VA Research and Development LOGO

Logo for the Journal of Rehab R and D
Volume 41 Number 6A, November/December 2004
Pages 835 — 846

Qualitative profiles of disability

Roberta Annicchiarico, MD, PhD; Karina Gibert, PhD, Associate Professor; Ulises Cortés, PhD, Associate
Professor; Fabio Campana, MD, PhD; Carlo Caltagirone, MD, Professor

IRCCS (Istituto di Ricovero e Cura a Carattere Scientifico) Fondazione Santa Lucia, via Ardeatina 306, 00179 Rome, Italy; Department of Statistics and Operation Research, Universitat Politčcnica de Catalunya, Pau Gargallo 3, 08028 Barcelona, Spain; Software Department, Universitat Politčcnica de Catalunya, c/Jordi Girona 1-3, 08034 Barcelona, Spain; Centro Assistenza Domiciliare ASL RM E, Viale di Valle Aurelia 115/a, 00168 Rome, Italy; Department of
Neurology, Universitŕ Tor Vergata, via O. Raimondo 18, 00173 Rome, Italy
Abstract — This study identified profiles of functional disability (FD) paralleled by increasing levels of disability. We assessed 96 subjects using the World Health Organization Disability Assessment Schedule II (WHODAS II). Clustering Based on Rules (ClBR) (a hybrid technique of Statistics and Artificial Intelligence) was used in the analysis. Four groups of subjects with different profiles of FD were ordered according to an increasing degree of disability: "Low," self-dependent subjects with no physical or emotional problems; "Intermediate I," subjects with low or moderate physical and emotional disability, with high perception of disability; "Intermediate II," subjects with moderate or severe disability concerning only physical problems related to self-dependency, without emotional problems; and "High," subjects with the highest degree of disability, both physical and emotional. The order of the four classes is paralleled by a significant difference (<0.001) in the WHODAS II standardized global score. In this paper, a new ontology for the knowledge of FD, based on the use of ClBR, is proposed. The definition of four classes, qualitatively different and with an increasing degree of FD, helps to appropriately place each patient in a group of individuals with a similar profile of disability and to propose standardized treatments for these groups.

Key words: artificial intelligence, cluster analysis, disability profiles, functional disability, Knowledge Discovery, qualitative analysis, rehabilitation.

Abbreviations: AI = artificial intelligence, ClBR = Clustering Based on Rules, FD = functional disability, KB = knowledge base, WHODAS II = World Health Organization Disability Assessment Schedule II.

The work published in this manuscript was unfunded at the time of manuscript preparation.

Address all correspondence and requests for reprints to Roberta Annicchiarico, MD, PhD; IRCCS Fondazione Santa Lucia, via Ardeatina 306, 00179 Rome, Italy; +39-06-5150-1516; fax: +39-06-503-2097; email:

DOI: 10.1682/JRRD.2004.02.0016

Disability is usually defined as the difficulty or inability to independently perform basic activities of daily living or other tasks essential for independent living without assistance. Disability has no clear limits, and defining different levels in different patients is very difficult, particularly when referring to functional disability (FD). FD has to be intended as the result of the interaction of different individual components of compromised functions: physical, emotional, and cognitive aspects usually interact to produce a comprehensive disability, which is more than the simple addition of the single impairments, affecting the patient's global function and his or her self-dependency [1]. From a rehabilitation perspective, patients' functioning and health are associated with, but not merely a consequence of, a condition or disease. Furthermore, functioning and health must be seen in associated with a condition but also associated with personal and environmental factors and context. Rehabilitation interventions are targeted toward a person's functioning and health [2]. A multidisciplinary involvement in comprehensive rehabilitation programs has proven effective [3], and the positive interaction of mental and physical training has been studied [4-9].

Recent debate on the evaluation of treatment efficacy and effectiveness has requested that the treatments be objectively defined and protocol-based. These treatments cannot be defined without also specifying the nature of the population being treated [10]. The possibility of defining reliable profiles of disability represents a prerequisite for proposing standardized rehabilitative treatments for patients sharing the same pattern of FD-patients who at that point will be allowed to undergo a program of comprehensive rehabilitation instead of the treatment of single compromised functions.

However, the scientific community lacks consensus on how to measure disability and how to individuate subgroups of disabled individuals [11]. Many assessment scales have been proposed to correctly define disability. Clinicians' most widely used approach is to produce an absolute or relative score for each assessed patient [12-13] to classify him or her or his or her disability on a scale or a range of values. In these terms, the disability scores of each scale constitute an increasing or decreasing continuum of values related to individual degree of self-dependency. For one to better interpret data resulting from assessment scales, a possible solution is to identify different subgroups defined by cutoff values, which possibly can be associated with common characteristics. However, summarizing a multidimensional evaluation in a single result is difficult and these scores would then hardly play a substantial role in clinical practice and decision-making.

