Siemens Healthineers’ Dr Liana Romero reflects on her background in clinical biochemistry and public health epidemiology, her work on AI for clinical decision support, and the challenges of implementation and return on investment for artificial intelligence in healthcare.

You have a background in clinical biochemistry and public health epidemiology. How have you come to specialize in artificial intelligence?

Indeed, I started in clinical biochemistry and clinical diagnostic laboratory testing where I gained a deep appreciation for the value of testing in disease diagnosis, monitoring disease progression and efficacy of therapy, especially in population health applications.

Fast forward to my doctorate in public health and the epidemiology of chronic diseases. And It’s in this application that I see the tremendous value of digital solutions in healthcare, and specifically artificial intelligence, in bettering individual and population health – whether it’s because AI is expediting diagnosis, estimating risk of disease, personalizing therapy or combining data into new insights that can lead to better patient outcomes.

Your primary focus is on AI for decision support. In which clinical domains are you seeing the greatest impact?

AI is having a significant impact in supporting diagnostic and therapeutic decision making. From my perspective, one of the areas that has benefited the most from AI is radiology.

Artificial intelligence applications are significantly enhancing the ability of the radiologists to provide excellence in diagnostic image interpretation, particularly in differentiating between high- and low-risk lesions across a wide variety of imaging modalities.

Integration of radiographic imaging with other sources of data (e.g., clinical features and genetic/biochemical markers) to risk stratify image-detected lesions already exists and will likely be more commonplace in the future. And they can do so while also expediting workflow and removing the burden of repetitive tasks that impact cognitive abilities during long days and heavy caseloads.

Thus, AI is not only helping to reduce errors and decreasing time to diagnosis, but also enhancing the clinician and patient experience throughout the diagnostic process. We know this will only get better as machine learning and deep learning continue to improve the algorithms.

There has been great progress in the deployment of AI in cancer care over recent years. Do you have examples of successful applications?

AI algorithms have the potential to transform healthcare delivery through its inherent ability to solve complex problems. Therefore, the opportunity for AI to impact oncology-related problems is great because, as we know, treating oncology patients has become exceptionally complex.

I’ve already touched on the applications of AI in radiology and diagnostics.  Now, let’s focus on treatment decision making.

First, let’s consider data. The data set for just one patient is exceptionally large. And yet, optimal treatment planning requires the interpretation and synthesis of these large amounts of complex data from different sources, including patients’ history, pathology, laboratory, radiology, and advanced molecular diagnostics.  Aggregating the most significant data points into one dashboard and providing guidance for diagnostic and treatment planning is already a significant benefit of AI in oncology.

Secondly, cancer care is multidisciplinary – requiring coordinated input from multiple clinical stakeholders, as well as the patients and caregivers. Therefore, the aggregated data enables the multidisciplinary tumor boards in several key areas:  providing a comprehensive data set for decision making, creating efficiency in expediting the decision-making process, enabling multidisciplinary collaboration, and presenting a personalized mapping of the patient along the recommended treatment options.

Couple the above with the fact that oncology care is multi-site and administered across disparate settings, and we can see that the opportunities where AI is able to solve these hurdles will ultimately have a tremendous impact in improving the quality and value of cancer care.

What are the main challenges you have observed in the implementation of AI solutions, and how can they be mitigated?

There are several significant challenges I have personally observed in the implementation of AI solutions.

Although I truly believe clinicians appreciate the value AI can bring in supporting diagnostic and therapeutic decision making, there is an inherent trust factor they need to overcome. Ironically, it’s the chicken and egg situation – it’s only by using AI applications that the trust will be established. The volume of literature being published on this topic is certainly a great step in this direction, and I encourage our customers and those using AI in healthcare around the world to continue publishing their experiences and examples of AI implementations.

The digital maturity of healthcare systems is another challenge. Institutions with advanced digital maturity have the backbone of electronic health records upon which AI applications can integrate to pull the data that is necessary.  But when the digital maturity is low, and the institutions have numerous disparate data sources, that creates a significant burden towards AI implementation. This is where government programs and policy changes can really make a difference. If the government enables programs to facilitate the widespread implementation of electronic health records, it will also enable the adoption of AI by removing the burden of data standardization, collection, and management.

AI-based solutions must be integrated seamlessly into the clinician’s workflow, be intuitive to use, and provide value to the user. For example, clinical decision support systems for oncologists need to be delivered through environments that are highly interactive and provide explanations. And as a patient’s case evolves along the clinical pathway, any changes that the AI algorithms are making because of the case evolution, need to be apparent to the clinician. This goes back to the trust issue – the AI solutions need to be trusted by the clinician and this can be achieved if the clinician understands how the AI is adapting.

A major consideration for healthcare providers looking at AI solutions is around ROI. What sort of outcomes are you seeing in your sites around the world?

I’ve shared a number of examples where AI is having a significant impact on supporting diagnostic and therapeutic decision making in healthcare. Through these examples, we see return on investment in a number of ways.

For example, clinical ROI is coming from the increased accuracy in diagnosis and personalization and standardization of treatment decisions that are also aiding to reduce unwarranted variations in care. Operational ROI is coming from creating efficiencies, removing administrative burdens of workflows, improving multi-disciplinary engagement with complete data sets, and reducing the time from diagnosis to therapy.

Financial ROI comes from all of the above: improving workflow and reducing administrative tasks translates into operational efficiencies and costs savings; comprehensive patient data facilitates the efficacy of multi-disciplinary teams which leads to better decision making and reduction in errors; and accuracy in diagnosis leads to accuracy in treatment, which leads to improved patient outcomes, which is by far the most significant ROI for healthcare providers.

Liana Romero, PhD, MBA, MT (ASCP), is the Head of Global Marketing, Clinical Decision Solutions, Digital Health, for Siemens Healthineers GmbH. In this role, Dr Romero utilizes her vast clinical experience to align the AI technologies that aid diagnostic and therapeutic decision making solutions to oncology clinical pathways.

Dr Romero is a speaker for our Clinician Series on primary care and population health. Click here to find out more.