18. Di Carlo A, Lamassa M, Pracuccci G et al for the European
Biomed Study of Stroke Care Group. Stroke in the very old-
clinical presentation and determinants of 3-month functional
outcome: a european perspective. Stroke 1999; 30: 2313–9.
19. Appelros P, Nydevik I, Seiger A, Terent A. Predictors of
severe stroke:influence of preexisting dementia and cardiac
disorders. Stroke 2002; 33: 2357–62.
20. Henon H, Durieu I, Lebert F, Pasquier F, Leys D. Influence
of prestroke dementia on early and delayed mortality in
stroke patients. J Neurol 2003; 250: 10–6.
Published electronically 27 January 2011
© The Author 2011. Published by Oxford University Press on behalf
of the British Geriatrics Society. All rights reserved. For Permissions,
please email: firstname.lastname@example.org
Association of adverse drug reactions with
drug–drug and drug–disease interactions in
frail older outpatients
SIR—The most common type of medication-related
adverse events in older adults is Type A (‘augmented’)
adverse drug reactions (ADRs) [1–3]. Type A reactions are
an exaggeration of the expected pharmacologic effect of a
drug. These ADRs are more predictable, dose dependent
and potentially preventable than Type B (‘bizarre’) ADRs (i.
e. allergic reactions) [3,4].
The relationship of different elements of suboptimal
prescribing to ADRs in older outpatients has not been ade-
quately explored. Recently, Chrischilles et al.  examined
the association between multiple aspects of potentially inap-
propriate prescribing (defined by explicit criteria for
drugs-to-avoid, drug–disease interactions, drug–drug inter-
actions and therapeutic duplication) with self-reported
adverse drug events (ADEs). A recent study used a modi-
fied weighting system for the medication appropriateness
index (MAI), a validated measure that employs a standar-
dised implicit approach to determining prescribing appro-
priateness, to examine the association of potentially
inappropriate prescribing with self-reported ADEs [6, 7].
Neither of the above studies, however, had a specific focus
on Type A ADRs.
Given this background, the objective of this study was
to determine whether incorrect dosage, incorrect directions,
drug–drug interactions and drug–disease interactions, as
measured by the MAI, are associated with the Type A
ADRs among frail older veterans transitioning from the
hospital to the community.
Study design and study sample
This retrospective cohort study included a random sample
of 400 patients from the Geriatric Evaluation and
Management (GEM) Drug Study, which examined the
impact of GEM care on drug-related problems in 1,388
older veterans from 11 Veterans Affairs Medical Centers
(V AMC) . Details about inclusion and exclusion criteria
can be found elsewhere . We further restricted the
sample to those 359 patients taking one or more high-risk
medications (see Supplementary data available in Age and
Ageing online; http://www.ageing.oupjournals.org/) [3, 9,
10]. The study was approved by the V AMC Research and
Human Subjects Committees at each study site and the
Institutional Review Boards of Duke University and the
University of Pittsburgh.
Potential drug-related adverse events: data
collection, abstracted chart screening and
Detailed information about data collection and screening
for potential drug-related adverse events has been pre-
viously published [8, 11]. Briefly a trained research assistant
at each site prepared an abstract of each patients V AMC
inpatient and outpatient medical chart. A trained research
nurse reviewed the abstracted charts and screened for
potential drug-related adverse events using a standardised
approach. In addition, at the 12 month closeout a trained
research clinical pharmacist queried patients for self-reports
of potential drug-related adverse events using previously
validated methods . For each potential drug-related
adverse event identified by chart review and/or patient
interview, a trained clinical pharmacist created a detailed
narrative based on reporting information required by the
Food and Drug Administration MEDWatch program .
The primary outcome measure was any Type A ADR with
a causality rating of at least ‘possible’ . Blinded geriatri-
cian and geropharmacist pairs evaluated ADR causality
using the narrative and the validated Naranjo ADR causal-
ity algorithm . These ADRs were also assessed for type
of ADR (i.e. Type A or not) [3, 4]. Any discordances
among evaluators regarding the presence or type of ADR
were resolved by clinical consensus conference.
Primary independent variables
The primary independent variables were inappropriate
dosage, directions, drug–drug and drug–disease inter-
actions. Physician–pharmacist pairs evaluated each patient’s
medication regimen for these potential problems using the
MAI . Any discordances among evaluators were resolved
by clinical consensus conference.
