Multimorbidity patterns in Irish adults ageing with and without an intellectual disability using network analysis and association rules: A comparative analysis using data from the Intellectual Disability Supplement to the Irish Longitudinal Study on Ageing (IDS-TILDA)

Abstract

Multimorbidities; diseases and health conditions reported concurrently; are quantified and visualised herein for men and women who are ageing in Ireland with an intellectual disability data from the Intellectual Disability Supplement to the Irish Longitudinal Study on Ageing (IDS-TILDA). Network modelling, using association rules, provides insight into how likely certain diseases are to present together for each sex. Multimorbidity patterns including commonly reported pairs and triads of health conditions and clustering patterns of diseases presenting in people with an intellectual disability aged 50 and above are compared with the general population of the same age using previously published analysis from the Irish Longitudinal Study on Ageing (TILDA).

Introduction

The Intellectual Disability Supplement to The Irish Longitudinal Study on Ageing (IDS-TILDA) has gathered to date five waves of interview data at 3-year intervals from older adults with intellectual disability. This includes, but is not limited to, self-reported doctor’s diagnosis of a range of diseases and health conditions. The data provides rich information regarding the health status of individuals with intellectual disability and their demographic characteristics.

Multimorbidity is most commonly defined in multimorbidity network literature as two or more health conditions concurrently present within a person (Jones et al., 2023). Multimorbidity is an ongoing concern for the population with intellectual disability as they age. The IDS-TILDA Wave 5 report (Burke et al., 2023) points to some improvements across time in the levels of multimorbidity with an increase in the number of younger participants (40- to 50-year-olds) reporting no health conditions in the last 10 years (55.2% up from 36.7%) and a decrease in those reporting three or more health conditions (11.9% down from 23.3%). However, 72.4% of participants in the full cohort (aged over 40), reported multimorbidity and this is consistent with results from previous waves. Therefore, it is an important consideration for policy and service planning into the future.

The multimorbidity network presenting in the general ageing population in Ireland has been quantified and described in Hernández et al. (2019) using data from Wave 3 of The Irish Longitudinal Study on Ageing (TILDA). As these longitudinal studies are harmonised, a similar approach can be followed using data from IDS-TILDA at a close point in time, to compare multimorbidity patterns in the general population with the population with intellectual disability. Network analysis of multimorbidity is an emerging topic and to the authors’ knowledge, no methodological approach has been replicated in a second population. Moreover, there is no standard approach to modelling multimorbidity networks, and many articles are neither transparent nor replicable (Jones et al., 2023). Thus, one novelty of this article that it emulates a network approach to the analysis of multimorbidity patterns and compares multimorbidity network analysis of samples from two distinct populations.

The rates of multimorbidity in people ageing with intellectual disability in Ireland have been reported to be higher than the general population (Burke et al. (2023)). Analysis of the co-occurring pairs of conditions reported in IDS-TILDA and the prevalence of the combinations of health conditions suggested by Kirchberger et al. (2012) (cardiovascular/metabolic, liver/lung/joint/eye, mental/neurologic and gastrointestinal/ cancer) have been provided in McCarron et al. (2013). A network visualisation has not yet been produced for this cohort and neither has a comparison of the multimorbidity network or the groups of conditions that are likely to co-occur that can be compared with the general population. In McCarron et al. (2013), hypertension, eye disease, heart disease, endocrine disease, joint disease, lung disease, gastrointestinal disease, mental health conditions, stroke, cancer, neurological disease and liver disease were examined based on the literature and what we know to be important and prevalent conditions for people ageing with intellectual disability. In this article a different set of diseases and health conditions will be presented than that used in McCarron et al. (2013). More granularity is provided, to follow the multimorbidity network methodology used in the general population study (Hernández et al., 2019) as carefully as possible. However, mental health conditions and neurological conditions, both of whom are highly prevalent in the population ageing with intellectual disability are excluded. Nevertheless, this was necessary to be faithful to the comparative methodology which we felt important and of value to the field.

