HPHCs Antidepressant Compliance Program ACP: Preliminary Findings and Methods

1
HPHCs Antidepressant Compliance Program (ACP)
Preliminary Findings and Methods

• Kara Zivin Bambauer, PhD
• DPRG Seminar Presentation at DACP
• January 31, 2005

2
Co-investigators and acknowledgements

• This study was funded by an HPHC foundation
  faculty grant
• Dennis Ross-Degnan, ScD PI
• Alyce Adams, PhD Co-investigator
• Neil Minkoff, MD Co-investigator
• Rick Weissblatt, PhD Co-investigator
• Andrea Grande MedImpact liaison
• Stephen Soumerai, ScD Faculty mentor

3
Background and context

• Why do we care about depression?
• Where does current depression treatment fall
  short?
• Studies on depression management
• Physician-patient communication
• Physician feedback

4
Motivation for present study

• HPHC sought to improve its antidepressant HEDIS
  (Health Employer Data Information Set) scores
• DACP can evaluate data from this natural
  experiment / policy change

5
Antidepressant Medication Management HEDIS
measures
6
Antidepressant Compliance Program (ACP)

• Began 5/15/2003
• New users no antidepressant use in previous 100
  days
• Compliant if prescription filled within 10 days
  after previous refill ran out
• If not, fax sent to physician
• Adherent - if prescription filled within 30 days
  after previous refill ran out
• If not, treatment failure

7
Research questions

• How to measure antidepressant adherence using
  refill data?
• How to measure changes in adherence using refill
  data?

8
Dataset Construction Process

• Obtain pharmacy data
• Useful data includes patient id, date of fill,
  quantity of days supplied
• Create refill level variables for each pharmacy
  claim
• runout_date date_of_fill-days_supplied1
• left in hand MAX(0,runout_date-date_of_fill1)
• gap MAX(0,date_of_fill-runout_date-1)
• days_since_last date_of_fill-last_fill

9
Refill level file
10
Dataset Construction Process (Continued)

• Create episodes of care
• If gap gt 60 then episode ends and new episode
  begins at next refill
• Keep track of count of episodes per person
• Create outcome variables for each analysis
• How many gap days in first 84 days (if gt30 then
  gap841, else 0)
• How many gap days in first 180 days (if gt52 then
  gap1801, else 0)
• Would someone be eligible for a fax (gt10 day gap)
• Would someone be a failure (gt30 day gap)

11
Episode level file
12
Dataset Construction Process (Continued)

• Create qualifying episodes
• If gap lt 100 days then episode is not qualifying
  either at the beginning of study or after a
  previous episode ends (gap)
• Needs to be enrolled for at least 180 days before
  and after first antidepressant use
• Analyze first episode for each person who has a
  qualifying episode
• Only analyze 180 days after first antidepressant
  use for each person/episode

13
Data Analysis (1)

• How to measure antidepressant adherence using
  refill data?
• Use logistic regression analyses
• Did user have a gap in first 84 days?
• Did user have a gap in first 180 days?
• Predictors
• age
• gender
• previous antidepressant use
• comorbidity index
• Use data from before policy began (May
  2002-November 2002)

14
Results (1)

• Predictors of antidepressant adherence
• N5873
• Gap84
• Previous antidepressant use (OR 1.606, CI
  1.325, 1.947) comorbidity (OR 0.948, CI 0.904,
  0.995) significant
• Age and gender not significant
• Similar pattern for Gap180
• Previous antidepressant use, comorbidity
  significant
• Age and gender not significant

15
Data Analysis (2)

• How to measure changes in adherence using refill
  data?
• Use time series analyses to look at changes over
  time
• Would user be eligible to receive a fax (gap gt10
  days)?
• Would user be considered a treatment failure (gap
  gt 30 days)?
• Create indicator for month of first use of
  antidepressants
• Count each person in the month that they first
  used antidepressants
• Dont analyze data from transition period look
  at people who were and were not exposed to policy
  change

16
Segmented Autoregressive Error Model

• One Policy Model

Assumptions (1) Linearity
(2) Autocorrelation Structure
(3) Normality
17
Segmented Autoregressive Error Model (Data)
18
Segmented Autoregressive Error Model (Continued)

• Yt is the mean number of prescriptions per
  patient in month t
• Time (t) is a continuous variable indicating time
  in months at time t from the start of the
  observation period
• Policy is an indicator for whether time t occurs
  before (policy0) or after (policy1)
• Time after policy is a continuous variable
  counting the number of months after the policy
  change
• Error term et at time t represents the random
  variability not explained by the model,
  potentially correlated to errors at preceding
  time points

19
Percent of Patients with 10-Day Gaps in Treatment
out of all Patients who Initiate Antidepressant
Use
Percent adherence failures
Episodes span policy period
20
Percent of Patients with 10-Day Gaps in Treatment
Who Proceed to Adherence Failure
Percent adherence failures
Episodes span policy period
21
Percent of Patients who Proceed to Adherence
Failure out of all Patients who Initiate
Antidepressant Use
Percent adherence failures
Episodes span policy period
22
Results (2)

• Impact of policy on antidepressant adherence
• N23,411 first qualifying episodes
• Non-significant decrease in of people who get
  fax out of total qualifying episodes
• Significant decrease in people who fail out of
  people who got faxes
• Betas Intercept 0.61, time 0.001, policy
  -0.06, time after policy -0.01 (policy and time
  after policy significant)
• Significant decrease in people who fail out of
  all who qualify

23
Discussion Topics

• Role of enrollment data
• Fewer people get faxes / fail when enrollment
  included
• Amount in hand
• Differences in our findings than what
  HPHC/Medimpact might see (we use enrollment data,
  qualifying episodes, etc)
• Apparent effectiveness of faxed intervention

24
Next Steps

• Subgroup analyses gender, age
• Switching take into account switching from one
  antidepressant type to another
• Implement chronic disease score (CDS)
• Make sure confidence intervals calculated
  correctly
• Do separate analyses of people who are
  immediately non-adherent (no refills after first
  fill) and people who become non-adherent
  downstream