Predicting Patient Adherence: Why and How

• Cognizant 20-20 Insights

Predicting Patient Adherence:
Why and How
To contain costs and improve healthcare outcomes, players across
the value chain must apply advanced analytics to measure and
understand patients’ failure to follow treatment therapies, and to
then determine effective remedial action.
Executive Summary
The new healthcare regime — both in the U.S.
and across the world — can be boiled down to
reducing costs and enhancing the quality of
patient care. Improving patient adherence to a
treatment regimen can be a crucial factor. And
as with many other areas of modern healthcare
and life sciences, the role of data analytics is of
paramount importance.
The World Health Organization (WHO) defines
patient adherence as “the extent to which a
person’s behavior — taking medication, following
a diet and/or executing lifestyle changes — corresponds with agreed recommendations from a
healthcare provider.”
Adherence is a combination of compliance
(correct use of medication as prescribed) and persistence (continued effort in taking medication).
Adherence is what is expected of the patient.
There are multiple estimates of non-adherence
and its associated impact on cost reduction, all
of which are quite alarming. As per the Annals
of Internal Medicine, non-adherence impacts
the U.S. healthcare system between $100 billion

cognizant 20-20 insights | september 2013

and $289 billion annually, 20% to 30% of which
equates to medication prescriptions that are
never filled and 50%for medications that are not
taken as prescribed to treat chronic diseases.1 In
Europe, the situation is no better as non-adherence is said to cause 200,000 deaths per annum
and cost the economy €125 billion each year.2 And
if the increase in age of the global population is
factored in, the cost will likely spike manyfold in
coming years. So it is in the interest of payers,
drug manufacturers and policy makers to contain
this cost by applying advanced analytics.
On average, it costs six times more to attract
new patients than it does to retain existing ones.
Increasing patient adherence can boost profitability in the following ways:

• Payers: Reduction in healthcare expenses.
• Patients: Better health.
• Pharma: Enhanced patient loyalty and
retention.

This white paper lays out a framework for
enabling patient adherence management and
some general prescriptions on how to convert
lofty concepts to meaningful action.

2020: When Computers Outnumber Humans 10 to 1
• Historical Trend analysis by segment, therapy areas.
• ROI measurement of marketing and
educational activities.

Measure
and Track

Predict

Action

• Adherence prediction score.

• Designing segment-level targeting strategy.
• Multichannel optimization for
adherence strategies.

Figure 1

Analytics Framework for
Adherence Management
Players across the healthcare value chain need
to understand the patient journey. Patient
adherence is a complex issue, and requires a
thorough understanding of the root causes of
discontinuing treatment. Measuring and tracking
adherence may help to illuminate specific
strategies, but such tactics needs to be integrated
with prediction and action for the organization to
succeed.
Measure and Track
This phase primarily consists of:

• Developing a patient adherence index.
• Tracking the adherence progress

across
different therapeutic areas, by geography,
patients and physician segments.

• Assessing

the ROI of various marketing
channels
(through
advanced
analytics
techniques such as multivariate regression and
test and control analysis).

The most critical of all these, however, is
developing an adherence index. Multiple measures
exist for this:

• Direct or clinical methods such as the level
of medicine or biological marker in the blood,
etc. However, on a regular basis, measuring
and keeping track of non-adherence by direct
methods is not very practical since patients
require diagnostic procedures to obtain these
measures.

• Indirect

methods, based on prescription
refills or pill counts, are much more popular.

cognizant 20-20 insights

Depending on data availability, these two
measures can be interchangeably used. For
example, if the pharmacy refill data is available,
pill count is probably the best measure; however,
if the analysis is based on patient claims or longitudinal patient data, frequency of claims or
refills is probably the most dependable metric.
Two common measurements that can define an
adherence index are: (1) Medication possession
ratio (MPR) — i.e., the number of days of medicine
supplied within a time interval/total number of
days in the period; and (2) period of days covered
(PDC) — i.e., the number of days of medication
covered over a time period. For a single drug
medication, these measurements are actually
the same, but for a multidrug therapy they might
differ considerably.
In the case of multidrug therapy, MPR would be
calculated as (total days of supply) x (number of
days in the interval). For PDC, it is the proportion
of days when all the medications were available.
Let’s consider a hypothetical example of a
three-drug medication and 10-day interval time
period to understand the difference between MPR
and PDC. The first drug was available for all 10
days, drug two for the last seven days, and drug
three only for the last three days.
MPR = 67% ((10+7+3)/(3)/10 = 6.6) while
PDC (a more accurate measure)
= 33% (10/3 = 3.3).
Apart from the objective measures as described
above there are quite a few self-reported, and
hence subjective, measures that are based on
questions answered by patients. These include

