The recent systematic review published in JMIR synthesises empirical evidence on hospital artificial intelligence (AI) implementations and validates a hospital-specific five-layer AI platform architecture encompassing infrastructure, data, algorithm, application, and security and compliance,offering a practical roadmap for scaling AI beyond isolated pilots. According to the report by JMIR, the review mapped 29 peer-reviewed empirical studies conducted across 11 countries to this architecture and scored each layer on a 0–5 ordinal maturity scale to quantify readiness and gaps. [1]

The study followed PRISMA guidelines,searched Web of Science,Embase,PubMed,and Scopus to May 23,2025,and applied dual independent screening,standardised extraction,and Critical Appraisal Skills Programme quality assessment;studies were included only if they described real-world hospital AI use with measurable outcomes,excluding reviews,commentaries,bench studies,non-English papers and grey literature. The authors emphasise that this scope preserved coding reliability but may have omitted non-English or unpublished implementation reports. [1]

Quantitative mapping showed a consistent maturity pattern across the five layers:application (mean 3.17,SD 0.85) and data (mean 3.00,SD 0.76) were the most mature,followed by algorithm (mean 2.79,SD 0.77) and infrastructure (mean 2.79,SD 1.70),whereas security and compliance lagged markedly (mean 1.69,SD 1.89). These pooled scores indicate that many deployments have reached clinical feasibility while sustaining enterprise-wide,governed AI remains limited. [1]

Weighted co-occurrence and Jaccard similarity analyses identified a tightly integrated technical core formed by data,algorithm and application layers;Jaccard indices of 0.80–0.89 demonstrate strong cross-layer coupling,whereas security and compliance showed weak alignment with the technical core (Jaccard=0.43–0.46). The authors interpret this as evidence that data readiness,model development and workflow integration are typically developed together,while governance is often peripheral. [1]

The 29 included studies spanned radiology,emergency medicine,cardiology,gynecology,surgery,nursing,psychiatry and cross-specialty implementations,and originated predominantly from high-income countries,the United States accounting for 41.4% of studies. Study designs were heterogeneous,with diagnostic test studies,cohort and qualitative designs most common,and quality varied,with 37.9% of studies meeting over 80% of CASP criteria and 10.3% scoring below 40%. These patterns shape which layers are visible in published work because technical studies emphasise data,algorithm and application,whereas qualitative work better captures infrastructure and governance issues. [1]

The review presents four fielded deployment narratives that exemplify the five-layer model:emergency CT triage for intracranial haemorrhage,radiograph first-reader triage,ML-enabled large-vessel occlusion detection for stroke transfer coordination,and a 90-site multicentre imaging platform. Across these examples,workflow-native integration , such as automated reprioritisation and alerting , translated algorithm outputs into measurable clinical benefits,but success depended on robust infrastructure,data pipelines,and explicit governance mechanisms such as standard operating procedures,drift monitoring and audit trails. [1]

Infrastructure emerged as variably mature:some deployments reported enterprise-grade compute,containerised inference and EHR/PACS integration,whereas others remained departmental or pilot-scale with bandwidth and compute bottlenecks that hinder real-time performance and scalable monitoring. The authors highlight technological solutions being proposed in the literature,including HL7 FHIR–based data lakes,hybrid cloud–edge topologies,and automated retraining pipelines,but note these are unevenly adopted across hospitals. [1]

Security and compliance were the weakest and most inconsistent layer in the evidence base,with low maturity scores and high variability;common deficiencies included absent or ad hoc privacy safeguards,lack of model governance,and incomplete regulatory alignment. The review recommends compliance-by-design approaches,early regulator engagement,federated learning and immutable audit mechanisms as pathways to reposition governance from a peripheral add-on to a core platform function. [1]

Organisational and adoption barriers are emphasised alongside technical gaps:clinician skepticism,workflow misfit,lack of interdisciplinary teams,and misaligned procurement and budgeting cycles frequently limited uptake. The review notes that projects with clinical champions,interdisciplinary governance,and phased,low-risk use cases achieved better adoption and measurable benefits. Economic constraints , high upfront and maintenance costs , also recur as a critical obstacle,and the authors recommend funding models such as AI-as-a-service,shared consortia and phased investment to mitigate financial risk. [1]

The authors acknowledge limitations of their review that temper interpretation:excluding non-English and grey literature may have omitted relevant implementations;the 0–5 ordinal maturity scale simplifies complex contextual dynamics;and the framework has not yet been prospectively validated in diverse hospital environments. They call for multilingual evidence retrieval,refined cross-layer maturity metrics,and prospective multicentre evaluations to test scalability and generalisability. [1]

For hospital leaders,policymakers and vendors,the study offers actionable guidance:use the five-layer model to prioritise phased investments in infrastructure and governance,embed compliance-by-design into platform development,align vendor offerings with interoperability and monitoring needs,and target workflow-native pilots that demonstrate measurable clinical or operational impact before scaling. According to the report by JMIR,sustainable,hospital-wide AI will require stronger investment in infrastructure,data governance,and compliance rather than exclusively focusing on algorithmic performance. [1]

In summary,the JMIR synthesis validates a pragmatic five-layer hospital AI platform architecture and reveals a persistent maturity imbalance:a cohesive technical core of data,algorithm and application exists in many published implementations,while infrastructure resilience and security-and-compliance mechanisms lag and must be elevated to achieve sustainable,enterprise-scale AI in health care. The framework provides a structured lens for future research,policy and investment to move deployments from pilots to durable clinical and operational transformation. [1]

📌 Reference Map:

##Reference Map:

  • [1] (JMIR) - Paragraph 1, Paragraph 2, Paragraph 3, Paragraph 4, Paragraph 5, Paragraph 6, Paragraph 7, Paragraph 8, Paragraph 9, Paragraph 10, Paragraph 11, Paragraph 12

Source: Noah Wire Services