1 CONTEXT

Recent years have witnessed substantial efforts in furthering Team sciences.^{1, 2} In Team science, experts from multiple varied fields work on addressing a scientific challenge by leveraging their diverse strengths and expertise.³ Traditionally, the field of orthodontics is characterized by clinician scientists working in silos with a vast majority of studies originating from academic centers.⁴ There is a paucity of studies that have transcended traditionally segregated clinical specialties and scientific disciplines. Single center academic studies are well conducted but are not truly generalizable to the real world of clinical practice. One of the initiatives of NIH/NIDCR to further science in the real world of dental practice is the National Dental Practice-Based Research Network (NDPBRN).³ The NDPBRN is currently in the third cycle of funding and over 40 studies involving data gathered from over 70 000 patients have been conducted.³ The success of NDPBRN has clearly demonstrated the benefits of conducting Team sciences to further oral health research and help practitioners make every day clinical decisions.

Large scale TEAM sciences based collaborative projects that generate findings that are generalizable and can be applied to the population at large could immensely strengthen the empirical framework for clinical decision making in healthcare. However, a major barrier for conducting large-scale nationwide collaborative studies across different practice settings in the field of Orthodontics is the lack of large centralized databases that captures information from a number of practitioners across different settings drawn from all geographical locations in the United States. Nationwide collaborative initiatives would facilitate research that is generalizable and externally valid, enable orthodontists to compare their outcomes in real time with their peers, help with a data driven decision making process that will move the specialty towards achieving truly precision/personalized status, and can help us to improve the quality of care we deliver to our patients (especially with current tools that enable real time continuous feedback loops).⁵

Several barriers have precluded the development of large-scale nationwide collaborative initiatives in orthodontics. These include: costs of creating databases and registries that collates information from a large number of practices, issues with ownership of data and how it can be leveraged for research and clinical purposes, obtaining consent of patients, maintenance of data, interoperability of electronic health records, cybersecurity, lack of uniformity or consensus in data collection, etc.^5-8 Furthermore, our healthcare framework is traditionally based on an “Institutional (Practice) Centered Model” as opposed to a “Patient Centered Model”.^{9, 10}

Disruptive transformative technology such as blockchain and recent advances in artificial intelligence based tools such as machine learning and deep learning have shown much promise in medicine to enable real time large scale collaborative initiatives involving multiple stake holders (hospitals, clinics, practitioners, patients, insurance providers, etc.).^11-18 In the following sections, we will provide a high level overview of blockchain technology and how it can be leveraged with federated machine learning to help establish collaborative models for care and research in orthodontics.

2 BLOCKCHAIN TECHNOLOGY

In the traditional “Institutional Centered” collaborative healthcare models, data from a wide variety of sources (distributed health care records) are stored typically in a centralized database and are accessed for analysis.^{9, 15, 17-19} A single point of failure or a cyber-attack could lead to catastrophic consequences. Furthermore, having a centralized authority over records furthers the possibilities for data manipulation, unintended use of data, security/privacy breaches, and data leakages. Consequently, health records are not tamper resistant.^17-19 Blockchain technology is a new technological paradigm that is purported to address the trilemma of issues (lack of data neutrality, opaqueness of data/lack of transparency, and lack of an immutable audit trail) that collaborative models are typically faced with.⁵ The three major advantages amongst a host of other benefits that a blockchain technology can effectively confer to further collaborative care models include decentralization (neutrality) of data, transparency and traceability, and immutability. Consequently, blockchain technology can be used to create nationwide and even international data repositories where patient/research subject confidentiality will not be compromised while enabling multiple institutions and providers to share data.

