The ICOPE Intrinsic Capacity Screening Tool: Measurement Structure and Predictive Validity of Dependence and Hospitalization

Abstract

Objectives

To evaluate the measurement structure of the ICOPE screening tool (IST) of intrinsic capacity and to find out whether the IST as a global measure adds explanatory power over and above its domains in isolation to predict the occurrence of adverse health outcomes such as dependence and hospitalization in community-dwelling older people.

Design

Secondary analysis of a cohort study, the Toledo Study of Healthy Ageing.

Setting

Province of Toledo, Spain.

Participants

Community-dwelling older people.

Measurements

Items equal or similar to those of the IST were introduced as a reflective-formative construct in a Structural Equation Model to evaluate its measurement structure and its association with dependence for basic and instrumental activities and hospitalization over a three-year period.

Results

A total of 1032 individuals were analyzed. Mean age was 73.5 years (sd 5.4). The least preserved indicators were ability to recall three words (18%) and to perform chair stands (54%). Vision and hearing items did not form a single sensory domain, so six domains were considered. Several cognition items did not show sufficiently strong and univocal associations with the domain. After pruning the ill-behaved items, the measurement model fit was excellent (Satorra-Bentler scaled chi-square: 10.3, degrees of freedom: 11, p=0.501; CFI: 1.000; RMSEA: 0.000, 90% CI: 0.000–0.031, p value RMSEA<=0.05: 1; SRMR: 0.055). In the structural model, the cognition domain items were not associated as expected with age (p values 0.158 and 0.293), education (p values 0.190 and 0.432) and dependence (p values 0.654 and 0.813). The IST included as a composite in a model with the individual domains showed no statistically significant associations with any of the outcomes (dependence for basic activities: 0.162, p=0.167; instrumental: −0.052, p=0.546; hospitalization: 0.145, p=0.167), while only the mobility domain did so for dependence (basic: −0.266, p=0.005; instrumental: −0.138, p=0.019). The model fit of the last version was good (Satorra-Bentler scaled chi-square: 52.1, degrees of freedom: 52, p=0.469; CFI: 1.000; TLI: 1.000; RMSEA: 0.01, 90% CI: 0.000–0.02, p value RMSEA<=0.05: 1; SRMR: 0.071). The IST operationalized as the sum of non-impaired domains was not associated after covariate adjustment (dependence for basic activities: −0.065, p=0.356; instrumental: −0.08, p=0.05; hospitalization: −0.003, p=0.949) either.

Conclusion

The cognitive domain of the IST, and probably other of its items, may need a reformulation. A global measure of intrinsic capacity such as the IST does not add explanatory power to the individual domains that constitute it. So far, our results confirm the importance of checking the findings of the IST with a second confirmatory step, as described in the WHO's ICOPE strategy.

Introduction

Functional ability, the corner stone of healthy ageing as defined by the World Health Organization (WHO), is ‘the health-related attributes that enable people to be and to do what they have reason to value' and is composed of the intrinsic capacity (IC) of the person, the environment, and their interaction. IC is ‘the composite of all the physical and mental capacities that an individual can draw on'(1). It was operationalized (2) by determining the body functions (i.e., the physiological functions of body systems according to the International Classification of Functioning, Disability and Health) that were most strongly associated with an increased risk of incident functional loss and care dependence in the literature. The five body functions (domains in the IC literature) that were selected were cognition, emotion, sensory function (including vision and hearing), vitality (i.e., homeostatic regulation, or the balance between energy intake and energy utilization), and locomotion. With the aim of promoting healthy ageing by maintaining and promoting functional ability through intervention on IC, WHO has proposed the ICOPE (Integrated Care for Older People) program. It begins with the assessment of IC using the ICOPE screening tool (IST) that explores the five domains through specific questions or functional measures. In a second step, a person-centered assessment is carried out in primary care, in which the impairments of the screened domains are further assessed with validated standard tools. In addition, a comprehensive assessment of the person's life and health management values, priorities and preferences; underlying diseases and polypharmacy; and social and physical environments and need for social care and support, is conducted (3).

