Diabetes multimorbidity combinations and disability in the Mexican Health and Aging Study, 2012-2015

Abstract

Purpose:

The aim of this study was to investigate the relationship between specific combinations of chronic conditions and disability in Mexican older adults with diabetes.

Methods:

This was a prospective cohort study of Mexican adults (n=2558) with diabetes and aged 51 or older that used data from the 2012 and 2015 waves of the Mexican Health and Aging Study. The main outcome was an index that measured ability to perform activities of daily living and instrumental activities of daily living. The main independent variables were diabetes multimorbidity combinations, defined as diabetes and at least one other chronic condition. The authors calculated the prevalence of each multimorbidity combination present in the sample in 2012 and used negative binomial regression models to estimate the association of the most prevalent of these combinations with disability incidence in 2015.

Results:

The three most prevalent combinations were: 1) diabetes-hypertension (n=637, 31.9%) 2) diabetes-hypertension-depression (n=388, 19.4%) and 3) diabetes-depression (n=211, 10.6%). In fully adjusted models comparing participants with specific multimorbidity combinations to participants with diabetes alone, the combinations that had an increased association with disability were diabetes-hypertension-depression, diabetes-depression and diabetes-hypertension-arthritis-depression. In nested models, the addition of arthritis to combinations including depression increased this association.

Conclusions:

Consistent with prior studies, multimorbidity combinations including depression were associated with increased risk of disability. However, the effect size of this relationship was lower than what had been previously been reported internationally. This highlights the need for globally oriented multimorbidity research.

1. Introduction

The global burden of diabetes and other chronic conditions is rapidly increasing, and most people with diabetes have at least one other chronic condition (Bennett et al., 2018; Magnan, Bolt, Greenlee, Fink & Smith, 2018; Nowakowska et al., 2019; Shaw, Sicree & Zimmet, 2010). An accepted definition of multimorbidity is “two or more chronic conditions that are expected to last at least 12 months and that confer a significant risk of death, decompensation, or functional decline.” (Dattalo et al., 2017, p. 713). High rates of multimorbidity globally underscore the need for a better understanding of the interactions that occur between chronic conditions, including diabetes, as well as the ways in which these interactions vary across populations (Garin et al., 2016; Hajat & Stein, 2018; Academy of Medical Sciences [AMS], 2018). Worldwide, the age group most affected by these trends is older adults (Marengoni et al., 2011; Saeedi et al., 2019). Diabetes multimorbidity in older adults has complex clinical and public health consequences, and there is a recognized need for more research to characterize its epidemiologic profile (Kirkman et al., 2012; Piette & Kerr, 2006).

One important area for further investigation is the relationship between multimorbidity and disability. A significant association exists between diabetes and disability (Wong et al, 2013), but the importance of multimorbidity as a mediator in this relationship is not completely understood. Prior studies have shown that controlling for individual comorbid conditions attenuates the relationship between diabetes and disability (Bruce, Davis, & Davis, 2005; Kalyani, Saudek, Brancati, & Selvin, E 2010; Volpato et al., 2002), and there is also evidence that the total number of concurrent chronic conditions in individuals with diabetes may be associated with disability (Gray et al., 2016). One recent study examined the relationship between disability and specific multimorbidity combinations in persons with diabetes (Quiñones, Markwardt & Botoseneanu, 2019). The authors found that persons with diabetes multimorbidity combinations that included depression or stroke were at higher risk for disability than persons with diabetes alone or with diabetes multimorbidity combinations that included only somatic conditions. Other studies focused on multimorbidity in the general population have identified a similar association between disability and multimorbidity combinations that include neurologic or psychiatric conditions (Jackson et al., 2015; Quiñones, Markwardt & Botoseneanu, 2016; Quiñones et al., 2018; Rivera-Almaraz et al., 2018; Vetrano et al., 2018).

