METHODS

Between January 1995 and December 1996, 1037 consecutive patients under the age of 76 admitted to hospital for hip or knee joint replacement due to advanced radiographic OA were eligible for recruitment in four tertiary centres in southwest Germany. Patients with signs of RA (according to the American Rheumatism Association 1987 criteria⁸), inflammatory joint disease (with history or laboratory evidence), hip fracture, or previous contralateral joint replacement were excluded. Of those eligible, 212 (20%) were operated on before they could be examined and 16 (1.5%) refused to participate. All remaining 809 participants gave written informed consent and were interviewed and examined by trained doctors according to a standardised questionnaire and examination procedure. The study was approved by the ethics review committee of the faculty of medicine of the university of Ulm and has been described in detail elsewhere.⁹

After documentation of high inter- and intrarater reliability,^10,11 hip and knee radiographs (both sides) were graded according to the Kellgren and Lawrence (K&L) score using a standard reference atlas from 0 to 4.^12,13 All participants had radiographic OA of the joint being replaced because this was an inclusion criterion. If radiographic OA (K&L ⩾2) was present in the joint contralateral to the one being replaced, the participant was categorised as having bilateral OA.

In addition, hand radiographs were taken and all finger joints (proximal interphalangeal and distal interphalangeal) and first carpometacarpal joints were studied for joint space narrowing, osteophytes, and subchondral sclerosis according to the atlas of Altman et al.¹⁴ If OA was found in two or more finger joints and one or both 1st carpometacarpal joints in addition to the OA of the replaced joint, the participant had radiographically defined OA in at least three separate joint groups and was considered as having generalised OA.¹⁵

Current intensity of pain in the joint to be replaced and the contralateral joint was separately assessed using visual analogue scales (VAS) with a horizontal bar of 10 cm length with vertical bars at both ends marked “no pain” (left) and “extreme pain” (right).¹⁶ This question was not defined for any specific situation or time scale and was part of a module asking different questions about pain in the hip and knee joints, respectively, including duration of pain, pain with different activities, and diurnal variation. The VAS score was measured in millimetres from the left hand vertical bar to a vertical line through the midpoint of the corresponding marking (usually a cross). Pain, swelling, and stiffness in 23 different joint regions, including different parts of the spine (neck, thorax, and lower back), shoulders (both sides separately for all paired joints), elbows, hands (wrist, thumb base, metacarpophalangeal, proximal interphalangeal, and distal interphalangeal joints), as well as lower extremity (hip, knees, ankle, and 1st metatarsophalangeal joint), with the exception of the affected joint (and the contralateral joint), were separately asked for. Painful joint regions were first summed and then dichotomised into one or more painful joint region (at least one painful joint region in addition to the joint being replaced) versus no additional painful joint region. Pain during different tasks, stiffness, and functional impairment were assessed using the hip- and knee-specific Western Ontario and McMaster Universities OA index (WOMAC)¹⁷ on a five point Likert scale ranging from 0 (none) to 4 (extreme). According to the number of items, the corresponding maximum (most extreme) values were 20, 8, and 68 for pain, stiffness, and functional impairment, respectively.

The brand names of all drugs taken daily during the last three months before hospital admission were asked for during the interview. Drugs were classified according to the Anatomical-Therapeutical-Chemical classification (ATC classification, WHO Collaborating Centre for Drug Statistics Methodology, Oslo 1993) by a pharmacist. The coding allowed for up to three different active ingredients per drug and up to 15 different drugs per patient. Patients using systemic steroids (n = 6) were excluded from all analyses, because steroids are usually given for systemic inflammation and are known to affect serum CRP.

Body weight and height were obtained by interview during the physical examination. Body mass index was calculated as a measure of weight relative to the height by dividing the weight in kilograms (kg) by the squared height in metres (m).

Non-fasting serum samples were obtained preoperatively by standard venepuncture, stored at −80°C, and analysed together in a central laboratory. High sensitivity CRP was measured by nephelometric method using a commercial kit (NA-latex CRP, Behring Werke, Marburg, Germany) calibrated with the WHO reference standard 85/506.

Statistical analyses

We first described the patients with respect to sociodemographic factors as well as markers of severity and extent of OA. Because serum hsCRP values were not normally distributed, log transformed (geometric) means were then calculated for various markers of severity and extent of OA. Finally, bivariate (table 2) and multivariable (table 3) linear regression models with the log transformed serum hsCRP values as dependent variable were used to assess the independent impact of severity and extent of OA and other factors on the geometric mean of serum hsCRP concentrations. Regression coefficients from the multivariable model were exponentiated to allow presentation of the associations of the independent variables as a percentage change in serum hsCRP values.

Radiographic extent of OA (bilateral or generalised), intensity of pain in the operated joint, additional painful joint regions, and the WOMAC subscales (pain, stiffness, function) were considered as predictor variables in patients with OA.

