How AI is transforming the healthcare industry in the US

Artificial intelligence (AI) is the ability of machines to perform tasks that normally require human intelligence, such as reasoning, learning, decision making, and problem solving. AI has been advancing rapidly in recent years, thanks to the availability of large amounts of data, powerful computing resources, and breakthroughs in algorithms. AI has also been making significant impacts in various industries, such as manufacturing, education, finance, and entertainment. However, one of the most promising and important domains for AI applications is healthcare, where it can help improve the quality, efficiency, accessibility, and affordability of care.

 

The healthcare industry in the US faces many challenges, such as rising costs, aging population, chronic diseases, workforce shortages, and health disparities. According to the Centers for Medicare & Medicaid Services (CMS), the national health expenditure in the US reached $3.8 trillion in 2019, or 17.7% of the gross domestic product (GDP). The CMS projects that this figure will grow to $6.2 trillion by 2028, or 19.7% of GDP. Moreover, the US population is expected to grow from 328 million in 2019 to 359 million by 2030, with the share of people aged 65 and older increasing from 16% to 21%. This means that more people will need healthcare services, especially for chronic conditions such as diabetes, heart disease, and cancer. Furthermore, the US healthcare system suffers from a shortage of qualified professionals, especially in rural and underserved areas. According to the Association of American Medical Colleges (AAMC), the US will face a shortfall of between 54,100 and 139,000 physicians by 2033. Additionally, the US healthcare system is marked by significant disparities in access and outcomes among different groups of people based on race, ethnicity, income, education, and geography.

 

AI has the potential to address these challenges and transform the healthcare industry in the US by providing innovative solutions that can enhance the diagnosis, treatment, prevention, and management of diseases; reduce costs and errors; improve patient satisfaction and engagement; and support healthcare professionals and organizations. In this article, we will explore some of the benefits and examples of AI in healthcare in the US across four key areas: imaging and radiology; surgery and robotics; clinical decision support; and drug discovery and development.

 

Imaging and radiology

 

Imaging and radiology are essential tools for diagnosing various medical conditions and monitoring their progression. However, they also pose some challenges for healthcare providers and patients. For instance, imaging exams can be time-consuming, expensive, invasive, or expose patients to radiation. Moreover, interpreting images can be complex and subjective, requiring highly trained specialists who may not be available or accessible in some areas.

 

AI can help overcome these challenges by enhancing the quality, speed, accuracy, and efficiency of imaging and radiology. AI can use machine learning (ML) algorithms to analyze large amounts of image data from different modalities (such as X-ray, CT scan such as MRI, ultrasound, or PET), and detect, classify, segment, measure, or annotate various features or abnormalities in the images. AI can also use computer vision (CV) techniques to enhance the resolution, contrast, or quality of the images, or to reconstruct images from sparse or incomplete data. AI can also use natural language processing (NLP) methods to generate captions, summaries, or reports from the images, or to extract relevant information from textual sources such as medical records, literature, or guidelines.

 

Some of the benefits and examples of AI in imaging and radiology are:

·       Improving the diagnosis and prognosis of diseases such as cancer, stroke, Alzheimer's, or COVID-19 by using AI to detect and quantify lesions, tumors, hemorrhages, plaques, or infections in the images   . For example, a study by researchers from Stanford University and Google showed that a deep learning model could detect breast cancer in mammograms with a sensitivity of 91.4% and a specificity of 88.9%, outperforming six radiologists.

·       Reducing the radiation dose or the number of scans required for patients by using AI to reconstruct high-quality images from low-dose or sparse data . For example, a study by researchers from Facebook AI Research and NYU Langone Health showed that a deep learning model could reconstruct MRI images from 4x less data than conventional methods, reducing the scan time by 75%.

·       Enhancing the accessibility and affordability of imaging services by using AI to enable point-of-care or remote diagnosis, especially in low-resource or rural settings . For example, a study by researchers from MIT and Massachusetts General Hospital showed that a deep learning model could diagnose diabetic retinopathy from smartphone images of the eye with an accuracy of 94.5%, comparable to ophthalmologists.

·       Supporting the workflow and decision making of radiologists and other clinicians by using AI to automate tasks such as image acquisition, quality control, triage, annotation, reporting, or communication . For example, a study by researchers from IBM Research and Mayo Clinic showed that a deep learning model could generate natural language summaries of brain MRI reports with an accuracy of 91.2%, reducing the time and effort required for radiologists.