More research is needed to clarify the relationship between diabetes multimorbidity combinations and disability and explore how this relationship might vary across populations, especially in low- and middle-income countries where the prevalence of diabetes and multimorbidity is rapidly increasing (AMS, 2018). Studying this relationship across diverse populations will provide more points of comparison to enhance our understanding of diabetes multimorbidity generally and could also help tailor interventions to specific contexts. One country where such research is relevant is Mexico, where high rates of diabetes and other chronic conditions have a strong impact on the national burden of disability (Meza et al., 2015; Parra-Rodríguez et al., 2019). Investigators have previously tested the association between broad multimorbidity patterns and disability in the general Mexican population (Rivera-Almaraz et al., 2018), but to the authors’ knowledge no prior studies have examined the relationship between specific multimorbidity combinations and disability in persons with diabetes in Mexico.

Therefore, the aim of this study was to investigate the relationship between specific combinations of chronic conditions and functional status in older Mexican adults with diabetes. Chronic conditions considered were arthritis, hypertension, cardiac disease (consisting of heart attack or heart failure), stroke, diabetes, cancer, lung disease, depression and neurocognitive impairment. Authors calculated the prevalence of all combinations of these conditions present in the population and selected the most prevalent combinations for analysis.

2. Materials and Methods

2.1. Study Design

This study was a longitudinal secondary analysis of data from the Mexican Health and Aging Study (MHAS).

2.2. Data

Complete information regarding MHAS is available elsewhere (Wong, Michaels-Obregon, & Palloni, 2017). Briefly, MHAS comprises a cohort of 50-years or older Mexican individuals and their spouses, with a population-based sample. Currently the survey has five assessments, the first in 2001, and the most recent in 2018. The survey is composed of a set of questionnaires that thoroughly investigate social and health topics. For sub-samples of respondents, anthropometric measurements and biomarkers were collected and physical performance tests were performed. Beginning in 2003, assessments include information from those individuals who died at follow-up, gathered from next-of-kin. The present study used only data collected during 2012 and 2015; two of the survey’s five waves.

2.3. Population

Interviews were completed in both the 2012 wave and the 2015 wave for 13,628 respondents, including spouses. Authors excluded 1,378 respondents who did not participate in direct interviews in both waves as functionality questions were only asked during direct interviews and not during proxy interviews. Author additionally excluded 938 respondents who were less than 51 years of age in 2012 and 8,754 respondents who did not have a diagnosis of diabetes in 2012. The final sample included 2558 respondents, of which 1997 had a diagnosis of diabetes and at least one other chronic condition. The remaining 561 respondents had a diagnosis of diabetes only.

2.4. Variables

2.4.1. Functional status – Dependent Variable

The dependent variable was a disability index tabulated from respondents’ answers to questions about activities of daily living (ADLs) and instrumental activities of daily living (IADLs) in the 2015 wave of the survey. During data analysis, the authors also considered baseline disability from the 2012 wave of the survey and incorporated this into fully adjusted regression models. All participants were asked if they had difficulty dressing themselves. Respondents who replied “no” were asked no further ADL questions. Those who replied “yes” were also asked if they had difficulty walking across a room, bathing, eating, transferring or using the toilet (six ADL questions in total). Additionally, to assess difficulty with IADLs all respondents were asked if they had difficulty preparing a hot meal, grocery shopping, taking medications or managing money (4 IADL questions in total). Respondents were assigned a point for every yes answer and the total number of points from both ADL and IADL questions were added together to create a composite disability index with a range from zero to ten. This method of measuring disability has been validated in prior studies (Liang et al., 2008; Spector & Fleishman, 1998; Thomas, Rockwood & McDowell, 1998).

2.4.2. Multimorbidity – Independent Variable

Respondents with a diagnosis of at least one other chronic condition in addition to diabetes were included in the diabetes multimorbidity group, in agreement with accepted definitions of multimorbidity ((Dattalo et al., 2017). These diagnoses were used to tabulate comorbidity combinations as described below in the data analysis section.

2.4.3. Somatic conditions

Survey respondents were asked if a physician had diagnosed them with any of seven somatic chronic conditions: arthritis, hypertension, cardiac disease (consisting of heart attack or heart failure), stroke, diabetes, cancer and lung disease. Respondents who replied “yes” to any of these questions were assigned a diagnosis of that condition.

