https://orcid.org/0000-0003-3881-0087
Abstract
Chronic conditions often co-occur in specific disease patterns. Certain chronic diseases contribute to incident frailty or cognitive impairment (CI), but the associations of multimorbidity patterns and the order of frailty and CI occurrence remain unclear.
To determine multimorbidity patterns amongst older adults and their associations with the order of frailty and CI occurrence.
Prospective cohort study.
Using data from National Health and Aging Trends Study, 7522 community-dwelling participants were included and followed up for four years. Latent class analysis was conducted to identify multimorbidity patterns with clinical meaningfulness. Fine and Grey competing risks models were used to examine the associations between multimorbidity patterns and different orders of frailty and CI occurrence (frailty-first, CI-first, frailty-CI co-occurrence).
Four multimorbidity patterns were identified: cardiometabolic, osteoarticular, cancer-dominated and psychiatric/multisystem pattern. Compared to non-multimorbidity, all four multimorbidity patterns were associated with a higher risk of developing frailty-first, but not developing CI-first. Specifically, the psychiatric/multisystem pattern had the highest risk of developing frailty-first ( Sub-distribution hazard ratios [SHR] = 3.74, 95% confidence intervals = 2.96, 4.71), followed by osteoarticular pattern (SHR = 2.53, 95% CI = 1.98, 3.22) and cardiometabolic pattern (SHR =2.41, 95% confidence intervals = 1.96, 2.98). In addition, only participants from psychiatric/multisystem and cardiometabolic pattern showed a higher risk of frailty-CI co-occurrence.
Our findings highlight the etiological heterogeneity between physical frailty and CI. Clinician should be aware of multimorbidity clusters and thus provide more effective strategies for comorbid older adults to prevent the onset of these two geriatric syndromes.
Key Points
Frailty and cognitive impairment might have distinct etiologies and should not be regarded as a composite outcome.
Identification of multimorbidity patterns can provide improved prognosis of frailty and/or cognitive impairment onset.
Disease characteristics should be considered when designing prevention strategies for physical and cognitive decline.
Introduction
Physical frailty is characterised by increased vulnerability to stressors due to the reduced capacity across multiple physiological systems [1]. Differently, cognitive impairment (CI) refers to a decline in an individual's cognitive function, including memory, orientation and execution. Although frailty and CI often coexist and are frequently treated as a composite outcome due to their bidirectional relationship [2–4], this approach typically overlooks their distinct temporal order. Several studies have highlighted the significant differences in the prevalence and incidence of frailty alone, CI alone and both [5–7], suggesting that these conditions might have distinct etiologies and pathways. That is, some clinical conditions may be more predictive of physical decline rather than cognitive decline, or vice versa. Understanding which one comes first is important for early prevention and intervention, and tailor the treatment plans.
Chronic conditions often co-occur in specific patterns, that is multimorbidity patterns [8]. From a gerontological perspective, the progressive accumulation of chronic conditions is one of the signs of aging, accompanied by loss of resilience and multisystem dysregulation [9]. Previous studies have investigated the disease characteristics amongst older adults with frailty and/or CI. A cross-sectional study showed that multimorbidity was strongly associated with frailty with or without CI, but not with CI alone [6]. Another longitudinal study explored the hierarchical development of frailty and CI, and found that multimorbidity was an important determinant [7]. Associations between multimorbidity patterns and incident frailty alone or incident CI alone have been previously established [10–13]. However, it remains unknown whether different multimorbidity patterns affect the order of frailty and CI occurrence. By integrating knowledge of multimorbidity patterns into health monitoring, clinical workers can identify vulnerable individuals and their varying health needs, and then tailor care plans that address specifically physical decline alone, and cognitive decline alone, or both.
Therefore, using data from National Health and Aging Trends Study (NHATS), this study aims to examine the multimorbidity patterns at baseline and investigate their associations with the order of frailty and CI occurrence. We hypothesise particular multimorbidity patterns may induce physical frailty before a clinically meaningful decline in cognitive function occurs, whilst others may follow the reverse order.
Methods
Study design and participants
NHATS is a nationally representative longitudinal cohort study of Medicare beneficiaries aged 65 and older in the United States. The first round of data collection began in 2011 and the sample was replenished in 2015. Core interviews were annually administered. To ensure a sufficient sample size, we combined participants recruited in 2011 (N = 8245) and 2015 (N = 4182). Our initial sample was community-resident or non-nursing home care participants with baseline interviews (N = 11,558). Exclusion criteria were: (i) missing data on frailty or CI at baseline; (ii) prevalent frailty or CI at baseline; (iii) missing data on three or more chronic conditions at baseline. Excluded participants tended to be older, less educated, lived more often alone and had a higher burden of comorbidities (Table S1). A total of 7522 participants were finally included (Fig. 1). Over the 4-year follow-up, incident frailty and CI were observed, and their occurrence patterns were then identified (Table S2). This study followed the STROBE reporting guideline [14].
The flow chart of sample selection and follow-up. The response rate was 70.9% in 2011 to yield 8245 interviews and the response rate in 2015 was 76.8% to yield 4182 interviews.
