Data Analytics for Real-World Business Problems

DATA ANALYTICS FOR SOLVING BUSINESS
PROBLEMS:
SHIFTING FOCUS FROM THE TECHNOLOGY
DEPLOYMENT TO THE ANALYTICS METHODOLOGY
Alexander Kolker, PhD
March 7, 2017
Alexander Kolker. All rights reserved 1

"Making better business decisions using data“
a co-hosted event with Accelerate Madison and Big
Data Madison Meetup
Date: February 13, 2017
Time: 5:30 PM - 7:30 PM CST
Key point:
Focusing on business outcomes rather than on data and
technology per se is getting momentum …

Some professional highlights…
• 4 business consulting projects: US Bank, Boston Consulting Group,
Children’s Hospital of Wisconsin, Ohio Hospital Association
• 12 years at GE (General Electric) Healthcare: Data Scientist
• 3 years at Froedtert Hospital: Process Simulation Leader
• 5 years at Children’s Hospital of Wisconsin: Simulation and Data
Analytics
• UW-Milwaukee Lubar School of Business-Adjunct Faculty: A graduate
course Healthcare Delivery Systems-Data Analytics
• Lead Editor and Author of 2 books, 8 book chapters, 10 reviewed papers,
18 Conference presentation in the area of operations management,
process modeling and simulation, business analyticsAlexander Kolker. All rights reserved

BIG DATA AND ANALYTICS
BACKGROUND

A bold statement to start with:
Big data without actionable analytics and business
decision-making is a ‘sleeping giant’
Big Data is a 2-part deal
1. Technology for storing and managing large
amounts of data of various nature- the current
trend
2. Methodology for helping business decision-
making using modeling and data.
This is called Analytics, it is getting momentum…
This presentation focus

Key points:
• Analytics must help in developing:
 New products
 Operational efficiency
 Business Decision support
$$$

7
WHAT WILL BE COVERED NEXT…
1. The concept of simulation analytics for studying systemic
complex business problems
Use case 1:
Analysis of the optimal staffing of a team of medical providers
using simulation methodology (with a live demonstration)
2. Analytics methodology for identifying a few contributing
variables to the organization’s financial outcome:
Use case 2:
Principal components decomposition of the large
observational dataset and regression with principal
components
3. Appendix: Food for thought… from Pierre Laplace, 1795
Alexander Kolker. All rights reserved

SIMULATE!
• In general, simulation is a process of studying complex
systems using their mathematical representation called a
model or a digital twin, e.g.
• Flight simulator-the aircraft response to the cockpit input
controls
• Nuclear plant operators simulators-reactor output
response to the various operator inputs
• Surgical and physiology procedures simulators on
mannequins
•Our focus here is simulation of business operations

Key Point:
The most powerful and versatile simulation methodology for
analyzing manufacturing, finance, healthcare, military and
other business operations is Discrete Event Simulation
Taken from a LinkedIn post on Data Science Central

Discrete Event Simulation (DES) Methodology.
What is it?
•A discrete event simulation (DES) model mimics a
system’s dynamic behavior as the system transitions
from state to state
(compare to Data Science approach: map an output to the input through
a black box model or algorithm)

The validated model is used for predicting various
scenarios of the future system’s responses to the
random inputs in a virtual reality
Key points:
•The simulation model is not a ‘black box’. It is a scalable
digital twin of the reality
•The model reflects what’s actually happening in the system
• This capability gives a sense of the expected system’s
output before incurring the cost and risk of the
business solution implementation
(compare to Data Science validation and cross-validation of a ‘black box’ model for
predicting the future outcomes…)

Use case 1
Analysis of the performance and the
optimal staffing
in an Endoscopy Unit using
Discrete Event Simulation
Presented at the:
5-th International Conference on Healthcare Systems, October, 2008;
and
IEEE Workshop on HealthCare Modeling and Simulation, February 18-20, 2010,
Venice, Italy

