The study was conducted between April 2002 and December 2003 in the province of Vizcaya (Basque Country) in the north of Spain. Vizcaya has a population of 1,125,000 inhabitants, 23.6% of whom were aged 60 years or older. The province of Vizcaya is predominantly urban.
Recruitment
To recruit participants from the general population, we used the register of the Basque Department of Health. This register includes all people covered by the National Health System which, in the Basque Country, covers almost 100% of the population. From that register, we performed a stratified random sampling by gender and age (3 categories: 60 to 69, 70 to 79, and 80 to 90 years) of all adults over sixty in the province.
Based upon a prevalence of 10% for knee OA and of 5% for hip OA observed in previous studies for patients aged 60 years or older [8–13], and using α = 0.05, 1-β = 0.8, and an error in the prevalence of OA of less than ± 1.5%, we estimated that we would need to recruit more than 7,000 individuals. Estimating a 10% exclusion rate and a participation rate of at least 70%, we planned to include at least 11,000 individuals in our study.
Selection criteria included: age ≥60 years, resident of Vizcaya, ability to complete the questionnaire and give consent to participate in the study, and ability to attend an outpatient clinic. We excluded individuals who were younger than 60 years, who lived outside the province, who did not have valid postal addresses or telephone numbers, who had severe psychiatric, sensorial or physical illness, or language problems that made it difficult to complete the questionnaire, or who were unable to attend the outpatient clinic.
KHOA-SQ development
The KHOA-SQ was created to be a short, quickly completed questionnaire that included all relevant variables that alone or in combination would indicate, with a high sensitivity and specificity, possible OA of the knee or hip. To create the KHOA-SQ, we reviewed the available literature to determine if similar tools existed [3, 9–12] and selected from prior studies the variables most likely to identify patients with knee or hip OA.
The resulting questionnaire had 28 questions in three sections: 11 pertaining to knee osteoarthritis, 11 to hip osteoarthritis, and 6 common or general questions about both joints related to previous fractures, future interventions, and sociodemographic data. We conducted a pilot study with 400 patients undergoing urography or barium enema examinations to test which combination of questions offered the best balance of sensitivity and specificity for hip OA. The algorithms selected are presented in Figure 1. For the hip screening algorithm (algorithm #1, hip), we included 4 of the 11 original questions : "During the last 12 months, have you often had pain in one or both your hips for one month or more?"; "Has a doctor ever told you that you have osteoarthritis in one or both of your hips?"; "Do you have a prosthesis in one or both of your hips?"; and, "During the last 12 months, have you had frequent limitations or difficulties walking more than 4 blocks (500 m) because of pain or stiffness in one or both of your hips?"
For the knee screening algorithm (algorithm #1, knee), we also included 4 of the original 11 questions: "During the last 12 months, have you often had pain in one or both your knees for one month or more?"; "Has a doctor ever told you that you have osteoarthritis in one or both of your knees?"; "During the last 12 months, have you had any limitations rising from a chair or toilet because of pain or stiffness in one or both of your knees?"; and "During the last 12 months, have you had stiffness in one or both your knees for one month or more?" We based our election on previous studies [13].
In both cases, we created algorithms based on the previous questions to achieve the best possible sensitivity, while preserving reasonable specificity and positive predictive value (PPV). PPV positive was defined as OA real cases among those screening positive. Negative predictive value (NPV) is defined as no OA real cases among those screening negative. Sensitivity was defined as percentage of OA cases who screened positive, specificity as percentage of non-OA subjects who screened negative and accuracy as the percentage of screening results, both positive and negative, that were correct.
Data collection
Letters were sent to 11,002 randomly selected residents of Vizcaya province to invite them to participate in the study. The letter presented the goals of the study, invited the recipient to participate, and asked for their informed consent. It also included the screening questionnaire for knee or hip OA along with a stamped return envelope. A reminder letter was sent to those who had not replied after 15 days. After 30 days, a copy of the questionnaire was sent to those who had not responded. Finally, we contacted by telephone those who had not responded within 45 days of the initial mailing.
We applied the algorithm to those who answered the KHOA-SQ. Individuals identified by the questionnaire as likely to have knee or hip OA were contacted by mail (up to three times, if necessary) or telephone to invite them to be evaluated for osteoarthritis at one of three hospitals chosen to provide ready access to the bulk of the population, and thus facilitate participation. Three orthopedic surgeons who collaborated in the study evaluated participants at any of the hospitals' outpatient clinics. These surgeons were trained by the research team and provided with standardized questionnaires requesting sociodemographic data, comorbidities, and symptoms (pain, stiffness, and functional limitations) and signs related to hip or knee osteoarthritis, separately. Information from a full clinical examination of the hip, knee, and lower back was also recorded. If the clinical examination was suggestive of hip or knee disease and the patient had not undergone an x-ray within the preceding 6 months, study participants were invited to have an x-ray of the affected joint(s). All x-rays were evaluated by the Kellgren and Lawrence [14] scale for hip OA and the Ahlbäck [15] scale for knee OA. Each orthopedic surgeon provided a final diagnosis about the presence or absence of OA. We classified individuals as having knee or hip OA if they had symptoms and radiographic evidence of OA in either of the hip or knee articulations. Patients also completed the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire[16, 17].
