NLP PEOPLE PRIMER
1. ACOs
The affordable care act is not just a new way to get insurance and also a short of affordable care. The ACOs is short for accountable care organizations. ACO is the network of doctors and hospitals that shares a financial and medical responsibility for patients. The goal is to coordinate care and eliminate unnecessary spending. In the healthcare system today, patients are usually responsible for coordinating their medical care. Someone with heart disease may see a primary care doctor, a cardiologist, maybe even a heart surgeon. But the doctors may not talk much so they could order repeated tests or prescribe conflicting drugs. That is not good for the patients, and it is expensive. In ACO, doctors, imaging specialists, surgeons, and hospitals are all working together, sharing information to figure out the best way to fix the patient up and keep healthy afterward. MSSP ACOs notably earned a mean quality score of 90.5 percent under pay-for-performance measurements in 2017, CMS reported. Comparing ACO spending to the absence of ACOs found this important Medicare program lowered Medicare spending by $3.53 billion from 2013 to 2017 and saved $755 million after paying shared savings.
1.1 How did ACO pay?
One of the main ways the Affordable Care Act seeks to reduce health care costs is by encouraging doctors, hospitals and other health care providers to form networks that coordinate patient care and become eligible for bonuses when they deliver that care more efficiently. In Medicare’s traditional fee-for-service payment system, doctors and hospitals generally are paid for each test and procedure. ACOs don’t do away with fee for service, but they create an incentive to be more efficient by offering bonuses when providers keep costs down. Doctors and hospitals have to meet specific quality benchmarks, focusing on prevention and carefully managing patients with chronic diseases.
1.2 Difference between HMO and ACO
Stakeholders have likened ACOs to the health maintenance organizations (HMOs) that were popular in the 1990s. An HMO is a type of insurance plan that limits coverage to services performed by providers who work for or contract with the plan. HMO providers generally receive a fixed monthly or annual payment to deliver care to beneficiaries. When all the parts work together, providers in an ACO can bring down costs and improve care quality while earning incentive payments. HMOs, on the other hand, seek to cut costs by setting fixed prices for services, which may encourage providers to reduce utilization or skimp on care in an effort to stay under the cap.
1.3 Covid-19 pandemic on ACO:
In early April, the National Association of ACOs (NAACOS) emailed all Medicare Shared Savings Program (MSSP) and Next Generation ACO Model participants, including NAACOS members and nonmembers, with a request to complete an online survey focused on the effects of COVID-19 on ACOs. There were 304 responses from 226 ACOs across the country. The results of that survey send a clear and strong message that ACOs are very concerned about the effects of COVID-19 on their ACO. (Figures credit to NAACOS's report.)
​
Fig 1. A vast majority of ACOs are very concerned about the effect of COVID-19 on their ACO
Fig 2. A large portion of risk-based ACOs are likely to quit over concerns about COVID-19
Fig 3. ACOs anticipate notable increases in 2020 Medicare spending from COVID-19
2. PCMH
The Agency for Healthcare Research and Quality (AHRQ) recognizes that the patient centered medical home (PCMH) is a promising model for transforming healthcare organizations in achieving high-quality, accessible, efficient health care. The key building blocks and functionalities include:
​
(1) Comprehensive Care
The PCMH is supposed to provide a majority of services for meeting patient’s physical and mental health needs, by assembling a professional team with physicians, nurses, assistants, pharmacists, nutritionists, social workers, and care coordinators. For smaller practices, virtual teams linking themselves to providers and services in the community is a way to address the issue of inefficient resources.
Further reading: Ensuring That Patient-Centered Medical Homes Effectively Serve Patients With Complex Health Needs
(2) Patient-Centered
A key functionality of PCMH is patient-centered care, where the delivery of healthcare is oriented towards the patient. This could be achieved by practice’s collectively working with patients and their families to understand patients’ needs, culture, values, and preferences.
Further reading: AHRQ Health Literacy Universal Precautions Toolkit
​
(3) Coordinated Care
PCMH has the ability of coordinating patient care across all elements of the healthcare system, such as specialty care, hospitals, home health care, and community services. The transition of healthcare information is the key in this step, where high-quality and efficient transition of data would ensure that patient will get timely treatment in other places. And by working with home care IT, more comprehensive data could be collected for coordinating care with different physicians.
Further reading: Care Management: Implications for Medical Practice, Health Policy, and Health Services Research
(4) Accessible Services
Accessibility in PCMH focuses on minimizing waiting time, enhancing in-person visit, and easy access through telemedicine.
(5) Quality and Safety
Most important of all, the quality and safety should still be a key factor in PCMH, as in other healthcare models. PCMH is trying to achieve this through implementation of clinical decision support tools, shared decision-making system, evidence-based care, performance measurement, and population health management.
Further reading: Engaging Primary Care Practices in Quality Improvement: Strategies for Practice Facilitators
​
What are we seeing in 2020?
The National Committee for Quality Assurance (NCQA) promoted PCMH among primary care practices since 2008, in hope of transforming them into medical homes to improve quality of healthcare. Before 2019, CMS released its “final evaluations” on all three medical home demos, including PCMH, all of which were announced to be failure. The three demos are: the Federally Qualified Health Centers Advanced Primary Care Practice Demonstration (Nov 2011 to Nov 2014), Multi-payer Advanced Primary Care Practice demo (started in Oct 2012), and the Comprehensive Primary Care Initiative demo (Oct 2012 - Oct 2016). The evaluations of the three demos were finished by different organizations, with similar conclusion that PCMH raised total spending. Another study in 2014 on NCQA PCMH recognition level showed that even among innovative practices, the NCQA PCMH recognition level was still only 40%, and many practices questioned whether PCMH recognition was a meaningful credential.
