Summary

One Dataset Everyday won’t Keep the Doctor Away....................................................p.2

I. Data Fragmentation and Limited Skills Deteriorate the Data Analysis Process p.6

1. Multiplicity of Data Sources Makes Collection & Use ...................................................p. 8

2. Data Diversity Hinders Data Integration.......................................................................p. 10

3. Limited Human Skills Inhibit Effective Data Analysis...................................................p. 14

II. Data Management Challenges Illustrated by the No-Show Issue.....................p. 16

1. The Financial and Human Cost behind No-Shows......................................................p. 17

2. The 3 Main Barriers to Solving the No-Show Issue.....................................................p. 20

3. 4 Quick Fixes... That Don’t Work.................................................................................p. 23

III. Step by Step Methodology to Build your Scheduling Data Product.................p. 28

1. Defining the Perfect Framework..................................................................................p. 30

2. Order Out of Chaos: Collecting & Making Sense of Data............................................p. 32

3. A Predictive Model to Test your Hypothesis................................................................p. 34

4. From Theory to Practice: Deploying your Data Product..............................................p. 36

IV. Creating a Proper Data Structure for a Complete Analytics Methodology.......p. 40

1. To Know before you Go................................................................................................p.41

2. Developing System Interoperability.............................................................................p. 43

3. Fostering the Distribution of Skills & Knowledge.........................................................p. 44

4. A Shift from Retrospective to Prospective Analytics...................................................p. 45

5. Engaging with Patients: Now or Never........................................................................p. 46

Curing the Healthcare Industry One Data Product at a Time....................................p. 47

End Notes.......................................................................................................................p. 50

Acknowledgements.......................................................................................................p. 52

1
Introduction: One Dataset a Day Won’t Keep the Doctor
Away
The healthcare industry is currently suffering from a host of issues. Knowledge
sharing between hospitals, determination of patient adherence to medications,
and the efficient management of surgical procedures are just three topics in a
long list of areas that need improvement. All of these issues have the same thing
in common: the healthcare industry has a data problem.
The fact is that there is an abundance of raw data and no one really knows what
to do with it. From patient records to heart-rate monitors, hospitals produce
reams of raw data that, after an initial reference, is usually forgotten. The good
news is that all of this data can be used to solve a multitude of common, day-to-
day problems using predictive analytics.
In this ebook we’ll highlight a specific issue—no-show appointments—and
show how predictive analytics can be used to discover real-world solutions
to a multi-billion dollar problem.

From patient records to heart-rate monitors, hospitals produce

reams of raw data that, after an initial reference, is usually forgot-
ten

2
The No-Show Problem

The unfortunate reality is that no-shows have become extremely common —

one study reported that the no-show rate in U.S. primary care practice can vary
from as little as 5% to as much as 55% 1.Appointment cancellation rates are al-
so a systemic problem in addiction treatment centers where 29% - 42% of pa-
tients fail to begin treatment 2 and 15% - 50% of patients do not even return for
a second visit 3.
Dealing effectively with patient no-shows has been a challenge in the
healthcare industry, especially now that reimbursement is more closely
tied to performance measures surrounding physical appointments.

The long-term effect of this phenomenon is lowered reimbursement for provi-

ders and, more importantly, the health welfare impact on adherence, quality, and
clinical outcome measures on patients. For patients, spotty appearances with
healthcare providers results in less coordinated care, particularly in cases of chro-
nic diseases and preventive encounters. Patients suffering from chronic condi-
tions may require very regimented treatment plans — missing even one treat-
ment may have debilitating consequences.

3
Missing preventive care treatments leads to longer and more expensive care as
potential issues become real health problems. No-shows also have a direct finan-
cial effect on healthcare providers as expected revenue targets fall short, labor
hours are wasted, and inefficiencies are created.

The challenge has always been, "What do we do with all this data?
How do we add meaning to it?"

Dealing effectively with patient no-shows has been a challenge in the healthcare
industry, especially now that reimbursement is more closely tied to performance
measures surrounding physical appointments. Many providers are simply

4
overwhelmed with the problem and resort to traditional stopgap policies,
such as reminding patients the day before their appointments. The effect is
marginal and, ultimately, is short-sighted because it does not directly address
the problem itself.

Executive Summary

This ebook aims at providing healthcare professionals with a clear view of how
efficiency gains could be realized at little cost via the integration of data analy-
tics. We will start by having a look at what is wrong with the current implemen-
tation of data analytics in the healthcare ecosystem and how it applies to the no-
show issue. We will then offer an alternative approach to addressing no-show ap-
pointments that makes use of predictive analytics. Lastly, we will discuss how
this method could be applied to the healthcare industry.

5
CHAPTER 1

Data Fragmentation and Limited Skills Deteriorate the Data

Analysis Process
The healthcare industry is no stranger to data — they have an abundance of it, from clinical
measures and demographic information to lab results and staffing data . Figuring out what
to do with all of the data is a challenge that lies at the heart of medicine in the 21st century.
According to a recent survey released by the National Association of ACOs (NAACOS),
51% of Medicare Shared Savings Program (MSSP) ranked data-related issues, such as ac-
cess, inconsistency, and deployment, as the biggest roadblocks in their accountable care jour-
ney 4. Core obstacles of data management & usage include:
• translating data into actionable information for providers,
• acquiring the required skill-sets needed to analyze the data,
• and finding solutions that can report on the business aspect of clinical data.

