Knee pain is a common, and often persistent, problem in middle and old age.1^–6 It is also a major cause of locomotor disability.7^–9 However, significant functional limitation among older adults with knee pain is not inevitable nor does it follow a fixed course over time. Individual variation in the course of functional limitation has been found in community-based cohorts of adults with knee pain10 and knee osteoarthritis (OA),11 and in patient cohorts recruited in primary care12 and hospital settings.13 Identifying which people with knee pain will either develop significant functional difficulties or fail to recover from them is a crucial step in directing prevention and management in primary care—the healthcare setting where most older adults with knee pain will be assessed and managed.

Despite being the commonest presenting pain problem seen in older patients,14 there is little empirical evidence on how to identify patients with knee pain at high risk of progressive or persistent functional limitation. Recent research has drawn parallels between knee pain in older adults and other regional pain syndromes—for example, low back pain.15^–17 Indeed the same set of “generic” prognostic indicators seem to be associated with poor outcome across several different regional pain syndromes.18 These include presenting with severe pain, longer standing pain and mood disturbance. But do such generic factors provide a sufficiently accurate prediction of future functional outcome without the need for additional information on the state of the joint itself; local signs and symptoms gathered from clinical history, physical examination or other investigations?

In this study we attempt to predict progressive or persistent functional limitation at 18-month follow-up in a prospective cohort of community-dwelling adults aged ⩾50 years with knee pain. Our objectives were to determine (a) whether local signs and symptoms gathered from clinical history, physical examination and plain radiography provide additional prognostic information once “generic” prognostic indicators have already been taken into account, and (b) whether generic and local factors can identify groups at high or low absolute risk of poor functional outcome.

METHODS

The Clinical Assessment Study of the Knee—CAS(K)—is a population-based prospective observational cohort study of 819 symptomatic subjects, aged ⩾50 years, registered with three general practices (irrespective of their actual consultation pattern). The study was approved by North Staffordshire Local Research Ethics Committee. Full details of the study design and methods have been previously presented.19^–21 Between August 2002 and September 2003, respondents to two postal questionnaires who reported knee pain in the past 12 months were invited to attend a research clinic for a standardised clinical interview and examination and plain radiographs of both knees. Participants were sent a postal follow-up questionnaire 18 months after their baseline clinic attendance.21

As reported previously,22 of the 819 participants attending the research clinics, 621 were eligible for inclusion in the current analysis (74 exclusions at baseline: total knee replacement in index knee (16), existing diagnosis of inflammatory disease (14), incomplete x-ray data (12), no pain reported in the previous 6 months on attending clinic (32); 124 exclusions at 18 months: non-responders to 18-month questionnaire (37), incomplete Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) data at both baseline and 18 months (87)).

Outcome measurement

The primary outcome was change between baseline and 18-month follow-up in the Physical Functioning scale of WOMAC (WOMAC-PF) LK 3.0.23 The WOMAC-PF is a disease-specific (hip or knee OA) self-report instrument with 17 items, yielding a score between 0 (good function) and 68 (poor function). It has been extensively validated and is recommended for use in studies of subjects with knee OA.24 We used an adapted version that instructed participants to rate difficulty “due to their knee pain”.25

Using WOMAC-PF scores at baseline and 18-month follow-up, we applied the same definition of “poor outcome” as in a recent analysis by Sharma et al.11 Baseline and 18-month WOMAC-PF scores were categorised into five equal-sized groups, based on their distribution (0–7, 8–14, 15–22, 23–33, 34–68). For each participant a good functional outcome at 18 months was defined as moving down one group compared with baseline or remaining in one of the two lowest groups. A poor functional outcome was defined as moving into a higher group or remaining within the three highest groups. Three hundred and seventy-two (60%) participants were classed as experiencing poor functional outcome at 18 months (300 persistent functional difficulties, 72 incident functional difficulties).22

