Harnessing AI and ML Algorithms for Value-Based Care Models in Healthcare
Authors: Jessa Deckwa, BS, Mayur Bhakta, MD, & Zain Khalpey, MD, PhD, FACS
Abstract
The rapid advancements in artificial intelligence (AI) and machine learning (ML) have the potential to revolutionize healthcare delivery by enabling value-based care models that prioritize patient outcomes and cost efficiency. This white paper examines the role of AI and ML in enhancing value-based care models, providing examples of their applications, and discussing the challenges and future prospects. The paper demonstrates that AI and ML algorithms can contribute meaningfully to value-based care models by improving patient outcomes, reducing costs, and increasing operational efficiency.
Introduction
Healthcare systems worldwide have been transitioning from fee-for-service models to value-based care models, which focus on improving patient outcomes while reducing costs. This shift is driven by the need to address the rising costs of healthcare and the growing prevalence of chronic diseases. AI and ML algorithms have the potential to contribute significantly to the success of value-based care models by enhancing decision-making, streamlining operations, and personalizing care.
1. AI and ML Applications in Value-Based Care Models
1.1 Predictive Analytics and Risk Stratification
AI and ML algorithms can be used to analyze large volumes of patient data and predict outcomes, enabling healthcare providers to identify high-risk patients and deliver targeted interventions. By proactively identifying at-risk patients, healthcare organizations can focus their resources on those who need them most, ultimately improving patient outcomes and reducing costs.
One way AI and Machine Learning Applications can improve the outcomes of patients is to provide safety upon discharge from the hospital. Hospital readmissions can be defined as a subsequent hospital admission for any cause within 30 days following an initial stay (index admission) according to the National Health Insurance Administration (2). One study wanted to determine if hospital admissions were explainable, a machine learning algorithm was developed to predict a 14 day unplanned readmission. The study reviewed 24,722 patients with a readmission rate of 1.22%(1). Patients who were aged < 20 years, who were admitted for cancer-related treatment, who participated in pharmaceutical clinical trial, who were discharged against medical advice, who died during admission, or who lived abroad were excluded from the study(1). Among the 4 machine learning algorithms selected, Catboost had the best average performance in five fold cross-validation (precision: 0.9377, recall: 0.5333, F1-score: 0.6780, AUROC: 0.9903, and AUPRC: 0.7515). After incorporating 21 most influential features in the Catboost model, its performance improved (precision: 0.9470, recall: 0.5600, F1-score: 0.7010, AUROC: 0.9909, and AUPRC: 0.7711)(1). ML prediction models can help clinicians to accurately identify patients likely to experience early unplanned readmission.
1.2 Personalized Medicine and Treatment Optimization
AI and ML can facilitate the development of personalized medicine by analyzing complex patient data and identifying the most effective treatments for individual patients. This approach can lead to better patient outcomes and cost savings by minimizing trial-and-error and reducing the need for expensive, ineffective treatments.
Using a patient’s electronic medical record has proven to accurately predict patient outcomes such as mortality, readmissions, and length of stay. One study demonstrated using de-identified electronic health record data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 h. In the sequential format we propose, this volume of EHR data unrolled into a total of 46,864,534,945 data points, including clinical notes (3). Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality (area under the receiver operator curve [AUROC] across sites 0.93-0.94), 30-day unplanned readmission (AUROC 0.75-0.76), prolonged length of stay (AUROC 0.85-0.86), and all of a patient’s final discharge diagnoses (frequency-weighted AUROC 0.90) (3). These models outperformed traditional, clinically-used predictive models in all cases.
This study indicated a different data source (the electronic health record) can personalize healthcare at discharge, allowing a clinician to provide additional resources for high risk patients. Another finding from this study is what we use to formulate our databases for machine learning and AI applications can significantly impact how we treat patients in the future. With data continually at our fingertips and improvements in medical technology every day the ability of a clinician to apply all data points is seemingly impossible. Using ML or AI applications can help reduce the noise at the bedside and allow clinicians to treat not only the symptoms and disease of a patient at present, but also impact the risk of disease predicted prior to discharge.
1.3 Image Analysis and Diagnosis
AI and ML algorithms have shown great potential in enhancing the accuracy and speed of image analysis in various diagnostic procedures, such as radiology, pathology, and dermatology. Improved image analysis can lead to more accurate diagnoses, faster treatment initiation, and better patient outcomes.
