Is a “false-positive” clinical diagnosis of knee osteoarthritis just the early diagnosis of pre–radiographic disease?




In routine practice, diagnosis of knee osteoarthritis (OA) currently relies on the combination of conventional risk factors and the presence of cardinal signs and symptoms. However, their role in early diagnosis has received little attention compared with biomarker research.


Using data from 122 adults ages ≥50 years with knee pain but no definite radiographic OA, we tested whether the clinical diagnostic probability of OA, based on risk factors, signs, and symptoms, was associated with subsequent incidence of radiographic OA 3 years later.


Clinical diagnostic probability performed only modestly in discriminating incident radiographic knee OA (area under the receiver operating characteristic curve = 0.59, 95% confidence interval 0.49–0.70).


Improving the measurement of conventional markers and using study designs that test the ability of new biomarkers to add to or replace conventional markers are priorities for research in the early diagnosis of OA.


For primary care physicians, a working diagnosis of knee osteoarthritis (OA) can normally be made on clinical grounds without the routine need for imaging or laboratory tests (1). A “full house” of risk markers and indicative symptoms and signs identifies patients with a very high probability of having definite structural features of OA on plain radiography (2). However, most patients seen in primary care are unlikely to have such an unambiguous presentation and their radiographs may appear normal. Since plain radiography is known to be relatively insensitive to early disease (3), it is possible that a clinical diagnosis of knee OA in the absence of definite radiographic disease represents an early diagnosis of pre–radiographic disease. A recent cross-sectional study focusing on biomarkers found significant differences in simple risk markers (age, body mass index [BMI]) and clinical measures (pain duration and severity, degree of functional limitation) between no, pre-radiographic, and radiographic OA as defined by magnetic resonance imaging and plain radiograph (4). In this report we take a slightly different approach, using longitudinal data. Among symptomatic adults ages ≥50 years but with no radiographic knee OA at baseline, we tested whether a combination of selected risk markers and symptoms and signs indicative of radiographic OA (expressed as a “clinical diagnostic probability”) was related to a higher incidence of radiographic knee OA 3 years later.

Subjects and Methods


Subjects with current or recent knee pain were recruited from a 2-stage cross-sectional postal survey of all adults ages ≥50 years registered with 3 general practices in North Staffordshire (irrespective of actual consulting patterns). Respondents reporting knee pain within the previous 12 months were invited to attend a research clinic at a local National Health Service (NHS) Hospital Trust.

The study protocol and each phase of followup were approved by the Research Ethics Committee (projects 1430 and 05/Q2604/72), and details have been published elsewhere (5). All of the participants provided written informed consent to clinical and radiographic assessments and were asked for consent to medical record review to assist in excluding preexisting inflammatory disease.

The inclusion criteria for the current analysis consisted of: age ≥50 years, registered with one of the participating general practices at the time of the study, responded to postal questionnaires at baseline, consented to further contact, attended baseline and 3-year followup research clinics, and had full knee radiograph data at baseline and 3-year followup. The participants were excluded if they had not experienced knee pain within the 6 months prior to baseline clinic attendance (reported no days of knee pain in the last 6 months and/or a Chronic Pain Grade of 0) (6), had a preexisting diagnosis of inflammatory arthropathy in the medical records, had a total knee replacement in their most affected (“index”) knee, or had definite radiographic OA in their index knee at baseline. Therefore, the target population for this study was adults with current or recent knee pain in an age range at which knee OA would be one of the main diagnoses considered but in whom there was no definite radiographic OA at the time of clinical diagnosis.

Data collection.

All of the data were planned and gathered prospectively. At baseline, the participants underwent a standardized clinical interview and physical examination by 1 of 6 research therapists blinded to the findings from radiography, postal questionnaires, and medical records. Inter- and intrarater reliability have been reported elsewhere (7, 8). Training of the assessors took place prior to the study and was updated after every 100 participants recruited. Training included comparisons against rheumatologists, open and blinded comparisons against each other using “expert patients,” and peer observation. The assessors were issued a manual of detailed protocols for assessing each sign and symptom.