2.4.4. Depressive symptoms

Depressive symptoms were tabulated using a depression scale included in the survey composed of nine yes or no questions. The scale was based on the depression scale used in the Health and Retirement Survey (HRS) and was modified for use with Spanish speaking populations (Steffick, 2000). A depression score ranging from one to nine was tabulated and participants with a score of five or higher were defined as having high depressive symptoms based on a prior validation study (Aguilar-Navarro et al., 2007)

2.4.5. Neuro-cognitive impairment

Neurocognitive symptoms were evaluated using an instrument built into the survey that was based on the Cross-Cultural Cognitive Examination (Glosser et al., 1993). As part of all in-person interviews participants were asked to complete a series of tasks evaluating different cognitive domains. For the present study, scores for the verbal learning, visuospatial memory, verbal fluency, visual memory and verbal recall domains were considered and previously validated cutoffs based on age, years of education and sex were used to identify respondents with possible neurocognitive impairment (Mejia-Arango & Gutierrez, 2011).

2.5. Covariates

The demographic variables of age, sex, body mass index (BMI), and years education from the 2012 survey wave were used as covariates in partially adjusted models. BMI was calculated using self-reported weight and height. In fully adjusted models, a baseline disability score was also included as a covariate. This baseline score was calculated as above but using responses from the 2012 wave of the survey.

2.6. Analysis

Baseline characteristics of respondents in the multimorbidity group and the diabetes only control group were described using frequencies and means.

2.6.1. Multimorbidity combination prevalence

Data analysis was conducted in two phases. In the first phase, comorbidity combinations were assigned to each participant and the prevalence of each combination in the study sample was calculated. Combinations were assigned by creating a variable that included all chronic conditions diagnoses a participant had been assigned. Prevalence was calculated using the number of participants in the study sample who had been assigned each specific combination of chronic conditions. All specific combinations with a prevalence above a cutoff of 1.5% were selected for further analysis. This a priori cutoff was chosen to ensure sufficient sample size for regression analysis.

2.6.2. Multimorbidity combinations and functional status

Sequentially adjusted negative binomial regression models were used to study the association between specific comorbidity combinations and functional status. Unadjusted, partially adjusted and fully adjusted models were constructed comparing the risk of disability associated with each of the most prevalent multimorbidity combinations previously identified with the risk of disability associated with the diabetes only control group. The independent variable in each model was dichotomous with two possible values: a diagnosis of diabetes only and no comorbidities (the diabetes-only comparison group) or the presence of one of the diabetes comorbidity combinations. The dependent variable was the combined ADL-IADL disability score. Subsequently, nested models were constructed to compare risk of disability between selected multimorbidity combinations to assess for possible synergistic interactions between conditions. Exponentiated coefficients were estimated for each model. These should be interpreted as the incident rate ratio for the multimorbidity group compared to the diabetes comparison group. Coefficients were tested for statistical significance using an α-level of 0.05. All analyses for this study was conducted using SAS software, Version 9.4 of the SAS System for Windows (SAS Institute Inc., Cary, NC, USA).

3. Results

The diabetes multimorbidity group included 1997 participants and the diabetes only group included 561 participants. Baseline data from 2012 is shown in Table 1. The most notable demographic difference between the two groups was the higher proportion of females in the diabetes only group.: 51.2% (n=287) compared to 32.3% (n=645) The mean age and BMI was higher in the multimorbidity group while the mean years of education was lower. Mean 2012 baseline disability scores were higher in the multimorbidity group for the ADL, IADL and combined ADL-IADL indices. More respondents in the diabetes only control group had a disability index of zero, although the proportion of respondents with a baseline disability index score of zero was high in both groups: 77.9% (n=1428) for the multimorbidity group and 94.0% (n=481) for the diabetes only group.

Table 1.

