Transcript Longitudinal study: from 1993 to 2004

Effects of introducing then
removing cost-sharing for drugs
among people with schizophrenia
in Quebec: A natural experiment
Eric Latimer, Ph.D.
Canadian Health Economics Study Group
May 27 2010
Co-authors






Willy Wynant, M.S.1
Adonia Naidu, M.Sc.2
Robin Clark, Ph.D.3
Ashok Malla, M.D.2,4
Erica Moodie, Ph.D.1
Robyn Tamblyn, Ph.D.1
Department of Epidemiology, Biostatstics and Occupational Health, McGill University
Douglas Mental Health University Institute
3 Department of Psychiatry, McGill University
4 Family Medicine and Community Health, Center for Health Policy and Research,
University of Massachusetts Medical School
1
2
Acknowledgement: Michal Abrahamowicz for contribution to original study
design
Study funding

Fonds de la recherche en santé du
Québec
BACKGROUND
Schizophrenia

Disabling mental illness




Several subtypes
Positive and negative symptoms
Usually develops around 18 for males,
25 for females (plus or minus several
years)
About 1% of the population
Antipsychotics




Help control positive symptoms
(psychotic episodes)
Reduce re-hospitalisations
Significant side-effects
Ineffective for 20 to 30% of people with
schizophrenia
Introduction of cost-sharing in
August 1996

For welfare recipients and seniors

Welfare recipients: ceiling of $50 per
quarter


$16.67 per month for people with mental
illness
Tamblyn et al. 2001:


Reduction in use of medications
Increase in adverse events (deaths,
hospitalizations and nursing homes) and
ER visits
Consistent with other studies

Ward et al. 06 (and others): Antipsychotics
compliance:



Hospitalisations
Suicides, mortality
Soumerai et al. 94: Capping prescriptions for
people with schizophrenia in NH:



Antipsychotics
Emergency psychiatric services
Government costs
Selective removal of costsharing in October 1999



For welfare recipients classified as
disabled
Includes people with schizophrenia
classified as disabled, who typically
consume antipsychotics
No studies of effects of removing costsharing for antipsychotics identified
Qualitative interviews
Qualitative interviews




In 2004-2005, 23 interviews with psychiatrists,
nurses and social workers were conducted at 8
different sites in 6 Québec cities or towns
Consumers considered, but not included for reasons
of efficiency (many interviews needed to obtain
representative sample)
Urban and rural, teaching and non-teaching sites
included
Questions on various topics, including of relevance
here: Effects of introducing, then removing costsharing on people with schizophrenia
Main comments from
qualitative interviews


Some schizophrenia patients more
closely followed than others – costsharing would have bigger impact on
them
Removal of cost-sharing expected to
have smaller impact
OBJECTIVES


Re-evaluate impact of introducing costsharing on use of medications, for
people with schizophrenia, with larger
sample
Evaluate impact of removing costsharing 39 months later
Data
Identification of patients

Data extracted for people who had at least one
prescription of antipsychotics between Jan, 1st
1993 and Dec, 31st 2004 while on welfare status


Extracted from RAMQ:




107,005 individuals
Welfare status
Prescription data (DIN, duration, dose, charge, etc.)
Medical service data (type of service, Dx, etc.)
Extracted from Med-Echo:

Hospitalization data (Adm. & discharge dates, Dx, etc.)
Data cleaning
Data cleaning procedures on
prescription data

Conservative methods to ensure that all the
corrections are plausible.

When a value seems incorrect, either:



At least 2 arguments concur to correct a value and
we make this correction
Or we drop this prescription
Focus on cost, quantity and duration fields
Numbers of prescriptions affected by
data cleaning (based on 03 and 04
data only)
 Problems of duration of prescription (number of days) = 0
& quantity of drug (i.e., total number of pills or ml) = 0
& drug cost = 0 when all not equal to zero but at least one equal to zero
 442 (0.02%) prescriptions are concerned
Duration of prescription > 270 days
 131 (0.01%) prescriptions, only 7 could be corrected
Problem with the ratio cost to quantity
 91 (<0.01%) prescriptions were concerned, no one could be corrected
Problem of low dose
 34 (<0.01%) prescriptions were concerned, only 4 could be corrected
Patients with prescriptions that could not be corrected were eliminated from the
study
Adjustment of prescription
durations
Adjustments of the
prescriptions: why?

If we draw successions of prescriptions for some patients we
observe different patterns:
Jan, 1st
Jan, 15th
1/
2 prescriptions of the same DIN:
Jan, 1st
Jan, 14th
Jan, 1st
Could be
Pills are lost
2/
Jan, 12th
Jan, 25th
interpreted as
Jan, 12th
3/
2 pills these days
Bases for adjustment of the
prescription start dates and durations

Consulted community pharmacist near Douglas Institute

A renewal less than 20% ahead of end of previous prescription is assumed
to be an early refill

But, since a pharmacist must justify to the RAMQ why s/he would have
accepted to fill a renewal prescription if the patient asks for a refill more
than 20% too early, we do not do this automatically in such a case.



