In-depth interviews were conducted with a pre-determined sample size of 35 patients to elicit how their knee injury affects their lives. Purposive sampling was used to select a stratified sample in relation to age, sex, severity, stage of management and condition – meniscal or ligamentous injuries. Interviews were conducted in two centres, Cardiff and York. They were audio-recorded and transcribed. Thematic analysis of the transcripts was undertaken by two researchers independently.

The resulting items were reviewed by the trial management group and piloted through a postal survey of 80 patients recruited from orthopaedic and physiotherapy departments at York Hospitals NHS Trust. Questionnaires were assessed for data quality including missing data and response frequencies. The questionnaire also included a question asking patients if they would be willing to attend an interview to discuss the questionnaire in more detail. These cognitive debriefing interviews conducted with twelve patients, were designed to assess whether they had any difficulties with particular items and if there were important areas not covered by the questionnaire.

Focus groups held with clinicians and patients in Cardiff and York were designed to assess content and face validity; the extent to which the items address the intended subject matter and whether relevant aspects of health and quality of life are adequately covered. The clinician focus groups included two GPs and three physiotherapists (Cardiff) and a GP, physiotherapist and consultant radiologist (York). The patient focus groups included three (Cardiff) and six patients (York). All focus groups followed a semi-structured topic guide and patients completed the questionnaire.

The instrument was evaluated concurrently within the randomised trial designed to evaluate whether GPs should have access to MRI for patients with a suspected ligamentous or meniscal injury. Patients completed a questionnaire at trial enrolment that also included the EQ-5D 11, SF-36 12 and questions relating to knee symptoms. Patients who presented in general practice were eligible for inclusion in this trial if aged between 18 and 55 years inclusive and the GP was considering referral to an orthopaedic specialist for suspected meniscal or ligamentous injuries. The main exclusion criteria were that the patient needed urgent orthopaedic referral or had non-traumatic arthropathy, chronic instability of the knee, a previous MRI examination within the same episode of care, or previous surgical intervention (excluding diagnostic arthroscopy) on the same knee. These eligibility criteria were designed to help ensure a valid sample of patients were included in the trial. For example, the upper age limit of 55 years inclusive was used as this is when the onset of osteoarthritis occurs 13. The injuries consistently referred to are those of the menisci, and cruciate or collateral ligaments 1415 so these are the type of patients GPs were asked to recruit. GP's also attended a two-hour education seminar that included the clinical assessment and management of patients with knee problems 16.

Questionnaires were also sent to patients invited to take part in a postal survey in Cardiff which comprised two cohorts: all recent GP referrals to the University Hospital of Wales and Llandough Hospital for MRI or orthopaedic consultation; and patients listed for an arthroscopy of the knee. These questionnaires included the Lysholm Knee Score 17, a widely used clinically-derived instrument for assessing ligament injuries of the knee. Postal reminders were sent at two and four weeks, and patients were contacted by telephone at six weeks.

Individual items were assessed for missing data. Principal component analysis (PCA) was used to identify interrelated variables or components of quality of life within the questionnaire items 18. Items with poor component loadings were considered for removal. Items were further assessed for their contribution to the identified components with Rasch analysis 19, a form of item response theory (IRT) using the WINSTEPS programme 20. IRT assumes that a measurement construct such as knee-related quality of life can be represented by an ordered continuum that ranges from the minimum to maximum level of functioning. IRT models produce a hierarchy in which items are positioned according to the proportion of patients who report a limitation, and patients are positioned according to the total number of items in which they report a limitation 21. Two mean square fit statistics assess the extent to which unpredicted responses to an item are given by patients whose position in the hierarchy as determined by their knee-related quality of life, is either close to the items position (Infit statistic) or far from the items position (Outfit statistic) in the hierarchy of items. It is generally recommended that the two fit statistics range from 0.7–1.3 2122.

The instrument was assessed for internal consistency reliability using item-total correlation and Cronbach's alpha 10. Test-retest reliability was assessed through a second questionnaire mailed to a random sample of every second patient taking part in the GP referral cohort of the Cardiff survey at two weeks. Patients reporting no change on the knee-specific transition question "Compared with 2 weeks ago, how is your knee now?" were included in the test-retest analysis 1023. To be acceptable for use in groups of patients the estimates of reliability should exceed 0.70 1024.

