pharmacoepidemiology and drug safety 2014; 23: 128–139

Published online 30 August 2013 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3488

ORIGINAL REPORT

Selection bias in pharmacy-based patient surveys

P. Frisk1,2, S. Kälvemark-Sporrong1 and B. Wettermark3
1
Department of Pharmacy, Uppsala University, Sweden
2
National Corporation of Swedish Pharmacies (Apoteket AB), Stockholm, Sweden
3
Centre of Pharmacoepidemiology, Karolinska Institutet, Stockholm, Sweden

ABSTRACT
Purpose To evaluate if there is a selection bias in drug utilization surveys on prescription drugs regularly conducted in Swedish
pharmacies, to describe the direction of this potential bias and discuss the implications for the results.
Methods Age and gender distributions within patient survey samples from drug utilization surveys conducted during 2006–2010 are
compared to the age and gender distribution of all Swedish patients, receiving the same drug or drugs, as given by the Swedish Prescribed
Drug Register. The differences between the proportions of patients within the age and gender segments of each pair of survey/register data
were calculated.
Results In 25 (81%) out of 31 included surveys, patients aged 75 years or older are significantly underrepresented, as they are less likely to
visit the pharmacy to collect their prescription drugs themselves and thus disqualify for the interviews. Data on women show similar results
as overall survey data, whereas the underrepresentation of the oldest age group among men appears in a lower proportion of the surveys,
67%. The general consequence is a selection towards a healthier survey sample, but the consequences in the individual surveys vary,
depending on what drug is being studied.
Conclusion Pharmacy-based patient surveys provide a convenient data collection method for patient self-reported data, but patients aged
75 years or older are consistently underrepresented. In surveys where this may influence the main research question, data should also be
collected with other methods reaching the oldest patients. Copyright © 2013 John Wiley & Sons, Ltd.

key words—patient survey; pharmacy; selection bias; drug utilization; self-report; pharmacoepidemiology

Received 05 November 2012; Revised 29 April 2013; Accepted 27 June 2013

INTRODUCTION newly introduced and existing therapies, post-marketing

observational studies with data from different sources
Ageing populations, increased disease prevalence, are needed.3–6
demands from health-care consumers and the introduc- Large claims databases and prescription registries
tion of new drugs have contributed to increasing expen- are increasingly used in pharmacoepidemiological
diture within the health sector over the last decades.1,2 studies.6 They are considered to provide data
This has resulted in a growing interest in drug utilization representative of routine clinical care, at relatively
studies and outcomes research. The scientific value of low cost. Record linkage to other health data registers
well-designed randomized controlled trials is indisput- and electronical medical records may be conducted to
able, but it is also accepted that their findings not further study the association between drug exposure
necessarily reflect the effectiveness and safety demon- and clinical outcomes.6,7 Nevertheless, these prescrip-
strated once a drug has passed regulatory approval and tion databases have some important limitations when
reached larger, unrestricted patient populations in assessing drug exposure. They normally lack informa-
routine care.3–6 To study long-term risks and benefits, tion on OTC drugs, and the sensitivity may be further
reduced due to drug dispensations abroad, drugs
rare adverse events, treatment results in vulnerable provided through Internet sites or drugs shared between
patient groups and to make direct comparisons between relatives. The specificity may be reduced due to poor
compliance, i.e. the registers give valid information on
*Correspondence to: P. Frisk, Department of Pharmacy, Uppsala University, drug exposure only when dispensed drugs actually are
P.O. Box 580, SE-751 23 Uppsala, Sweden. E-mail: [email protected] taken.7,8 Methodological challenges in using register

Copyright © 2013 John Wiley & Sons, Ltd.

 selection bias in pharmacy-based surveys 129
data also include difficulties defining drug exposure3,9 pharmacies. The access to population-based dispensing
and estimating prescribed dose from complex dosing data with complete coverage for the entire Swedish
regimen data.9 population makes it possible to measure how well the
Prescriber data from medical records normally sample populations selected for pharmacy surveys
contain the clinical information missing in the pharmacy represent all patients having the same drug or drugs dis-
dispensing databases. However, most health systems pensed during the same time period.
face the challenge of maintaining uniform and complete The purpose of this study was to evaluate if there is a
records with a quality that allows extraction of reliable selection bias in patient surveys on prescription drugs
data for epidemiological studies. 10,11 Furthermore, in regularly conducted in Swedish pharmacies as part of
some settings, medical records are still not computer- the drug dispensing process, to establish the direction
ized, making data access and extraction very time of this potential bias and to discuss the potential implica-
consuming.11 Several studies have compared the informa- tions for the interpretation of the survey results.
tion on prescribed drugs found in medical records with the
corresponding information reported by patients and often
found the medical records to be discrepant.12–14 METHODS
The rationale behind patient surveys is the assump-
tion that the patient ought to be the primary source of The pharmacy-based surveys included in this study
information in matters regarding his or her health, were conducted at 25–71 pharmacies distributed all
especially when it comes to their subjective view and over the country, including hospital pharmacies with
experiences as users of health care services and drugs. dispensing services in ambulatory care, pharmacies
However, the collection of self-reported data is often located in primary health care centres and pharmacies
time consuming and logistically complicated. Since in urban and rural areas. Based on information
patient surveys are based on samples rather than whole provided on the prescription, all patients presenting a
populations and rely on patient recall, they are subject prescription on a drug intended for themselves,
to both selection and recall bias, a threat to their meeting the entry criteria and consenting to participate
validity.6 Nevertheless, patient self-reported data are were automatically enrolled and asked to participate in
frequently used in drug utilization research, either as a short, structured questionnaire at the end of the drug-
a supplement to data from other sources or when drug dispensing process, with the pharmacist as the
exposure data are validated, often with self-reported interviewer. In surveys including minors, the caregiver
data as the “gold standard”.8,9,12–17 was offered to participate. General entry criteria were
One source for patient self-reported data is to collect age and gender of the patient, prescribed drug and
data from patients when they collect their prescriptions dispensing pharmacy. These were checked automatically
at pharmacies. The National Corporation of Swedish when the prescription was registered in the dispensing
Pharmacies (Apoteket AB) developed a method where system. In addition, criteria such as indication for use,
computer-aided surveys are used to collect drug- co-payment willingness and reimbursement status were
related information from the patient at the pharmacy sometimes used to select patients. This information was
counter. The method has mainly been used to collect captured in the questionnaires. The purposes of the
data which is more difficult or expensive to collect surveys differed but can, so far, be categorized into one
from other data sources. Such data include the indica- of the following: (i) to study prescribers’ adherence to
tion for use,18 treatment preferences among patients, prescribing guidelines or reimbursement restrictions for
attitudes towards co-payment and experiences of prescription drugs, (ii) to study patient preferences and
generic substitution.19 So far, no unique identifiers of other drug related issues relevant from a patient perspec-
any patients have been stored enabling record linkage tive and (iii) to study prescribing issues relevant from a
to other data to validate the method. Since both patient marketing and/or business development perspective.
selection and survey interviews take place in the phar-
macy with the pharmacists as interviewers, there is a
potential risk of selection bias and recall bias.20 Data Collection
Pharmacies have occasionally been used as a In this study, all drugs were classified according to the
recruitment or interview setting in drug utilization Anatomical Therapeutic Chemical (ATC) classification
surveys, but reports from systematic and nationwide system.25 As part of the review of all pharmacy surveys
surveying as described above are scarce.21–24 Conse- of this kind conducted in Sweden between 2006 and
quently, it is important to validate the method, i.e. 2012, the selection of surveys for this study was based
recruiting and interviewing survey respondents in on the following criteria:

