BMI2207C CA2 Flashcards
Difference between the EHR and EMR.
Electronic Medical Records (EMR) :
- Digital version of the paper charts and structured data in hospitals or clinical office
- System to capture medical records of a patient
- Does not travel easily out of hospitals
- Examples : EPIC and Allscripts
Electronic Health Records (EHR) :
- Total health of the patient going beyond standard clinical data collected in the provider’s office or hospital
- Reach out beyond the health organisation that originally collects the information
- Includes information from all clinicians in the patient care
- Built to share information with other healthcare providers like labs and specialists
- Examples : NEHR
Explain what the term “Meaningful use” means.
It is a term to get HCPs to begin sorting and sharing of health data electronically to be able to better improve clinical processes and health outcomes for patients
Functions of “meaningful use”.
- Improve quality of patient care
- Engaging patients in health
- Easier to coordinate care
- Improve overall health of a given patient population
- Secure and protects people’s health information
Explain the stages in meaningful use, and state which stage Singapore’s EHR is in.
Stage 1 : Focused on getting healthcare providers to adopt EHRs and store clinical data electronically
Stage 2 : Encourage healthcare professionals and institutions to then use the data and technology to improve the quality of care for their patients and make it easier to exchange information within and between organisations
Stage 3 : Centred on leveraging EHRs and clinical data to improve health outcomes, and ease reporting requirements to align with other government health programs
**Singapore’s EHR meets Stage 1 definition of “meaningful use”
Explain what “Directed Exchange” means.
Health information is sent directly to other providers over an encrypted secure connection.
Explain what “Query-based exchange” means.
Data is requested by providers to a central Health information exchange (HIE).
Explain what “Consumer mediated exchange” means.
Patients are involved in the collection and transmission of healthcare data to providers.
List the 12 persistent risks in AI.
- Disinformation
- Safety and security
- Black box problem (cannot see what process AI has undertaken before it has generated data output)
- Ethical concerns (Non-maleficence, Beneficence, Autonomy, Manipulation)
- Bias
- Instability
- Hallucinations in LLMs (AI giving false outputs and causing unnecessary panic)
- Unknown unknowns (we are not really sure how AI is going to react, blindspots preventing anticipation of AI behaviour)
- Job loss and social inequalities
- Environmental impacts
- Industry concentration
- State overreach
Explain the relationship between AI, ML, DL, NLP, LLM and Conversational AI.
- Artificial Intelligence includes ML, DL, Conv. AI, LLM, NLP
- ML includes DL, LLM, Conv. AI, and can intersect with NLP
- DL can intersect with Conv. AI and LLM
- Conv. AI and LLM are both subsets of ML and NLP
- NLP overlaps with ML, but has independent parts as well
**Data science, data management, descriptive analytics and visualisation is NOT AI and is a complete other subset
List the 2 roadblocks of modern AI.
- Common sense problems and implicit knowledge – Cognitive and contextual limitations require excessive exceptions to be inputted to model interactions
- Lack of training/labelled data – Amount of inputs are never enough, especially for cellular images
Differentiate supervised and unsupervised learning in machine learning.
Supervised learning :
- Classification is the problem of predicting the correct category/label for a input object
- Regression is the prediction of continuous response
Unsupervised learning :
- Clustering is the problem of identifying implicit groupings in the data
Explain regression analysis.
Regression analysis is a set of statistical processes for estimating the relationship between a dependent variable and one or more independent variable
- Linear regression : Models relationship b/w 2 variables and estimates the value of a response by using line of best fit
- Multiple regression : Models relationship between variables (with >1 independent variable) and method of least squares is used to find p-dimensional plane
- Logistic regression : Predicts a binary outcome and is a sigmoid function to map predictions to probabilities ; independent variables can be categorical/numeric, but dependent must ALWAYS BE CATEGORICAL due to binary output
Define artificial neural network (ANN).
ANN represents a prediction model as a series of multi-layered interconnected nodes functioning similar to neurons.
- Learning in ANN involves adjusting the weights of various inputs to minimise errors in output
- Model will back propagate to correct weights until the error rate converges
Explain Deep Learning.
- Subset of ANN that uses multi-layered neural networks
- Allows for more complex feature detections and requires more training data
- Image recognition, natural language, processing and autonomous driving
Define clustering and the 2 categories.
Clustering is defined as the classifying of data into points based on similarities, extracting underlying groupings without labels.
K means clustering : Clustering based on centroids, which changes its centroids until intra-cluster distance is minimised and inter-cluster data is maximised
Density based clustering (DBscan)
How do we conduct evaluation of ML models?
- Choose the best model in the market and use it as the benchmark for comparison with other models
- Ensures models are generalised well, not overtrained or under-trained
- Hyper-parameter tuning where hyperparameters can significantly affect model performance
- Interpreting pitfalls of models
