Statistical validation of the brief International Classification of Functioning, Disability and Health Core Set for osteoarthritis based on a large international sample of patients with osteoarthritis†
Presented by Ms Oberhauser in partial fulfillment of a PhD degree at the medical faculty of Ludwig-Maximilians-Universität, Munich, Germany.
With the International Classification of Functioning, Disability and Health (ICF), the impact of health conditions on functioning and disability can be described. A Comprehensive and a Brief ICF Core Set for osteoarthritis (OA) were developed. Our specific aims were to validate the Brief ICF Core Set for OA using a novel approach, i.e., to identify ICF categories that best explain patients' functioning, and to propose a statistically validated version.
We performed a psychometric study using group lasso regression on data from a convenience sample of 879 OA patients from 20 countries. The general health subscale of the Medical Outcomes Study Short Form 36 was used as a dependent variable, and all ICF categories of the Comprehensive ICF Core Set for OA and some sociodemographic and disease-specific characteristics were used as independent variables. The most relevant ICF categories were identified as those showing a significant effect based on the pointwise 90% confidence intervals resulting from 1,000 bootstrap trials.
The set of ICF categories that best explains patients' functioning consists of 4 body functions, 2 body structures, 2 activities and participation, and 3 environmental factors. Twelve of the original 13 Brief ICF Core Set categories were confirmed, and 3 additional ICF categories were identified.
Our findings reassure the validity of the Brief ICF Core Set for OA. The statistically validated Core Set with the additional statistically derived ICF categories will likely perform better with regard to discrimination and sensitivity to change in studies and trials, and should be further explored in the future.
Osteoarthritis (OA) can affect the functioning of an individual and has considerable socioeconomic burden (1, 2). At the individual level, OA could negatively impact work participation and involvement with life activities such as recreation and leisure and family activities (3, 4). At the societal level, OA-related burden can include increased health care cost, work productivity loss, and the need for formal or informal care to substitute for domestic responsibilities (5–8). With the increasing prevalence of obesity and the aging of the population, the burden of OA is expected to increase.
The World Health Organization (WHO) approved the International Classification of Functioning, Disability and Health (ICF) in 2001 as a framework and classification system to understand and describe the impact of health conditions on functioning and disability (9). The ICF is a conceptual framework and classification system on functioning, where different components of body functions, body structures, and activities and participation interact in relation to a health condition. This interaction can be influenced by contextual factors, specifically, personal and environmental factors. ICF categories, which are functioning domains, make up each component and are represented by alphanumerical codes (arranged in a hierarchy).
A condition-specific ICF Core Set represents a selection of ICF categories that are specific and relevant to understand the impact of a health condition on functioning (10–13). An ICF Core Set makes the ICF practical and manageable to users. The ICF Core Set for OA has 2 versions: the Comprehensive Core Set and the Brief Core Set (14). The development of both Core Sets is based on evidence from patients, clinicians, experts, and systematic review studies, and finally integrated by experts when they achieved consensus via a nominal group consensus process (15). In the Comprehensive ICF Core Set, which was mainly developed for use in multidisciplinary settings, there are 55 ICF categories. In the Brief ICF Core Set, there are 13 ICF categories, which represent the minimal list of ICF categories that should be addressed when the impact of OA on functioning is to be measured in studies or single clinical encounters. It is important to note that the brief version is developed out of the comprehensive version.