Problem Description
• The inevitable variability of the admission, recovery and
procedure time due to unforeseen medical complications
and delays result in some unit performance issues:
a long patient wait time to schedule procedures
not meeting daily patient demand for procedures
underutilization of the available capacity and staff
overtime
dissatisfaction of patients and medical staff
There has also been a lower than anticipated revenue
stream

The objectives of this work were:
(i) to analyze the main factors that contribute to the
inefficient patient flow and process bottlenecks,
and
(ii) to propose a more efficient patient scheduling and
staffing allocation aimed at increasing the number of
served patients, reducing procedure delays, and staff
overtime
Business Problem - Project Goal

The Endoscopy Unit High Level Process
Patients arrive at the
admission area
Patients are seen by the
admission nurse
Patients are attended by the
procedure nurse
Assigned doctors perform
procedures
Patients move to the
recovery area where they
are attended by the
recovery nurse

High Level Model Outline
• Admission, procedure and the patient recovery
duration are random variables
• These variables are represented as the best fit
statistical distributions built into the simulation model
• Each patient is assigned his/her attributes:
 scheduled arrival time
 procedure type
 assigned doctor’s name

Baseline Simulation Model Layout

What happens in the Exam Rooms?
if Proc_Type=col AND Doc_name=Bajaj AND Wk_Day=Fri Then
{
jointlyget (RN_WF and Tech_TF and D_Bajaj) OR (2 RN_WF and D_Bajaj)
Time (T(30,40,40) min)
Free all
}
else
if Proc_Type=egd AND Doc_name=Bajaj AND Wk_Day=Fri Then
{
jointlyget (RN_WF and Tech_TF and D_Bajaj) OR (2 RN_WF and D_Bajaj)
Time (T(10,20,20) min)
Free all
}
else
if Proc_Type=ERCP AND Doc_name=Dua AND Wk_Day=Fri Then
{
jointlyget (RN_WF and Tech_TF and D_Dua) OR (2 RN_WF and D_Dua)
Time (T(70,80,80) min)
Free all
}
Key Point:
Capturing multiple resources with different time distributions for different
procedures requires some coding…

Typical Input Data Format
Annual patient volume is ~10,000 patients
Key
Source
Destination
Nam
e
Action
Logic
W
eek
W
eekday
Tim
e
Quan
10 patient Late_Pt_arrival_adjustment_
Proc_Type=col
Doc_name=bajaj
Wk_day=Mon
1 Mon 7:00 AM 1
Proc_Type=egd
Doc_name=massey
Wk_day=Mon
1 Mon 7:00 AM 1
Proc_Type=col
Doc_name=johnson
Wk_day=Mon
1 Mon 7:00 AM 1
Proc_Type=egd
Doc_name=massey
Wk_day=Mon
1 Mon 7:20 AM 1
Proc_Type=col
Doc_name=bajaj
Wk_day=Mon
1 Mon 7:40 AM 1
Proc_Type=col
Doc_name=johnson
Wk_day=Mon
1 Mon 7:40 AM 1
Proc_Type=col
Doc_name=massey
Wk_day=Mon
1 Mon 7:40 AM 1
Proc_Type=egd
Doc_name=bajaj
Wk_day=Mon
1 Mon 8:20 AM 1

Typical information (data) usually required to populate a
DES model:
• Arrival pattern and quantities: periodic, random,
scheduled, daily pattern, etc.
• The time that the entities spend in the activities, i.e.
service time.
This is usually not a fixed time but a statistical distribution.
• Capacity of each activity, i.e. the max number of entities
that can be processed concurrently in the activity
• Routing types that connect structural elements: %,
conditional, alternate, create, renege, etc.
•Resource assignments: quantity and scheduled shifts

Live simulation demonstration is included
here: call simulation ProcessModel
patient arrivals, shifts for nurses,
technicians and doctors, stat-fit distribution