Baseline Characteristics of the Study Population, Mexican Health and Aging Survey 2012–2015

Variables	Diabetes multimorbidity group (n=1997)	Diabetes only group (n=561)
Female, n (%)	645 (32.3)	287 (51.2)
Years of education, mean (SD)	5.0 (4.2)	6.2 (4.7)
Age in 2012, mean (SD)	64.9 (7.9)	63.2 (7.6)
Body mass index, mean (SD)	30.0 (5.4)	27.2 (4.6)
Chronic diseases, n (%)
Arthritis	383 (19.2)	0 (0.0)
Hypertension	1581 (79.2)	0 (0.0)
Cardiac disease	137 (6.9)	0 (0.0)
Stroke	79 (4.0)	0 (0.0)
Diabetes	1997 (100)	561 (100.00)
Cancer	74 (3.7)	0 (0.0)
Lung disease	150 (7.5)	0 (0.0)
High depressive symptoms	988 (49.5)	0 (0.0)
Neuro-cognitive impairment	120 (6.0)	0 (0.0)
Number of conditions, mean (SD)	2.8 (1.5)	1.0 (0.0)
2012 ADL* index, mean (SD)	0.3 (1.0)	0.1 (0.4)
2012 ADL* index, n (%)
0	1696 (85.6)	542 (96.6)
1	108 (5.5)	10 (1.8)
2	69 (3.5)	3 (0.5)
3	46 (2.3)	3 (0.5)
4	32 (1.6)	2 (0.4)
5	19 (1.0)	1 (0.2)
6	11 (0.6)	0 (0.0)
2012 IADL index, mean (SD)	0.1 (0.3)	0.1 (0.3)
2012 IADL index, n (%)
0	1575 (85.4)	488 (95.31)
1	172 (9.3)	21 (4.10)
2	61 (3.3)	0 (0.0)
3	21 (1.1)	2 (0.4)
4	16 (0.9)	1 (0.2)
2012 ADL-IADL index, mean (SD)	0.5 (1.3)	0.1 (0.6)
2012 ADL-IADL index, n (%)
0	1428 (77.9)	481 (94.0)
1	183 (10.0)	20 (3.91)
2	93 (5.0)	3 (0.6)
3	49 (2.7)	3 (0.6)
4	30 (1.6)	2 (0.4)
5	15 (0.8)	2 (0.4)
6	14 (0.8)	0 (0.0)
7	9 (0.5)	0 (0.0)
8	5 (0.2)	1 (0.2)
9	4 (0.2)	0 (0.0)
10	4 (0.2)	0 (0.0)

Note. ADL = Activities of Daily Living, IADL = Instrumental Activities of Daily Living

All diabetes multimorbidity combinations with a prevalence greater than 1.5% are shown in Table 2. The three most prevalent multimorbidity combinations are 1) diabetes-hypertension (n= 637, 31.9%); 2) diabetes-hypertension-depression (n = 388, 19.4%); 3) diabetes-depression (n = 211, 10.6%). Of the nine multimorbidity combinations with a prevalence greater than 1.5%, four include depression and have a high combined prevalence (n = 742, 37.2%). The three combinations with the highest ADL-IADL index scores in 2015 all included depression. The diabetes only group had the lowest mean 2015 ADL-IADL index score (0.23, STD 0.85).

Table 2.

Prevalent Multimorbidity Combinations (2012) and Mean ADL-IADL (2015).

Group Rank	n (%)	Diabetes	Arthritis	Hypertension	Cardiac Disease	High Depressive Symptoms	Neuro-cognitive Impairment	Mean (SD) ADL-IADL
1	637 (31.9)	X		X				0.46 (1.36)
N/A+	561	X						0.23 (0.85)
2	388 (19.4)	X		X		X		0.88 (1.67)
3	211 (10.6)	X				X		0.54 (1.30)
4	110 (5.5)	X	X	X		X		1.78 (2.52)
5	79 (4.0)	X	X	X				0.72 (1.43)
6	43 (2.2)	X	X					0.76 (1.33)
7	35 (1.8)	X		X	X			0.32 (1.62)
8	33 (1.7)	X	X			X		0.87 (1.69)
9	33 (1.7)	X		X			X	0.32 (1.71)

Note. ADL = Activities of Daily Living, IADL = Instrumental Activities of Daily Living

^*Prevalence calculated using diabetes multimorbidity group (n=1997) as denominator.