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
130 p. frisk et al.
(a) The surveys should be completed, i.e. ongoing Swedish National Board of Health and Welfare which
surveys were excluded. is a government agency that, in accordance with
(b) For calculation of the relative sample size of Swedish law, may use population-based registers to
the survey sample, dispensing data from 2010 of follow and analyse health and social conditions among
the Swedish population should be available in the the general population. All data were made anonymous
Prescribed Drug Register. To provide survey and aggregated for the research team with no possibility
samples likely to be representative of all users of to identify any individual patient.
the specific drug in question, the number of
responders in each survey should constitute at RESULTS
least 0.5% of the period prevalence rate of 2010
for the corresponding drug on the ATC fifth level. Out of 49 pharmacy surveys conducted all over Sweden
(c) To avoid duplicate registration of patients in each between 2006 and 2012, 31 surveys were included.
study, the surveys should only include one pharma- (See Figure 1 and Table 1). The survey samples as
cological group, ATC fourth level. well as their corresponding register populations
(d) Population dispensing data from 2011 in the included all users, i.e. both prevalent and incident users.
Prescribed Drug Register should be available to The surveys included belong to the therapeutic areas
assess the potential implications of the study results. diabetes (survey 1–6), asthma/chronic obstructive
For each included survey, age and gender distribu- pulmonary disease (COPD) (surveys 7–9), oncology/
tions of all patients collecting their prescription drugs immunology (surveys 10–18), neurology/psychiatry
themselves and therefore being offered to participate (surveys 19–25) and a mixed group with different
were calculated. The age categories used were 0–44 therapeutic areas, no 26–31. Children/adolescents were
years, 45–64 years, 65–74 years and >75 years, one only included in surveys 2, 26 and 31, where their
of the intervals for population statistics recommended caregivers were interviewed.
by the WHO also applied in the National Swedish The most consistent difference between survey and
Drug statistics.26 register samples was an underrepresentation of all
To measure the overall representativity of the patients aged 75 years or older, seen in 25 surveys
sample populations selected for surveys, age and
gender distributions of all Swedish patients having
45 surveys 2 surveys
the same drug or drugs dispensed during the same or closed/finalised discontinued
approximately the same time period were extracted surveys 2006-2011 prematurely
from the Swedish Prescribed Drug Register.7

43 surveys 4 surveys including

Statistical Analysis completed > 1 ATC fourth level subgroup
The differences between the proportions in each age and
gender segment of each pair of survey/register data 1 survey lacking
were calculated. Proportions with non-overlapping 39 surveys register data for
confidence intervals on the 95% level of confidence for potential calculation of adequate
were considered statistically significant.27 inclusion sample size

38 surveys 6 surveys with sample

Ethics size < 0,5% of period
for sample
prevalence rate in 2010a
Due to the nature of the individual surveys they were not size analysis
considered to require ethical approval. Nevertheless, 1 survey lacking
each survey was conducted with attention to patient register data for
integrity and data safety issues through all stages of data 32 surveys calculation of current
collection, handling, analysis and presentation. Patients age distribution
participating in the surveys gave their informed consent
prior to the interview, the consent was documented in 31 surveys
included
the questionnaire form and all data were made anony-
mous with no possibility to identify individual patients. a
Figure 1. Inclusion of performed surveys in study. Prevalence data from
The register data analyses were undertaken by the the Swedish Prescribed Drug Register

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
 selection bias in pharmacy-based surveys 131
Table 1. Included surveys
Survey no. ATC code No. of eligible patients* Response rate† Survey duration‡ Age of survey sample (years) Purpose

Insulins and analogues for injection, long acting

∥
1 A10AE 272 93% 10 Jan–28 Feb 2006 18 <
∥
2 A10AE 275 91% 10 Jan–13 May 2006 All ages
††
3 A10AE 663 84% 16 Apr–16 Jun 2007 18 <
††
4 A10AE 400 76% 15 Nov–31 Dec 2007 18 <
††
5 A10AE 427 70% 26 Nov 2008–29 Dec 2008 18 <
††
6 A10AE 375 81% 22 Nov–16 Dec 2010 18 <
Adrenergics and other drugs for obstructive airway diseases
††
7 R03AK 948 89% 23 Nov 2006–28 May 2007 18 <
††
8 R03AK 1192 77% 13 Nov 2008–17 Apr 2009 18 <
††
9 R03AK 1321 76% 28 Dec 2009–26 May 2010 18 <
Antineoplastic and immunomodulating agents
††
10 L01XE 69 86% 31 Aug–1 Dec 2007 18 <
††
11 L01XE 80 83% 1 Oct 2008–11 Feb 2009 18 <
††
12 § L02AE 234 73% 27 Aug–31 Dec 2007 18 <
††
13 ¶ L02BA 142 88% 19 Apr–30 Jun 2007 52 <
††
14 ¶ L02BG 142 91% 19 Apr–30 Jun 2007 52 <
††
15 ¶
L02BG 139 83% 31 Aug–1 Dec 2007 18 <
††
16 ¶ L02BG 152 89% 31 Oct 2008–9 Feb 2009 18 <
††
17 L04AB 287 73% 5 Mar–9 Sep 2007 18 <
††
18 L04AB 280 71% 25 Feb–17 Aug 2008 18 <
Antiepileptics, anti-parkinson drugs and psychoanaleptics
††
19 N03AG 383 74% 18 Oct 2007–31 Jan 2008 18 <
††
20 N03AX 291 69% 24 Apr – 3 Dec 2008 18 <
21 N03AX 515 80% 8 Dec–31 Dec 2009 18 < **
22 N03AX 2571 56% 1 Dec 2010–3 Apr 2011 18 < **
††
23 N04BC 594 87% 11 Dec 2006–7 Mar 2007 18 <
††
24 N06AX 534 75% 1 Aug–3 Oct 2008 18 <
††
25 N06AX 621 64% 14 Apr–19 Jun 2009 18 <

Other drugs
††
26 A01AA 567 89% 12 May–26 Jul 2006 All ages
∥
27 A02BC 598 84% 4–23 Dec 2006 18 <
††
28 B01AC 577 83% 24 May–22 Jul 2007 18 <
29 C10AA 594 91% 19 Jan–25 Feb 2007 18 < **
∥
30 D11AH 39 79% 8 Nov 2006–30 May 2007 All ages
††
31 M01AX 590 86% 8 Dec 2009–13 Jan 2010 18 <

*No of patients presenting a prescription on a drug intended for themselves.

†
No of respondents/No of eligible patients.
‡
Register data were collected from the months corresponding to when each survey was conducted.
§
Only men included.
¶
Only women included.
∥
To study adherence to reimbursement restrictions.
**To study adherence to prescribing guidelines.
††
To study different aspects related to prescribing and/or actual drug use, e.g. daily dose taken by the patient, recommended dosing scheme and indication for use.

