CARE COLLEGE OF ENGINEERING, TRICHY
DEPARTMENT OF AI&DS
Acad year : 2023-24 (Even) III year/ VI semester Elective paper
AD3002 Health Care Analytics
Lecture
Unit -1 Introduction to HealthCare Analysis 7 &8
No
Topic Probabilistic reasoning and Bayes Theorem
Bloom’s
Learning Outcome (LO) At the end of this lecture, students will be able to
Knowledge Level
LO1 Define causal reasoning Remembering
LO2 Explain the causal reasoning Understanding
LO3 Describe the probabilistic reasoning Understanding
LO4 Explain the Bayes theorem Understanding
Lecture Notes
The Clinical Decision Support Systems (CDSS) were initially designed with “artificial
intelligence” in mind, early medical expert systems usually focused on imitating the decision-
making processes of real-life experts. That is ironic, given that most early CDSSs such as MYCIN
(MYCIN was an early expert system designed to provide diagnostic support for infectious
diseases. Developed in the 1970s at Stanford University, it was one of the pioneering systems in the
field of artificial intelligence for healthcare) and Pathfinder do not “reason” the way humans do.
They have no significant comprehension of human anatomy/physiology, are unable to recognize
temporal concepts and (most importantly) have no ability to learn/deduce new facts. However,
within their narrow range of knowledge, they have been demonstrated to successfully make
decisions on par with a real human expert.
Unfortunately, as their domain of knowledge is broadened, performance decreases.
Particularly, the ability to make inferences from “first principles” and understand the effects time
may have on disease processes are crucial to building robust systems with more human-like
capabilities. Fortunately, many promising methods for enabling inferential/temporal reasoning
within CDSSs exist. In addition to giving the system more human-like capabilities, they have also
reduced the burden of large-scale calculations in networks and have helped in handling conflicting
rules within the knowledge base.
Causal Reasoning:
Causal reasoning and probabilistic reasoning are related but distinct concepts. Causal reasoning
refers to the process of understanding and explaining cause-and-effect relationships between events
or variables. Probabilistic reasoning, on the other hand, involves reasoning under uncertainty by
assigning probabilities to different events or outcomes.
Probabilistic Reasoning:
One of the popular reasoning is probabilistic reasoning. As Bayesian reasoning fell out of
fashion during the 1970s, new reasoning systems known as “belief networks” began to develop.
These were directed acyclic networks consisting of nodes containing both conditional and
probabilistic data. These nodes had parent-child relationships (with the parent nodes “pointing” to
their children). The effects the parent nodes had on their children were represented by conditional
probability tables.
What made these networks popular was their ability to recognize the conditional distinctness
of different findings—a task that was very difficult with Bayesian networks (particularly in large
domains). This was accomplished through the use of causal relationships when designing the
networks along with catch-all probability estimates (or “noise parameters”), allowing for a
significant amount of leeway in terms of “correctness.”
Probabilistic reasoning is a versatile and powerful tool that allows decision-makers to make
informed choices in situations where uncertainty is present. It has applications in various fields,
including artificial intelligence, statistics, finance, and healthcare. Probabilistic reasoning concepts
are integral to various aspects of healthcare, contributing to informed decision-making and risk
assessment. Here's an explanation of some key probabilistic reasoning concepts in healthcare:
1. Bayesian Inference:
• Concept: Bayesian inference involves updating the probability of a hypothesis as
new evidence becomes available. It combines prior beliefs with new data to obtain a
posterior probability.
• Healthcare Application: Bayesian inference is used in diagnostic reasoning. For
example, updating the probability of a disease based on the patient's symptoms and test
results over time.
2. Probabilistic reasoning:
• Concept: Probabilistic reasoning is crucial in assessing the likelihood of various
outcomes or events. It involves quantifying the probability of specific risks associated with
patient health or treatment options.
• Healthcare Application: Risk assessment models help healthcare professionals
estimate the likelihood of complications, side effects, or treatment success. For instance,
predicting the risk of complications after surgery.
3. Decision Trees:
• Concept: Decision trees are graphical models that represent decision-making
processes, incorporating probabilities at each decision point. They help visualize and
quantify the likely outcomes of different choices.
• Healthcare Application: Decision trees are used in treatment planning, guiding
clinicians through a series of decisions based on the probability of success or failure at each
step.
4. Markov Models:
• Concept: Markov models are used to analyze and predict the progression of events
over time, considering transition probabilities between different states.
• Healthcare Application: Markov models are applied in chronic disease
management, such as modeling the progression of a disease over time and estimating the
impact of interventions.
5. Monte Carlo Simulations:
• Concept: Monte Carlo simulations involve generating random samples to model the
probability distribution of possible outcomes. This approach is used when deterministic
solutions are challenging to obtain.