Although the IST seems a conceptually appropriate measure of IC, its reliability and validity remain to be demonstrated. To determine its reliability, the measurement structure of the scale must first be established (4). Predictive validity assessed through the ability to predict relevant adverse health outcomes, such as dependence, is essential in a tool that has been developed primarily to help prevent and revert losses in functional ability. Previous attempts to assess the reliability and validity of IC measures (5, 6, 7) have used all domain-related variables available in existing surveys and have confirmed their grouping into five domains. Beard et al (5) found that IC had a direct relationship with two-year incident limitation of basic (BADL) and instrumental (IADL) activities of daily living and strongly mediated the effect of age, sex, wealth and education in English community-dwellers. The reliability of the IC score (set out as a bi-factor model) was good. Yu et al (6) reported in Hong Kong that IC predicted increased IADL difficulties during a seven-year follow-up. Aliberti et al (7) showed a cross-sectional association of IC with preserved ADL, IADL and more advanced functioning tasks among Brazilians. Studies that have specifically used the IST items (8, 9, 10, 11, 12) have not checked its measurement structure, but have shown its ability to predict falls and increased dependence in BADL (8), incidence of BADL and IADL dependence (11), health care costs (9), and its cross-sectional associations with physical and mental health, IADL disability and frailty (10).

In relation to the measurement structure of latent variables like IC, two conceptual approaches can be followed, the reflective or the formative. Reflective measurement models (also known as common factor models) assume that each observable variable (indicator) is a measurement-error-prone consequence of the theoretical, unobservable concept. There is a presumed causal relationship that flows from the latent variable representing the theoretical concept to its pertaining indicators. An example would be the items of a depression scale, where each item is the manifestation of the latent variable ‘depressive mood' plus error. In contrast, in the alternative conceptualization of formative measurement, the relationship between the indicators and the construct is definitional and not causal. The composite ‘emerges' from the indicators because it helps better than its isolated parts to explain certain relationships. The indicators jointly determine the conceptual and empirical meaning of the construct. The relationship is modeled from the indicators to the construct, rather than from the construct to the indicators (13, 14).

The specification of the measurement model is a theoretical issue of major importance because of its implications for the selection of the statistical approach used to study the relationships between the items and the construct to which they belong and to operationalize it. All (5, 6, 7) but one (15) of the papers that have investigated it have considered IC as a reflective construct. The preference for a reflective measurement model is no surprise, as this is the dominant approach in the Structural Equation Modelling (SEM) literature (16), but we consider that it is not the most appropriate in the case of IC. This discussion does not apply to the relationship between items and domains which is undoubtedly reflective.

Both reflective and formative approaches can be combined in second-order models (constructs with two levels of measurement). In this study we use the IST items and use a SEM approach which combines reflective and formative structures. In this sense, we consider that the correct specification of IC measurement is a type-II second-order model according to Jarvis typology (17), where at the first level there is a reflective relationship between indicators and domains while at the second level there is a formative relationship between domains and IC (Figure 1). This specification reflects that IC is a composite, an artificial construct whose existence could only be justified if it provides additional information to that of its isolated specific components.

Figure 1.

Proposed reflective-formative model for intrinsic capacity and its domains

Our aim was, therefore, to evaluate the measurement structure of the IST and to find out whether the IST as a global measure adds explanatory power over and above its domains in isolation to predict the occurrence of adverse health outcomes such as dependence and hospitalization in community-dwelling older people.

Methods

The population for this study came from the Toledo Study of Healthy Ageing (TSHA). This is a population-based, prospective, longitudinal cohort study of individuals aged >65 years residing in the province of Toledo (Spain). The study has been described elsewhere (18). The study obtained local ethics committee approval and have therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. Participants signed written informed consent.

Questionnaires, objective measures and laboratory results were used to describe variables in several areas in three waves: 2006–2009, 2011–2013 and 2014–2017. Baseline data for this study came from the second wave because vitality domain items were measured only then, whereas follow-up data were obtained from the third one. For this analysis, individuals living in nursing homes and those with basic (BADL) or instrumental (IADL) activities of daily living dependence in the second wave were excluded.

Table 1 presents the correspondence of the IST and TSHA items. Some TSHA items had to be adapted. Any self-declared response involving impairment of vision or hearing on the TSHA was considered an impairment of these domains. In the IST, hearing is measured objectively. The vitality domain questions were taken from the Mini Nutritional Assessment (MNA) (19, 20) on both the IST and the TSHA, but on the TSHA question on weight loss, the adverb ‘unintentionally' was not included. The time orientation and three-word recall items came in both cases from the Mini-mental State Examination (21, 22), but the IST item on memory and orientation problems was approximated by an item of the Geriatric Depression Scale (GDS) (23, 24) that only mentions memory problems. Four other GDS items from the TSHA were used to mimic the two IST depression items.