(81%) (Table 2). This difference was found across all underrepresentation of patients in the oldest age
therapeutic areas investigated, but it was most group in 24 (80%) of the surveys (Table 3)
consistent within the therapeutic areas of asthma/COPD (Figure 2b) and a significant overrepresentation of
(surveys 7–9) and neurology/psychiatry (surveys 19–25), patients in at least one of the younger age groups
where it appeared in all surveys. The underrepresenta- in 19 (63%) of these surveys.
tion of the oldest age group was accompanied by a Data on men were less consistent, with the oldest
significant overrepresentation of patients in at least age group being significantly underrepresented in
one of the younger age groups in 19 (61%) of the 18 (67%) of the surveys (Table 4) (Figure 2c).
surveys (Table 2) (Figure 2a). The concomitant overrepresentation of at least one
On the age and gender level, data on women of the younger age groups was seen in 11(41%)
showed a similar pattern, with a significant of the surveys.

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
132 p. frisk et al.
Table 2. Age distributions within each survey and register sample, proportions presented within brackets
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop ± C.I.95%

1 272 85 (31.3%) 120 (44.1%) 40 (14.7%) 27 (9.9%)

19 388 5551 (28.6%) 7978 (41.1%) 3328 (17.2%) 2531 (13.1%)
2.6% ±5,5% 3.0% ± 5.9% 2.5% ± 4.2% 3.1% ± 3.6%
2 275 162 (58.9%) 80 (29.1%) 24 (8.7%) 9 (3.3%)
5693 3049 (53.6%) 1817 (31.9%) 555 (9.7%) 272 (4.8%)
5.4% ± 6.0% 2.8% ± 5.5% 1.0% ± 3.4% 1.5% ± 2.2%
3 663 167 (25.2%) 302 (45.6%)* 143 (21.6%)* 51 (7.7%)*
29 755 8477 (28.5%) 11 975 (40.2%) 5296 (17.8%) 4007 (13.5%)
3.3% ± 3.3% 5.3% ± 3.8% 3.8% ± 3.2% 5.8% ± 2.1%
4 491 135 (27.5%) 229 (46.6%)* 94 (19.1%) 33 (6.7%)*
23 091 6222 (26.9%) 9345 (40.5%) 4350 (18.8%) 3174 (13.7%)
0.5% ± 4.0% 6.2% ± 4.5% 0.3% ± 3.5% 7.0% ± 2.3%
5 427 125 (29.3%) 173 (40.5%) 96 (22.5%) 33 (7.7%)*
25 808 6622 (25.7%) 10 155 (39.3%) 5231 (20.3%) 3800 (14.7%)
3.6% ± 4.3% 1.2% ± 4.7% 2.2% ± 4.0% 7.0% ± 2.6%
6 375 120 (32.0%)* 153 (40.8%) 71 (18.9%) 31 (8.3%)*
27 673 6654 (24.0%) 10 392 (37.6%) 6183 (22.3%) 4444 (16.1%)
8.0% ± 4.7% 3.2% ± 5.0% 3.4% ± 4.0 7.8% ± 2.8%
7 948 240 (25.3%) 338 (35.7%) 211 (22.3%) 159 (16.8%)*
45 059 11 882 (26.4%) 14 991 (33.3%) 8943 (19.8%) 9243 (20.5%)
1.1% ± 2.8% 2.4% ± 3.1% 2.4% ± 2.7% 3.7% ± 2.4%
8 1192 178 (14.9%)* 476 (39.9%)* 317 (26.6%)* 221 (18.5%)*
44 756 8467(18.9%) 14 936 (33.4%) 10 477 (23.4%) 10 876 (24.3%)
4.0% ± 2.1% 6.6% ± 2.8% 3.2% ± 2.5% 5.8% ± 2.2%
9 1321 305 (23.1%)* 464 (35.1%)* 331 (25.1%) 221 (16.7%)*
42 941 8250 (19.2%) 13 815 (32.2%) 10 290 (24.0%) 10 586 (24.7%)
3.9% ± 2.3% 3.0% ± 2.6 1.1% ± 2.4% 7.9% ± 2.1%
10 69 10 (14.5%) 27 (39.1%) 21 (30.4%) 11 (15.9%)
685 118 (17.2%) 276 (40.3%) 175 (25.5%) 116 (16.9%)
2.7% ± 8.8% 1.2% ± 12.1% 4.9% ± 11.3% 1.0% ± 9.1%
11 80 17 (21.3%) 31 (38.8%) 19 (23.8%) 13 (16.3%)
750 123 (16.4%) 271 (36.1%) 198 (26.4%) 158 (21.1%)
4.9% ± 9.4% 2.6% ± 11.2% 2.7% ± 9.8% 4.8% ± 8.6%
12 229 0 (0%) 20 (8.7%) 57 (24.9%) 152 (66.4%)*
11 792 6 (0.1%) 719 (6.1%) 2422 (20.5%) 8635 (73.2%)
2.6% ± 3.7% 4.4% ± 5.7% 6.9% ± 6.2%
13 142 NA 59 (41.5%) 46 (32.4%) 37 (26.1%)*
8318 3062(36.8%) 2322 (27.9%) 2934 (35.3%)
4.7% ± 8.2% 4.5% ± 7.8% 9.2% ± 7.3%
14 142 NA 68 (47.9%)* 48 (33.8%) 26 (18.3%)*
6475 2546 (39.3%) 1906 (29.4%) 2023 (31.2%)
8.6% ± 8.3% 4.4% ± 7.9% 12.9% ± 6.5%
15 139 9 (6.5%) 63 (45.3%)* 39 (28.1%) 28 (20.1%)*
2080 69 (3.3%) 761 (36.6%) 490 (23.6%) 760 (36.5%)
3.2% ± 4.2% 8.7% ± 8.5% 4.5% ± 7.7% 16.4% ± 7.0%
16 152 6 (3.9%) 85 (55.9%)* 39 (25.7%) 22 (14.5%)*
2564 57 (2.2%) 874 (34.1%) 670 (26.1%) 963 (37.6%)
1.7% ± 3.1% 21.8% ± 8.1% 0.5% ± 7.2% 23.1% ± 5.9%
17 287 97 (33.8%) 140 (48.8%) 39 (13.6%) 11 (3.8%)
2528 758 (30.0%) 1245 (49.2%) 384 (15.2%) 141 (5.6%)
3.8% ± 5.8% 0.5% ± 6.1% 1.6% ± 4.2% 1.7% ± 2.4%
18 280 114 (40.7%)* 113 (40.4%)* 48 (17.1%) 5 (1.8%)*
3375 1163 (34.5%) 1571 (46.5%) 490 (14.5%) 151 (4.5%)
6.3% ± 6.0% 6.2% ± 6.0% 2.6% ± 4.6% 2.7% ± 1.7%
19 383 161 (42.0%) 153 (39.9%) 50 (13.1%) 19 (5.0%)*
19 075 7641 (40.1%) 6706 (35.2%) 2262 (11.9%) 2466 (12.9%)
2.0% ± 5.0% 4.8% ± 5.0% 1.2% ± 3.4% 8.0% ± 2.2%
20 291 125 (43.0%) 128 (44.0%)* 35 (12.0%) 3 (1.0%)*
5577 2315 (41.5%) 2003 (35.9%) 735 (13.2%) 524 (9.4%)
1.4% ± 5.8% 8.1% ± 5.8% 1.2% ± 3.8% 8.4% ± 1.4%

(Continues)

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
 selection bias in pharmacy-based surveys 133
Table 2. Continued
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop ± C.I.95%