The validity of the new instrument was assessed through comparisons with self-reported health status instruments and questions about knee symptoms. It was hypothesised that the new instrument would have a higher level of correlation of around 0.7 with the Lysholm Knee Score than the two generic instruments the EQ-5D and SF-36. Moderate levels of correlation of 0.5 were expected with the EQ-5D scores. Higher levels of correlation were expected for the EQ-5D items relating to mobility and pain than for self-care or anxiety-depression. Similarly, higher correlations in the range 0.5–0.7 were expected for the SF-36 scales of physical function, role-physical and bodily pain than for the scales of general health, vitality and mental health. Low to moderate levels of correlation above 0.4 were expected with patient responses to questions about number of days off work or normal activities in the past four weeks. It was hypothesised that instrument scores would reflect the different categories of symptoms in the form of a linear association, the highest and lowest KQoL-26 scores being for patients having symptoms none of the time and all of the time respectively. It was hypothesised that patients who had or were considering changing job because of their knee would have lower scores than their counterparts.

Responsiveness was assessed on return of a follow-up questionnaire at six months for the trial patients. This questionnaire included a knee-specific health transition question "Compared with six months ago, how is your knee now?" which uses a 15 point scale from "a very great deal worse" to "a very great deal better". An improvement of a little or somewhat better has been defined as important 23. Responsiveness was compared using effect sizes and responsiveness statistic. The former is equal to the mean change in score divided by the standard deviation of the baseline scores 10. The latter is equal to the mean change in scores divided by the standard deviation of score differences in stable patients 25. The larger the responsiveness statistic, the smaller the sample size needed for purposes of evaluation. The denominator was the standard deviation of the score changes for those patients in the test-retest analysis.

The study was designed to comply with the Declaration of Helsinki as adopted by the World Medical Association. Northern and Yorkshire Multi-Centre Research Ethics Committee reviewed and approved the study (reference number MREC/1/3/59).

Analysis of the interview transcripts produced a 38-item instrument which was scaled with three types of five-point adjectival scales of "totally limited/unable to do" to "not limited at all", "all of the time" to "none of the time" and "extremely" to "not at all". The instrument was piloted using a postal survey and 47 (58.8%) were returned completed. Twelve questionnaires were also completed by patients attending the focus groups. With regards to missing data, when piloting the questionnaire 52 (88.1%) patients completed all items. The focus groups and follow-up interviews confirmed the content and face validity of the questionnaire following the addition of two items about financial concerns and feelings of getting old relating to the knee problem.

The 559 patients who took part in the trial completed the 40-item questionnaire. Of the 547 patients invited to take part in the Cardiff survey, 323 (59.0%) returned a completed questionnaire. For the trial and survey patients the mean ages were 39.67 (sd = 10.23) and 47.02 (sd = 14.29) and the number of females was 204 (36.6%) and 146 (43.5%) respectively (Table 1).

The majority of items have four or fewer missing responses in total, the mean number of missing responses being 0.12 (sd = 0.67) with a range of 0 to 9 (2.3%) patients that did not respond to a particular item. All forty items were completed by 751 (85.1%) patients. The largest number of missing responses was eleven items for one patient which is just over one quarter of the items. Across the 0–4 response scale, means ranged from 1.05 to 3.47 for the items "avoiding turning, twisting or sideways movements" and "crossing the road" respectively. The former item had the highest floor effect of 35.4%. The item "staying seated for 15 minutes" had the highest ceiling effect of 62.0%. The spread of the item means indicates that items are measuring different degrees of knee-related quality of life.

Principal component analysis gave five components with eigenvalues above 1.0 explaining 64.1% of the total variation between patients. The great majority of the items were making a strong contribution to the first component of the unrotated solution. Rotation served to identify clinically recognisable aspects of knee-related quality of life; the first, second and fourth components comprising items relating to physical functioning, activity limitations and emotional functioning respectively. The third and fifth components included six items relating to getting into or out of a lowered position and two items relating to bending and kneeling down respectively. The majority of these items and items within the physical functioning component had higher loadings within the first factor of the unrotated solution and the majority had loadings over 0.4 within the first component of the rotated solution. They were therefore assessed for their contribution to the hypothesised scale of physical functioning. As shown in Table 2, all items have component loadings above 0.5, the great majority above 0.6.