Several factors may confound any relationship between
potentially inappropriate prescribing and ADRs and were
Demographic factors included categorical variables for age,
race and marital status. Health status factors included con-
tinuous measures for the number of high-risk medications,
chronic disease status (Charlson Comorbidity Index) and
for basic activities of daily living, and a categorical variable
for self-rated health [14,15].
forinmultivariableanalyses[5, 7, 9].
Baseline patient characteristics are presented as either
means and standard deviations or frequencies and percents
of the respective totals. We used backward selection (alpha
=0.15) multivariate logistic regression to determine covari-
ates to be added along with all four MAI variables in the
final model . Hosmer and Lemeshow  testing for
goodness of fit was conducted. We also conducted colli-
nearity diagnostic testing. Post hoc we reran the final multi-
individual variables for drug–drug interactions and drug–
disease interactions with one composite variable that sum-
marised the occurrence of either type of drug interaction.
SAS 9.1 software (SAS Institute Inc., Cary, NC, USA) was
used to perform all analyses.
Table 1 displays the characteristics of the study sample and
Supplementary data available in Age and Ageing online (http://
www.ageing.oupjournals.org/) display the rate of high-risk
Overall, 31.8% of patients experienced one or more Type
A ADRs during the follow-up period (median=1; range 1–
7). Only 14% of those with ADRs had more than two.
Table 2 provides information about the frequency of the
four MAI prescribing problems and the univariate and
multivariate results. Neither dosage nor directions problems
were significantly associated with Type A ADRs (P>0.05).
However, there was some evidence (P<0.10) that both
drug–drug interactions (adjusted odds ratio [AOR] 2.37,
95% confidence interval [CI] 0.91–6.11) and drug–disease
interactions (AOR 1.93, 95% CI 1.00–3.72) separately were
associated with Type A ADRs. Moreover, post hoc analyses
revealed that having either type of drug interaction problem
increased the risk of Type A ADRs nearly 2-fold (AOR
1.83, 95% CI 1.03–3.25).
To the best of our knowledge this is one of the first studies
using standardised implicit methods to examine types of
inappropriate prescribing and their association with Type A
ADRs as determined by a structured causality algorithm.
Our findings provide some evidence of an association of the
occurrence of Type A ADRs with both drug–drug and
drug–disease interactions. Chrischilles et al.  reported that
the use of drugs to avoid, and the occurrence of drug–
disease interactions both independently and significantly
increased the risk of the occurrence of one or more self-
reported ADE. They also reported a trend towards increased
risk of ADEs with drug–drug interactions and therapeutic
duplication. When these four explicit measures of inappropri-
ate prescribing were combined, a statistically significant
relationship with ADEs was demonstrated .Lund et al. ,
found that a modified summated MAI score, with the
highest weights applied to drug–drug and drug–disease inter-
actions, increased the risk of self-reported ADEs.
We did not find a significant relationship between
dosage and direction problems and Type A ADRs. One
possible explanation is that these two items measured by
the MAI only assess whether they are incorrect. Therefore
some of these incorrect ratings were due to dosage being
too low or directions for use too infrequent which in both
cases could lead to decreased drug concentrations and be
more likely to be associated with therapeutic failure than
with ADRs. This is consistent with the results from the
Lund et al. , who assigned these two items a weight of
zero and found they were not associated with ADRs.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Health status factors
Number of high-risk medications
Charlson Comorbidity Index
Basic activities of daily living
Fair/poor self-rated health
Table 1. Patient characteristics of frail older patients taking
high-risk medications at hospital discharge (n=359)
Variablesn Per centMean (SD)
SD, standard deviation.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Any dosage problem14841.2
Table 2. Prevalence of prescribing problems at hospital
discharge and their relationship with Type A ADRs in the
subsequent 12 months (n=359)
ratio (95% CI)
ratio (95% CI)a
2.37 (0.91–6.11)21 5.9
4813.4 2.09 (1.14–3.85)1.93 (1.00–3.72)
aControlling for basic activities of daily living score and self-rated health.
Neither the number of high-risk medications nor the comorbidity index were
retained in the backward selection (alpha=0.15) multivariable logistic
regression model. Hosmer–Lemeshow goodness-of-fit test (χ2=7.05; df=8;
P=0.53) suggests adequate model fit. No collinearity problems were detected
with the final multivariable model.
This study has a number of limitations. First, the associ-
ation did not achieve statistical significance. Second, the
study relied primarily on chart review of information to
assess prescribing and ADR causality. We may have under-
estimated problems if the information was not recorded or
was erroneously recorded in the medical chart.