21 515 214 (41.6%)* 207 (40.2%) 57 (11.1%) 37 (7.2%)*

19 199 5977 (31.1%) 6897 (35.9%) 2423 (12.6%) 3902 (20.3%)
10.4% ± 4.3% 4.3% ± 4.3% 1.6% ± 2.8% 13.1% ± 2.3%
22 2571 1053 (41.0%)* 1020 (39.7%)* 289 (11.2%)* 209 (8.1%)*
38 663 11 103 (28.7%) 14 478 (37.4%) 5683 (14.7%) 7399 (19.1%)
12.2% ± 2.0% 2.2% ± 2.0% 3.5% ± 1.3% 11.0% ± 1.1%
23 594 51 (8.6%)* 241 (40.6%)* 173 (29.1%) 129 (21.7%)*
14 785 868 (5.9%) 4821 (32.6%) 3832 (25.9%) 5264 (35.6%)
2.7% ± 2.3% 8.0% ± 4.0% 3.2% ± 3.7% 13.9% ± 3.4%
24 534 257 (48.1%)* 225 (42.1%) 38 (7.1%)* 14 (2.6%)*
19 160 6967 (36.4%) 8644 (45.1%) 1939 (10.1%) 1610 (8.4%)
11.8% ± 4.3% 3.0% ± 4.3% 3.0% ± 2.2% 5.8% ± 1.4%
25 621 278 (44.8%)* 289 (46.5%) 39 (6.3%)* 15 (2.4%)*
20 265 7096 (35.0%) 9129 (45.0%) 2299 (11.3%) 1741 (8.6%)
9.8% ± 4.0% 1.5% ± 4.0% 5.1% ± 2.0% 6.2% ± 1.3%
26 567 50 (8.8%)* 209 (36.9%)* 150 (26.5%)* 158 (27.9%)*
21 940 2478 (11.3%) 6105 (27.8%) 4847 (22.1%) 8510 (38.8%)
2.5% ± 2.4% 9.0% ± 4.0% 4.4% ± 3.7% 10.9% ± 3.7%
27 598 93 (15.6%) 299 (50.0%)* 122 (20.4%) 84 (14.0%)*
23 754 3518 (14.8%) 10 048 (42.3%) 4663 (19.6%) 5525 (23.3%)
0.7% ± 2.9% 7.7% ± 4.1% 0.8% ± 3.3% 9.2% ± 2.8%
28 577 6 (1.0%) 201 (34.8%)* 189 (32.8%)* 181 (31.4%)*
29 304 412 (1.4%) 7984 (27.2%) 8262 (28.2%) 12 646 (43.2%)
0.4% ± 0.8% 7.6% ± 3.9% 4.6% ± 3.9% 11.8% ± 3.8%
29 594 18 (3.0%) 259 (43.6%) 204 (34.3%) 113 (19.0%)*
51 048 1303 (2.6%) 20 364 (39.9%) 16 837 (33.0%) 12 544 (24.6%)
0.5% ± 1.4% 3.7% ± 4.0% 1.4% ± 3.8% 5.5% ± 3.2
30 39 30 (76.9%)* 7 (17.9%) 0 (0%) 2 (5.1%)
1033 640 (62.0%) 273 (26.4%) 81 (7.8%) 39 (3.8%)
15.0% ± 13.6% 8.5% ± 12.3% 1.4% ± 7.0%
31 590 7 (1.2%) 190 (32.2%) 238 (40.3%)* 155 (26.3%)*
35 555 674 (1.9%) 10 692 (30.1%) 12 251 (34.5%) 11 938 (33.6%)
0.7% ± 0.9% 2.1% ± 3.8% 5.9% ± 4.0% 7.3% ± 3.6%

*Proportions with non-overlapping confidence intervals on the 95% level of confidence.

DISCUSSION Combining self-reported patient data with data

from other sources may be a suitable method of
This study aimed to investigate whether self-reported studying different aspects of drug prescribing and
patient data collected through interviews performed utilization, provided the self-reported data are
in Swedish pharmacies are subject to a selection valid and reliable. The agreement between
bias, and if so, to discuss the potential implications interview data regarding drug use and pharmacy
for the survey results. The overall analysis indicates record data has been evaluated in previous studies,
that patients aged 75 years or older are underrepre- showing a relatively high level of agreement for
sented in more than 80% of the surveys. With chronic or serious conditions,8,15,28,29 but also
respect to this deviation in age distribution, data on indicating differences between therapeutic areas or
women show the same results as overall survey data, drugs/drug classes,8,12,15,28,30 and validation methods
whereas data on men are less consistent, with used.8,9,15
the underrepresentation of the oldest age group Most drugs show an increased prevalence with
appearing in a somewhat lower proportion of the increasing age.31–33 reflecting the underlying
surveys. This selection bias is most likely introduced association between age and disease. The systematic
by the fact that only patients who appear at the exclusion of the oldest patients seen in this study
pharmacy themselves are interviewed. Patients who, may thus contribute to a healthy selection effect,
for any reason, are unable to go to the pharmacy underestimating findings associated with multiple
themselves and therefore have a representative to and/or more severe diseases. When studying cardio-
collect their prescription drugs are excluded. vascular drugs, this may e.g. have resulted in an

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
134 p. frisk et al.

30 analysing individual surveys and access to

complementary data sources in surveys where the
25
selection bias is expected to have an impact on the
No. of surveys

20 main research question. Conducting home interviews

survey sample
overrepresented
15
survey sample
or interviews in other settings, such as primary care
10
underrepresented centres and hospitals, is an alternative method for
collecting self-reported drug utilization data that
5
potentially provides representative data from all
0
0-44 45-64 65-74 75+
relevant age groups.8,12,35
Age group The design of this study has several limitations. It is
important to distinguish between patients eligible for
a inclusion, i.e. all patients coming to the pharmacy to
30 redeem their own prescription, and respondents,
25
patients who actually agree to participate and complete
No. of surveys

20
the survey interview. In this study, we focus on
survey sample

15
overrepresented patients eligible for inclusion and do not analyse
survey sample
underrepresented
any other factors than age and gender that could
10
influence whether a patient purchased his or her own
5
medication, e.g. other demographic or behavioural
0
0-44 45-64 65-74 75+ characteristics. Nor do we study the reasons why
Age group eligible patients refused to participate in the
b
individual surveys. Hence, the discussion on the
described selection bias and its implications
20 remains simplified since it assumes that the age and
18 gender distributions of all eligible patients are the same
16
No. of surveys

14 as for all respondents. However, a high response rate

12 survey sample
overrepresented
among those eligible for participation in the included
10
survey sample surveys indicates that the differences in age and
8
6
underrepresented
gender distribution between eligible patients and
4 respondents most likely are minor (Table 1).
2 The results from the analysis on the age and
0
0-44 45-64 65-74 75+ gender level reveal a less pronounced bias among
Age group men. Recent research has shown that, in patients
aged 75 years or older, women show higher
c
disability rates than men in the domains of mobility
Figure 2. a. Number of surveys with over- and underrepresented age and activities of daily living, something that is likely
groups (n = 31). b. Number of surveys with over- and underrepresented to have an impact on their ability to visit the
age groups, women (n = 30). c. Number of surveys with over- and under-
represented age groups, men (n = 27) pharmacy and purchase their prescription drugs
themselves.36 Whether our results reflect this differ-
ence or are due to the fact that stratification on both
underestimated prevalence of heart failure.18 Never- age and gender levels creates too small samples to
theless, several drugs or drug groups show a different detect existing differences among both men and
pattern with varying prevalence rates in different age women, respectively, remains unclear without access
groups, depending on their labelled indications, to larger survey samples.
prescribing guidelines and reimbursement restrictions. The general results found in the study were
Accordingly, the consequences of a consistent most consistent within the therapeutic areas of
exclusion of some of the oldest patients may vary asthma/COPD and neurology/psychiatry, since
considerably. Based on dispensing data from 2011, the detected selection bias appeared in all surveys
the proportion of patients aged 75 years or older within these areas. These two areas contain some
among prevalent users receiving the drugs included of the largest surveys in the study; therefore, it
in this study currently ranges from 4.0 to 73.1%.34 De- is not clear whether it is the drug group per se
spite the general result presented here, this illustrates or the large sample sizes that produce the consis-
the importance of an unbiased approach when tent results.

