Journals of Gerontology: Medical Sciences
cite as: J Gerontol A Biol Sci Med Sci, 2016, Vol. 71, No. 5, 625–631
doi:10.1093/gerona/glv181
Advance Access publication October 27, 2015
Research Article
Association Rules Analysis of Comorbidity and
Multimorbidity: The Concord Health and Aging in Men
Project
Fabian P. Held,1 Fiona Blyth,2 Danijela Gnjidic,2,4 Vasant Hirani,2,3,5 Vasikaran Naganathan,2,6
Louise M. Waite,2,6 Markus J. Seibel,6 Jennifer Rollo,1 David J. Handelsman,6
Robert G. Cumming,2,3,5,6 and David G. Le Couteur1,2,6
Charles Perkins Centre, University of Sydney, New South Wales, Australia. 2Centre for Education and Research on Ageing and the
Ageing and Alzheimers Institute, University of Sydney and Concord Hospital, New South Wales, Australia. 3School of Public Health,
4
Faculty of Pharmacy, 5ARC Centre of Excellence in Population Ageing Research, 6ANZAC Research Institute, University of Sydney, New
South Wales, Australia.
1
Address correspondence to David G. Le Couteur, MD, PhD, Centre for Education and Research on Ageing, Concord RG Hospital, Sydney, NSW
2139, Australia. Email: [email protected]
Received February 8, 2015; Accepted September 24, 2015
Decision Editor: Stephen Kritchevsky, PhD
Abstract
Background: Comorbidity and multimorbidity are common in older people. Here we used a novel analytic approach called Association Rules
together with network analysis to evaluate multimorbidity (two or more disorders) and comorbidity in old age.
Methods: A population-based cross-sectional study was undertaken where 17 morbidities were analyzed using network analysis, cluster analysis,
and Association Rules methodology. A comorbidity interestingness score was developed to quantify the richness and variability of comorbidities
associated with an index condition. The participants were community-dwelling men aged 70 years or older from the Concord Health and Ageing in
Men Project, Sydney, Australia, with complete data (n = 1,464).
Results: The vast majority (75%) of participants had multimorbidity. Several morbidity clusters were apparent (vascular cluster, metabolic cluster,
neurodegenerative cluster, mental health and other cluster, and a musculoskeletal and other cluster). Association Rules revealed unexpected
comorbidities with high lift and confidence linked to index diseases. Anxiety and heart failure had the highest comorbidity interestingness scores while
obesity, hearing impairment, and arthritis had the lowest (zero) scores. We also performed Association Rules analysis for the geriatric syndromes of
frailty and falls to determine their association with multimorbidity. Frailty had a very complex and rich set of frequent and interesting comorbidities,
while there were no frequent and interesting sets associated with falls.
Conclusions: Old age is characterized by a complex pattern of multimorbidity and comorbidity. Single disease definitions do not account
for the prevalence and complexity of multimorbidity in older people and a new lexicon may be needed to underpin research and health care
interventions for older people.
Key Words: Multimorbidities—Epidemiology—Frailty
The prevalence of co-occurring chronic diseases and disorders
increases dramatically with old age. This has significant and obvious
impact on the quality of life, function, hospitalization, and mortality
of older people (1–3). Health services and medical research, with
their focus on single diseases, are poorly equipped to support and
understand older people with complex health issues, leading to calls
for more research on comorbidity and multimorbidity to deal with
the aging population (2–5).
While comorbidity has been the focus of research for over
40 years (6), multimorbidity has been a more recent construct for
study (3,7). Most research on multimorbidity has utilized large
health system databases or smaller epidemiological studies to
© The Author 2015. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved.
For permissions, please e-mail: [email protected].
625
626
determine the following: (a) the prevalence and common patterns
of multimorbidity; (b) patterns of morbidities that cluster together
and provide insights into disease pathogenesis; and (c) the impact
on clinical outcomes and utilization of health services (1,3,5,7–10).
These studies have been very heterogeneous in terms of definitions
of diseases and multimorbidity, in part because they are typically
performed on convenience populations and sample sets (8,11) and
disregard the importance of geriatric syndromes (12). A heterogeneous group of statistical approaches has also been employed to
identify common clusters of morbidities—or those that occur more
frequently than by chance—including cluster analysis, principal
components analysis, and simple ratios of observed to expected
prevalence (7,13–17). These approaches all attempt to simplify and
interpret large and complex disease data sets.
