- Peer groups are useful because they allow healthcare organizations to compare themselves to others to identify performance outliers.
- Healthcare leaders frequently seek peer groups that resemble their organization. While having peers that are similar is important, finding exact matches isn’t practical.
- A data-driven strategy can be used to ensure a peer group will permit accurate comparisons. With data to prove that performance improvement opportunities exist and evidence that their peers are executing on those opportunities, leadership teams are motivated to pursue targeted corrective actions.
Benchmarking operational performance in healthcare has been a mainstay for many decades. However, the debate remains as to what constitutes a “good” basis of comparison, especially when looking at departmental peer groups. Below, we explore the key factors that should be taken into consideration when selecting peers for benchmarking purposes.
What is Benchmarking?
Benchmarking is the process of standardizing and comparing performance metrics of identified “like operations” to your own performance. This method allows healthcare organizations to learn from one another and ultimately incorporate better practices to improve outcomes.
It’s our recommendation to incorporate benchmarking into the larger organizational workforce management (WFM) program and associated performance goals. Here are a few key components of a best practice approach to an effective WFM program:
- Develop an integrated labor benchmarking and productivity approach with a tactical day-to-day management capability.
- Inform labor productivity standards with annualized benchmarks that, in turn, align with the budget targets.
- Utilize a peer group methodology to support your organization’s WFM strategy.
Often, when first starting out, department leaders gravitate towards peer groups that look just like their departments. While it’s important to have like peers, exact matches are neither realistic nor practical. Traditionally, when selecting peers, the following five steps are taken:
Although effective, this method of forming peer groups has some drawbacks. Here’s why you should rethink the traditional approach to peer groups:
- Choosing Criteria: It’s difficult to know what characteristics matter to performance – typical criteria include teaching status, size, geography – but do these always explain variation in performance?
- Searching and Filtering to Find Peers: Sometimes, very few peers meet all criteria, resulting in a small sample size.
- Comparing Performance: Is your peer group any more appropriate than a random group of hospitals? Does it provide a meaningful comparison? How do you know?
How Data Can Help Fix Peer Group Challenges
There is an alternative to this traditional approach to selecting peers that relies on data science. This data-driven methodology takes the following three items into consideration:
- Feature Selection: Use a statistical technique that identifies which characteristics explain variation in the performance metric of interest. It finds a short list of hospital characteristics (features) that explain the most variation in performance and estimates the relative importance (coefficient/weight) of each feature.
- Clustering: Statistical clustering techniques identify peers that are similar in ways that matter to performance. Weighted k-nearest neighbor clustering incorporates the relative importance of features (regression coefficients/weights) to find a fixed number of peers that are similar in the ways that matter most.
- Quantify the Usefulness of the Peer Group: Ask a few questions to determine the measures of peer group quality. How much variation in performance is explained by the feature selection? How similar are my peers to each other, relative to similarity with all hospitals?
- High R-squared (regression): a peer group similar in ways that explain most of the variation in performance provides an interesting basis for comparison.
- Low R-squared (regression): a peer group similar but in ways that do not explain much variation in performance, not much better for comparison than an arbitrary or random selection.
As an example of how this data-driven approach can be used, PINC AI™ data scientists compared a peer group consisting of large academic medical centers with a suggested peer group (data- and AI-generated) that includes non-academic hospitals. The PINC AI™ team found that the suggested (non-academic) peers are actually more similar in ways that explain performance. A peer group of only academics is more scattered (dissimilar in ways that explain performance) – and has a wider range of productivity – than the suggested peer group.
A data-driven approach provides a way to ensure that a peer group will allow for a meaningful comparison of the outcome of interest. Peers selected based on criteria that don’t explain variation in performance results in:
- A peer group that is no different from a random sample of all hospitals.
- A peer group that is different in arbitrary ways from all hospitals.
- The possibility of a hospital focusing on improvement opportunities where none exist due to misleading results.
A data-driven approach avoids selection bias when choosing peers and ensures performance efforts are focused on the right things. Ideally, it is recommended to combine data and machine learning with human expertise for best results.
For more on this topic:
The insights you need to stay ahead in healthcare: Subscribe to Premier’s Power Rankings newsletter and get our experts’ original content delivered to your inbox once a month.