Data Science-Data Analytics

© 2010 API Healthcare Corporation. Confidential Information. All Rights Reserved.
Enterprise Data Science, Data Analytics
&
Hospital Operations Management:
Concepts, Values & Interconnections.
July, 2014
Alexander Kolker, Data Scientist

“Data is the energy source (fuel) of business transformation and
Data Science/Business Analytics is the engine for its delivery.”
Data focus:
How to capture (massive) amount of raw data in different formats
(numerical values, descriptive texts, videos, audio, etc.), how to store,
retrieve, and manage it.
Business Analytics focus:
What data to collect, what metrics are needed to solve a business problem,
how to clean the raw data to make it usable, what analytic
method/technique to use, how to turn analysis into actionable managerial
decision.
 What do Data Science and Data/Business Analytics strive
to accomplish?

Take-away:
Data Science and Data/Business Analytics strive to transform the raw
data into actionable business decisions using a set of quantitative
methods, processes, and computer technologies.
Key points:
• Business analytics is NOT an extension or a side effect of collecting, keeping
and presenting / querying data in general.
• Data have business value only in a specific business context.
• Currently there is a bias to looking at data itself instead of focusing on
business context and data analysis methods and techniques.

Data Analytics:
• Seeks to provide operational observations into issues that we
- (i) either know that we know, or
- (ii) know we don’t know
by means of the simultaneous application of statistics, computer
programming and operations research to quantify system / process
performance.
Examples:
Descriptive analytics quantitatively describes the main features
of a collection of data.
Predictive analytics strives to predict relationships between
known random variables or sets of data in order to identify
(project) how an event will occur in the future.

Data Science, on the other hand, seeks to provide strategic actionable
insights in the situations where we don’t know what we don’t know.
• The primary means through which actionable insights can be found
are:
• Predictive analytics in the area of causation
• Prescriptive analytics, i.e. predictive + decision science
• Machine learning
Thus, the overall value of Data Science is exponentially increased when it is
coupled with its ‘cousin’, Data Analytics, and both are integrated into an
end-to-end Enterprise Value Chain.

Advanced Business Analytics
Predictive Analytics:
is the application of mathematical techniques to predict data that are not
currently available using some currently available data.
Predictive analytics supports business decision-making but it does not
make decisions itself.
Example:
Given my current census, what will my census be tomorrow?
Predicting the tomorrow’s census is helpful for making staffing decision but it does
not make staffing decision in itself.

Advanced Business Analytics
Prescriptive Analytics:
Recommends business decisions based on
(i) business decision rules, or (ii) mathematical optimization models.
Example:
Given predicted tomorrow’s census and acuity, how many nurses should I staff
based on this information?
Decisions are made using:
Business rules, e.g. if patient : nurse ratio mandated to 3 then staff at the
level: census/3.
or
Optimization, e.g. given a set of constraints, find the staffing level that
minimizes the overall staffing cost within this set of constraints

 Time series analysis of the past / historical data to answer the question:
“What happened- trends, seasonality, correlation?”
 Multivariate statistical analysis to answer the question:
“What factors played the most important role: significant vs. insignificant factors?”
 Predictive modeling / forecasting to answer the question:
“What will happen next?”
 Data Science / Prescriptive analytics to answer:
“Where are the opportunities- factors, resources, new directions?
What actions should be made to steer in the desired business direction?”
Examples of Business Analytics Applications

Examples of Business Analytics Problems
Business Problem The most appropriate methodology
Long-term budgeted staffing: Given variable census
and the cost of under- and over-staffing per nurse, what
is the optimal staffing that minimizes the total staffing
cost?
The “Newsvendor framework” with nurse:
patient ratios, and/or Patient Classification
System (PCS)
(Kolker, A., API Presentation, 2013)
Capacity: Given variable patient arrivals, how many
beds, procedure / operating rooms, or pieces of
equipment are needed for different services?
Discrete Event Simulation
(Kolker, A., Healthcare Management Engineering: What Does
This Fancy Term Really Mean? Springer_Briefs Series in
Healthcare Management & Economics. Springer, NY, 2012)
Staffing: Given variable patient arrivals, how many
nurses, physicians and other providers are needed for a
particular shift in order to best achieve operational and
service performance objectives?
Discrete Event Simulation/Linear
optimization
Scheduling: Given variable patient census, what is the
minimal-cost staff schedule that includes mandatory
days-off, full- and part-time, and shift rotations?
Linear optimization
Patient flow: How to organize patient move through
hospital system without undue delay in order to achieve
the system patient throughput goals?
Discrete Event Simulation
Kolker, A., Interdependency of Hospital Departments and Hospital-
Wide Patient Flows. Chapter 2, pp. 43-63. In: Patient Flow: Reducing
Delay in Healthcare Delivery. 2-nd Edition, R. Hall (Ed), Springer,
NY, 2013.

Examples of Business Analytics Problems (Cont.)
Business Problem The most appropriate methodology
Resource allocation: Is it more efficient to use
specialized resources or pooled (interchangeable)
resources (operating/procedure rooms, beds,
equipment, and staff)?
Does it make economic sense to keep some patient
service lines, or drop them at all?
Linear Optimization
Forecasting: How to forecast the future patient
volumes, unit census (demand), or transaction volumes
for short- and long-term budget planning and other
planning purposes?
Markov chains; Time series analysis
(stationary and non-stationary).
Autocorrelations. Fractal Analysis.
Performance / Productivity scoring: Given multiple
performance inputs and outputs for the unit, estimate its
productivity score relative to other similar units, and
decide on the factors that can be leveraged to improve
the productivity score of this unit.
Data Envelopment Analysis (DEA) based on
linear optimization.
(Kolker, A., API Presentation, 2014)
Discharge Planning: Given variable patient length of
stay (LOS) by acuity and/or other factors, predict the
expected number of discharges in the next time period
(say, 24 hours)
Estimation of conditional probabilities;
hazard functions.