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
 selection bias in pharmacy-based surveys 135
Table 3. Age distributions (women) within each survey and register sample, proportions presented within brackets
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop ± C.I.95%

1 113 39 (34.5%) 44 (38.9%) 17 (15.0%) 13 (11.5%)

8552 2365 (27.7%) 3257 (38.1%) 1474 (17.2%) 1456 (17.0%)
6.9% ± 8.8% 0.9% ± 9.1% 2.2% ± 6.6% 5.5% ± 5.9%
2 122 67 (54.9%) 40 (32.8%) 9 (7.4%) 6 (4.9%)
2699 1469 (54.4%) 818 (30.3%) 256 (9.5%) 156 (5.8%)
0.5% ± 9.0% 2.5% ± 8.5% 2.1% ± 4.8% 0.9% ± 3.9%
3 283 71 (25.1%) 128 (45.2%)* 56 (19.8%) 28 (9.9%)*
12 891 3555 (27.6%) 4801 (37.2%) 2292 (17.8%) 2243 (17.4%)
2.5% ± 5.1 8.0% ± 5.9% 2.0% ± 4.7% 7.5% ± 3.5%
4 209 61 (29.2%) 87 (41.6%) 44 (21.1%) 17 (8.1%)*
9860 2556 (25.9%) 3626 (36.8%) 1877 (19.0%) 1801 (18.3%)
3.3% ± 6.2% 4.9% ± 6.8% 2.0% ± 5.6% 10.1% ± 3.8%
5 183 61 (33.3%)* 64 (35.0%) 38 (20.8%) 20 (10.9%)*
10 847 2635 (24.3%) 3941 (36.3%) 2170 (20.0%) 2101 (19.4%)
9.0% ± 6.9% 1.4% ± 7.0% 0.8% ± 5.9% 8.4% ± 4.6%
6 149 54 (36.2%)* 54 (36.2%) 26 (17.4%) 15 (10.1%)*
11 520 2661 (23.1%) 3927 (34.1%) 2552 (22.2%) 2380 (20.7%)
13.1% ± 7.8% 2.2% ± 7.8 4.7% ± 6.1% 10.6% ± 4.9%
7 570 135 (23.7%) 216 (37.9%) 133 (23.3%) 86 (15.1%)*
25 697 6200 (24.1%) 9129 (35.5%) 5248 (20.4%) 5120 (19.9%)
0.4% ± 3.5% 2.4% ± 4.0% 2.9% ± 3.5% 4.8% ± 3.0%
8 723 106 (14.7%)* 313 (43.3%)* 184 (25.4%) 120 (16.6%)*
26 730 4647 (17.4%) 9063 (33.9%) 6418 (24.0%) 6602 (24.7%)
2.7% ± 2.6% 9.4% ± 3.7% 1.4% ± 3.2% 8.1% ± 2.8%
9 820 172 (21.0%)* 297 (36.2%)* 219 (26.7%) 132 (16.1%)*
25 718 4516 (17.6%) 8357 (32.5%) 6328(24.6%) 6517 (25.3%)
3.4% ± 2.8% 3.7% ± 3.3% 2.1% ± 3.1% 9.2% ± 2.6%
10 28 4 (14.3%) 11 (39.3%) 9 (32.1%) 4 (14.3%)
303 48 (15.8%) 122 (40.3%) 80 (26.4%) 53 (17.5%)
1.6% ± 13.6% 1.0% ± 18.9% 5.7% ± 18.0% 3.2% ± 13.7%
11 34 9 (26.5%) 12 (35.3%) 9 (26.5%) 4 (11.8%)
338 53 (15.7%) 121 (35.8%) 90 (26.6%) 74 (21.9%)
10.8% ± 15.3% 0.5% ± 16.9% 0.2% ± 15.6% 10.1% ± 11.7%
13 142 NA 59 (41.5%) 46 (32.4%) 37 (26.1%)*
8318 3062(36.8%) 2322 (27.9%) 2934 (35.3%)
4.7% ± 8.2% 4.5% ± 7.8% 9.2% ± 7.3%
14 142 NA 68 (47.9%)* 48 (33.8%) 26 (18.3%)*
6475 2546 (39.3%) 1906 (29.4%) 2023 (31.2%)
8.6% ± 8.3% 4.4% ± 7.9% 12.9% ± 6.5%
15 139 9 (6.5%) 63 (45.3%)* 39 (28.1%) 28 (20.1%)*
2080 69 (3.3%) 761 (36.6%) 490 (23.6%) 760 (36.5%)
3.2% ± 4.2% 8.7% ± 8.5% 4.5% ± 7.7% 16.4% ± 7.0%
16 152 6 (3.9%) 85 (55.9%)* 39 (25.7%) 22 (14.5%)*
2564 57 (2.2%) 874 (34.1%) 670 (26.1%) 963 (37.6%)
1.7% ± 3.1% 21.8% ± 8.1% 0.5% ± 7.2% 23.1% ± 5.9%
17 193 56 (29.0%) 97 (50.3%) 30 (15.5%) 10 (5.2%)
1688 464 (27.5%) 841 (49.8%) 270 (16.0%) 113 (6.7%)
1.5% ± 6.8% 0.4% ± 7.4% 0.5% ± 5.4% 1.5% ± 3.3%
18 188 73 (38.8%)* 75 (39.9%)* 35 (18.6%) 5 (2.7%)*
2141 654 (30.5%) 1020 (47.6%) 346 (16.2%) 121 (5.7%)
8.3% ± 7.2% 7.7% ± 7.3% 2.5% ± 5.8% 3.0% ± 2.5%
19 196 87 (44.4%) 74 (37.8%) 27 (13.8%) 8 (4.1%)*
9364 3515 (37.5%) 3384 (36.1%) 1093 (11.7%) 1372 (14.7%)
6.9% ± 7.0% 1.6% ± 6.9% 2.1% ± 4.9% 10.6% ± 2.9%
20 150 68 (45.3%) 67 (44.7%)* 14 (9.3%) 1 (0.7%)*
2795 1177 (42.1%) 979 (35.0%) 369 (13.2%) 270 (9.7%)
3.2% ± 8.2% 9.6% ± 8.2% 3.9% ± 4.8% 9.0% ± 1.7%
21 327 130 (39.8%)* 133 (40.7%)* 38 (11.6%) 26 (8.0%)*
11 689 3366 (28.8%) 4106 (35.1%) 1435 (12.3%) 2782 (23.8%)
11.0% ± 5.4% 5.5% ± 5.4% 0.7% ± 3.5% 15.8% ± 3.0%

(Continues)

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
136 p. frisk et al.
Table 3. Continued
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop ± C.I.95%

