Which PCI Operator Are You? Four Phenotypes Emerge From Big Data

Categorizing PCI operators by “practice patterns” can provide feedback on how they stack up against their peers, researchers suggest in a new paper.

Investigators used an algorithm based on details captured by the National Cardiovascular Data Registry (NCDR) to identify four unique phenotypes—operators who have similar procedural complexity, volumes, patient presentations, and other characteristics—that in turn were associated with patient outcomes.

Lead author Jacob A. Doll, MD (VA Puget Sound Health Care System, Seattle, WA), told TCTMD that the study, published online this week in Circulation: Cardiovascular Interventions, was a “first attempt to use an objective, machine-learning statistical technique to try to do the thing we all do a little intuitively, which is to put people in groups.”

It’s widely accepted that “different physicians care for different patient populations and have different practice patterns, but finding ways to do that objectively has been really hard,” he explained.

“When I talk with my colleagues, I might say, ‘That interventional cardiologist is really skilled,’ and, ‘This other colleague is really conservative in their management,’ and, ‘This other one’s a little out of date with what he does in the cath lab.’ But can we really measure that? And beyond that, how do I know which of those groups I’m in? Because in general we don’t do a great job as physicians, or as humans honestly, at self-assessing,” said Doll.

Profiling is meant to complement existing quality measures, the researchers say. The hope is that clinicians can use these phenotypes as a lens to better understand their own characteristics, with the goal of improving patient care.

Doll stressed that they didn’t set out with particular groups in mind; rather, these were uncovered by their algorithm, with no judgement about which practices were more desirable than others. “One of the strengths of this research is that the algorithm doesn’t care: [it] doesn’t have any skin in the game and doesn’t have any preconceived notions about who interventional cardiologists are,” he noted.

NCDR CathPCI Registry

Doll and colleagues analyzed data from the NCDR CathPCI Registry on 7,706 operators who performed at least 25 procedures annually between January 2014 and March 2018. Together, these operators did a total of 2,937,419 PCIs.

The investigators then used an agglomerative hierarchical clustering algorithm to explore whether they could tease out any distinct patterns. Their calculations revealed four groups:

• Cluster 1: “Mainstream” operators (43.4%), who had case mixes and practice patterns similar to the national median; were most likely to use radial access; and were second-most likely among the groups to treat patients with high-risk anatomy (eg, left main stenosis)
• Cluster 2: “Low-Risk Elective” operators (25.9%), who treated patients with lower clinical acuity; were less likely to do PCI off-hours; treated the least-complex lesions; less often used intracoronary imaging, atherectomy, and mechanical circulatory support (MCS); and were least likely to use radial access
• Cluster 3: “Acute Access” operators (19.6%), who had the lowest case volume; were most likely to practice at rural hospitals, perform PCI at multiple centers, and treat uninsured patients; had the highest proportion of on-call cases, STEMI, and cardiogenic shock; rarely did elective PCI; were second-most likely to use MCS; and (like cluster 2) had low use of radial access, intravascular imaging, and atherectomy
• Cluster 4: “High Complexity” operators (11.1%), who had the highest case volume; tended to treat patients with high clinical and anatomic complexity; had high use of radial access and low use of bivalirudin; were most likely to use atherectomy, intracoronary diagnostics, and MCS; and were most apt to practice at a university hospital

As for what to call the clusters, Doll advised PCI operators to focus less on the terminology and more on the characteristics within each one. “Honestly, we debated a lot about whether we should even put [shorthand names like “Mainstream”] in there because it would intentionally bias people about how they think about the clusters,” he noted.

The researchers also looked at how these categories were associated with outcomes, though Doll pointed out the differences they found were “not huge,” in part because PCI mortality is so low.

Mean operator-level in-hospital mortality was 2.0%, ranging from 1.3% in group 2 (Low-Risk Elective) to 2.8% in group 3 (Acute Access).

After adjustment based on the CathPCI mortality risk score, rates were similar in groups 1 and 2 (Mainstream and Low-Risk Elective), highest for group 3 (Acute Access), and lowest for group 4 (High Complexity). “The cluster that most closely corresponds to proposed standards for optimal percutaneous coronary intervention practice—high annual volume, frequent use of contemporary procedural techniques, high clinical, and anatomic complexity—had the lowest adjusted in-hospital mortality rates,” the researchers point out.

Group 2 also had lower procedural success compared with group 1, while group 4 had a lower rate of inappropriate PCI according to Appropriate Use Criteria.

What’s needed next is more and better evidence that this strategy “meaningfully measures practice,” said Doll, “or alternatively that feeding these back to people helps them to be better.” He predicted that, as larger and more-complex data sets become available, the strategy of profiling physicians’ practice patterns will become increasingly common.

This study, said Doll, “is a proof of concept.” Even now, though, their results can inspire introspection, he added.

“I really do hope that interventional cardiologists will look at the paper and ask themselves honestly: which cluster do I fit into, and then subsequently: is that the cluster I want to be in?” said Doll. Operators who realize they’re not where they want to be, he suggested, can consider what resources or training they need to make that shift.