Participants were recruited from a recent Spanish epidemiological survey on aging conducted in June 2005. The study sample was composed of probabilistic and geographically-defined subsamples. More specifically, we obtained data on the prevalence of chronic geriatric, neurological and mental disorders from the principal investigators of nine studies conducted in Spain. Each location contributed geographically-defined or census-based random samples from the population originally surveyed in their respective studies (age ≥75).13–21 Subsamples were obtained from survivors residing in the following study areas, where the original surveys were conducted: Arosa, Bidasoa, Cantalejo, Gerona, Central Spain (NEDICES), Prat de Llobregat (PRATICON), Santiago de Compostela, Toledo and Zaragoza (fig. 1). Sampling continued until an average of 60 participants per location were recruited. Additional details of the sampling process in each location are provided in table 1. The number of participants to be sampled in each location was estimated by means of the mortality rates for the birth cohorts under study.22 Mortality was proportional to the delay from the original survey, as was the number of participants to be sampled. To avoid selection bias, the groups used their original census-based sampling procedure. In locations with a highly limited number of survivors, a new geographically-defined sample was obtained from selected city neighborhoods. Additional details on the sampling process are shown in table 1. Cohorts consisted of elderly people living at home and in residential care and in rural and urban settings. Significant losses were caused by death and the inability to locate individuals during the sampling process (fig. 2). These circumstances (particularly death) may not cause specific selection biases in a geographically-defined prevalent sample. Dead and non-located individuals (due to death or change of residence outside the geographic area) were not considered as part of the sampling frame. The duration of the follow-up from the prevalence date varied from 3 years for El Prat to 15 years for Gerona. Assessments were conducted in two successive visits to the individual's home. In the first visit, we conducted all health-related assessments. In the second visit, we retrieved all the information pertaining to personal and social factors (see the assessments section).

Figure 1.

Geographical location and size of samples comprising the Spanish Epidemiological Survey on Aging. 1) El Prat de Llobregat (Barcelona), n = 59. 2) Irún-Hondarribia (Guipúzcoa), n = 57. 3) Zaragoza, n = 31. 4) Gerona, n = 75. 5) Isla de Arosa (Pontevedra), n = 53. 6) Getafe (Madrid), n = 98. 7) Santiago de Compostela (La Coruña), n = 33. 8) Cantalejo (Segovia), n = 24. 9) Toledo, n = 73. With permission from Acta Neurologica Scandinavica.

(0.18MB).

Figure 2.

Sample attrition. With permission from BMC Public Health.

(0.17MB).

We implemented a two-phase screening survey within a cross-sectional design. The 12-item WHO-DAS II was administered to all eligible participants for screening purposes. Participants scoring above 0 were administered the 36-item WHO-DAS II.

Since dementia is the health condition contributing the most to disability in the elderly23 and is highly underdiagnosed in Spain,24 simultaneous cognitive screening was also administered, using the Spanish version of the Mini-Mental State Examination (MMSE).25,26

The human subjects review board of the Carlos III Health Institute (Madrid, Spain) approved the study protocol. All participants signed a written informed consent document drafted in accordance with the guidelines in the Helsinki Declaration. The participants were visited twice at their home or nursing home. In the first visit, cognitive screening through the Spanish version of the MMSE was administered. Further details on the diagnostic procedure for dementia and the study design are available in a methodological study published as part of the present project11 and in a subsequent study focused solely on the prevalence of disability.12

• 1)

  Mini-Mental State Examination (MMSE)

  The MMSE is a standard screening measure of cognitive deterioration and has been validated in the Spanish elderly population25,26 (cut-off: < 24).

• 2)

  Disability assessment (WHO-DAS II)

  The WHO-DAS II (36-item) is a self-reported scale assessing six disability domains referring to the month preceding administration:10 understanding and communication (UAC), getting around (GAR), self-care (SCA), getting along with people (GAP), life activities (LAC), and participation in society (PSO) (table 2). Each item is scored on a 5-point Likert scale and establishes the difficulty experienced by the respondent in performing a given activity (no problem, mild, moderate, severe, and extreme difficulty).The scale has been translated to Spanish and adapted to the Spanish elderly population27 showing appropriate data quality, acceptability, scale scores and internal consistency (Cronbach's alpha by domain: 0.71-0.96; Cronbach's alpha for the summary index: 0.93). Similar findings were reported in a validation study incorporating a global sample with chronic diseases, which also evaluated responsiveness and latent structure.28 The empirical literature on the WHO-DAS II has been reviewed by Federici et al.29

  Table 2.

  Qualitative definition of WHO-DAS II subscales and ICF disability levels. With permission from BMC Public Health.