A network consists of a collection of nodes that are connected by edges representing a relationship (or lack thereof) between the nodes. Multimorbidities reported by the IDS-TILDA cohort can be viewed as a network dataset. In this study, a multimorbidity network is created where the nodes of the network are the health conditions / diseases. The nodes are connected by edges that link the disease pairs and are given a value between 0 and 1, where the value equates to the prevalence of that disease pair. For example, if the 2 health conditions are frequently co-occurring i.e. reported together by many survey participants, the weighted edge would have a value close to one, whereas a rarely co-occurring disease pair would have an edge value close to 0.

The size of the nodes is visualised in this article with their size proportional to the overall prevalence of the health condition. The prevalence of individual health conditions is likely to affect the prevalence of their co-occurrence. A highly prevalent condition is naturally more likely to occur with another highly prevalent condition than a pair of diseases with low prevalence. Thus, two very prevalent conditions have a greater chance of co-occurring by random chance and not because of anything physiological/biological linking the diseases. It is important to correct for this potential bias to avoid bias against low prevalence diseases that co-occur less frequently than high prevalence diseases. To this end, the effect of the prevalence of individual health conditions is adjusted for using the standardised lift of an association rule (McNicholas et al., 2008) as per Hernández et al. (2019). This provides a ‘measure of interestingness’ of pairs of health conditions that occurred more often than due to chance, while also taking into account the prevalence of the diseases.

Assumptions are required to compare the multimorbidity network patterns in people with intellectual disability to those of the general population aged over 50 and living in Ireland and to explore the network and patterns of co-occurring diseases and conditions. Although the TILDA study did not specifically exclude people with intellectual disability, they did exclude people living in institutional care in their initial sampling strategy in Wave 1. At that time, according to the National Intellectual Disability Database (Kelly and O’Donohoe, 2014), only 17.3% of those aged 55 years or over were living in a home setting. Thus, few people with intellectual disability lived in the community, and as a result, although TILDA did not specifically exclude intellectual disability at baseline, no one with intellectual disability was selected into their sample. Thus, these 2 samples are assumed to be coming from distinct populations that are living in the same country: Ireland.

Another difference between the samples is their age distributions. In the TILDA sample, participants are aged over 50 whereas in the IDS-TILDA sample, participants are included from age 40. For the purposes of this study, we exclude 197 IDS-TILDA participants aged 49 and under in our sample. A further contrast is that in most cases people with intellectual disability live very different lives than the general population. People ageing with intellectual disability in Ireland currently typically do not work (Wave 5 reported just one in 10 participants (10.9%) in paid employment) and thus do not retire from work as they age. They do not typically marry or have children (≤ 1%), and they often are not educated beyond primary level. It was reported in IDS-TILDA Wave 1 that half of respondents completed primary level but one third of participants did not receive any formal education at all (McCausland et al., 2016). This is in stark contrast with the general population where a majority (62%) complete second level education (Barrett et al., 2011). Moreover, some risk factors for poor health outcomes such as obesity (Ryan et al., 2021) and sedentary behaviour (Lynch et al., 2024) are typically higher than reported in the general population. On the other hand, rates of over-use of alcohol and smoking, often associated with poor health outcomes, are lower for people with intellectual disability (Wormald et al., 2023), than reported in the population ageing without intellectual disability.

Alongside comorbidity and multimorbidity comes polypharmacy; the prescription of many medications. This also presents differently for people ageing with intellectual disability when compared with the general ageing population. A recent study found that one in three participants ageing with intellectual disability took five to nine medicines (polypharmacy) and over one in five took ten or more medicines (hyper-polypharmacy) (O’Dwyer et al., 2016). This compares to one in five of the TILDA sample reporting polypharmacy and just 2% or one in fifty reporting hyper-polypharmacy in the general population (Richardson et al., 2012). This increases the risk of drug-drug interactions and adverse events such as hospitalisation and death. In a cross-sectional study of adults ageing in care, at least one potential drug-drug interaction was detected in 54% of the 197 individuals using two or more drugs (Hermann et al., 2021). Moreover, adverse events related to drug use are estimated to cause around 10% of hospital admissions in older people (Oscanoa et al., 2017). Thus, polypharmacy is an issue of concern.