Apart from the structural dysfunctions, healthcare suffers at the IT

level: technologies are out-of-date compared to other high-tech sec-
tors, institutions use proprietary platforms that are incompatible
with other systems, and the IT skill-sets of employees are highly dis-
parate.

The “data issue” is exacerbated by the nature of the U.S. healthcare ecosystem: it is a highly
fragmented industry across multiple sectors. Healthcare is rarely coordinated, incentives are
misaligned, and variation is ubiquitous.

6
Apart from the structural dysfunctions, healthcare suffers at the IT level: technolo-
gies are out-of-date compared to other high-tech sectors, institutions use proprietary plat-
forms that are incompatible with other systems, and the IT skill-sets of employees are highly
disparate.

7
CHAPTER 1: DATA FRAGMENTATION AND LIMITED SKILLS

Multiplicity of Data Sources Makes Collection & Use

Difficult
Too Much Data

As mentioned, there is an abundance of data in the healthcare industry and all of

it flows from multiple sources: 5
• Prescription, diagnostic (lab, vitals measurements), and demographic sources;
• Social media / Web-based and machine-to-machine data (e.g., remote devices);
• Transactional data (e.g., claims and billing activities);
• Biometric data; and,
• Human-generated data (e.g., Electronic Health Records [EHR], physician no-
t e s ) .


According to the Institute for Health Technology Transformation, the amount of

U.S. healthcare data reached 150 exabytes in 2012 and is estimated to double by
2022 6. One exabyte equals 1 billion gigabytes and, given that the human brain
can only process around 7 variables at once, it’s obvious that the sheer size of
the data involved poses a significant challenge.

Collecting the Data

A large percentage of human-generated and biometric data is transcribed into

Electronic Health Record (EHR) systems; many of these platforms were inven-
ted in the 1960s and, frequently, little has changed in the basic approach used to
categorize incoming data. These one-size-fits-all systems are not well-suited to
individual workflows and they also lack the personalization needed to truly un-
derstand the needs of the patient.

8
Familiar Data Sources

Many organizations use data sources that are comfortable, familiar, and accessi-
ble. Over time the usage of these data sources become increasingly entrenched
in healthcare environments, to the point where other sources of data are not
even considered. The problem with this approach is that it only provides a par-
tial picture and does not provide access to the value that big data analytics can
offer.

9
CHAPTER 1: DATA FRAGMENTATION AND LIMITED SKILLS

Data Diversity Hinders Data Integration

Lack of Standards and Use of Multiple Standalone Silos of Data

In U.S. hospitals, the documentation of incoming data is mandated via the use
of EHR solutions. Ideally, an organization would use a common data entry inter-
face for all departments, from the emergency room to the finance division. Such
a framework would enable an analytics solution to access multiple data point ori-
ginations for comparison and analysis, effectively providing a holistic view at the
operations & management levels.
The reality, however, is much different: currently, 72% of healthcare organi-
zations use more than 10 electronic interfaces to collect data.

10
This level of disparity between data sources is a product of environments that
use individual silos of data: the accounting department collects data their way,
patient biometrics are collected a different way, and so on. Consequently, there
is no real standardization of data across an organization.
In addition, single-function EHR systems do not have the capability to aggre-
gate, transform, or create actionable analytics. In fact, intelligence is largely de-
legated to retrospective reporting which is insufficient for forward-looking
healthcare data analytics initiatives.

11
No Real-Time Data Integration

There may be some healthcare organizations with advanced data collection capa-
bilities, but there are few that possess advanced data integration at the intra and
inter-organization level. Meaning, there are no mechanisms in place to support
the sharing of data between healthcare institutions. There is an anecdotal story
about one hospital that was unable to share data with another hospital located
just across the street — data had to be printed and manually entered into the
other hospital's EHR.
At the U.S. government level, there is much concern over a plan to share the
EHRs of 10 million military service members from hundreds of hospitals and cli-
nics across multiple public & private agencies — a monumental task estimated
to cost at least $11 billion over a decade 7.

The focus of decision-makers—in terms of the use of data—has tra-

ditionally been applied to identifying volume & cost trends within
fiscal reporting periods rather than the actual use of real-time data
at the operations level.

Data integration issues are also present within healthcare institutions. For exam-
ple, most internal HIT (Hospital Information Technology) systems do not
offer real-time data APIs; typically, this data is processed overnight and availa-
ble in the data warehouse on the following day. The avoidance of offering real-

12
time data analysis is indicative of the overall approach of the healthcare ecosys-
tem to data.
The reality is that organizations are rarely data-driven — there is little inter-
nal incentive to evangelize real-time data analysis. The reasons for this stem
from the nature of healthcare: administration-level decisions must take into ac-
count a host of contractual, regulatory, and political decisions before being imple-
mented.
In addition, the focus of decision-makers—in terms of the use of data—has tradi-
tionally been applied to identifying volume & cost trends within fiscal reporting
periods rather than the actual use of real-time data at the operations level.