To study these issues further and from a different perspective,
Association Rules and Frequent Set analyses were harnessed to interpret patterns of comorbidity and multimorbidity in a well-characterized cohort of older Australian men (Concord Health and Ageing in
Men Project, CHAMP) (18–21). Frequent Set analysis was developed
in the area of marketing research with the goal of identifying those sets
of items that are frequently purchased together from large databases
of transactions. An extension of Frequent Set analysis is Association
Rules analysis which is used to discover strong and often directional
associations between items that are purchased (22,23). This type of
analysis relies on measures of “interestingness,” which is a term related
to the effect size of a pattern, as opposed to simple tests of statistical
significance. Association Rules also allow analyses of the frequency
and interestingness of sets of items of any size and is not limited to
dyads. The results are presented here in the form of heatmaps that
provide the opportunity to visualize the prevalence and interestingness of the different patterns of comorbidities. We also developed a
novel metric, the comorbidity interestingness score, to compare the
complexity of comorbidities between diseases and syndromes.
The aims of this study were to determine the prevalence of multimorbidity and comorbidity in an older population and to evaluate the
relationships between their morbidities. We introduced Association
Rules as a novel method to evaluate comorbidities and developed a
comorbidity interestingness score to rank index morbidities according to the variety and complexity of their associated comorbidities.
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
because of the importance for older people. Data on 17 conditions
were obtained using self-report and/or objective measures: asthma,
chronic obstructive pulmonary disease or emphysema, arthritis,
osteoporosis, angina, congestive heart failure (or heart disease), myocardial infarction, anxiety or panic disorders, visual impairment,
hearing impairment, neurological disease (such as multiple sclerosis
or Parkinson’s disease), stroke, peripheral vascular disease, diabetes
types I and II, upper gastrointestinal disease, depression, self-reported
presence of back pain and obesity (defined as body mass index over
30 kg/m2). The 15-item Geriatric Depression and Goldberg Anxiety
scales were used to assess depressive (≥5 indicative of depressive
symptoms) and anxiety symptoms (with ≥5 considered as presence
of anxiety). The Baily-Lovie Chart was used to assess corrected visual
acuity (poor vision was defined as 6/19). Participants were screened
for cognitive impairment, and those who tested positive underwent
full clinical assessment after which the diagnoses of dementia or mild
cognitive impairment were made.
We also performed additional Association Rules analyses on
frailty, which was defined according to the Cardiovascular Health
Study and a history of falls as described previously (19,28).
Many different criteria to define diseases and conditions have
been used in past studies of multimorbidity. On average 18.5 different
conditions have been reported (range 4–102) (8). Here, the following morbidities were analyzed: arthritis, chronic obstructive pulmonary disease/asthma, angina, heart failure, heart attack, Parkinson’s
disease, stroke, peripheral arterial disease, diabetes mellitus, hearing
impairment, depression, anxiety, visual impairment, back pain, obesity, osteoporosis, and cognitive impairment. These conditions were
chosen because they are common, usually chronic, symptomatic, and
similar to the most commonly reported conditions in other studies
(8) except that disorders such as hypertension, hyperlipidemia, atrial
fibrillation, and reduced creatinine clearance were excluded because
they are largely asymptomatic risk factors rather than diseases and
are often effectively treated. Obesity was included on the basis of
the recommendation of Britt et al. (29). Association Rule analysis
is sensitive to combinations of items of very high with items of very
low frequency. To avoid detecting spurious associations, we limited
our analysis to these more common conditions.
Statistical Analysis
Methods
Study Population and Data Collection
Participants were community-dwelling men enrolled in the
CHAMP cohort, Sydney, Australia. Eligible participants were
≥70 years sampled via the Electoral Roll with a 53.7% participation rate, as described (18,19,24,25). The study was approved by the
Sydney Local Health District Human Research Ethics Committee.
Participants underwent assessments that comprised self-completed
study questionnaires and a clinical assessment with physical performance measures, neuropsychological testing, and medication inventory undertaken at hospital study clinic.
Morbidities and Geriatric Syndromes
“Comorbidity” was defined as the presence of two or more diseases
in a subject with a specific index disease, and “multimorbidity” was
used to describe the presence of two or more diseases in an individual
without reference to any index disease (5,26). We included conditions
that are common and captured in the Functional Comorbidity Index
that is associated with poor outcomes (27); and cognitive impairment
Data are presented as mean ± SD. Network analysis was performed
using Gephi (version 0.8.2) with a ForceAtlas 2 layout and modularity determined using a resolution of 0.6 (30). Analysis was restricted
to cases that had complete assessments for each condition (n = 1,494
of the total number recruited in the study n = 1,705).