This article provides a novel description of the multimorbidity patterns presenting in the IDS-TILDA sample. It compares the IDS-TILDA sample with previously published results from the general ageing study TILDA. Due to the differences described, as well as the defining difference in the two samples; intellectual ability levels, we posit that the multimorbidity picture may be quite different for the two samples. This study aspires to provide useful evidence of these differences as well as information to support the planning of service provision into the future for people ageing with intellectual disability. It may be useful for clinicians to guide efficient assessment of patients for further health conditions based on those already identified. It may provide evidence to suggest potential common drug-drug interactions that may cause harm based on the most common comorbidities presenting. It also may provide insight and invoke further discussion about underlying mechanisms that may explain differences in co-occurrence of diseases for two very different populations of people ageing in Ireland.

Methods

Data was obtained from the second wave of data collected by the Intellectual Disability Supplement to the Irish Longitudinal Study on Ageing (IDS-TILDA). The IDS-TILDA protocol is harmonized with other international longitudinal studies of ageing including the Irish Longitudinal Study on Ageing (TILDA) upon which this methodology is based (Hernández et al., 2019). Both studies consist of a pre-interview questionnaire (PIQ) and a computer assisted personal interview (CAPI). The latter is carried out by trained field researchers. In the case of participants with intellectual disability, the answers may be provided by a proxy; a person who knows the participant well such as a family member or a key healthcare worker. Participants may also be partially or fully supported by a proxy throughout the interview. In the general population study, proxy report is also allowed if the participant is not capable of answering (usually due to cognitive decline).

The closest timepoint to the analysis presented in Hernández et al. (2019) (Wave 3 of TILDA interviews conducted in 2014 ₋ 2015) is Wave 2 of IDS-TILDA (interviews conducted in 2013 ₋ 2014). Thus, the use of the Wave 2 IDS-TILDA data ensures optimal temporal comparability in this analysis. Participants were part of the original IDS-TILDA sample (Wave 1, reported in 2011) which consisted of 753 randomly selected persons aged over 40 registered with Ireland’s National Intellectual Disability Database (NIDD). Participants were followed up in the Wave 2 interviews. A total of 45 participants did not take part in Wave 2, 34 of whom had died. Participants aged 50 and over from this wave who responded to questions related to the health conditions defined in this study comprise the sample of 310 participants analysed in this study.

The analysis herein is a replication of analysis carried out in the multimorbidity network of the general population (Hernández et al., 2019). Thus, the analysis is restricted to the health conditions and diseases listed in Hernández et al. (2019) such that we can reproduce the methodological approach. For the vast majority of the health conditions, the questions asked of participants were identical. However, there are some differences of note. There were two health conditions; kidney disease and anaemia, which were listed in (Hernández et al., 2019) but were not asked by IDS-TILDA in Wave 2. These conditions have been excluded from this comparative analysis. A further difference is that depressive symptoms were measured in the TILDA sample (from the general population) using the Centre for Epidemiological Studies Depression Scale (CES-D) whereas for the IDS-TILDA sample (from the population with intellectual disability), the Hospital Anxiety and Depression Scale (HADS) was used. These are not directly comparable.

Obesity is a prevalent health condition in both the general population and the population with intellectual disability. In the IDS-TILDA study, it was necessary to use a surrogate height measure (ULNA length measured from the olecranon process and the styloid process) for participants who were unable to have their height measured. Further, for those with mobility issues, the mid-upper arm circumference (MUAC) was used as a proxy for BMI, where, as per the Health Service Executive Policy (HSE, 2021) for Adults Accessing Disability Services, if MUAC is greater than 32 cm, the participant is considered likely to be obese.

In both the TILDA and IDS-TILDA studies, urinary incontinence was defined as any involuntary loss of urine from the bladder within the last 12 months. In IDS-TILDA this includes several participants that were incontinent from birth.

To ensure data protection and prevent identifiability of individuals, it was necessary to exclude or combine less common health conditions. Parkinson’s disease, varicose ulcers and liver disease were deemed inappropriate for recategorisation and were excluded from this analysis as less than 5 participants reported these conditions. The heart conditions: angina, heart attack, heart failure, stroke and transient ischaemic attack were required to be combined and renamed as ‘other heart conditions’ as the number of participants reporting these individual conditions was small (n ≤ 5). For the same reason, glaucoma and age-related macular degeneration were also combined and renamed ‘other eye conditions’. Asthma and other lung disease were combined and renamed ‘lung disease’.