A blockchain is a distributed ledger (database that is consensually shared and synchronized by multiple sites) and each block comprises information on all transactions that is time stamped, cryptographic hash of the current block, and a hash of the previous block.^{11-18, 20-25} Each block is connected through the hash. Each block of data is authenticated and verified and consequently is tamper proof.^20-25 Once a block is created and authenticated, it is almost impossible to change the data in the block as it would necessitate all subsequent blocks to be changed. Altering the blocks would be extremely challenging computationally and requiring enormous amounts of power. Consequently, a blockchain is immutable. Unlike the traditional databases that are used in the healthcare landscape which are controlled by one single organization, a blockchain network is a distributed consensus algorithm and there is no centralized authority.^20-25 Therefore, a single point of failure cannot happen. Each and every transaction in a blockchain can be traced with verifiable time stamps and all stakeholders that participate in the network can view the transactions.

A blockchain can be public, private or hybrid.^{24, 25} Public blockchain networks are permission less, transparent, and opensource where anyone can participate. They typically operate on behavioural economic principles. Two of the most popular public blockchain networks are Bitcoin and Etherium. On the other hand, private block chain networks require permissions to access, and only permission-enabled stakeholders can participate and these are most relevant to healthcare.^{24, 25} In hybrid blockchains, the private network is tethered to a public network and depending on the application, they can share features of both private and public blockchain networks.

It should be kept in perspective that a blockchain is not a replacement for any of the current electronic health record systems that we use (including Epic, AxiUm etc.). Instead, a blockchain is an added layer and is modular. Modularity is a key advantage of blockchain technology as it enables implementation of smart contracts.^{9, 11-18, 26} For example, smart contracts can be used for patient/research subject consensus. Smart contracts are generated when a pre-defined set of rules are met.

Blockchain technology facilitates federated machine learning which could have enormous implications for clinical care and multi-center research.^11-16 Individual clinics and practitioners can place their data (or pointers to data) on a blockchain without uploading it to a centralized data repository. Permissions to access the data are granted to participants in the blockchain network (private network). This renders the data transparent, secure, and immutable. All participants can access the data. Cryptographic encryption enables the data to be secure. For any node (participant) to change the data, it will be almost impossible as all subsequent blocks have to be changed because each block of data is connected via a cryptographic hash and time stamped. Due to all of the above factors, a blockchain can be used for federated machine learning especially when numerous participants and stakeholders are involved.

3 FEDERATED MACHINE LEARNING

Machine learning, an Artificial Intelligence based approach to examine patterns and discern outcomes has been increasingly used in orthodontics in recent years.²⁷ To date, machine learning models in orthodontics have been built from data drawn from single centers or a few centers and have a single data storage point or server. As a result, these individual machine learning models are not adequately trained to reflect practice patterns at large and are not generalizable. Furthermore, data privacy and security could be compromised. To address these concerns, Federated Machine Learning models are being developed.^28-32 Federated machine learning is a distributed and decentralized approach to analysing data wherein different clinics can contribute their data (parameters) through a blockchain network without having to upload to a central server. A shared global machine learning algorithm is then developed using parameters from different clinics.^28-32 This federated learning approach mitigates data privacy and security risks while developing a prediction model that is far superior to models developed by traditional machine learning approaches.^27-32 Several privacy-protected data sharing approaches are available to minimize data privacy and security issues. The participating clinics need not share their entire datasets but instead provide access to a partial gradient of parameters through the blockchain network.^28-32

Federated machine learning can facilitate participation of a large number of orthodontics clinics in nationwide studies which would generate findings that are generalizable. Clinicians can compare their outcomes in real time with their peers' and make changes to their clinical practice accordingly. A major issue plaguing clinical outcomes research and assessments is that our empirical framework is bereft of data from underrepresented study cohorts and tends to include homogenous study populations. The current machine learning models which are based on homogenous single center datasets can further exacerbate health disparities and racial/ethnic inequities.^{11, 16-18, 26-32} With federated machine learning, we can increase inclusion of underrepresented cohorts since data are drawn from a large number and types of clinical practice settings and heterogeneous study population groups.^{11, 16-18, 26-32} As a result better prediction models that can account for a wide range of heterogeneity can be developed to establish individualized treatment plans and assess outcomes.