Frequent Set and Association Rule analysis was undertaken using
R (version 3.1.0, R Development Core Team, 2014) and the library
arules (version 1.1–6). The code for analysis is available online at
https://bitbucket.org/FPHeld/champ_arules. The goal of this analysis is to identify patterns and combinations that meet a minimum
requirement for prevalence and at the same time occur much more
frequently together than would be expected under statistical independence. There are a number of terms used in Association Rules
analysis: (a) an “itemset” refers to a group of items; (b) “support”
refers to the frequency of a particular itemset; (c) “association rule” is
a relationship between sets of items ({A}→{B}) with an “antecedent”
{A} and a “consequent” {B}; (d) “lift” is a measure of the interestingness of an association rule that refers to how much more frequently
two sets of items occur together compared to how often would be
expected under statistical independence; (e) “confidence” of an
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
association rule refers to how frequently an item occurs, conditional
on an index item or item set; and (f) “interestingness” is a broad term
that refers to items that are found to be interesting as a result of combinations of their support, confidence, lift, or other statistic.
For each subject, each combination of two or more morbidities
was considered as a separate cluster or itemset. There are 131,072 (2k
when k is number of items) possible clusters for the 17 morbidities we
included and many more possible Association Rules. To reduce and
simplify analysis, minimum thresholds were used that define interestingness as follows: lift ≥2 and confidence >10%, while support was
unbounded. This means that for all interesting rules ({A}→{B}) presented here, the joint set {A, B} occurs at least two times more frequently
than we would expect under statistical independence, the consequent
({B}) occurred in at least 10% of all cases that show the morbidity in
the antecedent ({A}). However, we did not exclude rules with low support of the antecedent, which affects the results for morbidities of low
prevalence, such as Parkinson’s disease. Analysis was limited to “closed
itemsets” which means that if there are two nested itemsets with the
same level of support, only the one with the larger number of items will
be included in our analysis. Lastly, to better structure the discussion, we
present only rules with one index morbidity in the antecedent, while
considering clusters of all possible sizes in the consequent.
The results are presented as heat maps for each index morbidity,
with the comorbidities listed on the x-axis, while each row on the
y-axis represents a different cluster of comorbidities that occurred
with the index morbidity and exceeded the lift and confidence
627
thresholds. Color coding is used to represent measures of interestingness, with darker colors indicating clusters with higher frequencies (support) and how much more frequent they are than would be
expected under independence (lift).
Morbidities were ranked by the complexity and variety of their
associated comorbidities (comorbidity interestingness score). This
score was the percentage of participants with an index disease that
also belonged to a cluster of morbidities that exceeded the threshold criteria for interestingness. It provides an aggregate ranking of
individual morbidities and their tendency to co-occur with others.
The resulting ranking is more robust toward spurious occurrence of
combinations of morbidities.
Results
Subject Characteristics
Summary multimorbidity data are shown in Figure 1A and B; 75%
of participants had two or more morbidities while only 18% had
one and 7% had no morbidities. Very few participants with one disorder had no other associated comorbidities (1%–15%, Figure 1C).
The number of comorbidities that were associated with each disorder varied between approximately two and five (Figure 1D).
Network and Cluster Analyses
Figure 2 shows a network analysis of the 17 morbidities. This
demonstrates graphically the complicated nature of interactions
Figure 1. The prevalence of specific morbidities in older male participants of the CHAMP study population (A), and the prevalence of the number of comorbidities
(B). There were few participants with any one morbidity who had no associated comorbidities (C), while average number of comorbidities associated with most
diseases was approximately 2–5 (D).
628
Figure 2. Network analyses of multimorbidities in older male participants
of the CHAMP study population. The network was generated using the
ForceAtlas2 algorithm. The diameter of the nodes corresponds to the number
of connections (“edges”) it has with other nodes. An edge of weight = 1
between two nodes represents a single subject with both comorbidities. The
network is dominated by hearing impairment and arthritis.
between comorbidities and the dominance of hearing impairment
and arthritis in this cohort. Modularity analysis with a resolution of
0.6 revealed five plausible networks: (a) heart attack, heart failure,
angina, and peripheral arterial disease; (b) diabetes mellitus and obesity; (c) visual impairment, cognitive impairment, and Parkinson’s
disease; (d) depression, anxiety, and stroke; and (e) arthritis, osteoporosis, back pain, chronic obstructive pulmonary disease, and hearing impairment.