Interestingly, as addressed by NCQA later for the criticism they received, they gave out their justification that such small time frames (1-2 years) would not usually be sufficient for a full transformation for clinics into PCMH model. I personally think this is a good point, but that’s all they gave.
​
3. Future of the Patient-Centered Medical Home In Regards to Health Information Technology
3.` The future of Health Information Technology (HIT) in the Patient-Centered Medical Home (PCMH) is integral to the future of the PCMH. In the figure below we see that in the PCMH requires that records and technology combine to inform the Primary Care Physician (PCP) and Specialists in order to best take care of the patient. While the PCMH doesn't pertain to a physical location, it does imply that the patient is the center of care. Both PCP's and specialists need to work together in order to maximize health for the individual patient. This requires communication with between the providers and the increased use of the EHR.
3.2 Bates and Bitton elucidate how the EHR can be used in the PCMH in their paper The Future of Health Information Technology in the Patient-Centered Medical Home. The figure below illustrates the 7 key issues they see the EHR playing in the future of the PCMH. They state that all the issues can be improved to benefit the PCMH. However, not all issues are equally developed. They state that, "with respect to these seven areas, we believe that today’s electronic health records perform most poorly in the domains of team care and care transitions, for which almost no functionality is in place. Major work is also needed to improve clinical decision support, especially for chronic diseases, registries, and measurement. By comparison, personal health records and telehealth technologies are reasonably well developed.(Bates and Bitton)"
3.3 NLP in the Future of the PCMH
Natural Language Processing (NLP) will primarily be integrated into the PCMH through the implementation of clinical decision support (CDS). One way the article mentions to integrate CDS into the EHR is the reduction of adverse drug events. This is a subject that trails into computational linguistics. The vectorization of MEDLINE literature through skip-gram models that are very similar to the one's implemented by the software word2vec. One paper I'd like to talk in detail about it the Mower et al. paper titled Classification by Analogy: Using Vector Representations of Implicit Relationships to Identify Plausibly Causal Drug/Side-effect Relationships.
In this paper, Mower et al. attempts to solve the problem of picking out adverse drug events (ade's) from noisy data that would otherwise be time intensive for manual review. The primary method Mower et al. uses is called Literature Based Discovery (LBD). This entails taking the MEDLINE literature and turning them into vectors. This takes into account the co-occurence of words in the MEDLINE literature. Not only does it take into account the co-occurence but it also "exploits implicit relationships between biomedical entities" such as predicates. It builds on past work on this done by Cohen et al. in 2010 titled EpiphaNet: An Interactive Tool to Support Biomedical Discovers which is a graphical tool that encompasses the relationships between drugs, predicates, and side-effects.
Mower et al. takes these implicit relationships between side-effects and creates "bound vectors". Bound vectors are the bound product of the random vectors for the predicate and argument of each predication it occurs in. The example Mower et al. uses is "the predication 'ibuprofen-TREATS-pain' would be encoded into the semantic vector for the concept ibuprofen by superposing the bound products of the random vectors for TREATS and pain." Once these bound vectors are created they are fed into a machine learning model to predict whether they are a positive drug/side-effect relationship or a negative drug/side-effect relationship. The type of model used by Mower et al. is a K-nearest neighbors classification model. This model works well with vectors; it takes into account the label (ground truth positive or negative) of the closest n vectors and uses that as the predicted label for the drug/side-effect bound vector in question. They used 5, 10 and 15 for the values of n, but 5 performed best. They also used logistic regression and support vector models. Mower et al. found found that both SVM and LR had 100% accuracy when trained on the full set and there was an F1 measure of .80 with cross-validation. All in all, Mower et al. proved that a combination of machine learning and computational linguistics can predict drug/side-effect relationships that can be translated into the clincial setting.
References
Hahn, K. A., Gonzalez, M. M., Etz, R. S., & Crabtree, B. F. (2014). National Committee for Quality Assurance (NCQA) patient-centered medical home (PCMH) recognition is suboptimal even among innovative primary care practices. The Journal of the American Board of Family Medicine, 27(3), 312-313.
AHQR: https://pcmh.ahrq.gov
Patient-Centered Medical Home. NCQA. https://thehealthcareblog.com/blog/2018/06/07/the-verdict-is-in-all-three-of-cmss-medical-home-demonstrations-have-failed/
The verdict is in: All three of CMS’s “medical home” demonstrations have failed. https://thehealthcareblog.com/blog/2018/06/07/the-verdict-is-in-all-three-of-cmss-medical-home-demonstrations-have-failed/
(The Future of Health Care: PCMH, n.d.) https://healthcareplainandsimple.com/understand-health-care/the-future-of-health-care-pcmh.aspx
Bates, D. W., & Bitton, A. (2010). The Future Of Health Information Technology In The Patient-Centered Medical Home. Health Affairs, 29(4), 614–621. https://doi.org/10.1377/hlthaff.2010.0007
Mower, J., Subramanian, D., Shang, N., & Cohen, T. (2016). Classification-by-Analogy: Using Vector Representations of Implicit Relationships to Identify Plausibly Causal Drug/Side-effect Relationships. AMIA ... Annual Symposium Proceedings. AMIA Symposium, 2016, 1940–1949.
Cohen, T., Whitfield, G. K., Schvaneveldt, R. W., Mukund, K., & Rindflesch, T. (2010). EpiphaNet: An Interactive Tool to Support Biomedical Discoveries. Journal of Biomedical Discovery and Collaboration, 5, 21–49.