Plain knee radiographs were taken on the day of clinic attendance. Three views were taken of each knee: a weight-bearing semiflexed (metatarsophalangeal joint) posteroanterior (PA) view according to the Buckland-Wright protocol (9) and lateral and skyline views, both in a supine position with the knee flexed to 45°.

The participants filled in a brief self-complete questionnaire about their knee symptoms on the day of their clinic attendance. All of the participants were invited to return to the same local NHS Hospital Trust for repeat plain radiographs 3 years later.

Radiographic scoring and definition of incident radiographic knee OA.

An incident case of radiographic OA was defined as the new appearance at 3-year followup of mild or moderate/severe radiographic OA in a knee that had no definite radiographic OA at baseline. Only one knee per individual was analyzed: the “index knee.” In patients with unilateral knee pain at baseline, the index knee was this single painful knee. In individuals with bilateral knee pain, the most painful knee was the index knee. In situations where participants thought both knees were similarly painful, the index knee was selected at random. All of the participants with definite radiographic OA in their index knee at baseline were excluded from the current analysis.

A single reader (RD) read all of the films at baseline and 3 years, and was blinded to all clinical data collection at both time points. When reading the 3-year followup films, the reader was blinded to the participants' baseline radiograph results but obviously was aware that the films were from the 3-year followup. A Kellgren/Lawrence scale score, based on their original written description (10), was assigned to both PA and skyline views and graded as follows: grade 0 = no features of OA; grade 1 = minute osteophytes, doubtful significance; grade 2 = definite osteophytes, unimpaired joint space; grade 3 = moderate diminution of joint space; and grade 4 = joint space greatly impaired with sclerosis of subchondral bone. In the lateral view, superior and inferior patellar osteophytes were scored using a standard atlas (11). A score between 0 and 3 was assigned, where 0 = no osteophytes, and increasing scores from 1 to 3 = increasing osteophyte size. Osteophytes on the posterior tibial surface were judged on the same basis as patellar osteophytes in the lateral view. Posterior osteophytes do not appear in standard atlases but were included to establish a comprehensive radiographic assessment of the tibiofemoral joint. Posterior osteophytes were judged on the same basis as those for superior and inferior patellar osteophytes found in the Burnett atlas.

The definitions of mild and moderate/severe radiographic knee OA are given in Table 1. Mild radiographic OA essentially equates to the presence of a definite osteophyte and moderate/severe radiographic OA equates to joint space narrowing, although a very small number of individuals with grade 3 osteophytes but no joint space narrowing were classified as having moderate/severe radiographic OA (12).

Table 1. Definitions of radiographic knee OA*
 No definite radiographic OADefinite radiographic OA
  • *

    Incident radiographic knee OA was defined as having no definite radiographic OA in the index knee at baseline and mild or moderate/severe radiographic knee OA in the same knee at 3-year followup. OA = osteoarthritis; K/L = Kellgren/Lawrence; PA = posteroanterior.

Patellofemoral jointSkyline K/L scale grade = 0–1 AND lateral osteophytes = 0Skyline K/L scale grade = 2 OR lateral osteophytes = 1–2Skyline K/L scale grade = 3–4 OR lateral osteophytes = 3
Tibiofemoral jointPA view K/L scale grade = 0–1 AND posterior osteophytes = 0PA view K/L scale grade = 2 OR posterior osteophytes = 1–2PA view K/L scale grade = 3–4 OR posterior osteophytes = 3

Baseline clinical diagnostic probability.

In our previous cross-sectional diagnostic study, multivariable logistic regression was used to calculate a predicted probability of radiographic knee OA for each participant, based on the combination of selected risk markers and clinical signs and symptoms at baseline (13). These included: age (50–59/60–69/≥70 years), sex (male/female), BMI (<25/25–29.9/≥30 kg/m2), previous injury (yes/no), pain in the entire leg (yes/no), difficulty descending stairs (none/mild/moderate/severe/extreme), palpable effusion (none/mild/moderate/severe), fixed flexion deformity (yes/no), restricted knee flexion range of motion (yes/no), and coarse crepitus (none/possible/definite). These 10 diagnostic indicators originally had been selected using univariable and multivariable analysis of cross-sectional data from a total of 57 candidate indicators that were either known risk indicators for radiographic knee OA, clinical signs and symptoms with a known or putative link to the occurrence of radiographic knee OA, or clinical manifestations of differential diagnoses. This clinical diagnostic probability could range from 0% (rule out radiographic OA with certainty) to 100% (rule in radiographic OA with certainty).