  WHO-DAS II subscales	Definitiona
  UAC (6, 0.81)b	Difficulty concentrating on something for more than 10minutes and learning new tasks
  GAR (5, 0.88)	Difficulty standing for long periods, moving around the house and getting out of the house
  SCA (4, 0.71)	Difficulty in bathing, getting dressed, feeding and being independent while being alone
  GAP (5, 0.77)	Difficulty in social activities such as starting and maintaining a conversation, dealing with unknown people and maintaining or making new friends
  LAC (4. 0.96)	Difficulty in performing instrumental activities quickly and effectively, particularly household duties
  PSO (8, 0.95)	Difficulty in joining community activities such as festivities or religious events. Lack of self-confidence due to health problems
  Summary Index (32, 0.93)	Average score across the six domains
  ICF disability levels	Definitionc
  No problem (0-4%)	No problem as measured by standardized instruments (5% error allowed)
  Mild disability (5-24%)	Problem that occupies up to a fourth of the time or, alternatively, the first fourth of the score range of a standardized instrument on self-reported difficulty for an activity/participation
  Moderate disability (25-49%)	Problem that occupies up to a half of the time or, alternatively, half of the score range of a standardized instrument on self-reported difficulty for an activity/participation
  Severe disability (50-95%)	Problem that occupies up to 95% of the time or, alternatively, a score on a standardized instrument on self-reported difficulty for an activity /participation up to 95% of the score range
  Extreme disability (96-100%)	Complete problem as measured by standardized instruments; 5% error allowed

  UAC: understanding and communication; GAP: getting along with people; LAC: life activities; GAR: getting around; PSO: participation in society; SCA: self-care. Activity: execution of a task or action by an individual; Participation: involvement in a life situation; Problem: self-reported difficulty/number of days with disability over the last month in actual performance or abstract capacity (no environmental aids) to complete an activity.

  a

  According to Vázquez-Barquero et al., 2006, p. 88.

  b

  Number of items, Cronbach's α.

  c

  According to WHO, 2001, p. 34.

  Work items were omitted. Life activities items were not computed in participants with no household duties. Finally, we decided not to compute item D4.5 (sexuality) due to the high number of refusals to respond. Missing data per item (criterion ≤ 10%) and fully computable scores were within the standard range of usable values (criterion ≥ 90%).30

  Each domain and the complete scale generate a summary index ranging from 1 to 100, with higher scores indicating greater disability. Scale scores were transformed into ICF disability categories before any statistical analyses were conducted: no problem (0%-4%), mild problem (5%-24%), moderate problem (25%-49%), severe problem (50%-95%), and extreme problem (96%-100%).9

• 3)

  Health conditions affecting disability

  Morbidity was identified by licensed physicians through direct medical examination and perusal of medical records on the basis of a pre-established list of prevalent diseases and health conditions in the elderly (cf. Swedish National Study of Aging and Care in Kungsholmen; www.aldrecentrum.se/snack). These conditions were used as independent variables both individually and grouped under their respective International Classification of Diseases, 10th Revision (ICD-10) chapter: circulatory system (angina pectoris, arrhythmia, myocardial infarction, heart failure, hypertension, other), respiratory system (asthma, chronic obstructive pulmonary disease, other), infectious diseases (meningitis, poliomyelitis, tuberculosis, Lyme disease, other), nervous system (epilepsy, Parkinson's disease, other), musculoskeletal and injuries of external cause (arthritis, arthrosis, osteoporosis, polymyalgia rheumatica, renal failure, bone fracture, vertebral lesion), skin and subcutaneous tissue (lupus, vasculitis, other), endocrine system (diabetes type I and II, hyperthyroidism, hypothyroidism, other), eye (cataract, glaucoma, age-related macular degeneration, other), neoplasm, and mental and behavioral disorders (depression, psychosis, dementia, psychological distress, developmental disabilities, other). Smoking was grouped with respiratory diseases. Cerebrovascular disease, brain damage and head trauma were grouped with neurological diseases. Morbidity was defined as the total number of identified conditions. Since very few participants had more than one condition in a given category (e.g., neurological), disease groupings were computed as binary variables when used as independent variables in logistic regression analyses (0, no condition; 1, one or more conditions in that category).

  Given the strong impact of depression on disability,31,32 the presence of depressive symptoms was assessed formally using the EURO-D scale.33,34 In addition, depression was assessed clinically by the licensed physician conducting the health survey (gerontologist, neurologist or psychiatrist).