Third, we utilised a modification of the MAI incorporat-
ing only four of the original 10 items. This modification
has not been independently validated. Moreover, our rate
of ADRs discovered only by self-reported potential ADEs
may be an underestimate especially in those with cognitive
impairment and those who died (nearly 8% of subjects)
before the 12 month follow-up period. It is reassuring that
a previous study by our group found that nearly 60% of
self-reported ADEs were also found during chart screening
. It is also possible that we overestimated the use of
high-risk drugs as only prescribing and not adherence was
assessed. Also the generalisability of our findings is
unknown as it involved mostly male frail older veteran out-
patients recently discharged from hospital and thus may
differ from other ADR studies of older outpatients who
were not hospitalised.
Despite these limitations, our results confirm that Type
A ADRs are common in frail older outpatients, and
provide evidence of an association with drug interactions.
Further studies, possibly with larger cohorts, are required to
test the reproducibility of these findings. In the meantime,
quality improvement activities to reduce ADRs by improv-
ing prescribing appropriateness should continue to include
a focus on the identification of potential drug–drug and
• Type A ADRs are common in frail older outpatients.
• This study provides evidence of an association between
drug interactions and Type A ADRs in frail older
• Future quality improvement activities should include a
focus on clinically important drug interactions.
Supplementary data mentioned in the text is available to
subscribers in Age and Ageing online.
Conflicts of interest
This study was not funded by outside sources. The original
GEM Drug Study  was supported by the National
Institutes of Health [R01-AG-15432] and the Veterans
Affairs Cooperative Study Program 006. J.T.H. was sup-
ported by the following: National Institute of Aging grants
MH082682), a National Institute of Nursing Research grant
(R01 NR010135), an Agency for Healthcare Research and
Quality grants (R01 HS017695 and R01HS018721) and a
V A Health Services Research grant (IIR-06-062).
JOSEPH T. HANLON1,2,3,4,*, RICHARD J. SLOANE5, CARL F. PIEPER5,6,
KENNETH E. SCHMADER5,7,8
1Department of Medicine (Geriatrics), University of Pittsburgh,
Kaufman Medical Building-Suite 514, 3471 5th Ave, Pittsburgh,
Tel: (+1) 412 692 2361; Fax: (+1) 412 692 2370.
2Department of Pharmacy and Therapeutics, University of
Pittsburgh, Pittsburgh, PA, USA
3Department of Epidemiology, University of Pittsburgh, Pittsburgh,
4Geriatric Research, Education, and Clinical Center and Center for
Health Equity Research and Promotion, Veterans Affairs Pittsburgh
Health Care System, Pittsburgh, PA, USA
5Center for the Study of Aging and Human Development, Duke
University Medical Center, Durham, NC, USA
6Department of Biostatistics and Bioinformatics, Duke University
Medical Center, Durham, NC, USA
7Department of Medicine (Geriatrics), Duke University Medical
Center, Durham, NC, USA
8Geriatric Research Education and Clinical Center, Durham
Veterans Affairs Medical Center, Durham, NC, USA
*To whom correspondence should be addressed
1. Tangiisuran B, Wright J, Van der Cammen T, Rajkumar C.
Adverse drug reactions in elderly: challenges in identification
and improving preventative strategies. Age Ageing 2009; 38:
2. Hamilton HJ, Gallagher PF, O’Mahony D. Inappropriate pre-
scribing and adverse drug events in older people. BMC
Geriatr 2009; 9: 5.
3. Hanlon JT, Handler S, Maher R, Schmader KE. Geriatric
Pharmacotherapy and Polypharmacy. In: Fillit H, Rockwood
KWoodhouse K, eds. Brocklehurst’s Textbook of Geriatric
Medicine, 7th edition. London, UK: Churchill Livingstone,
4. Edwards IR, Aronson JK. Adverse drug reactions: definitions,
diagnosis, and management. Lancet 2000; 356: 1255–9.
5. Chrischilles E, Van Gilder R, Wright K, Kelly M, Wallace RB.
Inappropriate medication use as a risk factor for self-reported
adverse drug effects in older adults. J Am Geriatr Soc 2009;
6. Hanlon JT, Schmader KE, Samsa GP et al A method for Download full-text
assessing drug therapy appropriateness. J Clin Epidemiol
1992; 45: 1045–51.
7. Lund BC, Carnahan RM, Egge JA, Chrischilles EA, Kaboli
PJ. Inappropriate prescribing predicts adverse drug events in
older adults. Ann Pharmacother 2010; 44: 957–63.
8. Schmader KE, Hanlon JT, Pieper CF et al Effectiveness of
geriatric evaluation and management on adverse drug reac-
tions and suboptimal prescribing in the frail elderly. Am J
Med 2004; 116: 394–401.