⁺Diabetes only group included for comparison.

Table 3 compares combined ADL-IADL scores of study participants with diabetes multimorbidity combinations to participants with just diabetes and no other chronic conditions. All combinations except diabetes-hypertension-cardiac had an association with disability in the unadjusted model. In the partially adjusted model that controlled for age, sex, years education and BMI three more combinations no longer have an association with disability: diabetes-hypertension-arthritis, diabetes-arthritis, diabetes-hypertension-NCI. In the fully adjusted model that also includes 2012 baseline ADL-IADL index scores, only three multimorbidity combinations continue to have a significant association with higher ADL-IADL scores: diabetes-hypertension-depression (IRR 2.44, CI 1.65–3.60), diabetes-depression (IRR 2.37, CI 1.34–4.21) and diabetes-hypertension-arthritis-depression (IRR 3.74, CI 2.08–6.73). All three of these combinations include depression. Additionally, the diabetes-hypertension-cardiac combination was found to have a negative association with ADL-IADL score in the final model (IRR .05, CI 0.00–0.91).

Table 3.

Negative Binomial Regression of ADL-IADL score on Multimorbidity Combinations Compared with Individuals with Only Diabetes

Multimorbidity Combination		Unadjusted model				Partially adjusted model+				Fully adjusted model+
		e	95% CI	P		e	95%CI	P		e	95% CI	P
DM + HTN		1.96	1.30–2.97	.001**		2.17	1.40–3.38	.001**		1.54	1.00–2.38	0.053
DM + HTN + DEP		3.75	2.62–5.37	<.001**		3.83	2.61–5.61	<.001**		2.44	1.65–3.60	<.001**
DM + DEP		2.32	1.39–3.87	.001**		2.69	1.57–4.61	<.001**		2.37	1.34–4.21	.003**
DM + HTN + ART + DEP		7.59	4.53–12.73	<.001**		4.72	2.72–8.19	<.001**		3.74	2.08–6.73	<.001**
DM + HTN + ART		3.10	1.38–6.99	.006**		1.50	0.67–3.34	.320		1.16	0.46–2.66	.805
DM + ART		3.22	1.19–8.76	.021*		2.34	0.92–5.97	.075		2.23	0.87–5.65	.092
DM + HTN + CAR		1.38	0.34–5.57	.654		0.23	0.04–1.27	.093		0.05	0.00– 0.91	.043*
DM + ART + DEP		3.71	1.17–11.76	<.001**		3.53	1.22–10.21	<.001**		2.18	0.68–6.96	.189
DM + HTN + NCI		2.23	0.51–9.55	<.001**		0.88	0.22–3.50	.851		0.82	0.20–3.30	.776

Note. ART = arthritis, CAR = cardiac disease, DM = diabetes mellitus, DEP = depression, HTN = hypertension, NCI = neurocognitive impairment, e = exponentiated coefficient

^*Statistically significant with α-level of 0.05

^**Statistically significant with α-level of 0.01

⁺Partially adjusted model includes age, sex, years education and BMI covariates. Fully adjusted model includes these as well as 2012 baseline ADL-IADL score as an additional covariate.

Table 4 shows nested models comparing respondents with a multimorbidity combination including depression to respondents with the same multimorbidity combination with the addition of one or more somatic conditions. The addition of arthritis and hypertension to the diabetes-depression combination and arthritis to the diabetes-hypertension-depression combination both had a significant effect on disability (IRR 2.43 CI 1.40–4.22 and IRR 1.71 CI 1.32–2.57). However, the addition of arthritis alone to the diabetes-depression combination did not have an effect. There was no increase in risk of disability when hypertension alone was added to either diabetes-depression or diabetes-arthritis-depression.

Table 4.

