In German hospitals, nearly every second case of severe acute kidney injury (AKI) remains completely undiagnosed. A similar pattern applies to other serious comorbidities—SIRS, delirium, and malnutrition—leading to poorer patient outcomes, longer hospital stays, and higher healthcare system costs.
At the webinar “The Diagnostic Gap in Serious Comorbidities in German Hospitals,” held on March 26, 2025, as part of the EU-funded AI2MED project, German expert Roland Trefftz (Klinikon GmbH) demonstrated how artificial intelligence is already helping to reduce these diagnostic gaps. Below, we present the key insights from this informative event.
1. Serious conditions often go unrecognized – because they are not coded
The central topic of Trefftz’s presentation was the diagnostic gap—specifically, the discrepancy between actual laboratory findings and what is recorded (coded) in hospital systems. According to a 2019 study conducted at Charité Hospital, as many as 49% of stage II and III acute kidney injury (AKI) cases remain uncoded. In some hospital specialties, this figure rises to 60%.
And without a code—there is no therapy, no follow-up, and no medical history.
“If there is no coding, then no therapy or diagnosis can be derived from the records,” Trefftz emphasized.
“If there is no coding, then no therapy or diagnosis can be derived from the records,” Trefftz emphasized.
2. AI systems can predict disease progression up to three days in advance
To address this issue, Klinikon developed an AI-based early warning system that has already been in use in German hospitals for over a year. The system continuously monitors laboratory data—such as changes in serum creatinine—and applies machine learning to calculate the probability of AKI development up to 72 hours in advance.
In the system’s visual interface, the rising risk is clearly displayed, along with precise probabilities for each stage:
“For stage III acute kidney injury, the system predicted a probability of 65.9% within the next 72 hours.”
3. Doctors don’t have time for manual calculations – AI brings automation
Trefftz explained very concretely why diagnostic oversights occur: physicians are overburdened and lack the capacity to manually track hundreds of parameters.
“I couldn’t believe that doctors are still calculating with calculators.”
This is why the AI system is designed to act like “a chart that tells you to take an umbrella”—if there’s an 80% chance of rain, you don’t wait to get wet.
4. Non-infectious SIRS, hypoactive delirium, and malnutrition – “invisible” threats
In addition to acute kidney injury, the early warning system monitors three further diagnoses that are common but systematically underrepresented in clinical practice:
- Non-infectious SIRS (Systemic Inflammatory Response Syndrome) often goes undetected, despite clinical expectations of prevalence similar to the infectious form. The reason lies in non-specific symptoms—such as changes in body temperature, respiratory rate, or leukocyte count—whose recognition requires continuous monitoring of multiple parameters.
- Hypoactive delirium, unlike the hyperactive form, is rarely recognized because it does not involve agitation. Patients appear calm, withdrawn, often drowsy—but behind this may lie a serious disturbance of consciousness with potential consequences for treatment outcomes.
- Malnutrition affects more than 20% of hospitalized patients, yet real hospital data show that it is diagnostically coded in only 1% of cases. This condition significantly impacts immune response, recovery, and length of hospitalization, but remains rarely monitored on a regular basis.
These are all conditions where automated detection can make the difference between life and death.
5. Technological framework: simple, transparent, and compliant with EU regulations
From a technical standpoint, the system uses random forest models, rather than “black box” neural networks whose decisions are difficult to interpret—because it is essential that physicians understand why a certain assessment was made. Data are scaled, relevant features are selected, and the system operates locally on hospital servers.
The analysis requires only anonymized data—no patient names, just ICD-10 codes and laboratory data. Data sharing complies with strictly defined EU regulations.
Strong interest from participants
Participants from Croatia showed strong interest, particularly in the technical aspects. Some of the questions included:
- Can this analysis also be implemented in primary healthcare?
- How are features selected for inclusion in the model?
- How is data correlation handled?
- How are data scaled to enable outcome prediction?
These questions demonstrated that the audience is already thinking about implementation within their own institutions.
Call for collaboration: “We can run this analysis with you within a week”
In closing, Trefftz extended an open invitation to hospitals in Croatia:
“We can run this analysis with you within a week.”
All that is required is anonymized hospital data, and the result is a clear insight into diagnostic discrepancies—and an opportunity to reduce them.
For physicians and hospital managers seeking to improve quality of care and optimize processes, this represents a concrete opportunity to take a first step—quickly and without technical barriers.
Time to take action
The diagnostic gap is not merely an administrative issue—it is a matter of patient safety and healthcare system efficiency. AI-based early warning systems are not a distant future—they are already here and available.
If you have access to hospital data, don’t wait for the rain—take the umbrella in time.
About the AI2MED project
The webinar was organized by Smion as part of the AI2MED project, which brings together 11 partners with the goal of modernizing education and integrating artificial intelligence into medicine. The project combines innovation, practice, and education to help transform the healthcare industry—making it more precise, efficient, and accessible to all.
If you are interested in similar events, subscribe to our Newsletter, where we regularly share updates about upcoming opportunities and initiatives.