Some Simulation Scenarios
Scenario 1- The Original Model –Baseline-used for model
validation and testing
Scenario 2 - One additional doctor scheduled part time for
11 hours per week
Scenario 3 - Change in the patient arrival schedule with
10% reduction in inter-arrival time with one additional
doctor
Scenario 4 - Cross-training of the admission and recovery
nurses
Scenario 5 - Adding a part-time nurse
Scenario 6 - Adding a part-time scope-cleaning tech
Scenario 7 – ladder nurse shifts, change breaks and lunch
time
Scenario 8 – combined Scenarios 2, 3 and 4, and all together

Simulation outcome example:
Scenario 1 vs. Scenario 2+Scenario 4 (additional part-time
doctor for 11 hours/week + cross-trained nurses):
39
34
29
35
23
44
40
30
40
23
0
5
10
15
20
25
30
35
40
45
50
Monday Tuesday Wednesday Thursday Friday
Days of the Week
NumberofPatients
Scenario I Scenario II
Weekly Total
Scenario I 160
Scenario II 177
The number of patients
increase: 17
Overtime, hours
Scenario I 28.2
Scenario II 20.9
Reduced doctors’
overtime: 7.3 hrs

Financial Cost-Benefit Estimate
Typical average colonoscopy patient charge is about $2,500
(Colonoscopy is a major GI procedure)
Nurse overtime rate is 1.5 times of the regular pay (about $30/hr)
Typical GI doctor’s annual pay is about $360,000, i.e. ~$360 / hr
Weekly revenue from additional 17 patients is 17 *$2,500 = $42,500
Reduced overtime cost for nurses and doctors is
7.3 hrs*($30*1.5+$360)= $2956
Cost of additional doctor (working 11 hrs): $360*11= $3960
Additional revenue that the additional doctor brings in is about
$42,500 + $2956 - $3960 = $41,496 per week

27
Concluding Key Points:
So how can you tell if simulation is right for you?
• This is methodology of choice for analyzing the dynamic behavior
of the complex systems/processes with random components
• There is a big decision to make with high potential for failure or
reward
• Provides a framework for experimenting with the system
and testing various business scenarios
• Reveals unintended consequences of business solutions
• Commitment to use the findings and recommendations, even if
they are not what you want to hearAlexander Kolker. All rights reserved

Use case 2
Analytics methodology for identifying a few
contributing variables to the organization’s financial
outcome:
Principal components decomposition of the large
observational dataset and regression with Principal
components
Reference:
A. Kolker. Management Engineering for Effective Healthcare Delivery: Principles and
Applications, IGI-Global, 2011, Chapter 1.
A. Kolker. Healthcare Management Engineering. What Does this Fancy Term Really
Mean? Chapter 5. Springer-Briefs in Healthcare Management & Economics, NY, 2012

• The large local hospital plans a major market share
expansion to improve its long-term financial viability
Business Problem - Project Goal
• The management wants to know what population
demographic factors and population disease prevalence
specific to the local area zip codes are the most important
contributors to financial contribution margin (CM $)?
Note: Contribution margin is defined as the difference between all
payments collected from patients and the patient variable costs.

Plan of the problem attack
• Step 1
Demographics data matrix (total 38 variables) to be analyzed
for the top 10 ZIPs using Principal Component decomposition.
• Step 2
Regression analysis to be performed that relates $ CM and
principal components of the original data matrix.
• Step 3
By analyzing eigenvectors for only statistically significant principal
components, conclusions to be made which demographic variables
are the biggest contributors for the top 10 ZIPs

Description of Data
A set of population demographic data was collected for
local area zip codes and the corresponding median
contribution margin for each zip code (CM $).
The following groups of demographic variables and
disease prevalence data were collected for each zip
code as percentage of the total zip code population:

• 4 Age categories:
 18-34
 35-54
 55-64
 65+
• 4 Educational categories:
 BS/BA degree and higher,
 Associate/Professional degree,
 high school diploma,
 no high school diploma

• 4 Income categories:
 less than $50K
 $50 - $75K
 $75K - 100K
 $100K +
• 5 occupational categories:
 Healthcare, Labor,
 Professional/Administrative,
 Public Service,
 Service industry
• Gender: male, female
• 5 Race categories: African American, Native American, Asian,
White, Other

• 14 disease categories:
 BMT
 Medical Oncology
 Surgical Oncology
 Cardiology
 Cardiothoracic surgical
 Vascular surgical, Digestive
 Medicine/Primary care
 Musculoskeletal
 Neurology
 Transplant
 Trauma, Unassigned
 Women Health
• There are total 38 data variables included in the data base.