Negative Binomial Regression of ADL-IADL score on Nested Combinations

Diabetes Multimorbidity Group Comparison		Unadjusted model				Partially adjusted model+				Fully adjusted model+
		e	95% CI	P		e	95%CI	P		E	95% CI	P
DM + DEP v DM + HTN +DEP		1.62	1.10–2.38	.014*		1.71	1.14–2.54	.001**		1.37	0.91–2.08	.131
DM + DEP v DM + ART +DEP		1.60	0.68–3.78	.282		1.66	0.70–3.95	.250		1.58	0.61–4.11	0.344
DM + DEP v DM + HTN + ART + DEP		3.27	2.06–5.21	<.001**		2.46	1.47–4.11	.001**		2.43	1.40–4.22	.002**
DM + HTN + DEP v DM + HTN + ART + DEP		2.02	1.36–3.02	.001**		1.66	1.08–2.54	.020*		1.71	1.32–2.57	.012*
DM + ART + DEP v DM + HTN + ART + DEP		2.04	1.01–4.13	.046*		1.42	0.67–2.98	0.352		1.27	0.59–2.70	.543

Note. ART = arthritis, CAR = cardiac disease, DM = diabetes mellitus, DEP = depression, HTN = hypertension, NCI = neurocognitive impairment, e = exponentiated coefficient

^*Statistically significant with α-level of 0.05

^**Statistically significant with α-level of 0.01

⁺Partially adjusted model includes age, sex, years education and BMI covariates. Fully adjusted model includes these as well as 2012 baseline ADL-IADL score as an additional covariate.

4. Discussion

These findings suggest that in Mexican older adults with diabetes, some multimorbidity combinations are associated with an increased risk of disability while others are not. Multimorbidity combinations that included depression appeared to be the only consistent predictors of disability, although in some models arthritis and depression together may have had a synergistic effect. Additionally, the authors observed a high prevalence of multimorbidity combinations that included depression.

4.1. Association between diabetes multimorbidity combinations and disability

In models comparing respondents with diabetes multimorbidity to respondents with diabetes alone, only diabetes multimorbidity combinations that included depression were associated with a higher prospective risk of disability. Those including somatic conditions alone were not. In direct comparisons of combinations containing depression to combinations containing only somatic conditions (depression vs somatic), two of three models predicted a higher risk of disability for respondents with combinations including depression. The third model had the smallest sample size and may not have been sufficiently powered to detect a difference in risk. In three out of five nested models (depression-somatic vs depression-somatic plus additional somatic condition), the addition of another somatic condition to a combination which included depression did not increase risk of disability, consistent with other findings indicating the importance of depression as an indicator of risk of disability. The elevated risk observed in the other two models may be the result of synergistic interactions with depression as discussed below.

These results reflect other recent studies indicating that some combinations of chronic conditions are more associated with disability than others and that the presence of neuropsychiatric conditions in patients with multimorbidity can be a predictor of disability. Most were conducted in high-income countries (Jackson et al., 2015; Quiñones, et al., 2018; Quiñones et al., 2016; Vetrano et al., 2018)., however several studies in low-and-middle income countries had relevant findings. One assessed a pooled sample including respondents from high income and low-and-middle income countries and found that ADL impairment was more likely in a respiratory-articular-mental class than in a cardio-metabolic class (Bayes-Marin et al., 2020). Similarly, another study in Mexico showed an association between disability and a multimorbidity cluster composed of depression and arthritis (Rivera-Almaraz et al., 2018). Investigators examining a population in rural Burkina Faso found that multimorbidity combinations including mental health conditions had a greater association with disability that those including somatic conditions alone. Finally, in a study using data from a group of low-and-middle income countries that did not include mental conditions and only considered dyads of somatic condition, the only diabetes dyad found to have an association with ADL impairment was diabetes-hypertension (Arokiasamy et al., 2015). This association was barely significant (OR 1.25; CI 0.98–1.57), and in this regard similar to the risk of disability associated with diabetes-hypertension in our study, which was the somatic combination that came closest to having a positive relationship with disability in the fully adjusted model and only narrowly failed to do so.

