Artificial Intelligence (AI) offers promising opportunities to reinvent the healthcare industry, from optimizing the diagnostic progress to improving quality of care as well as managing hospital capacities. Particularly during a global pandemic AI carries the hope to become more efficient and effective in healthcare related tasks and ultimately improve our health. However, the use of algorithms in healthcare does not come without some major challenges data scientists have to address.
With regards to our upcoming DSSVirtual for Healthcare, Finance & Technology, we had the chance to talk to leading data scientists from the industry who shared with us their AI predictions, main challenges of their daily work, as well as some useful advice for data scientists working in the industry.
How AI is transforming the healthcare industry in 2021
According to Vasileios Stathias, Lead Data Scientist at Sylvester Comprehensive Cancer Center, large amounts of data provided in the healthcare sector, combined with innovative AI technologies lead to promising applications towards more personalized medicine and more accurate diagnosis:
“One of the most exciting use cases is the application of AI towards precision medicine. The increase in the generation of large-scale biomedical datasets, together with the accessibility of high-performance computing gives us the unpresented opportunity to leverage AI to its full potential and to make substantial progress in the healthcare industry. The exponential growth of the field of AI, provides scientists with a plethora of available machine and deep learning algorithms that could readily be applied in the optimization of the detection, prognosis and diagnosis of patients at the individual level. Furthermore, the complexity and multi-dimensionality of the biological data, make them ideal for AI and can significantly advance the identification of patient-specific therapies and interventions.”
Value based care
A trend towards personalized care is also seen by Ayda Farhadi, Senior Data Scientist at UPS, who further highlights the great advantages of identifying people’s health at an early stage:
“I would say I can see two main transformations in healthcare going on in 2021. Last year hospitals were filled up with COVID-19 patients and AI has helped them manage capacity and prevent avoidable hospitalization. Definitely AI was useful to reduce the number of readmissions resulting in saving costs. The main transformation of AI in healthcare is shifting to value based care. Patients who need care will be more likely to have worse health in the future, increasing cost for healthcare. AI can be used to identify them and providers can then proactively engage their valuable patients in order to avoid hospitalization. Another transformation that is gaining more attention is personalized medicine. People are different and there is no one size fit all approach that would work for every patient. Genetic, diet, lifestyle and socioeconomic status are just a few variables that influence a person's health. With AI, providers can customize their care plans based on patients' unique circumstances. AI could be used to predict which medication will be most effective for individual patients. There are some other areas that AI would play an important role in healthcare but these two were mainstream and sensible due to pandemic.”
high risk COVID-19 patients Identification
Furthermore, the use of data and AI will keep playing an important role in 2021 to face COVID-19 challenges, as mentioned by Ankur Chaudhary, Senior Data Scientist at Healthfirst:
“AI can help to identify high risk individuals due to comorbidities, identify social isolation and availability of care. Improvements can also be done at geographical level and members who are at a disadvantage due to socioeconomic conditions can also be helped in a timely manner. This data can also be leveraged for vaccination and providing vaccines to the most frail members of our society.”
The biggest challenges of applying AI in healthcare
Innovation does usually not come without some major challenges, which in data science are mostly associated with imbalance datasets, according to Ayda Farhadi:
Data quality and imbalanced data
“The biggest challenges are imbalanced classes. Especially in healthcare, the number of patients who have the disease is usually much lower than the ones who test negative indicating that they do not have the disease. This was the case in most of the healthcare-related projects that I have been involved in in the past, such as predicting high intracranial pressure of children, predicting breast cancer, and predicting heart disease of patients. Another challenge is accessing data. Normally you would think that you will have access to more data when you are working in a large hospital but that is not the case. Because of privacy problems, the dataset that is provided to you is not that much and this causes low performance of machine learning models. However, using transfer learning the effect of small datasets or imbalanced datasets in healthcare can be reduced. Transfer learning has been utilized to solve several healthcare-related tasks using pre-trained state-of-the-art models from open-source datasets. It is still a very interesting research area in healthcare and it has the potential for getting better in this field.”
Ankur Chaudhary agrees with Ayda Farhadi but stays optimistic that the industry will be able to overcome data quality issues:
“The main challenge is quality of data since data in the healthcare industry is not as clean or organized as in other sectors like finance. This can be attributed to slower migration from legacy systems in a few cases and less governance, collection policies. However, the industry recognises this challenge and is working to resolve it in a speedy manner.”
Understanding the business problem
Besides data quality, Vaibhav Verdhan, Principal Data Scientist at Johnson & Johnson is highlighting additional challenges data scientists are facing when using algorithms AI in the healthcare industry:
“Healthcare is a dynamic field generating tons of data every moment and AI is making a huge impact and improving our health. But the application is AI faces some challenges:
- Defining the concise and clear business problem which is achievable.
- Getting a data set which is representative of the use case is another big challenge we face.
- Domain knowledge is something which is of paramount importance. Many times, data scientists might not have the required domain information.
- Data quality is a big concern we often face while trying to build AI models.
- Infrastructure can be a show-stopper in the progress.
These stumbling blocks can seriously impact the progress of AI projects. It is imperative these points are addressed and resolved before we commit to a project.”
Recommendations to data scientists working in the healthcare industry
In order to overcome the many challenges data scientists have to face in their daily work and to successfully implement AI projects in their organizations, the experienced data scientists provide some useful advice.
Get the data right
“Plan to spend 80% of your time getting the data right, always do an EDA to understand data and underlying relationships before starting work on model development. Your AUC will be vastly improved by getting the data right vs getting the best XgBoost model. Wear multiple hats of data engineer, data scientist and product manager when needed to reduce the gap between finished product and business problem.” —Ankur Chaudhary, Senior Data Scientist at Healthfirst
Keep yourself updated
“I would strongly highlight the importance of keeping up to date with the newest methodologies and technologies. The fields of data science and healthcare are both rapidly evolving, and therefore it's essential to be familiar with the latest approaches. The constant influx of new knowledge not only helps you with improving the design and execution of future projects, but can also retrospectively address bottlenecks in existing ones.
A second recommendation would be to not limit yourself in using only one programming language. In the long term, the advantages of being able to utilize the appropriate libraries for the right task, far outperform the initial learning curve and feeling of discomfort of coding in another language.” —Vasileios Stathias, Lead Data Scientist at Sylvester Comprehensive Cancer Center
Acquire domain knowledge
“Human body generates two terabytes of data daily. We have healthcare systems generating tons of data every-day, in the form of prescriptions, transactions, records and images. As a data scientist the onus lies on us to make the most of it. To achieve tangible and far-reaching results, having an in-depth domain knowledge is imperative. Understanding the business problems closely and converting them to analytical one is an art which will bear fruits in the long run. By upgrading one’s skill sets constantly and keeping an eye on upcoming trends, will allow us to constantly improve the craft. If you are a data scientist working in the healthcare domain or if you wish to make a career in this field, these points will enhance the impact you can make.” —Vaibhav Verdhan, Principal Data Scientist at Johnson & Johnson
As mentioned above, keeping up to date with the latest advancements is key to successfully execute AI projects. Data science events, such as the Data Science Salon for Healthcare, Finance & Technology from February 16-17 offer a great opportunity to learn from other practitioners and exchange best practices. The two day virtual event brings together a diverse community of data science managers and practitioners who share their experience and expertise in the field. Join us for the event and connect to like-minded people in the industry!