Access to this full-text is provided by SAGE Publications Inc.
Content available from Journal of Parkinson's Disease
This content is subject to copyright.
Journal of Parkinson’s Disease 1 (2011) 35–47
DOI 10.3233/JPD-2011-11024
IOS Press
35
Use of a Refined Drug Tracer Algorithm
to Estimate Prevalence and Incidence
of Parkinson’s Disease in a Large Israeli
Population
Orly Chillag-Talmora,1, Nir Giladib,c,∗, Shai Linna,d, Tanya Gurevichb,c, Baruch El-Ade, Barbara
Silvermane, Nurit Friedmaneand Chava Peretzc
aSchool of Public Health, University of Haifa, Haifa, Israel
bMovement Disorders Unit, Department of Neurology, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel
cSackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel
dEpidemiology Unit, Rambam Medical Center, Haifa, Israel
eMaccabi Healthcare Services, Tel Aviv, Israel
Abstract. Estimating rates of Parkinson’s disease (PD) is essential for health services planning and studies of disease determi-
nants. However, few PD registries exist. We aimed to estimate annual prevalence and incidence of PD in a large Israeli population
over the past decade using computerized drug purchase data.
Based on profiles of anti-parkinsonian drugs, age at first purchase, purchase density, and follow-up time, we developed a
refined algorithm for PD assessment (definite, probable or possible) and validated it against clinical diagnoses. We used the
prescription database of the second largest Health Maintenance Organization in Israel (covers ∼25% of population), for the
years 1998–2008. PD rates by age, gender and year were calculated and compared using Poisson models.
The algorithm was found to be highly sensitive (96%) for detecting PD cases. We identified 7,134 prevalent cases (67%
definite/probable), and 5,288 incident cases (65% definite/probable), with mean age at first purchase 69±13 years. Over the years
2000–2007, PD incidence rate of 33/100,000 was stable, and the prevalence rate increased from 170/100,000 to 256/100,000. For
ages 50+, 60+, 70+, median prevalence rates were 1%, 2%, 3%, respectively. Incidence rates also increased with age (RR= 1.76,
95%CI 1.75–1.77, ages 50+, 5-year interval). For ages 50+, rates were higher among men for both prevalence (RR = 1.38,
95%CI 1.37–1.39) and incidence (RR =1.45, 95%CI 1.42–1.48). In conclusion, our refined algorithm for PD assessment, based
on computerized drug purchases data, may be a reliable tool for population-based studies. The findings indicate a burden of PD
in Israel higher than previously assumed.
Keywords: Parkinson’s disease, Parkinson’s disease drug therapy, prevalence, incidence, drug tracer
1In partial fulfillment of the requirements for the PhD degree,
University of Haifa, Israel.
∗Correspondence to: N. Giladi, Department of Neurology, Tel
Aviv Sourasky Medical Center, 6 Weizmann Street, Tel Aviv 64239,
Israel. Tel.: +972 (0)3 6974790; Fax: +972 (0)3 6974153; E-mail:
nirg@tasmc.health.gov.il.
INTRODUCTION
Prevalence and incidence estimates of Parkinson’s
disease (PD) are essential for health services planning
and as a basis for studies of risk factors and poten-
tial disease modifying interventions. While regional
registries are the most accurate tool to follow PD mor-
bidity, few are currently in operation (e.g., Nebraska
ISSN 1877-7171/11/$27.50 © 2011 – IOS Press and the authors. All rights reserved
36 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
[1], California [2], and European [3] registries). PD
rates reported from ad-hoc observational studies vary
greatly due to differences in design and case defini-
tion [4–7]. The two classically-used designs to assess
PD rates, door-to-door surveys and clinic based studies
(i.e., of patients that have received medical atten-
tion, using combined sources of information, e.g.,
clinical records, medical claims, etc.), are costly and
time-consuming. Lack of long-term follow-up in door-
to-door studies impairs accuracy of PD diagnosis,
while PD documentation in medical records may be
inconsistent or inaccurate [5, 7]. Computerized phar-
macy purchasing databases are a relatively new and
reliable source of information, which enable utilization
of drug tracer methodology to estimate PD occurrence
based on consumption of specific anti-parkinsonian
drugs (APD) [8–11]. APD as a group, specifically the
dopaminergic agents (Anatomical Therapeutic Chem-
ical classification system (ATC) code N04B) [12], are
prescribed to all PD-diagnosed patients at some dis-
ease stage, and rather selectively for parkinsonism
in general [8]. The accuracy of rate estimates using
drug tracing depends on data completeness and case-
definition criteria. Only few studies previously used
this approach. Some employed aggregated purchase
data, estimating the number of PD cases based on total
APD sales divided by projected per-person utilization
[8, 13, 14]. Others used person-level data, but defined a
PD case as any person with at least one APD purchase
[10, 11].
The present study is based on the prescription
database of the second largest Health Maintenance
Organization (HMO) in Israel. Our aims were (1) to
develop and validate a refined drug-tracer algorithm
for assessment of PD cases at three levels of accuracy
– definite, probable and possible – based on patterns of
drug consumption, age and follow-up period; and (2)
to estimate the prevalence and incidence of PD in this
large Israeli population over the past decade.
PATIENTS AND METHODS
Design and study population
We conducted a retrospective cohort study of the
members of Maccabi Healthcare Services (MHS) –
over 1.8 million people nationwide (∼25% of the total
population), for the period 1.1.1998–31.12.2008. Since
January 1998, computer systems have captured all
pharmacy purchases covered by MHS. Each purchase
record includes the member’s identification number
(ID), purchase date and drug specifications. In Israel,
almost all APD are substantially subsidized for PD
patients through the National Health Plan. Thus, little
incentive exists for patients to purchase medications
outside the plan, and we could assume nearly com-
plete capture of the drug purchases of interest. APD are
dispensed for only one month of treatment, hence we
assumed each purchase represented treatment for the
following month. Treatment initiation served as proxy
for time of diagnosis.
Demographic characteristics of subjects – gender,
birth date, membership start-date and current status at
MHS (active/deceased/transferred to another HMO) –
were derived from MHS membership files.