Primary exposure
Twelve chronic conditions at baseline were included in this study, covering somatic and mental disorders [15]. Eight chronic diseases were ascertained through self-reported diagnoses, including heart disease (e.g. heart attack and heart failure), hypertension, arthritis, osteoporosis, diabetes, lung disease, stroke and cancer. Depressive and anxiety symptoms were assessed by Patient Health Questionnaire-2 (PHQ-2) and Generalized Anxiety Disorder-2 (GAD-2), respectively [16]. Vision impairment was determined by asking participants whether they were blind or unable to see well enough to recognise people across the street or read newspaper print. Hearing impairment was determined by asking participants whether they were deaf, wore the hearing aid or were unable to hear well enough to use the telephone or carry on a conversation whilst watching the television or listening to the radio. The operationalization of these chronic conditions has been previously validated [15, 17]. Multimorbidity was defined as the coexistence of at least two chronic conditions within one individual.
Primary outcomes
Frailty
Physical frailty was annually assessed by the Fried frailty phenotype [1] (Table S3): (i) exhaustion: defined as having low energy or being easily exhausted to the point of limiting activities in the last month; (ii) low physical activity: defined as never walking for exercise or engaged in vigorous activities in the last month; (iii) shrinking: defined as body mass index (BMI) < 18.5 kg/m2, based on self-reported weight and height, or unintentional weight loss ≥10 lb. in the last year; (iv) weakness: measured by the best of two dominant handgrip strength measurements. Participants with handgrip strength ≤20th percentile of the population distribution stratified by gender and BMI groups were defined as having weakness; and (v) slowness: measured by gait speed from the first of two 3 m walking trails. Gait speed ≤20th percentile of the population distribution stratified by gender and height. Individuals meeting more than two criteria are defined as ‘frail’, otherwise as ‘non-frail’ [18].
Cognitive impairment
In NHATS [19], participants were classified into 3 groups—probable dementia, possible dementia and no dementia, using three types of information. ‘Probable dementia’ was defined if at least one of three criteria was met: (1) self- or proxy-report of doctor’s diagnosis of dementia or Alzheimer’s disease; (2) a score of ≥2 on the AD8 Dementia Screening Interview [20]; or (3) having a cut-point of ≤1.5 SDs below the mean in at least two cognitive domains (memory, orientation and executive function). ‘Possible dementia’ (also referred to as CI not dementia, or CIND) was defined for self- and proxy-respondents not reporting a diagnosis with test performance scores ≤1.5 SD below mean in one domain. In our study, the definition of CI included not only ‘probable dementia’ but also ‘possible dementia’. More details are available in Table S4.
Order of frailty and cognitive impairment occurrence
According to the order of frailty and CI occurrence, we classified participants into four groups [7]: (i) neither frail nor cognitively impaired; (ii) incident frailty one year or more prior to CI (termed ‘frailty-first’ henceforth); (iii) incident CI one year or more prior to frailty (termed ‘CI-first’ henceforth); and (iv) concurrent frailty and CI within one year (termed ‘frailty-CI co-occurrence’ henceforth).
Covariates
Baseline covariates were collected as described in previous studies [7, 15, 21], including chronological age, gender, race/ethnicity, education, living status and smoking status.
Statistical analysis
Baseline characteristics were summarised using frequency (percentage) for categorical variables and mean (standard deviation [SD]) for continuous variables. These characteristics were compared with Chi-square for categorical variables and ANOVA for continuous variables.
Multimorbidity patterns were explored amongst those 5545 multimorbid participants at baseline using latent class analysis (LCA). LCA is a statistical model in which individuals can be classified into mutually exclusive and exhaustive latent classes. We conducted a sequence of LCA models (the one- through six-class model), beginning with a one-class model and then progressively adding one class at a time up to the six-class model [22]. We determined the optimal number based on goodness-of-fit criteria [23] —BIC, aBIC, cAIC and Lo–Mendell–Rubin Adjusted (LMRA) test, as well as clinical interpretability. The four-class model yielded the relatively optimal fit and the best clinical interpretability (Fig. S1). The four classes were labelled based on chronic conditions that exhibited a relatively higher prevalence within each class compared to the overall prevalence across all multimorbid participants (Table S5). Subsequently, participants were assigned to a certain class in which they had the highest estimated posterior probability membership.
The associations between multimorbidity patterns and risk of different orders of frailty and CI occurrence (frailty-first, CI-first, frailty-CI co-occurrence) were examined by Fine and Grey competing risks models [24]. These models treated three occurrence patterns as mutually exclusive outcomes whilst accounting for frailty-and-Cl-free death as competing risks [7]. Time to event was calculated the years from baseline until outcomes of interest, frailty-and-Cl-free death or the last interview. Loss of follow-up was treated as censoring in the competing risk model. Model 1 was an unadjusted crude model; Model 2 was adjusted for all covariates. No imputation was conducted because missing values on covariates were less than 1% and met the Little’s missing completely at random (MCAR) assumption [25].
In the sensitivity analysis, we: (i) assessed the contribution of disease count without adjusting multimorbidity patterns; (ii) repeated the analyses after excluding individuals who developed probable dementia during follow-up (n = 487); (3) assessed the impact of attrition after excluding those frailty-and-Cl-free participants at the time of drop-out (n = 2130); (4) repeated the analyses after excluding proxy-respondent individuals (n = 249). Two-tailed p values less than 0.05 were considered to be statistically significant. Data analyses were performed R 4.3.1 (R Foundation for Statistical Computing, Vienna) and StataSE 16.0 (StataCorp, College Station, TX).
Results
Participant characteristics at baseline
The baseline characteristics were reported in Table 1. Of overall participants, the average age was 75.3 years and 43.3% were male. The majority were non-Hispanic white (72.7%), and over half (55.1%) had some college or vocational school education or higher. Approximately one-third (31.1%) lived alone. Significant between-group differences were found in all covariates (P < 0.05).