Table 1.

Correspondence of the ICOPE screening tool and the Toledo Study of Healthy Ageing items, factor loadings of the ICOPE items and reasons to drop them from the measurement model (n=1032)

DOMAIN	ICOPE screening tool	Toledo Study of Healthy Ageing	Factor loading and decision
MOBILITY	Did the person complete five chair stands within 14 seconds?	Time to perform five chair stands. Preserved if it takes 14 seconds or less	Not applicable (only one item)
SENSORY: VISION	Do you have any problems with your eyes: difficulties in seeing far, reading, eye diseases or currently under medical treatment (e.g. diabetes, high blood pressure)?	Do you suffer from any visual impairment (how well can you see, allowing the use of glasses): Preserved vision: ‘Has no impairment, or this is minimal'. Impaired vision: ‘Has an impairment that makes difficult tasks like reading or sewing' or ‘has an impairment that makes difficult daily living activities, such as doing household chores, watching television' or ‘hardly sees, walks blindly around the house'.	−0.031 (p=0.831)→split sensory domain in two
SENSORY: HEARING	Hears whispers (whisper test) or screening audiometry result is 35 dB or less or passes automated app-based digits-in-noise test	Do you suffer from any hearing impairment? (how well can you hear, allowing the use of hearing aids): Preserved hearing: ‘Has no impairment, or this is minimal'. Impaired hearing: ‘Yes, suffers an impairment that prevents her from maintaining a normal conversation with a group in a noisy surrounding, but is able to maintain a normal conversation with a person' or ‘suffers an impairment that prevents her from maintaining a normal conversation with a person, there is a need to talk to him/her loudly and close to the ear' or ‘Yes, suffers an impairment that makes almost impossible to keep a conversation with her'	−0.537 (p=0.827))→split sensory domain in two
VITALITY	Appetite loss: Have you experienced loss of appetite? Weight loss: Have you unintentionally lost more than 3 kg over the last three months?	In the last 3 months, have you lost appetite? Have you lost more than 3 kg?	0.732 (p<0.001) 0.749 (p<0.001)
COGNITIVE	Do you have problems of memory or orientation (like not knowing where you are or what is today's date?) What is the full date today (year, month, day of the month, day of the week)? Recalls three words	Item from the Geriatric Depression Scale: Do you think you have more memory problems than most of the people? (Answer: Yes/no). Able to recall the year Able to recall the month Able to recall the day of the month Able to recall the day of the week Able to recall three words	A high modification index indicated a relationship with the vitality and emotion domains. Out of its nine correlation residuals, eight exceeded 0.1 in absolute value and three exceeded 0.2. All correlation residuals with the rest of the cognition items were negative - item dropped 0.628 (p<0.001) and correlation residuals of 0.151 with loss of weight and -0.163 with failing to recall the day of the week - item dropped 0.474 (p<0.001) and correlation residuals of 0.265 with weight loss and 0.161 with loss of appetite- item dropped 0.779 (p=0.072) 0.711 (p=0.089) <0.001 (p=0.996)→ item dropped
EMOTION	Over the past two weeks, have you been bothered by – feeling down, depressed or hopeless? – little interest or pleasure in doing things?	Items from the Geriatric Depression Scale (all answers: Yes/no): – Have you dropped many of your activities and interests? – Do you feel that your life is empty? – Do you feel happy most of the time? – Do you often feel helpless?	0.497 (p<0.001)→ item dropped 0.847 (p<0.001) 0.796 (p<0.001) 0.878 (p<0.001)

All items are scored 0/1 where 1 means preserved condition. In bold: factor loadings of the items that are kept in the final measurement model.

Outcome measures were defined as the occurrence of dependence in any of the six BADL (25) (excluding the need to tie shoelaces or have occasional continence ‘accidents') or eight IADL (26) at the third wave, and hospitalization for three years from the second wave, ascertained from the registry of the hospital to which individuals were referred (the Spanish health care system has a territorial structure where patients are referred to the hospital covering their area of residence and usually go to the same hospital for emergency care).

The covariates were age (in decades to avoid massive differences in the magnitude of variance), multimorbidity measured with the Charlson index (27), sex, and having completed at least primary education.

Statistical analysis

Variables are described by means (standard deviations) or counts (percentages).