The potential benefits of using the ICF Core Set for OA include having a standard to ensure comparability and standardization of measures in clinical trials, longitudinal observational studies, clinical practice, reimbursement, and health information technology (reports) (16, 17). However, before it can be accepted as valid for trials and therefore a true “standard,” its content validity has to be confirmed first. Two investigations have been performed to study the content validity of the Comprehensive ICF Core Set for OA, and both confirmed the relevance and use of the ICF categories in clinical practice in multidisciplinary settings (18, 19). Also for the Brief ICF Core Set, 1 study on construct validation has been conducted (20). The latter study on the Brief ICF Core Set was performed in persons with OA from Singapore using linear regression models. However, since the ICF Core Set is to be used as an international standard, additional studies including persons from different countries are needed. Moreover, recently, regression methods that allow simultaneously analyzing a large number of independent ordinal variables have been developed (21, 22). This development presents the opportunity of validating the content of the Brief ICF Core Set for OA with a statistical method tailored to the data collected with ICF categories and using the ICF qualifier scale. More concretely, the ICF categories that best differentiate among levels of functioning in persons with OA can be identified. Statistically validating the ICF categories of the Brief ICF Core Set using a large multinational data set on OA would be important for the robust and universal application of the ICF Core Sets in clinical practice and empirical studies.
The specific aims of this investigation were to identify ICF categories that best explain patients' functioning, to determine the content validity of the Brief ICF Core Set, and to propose a statistically validated version of the Brief ICF Core Set.
Significance & Innovations
This study applied a recently developed regression method to a large multinational study to identify the International Classification of Functioning, Disability and Health (ICF) categories that best explain the functioning and disability of patients with osteoarthritis (OA).
This study proposes a statistically validated version of the Brief ICF Core Set for OA.
The statistically validated Core Set with the additional statistically derived ICF categories would likely perform better with regard to discrimination and sensitivity to change in studies and trials.
PATIENTS AND METHODS
Study design and participants.
We performed a psychometric study using data from a convenience sample of 879 OA patients from 34 study centers in 20 countries. Cross-sectional data collection was performed in collaboration with the WHO using the Comprehensive ICF Core Set for OA (14).
The study protocol and informed consent were approved by the institutional ethics committee for each center and are in compliance with the Declaration of Helsinki. Inclusion criteria consisted of a diagnosis of OA according to the American College of Rheumatology criteria (23–25), age ≥18 years, sufficient knowledge of the official language of the country, comprehension of the purpose of the study, and signed informed consent. Patients were excluded after surgery if the wound healing process was not completely resolved.
We collected sociodemographic data on sex, age, and status of being pensioned due to OA; disease characteristics, including disease duration, type (primary versus secondary), and location (localized versus generalized; joints involved) of the OA; and a comorbidity score using the Self-Administered Comorbidity Questionnaire (26), which takes into consideration the number of comorbidities, the need for treatment, and the level of limitation due to the comorbidities.
The level of impairment and limitation (or restriction) for each ICF category included in the Comprehensive ICF Core Set for OA was rated with the ICF qualifier (where 0 = no problem, 1 = mild problem, 2 = moderate problem, 3 = severe problem, and 4 = complete problem). For the environmental factors, the extent to which the category is a barrier or facilitator was quantified on a scale ranging from −4 to +4 (with a similar description to above, i.e., mild, moderate, etc.), with negative values indicating the extent to which the category is a barrier and positive values indicating the extent to which the category is a facilitator. The qualifier 8 was applied when the available information was not sufficient to quantify the severity of the problem, and the qualifier 9 was used when the category was not applicable to the patient.
For the purpose of this validation study, the score on the general health (GH) subscale of the Medical Outcomes Study Short Form 36 (SF-36) (27) was used as a dependent variable in the regression analyses. The SF-36 is a generic measure of functioning that has been applied commonly in arthritis and has sound psychometric properties (28–33). We used the GH score because we adopted the WHO position that functioning is the operationalization of health, and therefore GH seems to be an appropriate dependent variable for identifying the ICF categories that best explain patients' functioning (34, 35). We used the GH score and not the single GH question because the former is a metric approximation of the single GH question and, therefore, more appropriate as a dependent variable. This score can take values ranging from 0–100, where higher values indicate better GH.
Patient recruitment and data collection were performed by health professionals in each study center. They were trained to perform interviews and to collect data from the patient records by attending a workshop or using a training video provided by the ICF Research Branch of WHO Collaborating Centre for the Family of International Classifications in Germany. The case record form that was completed by the health professionals was in English. The form completed by the patients was in the corresponding language of the country. Language-specific validated versions of the SF-36 were used.
