of these outliers in the data. A more meaningful analysis of trends in length of stay would adjust the data for these outliers after correctly identifying the patients affected by this new process.
Managing outliers in clinical analytics requires some expertise and judgment in order to make an informed decision. Whatever approach is taken, make sure to:
- Investigate the data
- Review the results of the analysis both with and without the outliers
- Communicate any assumptions to the users of that analysis.
For more information on managing outliers and how to leverage clinical analytics to improve clinical performance, contact [email protected] or call (888) 341-1014.
Sharon Carroll, Sr. Financial Consultant & Actuary with Clinical Intelligence, LLC