Algorithm for PD assessment
We developed a refined drug-driven algorithm
(Appendix 1) to assess PD patients at three accuracy
levels – definite, probable and possible, based on the
fact that PD therapy is chronic and generally involves
increasing number of drug-types and dosages with dis-
ease progression. Thus, those levels of accuracy were
assigned based on specific combinations of categories
of four factors: (a) APD types used; (b) age at first APD
purchase (c) follow-up period (FUP); and (d) APD pur-
chase intensity – number and continuity of purchases,
as follows:
Anti-parkinsonian drugs (APD)
We employed dopaminergic APD (ATC code N04B)
as tracers (see Appendix 2 for included medications).
We excluded anticholinergic agents (ATC code N04A),
since they are frequently used in Israel for indica-
tions other than primary PD (e.g., neuroleptic-induced
parkinsonism), and only few PD patients are treated
exclusively with anticholinergics for a long period of
time [15]. Selected drugs were categorized into seven
groups according to mechanisms of action and clinical
use (Appendix 2). Purchases of specific groups or com-
bination of groups were supportive of PD diagnosis
accuracy. Subjects who purchased only bromocriptine
and were most likely treated for non-PD indications
(e.g. hyperprolactinemia or termination of lactation)
were excluded (criteria shown in Fig. 1).
Age at first purchase
We included only subjects aged 20–84 years at first
recorded purchase to exclude cases of juvenile parkin-
sonism, and elderly people who are often prescribed
levodopa or amantadine empirically for slowness or
gait and postural disturbances. Further, in combina-
tion with purchase patterns, three categories of age at
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 37
18,546 Subjects
with at least one APD
purchase during 1998-2008
18,458
2,102 excluded
16,356
8,511
7,389
Date errors
Age at 1st purchase is
under 20 or 85+
Number of months
with at least one
purchase < 3
Only bromocriptine
purchases, likely for
non-PD indicationsa
88 excluded
7,845 excluded
1,122 excluded
255 excluded
Minimal number of
purchases within defined
period was lower than
required 7,134 PD cases
Prevalent cohort
4,018
definite
738
probable
2,378
possible
5,288 PD cases
Incident cohort
2,781
definite
601
probable
1,906
possible
1st purchase in 1998, n=1,846
Fig. 1. Selection criteria for prevalent and incident PD cases. APD
– anti-parkinsonian drugs. aExcluded: women with age at 1st pur-
chase < 50, subjects with ≥1 year between last purchase and end of
follow-up, subjects with 1st purchase after 2002 (PD treatment initia-
tion with bromocriptine unlikely due to availability of newdopamine
agonists).
first purchase were taken into account in PD assess-
ment – less than 65, 65–74 and 75–84 – assuming that
the likelihood of initiating APD treatment for non-PD
(misdiagnosis/empiric treatment) increases with age at
first purchase [16].
Follow-up period (FUP)
FUP was calculated as time elapsed from date of
first APD purchase to the earliest of the following
dates: end of study (31.12.2008), transfer out of MHS,
or death. FUP was categorized as “long” (≥3 years)
or “short” (<3 years). Longer FUP was considered
supportive of PD diagnosis accuracy. This concept is
backed by reports that clinical follow-up of three years
or more, particularly by a movement disorders special-
ist, improves the accuracy of the clinical diagnosis in
clinical-pathological confirmation studies [17–19].
Purchase intensity
A “purchase month” was defined as a month in
which at least one purchase was made. The number
of purchase months of any drug and of each APD
group was calculated for consecutive, 12-months long,
segments of FUP. For initial inclusion, at least three
purchase months during the entire FUP were necessary.
Furthermore, we required at least one FUP-segment
with a minimum of three purchase months (this crite-
rion was modified for short FUP cases, see Appendix
1). The algorithm accounted for both number of pur-
chase months and purchase continuity (i.e., number of
purchase months per time observed) in assigning PD
accuracy level.
Characteristics of cases assigned a definite level
of accuracy
Following are the major algorithm principles for
assigning a definite accuracy level (full algorithm
details appear in Appendix 1). For subjects with “long”
FUP (≥3 years), criteria were applied to the set of three
consecutive years with the highest purchase density,
and cases were defined as definite if records showed:
high purchase intensity (e.g., 9 purchase months out
of 12) of either levodopa or dopamine agonists or
monoamine oxidase inhibitors (MAOB-I) (the lat-
ter conditioned by age at first purchase < 75), OR;
extended purchase intensity (18/24 months) of either
amantadine or MAOB-I (the latter conditioned by age
at first purchase ≥75), OR; simultaneous purchase of a
combination of APD types (6/12 months). For subjects
with “short” FUP, criteria were applied to the entire
FUP. Cases were defined as definite if: they fulfilled
the long-FUP criteria for a definite accuracy level, OR;
they fulfilled the long-FUP criteria for a probable accu-
racy level, conditioned by age at first purchase ≤65
and purchase of either levodopa or dopamine agonists
or MAOB-I.
Selection of eligible patients for the PD cohorts
The prevalent PD cohort included all patients who
met the algorithm criteria during 1998–2008. The inci-
dent PD cohort excluded cases whose first purchase
was during 1.1.1998–31.12.1998, who may have been
treated prior to study initiation (Fig. 1).
Validation of PD assessment by the algorithm
We compared our algorithm-derived identification
of PD cases to diagnoses from a specialist outpatient
clinic in a tertiary medical center – the Movement
Disorders Unit (MDU) in the Tel Aviv Sourasky Med-
ical Center (TASMC). All four neurologists on the
team specialize in movement disorders, and have been
38 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
working in MDU over ten years; thus the TASMC-
MDU diagnoses were considered the gold standard
[19]. Diagnoses of MHS members who visited MDU
between mid-2003 and 2008 were retrieved from the
MDU electronic records and linked by ID to the
algorithm-driven PD assessment: patients with MDU
diagnosis of PD were employed to calculate sensitivity
of the algorithm (true positive rate), and patients with
other diagnoses (parkinsonism, gait disorders, essen-
tial tremor and non PD-related dyskinesia/spasticity)
were employed to calculate the false positive rate of
the algorithm within this patient population.
Data analysis
Annual PD prevalence and incidence rates (per
100,000), overall and gender- and age-specific, were
calculated for the years 2000–2007, since MHS mem-
bership data by age and gender were available as of
2000. The year 2008 was not included because our
algorithm was less likely to identify cases diagnosed
later in the study period, due to shorter follow-up time.