One demonstrated that CNN modeling utilizing deep neural networks could achieve dermatologist-level accuracy in classifying skin cancer using images of skin lesions. The test was performed against 21 board-certified dermatologists on biopsy-proven clinical images with two critical binary classification use cases: keratinocyte carcinomas versus benign seborrheic keratoses; and malignant melanomas versus benign nevi (4). This application of AI and ML has the potential to improve the accuracy of skin cancer diagnoses, expedite treatment initiation, and ultimately improve patient outcomes.
The future of AI integrated applications could be extended to mobile device applications. If these applications are outfitted with deep neural networks, mobile devices can potentially extend the reach of clinicians outside of the clinic. It is projected that 6.3 billion smartphone subscriptions will exist by the year 2021 (5) and can therefore potentially provide low-cost universal access to vital diagnostic care.
1.4. Natural Language Processing for Clinical Decision Support
Natural language processing (NLP), a subfield of AI, can be employed to process and analyze unstructured text data in electronic health records, clinical notes, and medical literature. NLP can enhance clinical decision support by providing evidence-based recommendations to healthcare providers, improving the quality of care and reducing unnecessary medical interventions.
One way to apply NLP AI in healthcare is utilizing the “human input” from the free text communications documentation regarding diagnostic imaging stored in the electronic health record for analysis. Although radiology reports are more commonly stored for communication and documentation of diagnostic imaging, harnessing their potential requires efficient and automated information extraction: they exist mainly as free-text clinical narrative, from which it is a major challenge to obtain structured data (6).
This systematic review highlights the potential of NLP techniques in radiology for extracting clinically relevant information from free-text radiology reports. The application of NLP can enhance clinical decision-making by providing timely access to critical patient data and evidence-based recommendations.
2. Challenges in Implementing AI and ML in Value-Based Care Models
Recent examples have demonstrated that big data and machine learning can create algorithms that perform on par with human physicians (15). Though machine learning and big data may seem mysterious at first, they are in fact deeply related to traditional statistical models that are recognizable to most clinicians. Despite their potential to revolutionize healthcare, the adoption of AI and ML algorithms in value-based care models faces several challenges, including:
2.1. Data Quality and Privacy
The success of AI and ML algorithms relies on the availability of high-quality, representative data. Ensuring data quality and protecting patient privacy are essential aspects of implementing AI and ML in healthcare. Development of AI applications may require updates to privacy and confidentiality laws and regulations, which vary widely. In the U.K., protection of health information centers on obtaining explicit consent from the patient in order to share information with any third party that is not in a direct care relationship with the patient. Researchers must apply to the Health Research Authority’s Confidentiality Advisory Group (CAG) for approval to access confidential patient information without patients’ consent (8). The U.K. consent requirement, and the definition of a “direct care relationship,” was challenged in 2017 in a published case study. The case study alleged that a technology company, Google DeepMind, did not have a direct patient care relationship with every patient included in the data shared and thus “held data on millions of Royal Free patients and former patients since November 2015, with neither consent, nor research approval (9).
In the U.S. the Privacy and confidentiality of protected health information are addressed in the Health Insurance Portability and Accountability Act (HIPAA) provides data privacy and security provisions for safeguarding medical information and allows for sharing protected health information without patient consent specifically for the purposes of “treatment, payment and operations (9).” The U.S. Subcommittee on Information Technology recommends that federal agencies conduct such a review and, where necessary, update existing regulations to account for the addition of AI (10). How the U.K. or U.S. approaches will be interpreted in cases related to data sharing for AI development is largely undetermined.
In addition to privacy and protection, there is also the question of liability and accountability for medical decisions based on AI applications. To advance deployment and acceptance of AI applications, developers will need to be able to produce the algorithm for inspection, support why the algorithm works, and ensure the application can meet expected outcomes in testing or certification procedures (7). If these initial steps are followed it will be a step in the right direction for clinical use of an AI algorithm recommendation of medical decision making, where the clinician or physician is liable and accountable for the treatment of a patient.
2.2. Algorithm Bias
Algorithmic bias can lead to disparate outcomes for different patient populations. It is crucial to ensure that AI and ML algorithms are developed and validated using diverse and representative patient data to avoid perpetuating existing healthcare disparities.