Data preparation and statistical analyses.
The 2 response options in the ICF qualifier not fitting the ordinal scale level, i.e., 8 = not specified and 9 = not applicable, were considered missing values. Missing values in ICF categories and some sociodemographic and disease-specific variables were replaced by multiple imputations with the assumption that values are missing at random. To do this, we used the R add-on package Amelia, where a latent multivariate normal distribution is assumed for the ordinal predictors, and a combination of the expectation maximization algorithm with a bootstrap approach was used to take draws from this posterior (36). We used descriptive statistics to present the characteristics of the study population.
Identification of ICF categories that best explain patients' functioning.
To identify the set of ICF categories that best explain patients' functioning, group lasso regression modified by Gertheiss and colleagues was applied using GH as the dependent variable and all ICF categories contained in the Comprehensive ICF Core Set for OA and some sociodemographic and disease-specific characteristics as independent variables (21, 22). Group lasso is a regression technique that, in addition to the estimation of regression coefficients, allows for the selection of dummy-coded categorical independent variables that best explain the variance of a dependent variable (37). Because of this characteristic of the group lasso regression, all of the response options of the ICF categories, even the negative values of the environmental factors, are treated as dummy-coded variables (1 = the response option applies or 0 = the response option does not apply). Therefore, there is no need for additional transformation. The sociodemographics, disease-specific variables, and comorbidities were forced to remain in the model, while variable selection was performed among the ICF categories.
To account for the uncertainty in the relationship between the ICF categories and the GH score due to the convenience sample of patients and missing values in the independent variables, especially in the ICF categories, the data set was imputed 1,000 times, and then a bootstrap sample (i.e., a random sample with replacement and of equal size to the original data set) (38) was drawn from each imputed data set and group lasso was applied to each of the 1,000 resulting data sets. Based on this procedure, the mean of the regression coefficients over the 1,000 trials and their pointwise 90% (percentile) confidence intervals were obtained. The mean regression coefficients provide an estimation of the size of the effect. The larger the absolute value of the regression coefficient, the stronger the relationship. The confidence interval indicates whether the effect is significantly different from 0 at the 90% (percentile) confidence level. Those ICF categories that showed a significant effect were the ones that best explain patients' functioning and were considered the “group lasso set.”
Brief ICF Core Set content validity.
To determine the content validity of the Brief ICF Core Set, the following 2 steps were followed: the Brief ICF Core Set categories contained in the group lasso set were considered directly confirmed content wise, and for the remaining Brief ICF Core Set categories, Spearman's correlation coefficient was calculated with the ICF categories contained in the group lasso set. If the correlation of a Brief ICF Core Set category with at least 1 ICF category from the group lasso set was high (>0.4), then that Brief ICF Core Set category was considered indirectly confirmed. This is a reasonable procedure, since among 2 highly correlated independent variables, group lasso selects only the one that is slightly better at contributing to the explanation of the dependent variable, while the other one is completely neglected. However, this does not mean that the independent variable not selected does not explain variance of the dependent variable or cannot be relevant for some individual patients.
Proposal for a statistically validated version of the Brief ICF Core Set.
The proposed statistically validated version of the Brief ICF Core Set for OA is made up of ICF categories from the Brief ICF Core Set that were directly or indirectly confirmed, as well as of those ICF categories from the group lasso set that were not contained in the Brief ICF Core Set and not correlated (>0.4) with any of the Brief ICF Core Set categories. For all of our computations, we used the statistical program R (39).
The descriptive statistics of our sample, including their GH score, are shown in Table 1. Most of the patients were women and the mean age was 61.9 years. The majority had primary (exact etiology unknown) and localized OA. The knees were the most affected by OA, followed by OA of the hips and hands. For the GH score, the mean ± SD observed value was 49 ± 21.8.