We also included in the evaluation by the orthopaedic surgeons of some individuals who were negative for knee or hip OA on the KHOA-SQ. We randomly selected 300 individuals whose screening questionnaires were negative for both joints. Of these, 158 accepted and were reviewed by an orthopedic surgeon. Since the orthopedic surgeon evaluated all 4 joints for each participant, we included in the estimations those who screened negative for one part of the questionnaire (hip or knee joint) but were positive for the other. In all cases the orthopedic surgeons were blinded to the study goals and the results of the screening algorithm.
The study was approved by the hospitals Research Committees. All data were kept confidential.
Statistical analysis
The unit of analysis was the person that was classified as positive OA for hip whenever any of his hips were positive and negative otherwise; or positive OA for knee whenever any of the knees were positive and negative otherwise. Frequencies and percentages were calculated as descriptive statistics of the sample. Hip and knee OA were treated separately.
In order to evaluate the possible presence of selection bias, we compared the answers on symptoms (see tables) to our screening questionnaire, by joint, between people who finally went, and those who did not, to the evaluation by the orthopaedic surgeon.
Evaluation of the KHOA-SQ
Selected OA screening algorithms for hip and knee were evaluated using sensitivity, specificity, and odds ratio by age and gender. For all of these estimates, 95% confidence intervals were calculated. Logistic regression models in which the dependent variable was the final diagnosis of kip or knee OA were used to estimate the c statistics of each question or algorithm, which were the independent variable in each model. In these analyses, the c statistic is a mathematical function of the sensitivity and specificity of our tool in classifying patients by means of a logistic regression model as either having OA or not. The c statistic is calculated as the fraction of patients with the outcome among pairs of patients where one has the outcome and one not, the patient with the highest prediction being classified as the one with the outcome (or c statistic estimate = .5 (1+Somer's D test). The null value for the C statistic is 0.5, with a maximum value of 1.0 (higher values indicating better).
Additionally, and as a way to provide with more information on the validity of our tool, we evaluated the KHOA-SQ results in relation to the WOMAC scores, by reported problems of hip and knee separately, only in those people who went to the revision by the orthopaedic surgeon and who completed the WOMAC. With those previous conditions, we were able to analyze 632 people who presented hip problems and 953 with knee problems. Means and standard deviations are provided. A Student t test was performed among those which gave positive for osteoarthritis, according to our KHOA-SQ, and those who did not.
Alternative screening algorithms to the KHOA-SQ were selected based on their sensitivity, specificity, positive predictive value, accuracy and odds ratio (OR: being OR the odds of having an OA). First, the association between individual questionnaire items and hip or knee OA was determined. Second, two alternative screening algorithms for hip and knee OA were created based on the ability of individual questions to predict OA. The simplest algorithm was based on two items, pain and previous OA status reported by any physician (algorithm #2). Another was based on three items related to OA symptoms – pain, stiffness, and the ability to walk more than 4 blocks (algorithm #3). Both were evaluated for hip and knee OA in the same sample.
Finally, we used Classification and Regression Tree (CART) analysis [18] to find the most informative way to classify subjects by their response to questionnaire items into successively more homogeneous groups with regard to symptomatic knee and hip OA. All the questions related to main symptoms (pain, stiffness, functional limitation, and insecurity [only for the knee]) and previous OA status reported by any physician were selected from the original questionnaire to find the optimal classification tree. Two different approaches of decision cost matrix were considered in the CART analysis. One penalized equally a false negative and a false positive, and the other penalized a false negative twice as much as a false positive. An optimal tree was created for hip OA with only pain and previous OA status reported by any physician, whereas for knee OA the items selected were limitation when going down steps or rising from a chair, previous OA status reported by any physician, and stiffness. Greater consistency between sensitivity and specificity was found when false negatives and false positives were penalized equally.
All effects were considered significant at p < 0.05 unless otherwise noted. Main statistical analyses were performed using SAS for Windows statistical software, version 8.2. (SAS Institute, Inc., Carey, NC). CART analysis was performed using CART with S-Plus software [19].