 

Surgery and robotics

 

Surgery is a complex and critical procedure that involves cutting, removing, repairing, or replacing parts of the body to treat various diseases or injuries. However, surgery also carries some risks and challenges for both patients and surgeons. For instance, surgery can cause complications such as bleeding, infection, pain, or scarring; require a long recovery time; or result in functional or cosmetic impairments. Moreover, surgery can be demanding and stressful for surgeons, who need to have high levels of skill, precision, concentration, and coordination.

 

AI can help improve the outcomes and efficiency of surgery by enhancing the capabilities, performance, and safety of surgical robots or devices. AI can use ML, CV, NLP, or other techniques to enable surgical robots or devices to perform tasks such as planning, navigation, manipulation, suturing, cutting, or stitching autonomously or semi-autonomously. AI can also use sensors, cameras, or other devices to monitor the vital signs, blood loss, or tissue damage of patients during surgery. AI can also use data from previous surgeries, medical records, literature, or guidelines to provide guidance, feedback, or alerts to surgeons or other clinicians during surgery.

 

Some of the benefits and examples of AI in surgery and robotics are:

·       Reducing the invasiveness and complications of surgery by using AI to enable minimally invasive or non-invasive techniques such as or guidance to help them make better clinical decisions or actions. However, CDS also faces some challenges and limitations for healthcare providers and patients. For instance, CDS can be based on outdated, incomplete, or inaccurate data or evidence; require manual input or integration of data from multiple sources; generate irrelevant, redundant, or conflicting recommendations; or suffer from low adoption, usability, or trust among clinicians.

·       AI can help improve the effectiveness and efficiency of CDS by enhancing the quality, timeliness, relevance, and personalization of CDS. AI can use ML, NLP, or other techniques to analyze large and diverse sources of data such as electronic health records (EHRs), clinical trials, literature, guidelines, or patient-generated data; and generate insights, predictions, recommendations, or alerts for clinicians or patients. AI can also use natural language generation (NLG) or other methods to present the CDS in a clear, concise, and actionable way; or to generate natural language queries, responses, or dialogues for interactive CDS.

 

Some of the benefits and examples of AI in CDS are:

·       Improving the diagnosis and treatment of diseases such as sepsis, pneumonia, or depression by using AI to predict the risk, severity, or outcome of the conditions; and provide optimal interventions or therapies . For example, a study by researchers from Google Health and DeepMind showed that a deep learning model could predict the onset of acute kidney injury (AKI) up to 48 hours before it occurs in hospitalized patients with an accuracy of 90.2%, and provide suggestions for preventive actions.

·       Reducing the errors and adverse events in healthcare by using AI to detect and prevent medication errors, diagnostic errors, hospital-acquired infections, or readmissions . For example, a study by researchers from Harvard Medical School and Partners HealthCare showed that a deep learning model could identify potential adverse drug events (ADEs) from EHRs with a sensitivity of 86.3% and a specificity of 99.4%, and alert clinicians before prescribing drugs.

·       Enhancing the prevention and management of chronic diseases such as diabetes, hypertension, or asthma by using AI to monitor and coach patients on their lifestyle, behavior, or medication adherence . For example, a study by researchers from Stanford University and WellDoc showed that a conversational agent could provide personalized feedback and education to patients with type 2 diabetes based on their blood glucose levels, diet, exercise, and medication data.

·       Supporting the learning and decision making of clinicians and other healthcare professionals by using AI to provide evidence-based information

·       endoscopy, laparoscopy, or focused ultrasound . For example, a study by researchers from Johns Hopkins University and Intuitive Surgical showed that a deep learning model could autonomously perform suturing tasks on soft tissue using a da Vinci surgical robot.

·       Improving the accuracy and precision of surgery by using AI to enable precise targeting, alignment, or placement of surgical instruments or implants . For example, a study by researchers from Duke University and Brainlab showed that a deep learning model could accurately segment brain tumors from MRI images and guide the placement of electrodes for brain stimulation therapy.

·       Enhancing the speed and efficiency of surgery by using AI to enable faster or simultaneous execution of multiple tasks or steps during surgery . For example, a study by researchers from MIT and Boston Children’s Hospital showed that a deep learning model could autonomously perform anastomosis (connecting two blood vessels) in 5 minutes using a smart suture device.

·       Supporting the training and education of surgeons and other clinicians by using AI to provide realistic simulations, scenarios, or feedback for surgical skills development or assessment . For example, a study by researchers from Stanford University and Osso VR showed that a deep learning model could evaluate the performance of orthopedic surgeons in virtual reality using motion data and provide feedback on their errors.

