Bridging the Diagnostic Gap in European Hospitals
At the AI2MED webinar “The Diagnostic Gap in Serious Comorbidities in German Hospitals”, German medical data expert Roland Trefftz (Klinikon GmbH) delivered an important insight: nearly 50% of severe acute kidney injury (AKI) cases remain undetected in hospitals.
Why? Because even when data is technically available, it’s not being meaningfully interpreted. Trefftz’s work sits at the intersection of medical data science and clinical practice, making him uniquely positioned to discuss both the systemic diagnostic blind spots and the real-world AI interventions designed to close them.
In this interview, Dr. Trefftz explains the roots of this diagnostic gap—and how AI can realistically support clinicians today.
Why Half of Severe AKI Cases Are Missed
“Trying to perform this ‘digitally’—in the original sense of the word: using only one’s own fingers in the hectic daily routine at the bedside is hardly feasible in a reliable manner.”
Despite being clearly visible in lab data, nearly 50% of severe acute kidney injury (AKI) cases are not identified during hospital stays. Detecting them requires doctors to manually analyze complex tables of serum creatinine values—often dozens of rows and columns—while interpreting time intervals, changes, and medical guidelines in the middle of a busy ward shift. Digital tools could help, but under EU regulations, any algorithm used for diagnosis must be certified as a medical device, which is costly and complex. As a result, many providers avoid developing such solutions.
How the AI Early Warning System Works
“The system analyzes the incoming stream of laboratory data for each patient, evaluating up to 135 different parameters or dimensions.”
The medlytics AI Early Warning System runs alongside hospital lab systems and monitors up to 135 lab parameters per patient in real time. When it detects abnormal patterns linked to early disease stages in known patient groups, it automatically generates a PDF alert and sends it to the patient’s digital records. This gives doctors a chance to intervene early—for example, by avoiding kidney-harming drugs or giving fluids in AKI cases.
AI Helps Staff Focus Where It Matters
“With the AI early warning system, the group of patients needing monitoring or screening can be reliably narrowed down, allowing staff resources to be much more effectively focused on those patients at risk.”
In diagnoses like SIRS (Systemic Inflammatory Response Syndrome), doctors typically need to monitor multiple vital signs for every patient—a task that’s often unmanageable due to limited staff. The AI early warning system acts as a filter, flagging only the most at-risk patients. Instead of tracking everyone, staff can focus on the 9% who receive an alert—of whom nearly half actually develop SIRS. This makes early detection both more practical and more precise.
GDPR Means Going Local
“This local installation and individual integration are relatively time-consuming and require extensive testing.”
Strict GDPR rules in the EU require that sensitive patient data stays within the hospital’s control. That’s why the medlytics AI system is installed and maintained on local servers—not in the cloud. While this approach is much safer from a data privacy standpoint, it also brings technical challenges. Hospitals have very different IT systems, and setting up each installation takes time and resources.
Two Paths for AI in Hospitals
Looking ahead, Dr. Trefftz highlights two major areas for AI in healthcare:
Medical AI (physician decision support):
These tools help doctors make clinical decisions, but require complex and costly medical device certification. Because of that, this type of AI is often far more expensive to implement than simpler alternatives.
Process AI (in organizations, including hospitals):
While digital workflows have existed in industries like banking for years, what’s new is that large language models (LLMs) like GPT-4, Claude, or DeepSeek can now be added to these processes. That means previously rigid systems can now analyze information, understand context, and act intelligently.
“These agentic workflows can already be introduced in a hospital for a three-or four-digit Euro amount. Setting up agentic workflows takes days, not months—unlike traditional software products.”
Final Thought
Even when the data exists, diagnosing isn’t guaranteed. AI can help fill this gap—not by replacing clinicians, but by giving them back the time and insight they need to act early
