Hospitals reported broadening use of predictive artificial intelligence (AI) across clinical and operational workflows between 2023 and 2024, with adoption rising from 66% to 71% among non-federal acute care providers. Integration within the electronic health record (EHR) remains the dominant route, reflecting the technology’s embedding in day-to-day practice. Growth has not been uniform, with system-affiliated, larger and urban hospitals moving faster than smaller, rural and independent peers. Meanwhile, use cases are shifting, with rapid gains in billing simplification and scheduling facilitation alongside continued reliance on models that predict inpatient health trajectories and identify high-risk outpatients. Alongside uptake, hospitals reported more systematic evaluation and multi-stakeholder oversight, signalling a maturing approach to model performance, bias and ongoing monitoring.
Adoption Rises but Gaps Persist
By 2024, about seven in ten hospitals reported using predictive AI integrated with the EHR, up from two-thirds in 2023. Adoption increased across all hospital types yet remained concentrated among medium and large facilities, urban locations and members of multi-hospital systems. System affiliation was strongly associated with higher use: system members reported 81% adoption in 2023 and 86% in 2024, compared with 31% and 37% among independent hospitals. Size also mattered, with large hospitals reporting the highest levels, followed by medium hospitals and small hospitals trailing.
Must Read: Global Trends in Digital Therapeutics
Geography and designation contributed to variation. Urban hospitals used predictive AI at higher rates than rural hospitals in both years. Critical access hospitals (CAHs) lagged behind non-CAHs, reflecting persistent differences linked to scale, resources and digital infrastructure. Ownership also aligned with uptake patterns, with non-profit and for-profit hospitals outpacing government-owned facilities. Technology stack played a role: hospitals using the market leading EHR vendor reported 90% adoption in 2024, compared with 50% among hospitals using other vendors.
Model sourcing shows EHR-embedded tools remain foundational. In 2024, 80% of hospitals used predictive AI furnished by their EHR developer, while 52% used third-party models and 50% used self-developed models. This mix underlines a dual track of platform-based capabilities supplemented by solutions procured from external vendors or built in house. Differences in adoption intensity by hospital characteristics suggest a continuing digital divide that may shape the pace and breadth of AI-enabled transformation across the sector.
Administrative Use Cases Expand Rapidly
Hospitals reported significant year-on-year growth in selected operational applications. The fastest gains from 2023 to 2024 were in simplifying or automating billing procedures, which rose by 25 percentage points and facilitating scheduling, up 16 percentage points. Identifying high-risk outpatients to inform follow-up care increased by 9 percentage points. Use of predictive AI for treatment recommendations rose slightly by 2 percentage points while predicting health trajectories or risks for inpatients remained high and stable, reinforcing the centrality of risk stratification for inpatient care.
Model source influenced how these use cases evolved. Hospitals relying on third-party or self-developed AI reported higher use for billing simplification in 2024 than those using EHR-developed AI, 73% versus 58%. Growth in identifying high-risk outpatients was also steeper for hospitals using third-party or self-developed AI, up 51 percentage points, compared with a 3 percentage-point decline among hospitals using EHR-developed models. In contrast, facilitating scheduling grew more among hospitals using EHR-sourced AI, up 19 percentage points, compared with 6 percentage points among those using third-party or self-developed tools.
These patterns suggest hospitals are supplementing EHR-integrated capabilities with targeted solutions for selected administrative pressures while continuing to rely on embedded models for core clinical risk tasks. The combination of high baseline use in inpatient risk prediction and rapid expansion in billing and scheduling indicates a balanced portfolio that spans clinical prioritisation and operational efficiency. Monitoring health through wearables integration and recommending treatments remained relatively less common, reflecting more limited uptake for these categories during the period.
Evaluation and Governance Mature
Evaluation practices became more robust in 2024. Most hospitals assessed predictive AI for model accuracy and bias and reported post-implementation evaluation or monitoring. Specifically, 82% evaluated accuracy, 74% evaluated bias and 79% conducted post-implementation evaluation or monitoring. While overall rates for accuracy and bias evaluation were steady year on year, a larger share of hospitals reported evaluating all or most of their models, pointing to increased depth and coverage in assurance processes.
Governance structures were also broadened. In 2024, 74% of hospitals indicated multiple entities were accountable for evaluating predictive AI. The most frequently cited were a specific committee or task force for predictive AI at 66% and division or department leaders at 60%. Clinical decision support committees and designated senior executives were also commonly involved while IT staff were least often cited at 41%. Only a small minority indicated that none of the listed entities were accountable. This distribution aligns with a multidisciplinary approach in which oversight spans leadership, clinical governance and specialised AI committees to align model deployment with organisational strategy and risk controls.
Differences in evaluation practices mirrored adoption disparities. Medium and large hospitals, urban sites, system-affiliated organisations, non-CAHs and hospitals using the market leading EHR vendor were more likely to report local evaluation of accuracy and bias. These trends indicate that scale and digital maturity support more comprehensive assurance though progress was visible across groups. Reference frameworks and safety resources, including guidance for risk management and EHR safety, provide additional structure for evaluation and monitoring without supplanting local governance.
Hospitals expanded predictive AI use from 2023 to 2024, consolidating EHR-integrated models for core clinical risk tasks while accelerating adoption for billing simplification and scheduling facilitation. Uptake remained uneven, with system-affiliated, larger and urban organisations moving ahead of smaller, rural, independent and critical access peers and hospitals on the market leading EHR reporting higher adoption. Evaluation and governance matured, with more hospitals assessing accuracy, bias and post-implementation performance and distributing accountability across committees, leaders and clinical governance bodies. For healthcare professionals and decision-makers, the trajectory points to broader operational and clinical integration of predictive AI supported by stronger multi-stakeholder oversight and monitoring.
Source: Assistant Secretary for Technology Policy
Image Credit: iStock