22 1546 603 (39.0%)* 615 (39.8%)* 180 (11.6%)* 148 (9.6%)*

23 103 6116 (26.5%) 8504 (36.8%) 3364 (14.6%) 5119 (22.2%)
12.5% ± 2.5% 3.0% ± 2.5% 2.9% ± 1.7% 12.6% ± 1.6%
23 375 36 (9.6%)* 163 (43.5%)* 103 (27.5%) 73 (19.5%)*
9000 503 (5.6%) 2892 (32.1%) 2221 (24.7%) 3384 (37.6%)
4.0% ± 3.0% 11.3% ± 5.1% 2.8% ± 4.6% 18.1% ± 4.1%
24 391 195 (49.9%)* 160 (40.9%) 24 (6.1%)* 12 (3.1%)*
13 394 4868 (36.3%) 6061 (45.3%) 1258 (9.4%) 1207 (9.0%)
13.5% ± 5.0% 4.3% ± 5.0% 3.3% ± 2.5% 5.9% ± 1.8%
25 454 204 (44.9%)* 210 (46.3%) 27 (5.9%)* 13 (2.9%)*
14 080 4928 (35.0%) 6372 (45.3%) 1473 (10.5%) 1307 (9.3%)
9.9% ± 4.7% 1.0% ± 4.7% 4.5% ± 2.2% 6.4% ± 1.6%
26 423 39 (9.2%) 155 (36.6%)* 110 (26.0%) 119 (28.1%)*
15 768 1538 (9.8%) 4475 (28.4%) 3499 (22.2%) 6266 (39.7%)
0.5% ± 2.8% 8.3% ± 4.6% 3.8% ± 4.2% 11.6% ± 4.4%
27 359 53 (14.8%) 174 (48.5%)* 81 (22.6%) 51 (14.2%)*
13 411 1920 (14.3%) 5458 (40.7%) 2583 (19.3%) 3450 (25.7%)
0.4% ± 3.7% 7.8% ± 5.2% 3.3% ± 4.8% 11.5% ± 3.7%
28 210 2 (1.0%) 62 (29.5%)* 59 (28.1%) 87 (41.4%)*
12 043 122 (1.0%) 2405 (20.0%) 3098 (25.7%) 6418 (53.3%)
0.1% ± 1.3% 9.6% ± 6.2 2.4% ± 6.1% 11.9% ± 6.7%
29 265 6 (2.3%) 98 (37.0%) 99 (37.4%) 62 (23.4%)*
22 724 333 (1.5%) 7518 (33.1%) 7882 (34.7%) 6991 (30.8%)
0.8% ± 1.8% 3.9% ± 5.8% 2.7% ± 5.9% 7.4% ± 5.1%
30 29 21 (72.4%) 7 (24.1%) 0 (0%) 1 (3.4%)
634 375 (59.1%) 181 (28.5%) 56 (8.8%) 22 (3.5%)
13.3% ± 16.7% 4.4% ± 16.0% 0.0% ± 6.8%
31 418 3 (0.7%) 145 (34.7%)* 157 (37.6%) 113 (27.0%)*
24 211 364 (1.5%) 7283 (30.1%) 8222 (34.0%) 8342 (34.5%)
0.8% ± 0.8% 4.6% ± 4.6% 3.6% ± 4.7% 7.4% ± 4.3%

*Proportions with non-overlapping confidence intervals on the 95% level of confidence.

The results are presented with register data as a not possible. Furthermore, data of the individual
reference, representing the whole population of actual surveys were not stored with unique patient identifiers
users. The survey is the data source being validated and to enable linkage between the different surveys
hence presented as either deviating from the register and register data. However, this study represents a large
population distribution or not. Register data indeed number of surveys with a variety in therapeutic areas
represent data collected from the whole population. and patient populations. This allows the results to be
Nevertheless, in this study, each set of register data is generalized to pharmacy-based surveys in general, with
limited by a specific time window, which only approxi- the limitations mentioned above taken into account.
mately corresponds to the data collection period for the
survey and therefore introduces a possible error.
The exact number and location of the pharmacies CONCLUSION
involved differ somewhat between the surveys, since
all pharmacies not always participate. This may also Conducting patient surveys in pharmacies is a convenient
contribute to deviations in the age and gender interview method, but the setting per se contributes to the
distribution of the survey population, especially in exclusion of some of the oldest patients, regardless of
surveys including drugs with regional differences in what drug group is being studied. In pharmacy
prescribing patterns. The surveys were not originally surveys where the selection bias is expected to have
designed for inclusion in this study. Consequently, an impact on the main research question, data
some pharmacologic groups or patient groups are should be collected alongside with survey data from
underrepresented. Data on minors are only included home interviews or other methods reaching the
in three surveys; thus, an analysis of the representativ- oldest patients and thus contributing with the data
ity of data from minors in pharmacy-based surveys is not captured in pharmacies.

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
 selection bias in pharmacy-based surveys 137
Table 4. Age distributions (men) within each survey and register sample, proportions presented within brackets
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop± C.I.95%

1 159 46 (28.9%) 76 (47.8%) 23 (14.5%) 14 (8.8%)

