Data Science and Healthcare

Source: Deep Learning on Medium

I recently went to a Data & Healthcare meetup to see how a renowned cancer research institute, Memorial Sloan Kettering Cancer Center (MSK), applies and uses Data Science.

Predictive Modeling

One of the first uses they mentioned was predictive modeling for post-hospital care. Whether it is from being discharged after an acute condition that required a hospital stay or perhaps going home after a chemotherapy treatment, MSK made it clear that their treatment does not end just because you left their premises. MSK collects data that tracks how often a patient may call, the reason of their call, and the outcome — whether it is to come into the hospital for immediate care or otherwise. With this data they can use predictive modeling to assess the needs and interventions a patient may need in the future. Data Science and machine learning can also be used to help predict pain crises. It was very inspiring to hear that MSK has “a lot of data” and they’re exploring how Data Science can be used to be beneficial and impactful to provide the best patient centered experience.

“How Many Clicks Is Too Many Clicks?” or A/B Testing.

There is a multitude of data generated in healthcare. Vital signs, past medical history, co-morbidities, drug interactions, allergies, treatments, treatment notes, interventions, test results and so on. A very important problem one of the speakers of MSK presented was “How many clicks is too many clicks?” Everything in a hospital must be documented, after all, “if you didn’t document it, you didn’t do it” is a very popular saying in healthcare. The problem posed is how do you present all the necessary and vital information without burying the user with too much information or having to take too many steps (or clicks) to finally arrive at the information they need. After all, it could be a life or death situation and the information must be accessed in the fastest and most efficient way possible.

Documenting also takes up a large amount of a healthcare professional’s time. Does the professional have to click through four, five, or six pages to document everything they have done for the patient? How many minutes does it take the professional to complete documentation for one patient? Then calculate the amount of time this takes since that professional may have several patients. Interventions and documentation needs to be done several times in one shift.

While the speaker from MSK did not specifically say how they addressed this problem, A/B Testing sounds like it would be one of the most ideal solutions to discovering how many clicks is too many clicks.

Natural Language Processing

Another speaker from MSK stated that some providers don’t like to click on checkboxes to document their notes and that they prefer to write it out in prose. The speaker said they are exploring ways to address this. I immediately thought how Natural Language Processing can be used to tokenize the notes of the provider, skipping the need for checkboxes and would allow the institute to store the information in their database in whatever form that would be the most purposeful and practical.

Scheduling/Staffing

A hospital is the sum of its parts, and a huge component of that is staffing. A hospital is made up for a multidisciplinary team consisting of providers, nurses, aides, nutritional services, environmental services, engineering, researchers, scientists, and so on. Data Science can be employed to find hiring and turnover rates, what is the appropriate amount of staff needed on a particular night and increase employee performance and retention.

Research and Clinical Trials

Finally, Data Science is used in research and clinical trials. The presenters were not too detailed on the research they were doing but they did emphasize that they do run clinical trials. The results of these trials can be expanded upon using machine learning to look for additional information or insights.

Challenges

If a machine learning algorithm comes to a different conclusion than a provider, what would it take to make that provider trust the machine learning result over their own call using their education and intuition?

In closing

In MSK’s About Us page, they have the following statement:

The close collaboration between our physicians and scientists is one of our unique strengths

While I don’t believe that quote was necessarily addressed towards Data Science, it is inspiring and interesting to see how a company such as MSK promotes and shares their interest and excitement in the field of Data Science and analytics to ultimately provide the best patient centered care.