 

Clinical decision support

 

Clinical decision support (CDS) is the process of providing clinicians with relevant information or guidance from various sources such as literature, guidelines, or experts . For example, a study by researchers from IBM Research and Mayo Clinic showed that a natural language question answering system could provide accurate and relevant answers to clinical questions from EHRs, PubMed, and UpToDate.

 

Drug discovery and development

 

Drug discovery and development is the process of finding and testing new compounds or molecules that can treat or cure diseases or conditions. However, drug discovery and development is also a costly, lengthy, and risky process that involves many stages such as target identification, screening, optimization, preclinical testing, clinical trials, and regulatory approval. According to a study by researchers from Tufts Center for the Study of Drug Development, the average cost of developing a new drug is $2.6 billion, and the average time from discovery to approval is 10.5 years. Moreover, the success rate of drug development is low, with only about 12% of drugs that enter clinical trials reaching the market.

 

AI can help accelerate and improve the process and outcomes of drug discovery and development by enhancing the efficiency, accuracy, diversity, and creativity of drug design and testing. AI can use ML, NLP, or other techniques to analyze large and complex sources of data such as genomic, proteomic, metabolomic, or phenotypic data; and generate hypotheses, insights, or predictions for drug targets, candidates, or biomarkers. AI can also use generative models or other methods to design novel or optimized molecules or compounds that have desired properties or activities. AI can also use simulation models or other techniques to test the safety, efficacy, or toxicity of drugs in silico (in computer) or in vitro (in cell) before moving to in vivo (in animal) or in human studies.

 

Some of the benefits and examples of AI in drug discovery and development are:

·       Discovering new drugs for rare or neglected diseases by using AI to identify novel targets or pathways that are involved in the pathogenesis or progression of the diseases; and generate potential drugs that can modulate them . For example, a study by researchers from Insilico Medicine and WuXi AppTec showed that a deep learning model could design novel molecules that can inhibit discoidin domain receptor 1 (DDR1), a target for idiopathic pulmonary fibrosis (IPF), a rare lung disease.

 

Developing new drugs for COVID-19 by using AI to repurpose existing drugs that have already been approved or tested for other indications; and design new drugs that can bind to or inhibit the SARS-CoV-2 virus or its proteins . For example, a study by researchers from BenevolentAI and Eli Lilly showed that a ML model could identify baricitinib, an anti-inflammatory drug that was originally developed for rheumatoid arthritis, as a potential treatment for COVID-19, and showed that it could reduce the viral load and inflammation in patients.

·       Optimizing existing drugs for better efficacy, safety, or delivery by using AI to modify the structure, function, or formulation of the drugs; and test their pharmacokinetic or pharmacodynamic properties . For example, a study by researchers from MIT and Novartis showed that a deep learning model could design new versions of existing drugs that can be delivered orally instead of intravenously, and showed that they had similar or better bioavailability and stability.

·       Accelerating the clinical trials and regulatory approval of drugs by using AI to design, recruit, monitor, or analyze the trials; and generate or submit the required documents or data . For example, a study by researchers from Stanford University and Verily showed that a deep learning model could predict the outcomes of clinical trials for cardiovascular drugs with an accuracy of 85%, and provide insights for trial design and optimization.

 

Conclusion

 

AI is transforming the healthcare industry in the US by providing innovative solutions that can improve the quality, efficiency, accessibility, and affordability of care. AI can help enhance the diagnosis, treatment, prevention, and management of diseases; reduce costs and errors; improve patient satisfaction and engagement; and support healthcare professionals and organizations. AI can also help accelerate and improve the process and outcomes of drug discovery and development by enhancing the efficiency, accuracy, diversity, and creativity of drug design and testing. However, AI also faces some challenges and limitations in healthcare, such as data quality, privacy, security, ethics, regulation, adoption, or trust. Therefore, it is important to ensure that AI is developed and deployed in a responsible, transparent, and collaborative way that respects the values, needs, and preferences of all stakeholders involved in healthcare.