Statistical analysis.

The incidence of any radiographic knee OA at 3 years was plotted against quintiles of baseline clinical diagnostic probability. This process was then repeated for the incidence of moderate/severe radiographic knee OA, patellofemoral joint radiographic OA, and tibiofemoral joint radiographic OA. To test whether clinical diagnostic probability was superior to age alone in predicting any incident radiographic knee OA, the area under the receiver operating characteristic curve was calculated for each model and compared using the algorithm by DeLong et al (14).


Of the 213 potentially eligible participants at baseline, 122 provided repeat plain radiographs at 3-year followup (mean ± SD age at baseline 62.0 ± 7.3 years; n = 80 [66%] women). In 21 participants (17%), their knee problem began within the previous 12 months, in 38 (31%) it began between 1 year and <5 years, in 33 (27%) it began 5 years to <10 years, and in 30 (25%) it began >10 years before baseline assessment. Sixty-six participants (54%) reported 1–30 knee pain days in the 6 months before baseline assessment, 34 (28%) reported 31–89 knee pain days, and 22 (18%) reported ≥90 knee pain days. Current pain intensity at the time of baseline assessment was generally mild (mean ± SD 2 ± 2 on a 0–10 numerical rating scale). Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) (15) pain scores (range 0–20) for the sample were essentially unchanged between baseline and 3-year followup (baseline: median 5 [interquartile range (IQR) 2–8], 3-year followup: median 5 [IQR 2–9]). WOMAC physical function scores (range 0–68) for the sample showed a small increase in functional limitation at followup (baseline: median 12 [IQR 3.5–21.5], 3-year followup: median 15.9 [IQR 3–26]). Those with 3-year radiograph data were not different in terms of sex, age, duration of knee symptoms, or baseline clinical diagnostic probability from those in whom repeat radiographs were not obtained at 3-year followup.

Forty-two participants (34%) had incident radiographic knee OA at followup, of whom 12 were classed as incident moderate/severe OA. The most common pattern of incident radiographic knee OA was isolated patellofemoral joint OA (n = 20), followed by combined patellofemoral/tibiofemoral joint OA (n = 12) and isolated tibiofemoral joint OA (n = 10).

Clinical diagnostic probabilities for the 122 participants ranged from 7.0–93.5%. There was no strong evidence of a positive association between baseline clinical diagnostic probability and the incidence of radiographic knee OA 3 years later (Figures 1 and 2). The possible exceptions were patients with a very low clinical diagnostic probability of knee OA. Probabilities below 40% (i.e., few risk markers and clinical signs or symptoms indicative of OA) were associated with a relatively low incidence of any radiographic knee OA at 3-year followup; above this, the incidence of any radiographic knee OA was uniformly higher. The threshold appeared to be higher for moderate/severe radiographic knee OA compared with mild radiographic knee OA (Figure 1) and for tibiofemoral joint radiographic OA compared with patellofemoral joint radiographic OA (Figure 2).

Figure 1.

Relationship between predicted probability of knee radiographic osteoarthritis (ROA) at baseline, based on risk markers and clinical signs and symptoms, and the incidence of knee ROA at 3-year followup, by severity. * = points on the x-axis show quintiles of predicted probability.

Figure 2.

Relationship between predicted probability of knee radiographic osteoarthritis (ROA) at baseline, based on risk markers and clinical signs and symptoms, and the incidence of knee ROA at 3-year followup, by compartment. PFJ = patellofemoral joint; TFJ = tibiofemoral joint; * = points on the x-axis show quintiles of predicted probability.