2

the Morisky scale, medication adherence report
scale (MARS), beliefs about medication questionnaire (BaMQ), etc.
Prediction
The measure and track stage is a completely retrospective analysis. It’s a good first step for any
company starting patient adherence analysis,
but is not enough. That is simply because once
a patient becomes non-adherent, bringing him
or her into the adherence zone is much more
costly. Hence, predicting the
The sole objective of probable non-adherence and
taking proactive actions is
the predict phase is to required to stay ahead of the
discover the factor/s curve. Given huge healthcare
influencing adherence data sets, and ever-escalating computing speeds, this is
and understanding actually not such a big deal.
why the patient is However, analysis should
entering the non- follow a systematic approach
and companies are likely to
adherence zone. encounter several challenges
along the way:

Generally, the data set is divided into two
components — development and validation
(a 50:50 or 60:40 split). While the model is
estimated on the development sample, validation
data is checked to see whether the model
equation works equally efficiently. Metrics for
efficiency include KS statistic, ROC, etc.
However, the challenges typically entail finding
the proper set of explanatory variables rather
than the techniques used to model them.
Primarily, there are five types of variables:

• Demographic variables: Age, sex, race, marital
status, number of people in household, etc.

>> Source: Patient enrollment file with healthcare payers, patient registry, longitudinal patient data and medical claims data.

• Treatment variables: Number of medications

used, duration of treatment, cost of medication
or treatment, frequency of dosing, complexity
of treatment regimen, etc.

>> Source: Medical claims data, EHR/EMR data,
longitudinal patient data, patient registry,
etc.

• Define

“bad” adherence: The sole objective
of the predict phase is to discover the factor/s
influencing adherence and understanding why
the patient is entering the non-adherence
zone. But what level of adherence is nonadherence or bad adherence? Many times this
is guided by past experience; however, a datadriven approach can be more illuminating. An
analysis of the association between non-adherence and the treatment outcome — whether
an adherence rate below a certain level has
significant impact on treatment success rate
— is helpful here. For example, in the case of
diabetes, a benchmark can be decided by the
MPR level below which we see a dramatic rise
in the blood sugar level.

• Disease

factors: State of disease, diagnostic
tests, time, etc.

>> Source:

EHR/EMR, patient registries and
medical claims.

• Patient

factors: Knowledge about disease,
poor communication by healthcare provider,
patient’s own belief in the effectiveness of his
therapy regimen, etc.

>> Source: There is no syndicated data source

to capture patient-related factors. Data is
primarily collected through questionnaires
presented to patients.

• Health insurance variables: Formulary status
of the drug being used, co-pay amount, etc.

• Predictive

modeling: Once bad adherence is
defined, the next order of business is to predict
whether a patient breaches the benchmark.
Typically, a statistical model with a binary
dependent variable is developed, using any of
three techniques: logistic regression, decisiontree analysis or a neural network model. There
are a plethora of statistical measure to judge
the model qualities: C-statistics, coefficient
P-values, Hosmer and Lemshow test, goodness
of fit test, to name a few. However, the final
decision in predictive modeling is based on
whether the result remains “valid outside the
modeling data set.”


>> Source: Medical claims data, IMS PlanTrak,
Source®Payer, etc.

Clearly, data for this analysis can come from
disparate sources. Hence, the challenge is to
integrate and collate relevant information from
all inputs and, in the absence of a primary joining
key, the best technique to utilize is name matching
or text matching.
Action
The actions based on the “prediction” stage are
primarily to design adherence strategies that

3

include changing dosage and delivery, education
and communication, such as behavioral coaching/
patient education, program reminders and
financial assistance (i.e., co-pay discounts, etc.).
However, these activities need to be assisted by
additional analysis and modeling exercises.

• Patient

segmentation based on adherence
score and rationale: Targeting strategy
needs to vary according to patient behavior.
For example, if the reason for non-adherence
is frequency of dosages or the method of
delivery, stakeholders need to adopt different
strategies. In the former case, it might be
moving toward longer-lasting medications
and hence less frequency, while in the latter
case it may be moving from injectible to oral
medication.

• Segmentation

of healthcare professionals
(HCPs) based on their patient profile: As the
first patient touch point, HCPs play a critical role
in adherence. An SDI (formerly Verispan) study3
suggests that physicians’ choice of language
can influence adherence. For instance, patients
who were coached by a physician to take their
medication regularly had a 21% greater Rx utilization. They need to be informed and coached
as well. They can be segmented based on their
prescribed population profile (classified as
referenced above) .

• Marketing

budget optimization: A multichannel optimization analysis is needed to
decide where investment should be made to
reinforce adherence. (This is largely influenced
by the ROI analysis undertaken as part of the
“measure and track” stage.)

Looking Forward

industry as a whole. Analytics can help rank the
patient and physician population, and act as an
aid for designing education and other targeting
strategies. However, analytics based on “garbage
in, garbage out” principles will not fly, given the
current immature state of healthcare data, a
situation exacerbated by these factors:

• A

good number of healthcare and pharma
companies do not have a properly functioning
data warehouse.