• Healthcare Application: Monte Carlo simulations are employed in healthcare
resource allocation, capacity planning, and evaluating the cost-effectiveness of healthcare
interventions.
These probabilistic reasoning concepts enhance the ability of healthcare professionals to
make informed decisions in the face of uncertainty, promoting personalized and evidence-based
care.
Belief network:
Belief networks, also known as Bayesian networks or probabilistic graphical models, are
powerful tools in healthcare for representing and reasoning about uncertain and complex
relationships among variables. These networks are particularly useful in situations where there is a
need to model and understand the interplay of various factors in a probabilistic manner. Here's how
belief networks are applied in healthcare:
1. Modeling Medical Knowledge:
• Representation of Relationships: Belief networks allow healthcare professionals to
model and represent the relationships between different medical variables, such as
symptoms, diseases, patient demographics, and test results.
• Probabilistic Dependencies: The edges in a belief network denote probabilistic
dependencies between variables. For example, the presence of certain symptoms might
increase the likelihood of a particular diagnosis.
2. Clinical Decision Support Systems (CDSS):
• Diagnostic Support: Belief networks can be employed in clinical decision support
systems to assist healthcare practitioners in the diagnostic process. By incorporating
probabilistic information, these systems can provide more nuanced and personalized
recommendations.
3. Treatment Planning:
• Effectiveness of Treatments: Belief networks can model the effectiveness of
different treatment options based on patient characteristics, medical history, and other
relevant factors. This helps in personalized treatment planning.
4. Epidemiological Studies:
• Disease Spread Modeling: In epidemiology, belief networks can be used to model
the spread of diseases in populations, considering factors such as social interactions,
geographical locations, and demographic information.
5. Risk Assessment:
• Patient Risk Profiles: Belief networks can assist in assessing the risk profile of
patients by considering multiple risk factors. This is valuable in preventive care and early
intervention strategies.
6. Genetic and Genomic Analysis:
• Genetic Risk Factors: Belief networks can incorporate genetic and genomic
information to assess the likelihood of specific genetic conditions or the risk of developing
certain diseases based on genetic markers.
7. Clinical Research and Trials:
• Patient Stratification: Belief networks help in stratifying patients in clinical trials
based on various factors, optimizing the selection of participants and improving the
efficiency of the trials.
8. Inference and Updating:
• Dynamic Updating: Belief networks support dynamic updating of probabilities as
new information becomes available. This is especially useful in healthcare scenarios where
patient data is continuously generated.
9. Explanatory Power:
• Explanatory Models: Belief networks provide explanatory models that can help
healthcare professionals and patients understand the reasoning behind a particular diagnosis,
prognosis, or treatment recommendation.
In the healthcare domain, belief networks enhance decision-making by combining clinical expertise
with probabilistic reasoning, thereby improving the understanding and management of uncertainty
in medical scenarios. They contribute to more informed and personalized healthcare practices.
Bayesian Models: The Bayes theorem is one of the most important principles in probability theory
and mathematical statistics; it provides a link between the posterior probability and the prior
probability, so we can see the probability changes before and after accounting for a certain random
event. The formulation of the Bayes theorem is
where Pr(Y|X) is the probability of event Y, conditional upon event X. Based on this theory, there
are two widely used implementations: na¨ıve Bayes and the Bayesian network. Both of these
approaches are commonly studied in the context of clinical prediction.
Bayesian Network:
Although the naive Bayes classifier is a straightforward implementation of Bayesian classifier,
in most real-word scenarios there are certain relationships that exist among the attributes. A
Bayesian network introduces a directed acyclic graph (DAG), which represent a set of random
variables by nodes and their dependency relationships by edges. Each node is associated with a
probability function that gives the probability of the current node conditional on its parent nodes’
probability.
If the node does not have any parents, then the probability function will be the prior probability
of the current node. More specifically, in decision making or prediction problems, this Bayesian
network can be viewed in terms of a hierarchical structure. Only the independent attributes that
have prior probability are in the top level. For example, in Figure 10.1, there are 5 attributes that
contribute to the output; among them “Smoking (Attribute 3)” and “Family history of heart disease
(Attribute 5)” do not have any predecessors, so we can compute the prior probabilities Pr(Smoking)
and Pr(Family history of heart disease) directly; “Aortic rupture (Attribute 1)” and “Hypertension
(Attribute 4)” are in the second level, and their conditional probabilities are Pr(Aortic
rupture|Smoking) and Pr(Hypertension|Smoking) respectively; “Stroke (Attribute 2)” is in the third
level and its conditional probability is Pr(Stroke|Aortic rupture, Smoking, Hypertension). “Heart
rate (Attribute 6)” and “Blood pressure (Attribute 7)” are two medical observations, and “Heart
attacks” is the disease that needs to be predicted.