A study reviewed potential bias in AI algorithm development in regards to contact mapping with CVOID-19. The analysis of previous literature shows that the main sources of biases identified in both triage or PRP and DCT AI systems for COVID-19 are mainly related to data source variability and inadequate data collection (11). The main biases are related to data collection and management. Ethical problems related to privacy, consent, and lack of regulation have been identified in contact tracing while some bias-related health inequalities have been highlighted. There is a need for further research focusing on social determinants of health and these specific AI apps.
2.3. Integration with Existing Healthcare Systems
Integrating AI and ML technologies with existing healthcare systems and workflows can be challenging due to technical, organizational, and cultural barriers. Collaboration between healthcare professionals, administrators, and technologists is necessary to ensure seamless integration and maximize the benefits of AI and ML.
User satisfaction/acceptance refers to user perceptions about system output (12). Validated satisfaction scales miss indicators for trust and competence of the personnel to deal with the decisions made by AI (e.g. to identify false results, and to record override and its contemporaneous justification) (13). The competence of professionals to use commonly agreed documentation standards, structures, and classifications may also be essential, since it impacts the quality of documentation (input for AI). From the viewpoint of patients, barriers to use, including trust and difficult instructions are likely to become increasingly important with growing numbers of AI implementations and wider settings. User competence to evaluate decisions offered by AI would gain crucial importance in future.
3. Future Prospects
As AI and ML technologies continue to evolve, their applications in value-based care models will likely expand. Future advancements may include:
3.1. Real-time Monitoring and Decision Support
Wearable devices and the Internet of Medical Things (IoMT) can facilitate real-time monitoring of patients’ health and provide decision support to healthcare providers, enabling more timely and personalized interventions. Clinical medicine has always required doctors to handle enormous amounts of data, from macro-level physiology and behavior to laboratory and imaging studies and, increasingly, “-omic” data (16). The ability to manage this complexity has always set good doctors apart. Machine learning will become an indispensable tool for clinicians seeking to truly understand their patients. As patients’ conditions and medical technologies become more complex, its role will continue to grow, and clinical medicine will be challenged to grow with it.
3.2. AI-Enhanced Telemedicine
AI and ML can enhance telemedicine by improving remote monitoring, diagnosis, and treatment planning, allowing patients to receive high-quality care regardless of their location. In medicine, this is beginning to have an impact at three levels: for clinicians, predominantly via rapid, accurate image interpretation; for health systems, by improving workflow and the potential for reducing medical errors; and for patients, by enabling them to process their own data to promote health (14). Empowering patients with accessibility to and ownership of their own medical data reverses the predominantly one-way dynamic of today’s health care system (19).
3.3. Interoperability and Data Sharing
Improved data sharing and interoperability between healthcare organizations can enhance the effectiveness of AI and ML algorithms by providing a more comprehensive view of patients’ health data. Current prognostic models (e.g. the Acute Physiology and Chronic Health Evaluation [APACHE] score and the Sequential Organ Failure Assessment [SOFA] score) are restricted to only a handful of variables, because humans must enter and tally the scores (16). In order to work well, AI systems need to be trained (continuously) by data from clinical studies. However, once an AI system gets deployed after initial training with historical data, continuation of the data supply becomes a crucial issue for further development and improvement of the system (17).
The current healthcare environment does not provide incentives for sharing data on the system. In order to change the environment and motivate payers, mostly insurance companies, a shift from rewarding the physicians from treatment volume to the treatment outcome has begun. Furthermore, the payers also reimburse for a medication or a treatment procedure by its efficiency. Under this new environment, all the parties in the healthcare system, the physicians, the pharmaceutical companies and the patients, have greater incentives to compile and exchange information (18).
Conclusion
AI and ML algorithms have the potential to significantly contribute to the success of value-based care models by enhancing patient outcomes, reducing costs, and increasing operational efficiency. By addressing the challenges associated with implementing these technologies and fostering a collaborative environment among healthcare professionals, administrators, and technologists, AI and ML can revolutionize healthcare and improve the quality of care for patients worldwide.