13
CHAPTER 1: DATA FRAGMENTATION AND LIMITED SKILLS

Limited Human Skills Inhibit Effective Data Analysis

Few Data Skills

In terms of data analysis at the human resources level, there is a severe discon-
nect between the skills required and the skills currently available. Healthcare
organizations require data professionals with a range of skills that are not solely
technical. Today's data scientists need to expand their skill-sets to include soft
skills such as communication, collaboration, creativity, and leadership. It is not
enough for a data scientist to know how to design and build analytical models —
they must be able to work with their peers to add meaning to the data and then
successfully convey that information to healthcare professionals who do not
have an IT background. A July 2014 survey of healthcare leaders stated that
60% of them were unsure of whether their organizations had the in-house
expertise necessary; in many cases, system development skill-sets were
outsourced 8.
In addition to the type of skill-sets needed, there is a shortage of talent. The
McKinsey Global Institute estimates that there will be a 10,000+ analytic talent
shortage through 2020; the end-result of this shortage means that 50% - 60% of
data scientist positions could go unfilled 9.

It is not enough for a data scientist to know how to design and build
analytical models — they must be able to work with their peers.

14
Reactive vs ProActive: A Shift in Industry Culture

As organizations grow and mature, there is an increasing reliance on established

methodologies — even if those methodologies are inefficient. Long-standing pro-
cedures are given merit simply due to their age; in these situations, "this is the
way it's always been done" is a common refrain.
The shift to an analytics-based culture is a significant one because it requi-
res the abandonment of long-standing reactive procedures and the adop-
tion of a proactive data-driven approach. In this environment, organizations
use a single source of truth to guide choices, they stop making "gut decisions,"
and avoid data shopping (i.e., finding data that supports conclusions that have
already been made).

15
CHAPTER 2

Data Management Challenges Illustrated by the No-Show

Issue
The ubiquity of no-shows has put a spotlight on a set of broader data management issues in
the healthcare industry. The inability of healthcare organizations to deal with the no-show
issue has had a profound effect on patient health, their experiences with healthcare provi-
ders, and on the financial bottom line. The problem is a difficult one to solve due largely to
industry practices that are both archaic and ineffective.

16
CHAPTER 2: DATA MANAGEMENT CHALLENGES ILLUSTRATED BY THE NO-SHOW ISSUE

The Financial and Human Cost behind No-Shows

A Costly Reality: Upward of $150b per Year

When a patient is unable to attend an appointment, there are multiple repercus-

sions that affect much more than the healthcare provider's bank account. The
U.S. healthcare system loses more than $150 billion per year in no-shows alone;
these costs stem largely from all of the associated issues that come into
play when a patient is unable to attend an appointment 10.

For example, missing an appointment means that the overhead related to that ap-
pointment is not reimbursed — items such as staffing costs, insurance, and utili-
ties remain on the books. In addition, a significant number of appointments are
made on a referral-basis — cancellations made at the primary care level means

17
that those referrals are never made, while cancellations at the specialist level
means that more revenue is lost and the patient's health may suffer. Ultimately,
no-shows have a significant impact on everyone, from physicians to patients, as
physician costs increase in order to bridge the financial gap caused by missed ap-
pointments.

Example
• A doctor is supposed to see 15 patients every day;
• 10% no-show rate = 1,5 missed appointments daily = 8 no-shows per week
• The doctor organizes appointments into 30-minute sessions at a cost of
$150/session.
Because of the 10% no-show rate, he loses $1,200 per week. This no-show
rate costs the practice around $62,400 per year.

Beyond the Financial Cost: Deterioration of the Patient Experience

An unintended consequence of no-shows are negative patient experiences (e.g.,

long waits or abbreviated visits). This is due to attempts by healthcare organiza-
tions to solve the no-show problem using ill-conceived methods.
For example, some health centers implement financial penalties for missed ap-
pointments. Another technique is to double-book patients; this results in short
appointments (e.g., 15 minutes instead of 30 minutes) that do not give patients
an opportunity to properly address their health concerns.
The people affected the most by no-shows are really the patients themsel-
ves. Preventive healthcare is responsible for discovering a wide array of potential-
ly life-threatening diseases, but is frequently eschewed when patients either pro-
crastinate or simply decide to cancel their appointments. For example, diabetic

18
patients with weakened immune systems can use wound clinics to treat minor
cuts... this does not cost much. If they decide to cancel their appointment,
though, a small issue may turn into much larger (and more expensive) problem.

19
CHAPTER 2: DATA MANAGEMENT CHALLENGES ILLUSTRATED BY THE NO-SHOW ISSUE

The 3 Main Barriers to Solving the No-Show Issue

Difficulty with the Incorporation of Data Sources

Why do patients cancel appointments? There are many factors that are responsi-
ble for no-shows and it's not always about something else "coming up."
Some of these factors include distance (i.e., geographically remote), transporta-
tion (e.g., lack of public transportation options), and scheduling (i.e., too early,
too late, etc.). Sometimes appointment destinations are selected based on pa-
tient data that may be outdated. The type and severity of the disease being trea-
ted is also a no-show factor.

In terms of predictive analytics, all of these factors represent data points

that could be used to decrease the likelihood of no-shows or, at least, bet-

20
ter anticipate them. The reality, however, is that HIT systems typically lack the
capability to take advantage of this data and are unable to access data stored
across different silos.