Table 1. Characteristics of 879 patients with OA*
|Sociodemographic data|| |
| Women, no. (%)||631 (72)|
| Age, mean ± SD years||61.9 ± 11.0|
| Pensioned due to OA, no. (%)||183 (21)|
|Disease characteristics, mean ± SD|| |
| SCQ score||6.3 ± 4.0|
| Disease duration, years||6.9 ± 8.6|
|Type of OA, no. (%)|| |
| Primary||791 (90)|
| Secondary||74 (8)|
|Location of OA, no. (%)|| |
| Localized||653 (74)|
| Generalized||198 (23)|
|Anatomic sites, no. (%)|| |
| Knee||697 (79)|
| Hip||268 (31)|
| Hands||123 (14)|
| Spine||42 (5)|
| Shoulder||23 (3)|
|General health score (range 0–100), mean ± SD||49.0 ± 21.8|
ICF categories that best explain patients' functioning.
The mean of the regression coefficients over the 1,000 trials and their pointwise 90% (percentile) confidence intervals for sociodemographics, disease characteristics, and comorbidities are shown in Table 2. There was a strong effect of the country in which the study center was located that also reflected large differences between patient groups included in the different countries. Women and patients pensioned due to OA showed slightly worse health than men and nonpensioned patients. An increased comorbidity score had a negative effect on GH. The type and location of OA, age, and duration of OA did not show a significant effect on the 90% level.
Table 2. Regression coefficients for non-ICF categories*
|Country|| || |
| Argentina||0||0, 0|
| Brazil||27.494†||7.314, 46.115|
| Chile||13.934||−1.346, 29.941|
| China||4.579||−9.440, 18.072|
| Cyprus||−13.482||−29.471, 1.901|
| Czech Republic||17.629†||1.142, 33.029|
| Germany||15.272†||1.797, 29.328|
| Hungary||8.653||−4.882, 21.982|
| Indonesia||13.122||−2.730, 28.086|
| Italy||17.197†||3.218, 32.315|
| Kuwait||0.091||−13.887, 14.397|
| Lithuania||5.035||−9.670, 19.086|
| Philippines||32.268†||19.580, 46.499|
| Poland||15.846†||1.723, 30.188|
| Serbia||14.937†||0.947, 28.539|
| Singapore||34.216†||19.824, 49.002|
| South Korea||20.504†||4.731, 34.978|
| Spain||18.301†||4.157, 32.441|
| Switzerland||22.705†||7.880, 37.867|
| Turkey||9.073||−6.890, 24.047|
|Sex|| || |
| Male||0||0, 0|
| Female||−2.423†||−4.561, −0.014|
|Pensioned due to disease|| || |
| No||0||0, 0|
| Yes||−4.325†||−7.997, −0.535|
|OA type|| || |
| Primary||0||0, 0|
| Secondary||1.766||−2.022, 5.855|
|OA location|| || |
| Localized||0||0, 0|
| Generalized||−1.905||−5.302, 1.326|
|SCQ score||−1.443†||−1.792, −1.104|
Table 3 shows the mean of the regression coefficients over the 1,000 trials and their pointwise 90% (percentile) confidence intervals of those ICF categories that showed a significant effect on the 90% (percentile) confidence level and therefore form the set of ICF categories that best explain patients' functioning (this set was referred to as the group lasso set). This set consists of 11 ICF categories (4 body functions, 2 body structures, 2 activities and participation, and 3 environmental factors).