Statistical analysis

Bivariate analysis was conducted using chi-square tests to determine significant associations between reported prevalence of health conditions and sex. The null hypothesis was that there was no difference between the health condition prevalence statistics of men and women in the study. Resulting p-values<=0.05 were considered significant evidence at the 5% level to reject the null hypothesis and conclude a significant difference between the two prevalence statistics. A false discovery rate (Benjamini and Hochberg, 1995) was determined to address the potential effect of multiple comparisons made using the same dataset, with false positive rates examined from 5% to 25%. Comparisons were made to prevalence statistics reported in Hernández et al. (2019) for men and women in the ageing general population.

The number of multimorbidities was summarised and displayed as a histogram of the percentage of the sample in this study reporting each number of health conditions. The prevalence of co-occurring pairs of health conditions (known as ‘support’) was calculated using association rules and the R package arules (Hahsler et al., 2005). A network visualisation was created for male and female multimorbidity separately, where the ‘nodes’ are circles representing each health condition and the ‘edges’ are curved lines between nodes. Node circles are larger for more prevalent diseases and thicker edges reflect more commonly occurring comorbidities. The ‘confidence’; the probability that health condition B occurs given that health condition A is already present, was calculated and visualised as a heatmap of probability values from 0 to 1. This aligns with the classical definition of comorbidity where diseases are considered in reference to an index condition of interest (Feinstein, 1970). Following the approach given in Hernández et al. (2019), a standardised version of the observed/expected ratio (known as the ‘standardised lift’) was also calculated and used to determine and rank pairs of diseases that occur more often one would expect at random. The standardised lift provides a value between 0 and 1 where values higher than 0.5 represent comorbidities that occurred more often than due to chance, while also taking into account the prevalence of the diseases. Clusters of health conditions were determined using the multimorbidity network data for males and females and the fast greedy search approach used in the general population study. Commonly occurring triads of disease were also calculated for each sex.

Results

Table 1 provides summary information on the sample prevalence of 19 health conditions stratified by sex. These 19 health conditions were reported by over 5% of the IDS-TILDA sample. The rows of the table are ordered by absolute differences in prevalence between the sexes. A majority; 14 health conditions out of 19; were reported by women more often than by men. Of these, the largest differences were found in the reporting of osteoporosis, obesity, thyroid conditions and diabetes. These 4 differences were statistically significant (chi-square p-values<=0.05). They were also significant at the 25% false discovery rate for multiple testing (Benjamini and Hochberg, 1995). Osteoporosis and thyroid conditions were also significantly more prevalent in women than men in the TILDA (general ageing population) sample. The converse was true for thyroid conditions and diabetes in the TILDA sample (Hernández et al., 2019), where both were significantly more prevalent in men than in women. Arthritis and urinary incontinence were more prevalent in women in both the IDS-TILDA and the TILDA samples but were found to be statistically significant in the general population sample only.

Table 1.

Overall health condition prevalence reported in the IDS-TILDA sample, stratified by sex and ordered by greatest to least absolute difference between male and females reported prevalence. Chi-square test p-values are provided and are in bold type where significant at both the 5% significance level and a 25% false discovery rate.

Disease	Total %	Male %	Female %	p-value
	(n = 310)	(n = 119)	(n = 191)
Osteoporosis	16.77	6.72	23.04	0.0003
Obesity	37.74	27.73	43.98	0.01
Thyroid	18.06	11.76	21.99	0.03
Diabetes	10.65	5.04	14.14	0.02
Arthritis	13.23	8.40	16.23	0.07
Poor hearing	12.26	7.56	15.18	0.07
Cataracts	15.81	11.76	18.32	0.17
Urinary incontinence	42.58	38.66	45.03	0.32
High cholesterol	38.71	35.29	40.84	0.39
Hypertension	22.26	19.33	24.08	0.40
Poor vision	19.03	17.65	19.90	0.73
Other eye conditions	5.16	4.20	5.76	0.73
Depression (HAD scale)	10.97	10.08	11.52	0.84
Lung disease	7.10	6.72	7.33	1.00
Cancer	5.16	5.04	5.24	1.00
Heart murmur	6.77	8.40	5.76	0.50
Other heart conditions	7.10	9.24	5.76	0.35
Heart arrhythmia	7.42	10.08	5.76	0.23
Stomach ulcers	5.48	8.40	3.66	0.13

Heart conditions (heart murmur, heart arrhythmia and other heart conditions) and stomach ulcers were reported more often by men than women in both samples. These differences were statistically significant in the TILDA sample only, perhaps due to the small numbers reporting these conditions in the IDS-TILDA sample.