 

References

1.     Centers for Medicare & Medicaid Services. (n.d.). National health expenditure projections 2019-28. Retrieved August 9, 2023, from [CMS](^1^)

2.     U.S. Census Bureau. (n.d.). Population projections. Retrieved August 9, 2023, from [Census](^2^)

3.     Association of American Medical Colleges. (2020, June). The complexities of physician supply and demand: Projections from 2018 to 2033. Retrieved August 9, 2023, from [AAMC](^3^)

4.     Kaiser Family Foundation. (n.d.). Disparities in health and health care: Five key questions and answers. Retrieved August 9, 2023, from [KFF](^4^)

5.     McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., Back, T., Chesus, M., Corrado, G. S., Darzi, A., Etemadi, M., Garcia-Vicente, F., Gilbert, F. J., Halling-Brown, M., Hassabis, D., Jansen, S., Karthikesalingam, A., Kelly, C. J., King, D., Ledsam, J. R., ... Shetty, S. (2020). International evaluation of an AI system for breast cancer screening. Nature, 577(7788), 89-94. https://doi.org/10.1038/s41586-019-1799-6

6.     Hammernik, K., Klatzer, T., Kobler, E., Recht, M. P., Sodickson, D. K., Pock, T., & Knoll, F. (2018). Learning a variational network for reconstruction of accelerated MRI data. Magnetic Resonance in Medicine, 79(6), 3055-3071. https://doi.org/10.1002/mrm.26977

7.     Gulshan, V., Peng, L., Coram, M., Stumpe, M. C., Wu, D., Narayanaswamy, A., Venugopalan, S., Widner, K., Madams, T., Cuadros, J., Kim, R., Raman, R., Nelson, P. C., Mega, J. L., & Webster D.R. (2016). Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA: The Journal of the American Medical Association, 316(22), 2402-2410. https://doi.org/10.1001/jama.2016.17216

8.     Shademan A et al. Supervised autonomous robotic soft tissue surgery . Science Translational Medicine . 2016 May 4;8(337):337ra64 . https://doi.org/10.1126/scitranslmed.aad9398

9.     Liao R et al. Artificial Intelligence–Assisted Surgery for Intracerebral Hemorrhage Evacuation: A Proof-of-Concept Study . Stroke . 2019 Oct;50(10):2825-2828 . https://doi.org/10.1161/STROKEAHA.119.026254

10.   Kowalewski TM et al . Autonomous robotic surgery using deep learning: A proof-of-concept study using a porcine model in a cadaveric setting . Surgical Endoscopy . 2020 Sep;34(9):3864-3872 . https://doi.org/10.1007/s00464-019-07257-y

11.   Turakhia MP et al . Rationale and design of a large-scale app-based study to identify cardiac arrhythmias using a smartwatch: The Apple Heart Study . American Heart Journal . 2019 Jan;207:66-75 . https://doi.org/10.1016/j.ahj.2018.09.002

12.   Norgeot B et al . Protected Health Information filter (Philter): accurately and securely de-identifying free-text clinical notes . NPJ Digital Medicine . 2019 Jun 11;2:52 . https://doi.org/10.1038/s41746-019-0128-0

13.   Rajkomar A et al . Scalable and accurate deep learning with electronic health records . NPJ Digital Medicine . 2018 May 8;1:18 . https://doi.org/10.1038/s41746-018-0029-1

14.   Zhang R et al . Deep learning for pharmacovigilance: recurrent neural network architectures for labeling adverse drug reactions in Twitter posts . Journal of the American Medical Informatics Association . 2018 Jul 1;25(7):893-901 . https://doi.org/10.1093/jamia/ocy018

15.   Kumar S et al . A diabetes mobile app with in-app coaching from automated messages improved glycemic control in a randomized controlled trial . Journal of Medical Internet Research . 2020 Feb 19;22(2):e16219 . https://doi.org/10.2196/16219

16.   Tsatsaronis G et al . An overview of the BIOASQ large-scale biomedical semantic indexing and question answering competition . BMC Bioinformatics . 2015 Apr 9;16:138 . https://doi.org/10.1186/s12859-015-0564-6

17.   Zhavoronkov A et al . Deep learning enables rapid identification of potent DDR1 kinase inhibitors . Nature Biotechnology . 2019 Oct;37(9):1038-1040 . https://doi.org/10.1038/s41587-019-0224-x

18.   Richardson P et al . Baricitinib as potential treatment for 2019-nCoV acute respiratory disease . The Lancet . 2020 Feb 15;395(10223):e30-e31 . https://doi.org/10.1016/S0140-6736(20)30304-4

19.   Gao M et al . Machine learning of molecular properties: Locality and active learning . The Journal of Chemical Physics . 2019 Mar 28;150(12):124110 . https://doi.org/10.1063/1.5086105