Baseline characteristics of participants by multimorbidity patternsa (N = 7522).
| Overall | No multimorbidity (n = 1977) | Cardiometabolic pattern (n = 2510) | Osteoarticular pattern (n = 1013) | Cancer-dominated pattern (n = 1131) | Psychiatric/multisystem pattern (n = 891) | P value | |
|---|---|---|---|---|---|---|---|
| Age, Mean (SD) | 75.3 (7.1) | 73.6 (6.6) | 75.2 (6.8) | 76.9 (7.5) | 76.45(7.0) | 75.6 (7.4) | <.001 |
| Male, n (%) | 3256 (43.3) | 975 (49.3) | 1103 (43.9) | 191 (18.9) | 656 (58.0) | 331 (37.1) | <.001 |
| Race/ethnicity, n (%) | <.001 | ||||||
| White, non-Hispanic | 5468 (72.7) | 1471 (74.4) | 1646 (65.6) | 842 (83.1) | 902 (79.8) | 607 (68.1) | |
| Black, non-Hispanic | 1391 (18.5) | 315 (15.9) | 622 (24.8) | 94 (9.3) | 168 (14.9) | 192 (21.5) | |
| Hispanic | 356 (4.7) | 98 (5.0) | 138 (5.5) | 42 (4.1) | 26 (2.3) | 52 (5.8) | |
| Othersb | 307 (4.1) | 93 (4.7) | 104 (4.1) | 35 (3.5) | 35 (3.1) | 40 (4.5) | |
| Education, n (%)c | <.001 | ||||||
| Less than high school | 1259 (16.7) | 270 (13.9) | 453 (18.3) | 160 (16.0) | 133 (11.9) | 243 (27.6) | |
| High school graduates | 2029 (27.0) | 476 (24.4) | 696 (28.1) | 286 (28.6) | 292 (26.1) | 279 (31.7) | |
| Some college or vocational school | 2155 (28.7) | 548 (28.1) | 743 (29.9) | 299 (29.9) | 339 (30.3) | 226 (25.7) | |
| Bachelor or higher | 1987 (26.4) | 654 (33.6) | 589 (23.7) | 256 (25.6) | 355 (31.7) | 133 (15.1) | |
| Living status, n (%)c | <.001 | ||||||
| Alone | 2351 (31.3) | 535 (27.1) | 790 (31.6) | 380 (37.5) | 321 (28.5) | 325 (36.6) | |
| With spouse/partner only | 3515 (46.7) | 1063 (53.9) | 1105 (44.1) | 429 (42.4) | 596 (52.9) | 322 (36.2) | |
| With others only | 954 (12.7) | 201 (10.2) | 368 (14.7) | 130 (12.8) | 102 (9.1) | 153 (17.2) | |
| With spouse/partner and with others | 682 (9.1) | 172 (8.7) | 240 (9.6) | 73 (7.2) | 108 (9.6) | 89 (10.0) | |
| Smoking status, n (%)c | <.001 | ||||||
| Never | 3648 (48.5) | 1019 (51.6) | 1179 (47.0) | 518 (51.2) | 517 (45.8) | 415 (46.6) | |
| Former | 3271 (43.5) | 780 (39.5) | 1149 (45.8) | 426 (42.1) | 544 (48.1) | 372 (41.8) | |
| Current | 597 (7.9) | 177 (9.0) | 179 (7.1) | 68 (6.7) | 69 (6.1) | 104 (11.7) |
| Overall | No multimorbidity (n = 1977) | Cardiometabolic pattern (n = 2510) | Osteoarticular pattern (n = 1013) | Cancer-dominated pattern (n = 1131) | Psychiatric/multisystem pattern (n = 891) | P value | |
|---|---|---|---|---|---|---|---|
| Age, Mean (SD) | 75.3 (7.1) | 73.6 (6.6) | 75.2 (6.8) | 76.9 (7.5) | 76.45(7.0) | 75.6 (7.4) | <.001 |
| Male, n (%) | 3256 (43.3) | 975 (49.3) | 1103 (43.9) | 191 (18.9) | 656 (58.0) | 331 (37.1) | <.001 |
| Race/ethnicity, n (%) | <.001 | ||||||
| White, non-Hispanic | 5468 (72.7) | 1471 (74.4) | 1646 (65.6) | 842 (83.1) | 902 (79.8) | 607 (68.1) | |
| Black, non-Hispanic | 1391 (18.5) | 315 (15.9) | 622 (24.8) | 94 (9.3) | 168 (14.9) | 192 (21.5) | |
| Hispanic | 356 (4.7) | 98 (5.0) | 138 (5.5) | 42 (4.1) | 26 (2.3) | 52 (5.8) | |
| Othersb | 307 (4.1) | 93 (4.7) | 104 (4.1) | 35 (3.5) | 35 (3.1) | 40 (4.5) | |
| Education, n (%)c | <.001 | ||||||
| Less than high school | 1259 (16.7) | 270 (13.9) | 453 (18.3) | 160 (16.0) | 133 (11.9) | 243 (27.6) | |
| High school graduates | 2029 (27.0) | 476 (24.4) | 696 (28.1) | 286 (28.6) | 292 (26.1) | 279 (31.7) | |
| Some college or vocational school | 2155 (28.7) | 548 (28.1) | 743 (29.9) | 299 (29.9) | 339 (30.3) | 226 (25.7) | |
| Bachelor or higher | 1987 (26.4) | 654 (33.6) | 589 (23.7) | 256 (25.6) | 355 (31.7) | 133 (15.1) | |
| Living status, n (%)c | <.001 | ||||||
| Alone | 2351 (31.3) | 535 (27.1) | 790 (31.6) | 380 (37.5) | 321 (28.5) | 325 (36.6) | |
| With spouse/partner only | 3515 (46.7) | 1063 (53.9) | 1105 (44.1) | 429 (42.4) | 596 (52.9) | 322 (36.2) | |
| With others only | 954 (12.7) | 201 (10.2) | 368 (14.7) | 130 (12.8) | 102 (9.1) | 153 (17.2) | |
| With spouse/partner and with others | 682 (9.1) | 172 (8.7) | 240 (9.6) | 73 (7.2) | 108 (9.6) | 89 (10.0) | |
| Smoking status, n (%)c | <.001 | ||||||
| Never | 3648 (48.5) | 1019 (51.6) | 1179 (47.0) | 518 (51.2) | 517 (45.8) | 415 (46.6) | |
| Former | 3271 (43.5) | 780 (39.5) | 1149 (45.8) | 426 (42.1) | 544 (48.1) | 372 (41.8) | |
| Current | 597 (7.9) | 177 (9.0) | 179 (7.1) | 68 (6.7) | 69 (6.1) | 104 (11.7) |
aBecause analytic weights were unavailable for the combined samples in our study, our statistical analyses proceeded without incorporating sample weights.