Cluster analysis is a more sophisticated approach that distinguishes (with higher definition) different groups of subjects on the basis of similar individual attributes. Cluster analysis allows the identification of groups of patients affected by different pathologies but with similar symptoms such as motor, cognitive, or mood alterations [14-16]. It has been applied in studies of psychiatric patients [17] and patients suffering from an alteration of pain perception [18]. However, the classical clustering techniques do not recognize the structure of certain sets of complex data and produce some nonsense classes [19]-in clinical terms-that experts cannot interpret and that do not correspond to real concepts in the medical domain. In fact, this occurs when dealing with "ill-structured domains" [20], in which relationships between variables are complex, either numerical or qualitative information has to be treated together, and additional semantic knowledge is available but not considered by classical clustering techniques.

Since the results of classical clustering methods are frequently difficult to apply clinically [21], we propose the interaction between statistical and artificial intelligence (AI) methods, which has already been successfully applied in the medical field [19,22-24]. The method is known as Clustering Based on Rules (ClBR) [25]. It is a hybrid technique that combines some inductive learning elements with statistical methods to improve clustering results on ill-structured domains. In fact, one of the main advantages of this approach is that it guarantees the semantic meaning of the resulting classes. Previously, it was successfully used for Knowledge Discovery in several real applications [19,26-27]. Discovering the underlying classes of a domain helps to organize and make explicit the knowledge about it, and it allows establishing what is known as ontology in AI environments [28].

The main goal of this research is to identify disability profiles on the basis of similar functional problems-possibly paralleled by increasing levels of disability-subsequently to propose an ontology for the target phenomenon. This attempt relies on the assumptions that individuating the multiple components of FD will help clinicians in making decisions about the most effective treatment strategies and that grouping patients in qualitatively defined profiles will provide the basis for standardized treatment.


The sample included 96 subjects, 58 males (60.4%) and 38 females (39.6%); mean age was 56 years. Of these 96, 76 were neurological patients, aged 18 to 80 years, recovering from October 1999 to February 2000 at the IRCCS (Istituto di Ricovero e Cura a Carattere Scientifico) S. Lucia hospital in Rome, Italy. They included 20 spinal cord injury patients (mean age 47.20 ± 17.6 standard deviation [SD]), 20 Parkinson disease patients (mean age 69.25 ± 6.53 SD), 20 stroke patients (mean age 63.40 ± 15.96 SD), and 16 depressed patients (mean age 46.56 ± 11.15 SD). The control group included 20 healthy subjects (mean age 55.05 ± 15.57 SD). A trained physician interviewed each subject according to the World Health Organization Disability Assessment Schedule II (WHODAS II) scale. The only inclusion criterion was the absence of cognitive disorders, according to a prior clinical evaluation by a neurologist.


We derived data from use of the WHO evaluation scale, which provides the necessary information about disability on the basis of quality and quantity. The WHO-DAS II [29] is a multidimensional evaluation scale for measuring level of functioning and disability across a variety of conditions. One added advantage of the WHODAS II is that it assesses functioning and disability at the individual level instead of the disorder-specific level. The WHODAS II is undergoing reliability and validity testing (in 16 centers across 13 countries) and health services research studies to test sensitivity to change and predictive validity [30]. The WHODAS II has shown an excellent internal validity and convergent validity in the primary care setting [31], and it has been used in the study of depressive disorders [32].

In the same set, the WHODAS II incorporates instruments for the detection of both physical and mental health factors related to disability. The WHODAS II version 3.1a (June 1999, interviewer-administered version) is a scale containing 96 items for the assessment of disability levels according to the ICF (International Classification of Functioning and Disability). This interview measures self-reported difficulty in functioning in six major domains considered important in most cultures: Understanding and Communicating (six items), Getting Around (five items), Self-Care (four items), Getting Along with People (five items), Life Activities (eight items), and Participation in Society (eight items). The WHODAS II employs a 5-point rating scale for all items in which "1" indicates no difficulty and "5" indicates extreme difficulty or inability to perform the activity. The WHODAS II standardized global score ranges from 0 (nondisabled) to 100 (maximum disability).


ClBR was used to analyze data collected from the sample of 96 subjects. Gibert and Cortés originally presented this method in detail [25]. It consists of a mixture of two main elements: on one hand, an AI process that manages a knowledge base (KB), which includes prior medical knowledge, even if partial, and on the other hand, a later clustering process that is biased on the basis of some induction on the KB.

The KB is processed and used to induce a first structure on the sample, mainly consisting of the identification of certain groups of patients, called rules-induced classes. If the KB is partial, it will not involve certain objects, and these will constitute what is called the residual class. Then clustering is carried out in every rule-induced class; a prototype of each rule-induced class is built and clustered together with the residual class, thus constructing a single global hierarchy including all the objects.

The additional knowledge provided by the expert is expressed by means of logical rules, with the structure:

ccurrence of A implies membership toC,

which is read as "occurrence of A implies membership to C," where A is a Boolean expression either on the variables of the database or some transformation on them, and C is a group label. The KB is a set of rules r, and it may be a partial description of the domain: in ill-structured domains, building a complete explicit KB for the domain is usually difficult because of the large quantity of implicit knowledge experts unconsciously use.