What Makes Next Generation a truly Innovative
Product?
Next Generation Product vision is:
• Integration of
• existing products (TASS, BA, Shift_Select, PCS)
&
• Advanced Prescriptive Business Analytics aimed at helping in business
decision-making process
• The use of a single source data repository in a unified format/structure
to feed all products

TA/SS ShiftSelect
PCS BA
HRP
Functions/ Process
Unified Data
Unified Data
Base as “Fuel”
for all products
Engine for conversion “fuel”
into justified actionable
business decisions
Output
Data
Science
• Predictive
analytics
• Prescriptive
analytics
Actionable
justified
decisions

1.
Identify/Explore
Business
Opportunity
2. Formulate
and Define
Business
Problem
3. Translate
Business
Problem into
Analytics
Problem
6. Implement
and validate
customer value
5. Create the
User story and
Execution Plan
4. Define Data
(Fuel)
Needed to solve
the Analytics
Problem
Main Steps in Business Decision-Making Flow

Example
 1. Business opportunity:
Helping the customer to plan ahead in determining the optimized
budgeted staffing
 2. Business problem:
Given the unit variable shift census, what is the optimal budgeted
staffing that minimizes the total costs of staff call-on
(understaffing) and call-off (overstaffing)?
 3. Translation into Analytics Problem:
“Newsvendor” framework that minimizes the total costs of under-
and overstaffing
 4. Define data (fuel) needs:
- PCS estimation for patient total hours of care per shift,
or patient : nurse ratio broken by acuity
- unit variable census
- the costs of understaffing and overstaffing over the base rate

• 5. User Story Creation for Optimized Staffing
 Recall Census & Acuity Data
 Input Cost of Overstaffing and Understaffing
 Calculating the Cost Ratio
 Convert PCS hours of care into variable required FTE
 Calculating the empirical cumulative distribution staffing function
 Solving equation for the optimal long-term budgeted staffing

Enterprise solution data value chain
Enterprise
Data Science
Enterprise
Information
Management
Enterprise
Data Analytics
Enterprise
Architecture
Enterprise
Solutions
InfIIInformation
Specialists
Knowledge Specialists Solutions Specialists
Strong Collaboration
Strong Collaboration
Capabilities for:
Managing current
large scale data
Assets.
Includes
-relational data base
-data warehouse
-data marts
-emergence of big
data
Leverages data
assets for day-to-day
operational insights:
-staffing
-scheduling
-capacity
-forecasting
-patient flow
-productivity
-LOS
Seeks to exploit
Information and
analytic capabilities
for developing
enterprise
strategic decisions
Seek to fuse operationally
Into the capabilities of the
larger enterprise through
Architecture and Solutions

Enterprise Data Science
3. Data Prep:
Selection &
cleaning
1. Define
Business
Objectives
2. Data
Inventory:
Collection &
understanding
Determine
type & volume
of data
required to
meet
business
objective
-Define data
selection
rational
-Determine data
availability
-If not available,
develop data
collection plan
-summarize
available/
collected data
-verify data quality:
how critical are
missing
values
4. Modeling
Methodology:
Design &
Development
5. Optimization
& Evaluation
6. Deployment
Solution
within the
system
-Select data
to clean
-Integrate &
format data
for the next
step
-Select
Modeling
techniques
-Generate
test &
validation
designs
-Build model
parameters
-Assess
model
performance
-Integrate
individual
modeling
modules
-Evaluate
results
stability
- Approve
model
-Assess
deployment
Environment
-Create model
integration
strategy
-Plan model
deployment,
monitoring &
maintenance

Enterprise Data Value Circular Diagram
• The sequence of the phases is not strict and moving
back and forth between different phases is typical.
• The arrows in the process diagram indicate the
most important and frequent dependencies between
phases.
• The outer circle in the diagram symbolizes the cyclic
nature of the process itself:
• It continues after a solution has been
deployed.
• The lessons learned during the process can
trigger new, often more focused business
questions and subsequent data analysis will
benefit from the experiences of previous ones.

Data Science processes for an overall enterprise solution

Overall Key Points:
 Current Business Analytics is usually reporting or querying, i.e. presenting
data in the convenient condensed form, such as graphs, dash-boards, etc.
across various applications (clinical, revenue cycles, time & attendance,
etc.).
 Reporting / querying provides some basic business information, but it does
not directly help in business decision-making (except for simplest
straightforward situations)
 Next Generation Business Analytics should provide a means for business
decision-making based on predictive and prescriptive quantitative models
using quantitative Data Science/Computer simulation methodologies.
 Advanced Predictive and Prescriptive business analytics is a main
component of the next generation Hospital Operations Management.

Data Science-Data Analytics

More Related Content

What's hot (20)

Viewers also liked (16)

Similar to Data Science-Data Analytics (20)

More from Alexander Kolker (18)

Data Science-Data Analytics