bIncludes American Indian or Alaska Native, Asian, and Native Hawaiian or other Pacific Islander.
cMissing variables: 92 missing on education, 20 missing on living status, and 6 missing on smoking status.
Baseline characteristics of participants by multimorbidity patternsa (N = 7522).
| Overall | No multimorbidity (n = 1977) | Cardiometabolic pattern (n = 2510) | Osteoarticular pattern (n = 1013) | Cancer-dominated pattern (n = 1131) | Psychiatric/multisystem pattern (n = 891) | P value | |
|---|---|---|---|---|---|---|---|
| Age, Mean (SD) | 75.3 (7.1) | 73.6 (6.6) | 75.2 (6.8) | 76.9 (7.5) | 76.45(7.0) | 75.6 (7.4) | <.001 |
| Male, n (%) | 3256 (43.3) | 975 (49.3) | 1103 (43.9) | 191 (18.9) | 656 (58.0) | 331 (37.1) | <.001 |
| Race/ethnicity, n (%) | <.001 | ||||||
| White, non-Hispanic | 5468 (72.7) | 1471 (74.4) | 1646 (65.6) | 842 (83.1) | 902 (79.8) | 607 (68.1) | |
| Black, non-Hispanic | 1391 (18.5) | 315 (15.9) | 622 (24.8) | 94 (9.3) | 168 (14.9) | 192 (21.5) | |
| Hispanic | 356 (4.7) | 98 (5.0) | 138 (5.5) | 42 (4.1) | 26 (2.3) | 52 (5.8) | |
| Othersb | 307 (4.1) | 93 (4.7) | 104 (4.1) | 35 (3.5) | 35 (3.1) | 40 (4.5) | |
| Education, n (%)c | <.001 | ||||||
| Less than high school | 1259 (16.7) | 270 (13.9) | 453 (18.3) | 160 (16.0) | 133 (11.9) | 243 (27.6) | |
| High school graduates | 2029 (27.0) | 476 (24.4) | 696 (28.1) | 286 (28.6) | 292 (26.1) | 279 (31.7) | |
| Some college or vocational school | 2155 (28.7) | 548 (28.1) | 743 (29.9) | 299 (29.9) | 339 (30.3) | 226 (25.7) | |
| Bachelor or higher | 1987 (26.4) | 654 (33.6) | 589 (23.7) | 256 (25.6) | 355 (31.7) | 133 (15.1) | |
| Living status, n (%)c | <.001 | ||||||
| Alone | 2351 (31.3) | 535 (27.1) | 790 (31.6) | 380 (37.5) | 321 (28.5) | 325 (36.6) | |
| With spouse/partner only | 3515 (46.7) | 1063 (53.9) | 1105 (44.1) | 429 (42.4) | 596 (52.9) | 322 (36.2) | |
| With others only | 954 (12.7) | 201 (10.2) | 368 (14.7) | 130 (12.8) | 102 (9.1) | 153 (17.2) | |
| With spouse/partner and with others | 682 (9.1) | 172 (8.7) | 240 (9.6) | 73 (7.2) | 108 (9.6) | 89 (10.0) | |
| Smoking status, n (%)c | <.001 | ||||||
| Never | 3648 (48.5) | 1019 (51.6) | 1179 (47.0) | 518 (51.2) | 517 (45.8) | 415 (46.6) | |
| Former | 3271 (43.5) | 780 (39.5) | 1149 (45.8) | 426 (42.1) | 544 (48.1) | 372 (41.8) | |
| Current | 597 (7.9) | 177 (9.0) | 179 (7.1) | 68 (6.7) | 69 (6.1) | 104 (11.7) |
| Overall | No multimorbidity (n = 1977) | Cardiometabolic pattern (n = 2510) | Osteoarticular pattern (n = 1013) | Cancer-dominated pattern (n = 1131) | Psychiatric/multisystem pattern (n = 891) | P value | |
|---|---|---|---|---|---|---|---|
| Age, Mean (SD) | 75.3 (7.1) | 73.6 (6.6) | 75.2 (6.8) | 76.9 (7.5) | 76.45(7.0) | 75.6 (7.4) | <.001 |
| Male, n (%) | 3256 (43.3) | 975 (49.3) | 1103 (43.9) | 191 (18.9) | 656 (58.0) | 331 (37.1) | <.001 |
| Race/ethnicity, n (%) | <.001 | ||||||
| White, non-Hispanic | 5468 (72.7) | 1471 (74.4) | 1646 (65.6) | 842 (83.1) | 902 (79.8) | 607 (68.1) | |
| Black, non-Hispanic | 1391 (18.5) | 315 (15.9) | 622 (24.8) | 94 (9.3) | 168 (14.9) | 192 (21.5) | |
| Hispanic | 356 (4.7) | 98 (5.0) | 138 (5.5) | 42 (4.1) | 26 (2.3) | 52 (5.8) | |
| Othersb | 307 (4.1) | 93 (4.7) | 104 (4.1) | 35 (3.5) | 35 (3.1) | 40 (4.5) | |
| Education, n (%)c | <.001 | ||||||
| Less than high school | 1259 (16.7) | 270 (13.9) | 453 (18.3) | 160 (16.0) | 133 (11.9) | 243 (27.6) | |
| High school graduates | 2029 (27.0) | 476 (24.4) | 696 (28.1) | 286 (28.6) | 292 (26.1) | 279 (31.7) | |