10 836 3186 (29.4%) 4721 (43.6%) 1854 (17.1%) 1075 (9.9%)
0.5% ± 7.1% 4.2% ± 7.8% 2.6% ± 5.5% 1.1% ± 4.4%
2 153 95 (62.1%)* 40 (26.1%)* 15 (9.8%) 3 (2.0%)
2994 1580 (52.8%) 999 (33.4%) 299 (10.0%) 116 (3.9%)
9.3% ± 7.9% 7.2% ± 7.2% 0.2% ± 4.8% 1.9% ± 2.3%
3 380 96 (25.3%) 174 (45.8%) 87 (22.9%)* 23 (6.1%)*
16 864 4922 (29.2%) 7174 (42.5%) 3004 (17.8%) 1764 (10.5%)
3.9% ± 4.4% 3.2% ± 5.1% 5.1% ± 4.3% 4.4% ± 2.4%
4 282 74 (26.2%) 142 (50.4%)* 50 (17.7%) 16 (5.7%)*
13 231 3666 (27.7%) 5719 (43.2%) 2473 (18.7%) 1373 (10.4%)
1.5% ± 5.2% 7.1% ± 5.9% 1.0% ± 4.5% 4.7% ± 2.8%
5 244 64 (26.2%) 109 (44.7%) 58 (23.8%) 13 (5.3%)*
14 961 3987 (26.6%) 6214 (41.5%) 3061 (20.5%) 1699 (11.4%)
0.4% ± 5.6% 3.1% ± 6.3% 3.3% ± 5.4% 6.0% ± 2.9%
6 226 66 (29.2%) 99 (43.8%) 45 (19.9%) 16 (7.1%)*
16 153 3993 (24.7%) 6465 (40.0%) 3631 (22.5%) 2064 (12.8%)
4.5% ± 6.0% 3.8% ± 6.5% 2.6% ± 5.2% 5.7% ± 3.4%
7 378 105 (27.8%) 122 (32.3%) 78 (20.6%) 73 (19.3%)
19 362 5682 (29.3%) 5862 (30.3%) 3695 (19.1%) 4123 (21.3%)
1.6% ± 4.6% 2.0% ± 4.8% 1.6% ± 4.1% 2.0% ± 4.0%
8 469 72 (15.4%)* 163 (34.8%) 133 (28.4%)* 101 (21.5%)
18 026 3820(21.2%) 5873 (32.6%) 4059 (22.5%) 4274 (23.7%)
5.8% ± 3.3% 2.2% ± 4.4% 5.8% ± 4.1% 2.2% ± 3.8%
9 501 133 (26.5%)* 167 (33.3%) 112 (22.4%) 89 (17.8%)*
17 223 3734 (21.7%) 5458 (31.7%) 3962 (23.0%) 4069 (23.6%)
4.9% ± 3.9% 1.6% ± 4.2% 0.6% ± 3.7% 5.9% ± 3.4%
10 41 6 (14.6%) 16 (39.0%) 12 (29.3%) 7 (17.1%)
382 70 (18.3%) 154 (40.3%) 95 (24.9%) 63 (16.5%)
3.7% ± 11.5% 1.3% ± 15.7% 4.4% ± 14.6% 0.6% ± 12.1%
11 46 8 (17.4%) 19 (41.3%) 10 (21.7%) 9 (19.6%)
412 70 (17.0%) 150 (36.4%) 108 (26.2%) 84 (20.4%)
0.4% ± 11.5% 4.9% ± 15.0% 4.5% ± 12.7% 0.8% ± 12.1%
12 229 0 (0%) 20 (8.7%) 57 (24.9%) 152 (66.4%)*
11 792 6 (0.1%) 719 (6.1%) 2422 (20.5%) 8635 (73.2%)
2.6% ± 3.7% 4.4% ± 5.7% 6.9% ± 6.2%
17 94 41 (43.6%) 43 (45.7%) 9 (9.6%) 1 (1.1%)
840 294 (35.0%) 404 (48.1%) 114 (13.6%) 28 (3.3%)
8.6% ± 10.5% 2.4% ± 10.6% 4.0% ± 6.4% 2.3% ± 2.4%
18 92 41 (44.6%) 38 (41.3%) 13 (14.1%) 0 (0%)
1234 509 (41.2%) 551 (44.7%) 144 (11.7%) 20 (2.4%)
3.3% ± 10.5% 3.3% ± 10.4% 2.5% ± 7.3%
19 187 74 (39.6%) 79 (42.2%)* 23 (12.3%) 11 (5.9%)*
9711 4126 (42.5%) 3322 (34.2%) 1169 (12.0%) 1094 (11.3%)
2.9% ± 7.1% 8.0% ± 7.1% 0.3 ± 4.8% 5.4% ± 3.4%
20 141 57 (40.4%) 61 (43.3%) 21 (14.9%) 2 (1.4%)*
2782 1138 (40.9%) 1024 (36.8%) 366 (13.2%) 254 (9.1%)
0.5% ± 8.3% 6.5% ± 8.4% 1.7% ± 6.0% 7.7% ± 2.2%
21 188 84 (44.7%)* 74 (39.4%) 19 (10.1%) 11 (5.9%)*
7510 2611 (34.8%) 2791 (37.2%) 988 (13.2%) 1120 (14.9%)
9.9% ± 7.2% 2.2% ± 7.1% 3.0% ± 4.4% 9.1% ± 3.5%
22 1025 450 (43.9%)* 405 (39.5%) 109 (10.6%)* 61 (6.0%)*
15 560 4987 (32.1%) 5974 (38.4%) 2319 (14.9%) 2280 (14.7%)
11.9% ± 3.1% 1.1% ± 3.1% 4.3% ± 2.0% 8.7% ± 1.6%
23 219 15 (6.8%) 78 (35.6%) 70 (32.0%) 56 (25.6%)*
5787 365 (6.3%) 1929 (33.3%) 1611 (27.8%) 1880 (32.5%)
0.5% ± 3.4% 2.3% ± 6.5% 4.1% ± 6.3% 6.9% ± 5.9%
24 143 62 (43.4%) 65 (45.5%) 14 (9.8%) 2 (1.4%)*
5766 2099 (36.4%) 2583 (44.8%) 681 (11.8%) 403 (7.0%)
7.0% ± 8.2% 0.7% ± 8.3% 2.0% ± 4.9% 5.6% ± 2.0%

(Continues)

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
138 p. frisk et al.
Table 4. Continued
Survey no. n (survey) 0–44 yrs 45–64 yrs 65–74 yrs 75+ yrs
n (register)

Diff prop± C.I.95%

25 167 74 (44.3%)* 79 (47.3%) 12 (7.2%)* 2 (1.2%)*

6185 2168 (35.1%) 2757 (44.6%) 826 (13.4%) 434 (7.0%)
9.3% ± 7.6% 2.7% ± 7.7% 6.2% ± 4.0% 5.8% ± 1.8%
26 144 11 (7.6%)* 54 (37.5%)* 40 (27.8%) 39 (27.1%)*
6172 940 (15.2%) 1630 (26.4%) 1348 (21.8%) 2254 (36.5%)
7.6% ± 4.4% 11.1% ± 8.0% 5.9% ± 7.4% 9.4% ± 7.4%
27 239 40 (16.7%) 125 (52.3%)* 41 (17.2%) 33 (13.8%)*
10 343 1598 (15.5%) 4590 (44.4%) 2080 (20.1%) 2075 (20.1%)
1.3% ± 4.8% 7.9% ± 6.4 % 3.0% ± 4.8% 6.3% ± 4.4%
28 367 4 (1.1%) 139 (37.9%)* 130 (35.4%)* 94 (25.6%)*
17 261 290 (1.7%) 5579 (32.3%) 5164 (29.9%) 6228 (36.1%)
0.6% ± 1.1% 5.6% ± 5.0% 5.5% ± 4.9% 10.5% ± 4.5%
29 329 12 (3.6%) 161 (48.9%) 105 (31.9%) 51 (15.5%)*
28 324 970 (3.4%) 12 846 (45.4%) 8955 (31.6%) 5553 (19.6%)
0.2% ± 2.0% 3.6% ± 5.4% 0.3% ± 5.1% 4.1% ± 3.9%
30 10 9 (90%)* 0 (0%) 0 (0%) 1 (10.0%)
399 265 (66.4%) 92 (23.1%) 25 (6.3%) 17 (4.3%)
23.6% ± 19.2% 5.7% ± 18.7%
31 172 4 (2.3%) 45 (26.2%) 81 (47.1%)* 42 (24.4%)*
11 344 310 (2.7%) 3409 (30.1%) 4029 (35.5%) 3596 (31.7%)
0.4% ± 2.3% 3.9% ± 6.6% 11.6% ± 7.5% 7.3% ± 6.5%

*Proportions with non-overlapping confidence intervals on the 95% level of confidence.

