>Business >ML in Healthcare – Twelve use-cases Part 5

ML in Healthcare – Twelve use-cases Part 5

In the previous part of this blog series, we took a look at risks and compliance considerations that are particular to AI/ML within the healthcare space. The fifth part of this blog series will go further into these risks and compliance considerations.

Risks and compliance considerations that are specific to AI/ML in healthcare

  • Contractual Risks

Contractual risks within the healthcare space are primarily in relation to the administration of the contract lifecycle. It was documented that it took 92-minutes for a human agent to review five legal documents in one of the latest researchers carried out by Stanford University, USC, and Duke University School of Law. On top of this, artificial intelligence AI was observed to demonstrate 10% increased precision in its review.

One can only imagine the massive influx of corporate agreements that any specific health care system is needed to handle. Ideally, to stay compliant with each and every contract, regardless of whether it is a physician affiliation, lease agreement, a contracted staff member, or a purchase contract – health care enterprises should review them on an annual basis. Owing to the massive volume and intricacy, carrying out a yearly review like this is considered to be uncommon. Currently, artificial intelligence and machine learning technologies are here to lend a helping hand.

A few instances of contract lifecycle administration that can reap advantages from the harnessing of artificial intelligence and machine learning technologies include the following:

  • Reviewing multiple contracts that are in effect with the same entity to reveal any incongruencies.
  • Undertake review of contracts to stay compliant with federal or multi-state regulations
  • Flag lease contracts that are due to be renewed
  • Review contracts regarding physician privileges due to be renewed and updated
  • Cross-reference physician qualifications to remain compliant with the Stark law.

It would be a lot better off if 5,000 contracts could be reviewed within the span of 7-days leveraging artificial intelligence and ML technologies, instead of redirecting 20 full-time staff members to undertake manual review of them over the course of numerous months. Or probably you have acquired, intend to partner, or have merged with a facility where a massive number of agreements require review for assessing risk. How much simpler would this procedure/process be with artificial intelligence to flag at-risk areas.

With the help of Artificial Intelligence to undertake review of documentation for routine information, human time and expertise can be harnessed in better ways on more critical tasks that actually need the touch of a human agent and their decision-making capacity.

  • Data privacy – compromises and breaches

Having access to massive volumes of high-quality is presently considered to be one of the biggest restrictions for ML tech within the healthcare space. Health information is to be treated with the utmost sensitivity, mandating increased regulations. This section of the blog looks into how this hurdle can be surpassed by harnessing federated learning, an ML technology that considerably minimizes privacy concerns by storing patient information recorded in a secure manner onsite over the course of model training.

The possibilities for ML in healthcare space is presently restricted by basic data access hurdles, like ‘siloed’ information storage within several differing hospitals. Worries regarding inadequate transparency of machine learning systems and unsatisfactory privacy settings to safeguard critically sensitive information and data makes accessing big data a stressful endeavour.

Federated learning is on course to ‘the next big thing’ in progressing medical research with no compromises to health data privacy. It facilitates a machine learning algorithm to learn on differing datasets without deleting the data from where they are recorded. Research and healthcare facilities maintain control with regards to data governance and GDPR compliance. Extra privacy-maintaining measures like differential privacy and secure aggregation facilitate for novel ways to safeguard data in FL.

  • Data utilization issues

Fragmentation

This is the dominant issue that requires to be tackled if EHRs have any hope of being utilized in any worthwhile clinical capacity. Fragmentation crops up when EHRs are not able to communicate in an efficient manner amongst each other – essentially locking patient data into a proprietary framework. While there are dominant players within the EHR domain in the United States of America, a few instances including entities like General Electric and Epic, there are additionally a plethora of lesser and niche enterprises that also generate their own products – several of those are unable to communicate in an effective and efficient fashion with each other.

The Clinical Oncology Requirements with regards to the EHR and the National Community Cancer Centres Program have both talked about the requirement for interoperability requirements with regards to EHRs and even released guidelines. On top of this, the Certification Commission with regards to Health Information Technology was developed to produce guidelines and standardize interoperability with regards to EHRs. Fast Healthcare Interoperability Resources (FHIR) is the present new standard with regards to data exchange put out by Health Level 7 (HL7). It develops upon historical standards from both HL7 and a plethora of other standards like the Reference Information Model. FHIR provides fresh principles upon which data sharing can occur via RESTful APIs – and projects like Argonaut are moving towards expansion of adoption to EHRs.

Data Ownership

One of the pressing hurdles with regards to present deployments of Big Data are the lacking presence of regulatory standards, systems, and incentives to handle ownership and accountabilities for data and information. Within the healthcare and clinical domain, specifically in the United States of America, this takes on the shape of compliance with HIPAA, a currently 10-year old law that intended to establish rules with regards to patient privacy and control for data.

As more variants of data are produced for patients and uploaded to digital platforms, HIPAA turns a dominant roadblock to information sharing as it gives room to considerable privacy considerations that hinder research progress. Currently, if a researcher is to for even a simple thing, such as disease and demographics statistics, they can quickly detect an otherwise de-identified patient.

Considerations with regards to breaking HIPAA prevent total and open data sharing agreements – blocking a bath to the specificity required for the next gen of research from being accomplished, and also throws a wrench into clinical deployment of these tech as information sharing is held back by nebulousness with regards to legacy regulations about patient privacy. Further, staying compliant with the General Data Protection Regulation (GDPR) within the European Union has negatively impacted global collaborations as compliance with regards to both GDPR and HIPAA is pending to be standardized.

Data sharing is further made intricate by the requirement to produce novel, and innovative technologies for the purposes of cross-provider integrations. Taking from the instance of the Informatics for Integrating Biology and the Bedside (i2b2) program backed by the National Institute of Health, it is tough, and incredibly expensive to overlay programs atop current EHRs. Instead a new strategy is required to be produced to find a solution to the issue of data sharing.

Blockchain furnishes a novel strategy and has lately been looked into the literature as a potential solution that centres patient control of their data, and additionally promotes safety and security in data sharing via data transfer transactions secured through encryption.

Conclusion

The fifth part of this blog series on ML within healthcare went deeper into risks and compliance considerations. The next part will wrap up the compliance discussion, speak about prospective use cases and issues ML can address, and also the outlook for the technology within the healthcare space.

Add Comment