The proportion of respondents in the IDS-TILDA sample reporting from 0 to 9 reported health conditions is displayed in Figure 1. Multimorbidity (2 or more health conditions) was reported by 83.5%. The median number of health conditions reported was 3 and 71.6% of people reported between 2 and 4 health conditions. A maximum of 9 co-occurring health conditions were reported. Only 6.5% of this sample ageing with intellectual disability reported no health conditions. Although not directly comparable due to differences noted in the health conditions included, it is of interest to compare the ageing general population TILDA sample, where the estimated proportion reporting no health conditions was greater (9.08%) and reported multimorbidity was lesser (73.25%). The proportion of the general population reporting 5 or more health conditions was similar for both samples (23.2% of the sample ageing without intellectual disability compared with 22.1% for the sample ageing without intellectual disability).

Figure 1.

Proportion of the sample according to number of reported health conditions.

Multimorbidity networks

The networks of multimorbidities reported by the women and men in this sample are presented in Figures 2 and 3. The size of the nodes (circles representing health conditions) is proportional to the overall prevalence of the disease. The width of the edge connecting the nodes is proportional to the number of participants reporting the comorbidity indicated by both nodes. The thickest lines in both multimorbidity network graphs represent the most prevalent comorbidities reported by both women and men in this sample and are those involving high cholesterol, obesity and urinary incontinence. This is different from the general population where urinary incontinence is paired with other diseases more often for women but not men. Hypertension is also a key health condition in the comorbidity network for the general population study of both sexes but not for those with intellectual disability.

Figure 2.

Network of multimorbidities reported by the female IDS-TILDA sample. The size of the nodes (circles representing health conditions) is proportional to the overall prevalence of the disease. The width of the edge connecting the nodes is proportional to the number of female participants with the comorbidity indicated by both nodes.

Figure 3.

Network of multimorbidities reported by the male IDs-TLDA sample. The size of the nodes (circles representing health conditions) is proportional to the overall prevalence of the disease. The width of the edge connecting the nodes is proportional to the number of male participants with the comorbidity indicated by both nodes.

In general, the nodes and the edges are thicker for the females than for males. This is because both the prevalence of the individual conditions and the comorbid pairs are typically higher in this sample for females than for males. Obesity and high cholesterol are reported together by 18.9% of women compared with 12.6% of men in the IDS-TILDA sample. The link between thyroid conditions and osteoporosis with urinary incontinence is relatively thick for the women (14.6% and 12.04% respectively) but not for the men (less than 5%). This result mirrors findings for osteoporosis and urinary incontinence in the general ageing population but not for thyroid conditions and urinary incontinence where the links are not thick for either sex (Hernández et al., 2019).

The most common comorbid pair in this study was obesity and high cholesterol and this was reported by 18.8% of females and 12.61% of males in the IDS-TILDA sample. The next most reported comorbid pairs were reported by both men and women; high cholesterol and urinary incontinence (17.28% of females and 12.61% of males), obesity and urinary incontinence (16.23% of females and 9.24% of males) and hypertension and high cholesterol (15.71% of females and 8.40% of males). This is different from the general population where the hypertension and high cholesterol comorbidity was the most reported comorbidity by both sexes (reported by c. one in four TILDA participants) but other commonly reported comorbidities were different between the sexes. In the male TILDA sample, the most reported comorbidities were hypertension and arthritis, high cholesterol and arthritis and obesity and hypertension which were experienced by at least 15% of the TILDA sample. These comorbid pairs were also commonly reported by females but other pairs such as osteoporosis and high cholesterol (16.67%); and urinary incontinence and high cholesterol were reported by more than 15% of the female TILDA sample.

