Last week, a group of five prominent practitioners and data science experts including Dr. Leo Anthony Celi, Associate Professor of Medicine (part-time) at Harvard Medical School and Principal Research Scientists at the Massachusetts Institute of Technology, who spoke at the recent AIMed webinar: Key trends in Critical Care AI, published an article in the British Medical Journal, urging hospitals to start clinical artificial intelligence (AI) departments to support related development for medical research and improvement of patient care.

A new proposal rooted in medical history

The authors began by highlighting a burning concern: will the implementation and usage of AI in clinical settings be as disappointing as the electronic health records (EHRs)? Indeed, EHRs were introduced to facilitate storage and sharing of patients’ information and in turn, improve care management and quality in the long run. Yet, over the years, many medical staff criticized EHRs for increasing their workload and undermine their time spent with patients, to the point that some detested having computers in their practice.

Conversely, AI has enjoyed some preliminary success in certain medical sub-specialties such as radiology and laboratory medicine. The authors believed it was probably due to all “relevant stakeholders” being housed under a single department. They explained the approach is not new; in fact, it was deeply rooted in medical history. When X-ray was invented in 1890, one of its creators had informally gathered interested clinicians to adopt the technology and this eventually led to the establishment of the radiology department.

Eventually, the department turns into a melting pot for different specialists to share their expertise and bring the technology forward. As such, the authors thought a clinical AI department may also encourage interested clinical staff to take the lead; so that they will not have to rely on external third-party developers which often lead to technical silos. These isolated algorithms, as the authors noted, tend to be trained using respective pools of health data and seldom undergo prospective evaluations on patients.

Even if some of them had done so, results are usually not imposing. However, most of them are still being marketed as game-changing, safe, and ready-to-use. The authors argued “a lack of coherence, leadership, and vision” is to be blamed. Thus, a change is inevitable especially if we do not wish to see AI following the fate of EHRs: bringing mainly corporate and administrative benefits but not having a real change for the patients and the field.

Making health centers “AI ready”

Nevertheless, implementation is still a challenge that needs to be overcome. So, the authors proposed the clinical AI department as the centralized body to support health centers to be “AI ready”. This means it will take up responsibilities from integrating clinical workflow to fostering related educational and training programs to prepare present and new generations of practitioners.

The department will also play a pivotal role to implement, utilize, and enhance infrastructures to facilitate AI adoption. At the heart of this change; access barriers to data will either be removed or reduced. At a higher level, these clinical AI departments will drive the country’s financial and regulatory bodies to understand AI’s potential in medicine and healthcare, so that they can modify existing reimbursement and regulatory structures to accommodate the progress.

Most importantly, a new set of standardization and guidelines underlying best practice are to be initiated so that AI models are continuously assessed to reassure safety and model calibration. AI models can be quite sensitive to small input changes, so if re-assessment or recalibration are not performed regularly, its reliability may be in question especially if researchers have the intentions to apply the model across different systems or institutions. At the end of the day, conclusions provided by an AI model should be transparent; one which medical staff can depend upon for decision making.

With that, the authors wrote, “Our patients are waiting for us to make use of these advances to improve their care, and every day wasted is a missed opportunity. Therefore, we ask – who will establish the first department of clinical AI?”


Author Bio

Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.