Issues with direct use of data for regression:
• In large observational data sets with the dozens variables
some of them are inevitably correlated
• Correlation means that some information is redundant
• This redundant information in the data makes it difficult
to attribute the contributions of each variable to the
output
This issue is called Multicollinearity!!

Illustration of some pairwise correlation:
Correlation coefficient of the variables
'No high school’ and ‘Annual income less $50K’: 0.93
vs.
Correlation coefficient of the variables
‘Professional Degree’ and ‘Annual income less $50K’: - 0.87

Illustration of the regression disaster with all original
data (38 variables)
CM $ =4130333+41195*18-24 years–39029*25-34 years+
11836*35-44years+2894*45-54 years+5507*55-59 years+
209919*60-64 years-142258*65-74 years+53373*75 years+ -
2665632*AD–2662185*BD-2620383*PhD- 2649374*HS - 2648440
Less HS - 2687756 MD - 2717506 ProD- 2665190 Some Coll -
2692213 Some HS - 2398380 Less $15K- 2386133 $15K to $25K
- 2493006 $25K to $35K - 2413833 $35K to $50K- 2398657
$50K to $75K - 2455023 $75K to $100K - 2434483 $100K to
$150K- 2404935*$150K to $250K - 2414342 $250K to $500K -
2393024 $500K+ 947225 Health Care + 954055 Labor + 966787
Professional/Administrative+ 954355 Public Service +
960649* Service Industry+………..
Regression diagnostics:
R-Sq = 67.1% R-Sq(adj) = 8.6%
Huge variances inflation factors VIF:

Predictor Coef SE Coef T P VIF
Constant 4130333 4378828 0.94 0.358
18--24 years 41195 32885 1.25 0.226 13.820
25--34 years -39029 24759 -1.58 0.132 23.274
35--44 years 11836 30294 0.39 0.701 9.458
45--54 years 2894 44603 0.06 0.949 25.180
55--59 years 5507 162937 0.03 0.973 89.682
60--64 years 209919 157301 1.33 0.199 65.101
65--74 years -142258 66336 -2.14 0.046 43.529
75 years+ 53373 36529 1.46 0.161 26.059
AD -2665632 3334182 -0.80 0.434 90827.662
BD -2662185 3342475 -0.80 0.436 2400778.419
PhD -2620383 3375609 -0.78 0.448 20953.952
HS -2649374 3333923 -0.79 0.437 1711185.583
Less HS -2648440 3329576 -0.80 0.437 575442.669
MD -2687756 3321036 -0.81 0.429 389134.963
ProD -2717506 3320805 -0.82 0.424 161574.141
Some Coll -2665190 3325834 -0.80 0.433 256129.161
Some HS -2692213 3334397 -0.81 0.430 1402053.683
Less $15K -2398380 2972893 -0.81 0.430 1398310.925
$15K to $25K -2386133 2983525 -0.80 0.434 429011.942
$25K to $35K -2493006 2994782 -0.83 0.416 281665.965
$35K to $50K -2413833 2973178 -0.81 0.427 253783.866
$50K to $75K -2398657 2980453 -0.80 0.431 371553.358
$75K to $100K -2455023 2994758 -0.82 0.423 541397.221
$100K to $150K -2434483 2980581 -0.82 0.425 953779.541
$150K to $250K -2404935 2982679 -0.81 0.431 330537.600
$250K to $500K -2414342 2994755 -0.81 0.431 71152.055
$500K+ -2393024 2989787 -0.80 0.434 36401.343
Health Care 947225 1810961 0.52 0.607 32674.125
Labor 954055 1801535 0.53 0.603 727911.597
Professional/Administrative 966787 1801311 0.54 0.598 501480.184
Public Service 954355 1807843 0.53 0.604 42387.891
Service Industry 960649 1803238 0.53 0.601 19069.682
VIF=1/(1-corr^2)
Corr is the
multiple
correlation of the
variable with the
remaining
independent
variables