4 REMOTE MONITORING

Recently, an aspect of teledentistry called remote treatment monitoring or telemonitoring has become increasingly utilized in our digital world. Telemonitoring in orthodontics is a two-way street requiring active participation from healthcare professionals and patients.³³ Remote monitoring can be used during treatment to monitor treatment and during the retention phase with previous orthodontic patients sending photos to assess the stability of their occlusion instead of travelling to the office for an in-person evaluation.^{34, 35} Telemonitoring can also mitigate emergency appointments by using a mobile phone or Internet services to determine if an in-office visit is warranted or if the patient can address the problem safely and effectively on their own.³⁶ An even more cutting-edge aspect of remote monitoring involves the use of artificial intelligence. Artificial Intelligence Driven Remote Monitoring (AIDRM) is an innovative new technology that allows patients to take photos or videos of their mouth at home that are then evaluated by an AI image analysis software.³⁷ Real time alerts can then be sent to the patient and doctor regarding the specific conditions the AI detects. AI algorithm can evaluate tooth movement, monitor oral hygiene, and detect appliance breakage.³⁸

Blockchain technology and remote dental monitoring are both powerful tools that would have significant potential to improve orthodontics and craniofacial health. By combining remote dental monitoring with blockchain technology, orthodontic practices can securely store and share patient data, including treatment plans, progress, and images. This can help orthodontists to provide more personalized and effective treatment plans and improve overall patient outcomes. Blockchain technology provides a secure and transparent way to store patient data, which can improve trust and transparency between orthodontists and their patients. In addition, remote dental monitoring and blockchain technology can also benefit craniofacial health. Patients with craniofacial conditions require regular monitoring and may need to see multiple specialists. Remote dental monitoring can reduce the need for in-person appointments, while blockchain technology can provide a secure and accessible way to share patient data between multiple healthcare providers. This can improve coordination and communication between healthcare providers, leading to better treatment outcomes for patients with craniofacial conditions.

5 GENOMICS AND PRECISION MEDICINE

The healthcare system is moving away from the old concept of “treating patients to an average” to a more modern approach of “personalized precision medicine”.⁵ This approach allows health care providers to deliver customized care in a patient centered fashion. In this context genomics technology plays a major role. The term genomics refers to the study of all the genes at once using advanced technologies such as next generation sequencing. The average cost of sequencing at the time of first human genome sequencing was ~$10 000 000, which has now significantly reduced to $1000. This falling cost of sequencing technology has greatly enabled incorporation of genomic tools in health care. For example, the application of sequencing technology in neonatal intensive care units facilitates rapid diagnosis of genetic conditions in new-born babies, which allows faster clinical management. The falling cost also has enabled the inception of direct-to-consumer companies such as 23andMe®.

This increased utilization of genomic technologies presents its own set of challenges. For example, when a private enterprise sequences a genome of an individual, it stores the coded and de identified data in a highly secured and backed-up servers owned by that enterprise and therefore the ownership falls in the hands of the enterprise. This also allows private entities to sell large scale genomic data, at a cost, to pharmaceutical companies and researchers. In this process, the actual owners of the data – the individual patients himself or herself – does not get benefitted as the companies do not provide direct access to the data nor do they share the revenue. Furthermore, as the data are housed in the servers of the private enterprise, sharing the data with other healthcare providers and researchers becomes a cumbersome task. This not only delays diagnosis and initiating treatment at the right time, but also the discovery of genetic risk for multiple diseases and the development of novel therapeutic approaches. In essence it hugely impacts the social well-being of the general population.

