Osteoarthritis (OA) is the most prevalent joint disease world wide and creates a massive economic burden.1 There is a paucity of data on the epidemiology of knee and hip OA in North America and Europe.2^–9 Recently, the Dutch Institute for Public Health produced data on the prevalence of radiological OA with pain in the Dutch population9: 2.5% for men and 5% for women for hip OA and 3.9% for men and 9.9% for women for knee OA. Another recent Greek study showed a symptomatic knee and hip OA prevalence of 6% and 0.9%, respectively.8 Epidemiological data are not available for France. Furthermore, a corollary to the rapid ageing of the population in many industrialised countries is a sharp rise in the prevalence of OA, which limits the validity of earlier epidemiological studies. Accurate prevalence estimates are needed to evaluate the costs of OA and the healthcare needs generated by the disease, knowledge of which is crucial to optimisation of resource allocation.

Prevalence estimates differ widely according to whether only moderate and severe radiographic changes are considered or mild abnormalities are also included. Moreover, several definitions of OA have been used in prevalence studies. Radiographic criteria are the most widely used4 5 10^–12 but hold little interest in terms of public health, as radiographic abnormalities may be asymptomatic. The use of clinical criteria probably leads to overestimation of the prevalence of OA.2 3 5 The best method consists in combining radiographic and clinical criteria,5 8 10 13^–17 which is more challenging and more costly than using either set of criteria alone.

Implementation of the clinical and radiological investigations that are needed to confirm OA is complex and difficult to consider on a large scale in studies designed to obtain population-based prevalence estimates.18 Screening using a specifically designed questionnaire is less costly and easier to perform. The goal of a screening questionnaire is to select a group of people in whom further investigations will yield a high rate of OA diagnoses (high specificity), without missing a substantial number of patients with OA (high sensitivity). Self-administered questionnaires designed for OA screening have shown inadequate sensitivity and specificity.18

Previously, we successfully developed a two-step telephone procedure for estimating the prevalence of rheumatoid arthritis and spondyloarthropathies in France.19^–21 Here, our objective was to examine the feasibility and validity of a two-step telephone screening procedure designed to estimate the prevalence of symptomatic hip and knee OA in a population-based sample.

POPULATION AND METHODS

Design

A pilot two-step survey was conducted in two geographical areas in France (Alpes Maritimes and Bretagne) (fig 1). The survey involved telephone interviews followed by confirmation testing in subjects with suspected OA.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Screening procedure and case ascertainment. OA, osteoarthritis.

Sampling

Assuming a 10% prevalence of symptomatic hip and knee OA in the general population aged 40–75 years, 1400 subjects were required to obtain 5% precision. Home phone numbers were taken at random from national phone company lists in the two selected geographical areas. Phone numbers of business offices and institutions were not used. At each number, a person aged 40–75 years was selected at random using the next birthday method22 and invited to participate in the study.

Ethical considerations

The study was carried out according to the Declaration of Helsinki and good clinical practice guidelines. The study protocol was approved by local institutional review boards and the independent ethics committees of Nice (France). Written informed consent was obtained from all patients who underwent investigations.

Screening questionnaire

The questionnaire evaluated signs and symptoms of knee/hip OA, American College of Rheumatology (ACR) criteria for OA,23 24 and whether the patient had a diagnosis of OA confirmed by a doctor. It was designed to be easily administered over the telephone by a trained interviewer who was not a health professional. To evaluate feasibility and make final improvements, preliminary testing of the questionnaire was performed in a small panel of healthy volunteers and patients with OA.

The questionnaire time frame was the past 4 weeks. The questionnaire comprised four items about the hip and four about the knee (box 1).

Screening and confirmation procedures

Screening (fig 1) comprised a telephone questionnaire administered by a non-healthcare professional. People who gave a positive reply to one or more items were interviewed again, by a rheumatologist, who used the same questionnaire but was not told which item or items were positive during the first interview.

Among subjects whose first screen was negative, a random sample was taken. People in this sample were interviewed over the telephone by a rheumatologist, who sought to confirm the absence of symptomatic hip or knee OA. Based on the results of this second interview, the person was either classified as free of OA or invited to undergo confirmation testing.

