Measuring Performance of the Diagnostic Process

“During the diagnostic process, it is not unusual, or incorrect, for working diagnostic labels to change as new information is acquired and as the patient’s condition evolves both naturally and in response to interventions. The language used to communicate risk of disease and uncertainty about diagnosis is not uniform and may be overly ambiguous (eg, “cannot rule out,” “consider the possibility”). Thus, attempts to standardize and measure diagnostic processes should avoid unrealistic expectations or overzealous judgments to be both accurate and fair in judgment (eg, driving performance not feasible under the conditions at the time, or expecting actions predicated on facts not available at the time of care).

Standards for diagnostic performance will change as the current understanding of disease evolves, new diagnostic technologies are introduced, and new therapies become available, and quality measures will need to be refined and updated to remain relevant, especially when diagnostic standards are in rapid evolution (eg, the rapidly changing landscape of genetic testing for cancer). Absolute requirements in measures that may lead to excess testing to optimize accuracy without consideration of potential harm from overdiagnosis should be avoided. The overuse of CT angiography for suspected pulmonary embolism is an example of a common practice that illustrates excessive testing that fails to improve outcomes, with recently reported yield rates of only 1% to 3%. [..]

Current efforts to measure quality performance tend to rely on readily available data sources, such as claims-based data, rather than rich clinical data present in electronic health records, found in narrative reports, or acquired through patient surveys; these sources of information may be more relevant to diagnosis. Reliance on a single data source may not provide the information necessary to determine the accuracy and quality of a diagnosis. Claims data were not designed to provide information on presenting symptoms, the diagnostic process, or the overall diagnostic trajectory. Although claims data can document that a diagnostic study was performed, those data do not capture the clinical logic that led to the order or the results of the examination. [..]

Ideally, a new measurement model for diagnosis will be built in partnership with clinicians and patients. Patients would drive measurement toward meaningful diagnostic outcomes of interest, including diagnostic errors and delays. Although it remains difficult to accurately measure diagnostic errors, a new measurement model built with patients will drive more open discussion about potential solutions, including patient-reporting systems. A well-designed diagnostic measurement system could provide value-added, actionable information to clinicians and patients, but be designed to detect and avoid potential unintended consequences to patients. To avoid these potential harms, this new diagnostic approach should be built with the capacity for real-time feedback to prospectively monitor for unintended harms.”

Full article, H Burstin and K Cosby. JAMA, 2022.6.23