Abstract
Missed and delayed cancer diagnoses are common, harmful, and often preventable. The Centers for Medicare and Medicaid Services have emphasized digital quality measures (dQMs) that leverage longitudinal electronic health record data across multiple settings of care to address quality. Despite measurement gaps in diagnostic quality, no dQMs are currently used to evaluate the quality of cancer diagnosis. We previously developed and validated a dQM of emergency presentation (EP) of lung cancer in two US health systems. EPs were shown to be associated with preceding missed opportunities for diagnosis, and independently associated with higher mortality. In this study, we applied this dQM to Epic Cosmos, a deidentified database covering 184 million individuals in the US.
Introduction: The Measurement Gap in Diagnosis
In contemporary medical quality improvement, clinical performance is frequently measured through treatment-oriented metrics, such as the timely administration of specific medications. However, the diagnostic process remains significantly under-measured. Missed or delayed cancer diagnoses represent a critical failure point in healthcare, yet automated systems to flag these failures are largely absent from standard practice.
A promising proxy for identifying "missed opportunities" is the Emergency Presentation (EP). An EP occurs when a patient is diagnosed with cancer shortly after an emergency department visit, suggesting that earlier, more subtle symptoms may have been overlooked during prior primary care encounters. This research explores the transition from manual chart reviews to automated, national-scale diagnostic surveillance using Digital Quality Measures (dQMs).
Methods: Scaling Diagnostic Surveillance
The study implemented a validated EP detection algorithm within the Epic Cosmos environment, the largest integrated electronic health record (EHR) repository in the United States, covering 184 million deidentified individuals.
The EP Logic and E-Triggers
An EP was defined as any emergency room visit followed by a new lung cancer diagnosis within 30 days. This approach builds upon the use of "e-triggers"—electronic flags that identify clinical scenarios with a high probability of diagnostic error, such as a return to the emergency department or inpatient admission following a primary care visit.
Social and Clinical Context
The analysis focused on a cohort of 58,274 patients newly diagnosed with lung cancer in 2019. To understand the drivers of diagnostic inequity, the researchers cross-referenced EP rates with the CDC’s Social Vulnerability Index (SVI) and patient-reported social determinants of health (SDOH), including income and transportation access.
Key Results: Uncovering Systemic Inequities
The application of the dQM revealed that diagnostic outcomes are heavily influenced by socioeconomic and racial factors.
- National Benchmarks: The overall EP rate for lung cancer was 19.5%, consistent with smaller-scale benchmarks from the Veterans Affairs (19%) and Geisinger Health (18%).
- Racial Inequity: Black patients experienced a significantly higher EP rate than White patients (24% vs. 19%, $p < 0.001$), indicating a systemic gap in early detection for minority populations.
- The Impact of Social Vulnerability: A clear socioeconomic gradient was observed; patients in the highest SVI category had a 22% EP rate, compared to 18% in the lowest.
- Transportation as a Barrier: Perhaps the most significant finding was the role of physical access. Patients reporting transportation difficulties had an EP rate of 43%, nearly double the rate of those without such hurdles (28%).
- Age Paradox: Patients younger than 69 had a higher EP rate (23%) than those aged 78 and over (17%), suggesting potential symptom dismissal in younger populations or differences in primary care engagement.
Discussion and Future Directions
This study demonstrates the feasibility of using dQMs for large-scale, near-real-time diagnostic surveillance. The findings suggest that "diagnostic error" is often a failure of social and systemic infrastructure rather than individual clinical reasoning alone.
Future Work
- Broader Application: Extending dQMs to other high-stakes malignancies, such as colorectal or pancreatic cancer.
- Interventional Strategies: Investigating whether social interventions, such as ride-share vouchers or mobile health clinics, can reduce EP rates and lead to earlier, more treatable diagnoses.
- Clinical Integration: Moving these measures from retrospective research into active dashboards for hospital administrators to monitor their own performance in real-time.
By leveraging big data and digital quality measures, healthcare systems can move toward a future where diagnostic excellence and equity are measurable and achievable goals.
Summary updated