Abstract
US hospitals are rapidly adopting artificial intelligence (AI), but there is a lack of knowledge about AI-adopting hospitals' characteristics, trends, and spread. This study aims to fill this gap by analyzing the 2022 American Hospital Association (AHA) data. The novel Hospital AI Adoption Model (HAIAM) is developed to categorize hospitals based on their AI adoption characteristics in the fields of (1) predicting patient demand, (2) optimizing workflow, (3) automating routine tasks, (4) staff scheduling, and (5) predicting staffing needs. Nearly one-fifth of US hospitals (1107 or 18.70%) have adopted some form of AI by 2022. The HAIAM shows that only 3.82% of hospitals are high adopters, followed by 6.22% moderate and 8.67% low adopters. Artificial intelligence adoption rates are highest in optimizing workflow (12.91%), while staff scheduling (9.53%) has the lowest growth rate. Hospitals with large bed sizes and outpatient surgical departments, private not-for-profit ownership, teaching status, and part of health systems are more likely to adopt different forms of AI. New Jersey (48.94%) is the leading hospital AI-adopting state, whereas New Mexico (0%) is the most lagging. These data can help policymakers better understand variations in AI adoption by hospitals and inform potential policy responses.
Keywords: artificial intelligence, hospital, technology adoption, innovation, HAIAM
Introduction
Healthcare has consistently been one of the top industries for artificial intelligence (AI) investment, receiving $31.5 billion in equity funding between 2019 and 2022.1 Numerous hospitals have designated chief analytics or chief data officers,2 and healthcare providers and professionals think AI technologies will be widespread in the next 3 years.3 Some have argued that AI could reduce $60-$120 billion, or 4%-10% of hospital costs.2 Economic models have demonstrated tremendous cost savings in diagnosis and treatment from AI tools.4 Artificial intelligence can impact hospitals through effects on clinical procedures, management practices, and hospital and patient outcomes.5 Considering these factors, the Biden administration has stressed the importance of AI in improving health outcomes.6 Understanding hospital AI adoption trends is thus important to optimize benefits and mitigate potential drawbacks. This paper provides the first comprehensive description of hospital AI adoption levels, identifies hospital characteristics associated with adoption, and proposes a novel framework, the Hospital AI Adoption Model (HAIAM), for measuring AI in the future.
Methods
The 2022 American Hospital Association (AHA) Annual Survey data are used to obtain binary indicators (0/1) of hospital AI adoption in the following fields: (1) predicting patient demand, (2) optimizing workflow, (3) automating routine tasks, (4) staff scheduling, (5) predicting staffing needs, and (6) any AI adoption—if a hospital adopted any of the 5 AI fields.
Hospital AI Adoption Model (HAIAM)
A novel HAIAM is developed to understand the extent of hospital AI adoption.7,8 The HAIAM categorizes hospitals based on the number of adopted AI fields: (1) “No Adopters”—no adoption; (2) “Low Adopters”—any 1 or 2; (3) “Moderate Adopters”—any 3 or 4; and (4) “High Adopters”—all 5. To understand the geographical spread of AI-adopting hospitals, the HAIAM is analyzed by zip code (Figure 1) and state (Figure SA1).
Figure 1.
Hospital AI Adoption Model (HAIAM) applied to US zip codes, American Hospital Association (AHA) 2022. The zip code level map is generated by HAIAM categories based on the quarterly percentile of the total artificial intelligence (AI) adoption fields for all hospitals in a zip code. Source: Authors’ analysis of the AHA data 2022.
The sample includes 5920 hospitals in all 50 states and Washington, DC, with active provider status reporting AHA data in 2022. Hospitals without Medicare numbers, geographic indicators, or closures during the study period are excluded from the sample. The analysis includes the following hospital characteristics: bed size, provider and ownership type, teaching status, designated trauma center, dedicated emergency department, outpatient department, and outpatient surgical service. The Rural-Urban Commuting Area Codes 2010 zip code level file is used to obtain rural–urban categorization of hospitals, along with state and regional indicators.