| Some college or vocational school | 2155 (28.7) | 548 (28.1) | 743 (29.9) | 299 (29.9) | 339 (30.3) | 226 (25.7) | |
| Bachelor or higher | 1987 (26.4) | 654 (33.6) | 589 (23.7) | 256 (25.6) | 355 (31.7) | 133 (15.1) | |
| Living status, n (%)c | <.001 | ||||||
| Alone | 2351 (31.3) | 535 (27.1) | 790 (31.6) | 380 (37.5) | 321 (28.5) | 325 (36.6) | |
| With spouse/partner only | 3515 (46.7) | 1063 (53.9) | 1105 (44.1) | 429 (42.4) | 596 (52.9) | 322 (36.2) | |
| With others only | 954 (12.7) | 201 (10.2) | 368 (14.7) | 130 (12.8) | 102 (9.1) | 153 (17.2) | |
| With spouse/partner and with others | 682 (9.1) | 172 (8.7) | 240 (9.6) | 73 (7.2) | 108 (9.6) | 89 (10.0) | |
| Smoking status, n (%)c | <.001 | ||||||
| Never | 3648 (48.5) | 1019 (51.6) | 1179 (47.0) | 518 (51.2) | 517 (45.8) | 415 (46.6) | |
| Former | 3271 (43.5) | 780 (39.5) | 1149 (45.8) | 426 (42.1) | 544 (48.1) | 372 (41.8) | |
| Current | 597 (7.9) | 177 (9.0) | 179 (7.1) | 68 (6.7) | 69 (6.1) | 104 (11.7) |
aBecause analytic weights were unavailable for the combined samples in our study, our statistical analyses proceeded without incorporating sample weights.
bIncludes American Indian or Alaska Native, Asian, and Native Hawaiian or other Pacific Islander.
cMissing variables: 92 missing on education, 20 missing on living status, and 6 missing on smoking status.
Multimorbidity patterns
Amongst 5545 participants with multimorbidity, 2510 (45.3%) were classified as cardiometabolic pattern, followed by cancer-dominated pattern (20.4%), osteoarticular pattern (18.3%) and psychiatric/multisystem pattern (16.1%). Of note, these labels reflect the diseases most prevalent within each cluster, rather than across all clusters (Fig. 2). The cardiometabolic pattern had a higher prevalence of diabetes, hypertension and heart disease, along with a slightly higher prevalence of stroke. The osteoarticular pattern had a higher prevalence of arthritis and osteoporosis. The cancer-dominated pattern was predominantly driven by cancer diagnoses. The psychiatric/multisystem pattern was primarily driven by anxiety and depressive symptoms, but many individuals of this group might suffer from other conditions involving multiple systems (e.g. lung disease, stroke, heart disease, vision impairment, hearing impairment).
Relative excess prevalence of chronic conditions by classes compared to population prevalence. Four multimorbidity patterns were labelled according to the relative excess prevalence (i.e. difference value of population prevalence and conditional probability), which indicates the likelihood of an individual within a certain class reporting the presence of a specific condition. Cases with missing values are retained and handled with a full-information maximum likelihood technique. The detailed class selection process, conditional probabilities and relative prevalence rate of chronic conditions by 4 classes are reported in the Appendix.
Multimorbidity patterns and order of frailty and cognitive impairment occurrence
Amongst 7522 participants, 2828 (37.6%) remained robust without frailty nor CI during the four-year follow-up; 1028 (13.7%) developed incident frailty before CI, with 175 (17.0%) of them subsequently developing CI; 979 (13.0%) developed incident CI before frailty, with 170 (17.4%) of them subsequently developing frailty; and 208 (2.8%) developed both within the same year. In addition, 349 (4.6%) participants died during follow-up, and 2130 (28.3%) dropped out before any of the outcomes could occur.