Standardised lift - adjusting for disease prevalence

The differences in multimorbidity networks described above and depicted in Figure 3 could be due to the higher prevalence of these diseases in the female sample compared to the male. This highlights the importance of considering the prevalence of comorbidities occurring together more often than would be expected to occur randomly together. Out of a possible 171 (19 choose 2) possible pairs of diseases, 158 of the comorbid pairs were reported by the female participants in this study. Of those disease pairs, 36 occurred together more often than would be expected due to randomness and disease prevalence (standardised lift ≥ 0.5). The disease pairs with the 10 highest standardised lift values for the female sample are presented in Table 2. Although stomach ulcers and obesity are only reported by 3.14% of the female sample, this comorbid pair is the most likely to occur more often than due to random chance.

Table 2.

Comorbidities reported by females which are most likely to occur together more often than due to random chance and disease prevalence (the 10 highest standardised lift values).

Health conditions	Prevalence (%)	Standardised lift
Stomach ulcers, obesity	3.14	0.84
Other eye conditions, cataracts	4.71	0.81
Thyroid, urinary incontinence	14.66	0.66
Hypertension, high cholesterol	15.71	0.65
Diabetes, obesity	8.38	0.58
Diabetes, high cholesterol	8.38	0.58
Depression, urinary incontinence	6.81	0.58
Obesity, diabetes	8.38	0.58
Obesity, arthritis	9.42	0.57
Obesity, hypertension	13.61	0.55

Out of a possible 171 (19 choose 2) possible pairs of diseases, 137 of the comorbid pairs were reported by the male participants in this study. Of those disease pairs, 24 comorbidities occurred together more often than would be expected due to randomness and disease prevalence (standardised lift ≥ 0.5). The disease pairs with the 10 highest standardised lift values for the male sample are presented in Table 3. Although cancer is quite a low prevalence condition in the male sample (5.04%), it occurred with urinary incontinence as the most likely comorbidity to occur more often than due to random chance as well as in 3 of the 10 most likely comorbidities to occur more often than due to chance (with hypertension and high cholesterol).

Table 3.

Comorbidities reported by males which are most likely to occur together more often than due to random chance and disease prevalence (the 10 highest standardised lift values).

Health conditions	Prevalence (%)	Standardised lift
Cancer, urinary incontinence	4.20	0.83
Heart murmur, high cholesterol	5.88	0.69
Diabetes, hypertension	3.36	0.65
Cancer, hypertension	3.36	0.65
Cancer, high cholesterol	3.36	0.65
Other heart conditions, urinary incontinence	5.88	0.62
Heart arrhythmia, heart murmur	5.04	0.59
Heart arrhythmia, high cholesterol	5.88	0.57
Depression, urinary incontinence	5.88	0.57
Other eye conditions, cataracts	2.52	0.57

Confidence measures - comorbidity in relation to an index condition

Figures 4 and 5 displays heatmaps of probability values corresponding to the ‘confidence’ measure of an association rule; the probability that the health condition marked on the x-axis occurs given that health condition indicated on the y-axis is already present. Looking at the squares coloured peach / orange, some examples of note include that the probability of a woman being urinary incontinent given that they identify depressive symptoms is 0.59 and the probability of a man being urinary incontinent given that they have reported cancer is 0.83. At the other end of the scale (green squares), given that a participant has obesity or high cholesterol, the probability of having many of the conditions is close to 0 for both males and females.

Figure 4.

Heatmap of confidence values for females in the IDS-TILDA sample. The values depicted are the estimated probability of having the condition denoted on the x-axis given that you have the condition on the y-axis.

Figure 5.

Heatmap of confidence values for males in the IDS-TILDA sample. The values depicted are the estimated probability of having the condition denoted on the x-axis given that you have the condition on the y-axis.

Health condition triads

The most prevalent health condition triads reported by women and men in this sample are presented in Table 4 and Table 5 respectively. Many of the disease triads experienced by over 5% of the female IDS-TILDA sample, were also reported by over 5% of the TILDA sample apart from those including thyroid conditions and the high cholesterol/obesity/urinary incontinence triad, which were experienced by over 5% of the IDS-TILDA sample but less than 5% of the TILDA sample. Only one triad is experienced by more than 5% of males in the IDS-TILDA sample and this is experienced by less than 5% of the TILDA sample. More than twice the number of disease triads were reported by the TILDA sample, indicating more heterogeneity in the multimorbidity patterns of the IDS-TILDA participants.