Association Rules and Frequent Set Analyses
Figure 3 shows the Frequent Set analysis of multimorbidities presented as a heatmap. There were 18 dyads and 1 triad of morbidities that were seen in more than 100 participants which reflected
the overall prevalence of these morbidities in the study population.
The most common dyad was arthritis with hearing impairment and
the most common triad was arthritis with hearing impairment and
obesity.
Figures 4 and 5 show the heatmaps for Association Rules analyses
of comorbidities for heart failure and anxiety (the other heatmaps
are shown in the Supplementary Material). Each row on the y-axis
represents a different combination of morbidities associated with the
index morbidity, ranked from the most frequent at the top. Obesity,
arthritis, diabetes, cognitive impairment, and hearing impairment had
no interesting comorbidity clusters above the threshold criteria we
chose for interestingness. While these morbidities occur frequently,
their associated comorbidities do not occur much more frequently
than expected. The combinations of comorbidities that are linked
to the index morbidity are colored in green and ranked from most
frequent occurrence (“support”) at the top in the darker green, to
the least common at the bottom in the lighter green. Beside these
green heatmaps are provided the blue heatmaps showing how much
more often they occurred compared to expected co-occurrence under
independence (“lift”) for the corresponding cluster of comorbidities.
Figure 4 shows the heatmaps for comorbidities associated
with congestive heart failure. The most common combination was
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
Figure 3. Heatmap for Frequent Set analysis of multimorbidities. The
combination of morbidities with the greatest frequency in the entire
population was arthritis and hearing impairment, while the second most
frequent was obesity and hearing impairment.
heart failure in association with heart attack and angina which is
not surprising, and this combination occurred about two to three
times more frequently than expected by chance (shown with asterisk). Other common clusters that occurred even more frequently
than expected included heart failure with depression and heart
attack; cognitive impairment and heart attack; arthritis, depression
and heart attack; and angina, depression, and heart attack. These
combinations all occurred four to eight times more frequently than
expected by chance.
Figure 5 shows the heatmaps for anxiety. Anxiety has a very
large number of combinations of comorbidities, the most common
including depression, hearing impairment, and arthritis. There are
many interesting clusters, those with the greatest lift included arthritis, chronic obstructive pulmonary disease/asthma, diabetes, hearing
impairment, and depression. Most of the clusters associated with
anxiety included depression.
We also performed Association Rules analysis for the geriatric syndromes of frailty and falls to determine their association
with multimorbidity (Supplementary Material). Frailty had a
very complex and rich set of frequent and interesting comorbidities, while there were no frequent and interesting sets associated
with falls.
Finally morbidities were ranked on the basis of the variability and complexity of their associated patterns of comorbidities
(Figure 6A). We created a comorbidity interestingness score, which
is the percentage of participants with the index condition who have a
comorbidity or cluster of comorbidities that exceeded the confidence
and lift thresholds. In essence, this score provides an insight into the
variety and interestingness of the comorbidities linked to each condition. The condition with the highest scores was anxiety, while the
common conditions arthritis, obesity, and hearing impairment had
the lowest (zero) scores. Figure 6B shows for each index morbidity,
the percentage of cases that are classified as interesting, for a range
of different thresholds along the axes. These patterns reflect the same
hierarchy of morbidities and their propensity to be associated with
other comorbidity clusters in interesting ways.
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
629
Figure 4. Heatmaps developed from Association Rules analysis for comorbidities co-occurring with congestive heart failure. Each row represents a different
combination or “itemset” of comorbidities. (A) The heatmap for the occurrence of the clusters that co-occurred with congestive heart failure, with the most
frequent combinations at the top of the figure and in the darkest color. The most frequent cluster was congestive heart failure + heart attack + angina (*). (B) The
heatmap for the lift of the clusters of comorbidities. The most frequent itemset (*) occurred with a lift of about 2–4.
Figure 5. Heatmaps developed from Association Rules analysis for comorbidities co-occurring with anxiety. (A) The frequency heatmap and (B) the lift heatmap.
Anxiety has a very large and complicated variety of clusters of comorbidities.