Table 3. Regression coefficients for those ICF categories for which at least 1 regression coefficient for the dummy-coded response options, i.e. the ICF qualifiers, is significantly different from 0 with regard to the pointwise 90% (percentile) CI*
|Energy and drive functions (b130)|| || |
| No problem||0||0, 0|
| Mild problem||−0.383||−2.315, 1.513|
| Moderate problem||−3.130†||−6.445, −0.095|
| Severe problem||−4.150†||−8.352, −0.173|
| Complete problem||−2.972||−7.049, 0.535|
|Emotional functions (b152)|| || |
| No problem||0||0, 0|
| Mild problem||−2.114†||−4.437, −0.160|
| Moderate problem||−4.037†||−7.383, −0.803|
| Severe problem||−3.341||−7.038, 0.000|
| Complete problem||−3.596†||−7.674, −0.002|
|Sensation of pain (b280)|| || |
| No problem||0||0, 0|
| Mild problem||−0.391||−2.693, 1.835|
| Moderate problem||−2.906||−6.641, 0.060|
| Severe problem||−4.321†||−8.796, −0.537|
| Complete problem||−3.712||−8.395, 0.045|
|Muscle power functions (b730)|| || |
| No problem||0||0, 0|
| Mild problem||−2.712†||−5.712, −0.224|
| Moderate problem||−3.089||−6.939, 0.000|
| Severe problem||−2.467||−7.040, 0.725|
| Complete problem||−2.066||−6.525, 1.133|
|Structure of shoulder region (s720)|| || |
| No problem||0||0, 0|
| Mild problem||−2.197||−4.766, 0.000|
| Moderate problem||−4.411†||−8.253, −1.023|
| Severe problem||−7.596†||−14.009, −1.970|
| Complete problem||−7.132†||−13.225, −1.728|
|Structure of lower extremity (s750)|| || |
| No problem||0||0, 0|
| Mild problem||−2.266†||−5.078, −0.046|
| Moderate problem||−4.219†||−7.617, −1.104|
| Severe problem||−2.233||−6.183, 1.149|
| Complete problem||0.546||−4.013, 6.237|
|Moving around (d455)|| || |
| No problem||0||0, 0|
| Mild problem||−2.747†||−6.305, −0.068|
| Moderate problem||−2.970||−7.220, 0.001|
| Severe problem||−4.329†||−8.955, −0.271|
| Complete problem||−2.461||−7.097, 1.467|
|Recreation and leisure (d920)|| || |
| No problem||0||0, 0|
| Mild problem||−0.220||−2.288, 1.843|
| Moderate problem||−3.950†||−7.130, −1.052|
| Severe problem||−3.168||−6.736, 0.142|
| Complete problem||−3.411||−8.252, 0.820|
|Products and technology for personal use in daily living (e115)|| || |
| Complete barrier||−0.910||−4.563, 2.358|
| Severe barrier||−1.062||−4.686, 1.943|
| Moderate barrier||−1.315||−4.834, 1.294|
| Mild barrier||−0.839||−3.180, 1.007|
| Neither barrier nor facilitator||0||0, 0|
| Mild facilitator||2.248†||0.052, 5.047|
| Moderate facilitator||1.685||−0.759, 4.711|
| Severe facilitator||2.025||−0.938, 5.887|
| Complete facilitator||2.744||−1.077, 7.680|
|Societal attitudes (e460)|| || |
| Complete barrier||−0.031||−4.454, 5.213|
| Severe barrier||−0.809||−4.765, 3.140|
| Moderate barrier||−1.639||−4.832, 0.754|
| Mild barrier||0.130||−1.895, 2.498|
| Neither barrier nor facilitator||0||0, 0|
| Mild facilitator||1.415||−0.340, 3.883|
| Moderate facilitator||2.240||−0.040, 5.102|
| Severe facilitator||3.351†||0.063, 7.139|
| Complete facilitator||3.510||−0.342, 8.314|
|Health services, systems, and policies (e580)|| || |
| Complete barrier||2.285||−1.428, 7.780|
| Severe barrier||2.565||−0.915, 7.997|
| Moderate barrier||1.752||−1.012, 5.844|
| Mild barrier||1.006||−1.015, 3.850|
| Neither barrier nor facilitator||0||0, 0|
| Mild facilitator||2.717†||0.321, 5.660|
| Moderate facilitator||4.629†||1.161, 8.417|
| Severe facilitator||3.962†||0.567, 7.761|
| Complete facilitator||4.065†||0.151, 8.337|
The regression coefficients in Table 3 show that higher problems in functioning (i.e., in body functions, body structures, and activities and participation) are associated with worse GH. For the environmental factors, the relationship to GH is more complex. On one hand, when products and technology for personal use in daily living (e115) and societal attitudes (e460) were rated as facilitators, the GH was better. If they were rated as a barrier, they were associated with worse GH. On the other hand, health services, systems, and policies (e580) was associated with better GH independently of whether it was rated as a facilitator or barrier.
