Horizontal AI platform or AI in specialist systems: what hospitals need

Which patterns are there?

Pattern	What it is	Examples (according to the vendor)
AI in the HIS	AI features in the screens of the hospital information system (HIS, German: KIS): documentation from the conversation, doctor’s letter, coding	ORBIS, CGM, NEXUS / KI, iMedOne
Specialist platform with AI	Software for one process, retrofitted with AI	Recare (discharge management), ID Berlin (coding), roXtra (quality management)
AI-native specialist vendor	Built as AI from the start, for one clinical task, in some cases a medical device	medicalvalues (laboratory diagnostics)
Horizontal AI platform	AI for many tasks in all departments, with access to documents and systems	basebox, ChatGPT Enterprise, Langdock

The patterns solve different problems. AI in the HIS and specialist platforms do not compete with a platform in the sense of “doing the same thing”; they cover the depth of one process, the platform covers the breadth of the organisation. Details: AI in the HIS, specialist platforms with AI, AI in laboratory diagnostics.

Not every “AI” in specialist systems is generative AI. Duty rostering and staff planning as with POLYPOINT or ATOSS, transport dispatch as with DYNAMED or the planning of operating theatres and beds as with Timerbee rely mainly on optimisation, forecasts and rules. That is a different task from drafting texts or answering questions on documents; the two do not exclude each other.

What can AI in a specialist system do that a platform cannot?

• It knows the process and its data. A doctor’s letter in the HIS is created from the diagnoses and medication of the case; a coding suggestion in the coding system knows the rule set and the checking rules.
• It brings what no software has on its own. Recare, for example, connects hospitals with a network of nursing and rehabilitation facilities.
• It may help with medical decisions if it is approved as a medical device for that purpose. medicalvalues describes its Diagnostic Intelligence as a class IIa medical device under the MDR.

What can a platform do that a specialist system cannot?

• Reach all departments. Purchasing, human resources, facilities, pharmacy, secretariats and management rarely have a specialist system with AI.
• Answer questions across several sources. Guideline, SOP, contract and e-mail at the same time, each with its source.
• Take on new tasks without a new purchase. A new task becomes an app that a group uses, instead of another tool with its own contract.
• One control for everything. One permission management, one log, one data protection impact assessment for the platform, extended for each new purpose (Art. 35 GDPR), one data processing agreement (Art. 28 GDPR).

How much AI does a hospital need in which pattern?

The answer depends on which specialist systems already bring AI. An example of how tasks can be split:

Task	Pattern
Doctor’s letter from the case data, bedside conversation documentation	AI in the HIS
Finding follow-up care, coding, document control in QM	Specialist platform
Evaluating laboratory results, suspected diagnoses	Specialist vendor, medical device
Querying guidelines and SOPs, letters, translations, e-mails	Platform
Purchasing, human resources, facilities, administration	Platform
Questions across several systems	Platform with connectors

Two consequences follow:

1. Many point solutions reduce the need for a platform, but only in their processes. If you already have AI in the HIS and in three specialist platforms, you do not need a platform for those processes.
2. Many point solutions increase the effort. Each vendor brings its own place of data processing, its own contracts, its own permissions and its own interface.

What does each point solution cost beyond its price?

In Germany, every new AI tool that processes patient data brings the same obligations, however small it is:

Obligation	Per point solution	With one platform
Data processing agreement (Art. 28 GDPR)	one per vendor	one
Confidentiality commitment (§ 203(4) StGB, German Criminal Code)	per vendor	once
Data protection impact assessment (Art. 35 GDPR)	per processing operation, usually per tool	one for the platform, extended for each new purpose
Record of processing activities (Art. 30 GDPR)	one entry per tool	entries per purpose, one system
Proof under § 393 SGB V for cloud services	check per vendor	check once
Sign-in and permissions	set up per tool	from your existing identity management
AI literacy (Art. 4 AI Act)	training per tool	training on one interface

On top of this come the security review, involving the works council or staff council, and maintaining each tool in operation. With three tools this is manageable; with twenty it becomes a permanent task for IT and data protection.

How do you proceed?

1. Take stock: Which specialist systems have AI, which have announced it, and which of it does the organisation use?
2. Name the gaps: Which departments and tasks remain without AI?
3. Use specialist systems for their processes: Keep AI in the HIS, in coding or in QM where it knows the data and workflows.
4. Put the breadth on one platform: everything else, with one permission management and one log.
5. Connect instead of duplicating: Where specialist systems hold knowledge that others need, such as QM documents, the platform connects them, with the permissions of the person concerned.
6. A rule for new needs: A new need is first checked as an app on the platform; a separate tool comes only if the platform cannot or may not do the task.

How does this work with basebox?

basebox is a horizontal AI platform for the whole organisation, in our cloud in Germany or on a server at your site. basebox does not replace an HIS or a specialist platform and runs alongside any HIS. It comes with 146 use cases for hospitals in 21 areas, permission management from your existing identity management and a log; sharing an app shares its documents, never your own access.

basebox connects specialist systems through connectors: a connector to roXtra is planned at customers’ request; it is to use the roXtra interface. For an HIS or another specialist system, we build the connector as soon as the vendor provides access, for example through the FHIR interface under ISiK (the German interoperability standard for hospital systems); the design is free of charge, followed by a fixed price. basebox is not a medical device; for tasks with patient data it delivers drafts and summaries with sources, no diagnosis and no therapy recommendation; every result is checked by the doctor or the nurse.