With respect to the underlying clustering method, in this particular application, the reciprocal neighbors algorithm is used, with Ward criteria (which is based on the concept of inertia and has the property of giving priority to the formation of homogeneous groups of objects). A particular property of the ClBR is that it allows treatment of heterogeneous databases (with both numerical and categorical variables), measuring distances between objects by means of "Gibert mixed metrics" [33]:

measuring distances between objects by means of Gibert mixed metrics

where χ ik is the value taken by variable χk for object i, χι'κ is the value taken by variable Χκ for object ι', and n2 is the number of qualitative variables with rewritten equation a rewriting of x2 metrics, calculated with qualitative variables and supporting symbolic representation. The rules used in this application concern the items of the WHODAS II that inquire about emotional behavior.

Four of the test questions are related to emotional behavior. They are-

1. B4 = How do you rate your mental or emotional health in the past 30 days?
2. B9 = How much worry or distress have you had about your health in the past 30 days?
3. S5 = How much have you been emotionally affected by your health condition?
4. R2 = How much have the difficulties been caused by mental health or emotional problems?

People who give higher values to these questions are considered to have an emotional problem of some kind. So, the proposed KB for biasing the cluster is where r1, r2, r3, and r4 are the "rules"; B4, B9, S5, and R2 are the WHODAS II items just reported; [4,5] are the highest (worst) scores for the items considered, meaning high (4) or extremely high (5) difficulties.

KB =


r1: If B4 is in [4,5], then emotional problems,
r2: If B9 is in [4,5], then emotional problems,
r3: If S5 is in [4,5], then emotional problems,
r4: If R2 is in [4,5], then emotional problems},

One can interpret classification in a nonautomatic way by analyzing conditional distributions of the WHO-DAS II items against classes and using medical expertise. The significance of differences among classes for different variables is assessed with the Kruskal-Wallis test (since nonnormality holds) for numerical variables and x2independence test for qualitative ones. Differences of mean standardized global scores among classes are analyzed with analysis of variance (ANOVA).


ClBR suggested a set of seven classes, three of which contained isolated patients confirmed as outliers (Figure). The four remaining classes were kept for interpretation. The classes identified four groups of subjects with different profiles of FD and could be ordered according to an increasing degree of disability. The most characteristic items in each class were also identified (fields in gray in the Table). The Table reports the mean scores ± SD of each item in the four classes.

Figure Dendrogram for ClBR (Clustering Based on Rules).

Taxonomy of four qualitative profiles of increasing disability. Mean scores, mean standardized global scores (standard deviations [SDs]), and qualitative rank of most characteristic items are included.
Functional Disability
Intermediate I
Intermediate II
Physical Problems (B1)
Emotional Problems (B3)
Physical Health (B2)
Mean score qualitative rank
2.19 good
2.83 moderate
2.83 moderate
3.25 extreme
Worry or Distress (B9)
Mean score qualitative rank
1.77 mild
3.50 mild/moderate
2.50 mild/moderate
3.93 extreme
Standing Up for 30 Minutes (S1)
(difficulty in)
Mean score qualitative rank
1.43 none
2.70 moderate
4.16 extreme
3.43 moderate
Walking for a Long Distance (S7)
(difficulty in)
Mean score qualitative rank
1.74 none
3.66 extreme
4.66 extreme
4.03 extreme
Toileting Whole Body (S8)
(difficulty in)
Mean score qualitative rank
1.19 none
2.75 moderate
3.50 extreme
3.46 extreme
Dressing (S9)
(difficulty in)
Mean score qualitative rank
1.16 none
2.16 mild/moderate
3.66 extreme
3.21 moderate/ extreme
Difficulty Due to Physical
Problems (R1)
Mean score qualitative rank
1.74 none
3.66 extreme
3.83 extreme
3.43 extreme
Difficulty Due to Mental
Problems (R2)
Mean score qualitative rank
1.32 none
2.45 moderate
1.33 moderate
3.21 extreme
Concentrating for
10 Minutes (S6)
(difficulty in)
Mean score qualitative rank
1.41 none
2.16 mild
1.16 none
2.81 extreme
Daily Work (S12)
(difficulty in)
Mean score qualitative rank
1.51 none
2.83 mild
3.33 moderate/ extreme
4.18 extreme
Learning a New Task (S3)
(difficulty in)
Mean score qualitative rank
1.22 none
1.41 mild
1.16 none
2.81 extreme
Participating in Community
Activities (S4)
(difficulty in)
Mean score qualitative rank
1.25 none
1.41 mild
2.0 moderate
3.46 extreme
Emotional Involvement in
Life Conditions
(S5) (perceived)
Mean score qualitative rank
1.67 mild
3.25 extreme
2.50 moderate
4.15 extreme
In Daily Work Related to
Interaction with
Other Persons (P4.1)
(difficulty in)
Mean score qualitative rank
1.29 none
2.29 moderate
2.16 moderate
3.00 extreme
Standardized Global Score
Mean ± SD
10.4 ± 10.5
27.3 ± 10.6
34.1 ± 15.4
48.6 ± 12.6
Note: Fields in gray refer to items that qualitatively characterize four classes. Codes in parentheses refer to original World Health Organization Disability Assessment Schedule II (WHODAS II) items:
B1 = Do you have any physical health problems?
B3 = Do you have any mental or emotional health problems?
B2 = How do you rate your physical health in the past 30 days?
B9 = How much worry or distress have you had about your health in the past 30 days?
S1 = How much difficulty did you have in standing for long periods such as 30 minutes?
S7 = How much difficulty did you have in walking a long distance such as a kilometer?
S8 = How much difficulty did you have in washing your whole body?
S9 = How much difficulty did you have in getting dressed?
R1 = How much have the difficulties been caused by physical health problems?
R2 = How much have the difficulties been caused by mental health or emotional problems?
S6 = How much difficulty did you have in concentrating on doing something for 10 minutes?
S12 = How much difficulty did you have in your day-to-day work?
S3 = How much difficulty did you have in learning a new task, for example, learning how to get to a new place?
S4 = How much of a problem did you have joining in community activities (i.e., festivities, religious, or other activities) in the same ways as anyone else can?
S5 = How much have you been emotionally affected by your health condition?
P4.1 = How much did these difficulties in getting along with people interfere with your life?