Differences in methods for selecting and grouping multimorbidity combinations limited more precise comparison between the present study and others. Additionally, the present study included only persons with diabetes whereas most included all persons with multimorbidity. However, one study was similar enough to allow for further comparison (Quiñones et al., 2019). This study employed the same design as the present study and used data from the HRS to examine the association between multimorbidity combinations and disability in community dwelling older with diabetes in the United States. The authors also found an association between disability risk and multimorbidity combinations including depression, but the magnitude of this association was larger than that observed in the present work. In the United States cohort, the exponentiated coefficients of significant, fully adjusted regression models comparing respondents with multimorbidity combinations including depression to respondents with diabetes alone ranged from 3.99 to 18.15. This compares to a range of 2.37 to 3.74 for Mexico.

In two of the nested models, the addition of arthritis to a disability model including depression increased the risk of disability even further. Interestingly, in models comparing respondents with multimorbidity combinations that included arthritis but not depression to respondents with only diabetes, there was no difference in risk of disability. This raises the possibility that depression and arthritis together might have a synergistic effect on risk of disability. Other authors studying multimorbidity have observed synergistic relationships between conditions (Hunger, 2011; Marventano, 2014). Synergistic interactions have implications for the way multimorbidity is conceptualized, complicating efforts to measure multimorbidity with indices and reorienting its study towards approaches that take specific multimorbidity combinations as the fundamental unit of analysis.

In direct comparisons to respondents with only diabetes, the only multimorbidity combination found to have an association with incident disability that did not include depression was diabetes-hypertension-cardiac. This combination was marginally associated with a decreased risk of disability (p=0.43). There is a possibility that this represents the adoption of protective behaviors or is a survivorship effect reflecting the death of respondents with severe cardiac disease.

4.2. Prevalence of diabetes multimorbidity combinations

Our study is the first to report on the prevalence of multimorbidity combinations in older adults with diabetes in Mexico. Two recent studies providing estimates of the prevalence of multimorbidity groupings in populations of Mexican older adults did not focus specifically on persons with diabetes (Dolores et al., 2017; Rivera-Almaraz et al., 2018). Additionally, their methods generated clusters that cannot be easily compared to those of the present study. Other studies from other low-and-middle income countries characterizing multimorbidity groupings and their prevalence also used cluster methods that present similar challenges for comparison (Craig et al., 2020; Zahra et al., 2020). Those that do allow for comparison show that combinations including hypertension or arthritis tend to have the highest prevalence, a finding reflected in our study (Garin et al., 2016; Wang et al., 2020).

In comparison to older adults with diabetes in the United States, one notable difference is the prevalence of multimorbidity combinations that include depression. Both the present study and the United States study discussed above examined multimorbidity combinations that had a prevalence of >1.5% in a sample of older adults with diabetes. In the present study, the combined prevalence of all analyzed multimorbidity combinations that included depression was 37.2% compared to 10.5% in the United States group.

4.3. Multimorbidity combinations that include depression

One possible explanation for this difference in prevalence is that depression was over reported in the present study. This would also explain the weaker association observed between disability and multimorbidity combinations including depression in comparison to the United States study discussed above. However, the depression scale used in MHAS is very similar to the abbreviated version of the CES-D used in HRS and was validated in Mexican older adults with comparable sensitivity and specificity (Aguilar-Navarro et al., 2007; Steffick, 2000). Additionally, other studies of Mexican diabetic populations have reported a high prevalence of depression ranging from 46% to 63% (Juárez-Rojop et al., 2018; Tovilla-Zárate et al., 2012) which is similar to the prevalence of 38.7% in all respondents and 49.5% in respondents with multimorbidity in the present study. Another possible methodological explanation for this difference is survivorship bias. The present study excluded respondents who died between the 2012 and 2015 waves of the survey. It could be that in the Mexican population a high proportion of older adults with depression in 2012 who would have developed worsening disability in 2015 instead died and were not included in the study.