The main intuition of the Bayesian classifiers is comparing Pr(Y = y|Xi) for different y ∈ Y where Y is the
label set and choosing the most possible label (ychosen) as the estimated label for individual Xi =
(xi1,xi2,...,xip). From Equation (10.23), we can see that, in order to calculate Pr(Y = y|Xi) we need to know
Pr(Xi |Y = y), Pr(Y = y), and Pr(Xi). Among these three terms, Pr(Y = y) can be easily estimated from the
training dataset; Pr(Xi) can be ignored because while comparing different y’s; the denominator in the
Equation (10.23) remains a constant. Thus, the main work in Bayesian classifiers is to choose the proper
method to estimate Pr(Xi |Y = y).
Thus, to classify a test record Xi based on the Bayes theorem and ignore the Pr(Xi), the conditional
probability for each possible output y in the label set Y can be represented as: Pr(Y = y|Xi) ∝ Pr(Y
= y) p ∏ k=1 Pr(xik|Y = y). (10.25) Finally, the class label ychosen, which maximizes the Pr(Y = y)∏
p k=1 Pr(xik|Y = y) is chosen to be the output.
Problems:
A diagnostic test for a rare disease has a sensitivity of 90% and a specificity of 95%. If the
prevalence of the disease in the population is 1%, what is the probability that a person testing
positive actually has the disease, according to Bayes' Theorem?
a) 9% b) 47.5% c) 90% d) 95%
Answer: b) 47.5%
Explanation: Bayes' Theorem is expressed as:
P(A∣B)=P(B)P(B∣A)⋅P(A)
Where:
After calculating this expression, we get approximately 0.475 or 47.5%. Therefore, the correct
answer is option b) 47.5%.
Qn BT.Knowledge
Questions Answer
No Level
A diagnostic test for a rare disease has a sensitivity
1. of 90% and a specificity of 95%. If the prevalence b. Understanding
of the disease in the population is 1%, what is the
probability that a person testing positive actually has
the disease, according to Bayes' Theorem?
a) 9%
b) 47.5%
c) 90%
d) 95%
2. In a Bayesian network representing a medical
diagnosis system with nodes for symptoms and b. Understanding
diseases, what does a directed edge between a
symptom node and a disease node indicate?
a) The probability of the disease given the symptom.
b) The probability of the symptom given the disease.
c) Mutual exclusivity between the symptom and the
disease.
d)No probabilistic relationship between the
symptom and the disease.
Students have to prepare answers for the following questions at the end of the lecture
Marks CO Bloom’s
Qn
Question Knowledge
No
Level
1. Discuss in detail of probabilistic reasoning 15 1 K3
2. Explain the Baysean network 15 1 K3
CARE COLLEGE OF ENGINEERING, TRICHY
DEPARTMENT OF AI&DS
Acad year : 2023-24 (Even) III year/ VI semester Elective paper
AD3002 Health Care Analytics
Lecture
Unit -1 Introduction to HealthCare Analysis 9
No
Weighted sum approach – Identify variables, assigned sum, Weight Initialization, Decision
Topic
making, Artificial Neural Network(ANN), Weighted approach in ANN
Bloom’s
Learning Outcome (LO) At the end of this lecture, students will be able to
Knowledge Level
LO1 Define weighted sum approach Remembering
LO2 Explain the process involved in weighted sum approach Understanding
LO3 Describe the Artificial Neural Network and layers Understanding
LO4 Explain the weighted sum approach used in various networks Understanding
Lecture Notes
The weighted sum approach is a common technique used in healthcare analytics to combine
multiple variables or factors into a single composite score or index. This approach is often
employed when there is a need to aggregate diverse information and prioritize or rank different
elements based on their relative importance. In the context of healthcare analytics, the weighted
sum approach can be applied in various ways, such as risk assessment, patient scoring, or resource
allocation.
Here's a general overview of how the weighted sum approach works in healthcare analytics:
Identify Variables:
Begin by identifying the relevant variables or factors that you want to consider in your analysis.
These could include patient demographics, clinical measurements, historical data, or any other
relevant metrics.
Assign Weights:
Assign weights to each variable based on their relative importance or impact on the outcome. The
weights represent the significance of each variable in contributing to the overall score. These
weights can be determined through expert opinion, statistical analysis, or a combination of both.
Normalize Data:
If the variables have different scales or units, it's essential to normalize the data to bring them to a
common scale. This ensures that each variable contributes proportionally to the final score.
Calculate Weighted Sum:
For each individual, calculate the weighted sum by multiplying the normalized value of each
variable by its assigned weight and summing up these products. The formula for the weighted sum
(WS) is typically:
Wi is the weight value assigned to variable x.