For each respondent with a positive screen, a rheumatologist contacted the general practitioner (GP) or rheumatologist for confirmation of the diagnosis. The GP or rheumatologist then examined the patient and obtained radiographs and erythrocyte sedimentation rate if not done previously to confirm the diagnosis. Respondents with positive screens who had no GP or rheumatologist were invited to the study centre (in each geographical area) for a physical examination and radiography. The diagnosis of hip/knee OA was based on “combined clinical and radiographic” ACR criteria23 24 (table 1) and on a radiographic Kellgren–Lawrence stage of ⩾225 in at least one knee or hip.

Radiographs

Radiographs of the target joint(s) were obtained in all subjects who had a positive screening questionnaire. When one or more knee items were positive, the weightbearing posteroanterior semiflexed view and lateral view were obtained. In subjects with one or more positive hip items, an anteroposterior view of the pelvis and Lequesne’s “false profile” view were obtained. All films were sent to a single reader (BM) for scoring using the Kellgren–Lawrence method,25 which distinguishes five stages (from stage 0: normal x rays of the hip/knee, to stage 4: advanced hip/knee OA).

Statistical analysis

The following sociodemographic and radiographic data were analysed: age, sex, number of subjects in the household, occupation and radiographic stage.

Various diagnostic criteria for OA were used to compute positive predictive values (PPVs) of the screening questionnaire directly from the data collected by the non-healthcare professional interviewer. The negative predictive value (NPV) was estimated from the sample of 140 subjects with negative screens. For both PPV and NPV, 95% CIs were computed.

Sensitivity and specificity of the screening questionnaire could not be computed directly from the data, because not all subjects with negative screens were evaluated. We computed direct estimates using the data from the random sample of subjects with negative screens who were evaluated,26 as well as corrected estimates taking verification bias into account.27

Prevalence estimates for radiographic symptomatic OA of the knee and hip were computed, with correction for sampling fractions, and their 95% CIs were determined. Subjects who underwent confirmation testing were considered to represent a random sample of those with positive and negative screens. Therefore, the corrected prevalence estimate was computed by taking into account the number of subjects with positive screens who underwent confirmation testing, the total number of subjects with negative screens and among them the number who underwent confirmation testing, the numbers of true and false positives and the sample size.28 29

RESULTS

Screening results

In 2005, 8341 phone numbers in the Alpes Maritimes (n = 4645) and Bretagne (n = 3696) regions of France were called by non-healthcare professionals. Of the 4879 calls that were answered, 1380 resulted in the random identification of a person aged 40–75 years, based on the next birthday method, who agreed to participate to the study (20% refused to participate, 30% were out of the age range and 21.7% were false telephone numbers). Among these 1380 subjects, 479 (211 in Alpes-Maritimes and 268 in Bretagne) had a positive screen for symptomatic knee or hip OA; mean age 58.3 years and 68% of the group were women. Among these 479 subjects, 309 (65%) accepted screening by a rheumatologist, which was positive in 306 subjects, for knee OA in 210, hip OA in 194 and both in 98.

Among these 306 subjects, because of loss to follow-up 109 only underwent confirmation testing. Sixty-nine patients showed OA (40 knees and 36 hips) with a Kellgren–Lawrence grade of ⩾1; 50 patients (20 hips and 35 knees) showed a grade of ⩾2. Fifty four per cent of this subjects had had a prior OA diagnosis made by a GP or a rheumatologist. No differences were found between the two geographical areas. Up to 2 months elapsed between the two screens.

Of the 901 subjects who had negative findings from the screen performed by the non-healthcare professional, 140 were selected at random and screened by a rheumatologist. The screen was positive in 10 subjects, who underwent confirmation testing. The radiographs confirmed the diagnosis of symptomatic OA in two of these subjects.

We compared the patients who were lost to the study either between the first and the second telephone screens or between the second screen and confirmation testing and the patients who completed the study procedures. No differences were found in age (p = 0.08) or sex (p = 0.9).