• Paired correlation analysis for all 38 variables (703
pairs!!) is impractical.
• Knowing paired linear correlation coefficient does not
help in reducing redundant information and extracting
meaningful information for separate contributing
factors.
• Regression analysis with dozens of the original
variables from observational data sets usually
fails.
Key Points:

• It allows removing the redundant
variables that carry little or no information
while retaining only a few mutually
uncorrelated principal variables.
Why Principal components
decomposition?

The main idea of PCD
The purpose of PCD is determining r new variables
PCr that can best approximate variation in the p
original X variables as linear combinations

The principle of information
conservation
• The total amount of information in the original data
set is not changed because of its PC decomposition
• Rather, it is rearranged in the form of a few linear
combinations of the original variables as main
information holders (PCs)
• This significantly reduces the number of
independent variables but retain the same amount
of information that is contained in the original data
matrix

What’s the eigen value?
• The eigen value λj is a measure of how much
information is retained by the corresponding PC.
• A large value of λj (compared to 1) means that
there is a substantial amount of information retained
by the corresponding PC
• A small value means that there is little amount of
information retained by the corresponding PC
Remainder:
If the product of the data matrix A and the vector p can be presented as
A * p = λj * p
then λj are eigen values and the vector p is eigen vector of the matrix A.

Eigen value analysis of the demographic
data correlation matrix
Eigen
value
16.44 11.19 4.63 2.73 1.15 0.853 0.63 0.307 0.067
Propo
rtion
0.433 0.295 0.122 0.072 0.03 0.022 0.017 0.008 0.002
Cumu
lative
0.433 0.727 0.849 0.921 0.951 0.974 0.990 0.998 1.000
Key Point:
Only 9 principal components (9 linear combinations of the
original variables) are required to account for all 38 original
variables.

Why Regression with Principal components?
• Because PCs are mutually uncorrelated, the
variation of dependent variable (CM $) is accounted
for by each PC independently of other PC
• Contribution of each PC is directly defined by the
coefficients of the regression equation
Key Point:
Regression with totally uncorrelated PC is one of the
most powerful methodologies for identifying significant
contributing variables (factors).

The Best Subset Regression
• Best subsets regression identifies the best-fitting
regression models that can be constructed with as
few predictor variables as possible
• All possible subsets of the predictors are examined,
beginning with all models containing one predictor,
and then all models containing two predictors, and so
on.
• The two best models for each number of predictors
are displayed

Best subsets regression with PCs
Varia
bles
R-sq
(adj)
Mallow
Cp
PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9
3 87.0 128 X X X
3 64.9 349 X X X
4 90.1 83.0 X X X X
4 88.1 99.6 X X X X
5 92.2 54.4 X X X X X
5 91.3 60.3 X X X X X
6 94.5 31.7 X X X X X X
6 93.4 37.2 X X X X X X
7 97.4 14.5 X X X X X X X
7 94.0 26.2 X X X X X X X
8 99.4 9.0 X X X X X X X X

Final regression equation with PC
CM $ = 12.8 + 0.201*PC2 - 0.387*PC3 + 1.95*PC8
(compare to the original regression…)
Key Points:
• This equation accounts for R-sq(adj) = 99.4% of the response
function (CM $) variability.
• It contains only statistically significant terms (at 5% confidence
level)