A SEM was used. First, the measurement part of the model was tested with a confirmatory factor analysis (CFA) on the reflective relationship between the domains and their indicators. To enable identification, the reference variable method was used. Standardized factor loadings above 0.7, which explain almost half to the total variance of the indicator, were considered adequate (28). For this analysis, vision and hearing impairments were considered components of a single sensory domain. The items that remained in the CFA plus single-items domains were entered into a structural model. At this stage, six domains were considered, separating the sensory domain in its vision and hearing components. In its first version, associations between the domains and the three outcomes, and between the covariates and domains' indicators and the outcomes, were considered. In a second version, the formative construct IC composed of the six domains (with a constrained weight of one for one of them) was added in the causal pathway between the domains and the three outcomes. Statistically significant associations between specific domains and outcomes from the first version were added to the second version. Model fit was considered good if a non-significant (p≥0.05) chi-square exact-fit test and no evidence of bad local fit (29), as indicated by absolute correlation residuals above 0.10 (30), were found. Approximate fit indices were used to reinforce the assessment: A Bentler Comparative Fit Index (CFI) close to 1, a Standardized Root Mean Squared Residual (SRMR) ≤ 0.1 and a Root Mean Square Error of Approximation (RMSEA) whose lower bound 90% confidence interval (90%CI) included 0, whose upper bound was below 0.1 and whose p value for a RMSEA value below 0.05 was non-significant ((p≥0.05), were considered indicative of good fit (29). Modification indices (MI) greater than 10 were considered indicative of a need for model respecification, because they are associated with a level of significance of 0.01 in a chi-square test with one degree of freedom. The change in fit between these two nested models was tested with the Satorra-Bentler test (31). Finally, a variable for IC was created as the sum of the domains (out of six) without impairments in any of their items and a model was constructed with the three outcomes regressed on this variable and the covariates, without the domains. This model was simpler and allowed the error variance of age (assumed reliability 99% (32)) and the Charlson index (assumed reliability 91% (33)) to be fixed.

Since in all models the items and outcomes were dichotomous, a Diagonally Weighted Least Squares (DWLS) estimator with Nonlinear Minimization subject to Box Constraints (NLMINB) optimization method was used. We employed the ‘lavaan' library (34) of the R package (version 4.2.1) which, so far, does not handle missing data, so only complete data were used.

Results

In total, 2224 community-dwelling individuals were included in the second wave of the TSHA. In 45 cases (2%) the level of dependence could not be determined for either BADL or IADL, and 1445 were independent for both. In 287 cases (19.9%), the level of dependence at follow up could not be determined and in another 126 cases information on covariables, mainly the Charlson index, was missing. Thus, the final sample size for analyses was 1032 individuals. The excluded individuals were one year older (p<0.001), but no differences were found for sex (p=0.95), proportion with primary education or higher (p=0.948), or multimorbidity (p=0.125).

Table 2 describes the sample. The age range was 66 to 92 years, with a mean of 74 years. The maximum Charlson Index score was seven, with a mean of 0.7 (sd=1.1). By far, the least preserved indicator was the ability to recall three words (18%) followed by the ability to perform chair stands (54%). The mean number of intact domains was almost five and no individual had all domains impaired. While only 5% developed dependence for BADL at three-year follow-up, 22% developed dependence for IADL. Around 13% were hospitalized during this period.

Table 2.

Sample description (n=1032)

	Mean (sd)/n (percentage)
Age	73.5 (5.4)
Women	557 (54)
Charlson index	0.7 (1.1)
Education level
Less than primary	523 (50.7)
Primary	293 (28.4)
More than primary	216 (20.9)
Able to complete five chair stands in 14 seconds	566 (53.9)
Preserved vision	983 (95.3)
Preserved hearing	933 (90.4)
No weight loss	963 (93.7)
No appetite loss	847 (82.1)
No memory problems	891 (86.3)
Able to recall the day of the week	1000 (96.9)
Able to recall the day of the month	843 (81.7)
Able to recall the month	995 (96.4)
Able to recall the year	982 (95.2)
Able to recall three words	186 (18)
Not dropped activities nor interests	784 (76)
Not feeling that life is empty	921 (89.2)
Not feeling helpless	940 (91.1)
Feeling happy	948 (91.9)
Number of domains preserved	4.8 (1)
All domains preserved	284 (27.5)
Dependent for BADL at follow-up	54 (5.2)
Dependent for IADL at follow-up	223 (21.6)
Hospitalized in 3 years-time	130 (12.6)

BADL: Basic activities of daily living. IADL: Instrumental activities of daily living.