Prevalence was based on number of PD patients active
in MHS on December 31st of each calendar year;
annual incidence referred to PD patients making their
first purchase during the calendar year.
Poisson regression models were applied to study the
effect (RR and 95% CI) of gender, age category (5-
year intervals) and calendar year on annual prevalence
and incidence rates of PD, for the entire group and for
patients aged 50+, 60+ and 70+.
Although representing a broad cross-section of the
Israeli population, MHS population is younger. Thus,
we estimated the number of prevalent and incident PD
cases for the entire Israeli population in 2005 based on
the calculated prevalence/incidence rates and on the
national gender- and age-distribution [20].
Ethics
The study was approved by the Institutional Review
Boards (IRBs) of both TASMC and MHS. It was
based on anonymous databases and involved no direct
interaction with patients, thus the IRBs approved that
informed consent was not required.
RESULTS
Study population: PD cohorts, 1998–2008
Based on 499,629 APD prescriptions dispensed
to 18,546 MHS members between 1.1.1998 and
31.12.2008, a cohort of 7,134 prevalent PD cases was
identified by our algorithm (Fig. 1). We excluded
11,412 subjects for the following reasons: apparent
errors in purchase dates, or age at first purchase less
than 20 or over 84 years (2,190); fewer than 3 purchase
months (7,845); probable treatment with bromocrip-
tine for other indications (1,122); and fewer than the
minimum purchase months required within an FUP
segment (255). Among the 7,134 cases of the PD
prevalent cohort, 56% (n= 4,018) were identified as
definite and 11% as probable cases. The incident cohort
included 5,288 cases over the entire study period with
distribution of accuracy level similar to that of the
prevalent cohort (Table 1).
Algorithm validation
Of 625 MDU patients (with different diagnoses)
identified, 621 (99%) were confirmed by MHS as mem-
bers. For MDU patients diagnosed with idiopathic PD,
the algorithm sensitivity was 96% (179/186). The algo-
rithm’s false positive rate among patients with other
movement disorders varied across different diagnoses:
82% of parkinsonism cases were falsely identified
as PD, while only 4% of patients with non PD-
related dyskinesia/spasticity syndromes received false
PD identification (Appendix 3).
Demographics and purchase characteristics
Men comprised 52% of prevalent PD cases. There
were more men than women among the definite and
probable accuracy level, but more women in the pos-
sible accuracy level.
Mean age at first purchase was 69 ±13 years,
slightly older among women in the probable and def-
inite accuracy groups. Mean number of total purchase
months during the study period was 32.3 ±31, and
mean FUP was 5 ±3 years, longer for women (5.3
years vs. 4.7). In accordance with algorithm require-
ments, definite PD cases had the longest FUP and the
largest number of total purchase months. Distribution
of gender and age at first purchase in the incident cohort
were similar to those of the prevalent cohort (Table 1).
Sensitivity analyses
Changing the inclusion criterion of minimum pur-
chase months during the study period from three to
four, or changing the cutoff point for long/short FUP
from three to two years had negligible or no effect on
accuracy level distribution, mean age at first purchase
and purchase characteristics distribution.
Exclusion of all subjects who purchased only
dopamine agonists (n= 186, 2.6%), in order to con-
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 39
Table 1
Characteristics of prevalent (A) and incident (B) PD cases by gender and accuracy level, 1998–2008a
Accuracy level A. prevalent cases (n=7134) B. incident cases (n= 5288)
men women total men women total
n(%) definite 2179 (59) 1839 (54) 4018 (56) 1533 (55) 1248 (50) 2781 (53)
probable 404 (11) 334 (10) 738 (11) 329 (12) 272 (11) 601 (11)
possible 1126 (30) 1252 (36) 2378 (33) 923 (33) 983 (39) 1906 (36)
total 3709 (100) 3425 (100) 7134 (100) 2785 (100) 2503 (100) 5288 (100)
age at 1st purchase,
mean (SD)
definite 69.6 (10.8) 70.3 (11.1) 69.9 (11.0) 69.4 (11.1) 70.6 (10.8) 69.9 (11.0)
probable 69.6 (11.4) 70.9 (10.6) 70.2 (11.1) 68.7 (11.8) 70.2 (10.8) 69.4 (11.4)
possible 68.9 (14.3) 65.5 (18.0) 67.1 (16.4) 69.2 (14.3) 67.9 (16.2) 68.5 (15.3)
total 69.4 (12.1) 68.6 (14.2) 69.0 (13.1) 69.2 (12.3) 69.5 (13.2) 69.4 (12.8)
follow-up time from
1st purchase in
years, mean (SD)
definite 5.4 (3.1) 5.8 (3.2) 5.6 (3.1) 4.5 (2.5) 4.7 (2.5) 4.6 (2.5)
probable 3.1 (2.6) 3.5 (2.7) 3.3 (2.6) 2.8 (2.3) 3.2 (2.4) 3.0 (2.3)
possible 3.9 (3.2) 5.0 (3.6) 4.5 (3.4) 3.4 (2.7) 4.1 (2.9) 3.7 (2.8)
total 4.7 (3.2) 5.3 (3.3) 5.0 (3.3) 3.9 (2.6) 4.3 (2.7) 4.1 (2.7)
total purchase
months, mean
(SD)
definite 48.0 (31.2) 48.2 (31.0) 48.1 (31.1) 39.2 (23.5) 39.1 (23.4) 39.2 (23.4)
probable 17.5 (13.8) 18.9 (14.2) 18.2 (14.0) 16.6 (12.7) 18.1 (13.3) 17.3 (13.0)
possible 10.4 (15.0) 9.4 (10.9) 9.9 (13.0) 8.5 (9.2) 7.9 (7.0) 8.2 (8.1)
total 33.3 (31.2) 31.1 (30.4) 32.3 (30.9) 26.3 (23.6) 24.6 (23.) 25.5 (23.4)
aCharacteristics distribution was similar for the prevalent and incident cases of 2000–2007, which were included in the rates calculations.
trol for potential bias due to inclusion of patients with
restless leg syndrome, did not alter the results.