On the basis of the conditional distribution of the scores of different variables versus classes, a qualitative description was made of features outlining major differences among classes (codes in parentheses refer to the original WHODAS II items):

· Cr93 (mean ± SD standardized global score: 10.4 ± 10.5): (B2) good physical health, (B9) mild worry or distress, (S1) can stand up for 30 minutes, (S7) no difficulty in walking for a long distance, (S8) no difficulty in toileting the whole body or (S9) dressing, (R1) no difficulty related to physical or (R2) mental problems, (B1) 45 percent of the subjects reported some physical problems, and (B3) 35 percent some emotional problems.
· Cd52 (mean ± SD standardized global score: 27.3 ± 10.6): (B2) moderate physical health, (B9) mild or moderate worry or distress, (S1) moderate difficulty in standing up for 30 minutes, (S6) mild difficulty in concentrating for 10 minutes, (S7) extreme difficulty in walking for a long distance, (S12) mild difficulty in daily work, (R1) extreme difficulty related to physical problems, (R2) moderate difficulty related to emotional problems, (B1) 91 percent of the subjects reported some physical problems, and (B3) 71 percent reported some emotional problems.
· Cr89 (mean ± SD standardized global score: 34.1 ± 15.4): (B2) moderate physical health, (B9) mild or moderate worry or distress, (S1) extreme difficulty in standing up for 30 minutes, (S7) extreme difficulty in walking for a long distance, (S8) extreme difficulty in toileting the whole body and (S9) dressing, (R1) extreme difficulty related to physical problems, (R2) moderate difficulty related to emotional problems, (B1) 100 percent of the subjects reported some physical problems, and (B3) none reported emotional problems.
· Cd53 (mean ± SD standardized global score: 48.6 ± 12.6): (B2) extremely bad physical health, (B9) extreme worry or distress, (S1) moderate difficulty in standing up for 30 minutes, (S3) extreme difficulty in learning a new task, (S4) extreme difficulty in participating in community activities, (S5) extreme emotional involvement in life conditions, (S6) extreme difficulty in concentrating for 10 minutes, (S7) extreme difficulty in walking for a long distance, (P4.1) extreme difficulty in daily work related to interaction with others, (B1) 75 percent of the subjects reported some physical problems, and (B3) 90 percent reported some emotional problems.

The mean standardized score shows a clear increase and statistically significant difference (<0.001) among the four profiles (Table).


The efficacy and effectiveness of rehabilitative treatment [34] are of critical concern for their implications for quality of life, healthcare policies, and healthcare expenditures [10]. The need for proving the cost-effectiveness of different treatments has forced the debate and the research toward a better definition of rehabilitative care. Main topics of the debate are the treatment itself, the treatment outcome, and the population to be treated. In these terms, for one to respond to both the needs of standardized treatments and individualized care, the question to ask becomes "What works for whom?" [35]. One of the terms of the problem (the population to be treated) becomes crucially important in the judgment of treatments: the value of a specific protocol is not to be considered as an absolute one, but it has to be related to the characteristic of the group of patients who undergo the procedure.

In rehabilitation, because several outcomes are usually relevant and relevant outcomes are affected by multiple factors [36-38], the definition of subgroups of patients to be treated must rely on the possibility of both quantitative and qualitative information being considered.

A preliminary analysis of the data collected in this study clarified that the amount of information provided was hardly manageable with the use of the proposed standardized global score, with the risk of missing the whole potential of the test. Only a clear separation between disabled and nondisabled subjects is supplied by this kind of analysis, as already presented in a previous paper [21]. This loss of information hinders any attempt to clearly define different levels and profiles of disability. The individual score adds too little knowledge to the classical clinical assessment. More importantly, major characteristics common to subgroups of patients are lost when all information is summarized in a single score.