Measurement model

The initial CFA model with the sensory, vitality, emotion and cognition domains had a very poor test of exact fit (Satorra-Bentler scaled chi-square=174.6, degrees of freedom=71, p<0.001). The approximate fit indices were also low (CFI=0.887; RMSEA= 0.038, 90%CI: 0.031–0.045, p value RMSEA<=0.05=0.998; SRMR=0.119). As presented in Table 1, several items were successively eliminated due to low standardized factor loadings or high correlation residuals with items from other domains. The sequence of item deletions was: recall of three words, activities and interests, splitting of visual and hearing items of the sensory domain, no memory complaints, recalling the year, and recalling the month. After these adjustments, the fit was excellent (Satorra-Bentler scaled chi-square: 10.3, degrees of freedom: 11, p=0.501; CFI: 1.000; RMSEA: 0.000, 90%CI: 0.000–0.031, p value RMSEA<=0.05: 1; SRMR: 0.055), all factor loadings were above 0.7 and no MI exceeded 10. We found correlation residuals of recalling the day of the week with weight of 0.156 and with appetite of -0.178, but only the latter had a standardized version above 1.96 in absolute value (−2.089). This was our final reflective-measurement model.

Structural model

Our initial structural model showed a poor fit: Satorra-Bentler scaled chi-square=72.5, degrees of freedom=47, p=0.01; CFI=0.967; RMSEA=0.023, 90%CI=0.011–0.033, p value RMSEA<=0.05=1; SRMR=0.13. The highest MI (11.4) corresponded to the associations of the mobility and vision items. Vision had correlation residuals of −0.965, −0.311 and 0.287 with the weight, appetite, and mobility items. After including a path from vision to mobility, which meant that vision difficulties would hinder the sit-to-stand test, the model achieved a non-significant exact fit test (Satorra-Bentler scaled chi-square=61.1, degrees of freedom=46, p=0.068), with approximate fit indexes: CFI=0.980; RMSEA=0.018, 90%CI=0.000–0.029, p value RMSEA<=0.05=1. The SRMR remained almost the same (0.127). No MI pointed to a plausible association, but the correlation residuals of vision with the vitality items remained the same, although that of mobility was now 0.011. These huge correlation residuals led us to introduce a covariance between vision and the vitality domain, which improved fit. The correlation residuals of vision with weight and appetite dropped to −0.724 and −0.056, respectively. One MI pointed to an association between hearing and vitality and this correlation was entered into the model, which improved the fit even more (Satorra-Bentler scaled chi-square=48.3, degrees of freedom=44, p=0.303; CFI=0.994; RMSEA=0.01, 90%CI=0.000–0.024, p value RMSEA<=0.05=1; SRMR=0.101). No MI exceeded 10, but correlation residuals were still very high. Exploration of the two-by-two tables of the dichotomous items revealed that no individuals were impaired in both weight and vision. Zero cells in two-by-two dichotomous tables are known to produce problems in SEM with categorical variables. We decided to change one value of the vision variable to obtain an individual with weight loss and visual impairment. Since visual impairment was more prevalent among women and older individuals, we assigned the oldest woman (90 years) the presence of visual impairment. The rest of the analyses were performed with this modified sample.

With this change, the model fit remained virtually the same, although the SRMR decreased (Table 3, first column). No MI exceeded 10 and the correlation residual of vision and weight was now −0.144. By removing the covariance between vision and vitality, the correlation residual of vision with weight and appetite became −0.414 and −0.265, respectively, and the Satorra-Bentler test was significant (Chi-squared=6.34, df=1, p=0.012), so we maintained this covariance. The high correlation residuals of day of the week with weight and appetite remained, and there were residuals of 0.23 and −0.189 between dependence for BADL and sense of fulfilment and happiness, respectively. Nevertheless, we considered this structural model to be good enough. Figure 2 presents the statistically significant regression coefficients and the factor loadings of this model. Notably, the only domain that influenced dependence for activities of daily living was mobility. No domain influenced hospitalization.

Table 3.