Table 2 presented the associations between the multimorbidity patterns and order of frailty and CI occurrence. Compared to non-multimorbidity, all four multimorbidity patterns were associated with a higher risk of developing frailty-first. These associations were attenuated but still significant in the adjusted model. Specifically, the psychiatric/multisystem pattern had an almost fourfold higher risk (Sub-distribution hazard ratios [SHR] = 3.74, 95% confidence intervals = 2.96, 4.71), followed by osteoarticular (SHR = 2.53, 95% confidence intervals = 1.98, 3.22) and cardiometabolic pattern (SHR = 2.41, 95% confidence intervals = 1.96, 2.98). On the contrary, those in the cancer-dominated pattern (SHR = 0.78, 95% confidence intervals = 0.63, 0.96) had a decreased risk of CI-first occurrence in the adjusted model. In addition, only participants from the psychiatric/multisystem (SHR = 2.24, 95% confidence intervals = 1.38, 3.64) and cardiometabolic pattern (SHR = 1.82, 95% confidence intervals = 1.19, 2.78) showed a higher risk of frailty-CI co-occurrence.
Association between multimorbidity patterns and order of incident frailty and CI occurrence (N = 7522).
| Frailty-first SHR (95% confidence intervals)a | CI-first SHR (95% confidence intervals)b | Frailty-CI co-occurrence SHR (95% confidence intervals)c | |
|---|---|---|---|
| Model 1d | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.69 (2.19, 3.30)*** | 1.07 (0.91, 1.25) | 2.14 (1.41, 3.24)*** |
| Osteoarticular pattern | 2.97 (2.35, 3.75)*** | 0.95 (0.77, 1.17) | 1.74 (1.04, 2.92)* |
| Cancer-dominated pattern | 2.46 (1.94, 3.12)*** | 0.91 (0.74, 1.11) | 1.38 (0.81, 2.35) |
| Psychiatric/multisystem pattern | 4.40 (3.52, 5.51)*** | 1.21 (0.99, 1.49) | 3.38 (2.13, 5.34)*** |
| Model 2e | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.41 (1.96, 2.98)*** | 0.91 (0.78, 1.07) | 1.82 (1.19, 2.78)** |
| Osteoarticular pattern | 2.53 (1.98, 3.22)*** | 0.85 (0.69, 1.06) | 1.34 (0.78, 2.30) |
| Cancer-dominated pattern | 2.25 (1.77, 2.87)*** | 0.78 (0.63, 0.96)* | 1.10 (0.64, 1.90) |
| Psychiatric/multisystem pattern | 3.74 (2.96, 4.71)*** | 0.97 (0.78, 1.20) | 2.24 (1.38, 3.64)** |
| Frailty-first SHR (95% confidence intervals)a | CI-first SHR (95% confidence intervals)b | Frailty-CI co-occurrence SHR (95% confidence intervals)c | |
|---|---|---|---|
| Model 1d | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.69 (2.19, 3.30)*** | 1.07 (0.91, 1.25) | 2.14 (1.41, 3.24)*** |
| Osteoarticular pattern | 2.97 (2.35, 3.75)*** | 0.95 (0.77, 1.17) | 1.74 (1.04, 2.92)* |
| Cancer-dominated pattern | 2.46 (1.94, 3.12)*** | 0.91 (0.74, 1.11) | 1.38 (0.81, 2.35) |
| Psychiatric/multisystem pattern | 4.40 (3.52, 5.51)*** | 1.21 (0.99, 1.49) | 3.38 (2.13, 5.34)*** |
| Model 2e | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.41 (1.96, 2.98)*** | 0.91 (0.78, 1.07) | 1.82 (1.19, 2.78)** |
| Osteoarticular pattern | 2.53 (1.98, 3.22)*** | 0.85 (0.69, 1.06) | 1.34 (0.78, 2.30) |
| Cancer-dominated pattern | 2.25 (1.77, 2.87)*** | 0.78 (0.63, 0.96)* | 1.10 (0.64, 1.90) |
| Psychiatric/multisystem pattern | 3.74 (2.96, 4.71)*** | 0.97 (0.78, 1.20) | 2.24 (1.38, 3.64)** |
Notes: SHR, sub-distribution hazard ratios; Ref, reference comparison category; CI, cognitive impairment.
aFrailty-first competing risks: CI-first, frailty-CI co-occurrence, death.
bCI-first competing risks: frailty-first, frailty-CI co-occurrence, death.
cFrailty-CI co-occurrence competing risks: frailty-first, CI-first, death.
dModel 1 unadjusted.
eModel 2 adjusted for all covariates (age, gender, race/ethnicity, education, living status, and smoking status).
*P < .05.
**P < .01.
***P < .001.
Association between multimorbidity patterns and order of incident frailty and CI occurrence (N = 7522).
