Radiologist attrition rose across every subspecialty from 2014 to 2022, according to research from the Harvey L. Neiman Health Policy Institute published in the American Journal of Roentgenology. The analysis covered 159,490 radiologist-year observations and 29,770 subspecialist radiologists who submitted Medicare claims, giving the workforce warning real weight.
The headline finding was broad rather than isolated. Year-by-year attrition among subspecialist radiologists increased from 1.4% in 2014 to 2.7% in 2022. Every subspecialty saw an increase over time, though the size varied. The pattern raises a workforce concern because radiology demand has not slowed while the physician pool has become more difficult to sustain.
The Attrition Was Uneven
The study found an average unadjusted attrition rate of 2.2% across subspecialist radiologists during the study period. But the range was wide: vascular and interventional radiology was lowest at 1.0%, while cardiothoracic imaging reached 4.3%. Musculoskeletal imaging had the smallest increase over time, and cardiothoracic imaging had the largest reported percentage-point increase.
After adjustment for radiologist characteristics and practice setting, breast imaging and cardiothoracic imaging had higher odds of attrition compared with abdominal imaging, while vascular and interventional radiology had lower adjusted odds. That means workforce planning cannot stop at training more radiologists in general. The weak points are not identical across subspecialties.
Seniority Explains Part Of The Pattern
The Neiman researchers cautioned that subspecialty differences are not all caused by burnout or dissatisfaction. Seniority varied widely among subspecialties, and prior work has shown years in practice to be a major driver of attrition. Cardiothoracic imaging had relatively high seniority, while vascular and interventional radiology had lower seniority in the analysis.
That context is important because a workforce study should not become a single-cause story. Workload, call burden, reimbursement pressure, private-practice consolidation, academic demands, administrative tasks and the aftereffects of COVID-era strain can all shape career decisions. The study documents a trend. Explaining every reason for it requires more evidence.
Patient Care Feels The Loss
Radiologist shortages do not stay inside department staffing charts. They can lengthen report turnaround times, reduce local subspecialty access and push smaller hospitals toward more teleradiology coverage. Remote reads can help, but they do not replace every benefit of an on-site specialist who can speak directly with clinicians and participate in local care decisions.
The risk is especially acute in areas where expertise takes years to build. A department cannot replace a departing cardiothoracic, breast or pediatric radiologist as easily as filling a generic vacancy. The loss is not only headcount. It is institutional knowledge, consultation capacity and confidence on difficult cases.
AI Cannot Be The Whole Answer
Artificial intelligence may help with triage, measurement, prioritization and repetitive tasks, but it has not solved the workforce problem. If AI is deployed mainly to increase volume expectations, it can make burnout worse rather than better. Technology has to reduce cognitive load, not simply monitor productivity more tightly.
The practical response has to include retention, scheduling, call design, administrative relief, career flexibility and subspecialty-specific recruitment. Administrators should treat attrition as a quality metric, not only an HR metric. A department may appear staffed on paper while losing the precise expertise needed for complex studies.
Diagnostic medicine depends on human attention at scale. Hospitals cannot keep asking radiologists to read more, faster and under more administrative pressure while expecting attrition to remain stable. Imaging delays can delay cancer care, stroke workups, surgical planning and emergency decisions. The Neiman data describes a capacity warning for the medical system's diagnostic engine.