Of the 1,247 artificial intelligence (AI) systems cleared by the U.S. Food and Drug Administration as of May 2025, many function as classifiers to help screen or diagnose specific medical conditions. Yet, questions remain about how to best integrate AI into healthcare workflows, including whether AI should serve as a gatekeeper, determining which patients require human attention, or as a second opinion to complement medical consultations. Motivated by this question, we model a healthcare system in which patients can consult a specialist, an AI system, or both. The key design question is whether the patient should first consult AI or the specialist, corresponding to AI’s gatekeeper and second-opinion roles, respectively. We model a two-step decision-making process influenced by an initial signal, or anchor. Contrary to popular belief, we show using AI as a gatekeeper does not necessarily increase missed diagnoses; using AI as a second opinion, on the other hand, can increase missed diagnoses but can also increase false positives. In general, the gatekeeper approach is preferable in low-risk settings, whereas the second-opinion approach is better suited for high-risk patients for whom avoiding missed diagnoses is a primary concern. Notably, scenarios exist where AI should not be used for intermediate-risk patients for whom uncertainty is highest, challenging the premise that AI is most useful in reducing uncertainty. Finally, applying our model to glaucoma diagnosis, we numerically illustrate cost savings from optimizing patient pathways. Our work highlights the potential for AI to contribute to the United Nations’ Sustainable Development Goals by optimizing resource allocation and improving patient outcomes.
AboyMMinssenTVayenaE (2024) Navigating the EU AI Act: Implications for regulated digital medical products. NPJ Digital Medicine7(1): 237.
2.
AbràmoffMDDaiTZouJ (2024) Scaling adoption of medical AI—reimbursement from value-based care and fee-for-service perspectives. NEJM AI1(5): AIpc2400083.
3.
AbràmoffMDWhitestoneNPatnaikJL, et al. (2023) Autonomous artificial intelligence increases real-world specialist clinic productivity in a cluster-randomized trial. NPJ Digital Medicine6(1): 184.
4.
AdidaEDaiT (2024) Impact of physician payment scheme on diagnostic effort and testing. Management Science70(8): 5408–5425.
5.
AgarwalNMoehringARajpurkarP, et al. (2024) Combining human expertise with artificial intelligence: Experimental evidence from radiology. National Bureau of Economic Research Working Paper.
6.
AhmedMDaiTChannaR, et al. (2025) Cost-effectiveness of AI for pediatric diabetic eye exams from a health system perspective. NPJ Digital Medicine8(1): 3.
7.
American College of Radiology (2021) Integrating AI into the Clinical Workflow; Case.
8.
BachAKPNørgaardTMBrokJC, et al. (2023) “If I had all the time in the world”: Ophthalmologists’ perceptions of anchoring bias mitigation in clinical AI support (New York, NY, USA: Association for Computing Machinery). Available at: https://doi.org/10.1145/3544548.3581513.
9.
BalakrishnanMFerreiraKTongJ (2022) Improving human-algorithm collaboration: Causes and mitigation of over- and under-adherence, working Paper.
Centers for Medicare & Medicaid Services (2024) Nondiscrimination in Health Programs and Activities. Available at: https://bit.ly/fed-gov-non-discrim-hlth-24, effective date: July 5, 2024. RIN: 0945-AA17. Document Number: 2024-08711. Document Citation: 89 FR 37522, pp.37522–37703.
12.
ChenLKimSHLeeHL (2021) Vehicle maintenance contracting in developing economies: The role of social enterprise. Manufacturing & Service Operations Management23(6): 1651–1668.
13.
CuiRLiMZhangS (2022) AI and procurement. Manufacturing & Service Operations Management24(2): 691–706.
14.
DaiTAbràmoffMD (2023) Incorporating artificial intelligence into healthcare workflows: Models and insights. Tutorials in Operations Research: Advancing the Frontiers of OR/MS: From Methodologies to Applications, pp.133–155 (INFORMS).
15.
DaiTSinghS (2020) Conspicuous by its absence: Diagnostic expert testing under uncertainty. Marketing Science39(3): 540–563.
16.
DaiTSinghS (2025) Artificial intelligence on call: The physician's decision of whether to use AI in clinical practice. Journal of Marketing Research62(5): 854–875.
17.