Lack of Access to Patient Data

Frequently the scheduler, or the person responsible for scheduling doctor-pa-

tient appointments, does not have access to the global patient dataset, which
means that the scheduler may be working with either a lack of data or even incor-
rect data. This may be due to a lack of technical knowledge on the part of the
scheduler or it may be due to the underlying system being used (non-collabora-
tive). The end-result is that the scheduler cannot see the "big picture," such
as the patient's transportation needs, exact geographic location, and time availa-
bility. This lack of knowledge means that the scheduler is effectively working
blind and is unable to create appointments that reflect the patient's needs.

21
Lack of No-Show History

Profiling no-show patients is a difficult task because there is not too much rele-
vant data readily available. This is particularly true for patients with a health
plan who have no history of seeing a doctor within their network; predictions
are quite difficult because there is no historical data to analyze. The progres-
sive incorporation of new data, along with external data sources, may provide
the clues needed to determine which patients will likely not appear, but data sys-
tems that use such real-time data are not widespread.

22
CHAPTER 2: DATA MANAGEMENT CHALLENGES ILLUSTRATED BY THE NO-SHOW ISSUE

4 Quick Fixes... That Don’t Work

Double-Booking

Double-booking is when a healthcare practice sets appointments for more peo-

ple than they can viably handle with the expectation that a percentage of them
will not appear, effectively creating a fully-booked day with normal doctor-pa-
tient visit times. The high number of unknown variables, however, results in a
very negative patient experience.
If there are no no-shows, then all patients only have access to their provider for
50% of the expected time (e.g., 15 minutes instead of 30 minutes). As the day
progresses, the staff and physicians become frustrated, often wishing that
someone would not appear for their scheduled visit. The patients themselves
also become frustrated as they are rushed through their appointment

First-Come, First-Served

The first-come, first-served approach is a common method in 3rd world coun-

tries in order to alleviate the need for scheduling maintenance. The idea is that
care is given to whoever shows up first; a ranking system is kept by the front
desk personnel based on the patient's order of appearance.
The primary issue with this approach is that healthcare providers frequently
have little incentive to perform their job punctually; after all, they know that
their office will be full of eager patients every morning. This exacerbates the core
problem, which is the long wait — patients are kept waiting for hours until it is
their turn (and until the physician appears).

Financial Penalties

23
In an effort to negatively incentivize a patient's on-time appearance, some
healthcare providers implement a financial penalty to patients who do not ap-
pear (or appear late). This obviously has negative repercussions on patients
from all walks of life, as those without money will be unable to pay the pe-
nalty.  

Contractual Appointment Reminder Service

Appointment reminder services are typically call centers hired on a contractual

basis whose job is to call patients 24 hours before their scheduled appointment
and provide a reminder. The average no-show rate is 12%. This method decrea-
ses the no-show rate to 10%. The effect is minimal, the costs can be high,
and the experience is impersonal.

As a whole, all of the above methods are reactive by nature (trying to miti-
gate the problem) instead of being proactive (directly solving the core pro-
blem).

24
 Interlude: Guidelines to Conquer the No-Show Issue

A Painful Issue

It's been a typical, and frustrating, day. It's 5pm and 12% of today's 300 schedu-
led patients did not show up for their appointments. This means that 36 people
did not appear and your staff worked to 88% of their capabilities. At this rate,
you've been losing about $5,400 per day ($1.36 million annually) plus pay-
roll & infrastructure costs.
At the end of the day, you realize that you've been wasting money, frustrating
your staff, and losing efficiency. Your healthcare payers are dissatisfied with your
hospital's efficiency and the patients they are in charge of are not well cared for.

A Solution

Now, imagine if we could somehow reduce the 12% no-show rate by scoring the
patient likelihood of a no-show. For example, a scoring mechanism could isolate
the 5% of your patients that represent 40% of those most likely to not appear for
their appointment. Instead of using a reminder service to call all patients, which
costs both time and money, why not allow your scheduling staff to contact speci-
fic patients, remind them of their appointment and, if needed, arrange more flexi-
bility. This would reduce your no-show rate down to 7% and would save
your hospital $550k per year — happy patients and a less frustrated staff.
So, let's take this a step further. Instead of doing a one-time analysis of patient
no-shows, imagine if you could use a predictive analytics methodology to deter-
mine no-shows in real-time.

25
Processing your Data

The process may start with a computation of datasets to determine which ti-
mes have the highest no-show rates. This analysis may provide some surpri-
sing insights into exactly when your patients are not appearing. Secondly, given
the local time slot data combined with global dimensional data, determine the
reasons why patients are not appearing for specific time slots — possible cul-
prits could be the weather, the geography, the disease, transportation options,
and/or the patient. Defining these items would enable you to create & assign
time-based points to each of your patients, depending on their distinctive featu-
res.

Deploying your Strategy

Armed with this knowledge, your schedulers could be advised, in real-time, on

how likely your patients are to be a no-show based on time slot. The schedu-
ling process could suggest 3 specific time slots when patients' no-show sco-

26
ring would be at its lowest. (i.e., when they are the most likely to appear).
Combined with an overbooking strategy on specific time slots, this kind of proac-
tive scheduling would enable your scheduling staff to suggest relevant time slots
while also offering them flexibility. The end-result would be a no-show rate of
only 4% and an annual savings of almost $1 million.