Prevalence
Annual prevalence rates increased from 170/
100,000 in 2000 to 256/100,000 in 2007 (Table 2A),
6% per year (RR = 1.06, 95%CI = 1.04–1.08, gender-
adjusted). When considering only the definite cases,
prevalence rate increased by 5% per year (RR= 1.05,
95%CI = 1.03–1.06, gender-adjusted). Figure 2
presents the increase in prevalence rates over time of
the definite PD cases vs. all cases, demonstrating a
similar trend (accounting for the under-estimation in
the last study years due to our criteria for a definite
case which depends on longer FUP). We found that
the increase in prevalence rate varied significantly
across different age groups (significant age*year inter-
action effect, p < 0.01, in a hierarchical model which
included also year, age and gender as main effects),
e.g., it was 4% for ages 35–55, 2% for ages 55–85,
and 13% for ages 85+. Exclusion of the 85+ age
group yielded annual prevalence rates that increased
by 5% per year (RR = 1.05, 95%CI = 1.03–1.06,
gender-adjusted).
Prevalence rate significantly increased with age:
median annual rate was 1.0% for population aged
50+, 1.9% for ages 60+, and 3.3% for ages 70+. A
5-year increment in age resulted in a 50% increase
in prevalence rate (RR = 1.496, 95%CI = 1.493–1.499)
(Fig. 3A, Appendix 4A).
Annual prevalence rates were somewhat higher for
men compared to women, but did not differ signifi-
cantly. However, in the subgroup of patients aged 50
years and up (approximately 90% of cases), prevalence
rates among men were significantly higher (RR = 1.38,
95%CI = 1.37–1.39; Fig. 3A).
Based on age- and gender-specific rates and on
the age- and gender-distribution of the general Israeli
population for 2005, we estimated that there were
approximately 23,100 Israelis with PD that year, result-
ing in a standardized prevalence rate of 334/100,000.
Incidence
Between 2000 and 2007, annual incidence rate
remained stable at approximately 33/100,000, higher
by 20% for men (RR = 1.19, 95%CI = 1.00–1.41)
(Table 2B).
In a more stringently defined sub-group of incident
cases (92% of the incident cohort), from which we
excluded patients who joined MHS after 1.1.1998 and
made their first purchase within less than one year,
the mean incidence rate was reduced to 30/100,000,
but the men/women ratio remained similar. Stability
of the incidence rate along time was also found among
the definite cases (Fig. 2).
Incidence rate increased significantly with age
(Fig. 3B, Appendix 4B): over 8 years, median annual
incidence rates (per 100,000) for ages 50+, 60+ and
70+ (up to 85) were 165, 312 and 562 respectively.
In the subgroup of incident patients aged 50 years
and up at first purchase (∼90% of new cases), incidence
rates increased by 76% per 5-year increment in age
(RR =1.76, 95%CI = 1.75–1.77) and were 45% higher
for men (RR = 1.45, 95%CI = 1.42–1.48).
40 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
Table 2
PD prevalence (A) and incidence (B) rates per 100,000 by gender and year, 2000–2007
YearaA. prevalence rate /100,000 B. incidence rate /100,000
men women total men women total
rate (n) rate (n) rate (n) rate (n) rate (n) rate (n)
2000 170.8 (1232) 168.3 (1287) 169.5 (2519) 34.2 (247) 31.8 (243) 33.0 (490)
2001 187.0 (1406) 181.4 (1446) 184.1 (2852) 37.0 (278) 29.5 (235) 33.1 (513)
2002 202.1 (1557) 198.7 (1625) 200.4 (3182) 36.2 (279) 31.8 (260) 33.9 (539)
2003 213.5 (1685) 211.0 (1763) 212.2 (3448) 32.4 (256) 29.2 (244) 30.8 (500)
2004 228.2 (1844) 230.1 (1965) 229.2 (3809) 38.9 (314) 34.1 (291) 36.4 (605)
2005 241.3 (1988) 237.3 (2064) 239.2 (4052) 35.1 (289) 29.7 (258) 32.3 (547)
2006 251.1 (2109) 247.5 (2192) 249.2 (4301) 35.0 (294) 29.6 (262) 32.2 (556)
2007 260.6 (2237) 251.6 (2275) 256.0 (4512) 38.8 (333) 26.8 (242) 32.6 (575)
aPrevalent cases for a specific year were cases whose follow-up period included December 31st of that year.
Incident cases for a specific year were cases whose 1st purchase occurred in that year.
Fig. 2. PD prevalence and incidence rates per 100,000 by year, definite vs. all cases, 2000–2007. Numbers of definite cases in last years of the
study period are lower than expected due to the algorithm criteria – a definite accuracy level is generally dependent on a longer follow-up period.
Based on age- and gender-specific rates and on the
age- and gender-distribution of the general Israeli pop-
ulation for 2005, we estimated that the number of
incident PD cases that year in Israel was 3,100, result-
ing in a standardized incidence rate of 45/100,000.
DISCUSSION
Pharmacy purchase databases are a highly valid
source of drug utilization in populations with a uni-
versal drug benefit. They are very accurate and closely
monitored, since they are maintained for administrative
purposes. Thus, pharmacy databases enable observa-
tional studies of large populations with long follow-up,
reduced selection bias and increased generalizability.
The main limitation of this approach is inclusion
of patients with atypical and secondary parkinson-
ism who might receive APD-based treatment. This
difficulty in differentiating primary PD from other
parkinsonian syndromes is likely to cause over-
estimation of PD patients. Thus, our rates may be
over-estimation of the actual numbers. Additionally,
when using a drug tracing approach, undiagnosed
patients or patients not treated with medications are
not detectable.
In this study we developed a unique drug tracer
algorithm for PD assessment, which demonstrated a
very high sensitivity (96%), and a reasonable rate of
false identification of other movement disorders as PD
(except for parkinsonism, as expected). Our algorithm
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 41
Fig. 3. PD prevalence (A) and incidence (B) rates per 100,000 by year, gender and age, 2000–2007.
also accounted for major determinants of PD diagnosis
– age at first purchase and individual follow-up time,
and generated individual identification of PD accuracy
level. In order to reduce bias of over-identification of
cases, we employed a conservative approach in con-
structing the algorithm that gives balanced priority to
specificity, i.e., favors under-estimation. For example
we used criteria such as minimum three months of pur-
chases, exclusion of patients who began treatment at
85 years or later, purchase intensity requirements, etc.