To use all the information collected more comprehensively in this research, even summarizing it into meaningful specific characteristics, we decided to use a more sophisticated analysis, i.e., ClBR, which is a hybrid AI and statistics technique. In addition to the classical clustering analysis, ClBR allows combining database and additional semantic information, which can be considered when the clustering process is properly biased. In other words, through the intervention of experts who suggest "rules," their clinical experience represented in a KB can be considered in the analysis. In this specific case, medical experts' opinions have been used to generate the rules to be introduced to bias the clustering process. Different knowledge structures could, however, be used as KB. The introduction of diagnostic or treatment guidelines is also possible and of peculiar interest, since greater levels of adherence to rehabilitation guidelines have been shown to be associated with improved patient outcomes and the use of guidelines has been considered as a quality-of-care indicator [39]. An important property of the method is that semantic restrictions implied by the KB are satisfied by final clusters, and this guarantees interpretability of results and consistency with knowledge initially provided for the KB.

In this particular application, the classification obtained by classical clustering methods [21] was especially unsatisfactory for those items of the scale referring to emotional problems and depressive symptoms. As a matter of fact, a direct influence of depressive symptoms on general functioning is known: patients with either current depressive disorder or depressive symptoms tend to have worse physical, social, and role functioning and worse perceived current health than other subjects with no chronic conditions; depressed patient functioning is worse than or comparable to that of patients with major chronic medical conditions. Depression and chronic medical conditions have unique and additive effects on patient poor functioning [40].

The additional knowledge provided by the experts then focused on this topic, and the KB used in this application introduced a semantic bias referring to how to handle the items of the WHODAS II that investigate emotional behavior, and inducing the system to select at a first step all patients reporting high scores in such items. The final hierarchy, obtained after combining the KB management with the clustering, suggests a subdivision of these patients into two classes: one (Cd53) comprising patients with greater emotional and physical problems and the other (Cd52) patients with milder problems (Figure). The composition of the emerging classes corresponds to different profiles of FD. In particular, the classification of the emotionally compromised patients, influenced by the experts' rules, made possible the distinguishing between two intermediate degrees of disability that are qualitatively different mainly in terms of being or not being burdened by emotional problems, confirming the just-mentioned different performance of emotionally compromised, depressed patients.

The possibility to individuate a different self-perception of one's objective disability is important because rehabilitation implies (with the exception of passive treatments) active engagement and motivation in tasks that are supposed to improve function: patients with higher levels of depressive symptoms or history of depression use rehabilitative services less efficiently and progress slowly in regaining basic functional capacities [41]; depressive symptoms, even if mild, show a negative impact on functional recovery after stroke [42-43] with slower recovery and poorer final scores [44]. On the other hand, Loong et al. showed that the mood of depressed patients improved at the end of rehabilitative treatment [45].

Apart from the result of a higher definition in classifying depressed or emotionally impaired subjects, the analysis clarifies the most characteristic items in each class and, at the same time, defines the degree of impairment connected to single functions. As a consequence, the classes can be ordered according to the severity of the FD and a new taxonomy of four profiles of increasing disability can be proposed (Table): Low (Cr93), self-dependent subjects, with no physical or emotional problems (includes all control patients and a few patients without apparent FD); Intermediate I (Cd52), subjects with a low to moderate degree of both physical and emotional disability, with high perception of disability, worse than the objective (milder) clinically defined disability; Intermediate II (Cr89), subjects with moderate to severe disability concerning only physical problems related to self-dependency, without emotional problems; High (Cd53), subjects with the highest degree of disability, both physical and emotional. The proposed order of the four classes-in terms of worsening global function-is paralleled by a significant difference (<0.001) in the standardized global score (Table). Furthermore, the proposed profiles are not directly associated with underlying pathologies equally distributed over the four classes (not shown). We must note that the discrimination between patients with different perceptions of their impairments is no longer present in the high-level disability class. Perhaps in this group of subjects, the handicaps, diseases, and clinical conditions achieve such a state of seriousness to overwhelm a possible underlying different emotional response.

The method presented here offers the opportunity of a better classification of disabled patients-especially those with an intermediate degree of disability-and could help clinicians to evaluate a patient's potential for rehabilitation at an early stage, thus facilitating effective treatment strategies, as well as helping in deciding the most efficient use of services: both clinicians and administrators have a need to identify patients who are likely to benefit from rehabilitation. Determining in advance who will be a suitable candidate for which kind of rehabilitation should help to guarantee efforts and resources to be expended productively. [46-48]. The need to develop protocol-based treatments, the need to specify the nature of the population to be treated, and the definition of the outcomes represent basic requisites of the research on treatment efficacy and effectiveness [49] and, as a consequence, of the study of cost-effectiveness in rehabilitation medicine [50-52].

Of course, an appropriate analysis of data is a basic condition for obtaining good results. Indeed, one can obtain poor results by using either a univariate approach on the basis of the standardized global score or a multivariate approach using only classical clustering [21]. This paper shows once again that analyzing complex domains requires something more. We propose the use of a mixed technique (ClBR) that combines AI and statistical methods to improve quality of results. In this domain, integration of clinical knowledge in the analysis is fundamental for proper interpretation. None of the classical statistical methods allow expert knowledge to influence the data analysis. The use of ClBR greatly improved the results.