The blockchain technology provides many solutions to the challenges described above. However, some of the roadblocks in the application of blockchain technology in storing genomics data are the scale of the data itself, the speed of transaction, and the ability to query the data. A recent proof-of-concept study has demonstrated that storing genetic variants in variant call format and raw genomic reads are indeed feasible in a private blockchain network.³⁹ As blockchain is an immutable transaction ledger without a central authority it puts the ownership of the genomic data in the hands of the individual patients.^{24, 25} As the genomic data are contained in timestamped immutable transaction blocks, it allows distribution of data with other entities such as healthcare providers connected with that system. This would allow faster clinical utility in managing patients. Furthermore, if the individual owner chooses to sell the data to a third-party such as researchers or private companies, it cuts the middleman, and allow the individuals to reap not only financial benefit but also their contributions to accelerated scientific research. Furthermore, there are several companies already pioneering in blockchain technology in genomic data sharing and these are Nebula Genomics (USA), Zenome (Russia), Genecoin (Brasil), Gene-Chain (USA) and DNATIX (Israel).⁴⁰ The incorporation of blockchain technology in genomics will immensely change the landscape of delivery of precision healthcare as well as discovery of novel therapeutic approaches.

6 CHALLENGES FOR ADOPTION

There are several barriers to adopting blockchain technology and federated learning in orthodontics.^{4, 16, 26, 28-32} Adoption of a new technology which has shown much promise but still in infancy is likely to be a hard sell to orthodontists. The majority of orthodontists may not have the technological skills to use these new tools and may not have the time to invest to reskill themselves. Blockchain technology is evolving extremely rapidly and is a costly endeavour to implement. Consequently, traditional small-scale orthodontic clinics may not have the resources to use this technology. Scalability is an issue that has been reported as a major drawback of blockchain technologies because of the slow transaction processing times. There are multiple “implementation bottlenecks” to consider for using federated learning.^28-32 For example, the interoperability of various informational technology infrastructure specifications and electronic record systems that are in vogue in different clinics and how they will interact with each other is a key construct to take into consideration prior to implementation.^{26, 28-32} Participating clinics and practitioners should be able to seamlessly integrate and use the federated machine learning models in their daily clinical practice routine. The “entry point” for use of these technologies by practitioners should be as easy as possible but this is seldom the case because of the steep learning curve.^{26, 28-32} Finally, we have to consider the potential impact of blockchain technology and federated learning on further exacerbating existing health disparities due to different clinics and practitioners having lack of equitable access to technological skills and infrastructural resources. We could inadvertently further digital divide.^{14, 16, 26, 28-32}

7 CONCLUSIONS

Blockchain technology, a distributed consensus algorithm is an emerging technology that has the potential to facilitate sharing of data across multiple organizations and stakeholders (including clinicians, practitioners, insurance providers etc.). It is pivotal for implementation of federated machine learning which is key to furthering precision/personalized care and improving clinical outcomes. Federated machine learning can be used to develop robust models for developing individualized treatment plans and assessing outcomes in real time. At the present time there are multiple challenges for adoption. However, as technology evolves we anticipate blockchain networks to become more scalable, less costly, and more easy to adopt. We hope that in the near future, a new paradigm of collaborative care that puts the patient at the center of care will emerge.

AUTHOR CONTRIBUTIONS

Veerasathpurush Allareddy: Study idea, project management, securing funding, drafting of manuscript, and final approval of manuscript. Sankeerth Rampa: Study idea, drafting of manuscript, and final approval of manuscript. Shankar Rengasamy Venugopalan: Drafting of manuscript, and final approval of manuscript. Mohammed H. Elnagar: Drafting of manuscript and final approval of manuscript. Min Kyeong Lee: Securing funding, drafting of manuscript, and final approval of manuscript. Maysaa Oubaidin: Securing funding, drafting of manuscript, and final approval of manuscript. Sumit Yadav: Drafting of manuscript and final approval of manuscript.

FUNDING INFORMATION

Dr. Allan G. Brodie Craniofacial Chair Endowment. Dr. Robert and Donna Litowitz Fund. American Association of Orthodontists Foundation.