The world’s leading publication for data science, AI, and ML professionals.

What’s the Current State of Data Science in Healthcare?

Take a closer look at data science applications in healthcare and learn why the industry still lags behind in embracing data…

Photo by bongkarn thanyakij from Pexels
Photo by bongkarn thanyakij from Pexels

Nowadays, various healthcare institutions ranging from large governmental hospital networks to private physician offices are turning to data science consulting. This shouldn’t come as a surprise as healthcare is a natural data aggregator. Data-powered decision-making offers healthcare professionals an opportunity for more streamlined clinical operations, more cost-effective and efficient diagnosing, and more personalized patient care.

The AI possibilities

AI offers a spring of opportunities for healthcare. The most promising one, however, concerns the humanity’s ever-present quest for reliable cancer cure. In a nutshell, cancer is usually treated by manually analyzing gene mutations and then selecting treatments accordingly. Although this approach has proven to be effective, it’s very time- and resource-intensive. AI-assisted imaging enables healthcare professionals to identify tumors faster and more efficiently. At the beginning of 2020, Google announced that after being trained on thousands of mammograms, its DeepMind AI algorithm can outperform radiologists in breast cancer detection.

AI’s predictive capabilities have proven to be especially useful for early detection of patients’ life-threatening conditions in the ICU. For example, Philips has designed the eICU system, which allows physicians to proactively react to subtle signals of incoming deterioration in the patient’s condition. By leveraging various IoT devices, cameras, and powerful ML-powered predictive analytics systems, care teams can also quickly identify which patients are ready to be discharged, enabling better prioritization of patients based on the seriousness of their condition. This solution tackles typical problems in the ward: the scarcity of resources such as ICU beds in times of a pandemic, care costs, and high mortality rates. Hospitals that have utilized the eICU program report an overall 16% increase in patients’ chances to survive and 15% faster discharge.

Predictive analytics is also having a tremendous impact on medicine’s next frontier, genomics. This emerging field solves the long-known problem of variable medicine tolerance among different people. Every individual’s DNA contains thousands of different correlations between data points, which can now be quickly analyzed by predictive models. This enables caregivers to detect how exactly an illness will progress and more reliably identify appropriate treatment.

Natural Language Processing (NLP), another subdivision of AI, is poised to help healthcare professionals deal with the avalanche of unstructured data that is growing on a daily basis. Physicians commonly record the course of treatment into EHR systems, which poses numerous problems. First, EHR systems cause frustration among physicians, as they require a very standardized and structured approach to information input, which takes a lot of time. Secondly, physicians spend hours interpreting long medical histories to then identify the pertinent information from the EHR. NLP solves both problems by automatically converting physicians’ unstructured notes into EHR-suitable data and doing the same in reverse for doctors who need to quickly assess the patients’ medical history.

The problem with AI

Amid the AI hype and regardless of the technology’s huge potential, AI’s black box problem continues to persist. The most advanced AI-assisted tools can diagnose quicker and more accurately than a human being, but they can’t explain the logic behind their decisions.

The black box problem doesn’t allow for AI to unveil its full potential in the medical context. Until we find ways to deconstruct the methods these models use for decision-making, AI will require humans to intervene. Although it doesn’t seem to be much of a problem at the first sight, this obscurity factor completely eliminates the possibility for diagnosis automation, which would potentially have a critical impact on the healthcare industry as a whole.

Moreover, AI also has the mostly unsolved problem of dataset bias. Many modern models operate on generalized datasets, which means that minorities can be mistreated. Until solid regulatory frameworks regarding dataset discrimination will be put into practice, there is no chance for AI to reliably operate on a bigger scale. Furthermore, in the US, for example, any AI system would need an FDA approval. Essentially, all of AI’s problems in healthcare are linked to scalability, and there is no other industry that demands scalability like healthcare.

The future outlook

With all the advantages Big Data brings, the healthcare system can finally become preventive rather than reactive. Data allows healthcare professionals to have a much deeper understanding of image analysis and overall patient screening, so that physicians could reliably detect serious diseases like cancer before those escalate.

Big data also tackles one of the most critical healthcare aspects – personalization. Physicians can now use information that goes way beyond simple demographic and physical indicators, enabling patients to be treated individually.

However, it’s evident that healthcare is currently not yet ready for this transformation. Data-related regulations, a lack of governmental support, and data fragmentation are all serious bottlenecks to the widespread adoption of systems that rely on Data Science. On the other hand, we are now seeing that big data initiatives led by big governmental organizations are now getting sufficient financial support in the healthcare domain.

For example, in 2019, UK Biobank collected extensive healthcare datasets including sociological information, biological samples, and conventional physical metrics about half a million people, enabling researchers to find interconnections between disease outcomes and this rich medical information. In 2018, China also started to encourage the use of big data in healthcare by linking its personal identification system with medical data and establishing the National Central Cancer Registry. Given that China has the largest population size in the world, such programs can have an immense impact on cancer treatment.

The recency of these campaigns and the lack of interoperability between Healthcare institutions means that we can’t expect any dramatic results in the incoming years. Given that the penetration of big data in healthcare is most likely one of the biggest shifts this industry has experienced in the past decades, it’s not surprising that governments, healthcare institutions, researchers, and physicians need time to adapt.


Related Articles