Modern hospitals have access to more data than ever before. Laboratory results, vital signs, electronic medical records, medication orders and information generated by diagnostic devices can become available almost immediately.
The main challenge is no longer collecting information. It is determining which information requires action, how quickly someone should respond and who should receive the notification.
Healthcare professionals work in an environment filled with messages and interruptions. Patient monitors, medical devices and hospital information systems generate alerts. New laboratory results, medication warnings, orders and documentation updates appear throughout the day. Telephones, messaging platforms and nurse call systems compete for staff attention.
Each channel was introduced to improve communication. When the channels operate independently, however, physicians, nurses and pharmacists must assess the significance of dozens or even hundreds of signals.
Not every piece of medical information is equally urgent. A result outside the reference range does not always mean that a patient is in immediate danger. At the same time, a relatively small change may be highly significant when combined with earlier results, the patient’s diagnosis, current medication or a planned procedure.
The future of hospital information systems therefore cannot be based solely on generating more messages. It must focus on selecting, interpreting, prioritising and routing information.
When alerts stop helping
The phenomenon known as alert fatigue has long been recognised as a risk to patient safety and staff well-being. A 2025 scoping review describes it as a gradual reduction in responsiveness caused by repeated exposure to frequent or non-actionable alerts. Potential consequences include delayed responses, communication failures, increased stress and the risk of overlooking an important warning.
This does not mean that alert systems are unnecessary. Problems arise when systems cannot distinguish routine information from a critical event or fail to take account of the individual patient’s context.
The significance of this issue can also be seen in The Joint Commission requirements effective for United States hospitals from January 2026. Hospitals are expected to identify their most important alerts, assess which notifications unnecessarily contribute to information noise and establish criteria for recognising and responding to patient deterioration.
This represents an important change in perspective. The objective is not to make every system produce as many warnings as possible. The objective is to create an information environment in which a message requiring action is noticed, delivered to the appropriate person and handled through a defined process.
From an isolated result to the full patient context
Traditional warning systems often rely on predefined thresholds. When a parameter exceeds a selected value, the system triggers an alert. This approach is relatively simple and transparent, but it is not always sufficient.
The significance of a result may depend on:
- previous values and the direction of change;
- diagnosed conditions;
- the patient’s age and current clinical status;
- medication;
- other laboratory findings;
- completed or planned procedures;
- the unit and stage of the care process;
- the role of the person receiving the message.
Only by combining these elements can a system distinguish between a routine change, an event requiring observation and a situation that should be urgently communicated to clinical staff.
International healthcare is moving in this direction. According to a WHO/Europe report published in April 2026, 74% of European Union Member States reported using AI-assisted diagnostic tools. WHO also emphasises workforce preparedness, transparency and the continued accountability of healthcare professionals for AI-assisted decisions.
Can intelligent warning systems improve patient outcomes?
Early warning systems for patient deterioration are among the most widely studied clinical applications of artificial intelligence. Algorithms analyse available data and attempt to identify concerning patterns earlier than would be possible from an isolated measurement.
A 2025 meta-analysis included five studies in which AI models underwent prospective clinical validation. The authors observed lower in-hospital and 30-day mortality and a shorter overall hospital stay. They also stressed that the evidence base remains limited and that effectiveness partly depends on whether healthcare professionals act on the warnings they receive.
The TREWS sepsis early warning system provides an instructive example. A multi-site study monitored 590,736 patients across five hospitals. The analysis focused on 6,877 patients with sepsis identified by the system before antibiotic treatment began. Provider confirmation of an alert within three hours was associated with lower in-hospital mortality, reduced organ failure and a shorter hospital stay.
This was an observational study and therefore cannot attribute the entire effect to the algorithm itself. It does demonstrate that technology creates value only when an appropriate signal is connected to an effective clinical process.
The most difficult challenge begins after the risk is detected
Identifying a concerning pattern is not enough. A system must answer several practical questions:
Who should receive the information? How urgent is the response? How should the reason for the alert be presented? What happens if the recipient does not acknowledge it?
An attending physician may require different information from a nurse, laboratory professional, clinical pharmacist or rapid-response team. The significance of the same message may also change during regular working hours, a night shift or the transfer of a patient between hospital units.
Research on transporting AI models between healthcare organisations shows that even an effective algorithm should not be deployed automatically in every hospital. Local validation, threshold and workflow adaptation, data-quality management, user training and continuous monitoring are required. Model accuracy alone does not guarantee staff trust or improved outcomes.
MedAlert: from data to action
Infotower Business Solutions developed the MedAlert project in response to this challenge. Its official title is:
“Development of intelligent algorithms controlling the selection and routing of critical messages for medical processes in inpatient care.”
The objective of the research and development work was to create an intelligent, comprehensive model capable of generating real-time notifications about a potential health threat to a hospital patient, based on the analysis of key parameters obtained from hospital information systems.