• Many

players across the value chain cannot
match systems of records with internally
supplied data.

• Quality of third-party-supplied data is generally
poor.

Given these challenges, the approach that
healthcare and life sciences companies should
take is to not force complex analytics when and
where it is not possible. Successful implementation of analytics therefore becomes a function
of data quality and analytical maturity (i.e.,
management push to use a data-driven approach
to inform key decisions). While government initiatives such as “meaningful use” by CMS in the
U.S. will create incentives for better data quality,
pharma companies and healthcare players should
sincerely rethink data strategy — from collection
and storage through retrieval — as well as the ROI
of such strategies.
Hence, both basic data analysis solutions and
complex predictive analytics solutions can be
used to solve patient non-adherence challenges.
The precise approach will depend on the severity
of the problems faced and the ease and pace of
the implementation strategy considered by the
organization.

Improvement in adherence quality is necessary
for the well-being of patients and the healthcare

Footnotes
1

“Interventions to improve adherence to self-administered medications for chronic diseases in the United
States: a systematic review,” Annals of Internal Medicine, Dec 4, 2012.

2

European Federation of Pharmaceutical Industry and Associations.

3

“Understanding and Maximizing Prescriber Impact on Patient Compliance and Persistency,” SDI Case
Study.


4

About the Authors
Yashajit Saha is an Associate Director within Cognizant Analytics Practice specializing in life sciences
and healthcare analytics. He has more than 14 years of analytics experience working with global pharmaceuticals, healthcare and banking and financial services companies, on assignments spanning sales and
marketing through customer and strategic analytics. He can be reached at Yashajit.Saha@cognizant.com.
Rajesh Singal is a Senior Engagement Manager within Cognizant Analytics Practice focused on life
sciences and healthcare. He has 13 years of experience in analytics and the academics field, where he has
gained extensive experience in sales and marketing, customer and healthcare fraud and patient analytics.
He can be reached at Rajesh.Singal@cognizant.com.

About Cognizant Analytics
Within Cognizant, as part of the social-mobile-analytics-cloud (SMAC) stack of businesses under our
emerging business accelerator (EBA), the Cognizant Analytics unit is a distinguished, broad-based market
leader in analytics. It differentiates itself by focusing on topical, actionable, analytics-based solutions
coupled with our consulting approach, IP-based nonlinear platforms, solution accelerators and a deeply
entrenched customer-centric engagement model. The unit is dedicated to bringing insights and foresights
to a multitude of industry verticals/domains/functions across the entire business spectrum. We are a
consulting-led analytics organization that combines deep domain knowledge, rich analytical expertise
and cutting-edge technology to bring innovation to our multifunctional and multinational clients; deliver
virtualized, advanced integrated analytics across the value chain; and create value through innovative and
agile business delivery models. http://www.cognizant.com/enterpriseanalytics.

About Cognizant
Cognizant (NASDAQ: CTSH) is a leading provider of information technology, consulting, and business process outsourcing services, dedicated to helping the world’s leading companies build stronger businesses. Headquartered in
Teaneck, New Jersey (U.S.), Cognizant combines a passion for client satisfaction, technology innovation, deep industry
and business process expertise, and a global, collaborative workforce that embodies the future of work. With over 50
delivery centers worldwide and approximately 164,300 employees as of June 30, 2013, Cognizant is a member of the
NASDAQ-100, the S&P 500, the Forbes Global 2000, and the Fortune 500 and is ranked among the top performing
and fastest growing companies in the world. Visit us online at www.cognizant.com or follow us on Twitter: Cognizant.

World Headquarters

European Headquarters

India Operations Headquarters

500 Frank W. Burr Blvd.
Teaneck, NJ 07666 USA
Phone: +1 201 801 0233
Fax: +1 201 801 0243
Toll Free: +1 888 937 3277
Email: inquiry@cognizant.com

1 Kingdom Street
Paddington Central
London W2 6BD
Phone: +44 (0) 20 7297 7600
Fax: +44 (0) 20 7121 0102
Email: infouk@cognizant.com

#5/535, Old Mahabalipuram Road
Okkiyam Pettai, Thoraipakkam
Chennai, 600 096 India
Phone: +91 (0) 44 4209 6000
Fax: +91 (0) 44 4209 6060
Email: inquiryindia@cognizant.com

©
Copyright 2013, Cognizant. All rights reserved. No part of this document may be reproduced, stored in a retrieval system, transmitted in any form or by any
means, electronic, mechanical, photocopying, recording, or otherwise, without the express written permission from Cognizant. The information contained herein is
subject to change without notice. All other trademarks mentioned herein are the property of their respective owners.

Predicting Patient Adherence: Why and How

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Predicting Patient Adherence: Why and How

Similar to Predicting Patient Adherence: Why and How (20)

More from Cognizant

More from Cognizant (20)

Recently uploaded

Recently uploaded (20)

Predicting Patient Adherence: Why and How