9. Schmader K, Hanlon JT, Weinberger M et al Appropriateness
of medication prescribing in ambulatory elderly patients. J Am
Geriatr Soc 1994; 42: 1241–7.
10. Hajjar E, Artz MB, Lindblad CI et al Risk factors and preva-
lence for adverse drug reactions in an ambulatory elderly
population. Am J Geriatr Pharmacother 2003; 1: 82–9.
11. Hanlon JT, Maher R, Lindblad C et al Comparison of
methods to detect potential adverse drug events in frail
elderly inpatients and outpatients. Am J Health Syst Pharm
2001; 58: 1622–6.
12. Kessler DA. Introducing MEDWatch. A new approach to
reporting medication and device adverse effects and product
problems. J Am Med Assoc 1993; 269: 2765–8.
13. Naranjo CA, Busto U, Sellers EM et al A method for estimat-
ing the probability of adverse drug reactions. Clin Pharmacol
Ther 1981; 30: 239–45.
14. Charlson ME, Pompei P, Ales KL et al A new method for
classifying prognostic comorbidity in longitudinal studies:
development and validation. J Chron Dis 1987; 40: 373–83.
15. Katz S, Akpom CA. A measure of primary sociobiologic
functions. Int J Health Serv 1976; 6: 493–507.
16. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd
edition. New York, NY: Wiley; 2000.
Published electronically 21 December 2010
© The Author 2010. Published by Oxford University Press on behalf
of the British Geriatrics Society. All rights reserved. For Permissions,
please email: email@example.com
Comparison of retrospective interviews
and prospective diaries to facilitate fall
reports among people with stroke
SIR—Monitoring falls is an important aspect of stroke
rehabilitation. Retrospective methods include face-to-face or
telephone interviews, postal questionnaires and medical
note reviews [1–4]. Prospective methods include post cards,
medical records, diaries and calendars as well as surveil-
lance systems [1, 5–7]. Although prospective methods are
considered preferable [1, 8, 9], it remains difficult to ascer-
tain the accuracy of reporting methods as both may lead to
over- or under-reporting [8, 10–17] and many factors influ-
ence recall [1,8,18–20].
Recall of falls in the previous year has reasonable
sensitivity (80–89%) and specificity (91–95%) [5, 11, 16, 18]
and produced better results than recall over shorter periods
[16, 18] but recent studies have advocated short recall
periods and intensive prospective (weekly or monthly)
follow-up over longer periods [1, 5]. Little is known about
the accuracy of fall reports among people with stroke. The
present study examined the agreement between two fall-
reporting methods (retrospective interviews and prospective
fall diaries) over a 12-month period.
This study formed part of a larger project predicting fall
risk among stroke patients . Ethical approval was
obtainedfrom the Southampton
Consecutively hospitalised patients were recruited if they
were independently mobile prior to stroke, able to give
consent: those who failed a cognitive function test 
which might have affected fall recall, were excluded. Two
researchers carried out assessments: the first carried out
tests of balance, function, mood and attention and was
kept blind to participants’ fall status; the second collected
data concerning falls, the focus of this paper.
Retrospective falls data were collected during an inter-
view with participants and significant others at 12 months
following discharge from hospital to the community, using
an interview schedule .Over the same time period, pro-
spective falls data were self-completed in a diary: partici-
pants (and significant others) were asked to record falls as
and when they occurred and were reminded to do so by
regular telephone calls and letters. We defined a fall as an
event that resulted in a person coming to rest unintention-
ally on the ground or other lower level, not as a result of a
major intrinsic event or overwhelming hazard : partici-
pants were asked to adhere to this definition when report-
ing falls for either method. Participants were classified as
fallers if they experienced one or more falls and as repeat
fallers if they experienced two or more.
Agreement between the retrospective and prospective
methods of collecting numbers of falls was examined using
kappa statistics and Bland and Altman limits , which give
a range of values in which the difference is expected to lie
95% of the time. Response and falling rates are presented
with 95% confidence intervals calculated in CIA .
Of the 122 participants recruited to the main study, retro-
spective falls information during the 12-month period fol-
lowing hospital discharge was available for 112 (93%,
confidence interval 85–95%). Of these, 62 (55%, confi-
dence interval 46–64%) reported one or more falls, and 45
(40%, confidence interval 32–49%) reported repeat falling.
Using the prospective diary method, data for 76 (62%,
confidence interval 53–70%) cases were available. Missing
prospective information was due to the diary being lost
(n=26), death (n=4) and no reason was recorded for the