HCV and Kidney Transplant
In 2015, the FDA approved a direct-acting antiviral for use for people with HCV. Direct-acting antiviral safe and effective in over 95 percent of patients.
Before 2015, HCV+ kidneys were discarded at 2.5 times the rate of HCV- kidneys in the United States due to a lack of effective treatment. Between 2000-2015, about 17.3% of kidneys were discarded. However, 65% of all HCV infected kidneys between 2005 and 2014 were discarded. Historically, African Americans, regardless of their HCV status, have longer wait times for kidney transplants. Longer wait times are associated with increased morbidity and mortality.
The advent of direct-acting antiviral has increased the number of organs available to patients on the kidney transplant waitlist. Previously discarded HCV infected kidneys are now being used and this has increased the number of organs available to patients on the kidney transplant waitlist.
The goal of the study is to understand if there are the differences in time on waitlist for HCV+ kidney recipients and HCV- kidney transplant recipients. Also, I want to understand if there an association between HCV+ pre or post tx status and graft rejection post transplant. Last, I want to understand if there an association between HCV+ pre or post tx and hospitalization rates post transplant.
MACHINE LEARNING IN PSEUDO-REAL TIME
Using Python and Amazon Web Services, I built a system that takes in data from the OPEN FDA API, trains an ensembled machine learning algorithm to figure out tendencies for a specific pitcher, and then makes predictions on new data. Across a random sample of 50 pitchers, we saw an average improvement of about 5 percentage points beyond predicting the pitcher's most common pitch type every time. However, there was high variance: some pitchers were much more predictable than others. For some, we saw a bump of 15 percentage points on average. The algorithms do even better if you narrow the algorithm to certain game situations in which patterns emerge more clearly. Finally, we hooked the algorithm up to Twitter, so that the algorithm can watch a game and make predictions as the game progresses.
RESULTS(Insert screenshots as needed)
For more details, check out the project's website here.