Discussion
In this study of community-dwelling men aged 70 years and older,
75% of participants had multimorbidity as defined as two or more
diseases or conditions. While comparisons to other studies are difficult because of heterogeneity in definitions and populations (8)
our results are similar to some other Australian studies where the
prevalence of multimorbidity was found to be 83% in people aged
75 years and older (29) and 52% in people aged 50 years and older
(17). In a review of 21 international studies, 13%–72% of the general population aged 75 years and older had multimorbidity (11).
Moreover, the presence of a single morbidity without any other cooccurring morbidity was only seen in 1%–15% of cases, and the
presence of one morbidity was associated with approximately two to
five other morbidities. The data clearly indicate that old age is a life
stage characterized by comorbidity and multimorbidity.
The network analysis figure showed that hearing impairment and
arthritis are dominant as a result of their high prevalence, and therefore frequent co-occurrence. Modularity analysis showed five clusters
including a vascular cluster, metabolic cluster, neurodegenerative cluster,
mental health and other cluster, and a musculoskeletal and other cluster. Similar patterns could be interpreted for many of these conditions
from the Association Rules heatmaps—for example, there are obvious
linkages between heart attack, angina, and heart failure on one hand,
and depression and anxiety on the other hand. Most other studies have
reported groupings around cardiovascular and metabolic conditions,
mental health conditions, and musculoskeletal conditions (10), which is
somewhat similar to our result, although we also had a neurodegenerative cluster possibly because our study cohort was old. Such network
analytical approaches could also be applied to other geriatrics issues such
as the complexity and patterns of medication utilization in older people
630
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
Figure 6. (A) The ranking of morbidities by the comorbidity interestingness score. This score represents the percentage of participants with the index morbidity
who also belong to a cluster that exceeds the threshold confidence and lift. Anxiety has the highest scores. (B) For each index morbidity, the percentage of cases
that are classified as interesting, for a range of different thresholds along the axes.
Many studies including ours indicate that the use of a single
disease terminology may not be useful in older people—instead we
should be generating a new geriatric lexicon to describe these groups
of co-occurring morbidities. In the future, medical research investigating interventions should consider targeting common clusters of
morbidities rather than single disease outcomes. Another implication relates to our understanding of the pathogenesis of morbidities in old age. In some cases, the cluster of comorbidities might
have a common etiology based on current understanding of disease,
for example, vascular disease in any organ is related to smoking,
hypertension and hyperlipidemia (5). In other clusters, possible
explanations are less forthcoming and might reflect the bidirectional
interactions between the aging biology and what we currently define
as diseases in different tissue and organ systems.
In this study, Frequent Set and Association Rules analysis with
heatmaps were utilized to interpret the comorbidity and multimorbidity patterns. The heatmaps provide clinicians with visual method
to determine which comorbidities commonly cluster with their disease of interest. The “lift” heatmaps provide a quick visual method
to discover interesting combinations of comorbidities that are occurring more frequently than expected and might provide insights into
disease and aging mechanisms. We developed a comorbidity interestingness score to rank the richness of comorbidities associated
with index disease. This identified heart failure and anxiety as having highest comorbidity interestingness scores, which is not unexpected given the contribution of vascular disease to heart failure,
and the relationship between chronic physical diseases and anxiety.
Conditions with a high comorbidity interestingness score can be seen
as disorders that should rarely be considered as single standalone
conditions because they are so deeply and complexly intermeshed
with their comorbidities.
There are limitations to our study, including its cross-sectional
design. There are no standard operational definitions for diseases
or morbidities that should be used in such studies (8,11) so we
have used a group of conditions based on criteria that we have
justified on basis of functional and symptomatic importance and
are aligned with the Functional Comorbidity Index (27). There
are some conditions we did not include, such as atrial fibrillation for which we do not have information and chronic kidney
disease which was rare in our cohort (<3% (31)). Definitions for
many of the conditions relied on self-report; however, this is relatively common in other studies of multimorbidity (10). Subjects
who volunteered for this project might not be representative of the
true population and are likely to have less frailty and disability.
We have used Association Rules analysis which is well established
in marketing research but has rarely been applied to medical or
epidemiological research previously, although similar approaches
have been reported (15,16). It should be noted that this type of
analysis is designed to identify combinations of items that are interesting because they are common and occur more frequently than
expected, without addressing statistical significance. The purpose
of Association Rules is to identify combinations that are worthy
of further investigation, and has the additional benefit of addressing combinations of more than two items. This approach has confirmed results of other studies (such as the association of anxiety
and depression with multimorbidity), which provides support of
its face validity, but has identified novel issues such as the difference between the complexity of comorbidities associated with falls
and frailty, and the ability to rank comorbidities according to our
comorbidity interestingness score.