Consecutive refills that are more than 20% too early suggest a problem –
normally such events, if accepted by the pharmacist, are rare (e.g., going on
vacation, lost pills)
It could be an increase in dose
It could be an early renewal, concurrent with a new prescription, to synchronize
the prescriptions
Adjustment of the
prescriptions: algorithm

Two prescriptions of the same DIN and the same dosage
overlapped (even by more than 20%): we moved the start
date of the prescription forward, to make the prescription
begin when the previous one ended
Except if it was a too early renewal for the second time: we
supposed that this prescription began when it was filled and
that the remaining pills were lost.

Synchronized prescriptions = if there was a synchronization
(two or more DINs filled on the same day) the prescription
was considered as beginning when filled and the previous one
was stopped (considered as if the pills were lost)
Adjustments of the
prescriptions: hospitalizations

Sometimes a patient was supposed to fill a
prescription during a hospitalization (even when
the hospitalization was for a psychiatric reason).


We supposed that all these pills were lost
When a hospitalization occurred at a time when the
patient was on a prescription we supposed that all
the pills from that prescription were not taken
anymore and were considered as lost
Construction of the cohort





On welfare from 1993 to 2004 (ignoring interruptions < 1
month)
18+ in 1995 and alive in 2004
At least one prescription of antipsychotics every 180 days from
Jan 1st 1993 to July 31st 1996, removing hospitalization days
Schizophrenia Dx either on hospitalization records OR medical
records one or more times in the period 1993 – July 31st 1996
N=4,401
Proportion of days in month patient
had access to antipsychotics

Proportion of days in month that antipsychotics
available while in community


Adjustment for hospitalisations
< 10 days in community : Proportion undefined
First 9 months of 1993
excluded



No data from 1992
Don’t know when 1992 prescriptions end
Maximum prescription duration is 9 months
Subdivision of cohort into
3 sub-groups
0.6
0.5
0.4
0.3
Median coefficient of
variation
0.2
0.1
0
High stability
Moderate stability
Low stability
Estimation strategy


Test for fixed effects or random effects
Allow for different intercepts, linear and
quadratic time trends during pre-costsharing, cost-sharing, and post-costsharing periods
Results
Age and sex by stability
subgroup
45
40
35
30
25
Average age
20
% Female
15
10
5
0
High stability
Moderate stability
Low stability
Median Antipsychotic Possession
Ratio by Stability Subgroup
1.2
1
0.8
0.6
APR
0.4
0.2
0
High stability
Moderate stability
Low stability
.9
Change in Antipsychotic Posession Ratio Across Time, All
.85
Introduction of Cost Sharing
.8
Removal of Cost Sharing
0
50
100
150
Month
.8
.9
1
Change in Antipsychotic Posession Ratio Across Time, Stratified
.7
Introduction of Cost Sharing
.6
Removal of Cost Sharing
0
50
100
Month
High Stability Group
Low Stability Group
Moderate Stability Group
150
4
0
2
Density
6
8
Average APR in 6 months prior to cost-sharing
introduction minus average APR in 6 months after costsharing introduced (N=4401)
-1
-.5
0
drug_diff
.5
1
Median difference= 0.005 Mean difference= 0.046
0
5
Density
10
15
Average APR during 6 months after cost- sharing
removed minus average APR during 6 months prior
to cost-sharing removal (N=4401)
-1
-.5
Median difference= 0
0
delta_drug
.5
1
Mean difference= .0174596
Hausman test

Rejected at p<0.01

Use fixed effects
.9
Change in Antipsychotic Posession Ratio Across Time, All
.85
Introduction of Cost Sharing
.8
Removal of Cost Sharing
0
50
100
150
Month
.8
.9
1
Change in Antipsychotic Posession Ratio Across Time, Stratified
.7
Introduction of Cost Sharing
.6
Removal of Cost Sharing
0
50
100
Month
High Stability Group
Low Stability Group
Moderate Stability Group
150
.6 .7 .8 .9
1
Predicted Change in Antipsychotic Posession Ratio Across Time, All
Introduction of Cost Sharing
0
Removal of Cost Sharing
50
100
150
Month
.8
.9
1
Predicted Change in Antipsychotic Posession Ratio Across Time, Stratified
.7
Introduction of Cost Sharing
.6
Removal of Cost Sharing
0
50
100
Month
High Stability Group
Low Stability Group
Moderate Stability Group
150
Regression: High Stability
Subgroup (N=1466)
Regression: Medium Stability
Subgroup (N=1501)
Regression: Low Stability
Subgroup (N=1434)
Sensitivity analysis

Remove values 3 months before and 3
months after August 1 1996 and
October 1 1999


To mitigate any effects of stockpiling or
delaying purchasing of medications in
anticipation of policy change
Results qualitatively similar
Discussion
Conclusions


High-stability group: Permanent
reduction in APR, small effect of
removing cost-sharing
Other groups:


Long-term trends towards increased
consumption
Apparently greater effect of removing costsharing
Limitations




Non-experimental design: possible
confounding
CV classification crude
Schizophrenia Dx identification
Fixed cohort – drop-outs (welfare exit,
death) ignored, possible bias
Implications



Removing cost-sharing was effective
policy
Permanent effect of having introduced
cost-sharing – especially for high
stability group
Further evidence that cost-sharing for
antipsychotics undesirable