Content validity of the Brief ICF Core Set.
Table 4 shows that from the 13 ICF categories contained in the Brief ICF Core Set for OA, 5 were contained in the group lasso set and, therefore, directly confirmed. Seven of the remaining 8 were correlated (>0.4) with ICF categories contained in the group lasso set and, therefore, were considered indirectly confirmed.
Table 4. Spearman's correlation coefficients between the ICF categories from the group lasso set and those from the Brief ICF Core Set*
|Sensation of pain (b280)||0.355||0.394||1||0.233||0.341||0.305||0.314||0.250||0.184||0.046||−0.129||Directly|
|Mobility of joint functions (b710)||0.341||0.396||0.304||0.493||0.191||0.547†||0.482||0.461||0.253||0.027||0.102||Indirectly|
|Muscle power functions (b730)||0.361||0.37||0.233||1||0.163||0.383||0.361||0.451||0.157||0.069||0.148||Directly|
|Structure of upper extremity (s730)||0.265||0.204||0.357||0.190||0.589†||0.220||0.115||0.192||0.010||−0.010||−0.072||Indirectly|
|Structure of lower extremity (s750)||0.316||0.289||0.305||0.383||0.216||1||0.457||0.423||0.204||−0.013||0.067||Directly|
|Additional musculoskeletal structures related to movement (s770)||0.327||0.223||0.414†||0.284||0.361||0.349||0.322||0.333||−0.018||−0.038||−0.116||Indirectly|
|Hand and arm use (d445)||0.288||0.241||0.313||0.221||0.456†||0.133||0.052||0.219||0.049||0.051||−0.034||Indirectly|
|Products and technology for personal use in daily living (e115)||0.104||0.158||0.184||0.157||0.018||0.204||0.311||0.149||1||0.131||0.334||Directly|
|Design, construction and building products, and technology of buildings for public use (e150)||−0.034||0.012||−0.182||0.098||−0.104||−0.021||0.158||0.002||0.295||−0.033||0.275||Not|
|Immediate family (e310)||0.236||0.257||0.149||0.156||0.037||0.177||0.244||0.283||0.415†||0.267||0.306||Indirectly|
|Health services, systems, and policies (e580)||0.044||0.081||−0.129||0.148||−0.066||0.067||0.306||0.120||0.334||0.334||1||Directly|
|Additionally identified‡||Yes||Yes|| || || || || || || ||Yes|| || |
Only design, construction and building products, and technology of buildings for public use (e150) from the Brief ICF Core Set could not be confirmed directly or indirectly.
Statistically validated version of the Brief ICF Core Set.
Table 4 shows that energy and drive functions (b130), emotional functions (b152), and societal attitudes (e460) were in the group lasso set but were not included in the original Brief ICF Core Set and were not correlated (>0.4) with any of the Brief ICF Core Set categories. Therefore, based on our results, the proposed statistically validated version of the Brief ICF Core Set for OA is made up of the 5 ICF categories from the Brief ICF Core Set that were directly confirmed and the 7 categories that were indirectly confirmed, as well as b130, b152, and e460.
In this investigation, a recently developed regression method was applied to a large multinational study to identify the ICF categories that best explain the functioning and disability of patients with OA. Based on this novel approach and further analyses, the validity of the Brief ICF Core Set for OA has been confirmed and 3 additional ICF categories of the Comprehensive ICF Core Set are being suggested to be added. Thereby, a statistically validated version of the Brief ICF Core Set is proposed. Twelve of the 13 ICF categories contained in the original Brief ICF Core Set were confirmed, with the addition of 3 ICF categories. Overall, there are 15 statistically derived ICF categories that could be used as essential domains to capture the impact of OA on functioning in clinical and epidemiologic studies.