In conclusion, multimorbidity and comorbidity are very prevalent in older men, and particular combinations of morbidities cluster together frequently and more often than expected by chance.
Association rules provide a useful new method for evaluating multimorbidity. Heart failure and anxiety had the greatest comorbidity
interestingness scores and represent complex comorbid conditions.
Considering single diseases in isolation in old age ignores the complex and complicated array of multimorbidities that are characteristic of old age.
Supplementary Material
Supplementary material can be found at: http://biomedgerontology.
oxfordjournals.org/
Journals of Gerontology: MEDICAL SCIENCES, 2016, Vol. 71, No. 5
Funding
The CHAMP study is funded by the Australian National Health and
Medical Research Council (initial grant No 301916), Sydney Medical School
Foundation, and Ageing and Alzheimer’s Institute (AAAI). D.G. is supported
by National Health and Medical Research Council Early Career Fellowship.
The funding source had no involvement in the study design, analysis, interpretation, or decision to submit this work.
References
1. Tinetti ME, Fried TR, Boyd CM. Designing health care for the most common chronic condition–multimorbidity. JAMA. 2012;307:2493–2494.
doi:10.1001/jama.2012.5265
2. Banerjee S. Multimorbidity-older adults need health care that can
count past one. Lancet. 2015;385:587–589. doi:10.1016/S01406736(14)61596-8.
3. Wallace E, Salisbury C, Guthrie B, Lewis C, Fahey T, Smith SM. Managing patients with multimorbidity in primary care. BMJ. 2015;350:h176.
doi:10.1136/bmj.h176
4. Fortin M, Soubhi H, Hudon C, Bayliss EA, van den Akker M. Multimorbidity’s many challenges. BMJ. 2007;334:1016–1017. doi:10.1136/
bmj.39201.463819.2C
5. Valderas JM, Starfield B, Sibbald B, Salisbury C, Roland M. Defining
comorbidity: implications for understanding health and health services.
Ann Fam Med. 2009;7:357–363. doi:10.1370/afm.983
6. Feinstein AR. The pre-therapeutic classification of co-morbidity in chronic
disease. J Chronic Dis. 1970;23:455–468.
7. Lefèvre T, d’Ivernois JF, De Andrade V, Crozet C, Lombrail P, Gagnayre
R. What do we mean by multimorbidity? An analysis of the literature on
multimorbidity measures, associated factors, and impact on health services organization. Rev Epidemiol Sante Publique. 2014;62:305–314.
doi:10.1016/j.respe.2014.09.002
8. Diederichs C, Berger K, Bartels DB. The measurement of multiple chronic
diseases–a systematic review on existing multimorbidity indices. J Gerontol A Biol Sci Med Sci. 2011;66:301–311. doi:10.1093/gerona/glq208
9. Capobianco E, Lio’ P. Comorbidity: a multidimensional approach. Trends
Mol Med. 2013;19:515–521. doi:10.1016/j.molmed.2013.07.004
10.
Prados-Torres A, Calderón-Larrañaga A, Hancco-Saavedra J,
Poblador-Plou B, van den Akker M. Multimorbidity patterns: a systematic review. J Clin Epidemiol. 2014;67:254–266. doi:10.1016/j.
jclinepi.2013.09.021
11.Fortin M, Stewart M, Poitras ME, Almirall J, Maddocks H. A systematic review of prevalence studies on multimorbidity: toward a more
uniform methodology. Ann Fam Med. 2012;10:142–151. doi:10.1370/
afm.1337
12.Min L, Kerr EA, Blaum CS, Reuben D, Cigolle C, Wenger N. Contrasting
effects of geriatric versus general medical multimorbidity on quality of ambulatory care. J Am Geriatr Soc. 2014;62:1714–1721. doi:10.1111/jgs.12989
13.Marengoni A, Rizzuto D, Wang HX, Winblad B, Fratiglioni L. Patterns
of chronic multimorbidity in the elderly population. J Am Geriatr Soc.