A potential bias in this study is represented by the nonexplicit formulation of rules regarding cognitive impairment to be introduced in the KB to influence the clustering. Cognitive impairment has been correlated with limited functional gains and poor rehabilitation outcome [53-54]. Specific studies among elderly patients with stroke led to the assumption that these patients should not be considered suitable candidates for active rehabilitation programs [55-57]. However, other studies have shown that cognitive impairment may not interfere with effective rehabilitation of elderly stroke patients [58-60]. The preliminary analysis of the groups of patients individuated by the classical clustering method did not evidence any substantial bias with regard to the cognitive performance. This finding probably depends on the fact that a criterion of inclusion was the absence of a diagnosis of cognitive impairment, based on the answers to the questionnaire's items self-reported. Furthermore, the relatively young mean age of the sample (56 years) does not introduce biases connected to the subclinical presence of cognitive impairment, a problem which becomes typical in older ages. A future sample will include both older and cognitively impaired patients, and the analysis will specifically control for this parameter.

Another aspect of the study that needs further research is the possibility of immediately classifying a single patient in one of the four classes. This possibility bases itself on the use of the scores reported in those items resulted as characteristic for each class (i.e., the fields in gray in the Table) and on the weight assigned to each score for each class. Based on this concept, we are currently developing an interpretative grid to be used in clinical practice.

A substantial improvement of the model will be realized with the introduction of guidelines (for diagnosis and treatment) as a major component of the KB. A preliminary project for this development is currently under study.


In this paper, we have proposed the definition of four classes of disability, qualitatively different among them and of increasing degree of FD (Low, Intermediate I, Intermediate II, and High) that helps to appropriately place each patient in a group of individuals with a similar profile of disability and to propose standardized treatments for these groups-an approach that is occurring in the field of rehabilitation. The advantage of a multivariate approach that properly analyzes the rich source of information and produces useful results has been shown. A new way of analyzing assessment scales (ClBR) produced interesting results. The introduction of medical knowledge in the clustering process produced benefits and helped improve the actual knowledge on the domain of disability, proposing a new ontology for disabilities that can be used to support the decision-making process in this field. These results represent a first step toward defining a new ontology to improve the clinical knowledge about the multifaceted phenomenon of FD. This step forward will contribute to the efforts of other research groups [61-64] who are trying to define new taxonomies to be used operatively in the fields of research and clinical treatment of rehabilitative medicine, since classification is basic to the advancement of scientific understanding [65].