Clinical diagnostic probability discriminated incident radiographic knee OA slightly better than just age alone (area under the curve = 0.59, 95% confidence interval [95% CI] 0.49–0.70 versus area under the curve = 0.48, 95% CI 0.37–0.60), although the difference was not statistically significant (χ2 = 2.93, P = 0.0869).


We posed the question, “Is a ‘false-positive’ clinical diagnosis of knee OA just the early diagnosis of pre–radiographic disease?” Our analysis suggests that the answer is “no.” The possible exception is a case of knee pain where there is a low suspicion of OA at baseline. Such cases appear to be unlikely to develop incident radiographic knee OA in the medium term. This lack of association with baseline clinical diagnosis was also found by Thorstensson et al (16), who identified incident radiographic knee OA at 12-year followup.

But should we believe the current results? Several limitations need to be considered. First, our sample size was relatively small, permitting only a description as opposed to inferential testing. This particularly limited attempts to look at subsets of incidence. Second, despite no apparent bias on baseline characteristics, loss to followup was high and the effect of this on biasing the observed associations cannot be ruled out. Third, the model on which diagnostic probability was calculated contained several but not all known risk factors for the incidence of knee OA (17). Although age, sex, BMI, and previous injury were included, the presence of Heberden's nodes was not. This was investigated at baseline but found not to be associated cross-sectionally with the presence of radiographic knee OA. Furthermore, we did not gather data at baseline on previous intensive physical activity, certain occupational physical exposures (e.g., kneeling, squatting), or increased bone mineral density. The inclusion of these factors may improve the prediction of incident knee OA. Fourth, the diagnostic model has not been externally validated and relatively high levels of observer variability have been recorded for some of the clinical signs included in this model (8). This would lessen the chance of observing any true association between clinical markers and radiographic OA incidence, although we would argue that the reliability of clinical assessment in this research study was unlikely to be worse than that encountered in routine practice. Finally, our study highlights a key limitation in attempting to translate cross-sectional models based on prevalent cases to longitudinal models of incident cases: one explanation for the relatively poor predictive ability of baseline clinical diagnostic probability was that these were modeled on prevalent, predominantly combined patellofemoral/tibiofemoral joint radiographic OA, whereas the majority of incident cases involved radiographic OA in one compartment only. Clinical manifestations of established multiple compartment radiographic OA may not act as good indicators of early OA.

Despite these negative findings and the great interest in biomarkers, we would nevertheless support the inclusion of simple risk markers and clinical indicators in future attempts at early diagnosis of OA. They are likely to be accessible in all settings and have remained fundamental to risk prediction in other fields such as cardiovascular medicine (18). Improving the measurement of conventional markers and using study designs that test the ability of new biomarkers to add to or replace conventional markers (19) are priorities for research into the early diagnosis of OA.


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 submitted for publication. Dr. Peat 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. Peat, Thomas, Duncan, Wood.

Acquisition of data. Peat, Thomas, Duncan, Wood.

Analysis and interpretation of data. Peat, Thomas, Duncan.


We would like to acknowledge the contributions of Dr. K. Dziedzic, Dr. H. Myers, Dr. R. Wilkie, Dr. J. Hill, Ms J. Handy, and Ms C. Clements to the conception and design of the Knee Clinical Assessment Study (CAS[K]) and the acquisition of clinical data; the contributions of Dr. G. Carpenter, Dr. L. Coar, Dr. V. Cooper, Dr. P. Dawes, Dr. J. Greig, Dr. A. Hassell, Dr. M. Porcheret, Dr. M. Shadforth, and Dr. A. Rees to aspects of the conception and design of the CAS(K); Dr. J. Saklatvala, Ms C. Jackson, Ms J. Myatt, Ms J. Wisher, Ms S. Stoker, Ms S. Yates, Ms K. Hickson, Ms K. Wallbank, and Ms J. Bamford from the Department of Radiography; and Haywood Hospital, which conducted all study radiographs. We would like to thank the administrative and health informatics staff at Arthritis Research UK Primary Care Centre at Keele University and the staff of the participating general practices and Haywood Hospital. We also gratefully acknowledge the assistance of Professor C. Buckland-Wright for advice and training for the radiograph protocols.