Alternatively, one possibility is that there are social, cultural or other epidemiological differences between older adults with diabetes in Mexico and the United States that lead to differences in the prevalence and the effect of specific multimorbidity combinations in these two populations. Elsewhere, researchers have proposed several explanations for the lower risk of prospective disability observed in the general population of older adults in Mexico when compared to the United States (Gerst-Emerson, Wong, Michaels-Obregon, & Palloni, 2015). This discrepancy could reflect higher mid-century childhood mortality in Mexico so that only the “sturdiest” Mexican participants survived to older adulthood. Distinct patterns of institutionalization could lead to different rates of disability in the non-institutionalized population studied. Cultural differences in perception of disability could affect how respondents report disability. Evaluating these and other possible explanations for observed differences between these populations will require studies directly comparing both groups.

5. Strengths and Limitations

The present work makes significant contributions to the study of the epidemiology of multimorbidity and diabetes. To the authors knowledge, this study was the first to examine the prevalence of diabetes multimorbidity combinations and their association with incident disability in Mexico. At least one other study has examined the relationship between multimorbidity groupings derived from cluster analysis in the general Mexican population (Rivera-Almaraz et al., 2018), but as far as the authors know this is the first study to evaluate this relationship in a Mexican population using prevalent, specific multimorbidity combinations as the unit of analysis. Additionally, this study was designed to allow easy comparison to prior research on this topic (Quiñones et al., 2018).

One limitation of this study is that respondents who died between the 2012 and 2015 waves of the survey were excluded. This may have led to survivorship bias affecting some outcomes. Another possible source of bias was the exclusion of respondents assessed through proxy interviews. This was necessary as functionality questions were not asked during proxy interviews. However, the exclusion of these respondents may have led to underestimation of the prevalence of some diabetes multimorbidity combinations. This bias was likely most important for neurocognitive impairment as respondents with significant cognitive deficits are more likely to be unable to directly answer interview questions and to require a proxy interview. For chronic conditions other than neurocognitive impairment, the exclusion of proxy interviews may not have significantly affected results as only 1,378 respondents from an initial sample of 13,628 were excluded because of interview type.

An additional limitation is that most chronic disease diagnoses were self-reported by respondents. Depressive symptoms and neurocognitive impairment were assessed with scales administered as part of MHAS but were also not confirmed by a medical provider. About 15% of chronic conditions reported in 2012 were not reported in 2015. Given that the somatic conditions considered would be expected persist for the lifetime of respondents, this discrepancy points to potential limitations of working with self-reported diagnoses. Also, the small sample size of some multimorbidity combinations may have led to underpowered models in some instances and a failure to detect differences between groups.

Only respondents with diabetes were considered in this study, so the authors were unable to make any comparisons to respondents without diabetes. Other investigators have studied the relationship between frailty and disability in Mexican older adults (Rivera-Almaraz et al., 2018), which was not considered in the present study. Future research will have to address these limitations to allow for a better understanding of the interplaying factors affecting disability risk in this population.

6. Conclusions

The present study confirms that an important link exists between disability and diabetes multimorbidity combinations that include depression and generalizes this finding to a sample of Mexican older adults. One implication is that in Mexico, as in other populations, not all persons with diabetes and multimorbidity are at equal risk for poor health outcomes. In settings with limited resources and a high prevalence of diabetes and other chronic conditions, these results could be used to guide the development of interventions targeting high-risk subgroups such as persons with depression.

The present study also suggests that significant differences in the epidemiology of diabetes multimorbidity exist between countries. This highlights the need for a global approach to multimorbidity research. Exploring observed differences between populations could help investigators better appreciate the epidemiologic nuances of multimorbidity. Additionally, researching multimorbidity in diverse settings ensures that findings will better reflect the global population affected.

Highlights.

Multimorbidity combinations including depression are associated with disability.

Association with disability is lower in Mexico than reported internationally.

Prevalence of depression combinations is higher than reported internationally.

Global differences exist in the epidemiology of diabetes multimorbidity.