1. Guralnik JM, Simonsick EM. Physical disability in older Americans. J Gerontol. 1993;48:3-10.
2. Stucki G, Ewert T, Cieza A. Value and application of the ICF in rehabilitation medicine. Disabil Rehabil. 2003; 25(11-12):628-34.
3. Uhlig T, Finset A, Kvien TK. Effectiveness and cost-effectiveness of comprehensive rehabilitation programs. Curr Opin Rheumatol. 2003;15(2):134-40.
4. Crosbie JH, McDonough SM, Gilmore DH, Wiggam MI. The adjunctive role of mental practice in the rehabilitation of the upper limb after hemiplegic stroke: a pilot study. Clin Rehabil. 2004;18(1):60-68.
5. Page SJ, Levine P, Sisto S, Johnston MV. A randomized efficacy and feasibility study of imagery in acute stroke. Clin Rehabil. 2001;15(3):233-40.
6. Page SJ, Levine P, Sisto SA, Johnston MV. Mental practice combined with physical practice for upper-limb motor deficit in subacute stroke. Phys Ther. 2001;81(8):1455-62.
7. Page SJ. Mental practice: a promising restorative technique in stroke rehabilitation. Top Stroke Rehabil. 2001;8(3):54-63.
8. Stevens JA, Stoykov ME. Using motor imagery in the rehabilitation of hemiparesis. Arch Phys Med Rehabil. 2003; 84(7):1090-92.
9. Baker NA, Tickle-Degnen L. The effectiveness of physical, psychological, and functional interventions in treating clients with multiple sclerosis: a meta-analysis. Am J Occup Ther. 2001;55(3):324-31.
10. Whyte J, Hart T. It's more than a black box; it's a Russian doll: defining rehabilitation treatments. Am J Phys Med Rehabil. 2003;82(8):639-52.
11. Ostir GV, Volpato S, Kasper JD, Ferrucci L, Guralnik JM. Summarizing amount of difficulty in ADLs: a refined characterization of disability. Results from the women's health and aging study. Aging (Milano). 2001;13:465-72.
12. Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW. Studies of illness in the aged. The index of ADL: a standardized measure of biological and psychosocial function. JAMA. 1963;185:914-19.
13. Lawton MP, Brody EM. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist. 1969;9:179-86.
14. Graham JM, Sagar HJ. A data-driven approach to the study of heterogeneity in idiopathic Parkinson's disease: identification on three distinct subtypes. Mov Disord. 1999;14(1): 10-20.
15. Lundberg G, Gerdle B. Correlations between joint and spinal mobility, spinal sagittal configuration, segmental mobility, segmental pain, symptoms and disabilities in female homecare personnel. Scand J Rehabil Med. 2000;32(3): 124-33.
16. Clark MS, Smith DS. Psychological correlates of outcome following rehabilitation from stroke. Clin Rehabil. 1999; 13(2):129-40.
17. Lora A, Cosentino U, Rossini MS, Lanzara D. A cluster analysis of patients with schizophrenia in community care. Psychiatr Serv. 2001;52(5):682-84.
18. Cipher DJ, Clifford PA, Schumacker RE. The heterogeneous personality: diverse coping styles among sufferers of chronic pain. Altern Ther Health Med. 2002;8(6):60-69.
19. Gibert K, Sonicki Z. Classification based on rules and medical research. J Appl Stochastic Models Data Anal. 1999; 15(3):319-24.
20. Gibert K, Cortés U. Combining a knowledge based system with a clustering method for an inductive construction of models. Lecture Notes in Statistics No. 89. New York: Springer-Verlag; 1994. p. 351-60.
21. Annicchiarico R. New methodology for disability assessment: analysis of WHO-Disability Assessment Schedule II with clustering based on rules. AI Communications 2003; 16:213-15.
22. Mecocci P, Grossi E, Buscema, Intraligi M, Savare R, Rinaldi P, Cherubini A, Senin U. Use of artificial networks in clinical trials: A pilot study to predict responsiveness to donepezil in Alzheimer's disease. J Am Geriatr Soc. 2002; 50:1857-60.
23. Cacciafesta M, Campana F, Trani I, Annicchiarico R, Leonetti-Luparini R, Gianni W, Cicconetti P, Piccirillo G. A neural network study of the correlation between metabolic-cardiovascular diseases and disability in elderly people. Arch Gerontol Geriatr. 2000;31(3):257-66.
24. Cacciafesta M, Campana F, Piccirillo G, Cicconetti P, Trani I, Leonetti-Luparini R, Marigliano V, Verico P. Neural network analysis in predicting 2-year survival in elderly people: a new statistical-mathematical approach. Arch Gerontol Geriatr. 2001;32(1):35-44.
25. Gibert K, Cortés U. Clustering based on rules and Knowledge Discovery in ill-structured domains. Computación y Sistemas, Revista Iberoamericana de Computación. 1998; 1(4):213-27.
26. Gibert K, Sonicki Z, Martín JC. Impact of data encoding and thyroids dysfunctions. Health data in the information society, technology and informatics. 2001;V90:494-503.
27. Rodas J, Gibert K, Rojo JE. KDSM Methodology for knowledge discovery from ill-structured domains presenting very short and repeated serial measures with blocking factor. Lect Notes Artif Intell. 2002;2504:228-38.
28. Gruber T. A translation approach to portable ontology specifications. Knowl Acquis. 1993;5:199-220.
29. Rehm J, Ustun TB, Saxena S. On the development and psychometric testing of the WHO screening instrument to assess disablement in the general population. Inter J Meth Psychiatr Res. 1990;8(2):110-22.
30. Epping-Jordan J, T. Bedirhan Üstün TB. The WHODAS II: leveling the playing field for all disorders. 2001 [cited 2003] WHO Bulletin of Mental Health. Available from:
31. Chwastiak LA, Von Korff M. Disability in depression and back pain: evaluation of the World Health Organization Disability Assessment Schedule (WHO DAS II) in a primary care setting. Clin Epidemiol. 2003;56(6):507-14.
32. Kessler RC, Berglund P, Demler O, Jin R, Koretz D, Merikangas KR, Rush AJ, Walters EE, Wang PS. The epidemiology of major depressive disorder: results from the National Comorbidity Survey Replication (NCS-R). JAMA. 2003;289(23):3095-105.
