Learning objectives:
1) Clarifying terminologies: Diagnostics (process) vs Diagnosis (product)
2) Decision doesn't end at diagnosis: resolving therapeutic uncertainty (the final product)
3) Introduction to a real world, 24x7, AI LLM user driven diagnostic and clinical decision support workflow with it's past and current background
4) The AI and human driven user vs developer perspective in clinical decision making
Introduction:
Diagnostics (what is the process):
Conversational Transcripts from an interaction between a developer and user (to also understand the differences in their perspective)
[30/08, 22:06]hu1: Our wearable device captures a comprehensive set of physiological parameters, including:
Cardiac
• 4 Lead ECG
• Pulse Rate
• HR / HRV
• Stroke Volume
• Cardiac Output
• Cardiac Index
• Pulse Pressure
• Arterial Mean Pressure (MAP)
• SVR
• NiBP (under validation)
Respiratory
• Pleth-SpO₂
• Respiratory Rate
• Thoracic Impedance
Other
• Body Temperature
• Accelerometer (patient position)
All data is collected with *medical-grade accuracy*, seamlessly streamed to a bedside monitor (tablet) and securely transmitted to the cloud for *AI & algorithmic computation*. This enables near real-time alerts with *Early Warning Score (EWS)*, *Sepsis prediction*, and *Heart Failure (HF)* detection.
We are actively advancing toward *CDSCO* regulatory clearance
[31/08, 12:36]hu2: Medical grade accuracy is good but one needs to keep the broader process of diagnostics in mind along with the initial product (diagnosis) and final product (therapeutic outcome).
These data points are currently given a lot of weightage in diagnostics so much so that the meaning of the term "diagnostics" has changed to mean something limited to lab or imaging based parameters and this is something that needs to be debated.
Diagnostics is a collage of multimodal data points tied into an ontology:
It begins with patient history as in life events data points that can be discreet or continuous.
Trying to understand patient requirements from life events generated data can be analogous to prompting an AI LLM!
The other milestones in a diagnostic workflow process:
Clinical examination
Physician or Machine driven Objective vs patient driven Subjective data points
Labs
X ray
MRI
How much weightage can one give to each in the decision making workflow?
Diagnostics (what it isn't)!
Throwing a few lab reports (currently masquerading as objective data but actually suboptimal data) to a human posing as a physician agent and expecting him her to create magic with suboptimal data!
Your diagnostics and decision support will be as good as the data you share! Beware of GIGO!
The role of laboratory vs clinical encounter
Methods:
Clinical problem solving workflows
Medical cognition and procedural workflows toward diagnostic and therapeutic outcomes
Layers of diagnosis:
Anatomical
Etiologic
Past published discussion on critical realist heutagogy CRH and AI driven workflows:
Narketpally syn as a case based blended learning ecosystem CBBLE and it's PaJR pajr.in platform with detailed description of CRH and our medical cognition workflow
CRH first signposted here: https://medicinedepartment.blogspot.com/2024/10/critical-realist-pedagogy-and-theatre.html?m=1
CBBLE as an early progenitor of the current Narketpally syn:
Current regular methodological uploads archived:
2024:
2025:
2020:
Results:
Recent real patient outcomes:
An individual patient problem necessitating diagnostic and clinical decision support (human as well as AI driven): https://
More individual AI driven real patient diagnostics and decision making process regularly updated here: https://
Discussion:
Decision support toward diagnosis and therapy:
The pandemic of Over-testing and Overtreatment resulting from misinformation of diagnostics:
The process of Medical cognition (meta cognition, thinking about thinking to integrate medical education with practice):
System 1 cognition: default fast thinking mode and it's evolution in the dyadic doctor patient relationship
Past lecture: https://sites.pitt.
to
System 2, the slow thinking mode with the added advantage of collective sharing (blessing of dimensionality reduction) and the disadvantage of missing data (curse of dimensionality reduction)
AI and system 2 beginning with primordial asynchronous intelligence to current day automation unleashing an era of pre-tech singularity Orwellian data capture and sharing that will eventually merge with system 1 cognition with further
No comments:
Post a Comment