AI-based Automation to improve drug safety

In my previous posts, I had described the basics of Pharmacovigilance (PV)[i] and the role of consumer feedback in improving drug safety[ii]. In this post, I shall describe Navrotech’s journey towards achieving 99% automation in analyzing consumer feedback of medical products, from online media.

Why is consumer feedback so important for Pharmacovigilance?

During Phase-IV clinical trials (Post-Marketing Surveillance) it is critical to know about consumer experiences to uncover insights that cannot be found during the earlier Phases (0 — III) due to restrictions with sample size and variety of human participants. Below are examples of insights that our PV expert arrived at using Navrotech’s tool-YugenSocial®, that can help in improving drug safety

1. Adverse Events (AE) from medical products (drugs/vaccines/medical devices):
Did any consumer experience an adverse reaction after using the product? If yes, is the number/intensity of such cases statistically significant and was the reaction known during clinical trials?

2. Consumer Queries:
Is this drug/vaccine safe with the consumer’s current condition?

3. Out of Label use :
Is the drug being prescribed by physicians or used by consumers for reasons other than what it was intended for? Is there enough evidence of therapeutic benefit through out-of-label use or is it a case of misuse/abuse?

4. Possible Drug-Drug/Drug-Food Interactions:
Did any consumer experience an adverse reaction when the drug was consumed along with another drug or a specific food item?

5. Lack of Efficacy of Drug:
Did the drug provide sufficient benefit to the patient as intended or observed during clinical trials? If not, what did the patient do subsequently?

6. Medication Errors:
What are the most common dosage errors, drug package errors, missed dosage complaints etc.?

7. Clinically Important information:
Is there any other useful information that could be leveraged for additional use-cases? i.e Medico-Marketing, Medico-Legal etc.

Due to the lack of awareness and a direct incentive, most consumers do not formally provide this feedback the way they should[iii]. Instead, the natural tendency is to discuss their experiences with friends or fellow patients through in-person discussions, telephone calls or the most accessible (& impactful) medium- online media i.e healthcare social networks, online health forums, conventional social media etc. A few progressive pharmaceutical companies and regulatory bodies have realized the importance of tapping consumer feedback indirectly to benefit patients. Here are some examples

1. The US FDA actively monitors online media to uncover critical insights on drug safety. In the below article the FDA commissioner (Dr. Scott Gottlieb) outlines how they have been able to uncover the next wave of drug abuse post the opioid and stimulants crisis through “Proactive Pharmacovigilance[iv]

2. Using social media, Stanford scientists and doctors were able to uncover a hitherto unknown link between a cancer drug and an adverse reaction — hypohidrosis. They found that patients were discussing online about experiencing “loss of sweating” (hypohidrosis) while/after consuming this drug. They could identify a statistically significant number of such patients (23) for a deeper analysis [v]

3. Glaxosmithkline (GSK) was able to leverage consumer feedback from online media to learn that “consumers were unhappy with the ambivalence around vaccine safety and the timing of when shots should be administered.” Through proactive analysis of online media, GSK was able to provide “physicians with better educational materials to convey to parents vaccine safety information and the risk of diseases like measles”. [vi]

What does YugenSocial® do?

YugenSocial® (portmanteau: Yūgen (幽玄) + Social) is Navrotech’s proprietary AI solution that automates the process of extracting critical insights (mentioned above) from online media. Till date, YugenSocial® has achieved 90% automation and is on target to achieve 99% automation by November 2018. Currently, many life sciences companies employ a large number of resources to manually identify relevant insights from the massive amount of data available online. If we take the example of Twitter, on any given day for (say) 800 drug and generic product names, one can expect to get at least 7200 tweets mentioning the said drug/generic. Out of this 7200, approx. 75% (5400) tweets are irrelevant drug mentions. Further, approx. 10% (720) tweets are relevant for PV analysis and finally, a mere 1% (72) tweets need to be thoroughly analyzed and classified by clinical teams for actionable insights. The below image depicts the time required to perform this task manually (man hours) and how different levels of automation can expedite this task to provide real-time insights.

Image for post
Image for post

*For top 800 drug and generic names most mentioned with adverse reactions

How can AI based Automation improve the productivity and profitability of life sciences companies?

By embracing automation, the life sciences industry can free up their existing resources to take up more skilled tasks and leave the manual/repetitive tasks to advanced machines that can deliver a similar or better accuracy (> 85%) in much lesser time. This automation can lead to improved productivity and have a direct impact on a company’s bottom line through increased revenues (i.e can take up more projects) without additional expenses (i.e for the same resource cost). We, at Navrotech, experienced the benefit of automation by using our deep learning models to identify AE relevant tweets, with good accuracy and recall numbers. Currently, YugenSocial® can extract 72.5% of all AE related data available online for drugs, vaccines, and medical devices. With 90% automation, we need only 3 clinical experts (20 man hours) to clear our daily work backlog against 25 experts (200 man hours) without our AI solution. This way, our clinical experts are able to extract better insights from targeted and more relevant data. By November 2018, adopters of YugenSocial® would need only 1 clinical expert to review insights (on 800 drug/generics) from approx. 4 different data sources. The freed up clinical experts can be leveraged for additional revenue generating work such as extraction of insights on vaccines, medical devices, and other remaining drugs.

The richness of mined consumer feedback has surprised us, particularly in the case of vaccines. Currently, YugenSocial® captures approx. 10 relevant reports on medical devices and 60 relevant reports on vaccines, on a daily basis (globally). At a time when under-reporting of drug reactions is a major concern for the PV industry, such breakthrough solutions can play a vital role in monitoring the safety of existing drugs and the development of new drugs.


1. Dr. Ghanashyam Rao: Clinical Advisor

2. Dr. Tulsi Rajkumar: Pharmacovigilance Consultant & Expert

Note: Navrotech is committed to maintaining the highest level of data privacy through advanced encryption and de-identification techniques. YugenSocial® will only provide critical insights that can improve the safety of drugs and at no cost would any personally identifiable information be provided unless required by law for patient benefit.







Originally published at on September 6, 2018.

Building advanced AI solutions for improving drug safety

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store