An Aritificial Intelligence PhD and a Healthcare PhD walk into a bar. The Healthcare PhD starts telling the AI PhD tales from the trenches of clinical research.
The UVA/PADOVA Type 1 Diabetes Simulator is an example of a mathematical smulator used for in sillico trials of diabetes treatments. The researchers assessed a dataset of diabetes patients collected in a bihormonal closed-loop clinical trial and developed a set of equations that describe these training data. These equations have become the standard tool to forecast how a certain treatment affects a diabetes patient.
Workaround studies in Healthcare explore the deviations between prescribed clinical protocols and what doctors and nurses end up actually doing in real life practice. The reasons for these workarounds as well as their desirablitily are debated, but be they caused by individual clinicans' failures, organizational issues like understaffing and unrealistic expectations with regard to speed, poor quality of protocols themselves or (most likely) all of the above, workarounds have to be studied and understood. Such research can shed more light on the ongoing debate, as well as on particular protocols in question. To that end, researchers reverse engineer the de facto protocols that clinicans follow.
“So”, drawls the AI PhD, examining the bottom of his old fashioned glass, “these researchers are like a homo sapiens programming by example system?” AI PhD hiccups and looks up from the glass to give the Healthcare PhD a glance of bemusement “You have a dataset of trajectories and you are tasked with writing a protocol (a program) that can produce these trajectories. You write it, you test that it’s consistent with the data and you publish it. You know we have”, AI PhD pauses for another hearty hiccup, “apps for that, right?”
AI PhD is, at this point, too intoxicated to explain that programming by example
- studies automatic generation of programs based on a specification of input-output examples like those you would find at the bottom of a competitive programming problem statement
- has experienced explosive growth in recent years
- unlike other machine learning models, represents the induced algorithm as a program in a human-readable programming language, an approach that has a great potential in fields reluctant to adopt black box models
This post is a proposal for a new field of research at the intersection of Program Synthesis, Process Mining, Imitation Learning and Healthcare. Doctor parsing and patient parsing are tasks that can and should be solved algorithmically by applying PbE techniques to datasets like MIMIC-IV. Healthcare is, after all, one of the most exciting applications of Programming Synthesis.
Feel free to click around this diagram to learn more:
If you’re interested in working on this topic (or already are!) feel free to send me an email, let’s see if we can collaborate. If you’re interested and are a master’s student, there is also an opportunity to write your master’s thesis the field of doctor or patient parsing under my supervision. You are a great candidate if you know a lot about at least two of the topics and want to know a lot about all of them:
- Machine Learning
- Deep Learning
- Natural Language Processing
- Program Synthesis
If you’re an EU taxpayer, you have probably paid for the writing of this post. Thank you.