DaiTTayurS (2020) OM forum—healthcare operations management: A snapshot of emerging research. Manufacturing & Service Operations Management22(5): 869–887.
18.
DaiTTayurS (2022) Designing AI-augmented healthcare delivery systems for physician buy-in and patient acceptance. Production and Operations Management31(12): 4443–4451.
19.
DargahiHMonajemiASoltaniA, et al. (2022) Anchoring errors in emergency medicine residents and faculties. Medical Journal of the Islamic Republic of Iran36: 124.
20.
Das GuptaAKarmarkarUSRoelsG (2016) The design of experiential services with acclimation and memory decay: Optimal sequence and duration. Management Science62(5): 1278–1296.
21.
de VéricourtFGurkanH (2023) Is your machine better than you? You may never know. Management Science, ePub ahead of print.
22.
DvijothamKWinkensJBarsbeyM, et al. (2023) Enhancing the reliability and accuracy of ai-enabled diagnosis via complementarity-driven deferral to clinicians. Nature Medicine29(7): 1814–1820.
23.
FDA (2025) Artificial intelligence and machine learning (AI/ML)-enabled medical devices. Available at: https://bit.ly/FDA-AI-ML, published by the U.S. Food and Drug Administration (accessed 9 August 2025).
24.
FreemanMRobinsonSScholtesS (2021) Gatekeeping, fast and slow: An empirical study of referral errors in the emergency department. Management Science67(7): 4209–4232.
25.
FügenerAGrahlJGuptaA, et al. (2022) Cognitive challenges in human-artificial intelligence collaboration: Investigating the path toward productive delegation. Information Systems Research33(2): 678–696.
26.
GilbertS (2024) The EU passes the AI Act and its implications for digital medicine are unclear. NPJ Digital Medicine7(1): 135.
27.
GohEGalloRHomJ, et al. (2024) Large language model influence on diagnostic reasoning: A randomized clinical trial. JAMA Network Open7(10): e2440969.
28.
Grand-ClèmentJPauphiletJ (2023) The best decisions are not the best advice: Making adherence-aware recommendations. Available at: https://doi.org/10.48550/ARXIV.2209.01874.
29.
GurkanHde VéricourtF (2022) Contracting, pricing, and data collection under the AI flywheel effect. Management Science68(12): 8791–8808.
30.
HastingsWK (1970) Monte Carlo sampling methods using Markov chains and their applications. Biometrika57(1): 97–109.
31.
HathawayBAKaganEDadaM (2023) The gatekeeper’s dilemma: “When should I transfer this customer?”. Operations Research71(3): 843–859.
32.
HoSDoigGSLyA (2022) Attitudes of optometrists towards artificial intelligence for the diagnosis of retinal disease: A cross-sectional mail-out survey. Ophthalmic and Physiological Optics42(6): 1170–1179.
33.
JinLTangRYeH, et al. (2023) Path dependency in physician decisions. The Review of Economic Studies91(5): 2916–2953.
34.
KahnemanDMillerDT (1986) Norm theory: Comparing reality to its alternatives. Psychological Review93(2): 136–153.
35.
KovalykOMorales-SánchezJVerdú-MonederoR, et al. (2022) PAPILA: Dataset with fundus images and clinical data of both eyes of the same patient for glaucoma assessment. Scientific Data9(1): 291.
36.
LaiJXuLFangX, et al. (2024) Regulating adaptive medical artificial intelligence: Can less oversight lead to greater compliance? Working paper, Johns Hopkins University.
37.
LeeBPatelSFavoritoC, et al. (2025) Development and commercialization pathways of AI medical devices in the United States: Implications for safety and regulatory oversight. NEJM AI2(7): AIra2500061.
38.
LeeHLRammohanSVSeptL (2013) Innovative logistics in extreme conditions: The case of health care delivery in Gambia. In: Bookbinder JH (ed) Handbook of Global Logistics: Transportation in International Supply Chains. New York: Springer.
39.
LeeHLTangCS (1997) Modelling the costs and benefits of delayed product differentiation. Management Science43(1): 40–53.
40.
LeeHLTangCS (2018) Socially and environmentally responsible value chain innovations: New operations management research opportunities. Management Science64(3): 983–996.
41.