Prognostic indicators measured at baseline

Generic

In a previous paper exploring “generic” prognostic indicators in this cohort, four variables—age, body mass index (BMI), anxiety and high intensity knee pain—emerged from a total of 20 possible generic indicators as the best predictors of poor functional outcome at 18 months.22

Clinical history and physical examination

Assessment was based on simple, largely non-instrumented, measurement designed to be practicable for use within a routine primary care consultation. The content of the standardised clinical interview and physical examination was informed by medical textbooks, expert opinion and consensus development with general practitioners and rheumatologists.26 27 Potential prognostic indicators were selected for this analysis if they were clinical signs or symptoms with a known or putative link to functional status,11 27 28^–36 the development and progression of knee OA37 38 or were clinical manifestations of OA or other differential diagnoses.39^–44

Plain radiography

Weightbearing posteroanterior semiflexed and supine skyline and lateral views in 45° of flexion of both knees were taken. A single reader blinded to clinical status scored all films. Severity of radiographic OA was defined as none, mild or moderate/severe based on all three views.45 46 Our definition of radiographic OA essentially applies the same principle as earlier studies (ie, definite osteophyte47) but has extended this to all three views of the knee.

Box 1 lists all the prognostic factors.

Statistical analysis

Participants identified their affected or more affected knee in the clinical interview and clinical signs and symptoms and x-ray score relating to this “index knee” were used in the analysis. For continuous indicator variables or ordinal variables with several levels, cut-off points were chosen before univariable analysis on the basis of previous empirical evidence or, in the absence of this, frequency in the current sample.

Cox regression, applying a constant risk period,51 was used to quantify the association between potential prognostic indicators and poor functional outcome at 18 months. Results were presented as risk ratios (RRs) with 95% confidence intervals (95% CIs)—the same methodology used to derive the “generic” model.22

We evaluated the added prognostic value of three sets of variables: clinical history, physical examination and plain radiography, to the previously published “generic model”.22 For each set we performed univariate analyses to identify the strongest prognostic indicators of poor functional outcome at 18 months. Prognostic indicators with a univariate p<0.10 were entered into a backwards stepwise Cox regression (p_r ⩾0.05, p_e ⩽0.01) to produce a multivariate model that was refitted to all relevant participants without missing data for the variables selected. This process yielded four sequential models: generic only (G), generic + clinical history (G+CH), generic + clinical history + physical examination (G+CH+PE) and generic + clinical history + physical examination + plain x-ray examination (G+CH+PE+X). This sequence of analysis was designed to reflect the likely order of data gathering in clinical practice—that is, history, examination, simple investigation, complex investigation.52

The significance of each multivariate model was compared by performing a likelihood ratio (LR) test. The area under the receiver operating characteristic (ROC) curve and the associated 95% CI (comparable to the Harrell’s C statistic53) was used to describe the discriminative ability of each model. Model goodness-of-fit was assessed using calibration plots (splitting the sample into deciles and plotting predicted versus observed probability).54 For each multivariate model we described the proportion of participants who were predicted to have a high absolute risk of poor functional outcome at 18 months (⩾80% probability) and the proportion at low absolute risk (<20% probability).

Analysis was performed by Stata (version 7.0) and SPSS (version 14.0.2).

RESULTS

Prognostic value of generic indicators

As reported in our companion paper investigating the prognostic value of generic indicators, the LR test showed a significant additive value of the “generic” model compared with the null model (LR = 41.96, p<0.001). The area under the ROC curve for this model was calculated as 0.73 (95% CI 0.69 to 0.77).

Added prognostic value of clinical history variables

Poor functional outcome at 18 months was significantly associated in univariate analyses (p<0.10) with the following six baseline clinical history factors: bilateral knee pain, duration of morning stiffness, inactivity gelling, reported swelling, incident pain and reported giving way (table 1). After adjusting for each other, bilateral knee pain, duration of morning stiffness and inactivity gelling were independently associated with poor functional outcome at 18 months.

Table 2 shows the model refitted to all participants with complete data on these three factors. The model containing these three clinical history variables performed significantly better than the null model (LR = 29.03, p<0.001). The area under the ROC curve was calculated as 0.69 (95% CI 0.64 to 0.73).