We analyze descriptive statistics of the overall sample and perform χ2 tests to compare characteristics by HAIAM. Multivariable logistic regression models are used to estimate the determinants of hospital AI adoption. Standard errors are clustered at the hospital level, and the analyses are performed using Stata 17.
Results
About one-fifth (18.70%) of hospitals adopted at least one form of AI by 2022 (Table 1). Hospitals adopted AI to optimize workflow (12.91%), automate routine tasks (11.99%), predict patient demand (9.71%), predict staffing needs (9.68%), and staff scheduling (9.53%). According to HAIAM, hospitals are low (8.67%), moderate (6.22%), and high adopters (3.82%) of AI. Among states, New Jersey (48.94%) is leading in hospital AI adoption, followed by Utah (41.38%), Connecticut (35.00%), Pennsylvania (33.49%), and the District of Columbia (33.33%) (Table 1). New Mexico (0.00%) is the state most lagging in hospital AI adoption, followed by Mississippi (1.85%), Idaho (1.96%), Alabama (3.54%), and Wisconsin (4.79%). The ranking of all states is available in Table SA1.
Table 1.
Distribution of US hospital AI adoption in 2022.
| Characteristics | N (%) |
|---|---|
| Total US hospitals in the study sample | 5920 (100.00) |
| Total hospitals adopting AI | 1107 (18.70) |
| Fields of hospital AI adoptiona | |
| Optimizing workflow | 764 (12.91) |
| Automating routine tasks | 710 (11.99) |
| Predicting patient demand | 575 (9.71) |
| Predicting staffing needs | 573 (9.68) |
| Staff scheduling | 564 (9.53) |
| Hospital AI Adoption Model (HAIAM)a | |
| No Adopters | 4813 (81.30) |
| Low Adopters | 513 (8.67) |
| Moderate Adopters | 368 (6.22) |
| High Adopters | 226 (3.82) |
| States leading in hospital AI adoptionb | |
| New Jersey | 46 (48.94) |
| Utah | 24 (41.38) |
| Connecticut | 14 (35.00) |
| Pennsylvania | 71 (33.49) |
| District of Columbia | 4 (33.33) |
| States lagging in hospital AI adoptionb | |
| New Mexico | 0 (0.00) |
| Mississippi | 2 (1.85) |
| Idaho | 1 (1.96) |
| Alabama | 4 (3.54) |
| Wisconsin | 7 (4.79) |
Source: Authors’ analysis of the American Hospital Association (AHA) data 2022.
aCalculated as a percentage of total US hospitals.
bRanking according to the percentage of total hospitals in a state adopting any form of AI in 2022. A complete list of all state AI adoption rates is available in Table SA2, included in the Supplementary material.
Analysis of the determinants of hospital AI adoption is presented in Table 2. Comapred to small hospitals, large hospitals have a 1.48 times (P < 0.001) higher probability of adopting any form of AI, 1.34 times (P < 0.05) higher probability of adopting AI for optimizing workflow, and 1.31 times (P < 0.05) higher probability of adopting AI for staff scheduling. Compared to long-term hospitals, short-term hospitals are more likely to adopt AI—any form by 1.99 times (P < 0.05), predicting patient demand by 5.00 times (P < 0.01), optimizing workflow by 3.26 times (P < 0.01), and predicting staffing needs by 3.66 times (P < 0.01). Similarly, rehabilitation, children, and critical access hospitals are more likely to adopt different forms of AI than long-term care hospitals.
Table 2.