The following variables were considered as possible confounders in multivariable analyses: age (continuous), sex, body mass index (weight in kg divided by squared height in m, continuous), history of coronary artery disease, history of congestive heart failure, serum cholesterol level, use of antihyperlipidaemic drugs, hypertension (systolic blood pressure ⩾160 mm Hg and/or diastolic blood pressure ⩾95 mm Hg and/or use of a antihypertensive drug), diabetes (history or use of oral antidiabetics or insulin), gout (history or use of uricosuric or uricostatic drug), smoking (current and former), alcohol consumption (three categories), self reported current or former heavy physical exertion at work or at home, and use of non-steroidal anti-inflammatory drugs (NSAIDs), diuretics, angiotensin converting enzyme inhibitors, and oestrogen replacement therapy, as well as an impaired renal function (defined as an estimated serum creatinine clearance below 60 ml/min¹⁸).

All covariates associated with serum hsCRP in the bivariate analyses were included in the final multivariable model.

RESULTS

Frozen serum samples were available for 771 patients, and hsCRP could be measured in 770 patients (95.2% of all 809 patients). The mean age of these patients was 63 years, and 63% were women (table 1). More than half of the patients were hypertensive and 19% reported a history of congestive heart failure. Regular current use of NSAIDs was very common, followed by use of diuretics by almost a quarter of the patients. More than 40% were current or former smokers. Most patients consumed <10 grams of pure alcohol a day.

About half of the patients had OA of the hip or the knee, each, and mean serum hsCRP levels were almost identical in these two patient groups (table 2). Most patients had bilateral OA, associated with slightly (not significantly) higher serum hsCRP levels and more than a quarter had generalised OA, again associated with slightly (not significantly) increased hsCRP levels. Radiographic severity of OA as assessed by the K&L score was not associated with hsCRP levels in these patients with advanced OA.

The median severity of pain in the affected joint was 73 out of 100 on the VAS. A clear association between intensity of pain and hsCRP levels was seen with a monotonic increase in mean serum hsCRP with increasing pain severity. Patients reporting pain in a joint region in addition to the operated joint also had higher hsCRP levels. The median severity of pain during different tasks as assessed by the WOMAC was 12 out of 20. Neither pain, nor any of the five WOMAC pain items, nor any of the other dimensions of the WOMAC (data not presented) were associated with hsCRP levels.

Table 3 presents independent determinants of serum hsCRP levels, together with the mean percentage change in serum hsCRP levels associated with the corresponding trait as estimated by multivariable linear regression. Only characteristics with a significant association with serum hsCRP in bivariate analyses were included in this multivariable model controlling for all variables presented in the table. Age (9.3% higher mean hsCRP for every 10 year increase), body mass index (22.3% higher mean for every five units increase), female sex (25.9% higher mean), and current regular use of diuretics (34.0% higher mean) were all independently associated with higher mean serum hsCRP levels. The most important independent predictor of serum hsCRP levels in patients with OA was smoking, especially current smoking, but former smokers also had significantly increased levels of serum hsCRP. Because a monotonic increase in serum hsCRP was observed with increasing intensity of pain (see table 2), the intensity of pain was included as a continuous variable. Every increase of 10 mm on the VAS was associated with a 5.7% increase in mean serum hsCRP (95% CI 1.1 to 10.4). A separate analysis with intensity of pain categorised as quintiles again showed a monotonic increase in mean hsCRP values for increasing categories of pain. One or more painful joint region in addition to the region of the joint being replaced was associated with a 17.0% higher mean hsCRP (95% CI −1.8 to 39.5).

To further examine the association between pain, treatment of pain, and serum hsCRP values, we performed additional analyses stratifying the association between pain severity and hsCRP levels according to NSAID use. The patterns were essentially identical in patients without (adjusted elevation highest v lowest quintile = 34%, p = 0.07) and with current regular use of NSAIDs (adjusted elevation highest v lowest quintile = 35%, p = 0.1).

DISCUSSION

Of all markers of severity and extent of OA assessed, intensity of pain was most strongly associated with low level rises in the mean serum hsCRP levels in this group of patients with advanced OA, whereas neither radiographic severity or extent, nor any dimension of the WOMAC were associated with serum hsCRP levels.

There is growing evidence that systemic markers of inflammation are associated with severity or clinical course of OA. Serum CRP, probably the most widely used clinical marker of systemic inflammation, has been shown to correlate well with CRP in synovial fluid in patients with OA or RA.¹⁹ Spector et al observed increased CRP values in women with radiographically defined knee OA in a population based, cross sectional study of 845 women in Chingford, as well as higher CRP values in women with radiographic OA progression during a four year follow up.⁴ The median hsCRP value found in our study (2.5 mg/l) is almost identical to the 2.4 mg/l observed by Spector et al.