The addition of the “clinical history” model to the “generic” model was accompanied by a significant change in likelihood ratio (LR = 10.82, p = 0.029) (table 3). The area under the ROC curve was increased slightly from the generic model at 0.76 (95% CI 0.72 to 0.80).

Added prognostic value of physical examination variables

Significant associations with poor functional outcome at 18 months were seen for eight of the physical examination factors: patellofemoral joint compression, bony enlargement, local tender point count, knee flexion ROM, quadriceps strength, hamstring strength and single-leg balance (table 1).

After adjusting for each other, local knee tender point count and single-leg standing balance were independently associated with poor functional outcome at 18-months. Table 4 shows the model refitted to all participants with complete data on these two factors. The LR test showed a significant additive value of the “physical examination” model compared with the null model (LR = 24.84, p = 0.001). The area under the ROC curve was calculated as 0.68 (95% CI 0.64 to 0.72).

The addition of the “physical examination” model to the “generic and clinical history” model was accompanied by a non-significant change in likelihood ratio (LR  = 6.34, p = 0.50), suggesting no added prognostic value of the two physical examination variables (table 3). The area under the ROC curve was virtually identical to that seen in the simpler “generic and clinical history” model (0.77 (95% CI 0.73 to 0.81)).

Added prognostic value of plain radiography

Functional outcome at 18-months was univariately significantly associated with severity of knee radiographic OA at baseline (table 1).

The addition of prevalent knee radiographic OA to the “generic, clinical history and physical examination” model was accompanied by a non-significant change in likelihood ratio (LR = 1.86, p = 0.39), suggesting no added prognostic value of this plain x-ray variable (table 3). The area under the ROC curve did not change compared with the simpler model (0.77 (95% CI 0.74 to 0.81)).

Model calibration and proportion identified at high and low absolute risk of poor functional outcome

Each of the models showed a similar pattern on calibration plots with all models overestimating the risk of poor functional outcome in those at lowest risk (fig 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Observed and predictability of poor outcome. G, general prognostic indicators; CH, clinical history variables relating to the knee; PE, physical examination variables relating to the knee; X, severity of radiographic OA in knee x-ray findings.

Poor functional outcome at 18 months could be confidently ruled in (absolute risk ⩾80%) for very few participants—less than 1 in 10 people in the current study (range 6.4–9.3%). The models were unable to confidently rule out (absolute risk <20%) poor functional outcome in any participants.

DISCUSSION

Our study has found that clinical history, physical examination and severity of radiographic knee OA on plain x-ray examination are of limited value over generic factors when trying to predict which older adults with knee pain will report worsening or persistent functional difficulties 18 months later. Several clinical signs and symptoms were associated in univariate analysis with a modest increase in risk of poor functional outcome (eg, knee extensor muscle weakness). However, after adjusting for other predictors, few provided unique prognostic information. We were concerned to investigate the added prognostic value of obtaining information on these characteristics in the sequence in which this would occur in clinical practice and over and above generic information that might be obtained from all patients presenting with pain. The combination of four generic factors—age, BMI, anxiety and knee pain severity—provided acceptable discrimination. Although the addition of clinical history variables did improve the model fit (LR test significant), there was little improvement in discrimination (area under the ROC curve was essentially unchanged).

Comparison with other studies

Although there are differences in the specific measures used, the positive association between poor functional outcome and older age, being obese or overweight and higher pain severity, together with a protective effect of good mental health and the lack of association with severity of radiographic OA are all consistent with previous work.36 Few previous studies have investigated the longitudinal association between clinical assessment variables and functional outcome in older adults with knee pain. Much of the existing knowledge in this field comes from a single, high-quality, community-based study of older adults with knee OA.11 Using the same definition of poor functional outcome, that study reported a 52% risk of poor functional outcome at 3 years compared with the 60% risk at 18 months found in the present study. We did not gather information on self-efficacy, social support, or aerobic exercise, all of which were found in the study by Sharma et al11 to be associated with functional outcome. In contrast to their findings, varus or valgus laxity was not associated with poor functional outcome in our study. We assessed this in the current study using simple, non-instrumented manual testing whereas Sharma et al11 used an instrumented device to apply a fixed load resulting in angular deviation measured in degrees.