Performance of the full screening procedure

The two-step screening procedure had 87% sensitivity (95% CI 79% to 95%) and 92% specificity (95% CI 91% to 93%) for detecting knee OA. For detecting hip OA, sensitivity was 93% (95% CI 86% to 100%) and specificity was 93% (95% CI 92% to 94%).

Of the 1380 subjects interviewed by non-healthcare professionals, 1311 (95%) answered all eight questionnaire items. At least one item yielded a positive answer in 479 subjects, with no difference between the two geographical areas. Among 109 subjects who underwent confirmation process, 73 had a positive item for knee OA and 53 for hip OA. Screening by a rheumatologist was positive for knee OA in 70 (70/73, 95.9%) patients and for hip OA in 51 (51/53, 96.2%) patients detected “positive” by non-healthcare professional screening.

The NPV of the full screening procedure was 99% for the hip and 98% for the knee. The PPV of the full screening procedure was 56% for hip and 51% for knee (table 2).

The PPV of screening by non-healthcare professionals was 46% for hip OA and 62% for knee OA. The item with the highest PPV was a previous diagnosis of OA (78% for knee OA and 61% for hip OA). PPVs were lowest for knee swelling (58%) and hip pain (50%)

Age had no influence on performance characteristics (p = 0.18). A previous diagnosis of OA was significantly associated with radiographic scores of 3 or 4, both at the hip (p = 0.002) and at the knee (p = 0.01).

The PPV of screening by rheumatologists was 47% for hip OA and 65% for knee OA. The item with the highest PPV was a previous diagnosis of knee OA (86%) and the item with the lowest PPV was hip pain (45%) in the screening by rheumatologists.

Prevalence estimates of knee and hip OA

Corrected prevalence estimates for symptomatic OA in the general population were 9% (95% CI 7.7% to 10.3%) for knee OA and 9% (95% CI 7.8% to 10.2%) for hip OA. These estimates were computed using a Kellgren–Lawrence score of ⩾1 to define OA. When we used a radiographic score of ⩾2 to define OA, corrected prevalence estimates were 7.6% (95% CI 6.4% to 8.8%) for knee OA and 5% (95% CI 3.9% to 6.1%) for hip OA. Figure 2 shows the distribution of knee and hip OA cases by age group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Hip and knee osteoarthritis prevalences by age.

DISCUSSION

We developed and tested a two-step procedure for identifying cases of symptomatic hip and knee OA in the general population. Our results show that cases of symptomatic hip and knee OA can be detected using a limited set of items including a self-reported diagnosis of OA, symptoms and ACR classification criteria.23 24 We have successfully used the same method to screen for rheumatoid arthritis and spondyloarthropathies.19 20

Screening instruments for knee or hip OA have performed poorly, particularly when used in single-step procedures.18 The validity in the general population and in primary-care settings of questionnaires based only on ACR clinical criteria has been challenged.30^–33 ACR criteria were not designed as epidemiological tools, although they are often as such. Their usefulness depends on the objective being pursued.23 In epidemiological studies of OA, clinical ACR criteria may underestimate disease prevalence. They are as insensitive for use in community, though specific, an opinion expressed by several authors. Their performance seems linked to disease severity.14 33 Furthermore, restricting the definition of symptomatic radiographic OA to subjects with current symptoms excludes those with radiographic abnormalities but intermittent symptoms. The most important feature may be persistent pain accompanied by functional impairment during a defined period. The choice of the time frame used to assess knee pain may affect the results of prevalence estimates for symptomatic knee OA. The time frame varied across studies (6 weeks,13 1 month11 34) and current knee pain in other studies. We used a questionnaire on symptoms over the past 4 weeks and previous diagnosis, according to a two-step procedure with confirmation testing. This procedure allowed us to detect hip and knee OA, while ruling out specific inflammatory diseases and other non-articular causes of knee and hip symptoms.