Fit indexes and R² of the models with and without intrinsic capacity as a formative construct or as the addition of non-impaired domains

	Without intrinsic capacity	With intrinsic capacity	With intrinsic capacity as the sum of the non-impaired domains
Satorra-Bentler scaled chi-square, degrees of freedom, p value	48.6, 44, 0.292	52.1, 52, 0.469	84.3, 4, p<0.001
CFI	0.994	1	0.841
RMSEA (90%CI)	0.01 (0.000–0.024)	0.01 (0.000–0.02)	0.140 (0.115–0.166)
SRMR	0.067	0.071	0.000
R2 Dependency for basic activities of daily living	0.244	0.239	0.153
R2 Dependency for instrumental activities of daily living	0.265	0.262	0.251
R2 hospitalization	0.222	0.208	0.173

Figure 2.

Model of the domains of intrinsic capacity and adverse health outcomes (n=1032). Covariances among the exogenous constructs and the disturbance terms are omitted to preserve readability

BADL: Basic activities of daily living. IADL: Instrumental activities of daily living.

The model fit after introducing the formative latent variable IC between its six domains and the outcomes, and the R2 of the outcomes are presented in the second column of Table 3. It was not very different from the previous model and, in fact, the Satorra-Bentler test was non-significant (Chi-squared=3.59, df=8, p=0.892), meaning that IC did not improve the model fit. The regression coefficients of IC with dependence for BADL, IADL and hospitalization were 0.162 (p=0.167), −0.052 (p=0.546), 0.145 (p=0.167), respectively. The ‘lavaan' syntax of this model is given in the Appendix.

To perform sensitivity analyses, we changed one by one the vision status of the 22 women who were 75 years of age or older (about the median of that group) and experienced weight loss and compared the results. No differences were found that implied different conclusions.

Finally, Table 3 presents the model fit and the R² of the outcomes when IC was calculated as the sum of the domains. The fit was not good, but no MI pointed to plausible associations. The R² for BADL dependence and hospitalization were lower than in the other two models. Regression coefficients for IC on dependence for BADL, IADL and hospitalization were −0.065 (p=0.356), 0.08 (p=0.05) and −0.003 (p=0.949).

Discussion

We have found limitations in the way the IST measures the cognition domain: many of the items were not sufficiently related to it, it was not predicted by age or education, and it was not able to predict dependence. Furthermore, our results indicate that vision and hearing cannot be combined in the same domain. The only IST domain that explained dependence was mobility. A global IC measure such as the IST, operationalized either as a latent variable with different weights for each domain or as a sum of preserved domains, did not explain the development of dependence for BADL and IADL, nor hospitalizations over a three-year period above its domains.

The fit of our final measurement and structural models was good according to all fit indexes (including the most favorable possible values for CFI and RMSEA), but the SRMR, whose value was only acceptable due to the high correlation residuals of some indicators, especially the cognitive ones. We do not have an explanation for the poor performance of the cognitive items in our sample, because the MMSE and the way it was administered were standardized and supervised in the TSHA. First, the word recall item was not associated with the other items. Moreover, a very large proportion of the sample failed it. If, as proposed, failing any item of the IST would mean that the domain may be impaired, more than 80% of the population could have problems in the cognition domain and thus would have at least one impaired domain. Second, remembering the year or month did not correlate well with the items of remembering the day of the month or week. Admittedly, these items may reflect different severity of cognitive impairment, but one would expect them to be related. Our results warrant testing the factorial structure of the items of this domain in future research. That vision and hearing cannot be combined in the same domain is not surprising because of their different physiopathology and consequences, something that has been discussed previously (35, 36).

To our knowledge, this is the first time that an adequate specification of the measurement model of a tool similar to the IST has been evaluated. We believe that previous literature has not used the most appropriate IC measurement conceptualization. There are several reasons that favor the use of a formative conceptualization of IC. In the same way than smoking, hypertension and hypercholesterolemia constitute a latent variable called cardiovascular risk, or education, occupation and income determine socioeconomic status, the domains give rise to a given IC, not the other way around. For example, losing visual or hearing capacity decreases IC, but one does not begin to experience visual or hearing loss because of a decrease in IC. If the relationship between domains and IC were reflective, a change in IC would be transmitted to all domains simultaneously, whereas clinical experience shows that domains can change separately. If it were reflective, the construct validity of IC would not change by removing any of the domains (as when we remove only one item from a depression scale), because all facets of a unidimensional construct should be adequately represented by the remaining indicators. This is not the case for IC. Finally, indicators of a reflective construct should have positive associations (16). Although this is usually the case with IC domains, this does not happen always: impairment in any domain should not necessarily be accompanied by impairments in other domains.