| Frailty-first SHR (95% confidence intervals)a | CI-first SHR (95% confidence intervals)b | Frailty-CI co-occurrence SHR (95% confidence intervals)c | |
|---|---|---|---|
| Model 1d | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.69 (2.19, 3.30)*** | 1.07 (0.91, 1.25) | 2.14 (1.41, 3.24)*** |
| Osteoarticular pattern | 2.97 (2.35, 3.75)*** | 0.95 (0.77, 1.17) | 1.74 (1.04, 2.92)* |
| Cancer-dominated pattern | 2.46 (1.94, 3.12)*** | 0.91 (0.74, 1.11) | 1.38 (0.81, 2.35) |
| Psychiatric/multisystem pattern | 4.40 (3.52, 5.51)*** | 1.21 (0.99, 1.49) | 3.38 (2.13, 5.34)*** |
| Model 2e | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.41 (1.96, 2.98)*** | 0.91 (0.78, 1.07) | 1.82 (1.19, 2.78)** |
| Osteoarticular pattern | 2.53 (1.98, 3.22)*** | 0.85 (0.69, 1.06) | 1.34 (0.78, 2.30) |
| Cancer-dominated pattern | 2.25 (1.77, 2.87)*** | 0.78 (0.63, 0.96)* | 1.10 (0.64, 1.90) |
| Psychiatric/multisystem pattern | 3.74 (2.96, 4.71)*** | 0.97 (0.78, 1.20) | 2.24 (1.38, 3.64)** |
| Frailty-first SHR (95% confidence intervals)a | CI-first SHR (95% confidence intervals)b | Frailty-CI co-occurrence SHR (95% confidence intervals)c | |
|---|---|---|---|
| Model 1d | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.69 (2.19, 3.30)*** | 1.07 (0.91, 1.25) | 2.14 (1.41, 3.24)*** |
| Osteoarticular pattern | 2.97 (2.35, 3.75)*** | 0.95 (0.77, 1.17) | 1.74 (1.04, 2.92)* |
| Cancer-dominated pattern | 2.46 (1.94, 3.12)*** | 0.91 (0.74, 1.11) | 1.38 (0.81, 2.35) |
| Psychiatric/multisystem pattern | 4.40 (3.52, 5.51)*** | 1.21 (0.99, 1.49) | 3.38 (2.13, 5.34)*** |
| Model 2e | |||
| No multimorbidity | Ref | Ref | Ref |
| Cardiometabolic pattern | 2.41 (1.96, 2.98)*** | 0.91 (0.78, 1.07) | 1.82 (1.19, 2.78)** |
| Osteoarticular pattern | 2.53 (1.98, 3.22)*** | 0.85 (0.69, 1.06) | 1.34 (0.78, 2.30) |
| Cancer-dominated pattern | 2.25 (1.77, 2.87)*** | 0.78 (0.63, 0.96)* | 1.10 (0.64, 1.90) |
| Psychiatric/multisystem pattern | 3.74 (2.96, 4.71)*** | 0.97 (0.78, 1.20) | 2.24 (1.38, 3.64)** |
Notes: SHR, sub-distribution hazard ratios; Ref, reference comparison category; CI, cognitive impairment.
aFrailty-first competing risks: CI-first, frailty-CI co-occurrence, death.
bCI-first competing risks: frailty-first, frailty-CI co-occurrence, death.
cFrailty-CI co-occurrence competing risks: frailty-first, CI-first, death.
dModel 1 unadjusted.
eModel 2 adjusted for all covariates (age, gender, race/ethnicity, education, living status, and smoking status).
*P < .05.
**P < .01.
***P < .001.
In the sensitivity analysis, as the disease count increases, there was a higher risk of developing frailty-first, as well as frailty-CI co-occurrence (Table S6). The findings were consistent with the main analysis after accounting for ongoing cognitive decline towards probable dementia (Table S7). Moreover, our results remained largely unchanged after excluding drop-out participants (Table S8) and proxy respondents (Table S9).
Discussion
We identified four underlying multimorbidity patterns: cardiometabolic, osteoarticular, cancer-dominated and psychiatric/multisystem pattern, which have also been identified in previous studies [15, 26, 27]. In this study, approximately half of the participants were classified into cardiometabolic pattern, which was characterised by the co-existence of cardiovascular (e.g. heart disease, stroke) and metabolic diseases (e.g. diabetes). The cardiometabolic pattern has been observed as the predominant multimorbidity pattern in older adults and was associated with depressive symptoms [28], cognitive decline [29], and mortality [30].
To our best knowledge, this was the first longitudinal study to investigate the associations between multimorbidity patterns and the order of frailty and CI occurrence. All patterns were associated with a higher risk of developing frailty-first, but not developing CI-first. Only individuals from the psychiatric/multisystem pattern showed a higher risk of frailty-CI co-occurrence. In our study, older adults who developed frailty-first (13.7%) accounted for a slightly larger proportion than those who developed CI-first (13.0%). Nearly the same proportion of participants who initially experienced frailty or CI subsequently developed CI or frailty (17.0% vs. 17.4%) thus becoming cognitive frailty. This finding is inconsistent with a previous study, which found that the transition from CI to cognitive frailty was 8.7 times as likely to move from physical frailty to cognitive frailty [31]. This inconsistency might be explained by the differences in study sample (American vs. Chinese), and operationalization of variables (i.e. frailty and CI).
Previous studies have investigated the associations between multimorbidity patterns and incident frailty, but did not consider the competing risk between frailty and CI [10, 11, 32]. In our study, older adults within the psychiatric/multisystem pattern were associated with the highest risk of developing frailty-first. This elevated risk was primarily attributed to the higher reported comorbidities covering multiple systems (Table S6), which is in line with prior literature demonstrating the adverse impact of comorbidity burden on frailty progression [15, 33–35]. Accumulated systemic inflammatory markers resulted from cancer cachexia have underlying mechanisms to accelerate the progression of frailty [36]. Moreover, cardiovascular [37] and metabolic diseases [38] are well-established risk factors for frailty, and disease combinations could exert additive negative effects on physical performance, resulting in progressive functional decline [39, 40]. In addition, the osteoarticular pattern showed a higher risk of developing frailty first than other multimorbidity patterns, except for the psychiatric/multisystem pattern. A plausible explanation is that chronic musculoskeletal conditions like arthritis and osteoporosis often lead to significant physical limitations and pain, resulting in decreased mobility, muscle weakness and a greater vulnerability to frailty.
