As oncology becomes more data-driven, the burden of navigating complex digital workflows is growing. While “click fatigue” has long been cited as a byproduct of evidence-based pathways and clinical decision support tools, we’re now entering a new phase, one where pop-up fatigue may emerge as a more disruptive force.
Unlike passive documentation burdens, this wave is shaped by active, real-time alerts injected into the physician workflow, increasingly from AI platforms analyzing data in the background.
From Pathways to Pop-Ups
Companies like Tempus, Viz.ai, and OncoLens are embedding their AI engines into the EHR, surfacing alerts to identify clinical trial eligibility, treatment opportunities, or adverse events. These tools are powerful and have already demonstrated a positive impact for many cancer patients, often parsing unstructured data and labs to identify patterns a human might miss. As we say often in healthcare, however, these dynamics don’t operate in isolation.
Consider the recent partnership between Tempus and Stemline Therapeutics: Tempus’s AI scans patient records to flag potential candidates for Orserdu and generating real-time alerts that appear within a physician’s workflow¹. In a silo, that is incredibly valuable information. However, if these pop-ups layered onto simultaneous alerts from the EHR vendor, drug safety flags from integrated compendia, and institution-specific pathway nudges, it is easy to see how this can result in an overwhelming number of pop-ups.
In one JAMA Oncology study, researchers tested different EHR alert designs for prompting serious illness conversations in gynecologic oncology. They found that required alerts did improve response rates, but also increased workflow interruption and clinician frustration². In other words, even good alerts still contributed to fatigue when delivered at the wrong moment or without sufficient context.
Alert Fatigue, Amplified
The underlying risk isn’t just inconvenience, it’s cognitive overload. When clinicians see too many alerts, even important ones may get dismissed. A study on EHR-based trigger systems for cancer diagnosis delays highlighted this: although the triggers improved detection, the research team noted that without precise tuning and prioritization, alerts quickly became “background noise”³.
With multiple platforms potentially injecting signals into the same workflow, the probability of alert redundancy rises dramatically. If two different AI systems flag the same patient for two similar reasons, without coordination, providers may start tuning them all out.
The scale of the problem is already visible. A recent analysis of oncology specialists’ inbox use found a 19% increase in message volume and a 16% increase in EHR time from 2019 to 2022⁴ highlighting that clinician digital burden continues to climb.
Where Do We Go From Here?
The challenge isn’t just to limit alerts, but to make them smarter.
- Relevance Over Volume: Every alert should have a clearly defined purpose and a high likelihood of driving action.
- Context-Aware Timing: Alerts should be delivered at the right moment—not during documentation or order entry, but when clinicians are making related decisions.
- Precision Targeting: Instead of sending the same alert to every oncologist, systems should be able to route insights to the most relevant specialist based on role, disease state, or prior interaction with the patient.
Ultimately, the future of AI in oncology depends not just on what it can find, but on how, when, and why it chooses to speak up.
References
- Tempus and Stemline Therapeutics collaboration announcement. Tempus press release, Feb 2025.
- Patel MI, Sundaram V, Desai M, et al. Long-term Effect of Machine Learning-Triggered Behavioral Nudges on Serious Illness Conversations in Oncology. JAMA Oncology. 2022. Link
- Singh H, El-Kareh R, Thomas EJ, et al. Electronic health record-based triggers to detect potential delays in cancer diagnosis. BMJ Qual Saf. 2013. Link
- Tai-Seale M, Dillon EC, Yang Y, et al. National Trends in Oncology Specialists’ Electronic Health Record Inbox Work, 2019–2022. JAMA Network Open. 2025. Link