Data Science: How it really works
While it’s difficult to put a price tag on the global economic burden of diseases like asthma, it’s clear that improving access to preventative care is a key part of the solution. In the US alone 2 million asthma sufferers will land in the emergency room due to preventable attacks and 50 billion dollars a year is spent on asthma, with hospital visits as the largest contributor. So what if we could identify patterns behind
these attacks and prevent them? Saving billions of dollars in unnecessary suffering.
Meet Sophia she’s a data scientist at Apex analytics, studying trends in the field of biomedical research.
She’s been working on a project involving weather data and pollen: a known asthma trigger. So when Sophia’s colleague Jane mentions her son’s recent hospitalization for an asthma attack, Sophia wonders if the dramatic seasonal change over the weekend was a trigger and if so, could weather forecasts predict
when patients are at a higher risk of an attack? Jane is intrigued by Sophia’s hypothesis and as an offering manager, she gets approval to research an application that could be sold as an alert service.
Together Sophia and Jane gather preliminary business and academic information discovering that Sophia’s theory has not been validated but if they find a link this could be a breakthrough for patients and related healthcare businesses.
Sophia decides to move forward with her research and approaches her data engineer Greg to determine how to proceed. They’ll need a combination of data. Internal data can be sourced from earlier apex research using electronic medical records. The records on asthma-related emergency visits they’ll need are considered privileged and have to be extracted and kept private.
They will also look at external data
sources such as pollution and weather data that are publicly available. With the right datasets secured Sophia and Greg begin to clean and process the data, a task sometimes called data wrangling. This involves removing irrelevant data points and putting the right information into a format so it can be easily analyzed. The next step is data exploration Sofia looks for relevant location and time-specific correlations, for example, she compares wind patterns with past asthma records to see if the spread of pollen correlated to a high outbreak.
Sophia’s findings support her hypothesis, so she begins feature selection and model selection to create a predictive model. Based on the correlations she selects features in the data that are the most useful and relevant to predicting asthma outbreaks. Now that the model is set Sophia starts to cross-validate by dividing her data and conducting multiple tests for accuracy.
If those tests perform well Sofia runs a test
on an entirely new set of data, the model hasn’t seen. After successful testing, Sofia has quantified what specific weather conditions are strongly tied to the onset of severe asthma attacks and now that she has created a model to accurately predict those conditions she can deploy it to create an alert application that will benefit patients and businesses.
With the help of Greg and other software and DevOps engineers, they turn Sofia’s model into the accessible application programming interface or API that will service the alerts. Meanwhile, Jane as the offering manager starts telling her clients about the new application and how it can be of use to them and their business.
For insurance companies, it’s fewer expensive ER visits when their patients are prepared,
for hospital administrators its cost savings and smarter staffing of specialists, for pharmacies, it’s higher customer loyalty and engagement when they alert customers to get their prescriptions on time, and finally for people like Jane and her son it’s being able to enjoy the seasonal weather.
Now your take on this argument.
We would also like to hear what you feel about the topic we discussed today. Your feedback is very important to us. Feel free to drop your comments and recommendations. If you have a contrary opinion, you can drop that too.
You can also join our Facebook Page CRMNigeria for more updates. You can do that by clicking on the link or searching for our page on Facebook.