Determinants of hospital adoption of different forms of AI, AHA 2022.
| Characteristics | Any AI adoption | Predicting patient demand | Optimizing workflow | Automating routine tasks | Staff scheduling | Predicting staffing needs |
|---|---|---|---|---|---|---|
| Bed size (ref.: small) | ||||||
| Medium | 1.16 (0.13) |
1.02 (0.15) |
0.98 (0.12) |
0.85 (0.11) |
1.12 (0.16) |
1.14 (0.16) |
| Large |
1.48
a
(0.16) |
1.15 (0.16) |
1.34
b
(0.16) |
1.16 (0.15) |
1.31
b
(0.18) |
1.20 (0.17) |
| Type of hospital (ref.: long term) | ||||||
| Short term |
1.99
b
(0.60) |
5.00
c
(2.53) |
3.26
c
(1.45) |
2.25 (0.96) |
1.59 (0.57) |
3.66
c
(1.69) |
| Psychiatric | 0.93 (0.32) |
1.36 (0.86) |
1.24 (0.66) |
0.95 (0.51) |
0.98 (0.39) |
2.22 (1.19) |
| Rehabilitation |
3.99
a
(1.29) |
4.69
c
(2.80) |
6.25
a
(2.98) |
4.23
c
(1.98) |
3.22
c
(1.25) |
7.61
a
(3.95) |
| Children's hospitals |
2.22
b
(0.87) |
3.55
b
(2.30) |
2.77 (1.52) |
1.51 (0.86) |
1.15 (0.64) |
4.57
c
(2.69) |
| Critical access hospital | 1.65 (0.53) |
3.92
c
(2.05) |
2.31 (1.07) |
1.56 (0.70) |
1.43 (0.55) |
3.35
b
(1.60) |
| Ownership (ref.: public) | ||||||
| Church | 1.21 (0.22) |
1.19 (0.27) |
0.73 (0.17) |
1.05 (0.24) |
1.34 (0.30) |
1.56 (0.36) |
| Private not for profit |
2.43
a
(0.29) |
1.81
a
(0.29) |
2.10
a
(0.28) |
2.51
a
(0.37) |
1.87
a
(0.29) |
2.26
a
(0.38) |
| Private for profit |
0.37
a
(0.06) |
0.36
a
(0.08) |
0.37
a
(0.07) |
0.36
a
(0.08) |
0.46
a
(0.10) |
0.38
a
(0.09) |
| Physician ownership | 0.45 (0.20) |
0.28 (0.21) |
0.29b (0.18) |
0.66 (0.32) |
0.34 (0.25) |
0.37 (0.27) |
| Others |
2.11
a
(0.34) |
1.93
c
(0.40) |
1.84
a
(0.34) |
2.38
a
(0.45) |
1.71
c
(0.36) |
1.96
c
(0.43) |
| Health system |
2.54
a
(0.25) |
4.04
a
(0.59) |
2.14
a
(0.24) |
2.73
a
(0.33) |
3.02
a
(0.41) |
4.41
a
(0.67) |
| Teaching Stat. (ref.: no) |
1.35
a
(0.11) |
1.55
a
(0.17) |
1.54
a
(0.14) |
1.43
a
(0.14) |
1.31
c
(0.14) |
1.23 (0.13) |
| Trauma Cent. (ref.: no) | 1.07 (0.09) |
1.38
c
(0.14) |
1.02 (0.10) |
1.06 (0.10) |
1.17 (0.12) |
1.26
b
(0.14) |
| Emer. Dept. (ref.: no) | 0.82 (0.12) |
0.54
c
(0.10) |
0.90 (0.15) |
0.83 (0.14) |
0.90 (0.18) |
0.74 (0.14) |
| Otptnt. Dept. (ref.: no) | 1.03 (0.19) |
0.95 (0.25) |
1.32 (0.31) |
1.53 (0.39) |
0.71 (0.16) |
0.85 (0.22) |
| Otptnt. Srgcl. Ser. (ref.: no) |
1.90
a
(0.30) |
1.79
c
(0.38) |
1.99
a
(0.36) |
1.97
a
(0.37) |
1.60
b
(0.33) |
1.82
c
(0.40) |
| Rural (ref.: urban) |
0.46
a
(0.05) |
0.60
a
(0.08) |
0.50
a
(0.06) |
0.58
a
(0.07) |
0.60
a
(0.08) |
0.48
a
(0.06) |
Source: authors’ analysis of the American Hospital Association (AHA) data 2022. Statistically significant results are presented in bold. Exponentiated coefficients; standard errors in parentheses.
a P < 0.001.
b P < 0.05.
c P < 0.01.