27
CHAPTER 3

Step by Step Methodology to Build your Scheduling Data

Product
The process of developing a data analytics strategy to tackle the no-show problem requires a
comprehensive methodology. The approach should start with defining a tangible goal
and end with incorporating the output into business practices; between those two
end-points we have a complete agile-based process that includes data definition, gathering,
cleaning, processing, and improving.

Below is an agile roadmap that conveys how each process contributes to the eventual goal of
deploying a predictive service to your scheduling system.

28
29
CHAPTER 3: STEP BY STEP METHODOLOGY TO BUILD YOUR SCHEDULING DATA PRODUCT

Defining the Perfect Framework

Designing the Perfect Frame

The fascinating aspect of data analytics and no-show problem solving is

that all of this is possible. The data is there and the capability exists. Solving
the no-show issue would promote efficiency across your organization while pro-
viding substantial cost savings for the long-term. Tackling this problem is not
only a goal for healthcare providers, but also for healthcare payers who need to
keep a grip on their population.
We will explore the possibilities below and discuss how data analytics can be
used to negate the effect of no-shows and drive efficiency from the bottom-up.

First, though, we will discuss two critical aspects required when defining an ef-
fective project frame for healthcare analytics projects:

Collaborative Framework

As the saying goes, "No man is an island" — we are social beings who are most
effective when cooperating and working with others. In terms of healthcare ana-
lytics, this means that Health Departments and IT Departments need to work
hand-in-hand to effectively realize change. Likewise, an engaging analytics soft-
ware solution needs to be collaborative and available to data experts as well as
beginners.

Agile Framework

The Agile method is an iterative process whereby constant testing and incremen-
tal improvements lead to continuous improvements in the least amount of time.
Agile frameworks are particularly well-suited to healthcare data analytics, be-
cause it enables your teams to constantly test models and prototypes in an effi-

30
cient manner. The Agile team should be inclusive and collaborative; mem-
bers should be representative of both Health and IT Departments. For
example:
• Quality Director from the Medical Department;
• Data Scientists from the IT Department; and,
• Director of Primary Care from the Operations Department.

31
CHAPTER 3: STEP BY STEP METHODOLOGY TO BUILD YOUR SCHEDULING DATA PRODUCT

Order Out of Chaos: Collecting & Making Sense of Data

Define your Goal(s)

In order to keep costs within budget and to realize feasible results, it is necessa-
ry to specifically define the project goal. In this case, our goal is to score the
likelihood of patient no-shows in real-time. The scoring would be used to iden-
tify high-risk patients and schedule the best time slots for them in order to de-
crease the likelihood of subsequent no-shows.

Collect Historical Data (Appointment Dataset)

In order to create an algorithm, the predictive analytics solution needs to work

with data. If possible, provide 3 months’ worth of historical show/no-show
data; if not possible, you may need to collect this data for 3 months before begin-
ning the predictive modeling process.

Gather Workable and Clean Datasets

Next, we need to determine the datasets that will be used to establish patient
scoring. In other words, the factors that will determine whether or not a patient
is likely to appear for a given time slot. Some possibilities include:
• Appointment Dataset: historical data of shows and no-shows;
• Patient Datasets: age, location, health problems, diseases, children, status...
• External Sources: social mapping of geographic area, transportation data, di-
sease classification (i.e., effect of disease on the patient's lifestyle — for exam-
ple, wheelchair-bound? mobility? capabilities? limitations?), bank holiday ca-
lendars, weather, and so on.

32
Some key questions to answer: how frequently are these datasets updated?
Are they automated? Is accurate and up-to-date data available?

Combine and Clean your Sources

Combine all data sources, clean the data, delete empty/incorrect fields, and en-
sure that the same level of detail—in terms of granularity—is applied across all
data points (e.g., weather data may be available daily while appointment sheets
are created on a weekly basis). It is common for datasets to be available in dif-
ferent formats (xls, calendar files…), so one of the challenges of data collection
will be shaping them all in a common processing-friendly format.

33
CHAPTER 3: STEP BY STEP METHODOLOGY TO BUILD YOUR SCHEDULING DATA PRODUCT

A Predictive Model to Test your Hypothesis

Highlight and Pinpoint Distinct Features

The process of building a predictive model involves a series of normalization and

optimization steps designed to determine model accuracy. Some key steps in
this process include feature normalization, testing & optimization of models, de-
termination of model accuracy, and the specification of a user strategy. After the
model is defined, the data scientist needs to overfit the model, evaluate, and ulti-
mately validate it in order to isolate features.
The determination of accuracy is done by testing the underlying strategy in
practice; for example, given patients who are likely to appear for a given
time-slot, do they actually show up as expected? How accurate is the time-
slot scoring for patients who do appear? If overbooking is implemented, is it
being applied correctly? These questions all need to be addressed in order to de-
termine the accuracy of the underlying analytical model — this involves compa-
ring real-world results with the relevant predictions. This level of additional
analysis will enable a data analytics solution to further refine the model’s accura-
cy, if needed.
Of course if you are using an advanced software analytics solution, then many of
the above steps would be automated. It would be able to clean datasets, isolate
specific features, and automatically score the likelihood of patient no-shows.

These questions all need to be addressed in order to determine the ac-

curacy of the underlying analytical model — this involves comparing
real-world results with the relevant predictions.