Only a few studies have reported the associations between multimorbidity patterns and cognitive function [21, 41]. A longitudinal study found that older adults within the cardiovascular pattern were prone to develop cognitive risk syndrome, known as a pre-dementia syndrome. Combinations of cardiovascular conditions may have synergistic effects on cognitive decline, possibly through inflammation [42], vascular dysregulation [43], and brain pathology [44]. However, our study showed a lack of significant association between all multimorbidity patterns and the risk of developing CI-first. This finding is partially supported by another study indicating that history of single medical conditions failed to predict this occurrence pattern [7]. The cancer-dominated pattern showed a reduced risk of CI-first, possibly due to cancer-related pathways causing acute physical decline that overshadows CI. Whilst our results offer valuable insights into the temporal dynamics of frailty and cognitive decline, as well as their distinct underlying mechanisms, they should be interpreted with caution. These included chronic conditions may be more strongly linked to progressive physical decline and frailty, which could compete with the onset of CI. Future research should explore the relationship between multimorbidity patterns and the order of frailty and CI occurrence using more comprehensive clinical assessments.
Our definition of ‘frailty-CI co-occurrence’ is slightly different from the concept of ‘cognitive frailty’ [45]. Prior studies have identified comorbidity as a significant predictor of cognitive frailty [6, 46, 47]. However, these conclusions derived from either cross-sectional studies or endpoint assessment in longitudinal studies, thus overlooking the sequence of frailty and CI. In our main analysis, the cardiometabolic pattern was associated with an increased risk of frailty-CI co-occurrence. This can be explained by the interaction between cardiometabolic and brain pathology [44]. Evidence suggests that brain pathology contributes to simultaneous change in physical frailty and cognition in old age [48]. Some researchers argued that cognitive frailty might primarily be attributed to neurologic disease-related pathologies rather than normal cognitive aging or physical impairments [6]. Hence, further studies are warranted to fully consider neuropsychiatric conditions, which were underrepresented in this study.
This study has important clinical implications. Diseases within different clusters involve distinct physiological, neurological, and pathological processes and thus influence the hierarchical development of frailty and CI. Identification of multimorbidity patterns can help clinicians to identify older adults who are at risk of developing frailty either before, after or concurrent with CI. Furthermore, tailored prevention and management strategies need to be implemented for individuals with distinct multimorbidity patterns. For example, individuals within osteoarticular pattern should be more concerned about physical decline and keep physically active to delay frailty onset. However, disease patterns are not static. Clinicians should also consider the limitations of using cluster-based classifications and remain open to the multifaceted nature of each patient’s health. When conditions outside the identified clusters are present, they should be incorporated into care plans.
Limitations
One limitation is a risk of selection bias due to non-response over follow-up (n = 2130, 28.3%), resulting in a younger and relatively healthier sample. This bias could limit the generalizability of our findings to the broader population, particularly those with more severe frailty or CI, who were underrepresented in the study due to their higher attrition rates. However, our results were not much altered after accounting for attrition (Table S8). Second, all chronic conditions were self-reported, and consequently, the prevalence of multimorbidity may be underestimated. The NHATS collected data on only 12 common chronic conditions and may neglect others (e.g. hyperlipidemia, chronic obstructive pulmonary disease [COPD]). Further studies should consider more physical and neuropsychiatric diseases (e.g. Parkinson disease, schizophrenia), as well as their severity and duration. Third, the disease patterns of participants may have evolved since baseline. Individuals initially diagnosed with only one chronic condition could have developed specific multimorbidity patterns, leading to an underestimation of the associations between specific patterns and the outcomes of interest. Fourth, dichotomizing frailty and CI may fail to capture the progressive decline that occurs before reaching the thresholds used to define these conditions. Moreover, although LCA has a robust function in identifying distinct subgroups, misclassification is inevitable [49]. Whilst the multimorbidity patterns observed in our study are consistent with previous research, suggesting external validation, future studies need explore whether these patterns persist across various healthcare settings and populations, helping to establish their broader applicability in predicting frailty and cognitive decline.
Conclusions
In this prospective study, four multimorbidity patterns were identified (cardiometabolic, osteoarticular, cancer-dominated and psychiatric/multisystem pattern). All multimorbidity patterns were associated with a higher risk of developing frailty before CI, but not developing CI before frailty. Clinician should be aware of their occurrence sequence and thus provide tailored prevention for comorbid older adults at risk of these geriatric syndromes.
Acknowledgements:
We are grateful to the workers, researchers and participants involved in the NHATS.
Declaration of Conflicts of Interest:
None.
Declaration of Sources of Funding:
This work was supported by research grants from the National Natural Science Foundation of China [NSFC72174012], Humanities and Social Science Fund of Ministry of Education [24YJA840009], Key R&D Program Project of Hunan Province Grant [2023SK2009] and Changsha Science and Technology Program Soft Science Project [kh2302038]. The funders had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Data Availability:
The data relevant to this study are available from NHATS (https://www.nhats.org/).
Reference
Author notes
Hongyu Sun and Minhui Liu contributed equally as corresponding authors.
Comments