Private not-for-profits (compared to public hospitals) are more likely to adopt different forms of AI. However, private for-profit and physician-owned hospitals have a significantly lower likelihood of adopting AI into their operations. Hospitals part of health systems are more likely to adopt any form of AI by 2.54 times (P < 0.001), predicting patient demand by 4.04 times (P < 0.001), optimizing workflow by 2.14 times (P < 0.001), automating routine tasks by 2.73 times (P < 0.001), staff scheduling by 3.02 times (P < 0.001), and predicting staffing needs by 4.41 times (P < 0.001). Teaching hospitals and hospitals with outpatient surgical departments are more likely to adopt all forms of AI. Rural hospitals are significantly less likely to adopt any form of AI than urban hospitals.
Discussion
Results from this study show that hospitals are adopting AI at a much higher rate than previously reported.9 In a recent survey, 36% of hospital IT leaders indicated full or partial AI implementation, and 25% reported ongoing AI pilot projects.10 Learning about factors that enable or hamper hospitals' adoption of the new technology is essential. This study finds that teaching hospitals with large numbers of beds that are part of health systems adopt AI at a higher rate. Previous studies also found hospital size, age, and teaching status significantly associated with digital transformation.11,12 Contrary to evidence from previous technological adoption efforts, this study finds that not-for-profit ownership significantly predicts AI adoption, whereas for-profit ownership predicts AI nonadoption.12 This study finds that geographic location predicts hospital AI adoption, which has also been observed in previous technological adoption attempts.12
Due to the evolving nature of the field, it is essential to understand HAIAM. Technology adoption models are employed in various fields, such as telemedicine, electronic health records, devices, and applications.13 Since AI is a new arena of technology adoption, the proposed HAIAM will help researchers analyze and policymakers form policies regarding hospital AI utilization. Results from the HAIAM identify a small percentage of early adopters of AI, who may have an “advance use” advantage over other competitors.14
The Health Information Technology for Economic and Clinical Health Act has driven the availability of extensive healthcare data. Regarding the possession of data to train AI models, larger health systems with volumes of high-quality data have an advantage over smaller counterparts.15 Policymakers can consider policies to ensure equitable access to AI training data through interventions such as health information exchanges. In a recent survey, 35% of healthcare IT leaders mentioned limited budget and resources as the primary barrier to implementing AI in their organization, while 30% cited uncertain return on investment.10 Previous studies demonstrated that financial incentives for technological adoption speed up hospital uptake.16 If policymakers view lagging AI adoption as an important problem in certain settings, they may consider policies to lower costs or incentivize adoption at these hospitals.
States vary considerably in AI adoption environments, and several have enacted AI-related legislation pertaining to health, for example, Illinois' proposed legislation requiring hospitals to obtain certification from state entities before employing AI diagnostic algorithms.17 As demonstrated by this study, automating routine tasks and optimizing workflow are the fastest-growing AI adoption fields. Both are essential to easing the burden on providers. Hospital administrators must carefully evaluate institutional needs and state AI-related legislation before adopting AI into hospital operations.
Conclusion
This paper shows that only one-fifth of US hospitals adopted AI by 2022. Moreover, there is significant variation in the uptake of AI across hospitals. These data can help policymakers better understand the state of hospital AI adoption and inform relevant policies.
Supplementary Material
Supplementary material
Supplementary material is available at Health Affairs Scholar online.
Funding
The author received support from the Department of Health Management and Policy, University of North Carolina at Charlotte.
Data availability
The author has an active data user agreement with the American Hospitals Association (AHA) and cannot share the data publicly.