Compared with previous drug-tracer studies [9–11, 13,
14], our follow-up was long (11 vs. 2–8 years), and in
addition to levodopa, a wide yet specific range of med-
ications was used, allowing for a variety of treatment
combinations.
It is clear that the algorithm we used captures inci-
dent cases quite well, demonstrating rates higher than
expected, stable across the study period. Over the same
period a rise in prevalence rates was observed, similar
to findings of a recent Canadian study [21]. It should
be noted that the increase in the prevalence rate over
the study period differed significantly between the dif-
ferent age groups. The steep rise in the 85+ age group
was an artificial enhancement, due to algorithm crite-
ria, thus the trend must be interpreted with caution. The
increase in prevalence rate over time can be explained
by a combination of factors: the long duration of PD,
which is reflected in accumulation of cases; general
increase in longevity (life expectancy in the general
Israeli population increased by 2.8 years over the last
decade [22]); and possibly improvement in medical
care expressed in actual increase in disease duration. It
should be considered that non-inclusion in the cohort
of subjects aged 85+ at first purchase may have caused
over-estimation of the increase in prevalence over the
entire study period and under-estimation of rates dur-
42 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
ing the first study years. Our finding of an increment
of parkinsonian prevalence has major impact on med-
ical policy, but further research is needed to better
understand the causes of this trend.
In accordance with previous reports, prevalence and
incidence rates for ages 50 and up were significantly
higher (by ∼40%) among men compared to women,
and increased considerably with age, with prevalence
decreasing in the oldest age group [4, 6, 7, 16, 23, 24].
Age-specific rates in this study were higher than those
reported in several European studies [23], e.g., the
prevalence rate among persons aged 60 years or older
was almost 2-fold higher than the 1% rate traditionally
cited for this age group [7].
Our adjusted estimators of PD prevalence and inci-
dence in 2005 for the general Israeli population of
334/100,000 and 45/100,000, respectively, are very
high compared to other studies. For UK, Sweden, Italy,
Spain, USA and other developed countries, rate ranges
were: 60–350/100,000 for prevalence, 5–26/100,000
for incidence [4–7, 23]. Since data about ethnic origin
was unavailable from MHS databases, our estimations
do not account for ethnicity, although this factor may
be relevant in the mixed Israeli population: Among
Ashkenazi Jews rates may be higher, due to the
high frequency of PD-associated mutations in LRRK2
(G2019S) and GBA genes among this population in
Israel [25, 26]; This may account for the high occur-
rence we observed. However, Arab ethnicity also might
be of relevance – PD prevalence rates among Israeli
Arabs (∼10% of Israel’s population> 50 years) were
suggested to be lower [27, 28]. Our finding of a grow-
ing population of PD patients in Israel suggests that a
figure of 16,000 patients for 2005, based on extrapola-
tion of the findings of Anca et al. in Israel [20, 29], is
an under-estimate.
Mean age at first recorded purchase was 69 for
both prevalent and incident cases, higher than the
range 58–63 often reported in studies as age at
onset/diagnosis [6], but in accordance with several
population-based studies that reported age at onset
between 66 and 71 [6, 24, 29–31]. Evidently, age at
first medication purchase is a proxy of diagnosis age
rather than onset of symptoms, and also includes a lag-
time to treatment initiation, which may be longer than
one year [24, 32]. The high age may also result from
over-representation of patients with treatment initia-
tion at older age, although we tried to minimize this
bias by excluding cases with first purchase at 85 years
of age or later, when diagnosis is very challenging and
empiric treatment is common [16].
In conclusion, our proposed algorithm may be used
as a reliable and low-cost tool to establish PD cohorts
for epidemiological studies. Our findings of prevalence
and incidence higher than expected, and a rising num-
ber of PD patients in Israel reflect the growing burden
of PD morbidity on Israeli health and social systems,
and should be the basis for future national resource
planning.
ACKNOWLEDGMENTS
We would like to thank all members of the MDU
team for their help with the validation procedure. This
work was supported in part by grant SGA0902 from
the Environment and Health Fund, Jerusalem, Israel.
O. Chillag-Talmor received a scholarship (stipend) for
research students from the University of Haifa.
APPENDIX
Appendix 1A
Algorithm for identifying PD cases and assigning them to accuracy levels (definite, probable,
possible) based on drug purchase data
Definitions
Censoring Death, transfer to another HMO, or end of study (December 31st, 2008), whichever occurred first.
Follow-up period (FUP) Time from 1st purchase to censoring.
Observation segment The FUP is divided into consecutive observation segments of 12 months each, and a last observation segment
with the residual number of months.
Purchase month A month in which at least one purchase was made.
Final purchase gap Time from last purchase to censoring.
Examined interval The examined interval is the period upon which most criteria are applied. For subjects with FUP≥3 years (long
FUP), the examined interval is a period of 3 consecutive, 12-months long observation segments, in which the
purchase intensity (sum of purchase months of drug groups 1–6) was highest (see appendix 2 for list of drugs
and groups). For subjects with FUP < 3 years (short FUP), the examined interval is their full FUP.
Lag to 1st purchase Time from the later between start of study (January 1st, 1998) and the start date of membership in MHS to 1st
purchase. It is assumed that a lag to 1st purchase ≥1 year implies an actual 1st purchase, while a lag< 1 year
suggests that drug purchases may have occurred prior to the 1st purchase recorded in the data employed in the
study (i.e., before the study began or before the subject joined MHS).