Table 4.

Prevalence of health condition triads reported by more than 5% of the females in the IDS-TILDA sample.

Health conditions	Prevalence (%)
High cholesterol, hypertension, obesity	9.95
High cholesterol, obesity, urinary incontinence	7.85
High cholesterol, hypertension, urinary incontinence	7.33
Arthritis, high cholesterol, obesity	6.81
Arthritis, high cholesterol, hypertension	5.76
Osteoporosis, thyroid, urinary incontinence	5.76
Poor vision, thyroid, urinary incontinence	5.24

Table 5.

Prevalence of health condition triads reported by more than 5% of the males in the IDS-TILDA sample.

Health conditions	Prevalence (%)
High cholesterol, obesity, urinary incontinence	5.04

Clusters of health conditions

Clusters of health conditions were determined using the fast greedy search approach used in the general population study. Two clusters were identified in the female network of multimorbidities;

• Heart murmur, Heart arrhythmia, Other heart conditions, Cataracts, Other eye conditions, Osteoporosis, Cancer, Thyroid, Depression, Poor vision and Urinary incontinence

• Hypertension, Diabetes, High cholesterol, Lung disease, Arthritis, Stomach ulcers, Poor hearing, Obesity

Three clusters were identified in the male network as follows;

• Heart murmur, Heart arrhythmia, Lung disease, Arthritis, Depression

• Other heart conditions, Cataracts, Other eye conditions, Cancer, Stomach ulcers, Thyroid

• Hypertension, Diabetes, High cholesterol, Osteoporosis, Poor hearing, Poor vision, Obesity, Urinary incontinence

The diseases that clustered together as identified in the TILDA sample were more homogeneous e.g. cardiovascular/vision vs other in the male TILDA sample, again pointing to more heterogeneity in the IDS-TILDA sample compared with the TILDA sample.

Discussion

Multimorbidity is the norm for adult ageing with and without intellectual disability. Patterns of multimorbidity have previously been identified in the IDS-TILDA sample, but not to the level of granularity of health conditions presented herein. One novelty of the analysis presented here is this closer look, but more importantly the direct comparison to the general population enabled by this granularity. Although the overall multimorbidity picture is one of poor health and this is similar for both populations, this novel comparative network analysis identifies some key differences. Overall, the multimorbidity patterns for those ageing with intellectual disability are more heterogeneous than those without intellectual disability. The clusters of diseases identified in those with intellectual disability are not as ‘clear cut’ as in the general population. This supports the need for a personalized treatment and approach to healthcare provision. The influential disease nodes in the networks are also somewhat different for the two samples, with hypertension dominating the multimorbidity network in those without but not in those with intellectual disability. On the other hand, urinary incontinence is prevalent in the networks of both.

With regards to service planning, this article provides further evidence of complex healthcare needs of people ageing with intellectual disability. Adequate service provision is imperative to ensure that people with intellectual disability are ageing with access to appropriate care for their needs. An advantage of the network analysis approach presented is in providing information to clinicians regarding not only the prevalence of comorbid pairs, but also how likely a pair will occur more often than due to chance. Due to this nuanced approach, using the standardised lift, comorbid pairs of interest can be identified that may otherwise have been overlooked. For example, stomach ulcers have a relatively low prevalence here in the female sample (3.66%), but along with obesity, they are the most likely comorbid pair to occur more often than due to chance (standardized lift 0.84). This has been identified previously in an obesity specific analysis by Ryan et al. (2021) but other comorbid pairs of interest may not have been previously identified. Similarly, cancer has a relatively low prevalence in the male sample (5.05%), but its comorbidity with urinary incontinence is the most likely to occur more often than due to chance. Measures of ‘confidence’ provided may be considered an indication of the measure of the classical comorbidity definition; having a health condition in reference to an index disease (Feinstein, 1970). This provides further useful information for clinicians and healthcare workers. For example, one recommendation from this work is that a woman with intellectual disability who is presenting with depressive symptoms may benefit from attending a pelvic floor physiotherapist to address urinary incontinence (confidence 0.59). Similarly, a male with intellectual disability who is undergoing cancer treatment may also benefit from advice on the management and treatment of urinary incontinence (confidence 0.83).