34
Train Machine Learning Models on Test Datasets

If new features are added, then the models need to be re-trained. Additionally,
data visualization needs to be done in order to determine if the features are rele-
vant.

35
CHAPTER 3: STEP BY STEP METHODOLOGY TO BUILD YOUR SCHEDULING DATA PRODUCT

From Theory to Practice: Deploying your Data Product

Automate Data Preparation on Incoming Data

The hard work of data analytics is over. At this point, the outputs need to be in-
corporated into the scheduling process. The first step of deployment is to au-
tomate the preparation of new incoming data — this ensures that the solu-
tion continues to work effectively going forward.

Deploy Daily Results to Scheduling Team

After the predictive analytics solution creates the patient scores, there needs to
be a mechanism in place to integrate the results into the scheduling system. An
API should be used to ensure that schedulers can easily access scoring.
The goal here is to score patients in real-time when schedulers are creating ap-
pointments. Ideally, multiple time slots should be presented to the scheduler so
that there is room for scheduling flexibility. Each suggested time slot reflects the
time when the patient is most likely to appear for the appointment.

The Reality Check

Once the model has determined which time slots have the lowest no-show likeli-
hood, you are halfway there. Appointment scheduling lies at the intersection of
efficiency and timely access to health services. Timely access is important for
realizing good medical outcomes and is also an important determinant of
patient satisfaction. For example, if three patients with near-identical no-show
scores are scheduled for the same time slot, then the outcome may not be as ex-
pected: two of them will have to wait and the third will probably leave after half
an hour.

36
Scheduling issues are magnified when considering staffing optimization. If
multiple patients are scheduled for the same time slot, what will your staff do
with their remaining work hours?
No-show issues have an impact on multiple areas across your organization, fre-
quently in ways that are not expected. Establishing a scheduling optimization
system means taking into account no-show scoring in combination with staffing
optimization and timely access

One effective strategy is to develop an intelligent overbooking sys-

tem based on specific time-slots in order to decrease the no-show rate
as much as possible.

Your scheduling optimization model is all about being realistic: for obvious
reasons, you can’t schedule each of your patients on the same time slots. It’s un-
likely that your no-show rate will be equal to zero, so the best approach is to mi-
nimize it as much as possible. One effective strategy is to develop an intelligent
overbooking system based on specific time-slots in order to decrease the no-
show rate as much as possible. Such a system, powered by a machine learning
approach, could be enriched with previous results so that the overbooking rate
could be fine-tuned for specific time slots.

37
38
 Interlude - Predictive Analytics in Action

A major U.S. healthcare provider, responsible for 15+ hospitals and clinics, deci-
ded to deploy a predictive analytics solution in order to address the no-show
issue. In their situation, the most important factors to determine patient appea-
rance were time slots, location, and disease type. They discovered that there was
a significant relationship between public transportation schedules and benign di-
seases. In order to address this, the provider always scheduled benign disease-re-
lated appointments in the middle of the day in order to sync with the availability
of public transportation schedules.
They have now deployed real-time no-show scoring within their schedu-
ling process. Three time slots are suggested to high-risk no-show patients, ta-
king into account the disease, location, and the patient's mobility.
In addition, an overbooking strategy was established to reduce uncertainty
on time slots that are more likely to be skipped. As an example, the predic-
tive analytics system highlighted Thursday and Friday mornings as time slots
with the highest no-show rates. The scheduling system used precision overboo-
kings to make sure staff’s time was not being wasted.  
The no-show rate is now down to 4%, resulting in an annual savings of $3
million.

39
CHAPTER 4

Creating a Proper Data Structure for a Complete Analytics

Methodology
Data analytics in the healthcare industry has a bright future. This is due to a number of fac-
tors working together:

• A huge amount of data (150+ exabytes as of 2012);

• A desperate need to understand raw data and assign meaning;

• The significant number of business-oriented applications to which data analysis can be ap-
plied in the healthcare sector.


Right now we are only at the tip of the iceberg in terms of implementing data analysis in
healthcare. Going forward, however, there are some cautionary steps that could easily trip-
up any organization pursuing an analytics platform.

In order to successfully implement a predictive analytics solution in the healthcare industry,

it is necessary to have a clear vision of outputs, implement IT systems that are interoperable,
and have a commitment to knowledge sharing across the organization.

40
CHAPTER 4: CREATING A PROPER DATA STRUCTURE

To Know Before you Go

Identifying Business Needs

The application of data analytics in the healthcare industry is designed to make a

concrete impact on existing business processes and to improve efficiency. In
most industries the data already exists and, in the healthcare sector, it is in abun-
dance. The challenge has always been, "What do we do with all this data? How
do we add meaning to it?" Data analytics provides pathways to answer these
questions — your ultimate destination depends on your business needs.

Making all of this happen requires a comprehensive data analytics

platform that is capable of not only handling, automating, and visu-
alizing data, but can also be used as a collaborative tool for different
user profiles (e.g., IT, business, marketing).

Tackling the no-show issue is just one example of how data analytics can
transform entire business segments; predictive analytic methodologies could
be equally applied to physician profiling, precision medicine, disease
management, and so on.
Making all of this happen requires a comprehensive data analytics platform that
is capable of not only handling, automating, and visualizing data, but can also be
used as a collaborative tool for different user profiles (e.g., IT, business, marke-
ting).