Conclusion from the regression equation
• Eigen vector coefficients for PC2, PC3 and PC8
combined with PC coefficients represent the
contribution of each variable into the $CM output
Note:
In general, for not-normalized variables the relative contribution of the Xi is:
called the elasticity coefficient Ei= (dY/Y)/(∂Xi/Xi) = ai*Xi/Y

Variable PC2 PC3 PC8
Age 18-34 0.26 0.037 -0.034
Age 35-54 -0.084 0.331 0.037
Age 55-64 -0.229 -0.173 0.236
Age 65+ -0.058 -0.185 0.015
BS/BA+ degree -0.269 -0.137 0.049
Assoc/Prof degree -0.237 0.081 -0.18
High school 0.097 0.332 0.101
No high school 0.286 -0.084 -0.078
Income < $50K 0.275 -0.105 0.025
Income $50K-$75K -0.059 -0.013 0.256
Income $75-$100K -0.27 0.125 -0.183
Income $100K+ -0.259 0.097 -0.012
Occupation: Health -0.21 -0.176 -0.206
Labor 0.265 0.116 -0.133
Professional/Adm -0.275 -0.059 -0.104
Public Service 0.029 -0.328 0.463
Service Industry -0.125 0.264 0.542
% male 0.059 0.210 0.017
% female -0.059 -0.210 -0.017
Race: African American 0.235 -0.123 0.007
Asian 0.157 0.142 -0.337
Native American -0.033 -0.339 -0.253
Other 0.263 -0.114 0.158
White -0.252 0.128 -0.087
Disease: Cancer-BMT 0.012 0.108 0.002
Med Oncology 0.012 0.107 0.01
Surgical Oncology 0.011 0.108 0.012
Cardiology 0.014 0.103 0.012
Cardiothoracic Surgery 0.014 0.103 0.011
Vascular surgery 0.018 0.104 -0.001
Digestive disease 0.014 0.103 0.005
Medicine/Primary Care 0.015 0.103 0.01
Musculoskeletal 0.014 0.105 0.012
Neurology 0.014 0.104 0.013
Transplant 0.016 0.106 0.008
Trauma 0.015 0.104 0.006
Unassigned 0.014 0.103 0.000
Women Health 0.015 0.103 -0.002
Eigen vector coefficients
for PC2, PC3 and PC8

Conclusion from the regression with PC
The primary contributing variables (factors) to CM $ are:
 Age 55-64
 Annual income $50 K - $75 K
 Occupations: Public Service and Service Industry
 Race- Other
 Relative contributions of diseases are:
neurology, cardiology and musculoskeletal

Concluding Remarks and Reflections
• As analytics professionals we are rewarded for help in solving
business problems
• Building analytics that influences business decision-making
requires attention to the non-technical side of the project
(organization’s internal politics and power-sharing)
• Analytics has no practical value for the organization if it does
not affect business decision-making, regardless of how much
a new trendy technology is used
So, how much of your work is about understanding and
addressing real business problems vs. the technology
deployment, coding and finding insights in the data?

Alexander Kolker. All rights reserved
53
.
Appendix
“We may regard the present state of the universe as the effect of its past
and the cause of its future (Predictive analytics?!)
An intellect which at a certain moment would know all forces that set
nature in motion, and all positions of all items of which nature is
composed, if this intellect were also vast enough to submit these data to
analysis, it would embrace in a single formula (algorithm?) the
movements of the greatest bodies of the universe and those of the
tiniest atom.
For such an intellect nothing would be uncertain and the future
(predictive analytics?) just like the past would be present before its
eyes.”
- Pierre Simon Laplace, A Philosophical Essay on Probabilities, 1795
Food for Thought:
Can the contemporary Big Data Technology function as that ‘intellect’
capable of analyzing all data and getting a single formula for the future?

Data Analytics for Real-World Business Problems

More Related Content

What's hot (20)

Viewers also liked (16)

Similar to Data Analytics for Real-World Business Problems (20)

More from Alexander Kolker (12)

Recently uploaded (20)

Data Analytics for Real-World Business Problems