CONFLICT OF INTEREST REFERENCES

1. Lee TH, Emanuel EJ. Perspective. Tier 4 drugs and the fraying of the social
The authors declare no conflict of interest. compact. N Engl J Med 2008; 359: 333–335.
2. Zuvekas SH, Cohen JW. Prescription drugs and the changing concentration of
health care expenditures. Health Aff 2007; 26: 249–257.
3. Schneeweiss S, Avorn J. A review of uses of health care utilization databases for
KEY POINTS epidemiologic research on therapeutics. J Clin Epidemiol 2005; 58: 323–337.
• Pharmacy surveys on prescription drugs, 4. Eichler, HG, Abadie E, Breckenridge A, et al. Bridging the efficacy-
effectiveness gap: a regulator’s perspective on addressing variability of drug
conducted in pharmacies and enrolling only response. Nat Rev Drug Discov 2011; 10: 495–506.
those who appear at the pharmacy themselves, 5. Lu CY. Observational studies: A review of study designs, challenges and
strategies to reduce confounding. Int J Clin Pract 2009; 63: 691–697.
do not provide representative data on patients 6. Hartzema AG, Tilson HH, Chan KA (eds.). Pharmacoepidemiology and
aged 75 years or older. Therapeutic Risk Management. Harvey Whitney books: Cincinatti, USA, 2008.
• This selection bias appears, to a varying extent, 7. Wettermark B, Hammar N, MichaelFored C, et al. The new Swedish Prescribed
Drug Register - opportunities for pharmacoepidemiological research and experi-
in all therapeutic areas studied ence from the first six months. Pharmacoepidemiol Drug Saf 2007; 16: 726–735.
• For pharmacy-based surveys to provide valid 8. Haukka J, Suvisaari J, Tuulio-Henriksson A, Lönnqvist J. High concordance
between self-reported medication and official prescription database information.
data representative of all patients, data should Eur J Clin Pharmacol 2007; 63: 1069–1074.
be collected alongside with data from other 9. Nielsen MW, Søndergaard B, Kjøller M, Hansen EH. Agreement between self-
reported data on medicine use and prescription records vary according to method
methods reaching the oldest patients. of analysis and therapeutic group. J Clin Epidemiol 2008; 61: 919–924.
doi:10.1016/j.jclinepi.2007.10.021
10. Grimsmo A, Hagman E, Faikø E, et al. Patients, diagnoses and processes in
general practice in the Nordic countries. An attempt to make data from
computerised medical records available for comparable statistics. Scand J Prim
ACKNOWLEDGEMENTS Health Care 2001; 19: 76–82.
11. Wettermark B, Vlahovic-Palcevski V, Salvesen Blix H, Rønning M, Vander
Stichele RH. Drug Utilization Research. In Pharmacoepidemiology and
We gratefully acknowledge the staff at all the pharmacies Therapeutic Risk Assessment. Hartzema AG, Tilson HH and Chan KA (eds).
involved in the regular collection of survey data. Harvey Whitney books: Cincinatti, USA, 2008: 159–195.
12. Heerdink ER, Leufkens HG, Koppedraaijer C, Bakker A. Information on drug
This study was funded by the Swedish Academy of use in the elderly: a comparison of pharmacy, general-practitioner and patient
Pharmaceutical Sciences. The individual surveys data. Pharm World Sci 1995; 17: 20–24.
13. Orrico KB. Sources and types of discrepancies between electronic medical records
included were commissioned and funded by the and actual outpatient medication use. J Manag Care Pharm 2008; 7: 626–631.
pharmaceutical companies holding the marketing li- 14. Ekedahl A, Brosius H, Jönsson J, et al. Discrepancies between the electronic
medical record, the prescriptions in the Swedish national prescription repos-
cense for each drug and the Dental and Pharmaceutical itory and the current medication reported by patients. Pharmacoepidemiol
Benefits Agency, respectively. Drug Saf 2011; 11: 1177–1183. doi: 10.1002/pds.2226. Epub 2011 Aug 22.

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds
 selection bias in pharmacy-based surveys 139
15. Lau HS, de Boer A, Beuning KS, Porsius A. Validation of pharmacy records in 26. Swedish National Board of Health and Welfare. Läkemedel - statistik för år 2012.
drug exposure assessment. J Clin Epidemiol 1997; 5: 619–625. Socialstyrelsen: Stockholm, 2012 [Drug statistics for 2012] Available from: http://
16. Monster TB, Janssen WM, de Jong PE, de Jong-van den Berg LT. PREVEND www.socialstyrelsen.se/publikationer2013/2013-3-21 [27 April 2013].
Study Group Prevention of REnal and Vascular ENT Stage Disease Pharmacy 27. Altman DG, Machin D, Bryant NB, Gardner MJ. Statistics with confidence
data in epidemiological studies: an easy to obtain and reliable tool. (2nd ed). BMJ Books, 2000.
Pharmacoepidemiol Drug Saf 2002; 11: 379–384. 28. Caskie GI, Willis SL. Congruence of self-reported medications with pharmacy pre-
17. Noize P, Bazin F, Dufouil C, et al. Comparison of health insurance claims and scription records in low-income older adults. Gerontologist 2004; 44: 176–185.
patient interviews in assessing drug use: data from the Three-City (3C) Study. 29. Klungel OH, de Boer A, Paes AH, et al. Agreement between self-reported anti-
Pharmacoepidemiol Drug Saf 2009; 18: 310–319. hypertensive drug use and pharmacy records in a population-based study in The
18. Frisk P, Mellgren TO, Hedberg N, et al. Utilisation of angiotensin Netherlands. Pharm World Sci 1999; 21: 217–220.
receptor blockers in Sweden: combining survey and register data to study 30. Boudreau DM, Daling JR, Malone KE, et al. A validation study of patient
adherence to prescribing guidelines. Eur J Clin Pharmacol 2008; interview data and pharmacy records for antihypertensive, statin, and
64: 1223–1229. antidepressant medication use among older women. Am J Epidemiol 2004;
19. Frisk P, Rydberg T, Carlsten A, Ekedahl A. Patients´ experiences with 159: 308–317.
generic substitution – a Swedish pharmacy survey. J Pharm Health Serv Res 31. Jensen E, Schroll M. A 30-year survey of drug use in the 1914 birth cohort in
2011; 2: 9–15. Glostrup County, Denmark: 1964–1994. Aging Clin Exp Res 2008; 20: 145–152.
20. Bradburn N, Sudman S, Wansink B. Asking Questions: The definitive guide to 32. Bjerrum L, Sogaard J, Hallas J, Kragstrup J. Polypharmacy: Correlation with sex,
questionnaire design. John Wiley & Sons: San Francisco, USA, 2004. age and drug regimen. Eur J Clin Pharmacol 1998; 54: 197–202.
21. Carvajal A, Arias LH, Vega E, et al. Gastroprotection during the administration 33. Nobili A, Franchi C, Pasina L, et al. Drug utilization and polypharmacy in an
of non-steroidal antiinflammatory drugs. A drug utilization-study. Eur J Clin Italian elderly population: the EPIFARM-Elderly Project. Pharmacoepidemiol
Pharmacol 2004; 60: 439–444. Drug Saf 2011; 20: 488–496.
22. Motola G, Mazzeo F, Rinaldi B, et al. Self-prescribed laxative use: a drug- 34. Drug dispensing data from 2011 from the Swedish Prescribed Drug Register.
utilization review. Adv Ther 2002; 19: 203–208. 35. Rikala M, Hartikainen S, Sulkava R, Korhonen MJ. Validity of the Finnish
23. Gislason T, Olafson O, Sigvaldason A. Use of antiasthma drugs in Iceland: a Prescription Register for measuring psychotropic drug exposures among
drug utilization study. Eur Respir J 1997; 10: 1230–1234. elderly finns: a population-based intervention study. Drugs Aging 2010;
24. Mehuys E, Paemeleire K, Van Hees T, et al. Self-medication of regular head- 27: 337–349.
ache: A community pharmacy-based survey. Eur J Neurol 2012; 19: 1093–1099. 36. Virués-Ortega J, de Pedro-Cuesta J, Seijo-Martinez M et al. Prevalence of dis-
25. World Health Organization (WHO). Guidelines for ATC classification and DDD ability in a composite ≥ 75 year-old population in Spain: A screening survey
assignment. WHO Collaborating Center for Drug Statistics Methodology, Oslo, based on the International Classification of Functioning. BMC Public Health
2012. www.whocc.no 2011; 11: 176. http://www.biomedcentral.com/1471-2458/11/176

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2014; 23: 128–139
DOI: 10.1002/pds