Males with intellectual disability appear to be reporting less comorbidity than the general population. A more complex comorbidity pattern (based on prevalence of reported comorbid pairs) can be seen in the general population than in IDS-TILDA although this this could be a feature of TILDA’s larger sample size giving it more power to detect these complex patterns.

Men ageing with intellectual disability are also reporting less comorbidity than their female counterparts. Women in the IDS-TILDA sample are reporting more multimorbidity than men and higher prevalence of the majority of health conditions, comorbid pairs and triads of diseases. We posit that a larger sample of people ageing with intellectual disability would find further statistically significant differences between the sexes in many of the health conditions.

A strong recommendation of this article is a focus on women’s health and education. Although not a focus of this article, when considering women ageing with intellectual disability, the menopause is relevant. In the most recent Wave 5 report, just 41.6% of menopausal women in IDS-TILDA reported symptoms of the menopause, whereas women in the general population report upwards of 90%. One does not expect that women with intellectual disability are not experiencing menopause but that they are not aware of or identifying with menopause. In fact, only 48.8% of women in Wave 5 were aware of menopausal transition; only 37.9% of women said they discussed the menopause with someone and only 16.2% received easy read materials on the topic (Burke et al., 2023). Early menopause is associated with increased health concerns in all ageing persons (Shuster et al., 2010) and in those ageing with intellectual disability, early menopause has been identified as a cause for premature mortality (Robertson et al., 2021). Furthermore, for those ageing with Down Syndrome, early menopause has been linked with early onset dementia (Schupf et al., 1997, 2003, 2006, 2018) and the early age at menopause has been identified as a risk factor for post-menopausal health disorders Seltzer et al. (2001). Thus, it is important for society, services and family to address this issue and provide education and support in order to provide a more positive experience of ageing and better health for women ageing with intellectual disability.

Given the high levels of multimorbidity and associated polypharmacy in persons ageing with intellectual disability, we recommend that multimorbid persons with intellectual disability should have a regular review of the appropriateness of their medication use, including prescription and non-prescription drugs as the high levels of multimorbidity demonstrated here and the resulting polypharmacy places people ageing with intellectual disability at risk of adverse effects including feeling sleepy, falls, constipation and less energy (O’Dwyer et al., 2016). The potential consequences of drug-drug interactions while treating symptoms of different health conditions should be monitored for multimorbid persons with intellectual disability.

There are limitations to this work. Even though this network analysis is a comparative evaluation, it is based on the health conditions listed in the TILDA general population study. This is not a definitive or suitable list of health conditions or pairings of health conditions for people ageing with intellectual disability as it does not include mental health and neurological conditions which are highly prevalent. Given the high levels of neurological conditions such as epilepsy and dementia in this population (36.2% (McCarron et al., 2013)) as well as the high levels of mental health conditions (48.3% (McCarron et al., 2013)), a strong recommendation of the authors is to repeat this network analysis including both mental health and neurological conditions and with the disease list presented in (McCarron et al., 2013), in order to provide a more accurate and representative network analyses of multimorbidity for people with intellectual disability.

The network of multimorbidities along with the disease prevalences lend itself to the statistical modelling of weighted networks. This would allow covariates such as demographics to be included in the network models, along with the edges and node sizes modelled comprehensively. Moreover, the probabilistic clustering of same could provide further insight. This presents a challenge to the statistics methodology community driven by a real-world application. With advanced statistical modelling, clinicians would have a measure of uncertainty about which diseases cluster with which and this could guide examination, testing for diseases of high probability of a comorbidity in relation to the diseases already identified. Another opportunity presents itself given that five waves of data now exist at IDS-TILDA; this data could be harnessed further by analysing this multimorbidity network longitudinally and its relationship to mortality. This may even provide evidence for causality between health conditions over time. Again, this presents challenge and opportunity for meaningful transdisciplinary collaboration between statistics and intellectual disability researchers and a potential improvement in evidence-based healthcare for people ageing with intellectual disability.