41
Leveraging your Data

The "analytics" part of data analytics sometimes steals the show, but the hard
work occurs during the early stages: collecting and cleansing the data. The first
step on a data project is to define the inputs — where is the data coming from?
After data sources are defined, we progress to data cleansing which accounts for
more than 80% of a data scientist's work. These tasks revolve around standar-
dizing the dataset and dealing with issues such as missing data, redundant
data, and unformatted data... all of which needs to be parsed and format-
ted.
An advanced analytics platform should be able to automate all of the above
tasks, effectively freeing up a significant portion of labor hours spent doing mo-
notonous data cleansing work.

End-User Focus: Deploying to Existing Systems

All of the winning algorithms and awe-inspiring models in the world are useless
if the end-results cannot be effectively deployed to the relevant business process
or system. In the case of no-shows, it is critical that schedulers be able to easily
access scoring data so that they can make meaningful time slot suggestions to
high-risk no-show patients. In a healthcare environment, analytics outputs
have to be made available to those working in operations (e.g., nurses,
aids, physicians, insurance analysts).
It is therefore critical that an analytics platform should be highly collaborative,
easy-to-use, and accessible. It should not be a tool for data scientist alone but,
rather, an intuitive solution that can readily be used by those with both IT and
non-IT backgrounds. Projects should be shareable between users and editable in
a team-friendly interface.

42
CHAPTER 4: CREATING A PROPER DATA STRUCTURE

Developing System Interoperability

As mentioned, the healthcare industry is highly fragmented when it comes

to data. Every department seems to have its own standards and data entry tools.
These standalone silos of knowledge pose a significant challenge to data analy-
tics initiatives — data scientists face numerous obstacles in terms of securing ac-
cess to multiple data sources and then standardizing that data after access is ob-
tained.

System interoperability means that different datasets can be used, regard-

less of source. An advanced analytics platform should have the capability to con-
nect to multiple dataset sources and effectively combine them with other inter-
nal, or external, datasets. In other words, there should be no obstacles in-place
when it comes to accessing data.

Proprietary, or closed source, systems introduce a long host of limitations

and restrictions, particularly when it comes to scalability and data connecti-
vity. Never pursue a proprietary system and always ensure dataset interoperabili-
ty.

43
CHAPTER 4: CREATING A PROPER DATA STRUCTURE

Fostering the Distribution of Skills & Knowledge

All of those individual silos of knowledge represent more than actual datasets...
they also represent micro-cultures within large organizations. The larger the
organization is, the more likely it is that "This is how we do it!" attitudes are pre-
valent. Generally-speaking, there is often a hesitation to adopt new approa-
ches and solutions; sometimes there is also a disinterest in sharing information
across departmental lines. Even if there is an interest, it's more than likely that
the current technology does not support data sharing.

Data from all sources should be made accessible to the analytics plat-
form so that a single source of truth can be attained.

These limitations, whether human or IT-based, all represent barriers to

knowledge sharing. A good analytics platform should act as a catalyst in terms
of helping management to break down those barriers. Collaboration, content sha-
ring, and dataset connectivity are all features that can help healthcare payers &
providers implement data transparency across business segments. Data from all
sources should be made accessible to the analytics platform so that a single
source of truth can be attained. Ultimately, all parties benefit, particularly the pa-
tients.

44
CHAPTER 4: CREATING A PROPER DATA STRUCTURE

A Shift from Retrospective to Prospective Analytics

Being data-driven is not an option, as nearly every healthcare setting has too
much data to use effectively. The key is to transition from traditional retrospec-
tive analysis to the more forward-thinking prospective analysis.
The former tells you that there is an existing problem and delivers analytical con-
tent based on that problem — the latter predicts upcoming problems so that
they can be anticipated and their effects mitigated.

Healthcare providers are reacting to no-shows instead of proactively

addressing the reasons why they are occurring.

From a business standpoint, retrospective analysis provides high-level vi-

sual summaries while prospective analysis is highly focused on a single
well-defined business problem.

45
CHAPTER 4: CREATING A PROPER DATA STRUCTURE

Engaging with Patients: Now or Never

Data analysis has the capability to provide powerful insights into events that we-
re, in a previous life, comprised of raw unformatted data. Sometimes connec-
tions can be made between datasets and data points that were unanticipated.

There is no reason why these insights should be limited to healthcare providers

— why not share relevant data insights with patients in order to engage
and empower them?

For example, patients with special mobility needs may enjoy seeing data visuali-
zations that convey how transportation schedules are used to provide more rele-
vant appointment schedules.

46
Conclusion: Curing the Healthcare Industry One Data
Product at a Time
It’s clear that when a patient does not appear for an appointment, both time and
money are lost. The issue has now reached a stage where the healthcare
industry, as a whole, is losing billions of dollars each year. Attempts to fix
the problem are really stopgap measures designed to address the symptoms.

Data analytics enables organizations to stop the guesswork and un-

derstand exactly when specific patients are likely/unlikely to appear
for any given time slot.