Notes
- 1. CB Insights . The state of healthcare AI in 5 charts. 2023. Accessed August 1, 2023. https://www.cbinsights.com/research/healthcare-artificial-intelligence-ai-market/
- 2. Sahni N, Stein G, Zemmel R, Cutler DM. The potential impact of Artificial Intelligence on healthcare spending. Presented at “Economics of Artificial Intelligence Conference”, September 22-23, 2022; National Bureau of Economic Research Working Paper Series 30857. 2023. 10.3386/w30857 [DOI]
- 3. BRG. AI and the future of healthcare . 2024. Accessed August 10, 2024. https://media.thinkbrg.com/wp-content/uploads/2024/02/26113735/BRG-Report-AI-and-The-Future-of-Healthcare.pdf
- 4. Khanna NN, Maindarkar MA, Viswanathan V, et al. Economics of artificial intelligence in healthcare: diagnosis vs. treatment. Healthcare (Basel). 2022;10(12):2493. 10.3390/healthcare10122493 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56. 10.1038/s41591-018-0300-7 [DOI] [PubMed] [Google Scholar]
- 6. Brainard L, Tanden N, Prabhakar A.. Delivering on the promise of AI to improve health outcomes. Briefing Room blog. 2023. Accessed August 10, 2024. https://www.whitehouse.gov/briefing-room/blog/2023/12/14/delivering-on-the-promise-of-ai-to-improve-health-outcomes/
- 7. Bengel D. Organizational Acceptance of Artificial Intelligence. Springer; 2020. [Google Scholar]
- 8. McElheran K, Li JF, Brynjolfsson E, et al. AI adoption in America: who, what, and where. JEMS, 2024;33:375–415. 10.1111/jems.12576 [DOI]
- 9. Pham P, Zhang H, Gao W, Zhu X. Determinants and performance outcomes of artificial intelligence adoption: evidence from U.S. hospitals. J Bus Res. 2024;172:114402. 10.1016/j.jbusres.2023.114402 [DOI] [Google Scholar]
- 10. Healthcare IT Leaders . The state of AI adoption in healthcare: 2023 hospital and health system IT leadership survey. 2023. Accessed August 10, 2024. https://info.healthcareitleaders.com/healthcare-ai-survey-report
- 11. Raimo N, De Turi I, Albergo F, Vitolla F. The drivers of the digital transformation in the healthcare industry: an empirical analysis in Italian hospitals. Technovation. 2023;121:102558. 10.1016/j.technovation.2022.102558 [DOI] [Google Scholar]
- 12. Zhang NJ, Seblega B, Wan T, Unruh L, Agiro A, Miao L. Health information technology adoption in U.S. acute care hospitals. J Med Syst. 2013;37(2):9907. 10.1007/s10916-012-9907-2 [DOI] [PubMed] [Google Scholar]
- 13. Rahimi B, Nadri H, Lotfnezhad Afshar H, Timpka T. A systematic review of the technology acceptance model in health informatics. Appl Clin Inform. 2018;9(3):604–634. 10.1055/s-0038-1668091 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Adler-Milstein J, Holmgren AJ, Kralovec P, Worzala C, Searcy T, Patel V. Electronic health record adoption in US hospitals: the emergence of a digital “advanced use” divide. J Am Med Inform Assoc. 2017;24(6):1142–1148. 10.1093/jamia/ocx080 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Yaraghi N. Generative AI In health care: opportunities, challenges, and policy. Health Affairs Forefront. 2024. 10.1377/forefront.20231229.503206 [DOI] [Google Scholar]
- 16. Adler-Milstein J, Jha AK. HITECH Act drove large gains in hospital electronic health record adoption. Health Aff (Millwood). 2017;36(8):1416–1422. 10.1377/hlthaff.2016.1651 [DOI] [PubMed] [Google Scholar]
- 17. NCSL NCoSL . Artificial Intelligence 2024 legislation. Accessed September 7, 2024. 2024. https://www.ncsl.org/technology-and-communication/artificial-intelligence-2024-legislation
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The author has an active data user agreement with the American Hospitals Association (AHA) and cannot share the data publicly.