In addition, Beard et al (5), Yu et al (6) and Aliberti et al (7) have used a bi-factor model. In this operationalization, indicators are assumed to be influenced by both the domain to which they belong and a general factor (IC), which are orthogonal (28). Apart from the limitation that bi-factor models are based on a reflective measurement model, proposing that IC is a factor unrelated (orthogonal) to the domains and not composed of them has a difficult biological justification.

To our knowledge, only one paper has attempted to model an IC tool as a formative construct before (15). However, this study did not use the IST items, assembled the scale by selecting variables from the different domains that were longitudinally associated with function, and did not test the overall fit of the proposed model.

González-Bautista et al (11), using a similar operationalization of the IST in community-dwelling older adults attending memory clinics in France, also found that mobility was the only domain that predicted BADL dependence but did not predict IADL dependence. Yu et al (37) observed in community-dwelling Chinese older people that mobility, but also cognition and emotion domains, predicted incident IADL disability after one year, and that mobility, vision and emotion predicted BADL incident disability. No domain predicted hospitalization. Their sample consisted of volunteers, they used a different operationalization of the cognition and emotion domains, their statistical approach was more conventional, and their follow-up time was shorter but with a higher incidence for BADL disability and hospitalization than ours. All these results together suggest that the IST measures most IC domains in a way that hampers their ability to predict dependence and hospitalization. The one exception is mobility, which is measured objectively as the ability to perform chair-stands, a task that has shown good predictive value for dependence (38).

If the domains lack explanatory power, it is difficult for a global index composed of them to have any predictive capacity. González-Bautista et al (11) referred that the global IST score did not predict dependence for BADL but predicted dependence for IADL over 5 years, but their sample had worse health status, and their definition of incident IADL dependence and statistical models were different from ours. It has been suggested previously that the IST may be too short to assess IC comprehensively (35).

The main strengths of this study are the use of a scale very similar to the IST; a more appropriate conceptualization of the measurement model in accordance with the way IC was firstly proposed, i.e. a composite (2); and the analysis of its longitudinal association with relevant adverse health outcomes in a large representative cohort of community-dwelling older people. Our large sample size provides robustness to the fit indexes estimated with weight least squares.

A first limitation is that several items had to be adapted from the original survey. This applies specially to hearing, which is self-reported in the TSHA but not in the IST. This may limit the reliability of the TSHA version. There could be more measurement error in the weight loss item version of the TSHA because it includes people who lost weight on purpose. The four items of the GDS scale are not exactly the same as the two items of the IST and the cognitive complaints item of the TSHA does not take into account orientation problems. After splitting the sensory domain into its components, three domains had only one item, which prevents the study of their factorial structure. Another limitation is missing data due to missing items and mainly losses to follow-up. The sample for the analysis was one year older than the target sample. The statistical application we used does not allow imputation of missing data, so our results may be affected by selection and follow-up biases. We did not obtain enough deaths to study the prediction of mortality, another highly relevant adverse health outcome. We were unable to correct for measurement error of continuous variables in our main model because this approach produced identification problems. This mainly affects the measure of multimorbidity, whose reliability is lower.

Conclusion

The lack of explanatory power for relevant adverse health outcomes for most of the IST domains found in this retrospective study with an adaptation of the original IST items suggests the need for their reformulation. This should be confirmed in prospective studies using the actual IST. If this is the case, the fact that the only domain that is objectively measured, mobility, predicts dependence suggests that objective measures should be favored. When new items are found, a suitable reflective-formative measurement model should be used to test whether their aggregation into an overall measure of IC adds any predictive power over and above the isolated domains. In the meantime, a second confirmatory step in the assessment of IC remains clearly indicated.

Funding:

work was supported by the Thematic Area for Frailty and Healthy Ageing of the Network of Biomedical Research Centers (CIBERFES), Instituto de Salud Carlos III, Madrid, Spain.

Conflict of Interest:

Ángel Rodríguez-Laso, Francisco José García-García, and Leocadio Rodríguez-Mañas declare that they have no conflict of interest.

Electronic Supplementary Material

Supplementary material is available for this article at https://doi.org/10.1007/s12603-023-1985-y and is accessible for authorized users.

APPENDIX. Lavaan syntax of the final structural model with intrinsic capacity as a formative latent construct