Wolfe observed that disability (as assessed by the Health Assessment Questionnaire), number of tender joints, and pain were associated with higher mean CRP levels in 655 men and women with clinically defined OA of the hip or knee.² The result for pain is in agreement with the findings from our study. In contrast, we found no association between disability as assessed by the WOMAC nor radiographically defined generalised OA and mean hsCRP levels. Interestingly enough, the median CRP levels in the study by Wolfe were more than twice as high (5.9 mg/l) as in the study of Spector et al or in our study. This might be an indication that, despite the exclusion of patients with an erythrocyte sedimentation rate >30 mm/1st h, their clinically defined group of patients may still have included some patients with inflammation of unknown cause. Conrozier et al⁵ and Sharif et al³ found that low level increases in CRP were associated with progression of hip OA and knee OA in small samples, respectively, but their results were either unadjusted⁵ or inconsistent.³

Although OA pain and radiographic findings correlate to some extent, many patients with advanced radiographic OA have no pain and a fair proportion of patients with pain show no radiographic signs of OA.²⁰ Therefore, mechanical factors alone are insufficient to explain pain in OA. Because the sensation of pain in patients with OA is at least closely linked to, or may even be the result of, inflammatory processes,^21,22 the association between pain and hsCRP seems more plausible than an association between radiographic severity and hsCRP. Higher levels of both interleukin 1β and tumour necrosis factor α have also been detected in sera from patients with other painful conditions, like cervicogenic headache.²³ Tissue injury is known to result in the release of a variety of inflammatory mediators, including ions, bradykinin, histamine, 5-hydroxytryptamine, ATP, and nitric oxide.²⁴ The activation of the arachidonic pathway may then lead to the production of prostanoids and leukotrienes, and recruited immune cells may release further mediators, including cytokines and growth factors.²³

It is intriguing that we observed no association between any of the dimensions of the WOMAC and hsCRP levels, not even for the dimension of pain or any of the five pain items. In comparison with studies from Israel²⁵ and America,²⁶ which reported correlations between the WOMAC pain scale and pain assessed by a VAS of between 0.6 and 0.7, the association was weaker in our study (0.4), but still well within the range (0.3–0.4) seen in the American study for all other correlations of pain measures used, including the WOMAC. It seems as if the individual integration of overall pain, like the one assessed by the VAS, might be a better marker of the underlying systemic process than the assessment of pain associated with different tasks like walking or climbing stairs. Furthermore, it has been reported that the WOMAC captures more than just knee or hip pain and is influenced by the presence of fatigue, depression, and low back pain.²⁷ Because the WOMAC is sensitive to treatment of pain with analgesics or NSAIDs,²⁸ it might be speculated that the treatment of pain does not influence the association between pain as assessed by the VAS and serum hsCRP levels. In our study we found no association between NSAID use and hsCRP, and the association between VAS and hsCRP was in fact independent of NSAID use in an additional multivariable analysis.

Our study did not include healthy controls. We were able, however, to compare the geometric mean serum hsCRP concentration in our patients with OA (2.5 mg/l) with that of 567 age and sex matched population controls (1.7 mg/l, adjusted difference 36%, p<0.0001) who were recruited for a national health and nutrition survey conducted among healthy people in the western part of Germany in 1987–88.⁷ Because some of the people in the national sample are likely to have had OA (no specific information or radiographs were available), this observed difference in mean hsCRP levels between patients with OA and population controls is more likely to be an underestimation than an overestimation of the true difference between patients with and without OA.

Despite the observation of diurnal variation of CRP in patients with RA,²⁹ high sensitivity CRP has been shown not to vary during the day in healthy people³⁰ and to have a variability and classification accuracy similar to that of cholesterol screening.³¹ In combination with the use of a commercial kit calibrated with the WHO reference standard 85/506, we therefore think that the hsCRP levels used in our study are a consistent measure of the underlying low grade systemic inflammatory process.

The limitations of our study need to be taken into account when interpreting our results. The cross sectional study design does not allow assessment of the prognostic value of hsCRP or the association between pain and hsCRP when using repeated measurements within a person. Although widely used to assess the intensity of pain in clinical and epidemiological settings, the measurement of pain using a VAS is influenced by a variety of factors associated with perception of pain.^32,33 Nevertheless, pain is always subjective and the observed associations provide evidence for this subjective phenomenon to be better associated with markers of low level systemic inflammation than any of the objective markers of OA assessed in our study. Because of our study design with inclusion restricted to patients with advanced OA, we could not assess the association between radiographic severity of OA and hsCRP levels. In our study group of patients with advanced OA, however, we observed no difference in mean hsCRP levels between patients with K&L grades 3 and 4. Finally, the pathology behind anamnestic pain in additional joint regions remains elusive.

We conclude that OA and the subjective severity of OA pain are associated with low level systemic inflammation as assessed by hsCRP in patients with advanced OA. Longitudinal studies with repeated assessments of hsCRP and pain within the course of OA from diagnosis to joint replacement are needed to assess the possible value of low level increases of hsCRP for monitoring or predicting the clinical course of OA and, finally, to evaluate its possible role in determining the best timing of joint replacement.