MedAlert was not designed as another independent source of alerts. Its central idea is to use algorithms to:
- detect events that may require attention;
- analyse information in the context of an individual patient;
- determine the significance and urgency of an event;
- select the appropriate recipient;
- present information in a way that supports professional decision-making;
- record acknowledgement, response and subsequent handling of the message.
In practice, this means moving beyond a simple notification that a result has exceeded a threshold. The more important question is:
Does this particular situation, involving this patient at this stage of treatment, require action—and, if so, by whom and how urgently?
A real hospital workflow
One of the scenarios used during the practical validation of MedAlert at the Silesian Centre for Heart Diseases in Zabrze concerned the transfer of patients from a hospital ward to the haemodynamics laboratory.
Before a procedure, the laboratory should be informed of factors such as earlier positive bacteriological results. This information may affect the order of procedures, room preparation and the need to reserve additional time for disinfection.
In a traditional workflow, the information is often communicated by telephone. During an urgent transfer, however, it may arrive late or require several calls.
According to information published by Infotower Business Solutions, MedAlert analysed data available in hospital systems during the pilot and generated notifications related to patient transfers, including urgent cases. Communication through a mobile application made it possible to record the delivery of information and the subsequent actions of users.
The example demonstrates that intelligent alerting does not have to begin with a spectacular diagnostic scenario. Significant value can come from improving the flow of information that is already available but does not always reach the right person at the right time.
What can a hospital gain?
From a hospital management perspective, the importance of such solutions extends beyond the AI model itself.
Potential organisational benefits include:
- faster communication of significant information;
- fewer manual telephone calls;
- better use of data already held in hospital systems;
- a lower risk of critical information being overlooked;
- structured rules for escalating messages;
- records of when a notification was delivered, acknowledged and handled;
- the ability to analyse how effectively the organisation responds to events.
For a physician, the value does not come from receiving another alert. It comes from receiving concise, understandable and justified information. Nurses need a clear indication of whether an event requires immediate action. Similar mechanisms can help clinical pharmacists prioritise information relating to medication, laboratory results and pharmacotherapy safety.
Hospital executives can approach communication as a measurable process: how many events were detected, which required intervention, how quickly action was taken and where delays most frequently occurred.
Technology cannot replace human accountability
AI systems should not make autonomous treatment decisions. Their role is to support professionals by organising data, identifying potential relationships and drawing attention to situations that require assessment.
Every model can generate false-positive results or fail to detect a significant event. Its performance may change with data quality, documentation practices, patient populations and hospital workflows.
Responsible implementation should therefore include:
- local validation before production use;
- an initial silent mode that does not influence staff decisions;
- clinical oversight of rules and models;
- measurement of alert volume, accuracy and usefulness;
- records of user decisions and responses;
- monitoring for renewed alert overload;
- escalation and downtime procedures;
- data-protection and access-control rules;
- regular reassessment of algorithm performance.
The safest operating model is one in which artificial intelligence identifies information and provides its context, while a qualified professional evaluates the situation and makes the decision.
The most valuable hospital resource is staff attention
For many years, healthcare digitalisation focused primarily on collecting data. The next stage must focus on using those data more effectively.
Physicians, nurses and pharmacists do not need more information. They need support in finding the most important signal among hundreds of less significant messages.
MedAlert was developed around the belief that technology should protect the attention of healthcare professionals. It should help identify an event, determine its significance and deliver it to the person capable of taking appropriate action.
Patient safety sometimes depends not on whether the data exist, but on whether the right information reaches the right person at the right time.
About the project
- Project number: FENG.01.01-IP.02-1046/23
- Title: Development of intelligent algorithms controlling the selection and routing of critical messages for medical processes in inpatient care
- Beneficiary: Infotower Business Solutions sp. z o.o.
- Total project value: PLN 15,368,022.15
- European Funds contribution: PLN 11,727,902.80
- Status: completed
- Programme: European Funds for a Modern Economy
The project was co-funded by the European Union.
Sources and further reading
- Michels E.A.M. et al., a scoping review of alert fatigue in healthcare, BMC Nursing, 2025: PubMed
- The Joint Commission, “Joint Commission Requirements for Hospital Programs”, requirements effective January 2026: The Joint Commission document
- WHO/Europe, “New WHO/Europe report provides first-ever snapshot of AI in health care across European Union Member States”, 20 April 2026: WHO/Europe
- Yuan S. et al., “AI-Powered early warning systems for clinical deterioration significantly improve patient outcomes: a meta-analysis”, 2025: PubMed
- Adams R. et al., “Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis”, Nature Medicine, 2022: Nature Medicine
- Valan B. et al., “Evaluating Sepsis Watch generalizability through multisite external validation of a sepsis machine learning model”, npj Digital Medicine, 2025: npj Digital Medicine
- Infotower Business Solutions, official information about the MedAlert project: Infotower Business Solutions
- Infotower Business Solutions, company information about the pilot at the Silesian Centre for Heart Diseases, LinkedIn company profile: LinkedIn — Infotower Business Solutions
This article concerns technology and healthcare-process organisation. It does not constitute medical advice or a recommendation for the management of an individual patient.