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 43
Appendix 1B
Full-detail algorithm. Terms defined above (section A) are italicized in the algorithm table below
1. minimum 1 purchase during the study period 1.1.1998–31.12.2008 no−→ exclusion
yes↓
2. 20 ≤age at 1st purchase <85 no−→ exclusion
yes↓
3. minimum 3 purchase months within FUP no−→ exclusion
yes↓
4. purchases of GR 3 only (bromocriptine) AND pattern suggesting indications other
than PD, namely: subjects with 1st purchase in 2003 or later, or subjects with final
purchase gap ≥365 days, or women with age at 1st purchase <50
yes−→ exclusion
no↓
5. FUP ≥3 years no−→ go to step 15,
algorithm for
FUP < 3 years
yes↓
********algorithm for subjects with FUP ≥3 years********
6. at least 1 observation segment with a minimum of 3 purchase months no−→ exclusion
yes↓
7. at least 1 purchase month of GR 7 (apomorphine) yes−→ certainty level:
definite
no↓
Note: As of step 8, all criteria are applied to the examined interval
8. During the examined interval, at least 24 purchase months of GR 1, 2, 4 or 5 yes−→ certainty level:
definite
no↓
9. if age at 1st purchase <75: during at least 1 of the observation segments within the
examined interval, minimum 9 purchase months of GR 1, 2 or 4; OR during any 2
of the observation segments within the examined interval, minimum 18 purchase
months of GR 5
yes−→ certainty level:
definite
if age at 1st purchase ≥75: during at least 1 of the observation segments within the
examined interval, minimum 9 purchase months ofGR1or2;OR during any 2 of
the observation segments within the examined interval, minimum 18 purchase
months ofGR4or5
no↓
10. during at least 1 of the observation segments within the examined interval,
minimum 6 simultaneous purchase months of drugs of 2 groups or more, any
combination excluding (4+5)
yes−→ certainty level:
definite
no↓
11. if age at 1st purchase <75: during at least 1 of the observation segments within the
examined interval, minimum 6 purchase months ofGR1,2or4,or9purchase
months of GR 5; OR during any 2 of the observation segments within the
examined interval, minimum 16 purchase months of GR 5
yes−→ go to step 13
if age at 1st purchase ≥75: during at least 1 of the observation segments within the
examined interval, minimum 6 purchase months ofGR1or2,or9purchase
months ofGR4or5;OR during any 2 of the observation segments within the
examined interval, minimum 16 purchase months ofGR4or5
no↓
12. during at least 1 of the observation segments within the examined interval,
minimum 6 simultaneous purchase months of drugs of GR (4+5), OR minimum 3
simultaneous purchase months of drugs of 2 groups or more, any combination
excluding (4+5)
no−→ go to step 14
yes↓
13. final purchase gap < 365 days yes−→ certainty level:
probable
no↓
44 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
Appendix 1B
(Continued)
********algorithm for subjects with FUP < 3 years********
14. remaining subjects yes−→ certainty level:
possible
end
15. age at 1st purchase ≤65; OR lag to 1st purchase ≥1 year and deceased during the
study period
yes−→ go to step 6
no↓
16. at least 1 observation segment with a minimum of 3 purchase months, OR during
any 2 observation segments within the examined interval, minimum 4 purchase
months,OR – for subjects with FUP < 2 years – minimum 3 purchase months
within the entire FUP
no−→ exclusion
yes↓
17. lag to 1st purchase< 1 year and deceased during the study period yes−→ go to step 7
no↓
18. at least 1 purchase month of GR 7 (apomorphine) yes−→ certainty level:
definite
no↓
19. During the examined interval, at least 24 purchase months of GR 1, 2, 4 or 5 yes−→ certainty level:
definite
no↓
20. during at least 1 of the observation segments within the examined interval,
minimum 6 purchase months of GR 1, 2 or 4; OR during any 2 of the observation
segments within the examined interval, minimum 18 purchase months of GR 5
yes−→ certainty level:
definite
no↓
21. during at least 1 of the observation segments within the examined interval,
minimum 6 simultaneous purchase months of drugs of 2 groups or more, any
combination excluding (4+5), OR minimum 3 simultaneous purchase months of
drugs of 2 groups or more, any combination of GR 1, 2, 4, 6
yes−→ certainty level:
definite
no↓
22. during at least 1 of the observation segments within the examined interval,
minimum 9 purchase months of GR 5; OR during any 2 of the observation
segments within the examined interval, minimum 16 purchase months of GR 5
yes−→ certainty level:
probable
no↓
23. during at least 1 of the observation segments within the examined interval,
minimum 6 simultaneous purchase months of drugs of GR (4+5), OR minimum 3
simultaneous purchase months of drugs of 2 groups or more, any combination
excluding (4+5)
yes−→ certainty level:
probable
no↓
24. final purchase gap < 365 days no−→ go to step 26
yes↓
25. at least 1 observation segment with a minimum of 3 purchase months of GR 1, 2,
or 4 OR during any 2 observation segments within the examined interval,
minimum 4 purchase months and no purchases of GR 3 or 5, OR – for subjects
with FUP < 2 years – minimum 3 purchase months within the entire FUP and no
purchases of GR 3 or 5
yes−→ certainty level:
probable
no↓
26. remaining subjects yes−→ certainty level:
possible
end
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 45
Appendix 2
Generic drug list for tracing PD patients
ATC group name ATC code Generic namea(drug group in study) Mechanism of action
Dopa and dopa derivatives N04B A 02 Levodopa+carbidopa (group 1)
Levodopa+benserazide (group 1)
Dopamine precursor + inhibitor of dopa
decarboxylase
N04B A 03 Levodopa+carbidopa+entacapone
(group 1)b
Dopamine precursor + peripheral dopa
decarboxylase inhibitor + COMT inhibitor
Adamantane derivatives N04B B 01 Amantadine (group 5) Dopaminergic, anti-viral
Dopamine agonistscN04B C 01 Bromocriptine (group 3) Dopamine agonist
N04B C 02 Pergolide (group 2) Dopamine agonist
N04B C 04 Ropinirole (group 2) Dopamine agonist
N04B C 06 Cabergoline (group 2) Dopamine agonist
N04B C 07 Apomorphine (group 7) Dopamine agonist
N04B C 10 Lisuride (group 2) Dopamine agonist
MAO B inhibitors N04B D 01 Selegiline (group 4) MAO B inhibitor
N04B D 02 Rasagiline (group 4) MAO B inhibitor
Other dopaminergic
agents
N04B X 01 Tolcapone (group 6) COMT inhibitor
N04B X 02 Entacapone (group 6) COMT inhibitor
COMT – catechol-O-methyltransferase; MAO – monoamine oxidase.
aAll commercial preparations available during the study period were included, apart from specific preparations of amantadine (Influ-A®) and
low-dose cabergoline (Dostinex®, Cabotrim®) which are not indicated for PD in Israel.
bStalevo®was included only in Group 1.
cPramipexole was not included as it was not available in Israel until 2009.
