Medication adherence is a growing public health concern in the US. It is the extent to which patients are taking medications as prescribed by their healthcare providers. Simply put, are patients eating their pills on time?
We looked at patient data from Medicare part D program released by Centers for Medicare & Medicaid services. We built a prediction model to ascertain whether a patient would be adherent based on a variety of social, economic and behavioral aspects.
2. Our Team
Data by itself isn’t worth anything unless there’s a problem
to solve and a community to solve it.
- Beth Noveck, Founder, GovLab
Suresh Prasanth Bala Sakthi
3. Medication non-adherence is a major and growing public health concern, as
20% to 30% of medication prescriptions are never filled consistently.
Bad news is that we aren’t reaching 50% medication adherence
100%
88%
76%
47%
Rx Prescribed
Rx Filled
Rx Taken
Rx Continued
Medication Adherence
Source: American Heart Association: Statistics you need to know. http://www.americanheart.org/presenter.jhtml?identifier=107
Accessed November 21, 2007.
4. Investment in medication adherence can lead to dramatic improvements in
effectiveness of treatments
Why adherence matters?
• Results of failure to adhere to prescribed medications:
– Increased hospitalization
– Poor health outcomes
– Increased costs
– Decreased quality of life
– Patient death
About 30% to 50% of treatment failures are due to medication non-
adherence. These treatment failures are estimated to cause 125,000
deaths annually.
Source: Benner JS, Glynn RJ, Mogun H, Neumann PJ, Weinstein MC, Avorn J. Long-term persistence in use of statin therapy in elderly patients. JAMA
2002;288:455-461
5. Investment in medication adherence can lead to dramatic reductions in
overall cost of care
Diabetes Medication level of Adherence
$8,812
$6,959 $6,237 $5,887
$3,808
$55
$165
$285 $404
$763
1-19% 20-39% 40-59% 60-79% 80-100%
Rx $
Medical $
Healthcare expenditure ($/year)
Outcome is significantly higher than outcome for 80-100% adherence group (P<0.05). Differences were tested
for medical cost and hospitalization risk.
Source: Sokol M et al. Impact of Medication Adherence on Hospitalization Risk and Healthcare Cost. Medical Care.
Volume 43, Number 6, June 2005
6. Medication adherence needs to be addressed in Primary Care
Top reasons for nonadherence
• Cost of medications
• Side effects/fear of side effects
• Forget/can’t keep track of
medications/complexity
• Don’t think it works/don’t need it
Complex human behavior
Source: Sokol M et al. Impact of Medication Adherence on Hospitalization
Risk and Healthcare Cost. Medical Care.
Volume 43, Number 6, June 2005
Source: Nasseh K, et al. Cost of medication nonadherence associated with
diabetes, hypertension, and dyslipidemia.
Am J Pharm. 2012;4.2:e41–e47.
• Socioeconomic factors (age, race,
gender, economic status)
• Patient-related factors (knowledge,
attitudes, beliefs, and skills)
• Condition/treatment related factors
(disease severity, co-morbidity, regimen
complexity, side effects)
• Provider factors (skill, training, resources)
• Setting/policies (access to care, Rx
coverage)
7. Blending data from different sources was key to derive predictors &
classifying patients based on adherence
Source: www.cms.gov
www.nlm.nih.gov/research/umls/rxnorm/docs/rxnormfiles.html
www.census.gov
Medicare - Primary Data Source
Drug Information
Disease Information
rxNorm Data
Economic Information
Social Information
Household Information
Census Data
Patient Information
Medicare Part D Information
8. When we looked at this data, we kept asking this question often “What
diseases that this pill treat?”
• Five websites with 3 different datasets to
find this information
• Often information coding conventions are
not common or consistent
• We used Alteryx to help solve this problem
Simple question, but complex data
We were using this so often that we built
an Analytic App!
Published this on http://gallery.alteryx.com as well
9. Heart diseases contribute up to 50% of healthcare spends in US. We wanted
to look at medication adherence in the context of heart ailments.
Heart Disease and Strokes
• Cause 1 of every 3 deaths
• Over 2 million heart attacks and strokes each year
– 800,000 deaths
– Leading cause of preventable death in people <65
– $444 B in health care costs and lost productivity
– Treatment costs are ~$1 for every $6 spent
• Greatest contributor to racial disparities in life expectancy
The good news is that we know what works, and the medications, when required,
are low cost.
Source: Roger VL, et al. Circulation 2012;125:e2-e220
Heidenriech PA, et al. Circulation 2011;123:933–4
Chronic Diseases
10. Prediction Methodology - What factors influence nonadherence?
• Decision tree model to predict ‘Yes’/’No’
• Regression to fit adherence days. Can we predict for how many days does
this patient take his medication
• Models with 45 predictors
Modeling medication nonadherence
Socioeconomic Factors
Changing the Context
To make individuals’ behavior
Long-lasting
Protective Interventions
Clinical
Interventions
Counseling
& Education
Largest
Impact
Smallest
Impact
Poverty, education, housing, inequality
Hypothesis
11. Inferences from our predictive models are largely inline with our hypothesis
• Public Health factors – State, County, Area
• Personal factors – Age, Income, Race,
Sex, Education, Insurance
• Medication factors – Number of pills,
Patient payments, Cost of drugs, Other
chronic diseases
Factors that influence nonadherence
for cardio vascular diseases
Model Summary
Variables actually used in tree construction:
[1] Area.name Avg_Annual.Average.Pay BENE_BIRTH_DT
[4] BENE_RACE_CD BENE_SEX_IDENT_CD BENRES_CAR
[7] BENRES_OP MEDREIMB_CAR MEDREIMB_OP
[10] Num_Pills PPPYMT_CAR PTNT_Pays
[13] RX_Cost SomeColPct SP_ALZHDMTA
[16] State
Root node error: 1821/10308 = 0.17666
n= 10308
We can add more personal and behavioral factors to improve the
accuracy of our model further
Most nonadherence is not caused by
side effects or drug costs. Rather the
problem is behavioral, simple
procrastination and forgetfulness.
12. Effective interventions - Automated call helped increase the number of
prescriptions that were filled, indicating improved adherence.
Source: Derose SF, Green K, Marrett E. Automated outreach to increase
primary adherence to cholesterol-lowering medications
[published online November 26, 2012]. Arch Intern Med. 2013.
Automated call system in patients
with primary non-adherence to
statin medication
13. Each intervention must be tailored to individual patients. Incentives of all
stakeholders needs to be aligned to improve medication adherence.
Source: Derose SF, Green K, Marrett E. Automated outreach to increase
primary adherence to cholesterol-lowering medications
[published online November 26, 2012]. Arch Intern Med. 2013.
Improve medication adherence
• Behavioral – audible reminders, smart pill
boxes and auto-refills counteract
procrastination and forgetfulness.
• Financial – Lower cost pharmacies,
generics, and payment assistance make
medication more affordable.
• Clinical – Pharmacist consultations and
therapeutic resources help address
medical concerns.
Improve Medicare systems
• Medication Adherence rates are far from
optimal but CAN be improved through
collaboration and alignment of incentives
for plans, physicians and pharmacists
• Ongoing, consistent, measurement of
medication adherence rates is important to
gauge improvement and to identify “best
practices” across plans, physicians and
pharmacists
• What gets measured…. Gets improved!
Source: Nau D The importance of measuring adherence Pharmacy Quality
Alliance [published online November 7, 2011].