5 Things that Data Science Should Learn from Medical Science

Data science and medical science are two different fields. However, recently we have seen several applications of data science in medical science. For example, drug design, disease prediction, medical imaging, virtual assistant, intelligent diagnosis, and so on. The state-of-the-art data science algorithms along with cutting-edge technologies have enriched medical science significantly. 

Now the question is – can the reverse happen? What can data science learn from medical science?

Certainly, there are several areas that data science can learn and inherit from medical science. Let’s see those.

1. Global focus

Have you ever heard a heart specialist saying “You know – my job is to treat the patient’s heart! My concern is only the heart of the patient and not anything else. Other doctors should do that respectively”. Imagine, what will happen to the patient!

Fortunately, the doctors – irrespective of their specializations – care about the overall health and well-being of the patient. In other words, they prioritize the global objective of a patient’s well-being beyond the local concerns. This is something that data science should inherit from medical science – to focus on the global objective! 

On the other side, in many cases, data science models are built in silos and they usually do not talk to each other. This creates serious conflicts and counter-intuitive recommendations. For example, the analytics manager of procurement may advise buying in bulk for economies of scale and larger discounts from the suppliers. However, higher inventory holding and opportunity costs may often negate the benefits. Similarly, a good inventory model may save some dollars but can cause significant revenue loss due to material stockouts – if the right demand signals are not captured (link1).  

Only a holistic approach can stitch together all parts of information and provide the globally best recommendation. 

Data science can learn this aspect from medical science.

2. Going beyond predictive to prescriptive models  

A consulting doctor first carefully listens to the patient, predicts the disease, and provides recommendations to cure the disease. In other words, a doctor first discovers the root problem, predicts what bad is going to happen, and then thrives to eliminate the root cause with appropriate actions. Hence, in medical science, the intent is always to detect accurately (predictive) and go further to cure (prescriptive). 

Currently, there are sophisticated prediction models in data science. However, most of them are unexplainable BlackBox. Further, the models are built in silos. 

Such predictive models alone are limited and cannot solve the problems. For that, predictive models must be backed by actionable prescriptive models with clear recommendations. In its current form, the application of predictive modeling is very limited and has yet to see much traction in data science. 

We need more and more applied prescriptive models to create supreme value from data. 

3. Cleaner data

Have you seen how a doctor carefully collects a patient’s health records and other data? Right from taking periodic readings of a patient’s basic health stats to monitoring health status – everything is captured meticulously. There are sophisticated digital systems to capture the health data of critical patients in hospitals or special care units. All the information is collected, organized, and provided to the consulting doctor for better diagnosis. The important point to note is that the patients’ data are very clean, transparent, well organized, and less noisy.

On the contrary, the quality of data is always a concern for data science – even for large enterprises. Most data science projects spend 80% of their time collecting, organizing, and processing the data. Moreover, an organization has multiple interfaces and systems to source the data. This makes data reconciliation a tedious job. The collected data is not clean due to a lack of transparency. It is also challenging to organize and process the data for model building due to multiple data formats. With this, often the models are sub-optimal, imperfect, and fail badly on the ground!

This is one more area that data science should focus on – after all, quality data can produce quality output.

4. Exploratory data analysis

A well-experienced doctor will not jump to a conclusion suddenly. The doctor listens to the patient, gets some initial health stats to understand the problem in-depth, recommends a few medical tests, predicts the disease accurately, and finally writes a prescription. In short, medical science follows a meticulous step-by-step process of defining the problem, hypothesis testing, exploratory data analysis, prediction, and finally prescription.

On the other hand, I have seen many data analysts who lack the patience to follow all these steps. Given a dataset, they immediately jump on to fit a Random Forest or a Light GBM model on that – without any attempt to understand the business problem, validate the data relevance, check the quality of the data, and explore data to get an initial sense of the root cause investigation! Such models remain fundamentally very weak and often fail to meet the business expectation.

Hence, data science should follow a more structured step-by-step process like medical science to solve a problem.

5. Continuous improvement

Here comes the final thing from medical science – monitoring, tracking, and continuous improvement. There is a dedicated focus on tracking the patient’s overall health during treatment time. Appropriate actions are taken depending on the real-life health of the patient. Predictions and prescriptions are updated so that the patient gradually improves to the best of her/his health. It is a process and not just a day’s agenda.

Similarly, the job of a data science model is not over as soon as it is deployed. The process begins from here. There has to be performance tracking, managing uncertainties, model tweaking, re-tweaking, and redefining model strategy on the go. Hence, the data models are not like a still picture – but more like a live movie. 

The goal is continuous improvement.

Conclusion

To summarize, though data science and medical science are two different fields, they share a lot of mutual benefits. Data science certainly can inherit and adapt a few important philosophies from medical science:

• looking for system-wide benefits rather than local optima
• going beyond predictions and deploying prescriptive models to foster actionable insights 
• capturing cleaner data – after all, a sandwich is as good as the loaf
• extensive exploratory data analysis to understand the core challenges and root causes, and finally,
• continuous improvement.

Agree that it is very difficult to bring in the above philosophies from medical science to data science in its entirety. There are challenges. Even the largest corporations are struggling to do that. But a successful data-driven strategy will largely depend on how effective an organization is to manage the challenges. All that is required is a dedicated focus and a system-wide global vision.

Do you see any other fields that data science can learn from? Please let me know.