We used statistical methods that rely on the variability of the variables included in the models. Therefore, we can assume that the ICF categories selected are not only relevant from a clinical but also from a statistical point of view, in the sense that they are able to differentiate among varying levels of functioning. Therefore, the 15 ICF categories proposed in this investigation are reasonable candidates toward a statistically validated version of the Brief ICF Core Set for OA to address the impact of OA on functioning in clinical and epidemiologic studies or trials. Sensitivity to differentiate among levels of functioning is essential.
Three ICF categories are proposed that were not originally included in the Brief ICF Core Set for OA, specifically, energy and drive functions (b130), emotional functions (b152), and societal attitudes (e460). All of them are fundamental aspects of the life of a person with OA. Even though they are not addressed in the most frequently used health status measures, such as the Western Ontario and McMaster Universities Osteoarthritis Index and the Lequesne Algofunctional Index, the role of energy and drive and emotional functions in persons with OA has been frequently emphasized (40–42). It is interesting that social attitudes were identified in this study as an environmental factor that has a relevant influence on functioning. Our sample consists of patients from socially and culturally different countries and with intuitively different attitudes toward persons with health problems. The diversity of settings might have contributed to the variability of our data and to the selection of this variable. This variability is warranted, given that the ICF was intended for use internationally. To our knowledge, this is the first study that proved the influence of the social attitudes on the life of persons with OA. It will be important to confirm this result with future studies.
The fact that the ICF category design, construction and building products, and technology of buildings for public use (e150) could not be confirmed directly or indirectly is a finding that deserves further investigation. The expert participants of the consensus conference who decided on the ICF categories to be included in the Brief ICF Core Set, however, presented strong evidence for including this ICF category. The extent to which public buildings are accessible to persons with mobility limitations, which might be the case for OA patients, could determine consequent participation in society (14).
This study confirms that the modification of group lasso by Gertheiss et al is useful when analyzing ICF data collected with an ordinal scale, such as the qualifier scale (21, 22). The analyses of ICF data often have been presented as a challenge in the literature (35). We hope that other researchers working in the field would be encouraged by the analyses presented in this study. The modification by Gertheiss and colleagues permits us to exploit all of the information contained in the ICF qualifiers without artificially having to assume a metric scale or to collapse the response options to 0 and 1, as has been frequently done in previous investigations (19, 20, 39). With this method, it is possible to estimate a large number of coefficients with a moderate sample size. This is a robust property, since the ICF qualifier has 5 or 9 response options for the functioning components (body functions, body structures, and activities and participation) and environmental factors, respectively.
Within this context, it is important to mention that the Comprehensive ICF Core Set for OA, but not the Brief ICF Core Set, has already been validated from the perspective of patients (43). In contrast to the Comprehensive ICF Core Set for OA, which was developed for comprehensive multidisciplinary assessments, the Brief ICF Core Set was developed for clinical research. Therefore, we consider it more appropriate to study the content validity of the Brief ICF Core Set statistically, since in clinical studies it is important to assess those health domains that differentiate among different levels of health and functioning and are potentially sensitive to change.
This study also has some limitations. The cutoff criterion for the Spearman's correlation coefficient of 0.4 is arbitrary. However, it is difficult to decide what the appropriate criterion is without being too restrictive or too permissive. We decided to use the GH score from the SF-36 with the assumption that it provides an overall outcome of functioning and health as experienced by the patients themselves. The SF-36 has proven not only to have good psychometric properties when applied in patients with OA, but also it is one of the most widely used generic health measures and is translated in different languages, which further lends universality to our sample and the intention of the WHO with the ICF Core Sets. However, other outcomes specific to OA may be used in parallel. Therefore, it might be important to perform studies in the future applying the same methodology and using other disease-specific outcomes as well to see whether or not functioning constructs of other instruments give a similar or consistent finding. Even though our sample was large and included persons from 20 countries, there are still other countries or regions that are not represented.