In fact, the core issue is being ignored completely: healthcare providers are reac-
ting to no-shows instead of proactively addressing the reasons why they are oc-
curring. This knee-jerk reactionary approach has resulted in policies that not
only do not stop the financial loss, but cause needless patient discomfort, increa-
sed waiting times, and negative doctor-patient experiences. From charging fees
for no-shows to cutting precious appointment times in half, these misguided re-
medies have inadvertently created contentious doctor-patient relations
instead of fostering amicable & friendly relationships. Data analytics enables
organizations to stop the guesswork and understand exactly when specific pa-
tients are likely/unlikely to appear for any given time slot.
At its core, predictive analytics cuts through, clarifies, and conveys highly rele-
vant information based on a wide array of diverse data. Local data (e.g., patient
information and historical results of appointments) is combined with global di-
mensional data (e.g., transportation costs, traffic routes, weather, geographical

47
distances, and patient diseases) to create a holistic view of the variables that are
affecting no-show rates.
Models are created, tweaked, and refined until a clear picture emerges that ex-
plains why patients are not appearing and, more importantly, what your clinic or
hospital can do to directly address the core problem.

The possibilities for predictive analytics are endless and are indica-
tive of the world we live in: where vast quantities of raw data can be
accessed, cleansed, collected, parsed, formatted, and elegantly visual-
ized in a meaningful way.

Predictive analytics is rapidly changing the way business is done in the

21st century.
No-shows are a critical issue that have a negative impact industry-wide but, at
the end of the day, it is a singular problem in a vast sea of possibilities. The reali-
ty is that the healthcare industry faces a plethora of challenges whose solutions
revolve around vast amounts of untapped data.
What if patient satisfaction data could be correlated with healthcare fees? What
if Internet of Things (IoT) sensor data in hospitals could be used to predict medi-
cal appliance needs? What if EHR and global data could be used to predict pa-
tient non-compliance with medications?

The possibilities for predictive analytics are endless and are indicative of
the world we live in: where vast quantities of raw data can be accessed, clean-
sed, collected, parsed, formatted, and elegantly visualized in a meaningful way.

48
The future of predictive analytics in the healthcare industry is indeed bright and,
whether the subject is no-show issues or a different challenge all together, we
look forward to discussing the possibilities with you.

49
End Notes

1: Maalinii Vijayan, “No Shows: Effectiveness of Termination Policy and Review of Best Practices”,
Wright State University, 06 November 2013, 5.

2: Weisner C, Mertens J, Tam T, Moore C., “Factors affecting the initiation of substance abuse
treatment in managed care”, Addiction, 2001;96(5):705–716.

3: Mitchell AJ, Selmes T., “A comparative survey of missed initial and follow-up appointments to
psychiatric specialties in the United Kingdom”, Psychiatric Services, 2007;58(6):868–871.

4: Gaus, Clif, “National ACO Survey”, National Association of ACOs, 21 January 2014, 3.

5: Mathematica Policy Research, “Health Information Technology in the United States, 2015: Tran-
sition to a Post-HITECH World”, Harvard School of Public Health, et al., 2015, 54.

6: Hoover, Waco, “Transforming Health Care Through Big Data”, Institute for Health Technology
Transformation, 2013, 5.

7: Allen, Arthur, “Critics warn of $11 billion Pentagon health records fiasco”, Politico.com, 28 July
2015, retrieved 05 December 2015 from
http://www.politico.com/story/2015/07/pentagon-electronic-health-record-critics-120730.

8: Manyika, James, et al, “Big data: The next frontier for innovation, competition, and productivi-
ty”, McKinsey Global Institute, May 2011, 11.

9: Manyika, James, et al, “Big data: The next frontier for innovation, competition, and productivi-
ty”, McKinsey Global Institute, May 2011, 10.

50
 10: Toland, Bill, “No-shows cost health care system billions”, Pittsburgh Post-Gazette, 24 Februa-
ry 2013, retrieved 05 December 2015 from
http://www.post-gazette.com/business/businessnews/2013/02/24/No-shows-cost-health-care-system-
billions/stories/201302240381.

51
Acknowledgements

This ebook was made possible by the contribution of healthcare professionals,

researchers, industry experts, and our own customers.

Dataiku would like to thank Dr Martin Pusic, Director of the Division of Lear-
ning Analytics for the Institute for Innovations in Medical Education at NYU
School of Medicine. His views on data analytics tools were extremely educatio-
nal and helped us better understand the stakes at hand. If you’d like to find out
more, please read this interview, published on the 25th of November 2015.

We are also grateful to Dr William Tierney, President and CEO of the Regens-
trief Institute. He guided us in grasping Electronic Health Records’ limitations
in the healthcare industry.

A special mention for Eric Kramer, our Healthcare Data Scientist and expert in
the application of machine learning to discover genetic biomarkers. His precious
insights and in-depth knowledge of the healthcare industry assisted us in positio-
ning DSS as the perfect tool to help cure the healthcare system’s data problem.

Finally, without Romain Doutriaux, Dataiku’s Solution Marketing Manager, this

ebook would simply not exist. Romain led all the research, interviews, and de-
sign initiatives involved in the project.

Last but not least, we would also like to thank our healthcare customers, both
providers and payers, for their continuous support and the great work we mana-
ged to achieve together.

52
Healthcare is an information business where the difference between life and
death is at stake. We firmly believe that predictive analytics have a key role to
play in this regard and we hope you will join us soon in being a part of the solu-
tion.

53