Interpretation:
The resulting weighted sum provides a single composite score for each individual or case.
This score reflects the combined impact of the various factors, with higher scores indicating higher
risk, severity, or priority, depending on the context of the analysis.
Decision Making:
Use the composite score to inform decision-making processes, such as resource allocation,
patient prioritization, or intervention strategies. Organizations can set threshold values to categorize
individuals into different risk groups or priority levels.
It's important to note that the success of the weighted sum approach relies on the accuracy of
variable selection, appropriate assignment of weights, and the validity of the underlying
assumptions. Regular validation and refinement of the model based on real-world outcomes and
feedback are crucial for ensuring its effectiveness in healthcare analytics. Additionally, more
advanced techniques such as machine learning models may be used to optimize variable selection
and weight assignment in a data-driven manner.
The lexicographic method is a technique used to solve multi-objective optimization
problems. It involves arranging each of several cost functions in a hierarchy reflecting their relative
importance. The method then selects the solution that optimizes the most important cost function,
subject to the constraint that the other cost functions are optimized as well as possible1.
In other words, the lexicographic method is a way of prioritizing objectives in a decision-
making process. It is often used when there are multiple objectives to be optimized, and the
objectives cannot be optimized simultaneously.
An Artificial Neural Network (ANN) is an information processing paradigm that is inspired by the
brain. ANNs, like people, learn by examples. An ANN is configured for a specific application, such
as pattern recognition or data classification, through a learning process. Learning largely involves
adjustments to the synaptic connections that exist between the neurons.
Artificial Neural Networks (ANNs) are a type of machine learning model that are inspired by the
structure and function of the human brain. They consist of layers of interconnected “neurons” that
process and transmit information.
There are several different architectures for ANNs, each with their own strengths and weaknesses.
Some of the most common architectures include:
Feedforward Neural Networks: This is the simplest type of ANN architecture, where the
information flows in one direction from input to output. The layers are fully connected, meaning
each neuron in a layer is connected to all the neurons in the next layer.
Recurrent Neural Networks (RNNs): These networks have a “memory” component, where
information can flow in cycles through the network. This allows the network to process sequences
of data, such as time series or speech.
Convolutional Neural Networks (CNNs): These networks are designed to process data with a grid-
like topology, such as images. The layers consist of convolutional layers, which learn to detect
specific features in the data, and pooling layers, which reduce the spatial dimensions of the data.
Autoencoders: These are neural networks that are used for unsupervised learning. They consist of
an encoder that maps the input data to a lower-dimensional representation and a decoder that maps
the representation back to the original data.
Generative Adversarial Networks (GANs): These are neural networks that are used for generative
modeling. They consist of two parts: a generator that learns to generate new data samples, and a
discriminator that learns to distinguish between real and generated data.
The model of an artificial neural network can be specified by three entities:
The model of an artificial neural network can be specified by three entities:
Interconnections
Activation functions
Learning rules
Interconnections:
Interconnection can be defined as the way processing elements (Neuron) in ANN are connected to
each other. Hence, the arrangements of these processing elements and geometry of interconnections
are very essential in ANN.
These arrangements always have two layers that are common to all network architectures, the Input
layer and output layer where the input layer buffers the input signal, and the output layer generates
the output of the network. The third layer is the Hidden layer, in which neurons are neither kept in
the input layer nor in the output layer. These neurons are hidden from the people who are
interfacing with the system and act as a black box to them. By increasing the hidden layers with
neurons, the system’s computational and processing power can be increased but the training
phenomena of the system get more complex at the same time.
There exist five basic types of neuron connection architecture :
Single-layer feed-forward network
Multilayer feed-forward network
Single node with its own feedback
Single-layer recurrent network
Multilayer recurrent network
1. Single-layer feed-forward network
In this type of network, we have only two layers input layer and the output layer but the input layer
does not count because no computation is performed in this layer. The output layer is formed when
different weights are applied to input nodes and the cumulative effect per node is taken. After this,
the neurons collectively give the output layer to compute the output signals.
Multilayer Feed Forward Network:
Main layers in ANN:
Input Layer
Output Layer
One or Two Hidden Layer.
Output function is the is the weighted averages of all the inputs used in Input’s a
Assessment questions to the lecture
Bloom’s
Qn
Question Answer Knowledge
No
Level
1. What are the layers used in ANN
a. Input Layer d. Remembering
b. Input Layer, output layer
c.Hidden Layer, output layer
d. Input, Hidden layer, Output Layer
The weighted sum approach is used to combine Multiple,
2. _______variable into _________ index single Remembering
Students have to prepare answers for the following questions at the end of the lecture
Marks CO Bloom’s
Qn
Question Knowledge
No
Level
1. Explain the weighted sum approach in ANN? 15 1 K3