References
1. Lo, YT., Liao, J.C., Chen, MH. et al. Predictive modeling for 14-day unplanned hospital readmission risk by using machine learning algorithms. BMC Med Inform Decis Mak 21, 288 (2021). https://doi.org/10.1186/s12911-021-01639-y
2. Weiss AJ, Jiang HJ. Overview of Clinical Conditions With Frequent and Costly Hospital Readmissions by Payer, 2018. 2021 Jul 20. In: Healthcare Cost and Utilization Project (HCUP) Statistical Briefs [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2006 Feb-. Statistical Brief #278. Available from: https://www.ncbi.nlm.nih.gov/books/NBK573265/
3. Rajkomar A, Oren E, Chen K, et al. Scalable and accurate deep learning with electronic health records. NPJ Digit Med. 2018;1:18. Published 2018 May 8. doi:10.1038/s41746-018-0029-1
4. Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118.
5. Cerwall, P. & Report, E. M. Ericssons mobility report https://www.ericsson.com/res/docs/2016/ericsson-mobility-report-2016.pdf (2016)
6. Pons E, Braun LM, Hunink MG, Kors JA. Natural Language Processing in Radiology: A Systematic Review. Radiology. 2016;279(2):329-343. doi:10.1148/radiol.16142770
7. Stanfill MH, Marc DT. Health Information Management: Implications of Artificial Intelligence on Healthcare Data and Information Management. Yearb Med Inform. 2019 Aug;28(1):56-64. doi: 10.1055/s-0039-1677913. Epub 2019 Aug 16. PMID: 31419816; PMCID: PMC6697524.
8. NHS Health Research Authority. Confidentiality Advisory Group: NHS; 2018 [cited 2018 December 14]. Available from:https://www.hra.nhs.uk/approvals-amendments/what-approvals-do-i-need/confidentiality-advisory-group/
9. U.S. Department of Health & Human Services. The HIPAA Privacy Rule: HHS; 2015 [cited 2018 December 14]. Available from:https://www.hhs.gov/hipaa/for-professionals/privacy/index.html
10. Hurd H, Kelly R.Rise of the Machines: Artificial Intelligence and its Growing Impact on U.S. Policy.https://oversight.house.gov/wp-content/uploads/2018/09/AI-White-Paper.pdf: Subcommittee on Information Technology Committee on Oversight and Government Reform; 2018
11. Delgado J, de Manuel A, Parra I, Moyano C, Rueda J, Guersenzvaig A, Ausin T, Cruz M, Casacuberta D, Puyol A. Bias in algorithms of AI systems developed for COVID-19: A scoping review. J Bioeth Inq. 2022 Sep;19(3):407-419. doi: 10.1007/s11673-022-10200-z. Epub 2022 Jul 20. PMID: 35857214; PMCID: PMC9463236.
12. DeLone W H, McLean E R. Information systems success: The quest for the dependent variable. Inf Syst Res. 1992;3(01):60–95.
13. Magrabi F, Ammenwerth E, McNair JB, De Keizer NF, Hyppönen H, Nykänen P, Rigby M, Scott PJ, Vehko T, Wong ZS, Georgiou A. Artificial Intelligence in Clinical Decision Support: Challenges for Evaluating AI and Practical Implications. Yearb Med Inform. 2019 Aug;28(1):128-134. doi: 10.1055/s-0039-1677903. Epub 2019 Apr 25. PMID: 31022752; PMCID: PMC6697499.
14. Topol, E. J. (2019). High-performance medicine: the convergence of human and artificial intelligence. Nature Medicine, 25(1), 44-56.
15. Beam, A. L., & Kohane, I. S. (2018). Big data and machine learning in health care. JAMA, 319(13), 1317-1318.
16. Obermeyer, Z., & Emanuel, E. J. (2016). Predicting the future – big data, machine learning, and clinical medicine. New England Journal of Medicine, 375(13), 1216-1219.
17. Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., … & Wang, Y. (2017). Artificial intelligence in healthcare: past, present and future. Stroke and Vascular Neurology, 2(4), 230-243.
18. Kayyali B, Knott D, Kuiken SV. The big-data revolution in US health care: accelerating value and innovation. 2013. http://www.mckinsey.com/industries/healthcare-systems-and-services/our-insights/the-big-data-revolution-in-us-health-care (accessed 1 Jun 2017).
19. Steinhubl, S. R., Muse, E. D., & Topol, E. J. (2015). The emerging field of mobile health. Science Translational Medicine, 7(283), 283rv3-283rv3.