Strengths and limitations

The data for this analysis were collected from a large, prospective cohort study with a very high follow-up rate, which has previously been examined for potential selective non-participation during the recruitment and follow-up of these participants.21

Our use of simple, predominantly non-instrumented assessment methods was designed to reflect the type of information gathered in a general practice consultation. The benefits of this approach may be offset by higher random error and misclassification of simple clinical signs and symptoms and this may well be one reason for the lack of associations found in our study, particularly for physical examination variables.55 This limitation notwithstanding, we would argue that our findings represent what can be achieved with simple assessment of clinical signs and symptoms of the sort that might be undertaken in routine general practice.

Discriminating between patients with a poor functional outcome and those with a good functional outcome is challenging. It requires much stronger associations than are usually observed in aetiological research,56 57 and a combination of factors that each provide added prognostic information conditional upon the others.58 The multivariate prognostic models in this study had “acceptable” levels of discrimination (0.7⩽AUC<0.8).55 However, applying these models in the present sample resulted in fewer than 1 in 10 subjects being identified as having a high absolute risk (>80% probability) of poor functional outcome and none at low absolute risk (<20% probability).

The generalisability of our findings is supported by their comparability with the earlier study of Sharma and colleagues, also performed on a cohort of patients who came from a community base.11 The two studies found almost identical cut-off points to derive the groups based on the WOMAC scores, similar proportions experiencing a poor functional outcome and prognostic models which contained generic, clinical history and clinical examination factors. However, if this approach were to be applied to data collected from subjects in another setting, such as patients presenting with knee pain in primary or secondary care, the definitions used might not be as discriminatory and hence the performance of the model might be inferior.

Reducing or preventing disability are among the principal objectives for managing knee joint pain in older adults.59 However, patient ratings of self-reported function, even on condition-specific or site-specific measures like the WOMAC, reflect more than just difficulties attributable to the knee problem.60 Functional status is also likely to be determined by other morbidities and this may partly explain why markers such as age and BMI were found to be stronger predictors than many local signs and symptoms.

Implications

The need for research on the value of simple clinical signs and symptoms has been repeatedly stated.61^–63 Clinical history, physical examination and, to a lesser extent, plain radiographs, are commonly used in general practice to assess patients presenting with knee pain. A recent survey of Oxfordshire adults aged ⩾65 years found that among those who reported visiting the general practitioner about their knee, 63% recalled receiving a physical examination and a knee x-ray examination was requested for 51%.64 Although these clinical activities serve many purposes (eg, diagnostic, directing treatment, monitoring impairment), predicting the likelihood of worsening or persistent functional difficulties can be done as well by considering simple generic indicators. However, the failure of both generic indicators and local signs and symptoms to identify people at high or low absolute risk is problematic. Frustratingly, after taking account of all these factors, the risk of poor functional outcome remains, for most patients, in a grey zone of intermediate probability—somewhere between 20% and 80% probability. Improved discrimination is required to support more confident prognosis in clinical practice. More reliable assessment of prognostic indicators, the addition of new indicators—for example, biomarkers, extending follow-up to distinguish longer-term trajectories from short-term fluctuations, incorporating multiple outcomes, trying to account for prognosis-altering events and treatments and direct comparison between clinicians’ prognostic judgments and empirical models—should be considered in future research.

CONCLUSIONS

We have previously established that generic factors form part of the pattern of characteristics indicative of poor progressive or persistent functional difficulties that the generalist may find helpful for predicting the future course, irrespective of the site of pain. Our findings in this paper suggest that these generic factors alone provide information on prognosis to which subsequent more detailed clinical history, physical examination or radiographic severity add little.