Table 2 shows that for three scales the majority of items fit the Rasch model with Infit and Outfit statistics between 0.8 and 1.2, the great majority within the recommended range 0.7–1.3 22. This is further evidence that the instrument comprises three scales of knee-related quality of life. Within physical functioning, items 8 "bending down" and 17 "staying seated for two hours" have poor Infit and Outfit statistics which indicates that they do not fit the Rasch model very well and may not be sufficiently contributing to the construct of physical functioning. Similarly, three items within activity limitations and two items within emotional functioning have poor fit statistics. Item 22 is borderline.

For the three scales, all but one of the item-total correlations are above 0.40. Removal of items 6, 8, 17, 18, 19 and 20 from the physical functioning scale which includes those performing poorly in the Rasch analysis, gave final item-total correlations of 0.60–0.81 and a Cronbach's alpha of 0.94. Removing four items from the activity limitations scale including those performing poorly in the Rasch analysis, gave final items-total correlations of 0.70–0.82 and an alpha of 0.91. Removing four items from the emotional functioning scale including those performing poorly in the Rasch analysis, gave final item-total correlations of 0.74–0.81 and an alpha of 0.91.

Of the 109 randomly selected patients from the GP referral cohort of the Cardiff survey, 86 (78.9%) completed a two-week retest questionnaire. The intra-class correlations for the two sets of scores for the 27 patients stating that there was no change in their knee were 0.93, 0.87 and 0.80 for the physical functioning, activity limitations and emotional functioning scales respectively. These reliability estimates meet the criteria necessary for group level comparisons 1026.

The score distributions for the new instrument now referred to as the KQoL-26, are computed by summing the items where half or more are completed for the relevant scale and transforming to a 0–100 scale. The scores were approximately normal on a scale of 0–100 where 100 is the best possible knee-related quality of life. Means scores were 59.1 (sd = 18.8), 51.05 (sd = 25.2) and 40.4 (sd = 21.4) for the physical functioning, activity limitations and emotional functioning scales respectively. Therefore the instrument has the potential to measure differences in knee-related quality of life in cross-sectional or longitudinal studies.

The results of the tests of validity are shown in Tables 3 and 4. The correlations between KQoL-26 scores, the Lysholm, EQ-5D, SF-36 and knee-related questions are shown in Table 3. As hypothesised the highest correlations are with the Lysholm Knee Score. This is higher than that hypothesised which may reflect the high level of reliability of the new instrument 1026. As hypothesised, in general the KQoL-26 had the highest correlations with aspects of the EQ-5D and SF-36 relating to physical health. The KQoL-26 physical functioning and activity limitations had the largest correlations with the EQ-5D mobility and pain items. As expected the emotional functioning scale had the largest correlation with the EQ-5D anxiety and depression item. The KQoL scales had a similar moderate level of correlation with EQ-5D index scores. The KQoL-26 physical functioning and activity limitations scales had the largest correlations with the SF-36 scales of physical functioning, social functioning, bodily pain and role-physical. The emotional functioning scale had the largest correlations with social functioning and role-emotional. This KQoL scale also had the highest correlation with the SF-36 mental component summary scores, the other two scales having higher correlations with the physical component summary scores. The correlations with the number of days off work and on which normal activity was prevented because of the knee were of a low to moderate level. As expected, the activity limitations scale scores had the highest level of correlations with responses to these two questions.

Table 4 shows the results of the tests of validity in comparison to knee symptoms and change of job. The results are statistically significant for 26 of the 30 tests. The sample sizes are relatively small for those patients who have changed or are considering changing their job because of their knee, however three of the results are statistically significant, the largest differences being found for the activity limitations scale.

Of the 559 trial patients sent a six month questionnaire, 472 (84.4%) responded. Of these, 65 indicated that they had undergone an improvement of "a little better or somewhat better" on the knee-related health transition question. Table 5 shows the score changes, effect sizes and responsiveness index for these patients. With the exception of four SF-36 scale scores, all the instruments show a significant score improvement at six months. The KQoL-26 shows an improvement of 16.41, 23.38 and 23.46 points on scales of physical functioning, emotional functioning and activity limitations respectively. The KQoL-26 has the three highest effect sizes and responsiveness statistics.