• 4)

  Personal, and environmental factors

  In keeping with the ICF framework, personal and environmental factors were also assessed: (i) educational level: illiterate, primary incomplete, primary, and higher than primary; (ii) self-reported socioeconomic status over a 5-point scale (1: very low; 5: very high); (iii) frequency of social contacts, measured by the median of self-reported frequency of telephone and in-person contacts with children, extended family and friends (daily, weekly, bi-weekly, monthly, never); (iv) availability of a confidant, specifically, the self-reported presence of someone with whom the individual could share emotional experiences and life concerns (confidants were considered a form of social support); (v) availability of social and health resources, measured by the median of accessibility to the following six social and health resources on a 5-point scale (1: very accessible; 5: very hard to access): primary-care center/medical attention, public transportation, public leisure resources (garden, park), shops, social clubs/cultural resources, and religious services; (vi) municipality size: rural: 1–10,000 inhabitants; urban: > 10,000 inhabitants (source: Spanish National Institute of Statistics, 2007).

Binary logistic regression was used to identify disability-related factors, including health conditions (either grouped or as specific diagnoses) and personal and environmental variables. We defined the outcome variable of interest on the basis of the ICF level of disability: participants with severe or extreme disability levels, and participants with no disability, low, or moderate disability. This dichotomy divided the distribution of participants in two equally sized groups, thus maximizing statistical power. In addition, several preliminary trials using ordinal logistic regression failed to generate proportional odds across the levels of the outcome variable. Odds ratios (OR) and 95% confidence intervals (95%CI) were computed, controlling for age (continuous variable) and sex in all models. In addition, when examining personal and social factors, we also controlled for education, depression (EURO-D score), cognitive function (MMSE score), and morbidity.

The significance of sets of multiple independent variables was evaluated with the Wald χ2 tests, using design-adjusted, multiply-imputed coefficient variance–covariance matrices. Interaction effects among strong independent variables were analyzed to support the identification of specific groups of individuals that were more likely to be disabled. While we could have used several rank-ordered definitions of cases (e.g., mild, severe disability) to implement ordinal regression analyses, failure to meet proportional odds assumptions prevented this analytical strategy.

Our analytical strategy was guided by the ICF framework. Therefore, we developed two separate sets of models based on logistic regression analysis. These two sets of models explored the three groups of disability factors that are prominent in the ICF, namely health conditions, personal factors, and environmental factors of disability. The first set of models examined health factors across disability domains using groups of clinical diagnoses as putative indicators for health conditions. These models were subsequently replicated with specific prevailing diagnoses as independent variables; only prevalent conditions were used as independent variables. A second model incorporated independent variables of disability based on personal and environmental factors (e.g., educational level, access to social and health resources in the community). Both models were replicated across WHO-DAS II disability domains. The ICF model allows for complex interactions among disability factors. Therefore, we conducted additional logistic regression analyses incorporating health conditions, environmental and personal factors (including interaction factors between them) identified in the models initially obtained. With this strategy we aimed to identify specific subgroups that were more likely to be disabled.

We entered all variables in the model in a single step without checking for significant effects a priori. Categories of clinical diagnoses were grouped according to the ICD-10 classification. Comorbidity in each group was added as an independent variable. Specific health conditions used as independent variables in the subsequent model were selected based on the number of participants with the condition (20 or more). In the case of personal and environmental factors, we selected a set of independent variables that are known predictors for numerous health outcomes (education, social status, social contacts, municipality size, etc.). Finally, as indicated above, the examination of interaction effects was based on the performance of the selected variables in the initial model. Factors that were significantly associated with disability in single factor analyses were subsequently selected to study interaction. We examined the interaction of specific health conditions with environmental and personal factors in order to support the view that health conditions can act jointly with environmental and personal factors and affect disability. Interaction analyses were aimed at identifying groups of individuals that were particularly disabled. Factors that were significantly associated with disability in single factor analyses were subsequently selected to study interaction. First-degree interaction factors were checked in subsequent models. Finally, we entered all variables deemed relevant in the model providing results in a single step. Otherwise, priority was given to interaction factors and reclassification in broader exposure categories.

Etiological fractions and 95%CI for independent variables showing the highest contribution to prevalent disability status were calculated as:

For grouped conditions, only mental and behavioral (OR: 6.52; 95%CI: 3.31-12.85, number of individuals with a psychiatric disorder and severe/extreme disability [hereinafter n] = 146), and neurological disorders, including cerebrovascular disease (OR: 2.81; 95%CI: 1.33-5.92, n = 85), were significantly associated with severe/extreme disability (table 4). Notably, the category of mental and behavioral disorders was the only independent variable significantly associated with all domains of disability. With more specific conditions, cerebrovascular disease was a significant independent variable of disability status for GAR, SCA, PSO (OR range: 3.16 to 4.91, n = 49) and for disability status based on the WHO-DAS II summary index (OR: 4.42; 95%CI: 1.67-3.45; table 5). Among mental and behavioral disorders, diagnoses of depression and Alzheimer's disease were strongly associated with severe/extreme disability. A clinical diagnosis of depression was significant for the GAR, GAP, LAC and PSO domains (OR range: 2.36 to 3.40, n = 74). The presence of depressive symptoms (EURO-D) was significantly associated with GAR, SCA, LAC and PSO (OR range: 1.11 to 1.23, n [EURO-D > 0] = 249) and the WHO-DAS II summary index (OR: 1.18; 95%CI: 1.08-1.30). Alzheimer's disease proved to have the strongest effect across all disability domains (OR range: 4.86 to 29.44, n = 41), as well as for disability status based on the WHO-DAS II summary index (OR: 17.40; 95%CI: 6.85-44.22). Normal cognitive functioning established by the MMSE ≥ 24 was associated with a low level of disability in UAC, SCA, GAP and LAC (OR range: 0.82 to 0.90, n = 405) and the WHO-DAS II summary index (OR: 0.82; 95%CI: 0.77-0.88).

Osteoporosis acted as a disability factor but was significant only for the GAP (OR: 3.06; 95%CI: 1.01-9.29, n = 31) and PSO domains (OR: 3.41; 95%CI: 1.32-8.84). A diagnosis of hypertension was associated with low disability but was only significant for UAC (OR: 0.42; 95%CI: 0.21-0.84, n = 268) and the WHO-DAS II summary index (OR: 0.49; 95%CI: 0.25-0.97). Smoking was strongly associated with UAC disability (OR: 3.53; 95%CI: 1.16-10.79, n = 54) (table 3).

Etiological prevalence fractions indicated that mental (psychiatric conditions and dementia) and neurological disorders (major ICD categories) accounted for 59.76% (95%CI: 49.26-65.09) and 20.21% (95%CI: 7.78-26.07) of severe/extreme disability status, respectively. The specific conditions making the highest contribution to disability status were Alzheimer's disease and depression, with etiological fractions of 31.42% (95%CI: 28.47-59.76) and 18.62% (95%CI: 9.85-20.21), respectively.

Educational level was negatively associated with disability in the GAR and PSO domains (OR range: 0.21 to 0.26, n [illiterate] = 49, n [some secondary or higher] = 73). Frequent social contacts were associated with no/low disability in the SCA and GAP disability domains (OR range: 0.13 to 0.20, n [≤1 a month] = 105, n [daily] = 127). The availability of a confidant was associated with low disability, not only for SCA and GAP (OR range: 0.25 to 0.33, n = 405), but also for disability status based on the WHO-DAS II summary index (OR: 0.35; 95%CI: 0.13-0.91) (table 4).

Accessibility to social and health resources, and municipality size were associated with low disability. The effect of living in a municipality of 10,000 or more inhabitants was significant for the GAP disability domain (OR: 0.42; 95%CI: 0.22-0.82, n [>10,000 inhabitants] = 329). Accessibility to social and health resources was strongly associated with low disability in all domains. The effect showed some signs of being dose-dependent for all disability domains (difficult access, OR: 0.11-0.89; neither easy nor difficult, OR: 0.04-0.89; easy access, OR: 0.01-0.12; very easy access, OR: 0.01-0.09, n [very difficult] = 36, n [very easy] = 162) and the WHO-DAS II summary index (difficult access, OR: 0.18, 95%CI: 0.03-0.90; neither easy nor difficult, OR: 0.19, 95%CI: 0.03-1.07; easy access, OR: 0.02, 95%CI: 0.00-0.10; very easy access, OR: 0.05, 95%CI: 0.01-0.24). The effect was highly significant across all disability domains but was particularly so for GAR and PSO (table 6).

Additional binary logistic regression models were computed to analyze the potential interaction among major disease-related and environmental factors of disability. Specifically, the interaction among the presence of neurological disorders and accessibility to health and social resources, and the presence of mental disorders and accessibility to health and social resources was analyzed independently. For the purposes of this analysis, service accessibility was transformed into a binary variable (1: very difficult access; 0: not very difficult access). The interaction between very difficult access to services and having dementia or other psychiatric disorders was highly significant (OR: 66.06; 95%CI: 21.20-205.78, individuals with both characteristics = 19, reference, double negative individuals). In addition, there was also significant access to services by neurological disorder interaction effect (OR: 12.74, 95%CI: 2.78-58.64, n = 4). In terms of etiological fractions, these interactions accounted for 36.69% (95%CI: 35.50-37.07) and 7.23% (95%CI: 5.02-7.71) of severe-extreme disability status, respectively.