2009;57:225–230. doi:10.1111/j.1532-5415.2008.02109.x
14.Schafer I, von Leitner EC, Schon G, et al. Multimorbidity patterns in
the elderly: a new approach of disease clustering identifies complex
interrelations between chronic conditions. PLoS One. 2010;5:e15941.
doi:10.1371/journal.pone.0015941
631
15.van den Bussche H, Koller D, Kolonko T, et al. Which chronic diseases
and disease combinations are specific to multimorbidity in the elderly?
Results of a claims data based cross-sectional study in Germany. BMC
Public Health. 2011;11:101. doi:10.1186/1471-2458-11-101
16.Wong A, Boshuizen HC, Schellevis FG, Kommer GJ, Polder JJ. Longitudinal administrative data can be used to examine multimorbidity, provided
false discoveries are controlled for. J Clin Epidemiol. 2011;64:1109–1117.
doi:10.1016/j.jclinepi.2010.12.011
17.Islam MM, Valderas JM, Yen L, Dawda P, Jowsey T, McRae IS. Multimorbidity and comorbidity of chronic diseases among the senior Australians:
prevalence and patterns. PLoS One. 2014;9:e83783. doi:10.1371/journal.
pone.0083783
18.Cumming RG, Handelsman D, Seibel MJ, et al. Cohort profile: the Concord Health and Ageing in Men Project (CHAMP). Int J Epidemiol.
2009;38:374–378. doi:10.1093/ije/dyn071
19.Le Couteur DG, Blyth FM, Creasey HM, et al. The association of alanine
transaminase with aging, frailty, and mortality. J Gerontol A Biol Sci Med
Sci. 2010;65:712–717. doi:10.1093/gerona/glq082
20. Gnjidic D, Hilmer SN, Blyth FM, et al. High-risk prescribing and incidence
of frailty among older community-dwelling men. Clin Pharmacol Ther.
2012;91:521–528. doi:10.1038/clpt.2011.258
21.Hsu B, Cumming RG, Blyth FM, et al. Longitudinal and cross-sectional
relationships of circulating reproductive hormone levels to self-rated
health and health-related quality of life in community-dwelling older men.
J Clin Endocrinol Metab. 2014;99:1638–1647. doi:10.1210/jc.2013-3984
22.Agrawal R, Imieliński T, Swami A. Mining association rules between sets
of items in large databases. Proc ACM SIGMOD Intl Conf Mgt Data.
1993;207–216.
23.Tan PN, Steinbach M, Kumar V. Introduction to Data Mining. Boston,
MA, Addison-Wesley; 2006.
24.Le Couteur DG, Benson VL, McMahon AC, et al. Determinants of seruminduced SIRT1 expression in older men: The CHAMP study. J Gerontol
A Biol Sci Med Sci. 2011;66:3–8. doi:10.1093/gerona/glq158
25. Hirani V, Naganathan V, Cumming RG, et al. Associations between frailty and
serum 25-hydroxyvitamin D and 1,25-dihydroxyvitamin D concentrations in
older Australian men: the Concord Health and Ageing in Men Project. J Gerontol A Biol Sci Med Sci. 2013;68:1112–1121. doi:10.1093/gerona/glt059
26.Van den Akker M, Buntinx F, Knottnerus JA. Comorbidity or multi
morbidity: what’s in a name? A review of literature. Eur J Gen Pract.
1996;2:65–70.
27.Groll DL, To T, Bombardier C, Wright JG. The development of a comorbidity index with physical function as the outcome. J Clin Epidemiol.
2005;58:595–602. doi:10.1016/j.jclinepi.2004.10.018
28.Rochat S, Cumming RG, Blyth F, et al. Frailty and use of health and community services by community-dwelling older men: the Concord Health
and Ageing in Men Project. Age Ageing. 2010;39:228–232. doi:10.1093/
ageing/afp257
29.Britt HC, Harrison CM, Miller GC, Knox SA. Prevalence and patterns of
multimorbidity in Australia. Med J Aust. 2008;189:72–77.
30.Jacomy M, Venturini T, Heymann S, Bastian M. ForceAtlas2, a continuous
graph layout algorithm for handy network visualization designed for the Gephi
software. PLoS One. 2014;9:e98679. doi:10.1371/journal.pone.0098679
31.Gnjidic D, Bennett A, Le Couteur DG, et al. Ischemic heart disease, prescription of optimal medical therapy and geriatric syndromes in community-dwelling older men: A population-based study. Int J Cardiol.
2015;192:49–55. doi:10.1016/j.ijcard.2015.05.045