We included sociodemographic information and disease- specific characteristics in the regression model so as to control for external influences besides the influence of the ICF categories. Factors such as race, age, sex, body mass, presence of OA in multiple joints, knee alignment, and radiographic features are important considerations in OA (44–46). In our study, the country of the study center seemed to have a large influence on patients' functioning. This is probably related to the fact that the patients included in the study centers have different clinical characteristics, and also probably related to the fact that they are treated differently according to the health care system of the country or that cultural differences exist in the evaluation of how a disease impacts health. Finally, we cannot totally exclude the possibility that the different levels of English knowledge of the health professionals collecting the data may have influenced the results and the effect of the country observed in this study. The translation process of the ICF cannot be performed independently from WHO and is usually carried out by the collaborating centers in the different countries and regional offices. At the time that this study was performed, the ICF had not been translated into all of the languages of the countries participating in the study. Therefore, to homogenize the data collection procedures, we decided to keep the case record forms containing the ICF categories that were completed by the health professionals in English. For comparison purposes, we also performed a separate regression analysis that only included the ICF categories as independent variables and the results of which were consistent with the findings presented here. We recognize the differences among countries; some country samples were too small to explore country-specific differences and this might be interesting for future studies. However, we are confident the present data represent an international view of functioning and health in OA.
In conclusion, the statistical approach to select the core categories relevant to functioning and health largely reflected the original core categories proposed by expert opinion. This reassures the validity of the Brief ICF Core Set. While the Comprehensive ICF Core Set remains as the external standard for rehabilitation purposes and practice at the individual level, the statistically validated Core Set with the additional statistically derived ICF categories will likely perform better with regard to discrimination and sensitivity to change in studies and trials, and should be further explored in the future.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Escorpizo had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Oberhauser, Escorpizo, Boonen, Stucki, Cieza.
Acquisition of data. Boonen, Cieza.
Analysis and interpretation of data. Oberhauser, Escorpizo, Boonen, Stucki, Cieza.
The authors would like to give special thanks to the following participating study centers: Ankara University, Ankara, Turkey; Cattedra Reumatologia-Università Politecnica delle Marche, Jesi, Italy; Charles University, Prague, Czech Republic; Cipto Mangunkusumo General Hospital, Jakarta, Indonesia; Dr. Gero Istvan Egeszsegügyi Hospital, Senta, Serbia; First Affiliated Hospital of Nanjing Medical University, Jiangsu, China; Hanyang University, Seoul, South Korea; Hospital Clinico, Universidad de Chile, Santiago, Chile; Hospital das Clinicas, University of Sao Paulo, Sao Paulo, Brazil; Hospital San Juan Bautista, Catamarca, Argentina; Huazhong University of Science and Technology, Wuhan Hubei, China; Institute of Rheumatology, Warsaw, Poland; Istituto Scientifico di Montescano, Montescano, Pavia, Italy; Kaunas University of Medicine Hospital, Kaunas, Lithuania; Klinik am Regenbogen, Nittenau, Germany; Klinikum Bremen-Ost, Bremen, Germany; Kuwait Physical Medicine and Rehabilitation Society, Safat, Kuwait; Limassol Center of Physical and Rehabilitation Medicine, Limassol, Cyprus; Medical Park St. Hubertus, Bad Wiessee, Germany; Moorbad Bad Doberan, Bad Doberan, Germany; National Institute for Medical Rehabilitation, Budapest, Hungary; National Institute of Rheumatology, Budapest, Hungary; National University of Singapore, Singapore; Physioklinik im Aitrachtal, Mengkofen, Germany; Praxis für Physikalische Medizin und Rehabilitation, Goettingen, Germany; Rehabilitationsklinik Franken der BfA, Bad Steben, Germany; Rehazentrum Bad Eilsen, Bad Eilsen, Germany; Spanish Society of Rheumatology, Madrid, Spain; University Hospital Zurich, Zurich, Switzerland; and University of Santo Tomas, Manila, Philippines.