Appendix 3
Validation of the algorithm-driven assessment of PD patients based on drug purchases
Gold standard diagnosisa
PD Other movement disorders
Parkinsonismb(not PD) Gait disorders essential tremor non PD-related dyskinesia
or spasticity syndromes
Algorithm
assessment
PD 179 (96%) 60 (82%) 16 (27%) 4 (22%) 12 (4%)
not PD 7 (4%) 13 (18%) 44 (73%) 14 (78%) 272 (96%)
total 186 73 60 18 284
aDiagnosis was made by a movement disorders specialist from the Movement Disorders Unit at the Tel Aviv Sourasky Medical Center.
bIncludes: drug-induced parkinsonism, Parkinson plus, and other parkinsonian syndromes.
Appendix 4
Ageb(years) 2000 2001 2002 2003 2004 2005 2006 2007
A. PD prevalence rates per 100,000 by yeara, gender and age, 2000–2007
Men 50–55 144 151 158 170 147 146 157 170
55–60 281 270 248 274 341 320 299 317
60–65 536 659 664 583 494 523 567 593
65–70 1100 979 1097 1164 1222 1231 1272 1282
70-75 2090 2394 2404 2351 2490 2395 2132 2079
75–80 3853 3912 3923 3970 4041 4284 4400 4441
80–85 5666 5656 6028 6182 6666 6947 6737 6456
85+ 2552 3184 3567 4423 4318 4661 5191 5460
Women 50–55 124 107 118 120 131 133 136 130
55–60 192 201 215 224 255 244 239 249
60–65 419 402 382 379 367 414 428 454
65–70 775 789 797 832 898 881 836 788
70-75 1552 1691 1735 1701 1756 1782 1778 1732
77–80 2656 2738 2950 3067 3251 3230 3315 3217
80-85 4307 4499 4773 4878 4968 4817 4820 4903
85+ 1325 1597 1992 2475 3186 3476 3962 3842
46 O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel
Appendix 4
(Continued)
Ageb(years) 2000 2001 2002 2003 2004 2005 2006 2007
B. PD incidence rates per 100,000 by yeara, gender and age, 2000–2007
Men 50–55 49 32 22 29 21 14 27 38
55–60 48 28 34 39 57 50 37 61
60–65 116 194 145 118 71 89 113 111
65–70 279 192 212 141 183 148 218 208
70–75 333 601 453 374 517 457 371 358
75–80 924 745 732 666 817 702 644 753
80–85 817 1093 1218 1148 1333 1184 893 1007
Women 50–55 17 14 14 27 17 19 19 17
55–60 50 26 31 25 52 30 47 36
60-65 105 87 67 80 65 69 48 64
65-70 185 141 153 133 164 103 152 111
70–75 338 341 282 357 344 311 279 267
75–80 569 577 688 578 631 578 475 457
80–85 842 667 793 572 784 628 645 469
aPrevalent cases for a specific year were cases whose follow-up period included December 31st of that year. Incident cases for a specific year
were cases whose 1st purchase occurred in that year.
bData presented for cases aged 50+ on prevalence day for prevalence and on day of 1st purchase for incidence. These account for approximately
90% of all prevalent and incident cases, respectively.
REFERENCES
[1] Nebraska Department of Health & Human Services (2007)
Nebraska Parkinson’sDisease Registry, 2007 http://www.hhs.
state.ne.us/ced/parkinsons/, Accessed on March 23, 2008.
[2] The Parkinson’s Institute (2006) California PD Reg-
istry - Legislative History, 2006. http://www.thepi.org/site/
parkinson/section.php?id=92, Accessed on March 23, 2008.
[3] Sautter J, Wick R, Adlkofer F, & Baker MG (2003) Research
in the European Union. Lancet Neurol,2, 702-706.
[4] Zhang ZX & Roman GC (1993) Worldwide occurrence of
Parkinson’s disease: an updated review. Neuroepidemiology,
12, 195-208.
[5] Checkoway H & Nelson LM (1999) Epidemiologic
approaches to the study of Parkinson’s disease etiology. Epi-
demiology,10, 327-336.
[6] Twelves D, Perkins KS, & Counsell C (2003) Systematic
review of incidence studies of Parkinson’s disease. Mov Dis-
ord,18, 19-31.
[7] de Lau LM & Breteler MM (2006) Epidemiology of Parkin-
son’s disease. Lancet Neurol,5, 525-535.
[8] de Pedro Cuesta J & Rosenqvist U (1984) Tracers for paralysis
agitans in epidemiological research. Neuroepidemiology,3,
82-96.
[9] Menniti-Ippolito F, Spila-Alegiani S, Vanacore N, Bonifati
V, Diana G, Meco G, & Raschetti R (1995) Estimate of
parkinsonism prevalence through drug prescription histo-
ries in the Province of Rome, Italy. Acta Neurol Scand,92,
49-54.
[10] Lai BC, Schulzer M, Marion S, Teschke K, & Tsui JK (2003)
The prevalence of Parkinson’s disease in British Columbia,
Canada, estimated by using drug tracer methodology. Parkin-
sonism Relat Disord,9, 233-238.
[11] Brandt-Christensen M, Kvist K, Nilsson FM, Andersen PK,
& Kessing LV (2006) Use of antiparkinsonian drugs in Den-
mark: results from a nationwide pharmacoepidemiological
study. Mov Disord,21, 1221-1225.
[12] World Health Organization Collaborating Centre for Drug
Statistics Methodology (2008) Anatomical Therapeutic
Chemical Classification System with Defined Daily Dose
(ATC/DDD) Index 2008. http://www.whocc.no/atcddd/,
Accessed on January 30, 2008.
[13] Aquilonius SM & Hartvig P (1986) A Swedish county with
unexpectedly high utilization of anti-parkinsonian drugs. Acta
Neurol Scand,74, 379-382.
[14] Martinez-Suarez MM & Blazquez-Menes B (2000) [Esti-
mation of the prevalence of Parkinson’s disease in Asturia,
Spain. A pharmacoepidemiological study of the consumption
of antiparkinson drugs]. Rev Neurol,31, 1001-1006.
