OptiRX

Inspiration

Chronic Illnesses are quite common in the US with 6 in every 10 adults suffering from at least one of them. Additionally 5 out of 10 leading causes of deaths in the US are these chronic diseases. For at least 60% of these diseases, Narrow therapeutic index or NTI drugs are prescribed. But the critical nature of these drugs is the “narrow therapeutic range”, that a slight change in dosage can either make the drug toxic and fatal or make it entirely useless. Hence, such chronic care management requires continuous and vigilant care. Taking a different dose, changes in your personal life, missing a dose, forgetting what was the dose assigned in the first place, can affect the drug blood concentration levels by a lot and keep you at a high risk of either the drug not being effective or toxic.

What it does

We propose here a responsive app : OPTI-RX (on both PC and mobile devices) to track the drug effectiveness over the course of time, and continuously send the updates and alerts if any anomalies are found in the dosage effectiveness to the required physician. We can create personal profiles of the person and simultaneously save them to be accessed in future through login username and password, Additionally when we receive the prescription for the first time, the app can use the scanned photo to automatically fill in the details for the person like Name, age, body weight etc. as well as the drug profile including name and dosage. Once such parameters are added, one of the features of the app can reassure if the dose is safe or else toxic along with adding this datapoint to the time series plot as mentioned above, as well as finding the nearest dose for safe use.

How we built it

For building the application we utilized the Google Cloud’s Vision API, Flask and React. The personal profile creation in front end (React) takes in the data from the user and provides it to the back-end (Flask, hosted on the Google Compute Engine). Then it provides the extracted information to Google Cloud SQL platform and confirms the login ID and username or creates a new profile and saves to the database. The prescription recognition is done through Optical Character Recognition, the scanned photo is securely transferred from React to Flask to Google Cloud Vision APIs. The extracted drug and personal profile data from the prescription, is sent as an array of strings to the back-end model/algorithm. The algorithm behind estimating the drug concentration levels in the blood uses the CKD-EPI Creatinine equation, which accounts for the effect of personal factors like body weight, age, gender etc. on GFR (renal clearance). While the equation parameters and the therapeutic range limits are found from literature. The estimated drug blood concentration level is checked in the therapeutic range, which then classifies it as True if safe else False (toxic). Additionally an approximate dose is estimated for the safe drug blood concentration.

Challenges we ran into

One of the biggest challenges faced is developing a connection between Google Cloud SQL and VM Instance for personal profile creation. Other challenges include retrieving the pharmacokinetic based equation model for incorporating the personal factors affect on clearance rate along with the data for the parameters in those equations.

Accomplishments that we're proud of

We created an interactive web app, where the complete database was based on Google cloud APIs and app based tools like Flask and React. The model is also well constructed through extensive literature review and pharmacokinetic analysis.

What we learned

A number of new tools and concepts were learnt (many were mentioned above). With a project like this carried out in such a short time, a good team spirit, time management, problem solving skills all were very well grasped through this Hackathon.

What's next for OptiRX

The app we created is still in its initial stages, there are a number of developments which can take place over this. At this stage the app as well as the back-end algorithm uses only body weight, height, gender and age as the personal factors. Other time dependent lifestyle factors like nutrition intake, sleep times, sickness etc. can also affect the clearance rate and thus affecting the drug concentration. We use a deterministic/ theoretical model for drug concentration estimation; further we can use a regression based model which is trained from previous data of personal factors and advised therapeutic dose and hence predicts the effective dose based on the current personal and drug parameters. Lastly as a moonshot, we can try to detect the disease and aggression of the disease based on the present symptoms and personal characteristics, where the model is trained through WebMD datasets