33. Gibert K, Cortés U. Weighing quantitative and qualitative variables in clustering methods. J Mathware Soft Comput. 1997;4(7):251-66.
34. Fuhrer MJ. Overview of clinical trials in medical rehabilitation: impetuses, challenges, and needed future directions. Am J Phys Med Rehabil. 2003;82 Suppl 10:S8-15.
35. Krause MS, Howard KI. Between group psychotherapy outcome research and basic science revisited. J Clin Psychol. 1999;55:159-69.
36. Wade DT. Outcome measures for clinical rehabilitation trials: impairment, function, quality of life, or value? Am J Phys Med Rehabil. 2003Oct;82 Suppl 10:S26-31.
37. Keith RA. Conceptual basis of outcome measures. Am J Phys Med Rehabil. 1995;74(1):73-80.
38. Kane RL. Looking for physical therapy outcomes. Phys Ther. 1994;74(5):425-29.
39. Duncan PW, Horner RD, Reker DM, Samsa GP, Hoenig H, Hamilton B, LaClair BJ, Dudley TK. Adherence to postacute rehabilitation guidelines is associated with functional recovery in stroke. Stroke. 2002;33:167-78.
40. Wells KB, Stewart A, Hays RD, Burnam A, Rogers W, Daniels M, Berry S, Greenfield S, Ware J. The functioning and well-being of depressed patients: Results from the Medical Outcomes Study. JAMA. 1989;262:914-19.
41. Gillen R, Tennen H, McKee TE, Gernert-Dott P, Affleck G. Depressive symptoms and history of depression predict rehabilitation efficiency in stroke patients. Arch Phys Med Rehabil. 2001;82(12):1645-49.
42. Herrmann N, Black SE, Lawrence J, Szekely C, Szalai JP. The Sunnybrook Stroke Study. a prospective study of depressive symptoms and functional outcome. stroke. 1998; 29:618-24.
43. Morris PLP, Raphael B, Robinson RG. Clinical depression is associated with impaired recovery from stroke. Med J Aust. 1992;157:239-42.
44. Parikh RM, Robinson RG, Lipsey JR, Starkstein SE, Fedoroff JP, Price TR. The impact of poststroke depression on recovery in activities of daily living over a two-year follow-up. Arch Neurol. 1990;47:785-89.
45. Loong CK, Kenneth NK, Paulin ST. Post-stroke depression: outcome following rehabilitation. Aust N Z J Psychiatry. 1995;29(4):609-14.
46. Kaste M, Fogelholm R, Rissanen A. Economic burden of stroke and the evaluation of new therapies. Public Health. 1998;112:103-12.
47. Kramer AM, Steiner JF, Schlenker RE, Eilertsen TB, Hrincevich CA, Tropea DA, Ahmad LA, Eckhoff DG. Outcomes and costs after hip fracture and stroke. A comparison of rehabilitation settings. JAMA 1997;277:396-404.
48. Haldorsen EM, Grasdal AL, Skouen JS, Risa AE, Kronholm K, Ursin H. Is there a right treatment for a particular patient group? Comparison of ordinary treatment, light multidisciplinary treatment, and extensive multidisciplinary treatment for long-term sick-listed employees with musculoskeletal pain. Pain. 2002;95(1-2):49-63.
49. Whyte J. Clinical trials in rehabilitation: what are the obstacles? Am J Phys Med Rehabil. 2003;82 Suppl 10: S16-21.
50. Cardenas DD, Haselkorn JK, McElligott JM, Gnatz SM. A bibliography of cost-effectiveness practices in physical medicine and rehabilitation: AAPM&R White Paper. Arch Phys Med Rehabil. 2001;82:711-19.
51. Uhlig T, Finset A, Kvien TK. Effectiveness and cost-effectiveness of comprehensive rehabilitation programs. Curr Opin Rheumatol. 2003;15(2):134-40.
52. Torres-Gil FM, Wray LA. Funding and policies affecting geriatric rehabilitation. Clin Geriatr Med. 1993;9(4):831-40.
53. Heruti RJ, Lusky A, Dankner R, Ring H, Dolgopiat M, Barell V, Levenkrohn S, Adunsky A. Rehabilitation outcome of elderly patients after a first stroke: effect of cognitive status at admission on the functional outcome. Arch Phys Med Rehabil. 2002;83:742-49.
54. Heruti RJ, Lusky A, Barell V, Ohry A, Adunsky A. Cognitive status at admission: does it affect the rehabilitation outcome of elderly patients with hip fracture? Arch Phys Med Rehabil. 1999;80(4):432-36.
55. Cifu DX, Lorish TR. Stroke rehabilitation. 5. Stroke outcome. Arch Phys Med Rehabil. 1994;75(5 Spec No.):S56-60.
56. Diamond PT, Felsenthal G, Macciocchi SN, Butler DH, Lally-Cassady D. Effect of cognitive impairment on rehabilitation outcome. Am J Phys Med Rehabil. 1996;75:40-43.
57. Hajek VE, Gagnon S, Ruderman JE. Cognitive and functional assessments of stroke patients: an analysis of their relation. Arch Phys Med Rehabil. 1997;78:1331-37.
58. Alexander MP. Stroke rehabilitation outcome. A potential use of predictive variables to establish levels of care. Stroke. 1994;25:128-34.
59. Davies AD, Gargaro P, Dodd A, Smith C. Mental status and occupational therapy outcome in teaching dressing skills to the disabled elderly. Clin Rehabil. 1989;3:293-98.
60. Galski T, Bruno RL, Zorowitz R, Walker J. Predicting length of stay, functional outcome, and aftercare in the rehabilitation of stroke patients. The dominant role of higher-order cognition. Stroke. 1993;24:1794-800. Comment in: Stroke 1994;25:1295.
61. Leveille SG, Fried LP, McMullen W, Guralnik JM. Advancing the taxonomy of disability in older adults. J Gerontol A Biol Sci Med Sci. 2004;59(1):M86-93.
62. von Steinbuchel N, Poppel E. Domains of rehabilitation: a theoretical perspective. Behav Brain Res. 1993;56(1):1-10.
63. Weiner DK, Rudy TE, Gaur S. Are all older adults with persistent pain created equal? Preliminary evidence for a multiaxial taxonomy. Pain Res Manag. 2001;6(3):133-41.
64. Halpern M. Functional assessment taxonomy relevant to low-back impairments. J Occup Rehabil. 2001;11(3):201-15.
65. Harper AC, Harper DA, Lambert LJ, Andrews HB, Lo SK, Ross FM, Straker LM. Symptoms of impairment, disability and handicap in low back pain: a taxonomy. Pain. 1992; 50(2):189-95.
Submitted for publication February 5, 2004. Accepted in revised form May 21, 2004.

Go to TOP

Go to the Contents of Vol. 41 No. 6a

Last Reviewed or Updated  Tuesday, January 25, 2005 11:58 AM