[15] National Institute for Health, Clinical Excellence (2006)
Parkinson’s disease: Diagnosis and management in pri-
mary and secondary care. NICE clinical guideline 35, 2006.
http://www.nice.org.uk/CG035, Accessed on October 10,
2010.
[16] Kuopio AM, Marttila RJ, Helenius H, & Rinne UK (1999)
Changing epidemiology of Parkinson’s disease in southwest-
ern Finland. Neurology,52, 302-308.
[17] Gibb WR & Lees AJ (1988) The relevance of the Lewy body
to the pathogenesis of idiopathic Parkinson’sdisease. J Neurol
Neurosurg Psychiatry,51, 745-752.
[18] Gelb DJ, Oliver E, & Gilman S (1999) Diagnostic criteria for
Parkinson disease. Arch Neurol,56, 33-39.
[19] Hughes AJ, Daniel SE, Ben-Shlomo Y, & Lees AJ (2002)
The accuracy of diagnosis of parkinsonian syndromes in a
specialist movement disorder service. Brain,125, 861-870.
[20] Israel Central Bureau of Statistics (2006) Statistical abstract
of Israel, 2006: Table 2.18 - Population, by population group,
religion, sex and age. http://www.cbs.gov.il/reader/shnaton/
templ shnaton.html?num tab=st02 18&CYear=2006,
Accessed on October 29, 2009.
[21] Lix LM, Hobson DE, Azimaee M, Leslie WD, Burchill C, &
Hobson S (2010) Socioeconomic variations in the prevalence
and incidence of Parkinson’s disease: a population-based
analysis. J Epidemiol Community Health,64, 335-340.
O. Chillag-Talmor et al. / Drug-Tracing PD Occurrence in Israel 47
[22] Israel Central Bureau of Statistics (2010) Statistical
abstract of Israel, 2010: Table 3.24 - Life expectancy, by
sex, religion and population group. http://www.cbs.gov.il/
shnaton61/download/st03 24.xls, Accessed on May 8, 2011.
[23] von Campenhausen S, Bornschein B, Wick R, Botzel K, Sam-
paio C, Poewe W, Oertel W, Siebert U, Berger K, & Dodel
R (2005) Prevalence and incidence of Parkinson’s disease in
Europe. Eur Neuropsychopharmacol,15, 473-490.
[24] Wermuth L, Bech S, Petersen MS, Joensen P, Weihe P, &
Grandjean P (2008) Prevalence and incidence of Parkinson’s
disease in The Faroe Islands. Acta Neurol Scand,118, 126-
131.
[25] Orr-Urtreger A, Shifrin C, Rozovski U, Rosner S, Bercovich
D, Gurevich T, Yagev-More H, Bar-Shira A, & Giladi N
(2007) The LRRK2 G2019S mutation in Ashkenazi Jews with
Parkinson disease: is there a gender effect? Neurology,,69,
1595-1602.
[26] Sidransky E, Nalls MA, Aasly JO, Aharon-Peretz J, Annesi G,
Barbosa E.R, Bar-Shira A, Berg D, Bras J, Brice A, Chen CM,
Clark LN, Condroyer C, De Marco EV, Durr A, Eblan MJ,
Fahn S, Farrer MJ, Fung HC, Gan-Or Z, Gasser T, Gershoni-
Baruch R, Giladi N, Griffith A, Gurevich T, Januario C, Kropp
P, Lang A.E, Lee-Chen GJ, Lesage S, Marder K, Mata I.F,
Mirelman A, Mitsui J, Mizuta I, Nicoletti G, Oliveira C,
Ottman R, Orr-Urtreger A, Pereira LV, Quattrone A, Rogaeva
E, Rolfs A, Rosenbaum H, Rozenberg R, Samii A, Samaddar
T,Schulte C, Sharma M, Singleton A, Spitz M, Tan EK, Tayebi
N, Toda T, Troiano AR, Tsuji S, Wittstock M, Wolfsberg TG,
Wu YR, Zabetian CP, Zhao Y, & Ziegler SG (2009) Multicen-
ter analysis of glucocerebrosidase mutations in Parkinson’s
disease. N Engl J Med,361, 1651-1661.
[27] Glik A, Masarwa M, Abuful A, Deeb A, Strugatsky R, Far-
rer LA, Friedland RP, & Inzelberg R (2009) Essential tremor
might be less frequent than Parkinson’s disease in North Israel
Arab villages. Mov Disord,24, 119-122.
[28] Masalha R, Kordysh E, Alpert G, Hallak M, Morad M, Maha-
jnah M, Farkas P, & Herishanu Y (2010) The prevalence of
Parkinson’s disease in an Arab population, Wadi Ara, Israel.
Isr Med Assoc J,12, 32-35.
[29] Anca M, Paleacu D, Shabtai H, & Giladi N (2002) Cross-
sectional study of the prevalence of Parkinson’s disease in the
Kibbutz movement in Israel. Neuroepidemiology,21, 50-55.
[30] Elbaz A, Bower JH, Peterson BJ, Maraganore DM, McDon-
nell SK, Ahlskog JE, Schaid DJ, & Rocca WA (2003) Survival
study of Parkinson disease in Olmsted County, Minnesota.
Arch Neurol,60, 91-96.
[31] Alves G, Muller B, Herlofson K, HogenEsch I, Telstad W,
Aarsland D, Tysnes OB, & Larsen JP (2009) Incidence of
Parkinson’s disease in Norway: the Norwegian ParkWest
study. J Neurol Neurosurg Psychiatry,80, 851-857.
[32] Grosset D, Taurah L, Burn DJ, MacMahon D, Forbes A,
Turner K, Bowron A, Walker R, Findley L, Foster O, Patel K,
Clough C, Castleton B, Smith S, Carey G, Murphy T, Hill J,
Brechany U, McGee P, Reading S, Brand G, Kelly L, Breen
K, Ford S, Baker M, Williams A, Hearne J, Qizilbash, & N,
Chaudhuri KR (2007) A multicentre longitudinal observa-
tional study of changes in self reported health status in people
with Parkinson’s disease left untreated at diagnosis. J Neurol
Neurosurg Psychiatry,78, 465-469.
Available via license: CC BY-NC 4.0
Content may be subject to copyright.