It has been shown that minor changes in item content can modify prevalence estimates.33 The wording of our questionnaire items has been widely used to estimate the prevalence of symptomatic knee or hip OA.11 35 36 Knee pain was the single feature most strongly associated with the presence of definite radiological osteophytes in our study and in earlier work.37 Hip pain, however, was not associated with osteophytes in our study, possibly because many patients have difficult recognising hip pain as such. We specified the main locations of hip pain (groin or upper thigh), in keeping with earlier studies such as the SOF (Study of Osteoporotic Fractures) and the NHANES I.12 34

Sensitivity of ACR clinical criteria for knee OA is 95% and specificity is 69%.23 ACR criteria for hip OA are clinical, biological and radiological; they have 89% sensitivity and 91% specificity.24 Knee pain in subjects older than 50 years is nearly always caused by OA, whereas hip pain may reflect many other conditions.

Ideally, a screening tool should be both sensitive, so that few cases are missed (whether for clinical trials or healthcare planning) and specific, so that few of the confirmation tests are unnecessary. Our goal was to develop a screening procedure that was suitable for large populations and did not require an initial examination by a doctor. Sensitivity and specificity of our screening questionnaire are satisfactory and better than those of previously reported clinical questionnaires.18 38

Our study establishes that trained non-healthcare professionals can reliably administer our screening questionnaire. Of 309 subjects interviewed by rheumatologists, 306 had a positive screen (96.2% for knee OA and 95.9% for hip OA). Thus, rheumatologist screening could be omitted, which would substantially reduce the cost of screening. To improve the sensitivity of our questionnaire, we considered that a single positive item was sufficient to classify a person as having a positive screen. Of the 140 subjects with negative screens administered by non-healthcare professionals, only 10 (7.1%) had a positive rheumatologist-administered screen and of those 10 only 2 (2/140, 1.4%) had OA by confirmation testing. Other reports also found high concordance rates between trained metrologists and rheumatologists.38^–41

Our questionnaire had only eight items. Although a larger number of items might improve performance, a brief questionnaire is more likely to be accepted by the interviewee. Therefore, we discarded several of the items related to knee/hip pain and activities that were used in major OA studies. The excellent PPVs obtained with our questionnaire support this decision.

A potential limitation of our approach to screening for symptomatic OA is that, in addition to OA, several other conditions may cause local or referred knee or hip pain. Therefore, some of the subjects classified as having symptomatic OA in our study may in fact have had asymptomatic radiographic OA and pain due to another condition. These false positive cases would result in overestimation of the PPV of our questionnaire. This limitation is shared by similar reported studies. To limit its impact, we used an item about previous diagnosis of OA and we performed physical examinations in subjects who underwent radiographs. Another limitation of our study is that 35% of subjects with a positive screen administered by non-healthcare professionals failed to undergo the rheumatologist interview. Up to 2 months elapsed between the two screens in our study. Shortening this interval to 2 weeks might reduce the attrition rate. However, our results strongly suggest that the rheumatologist screen may be unnecessary. Several subjects were lost to follow-up between screening and confirmation steps with possible under- or overestimations of OA prevalence rates linked to possible selection of patients with knee/hip symptoms. But the statistical method used shows that losses to follow-up patients are not different from others. The absence of differences in sex and age between the two groups reinforces this hypothesis.

Whereas many studies report raw prevalence estimates, we computed corrected estimates that took into account the performance of the screening procedure. Methodological and age distribution differences complicate comparisons with earlier works. However, our results for knee OA are similar to those obtained in NHANES III13; neither study included patellofemoral OA. Prevalence estimates vary depending on whether Kellgren–Lawrence scores of 1 or 2 are included in the definition of OA. A clear and preferably standardised definition of OA is a prerequisite for comparisons of estimates across studies. To limit potential difficulties in differentiating Kellgren–Lawrence scores of 1 and 2 at the knee, radiographs should be read by experienced doctors using a standardised procedure.

The use of a questionnaire alone to identify subjects with knee OA for epidemiological surveys or clinical trials seems unrealistic.18 A physical examination and radiographs are needed for the definite diagnosis.

This pilot study shows that a two-step procedure is effective and feasible for identifying cases of symptomatic knee or hip OA. This approach is less costly than routine evaluations by a doctor. Furthermore, our results provide the first prevalence estimates for symptomatic hip and knee OA in two French regions. Finally, it yields valuable information for estimating the size of population-based samples needed to evaluate the nationwide prevalence of hip and knee OA. It is the tool used for the 3000 OA study which is currently taking place in France.