LiMLDaiT (2025) The future in sight: LumineticsCore and the first autonomous AI for diagnostics. Case 626-019, Harvard Business School, Boston, MA. Available at: https://hbsp.harvard.edu/product/626019-PDF-ENG, product # 626019-PDF-ENG.
42.
LiYDaiTQiX (2022) A theory of interior peaks: Activity sequencing and selection for service design. Manufacturing & Service Operations Management24(2): 993–1001.
LimJIRegilloCDSaddaSR, et al. (2023) Artificial intelligence detection of diabetic retinopathy. Ophthalmology Science3(1): 100228.
45.
LuanSSinghSDaiT (2024) Algorithmic bias and physician liability. Working paper, Johns Hopkins University.
46.
LyDPShekellePGSongZ (2023) Evidence for anchoring bias during physician decision-making. JAMA Internal Medicine183(8): 818–823.
47.
MathengeWWhitestoneNNkurikiyeJ, et al. (2022) Impact of artificial intelligence assessment of diabetic retinopathy on referral service uptake in a low-resource setting. Ophthalmology Science2(4): 100168.
MehtaKMRerolleFRammohanSV, et al. (2016) Systematic motorcycle management and health care delivery: A field trial. American Journal of Public Health106(1): 87–94.
50.
MiaoSYeungATangCS, et al. (2024) Does investment in AI talent pay? An examination of efficiency and spillover effects in manufacturing supply chains. Working paper.
51.
MullainathanSObermeyerZ (2021) Diagnosing physician error: A machine learning approach to low-value health care. Quarterly Journal of Economics137(2): 679–727.
52.
OgdieARReillyJBPangWG, et al. (2012) Seen through their eyes: Residents’ reflections on the cognitive and contextual components of diagnostic errors in medicine. Academic Medicine87(10): 1361–1367.
53.
OlsderWMartaganTTangCS (2023) Improving access to rare disease treatments: Subsidy, pricing, and payment schemes. Management Science69(9): 5256–5274.
54.
OrfanoudakiASaghafianSSongK, et al. (2022) Algorithm, human, or the centaur: How to enhance clinical care? HKS Working Paper.
55.
Porsdam MannSCohenIGMinssenT (2024) The EU AI Act: Implications for U.S. Health Care. NEJM AI1(11): AIp2400449.
56.
PriceWNGerkeSCohenIG (2019) Potential liability for physicians using artificial intelligence. JAMA322(18): 1765–1766.
57.
RastogiCZhangYWeiD, et al. (2022) Deciding fast and slow: The role of cognitive biases in ai-assisted decision-making. Proceedings of the ACM on Human-computer Interaction6(CSCW1): 1–22.
58.
RuamviboonsukPKrauseJChotcomwongseP, et al. (2019) Deep learning versus human graders for classifying diabetic retinopathy severity in a nationwide screening program. NPJ Digital Medicine2(1): 1.
59.
SagonaMDaiTMacisM, et al. (2025) Trust in AI-assisted health systems and AI’s trust in humans. NPJ Health Systems2: 10.
60.
ShumskyRAPinkerEJ (2003) Gatekeepers and referrals in services. Management Science49(7): 839–856.
61.
SodhiMSTangCS (2024) Seeking and exploiting synergies among the un sustainability development goals: Research opportunities for operations management. Production and Operations Management33(4): 962–978.
62.
TangCS (2016) OM Forum—Making OM research more relevant: “why?” and “how?”. Manufacturing & Service Operations Management18(2): 178–183.
63.
TverskyAKahnemanD (1974) Judgment under uncertainty: Heuristics and biases. Science (New York, N.Y.)185(4157): 1124–1131.
64.
van KolfschootenHvan OirschotB (2024) The EU Artificial Intelligence Act (2024): Implications for healthcare. Health Policy (Amsterdam, Netherlands)149: 105152.
65.
WolfRMChannaRAbramoffMD, et al. (2020) Cost-effectiveness of autonomous point-of-care diabetic retinopathy screening for pediatric patients with diabetes. JAMA Ophthalmology138(10): 1063.
66.
XieYNguyenQDHamzahH, et al. (2020) Artificial intelligence for teleophthalmology-based diabetic retinopathy screening in a national programme: An economic analysis modelling study. Lancet Digital Health2(5): e240–e249.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
