Subjects with IP and non-IP controls were recruited from the same population source: the Norfolk cohort of the European Prospective Investigation of Cancer (EPIC-Norfolk). Cases were individuals with inflammatory polyarthritis (IP) notified to the Norfolk Arthritis Register (NOAR) and who had also subsequently participated in EPIC. Two controls (age and sex matched) were selected for each case, being subjects who had been recruited to EPIC-Norfolk and who were not registered with NOAR. To determine the influence of IP on SOS and BUA, we compared these measurements between the cases and matched controls. To determine the influence of disease activity variables on these parameters we restricted our analysis to just the subjects with IP. Both EPIC-Norfolk and NOAR were in compliance with the Helsinki Declaration and approved by the ethics committee of the Norwich District Health Authority.

NOAR is a primary care based inception cohort of adults aged 16 years and over, registered with a local GP, with early IP in the former Norfolk Health Authority (UK). Subjects are included in the register if they had 2 or more swollen joints for a period of at least 4 weeks, with onset since January 1, 1989. Detailed methods for the study have been published previously 32. The current analysis includes subjects with an onset of symptoms before 2000. At baseline, subjects completed an interviewer administered questionnaire and Health Assessment Questionnaire 33. They were examined by a trained nurse for the presence of tender and swollen joints (51 joints), and had blood taken for rheumatoid factor measurement. As explained elsewhere, subjects had x-rays of the hands and feet performed if they satisfied the ACR criteria for RA or if the presence of erosions would make them satisfy these criteria 34. The ACR criteria were applied at the baseline visit.

Subjects were recruited in Norfolk, UK for participation in the European Prospective Investigation of Cancer – a large multicentre population based study with the primary aim of establishing the relationship between diet and the risk of developing cancer. In Norfolk, approximately 25,000 men and women aged 40–79 years were recruited between 1993 and 1997. Subjects completed a health questionnaire which included information about steroid therapy for 3 or more months. Height and weight were measured in light clothing without shoes. Body mass index (BMI) was calculated using measurements made at the baseline EPIC visit by dividing weight (Kg) by height squared (m²). Detailed methods have been published previously 35.

In 1998–2000 men and women from EPIC, now aged 42 to 82 years, were invited to attend a second visit where measurements of QUS of the heel were conducted by trained nurses working to standard protocols. Both speed of sound (SOS) and broadband ultrasound attenuation (BUA) were recorded for each subject. Measurement was made using a CUBA sonometer (McCue Ultrasonics, Winchester) at least twice on each foot. The mean of the measures (left and right foot) were used for analysis. The coefficient of variation was 3.5% 16. Five CUBA machines were used and each was calibrated daily with its physical phantom. A roving physical phantom was used monthly to check calibration between machines and room temperature measured daily.

For each individual notified to NOAR who had had heel ultrasound measurements performed, two age and sex matched controls were selected from EPIC-Norfolk. The analysis was divided into two stages. In the first, fixed effect linear regression was used to look at the association between heel ultrasound measurement and case status, with the bone ultrasound parameter (SOS or BUA) as the outcome and case status as the independent variable. As our aim was to examine the influence of IP on bone health, this approach allowed us to quantify the association in terms of the mean absolute difference (and 95% confidence interval) in the bone ultrasound parameters between the cases and controls which we felt was easy to interpret. Adjustments were subsequently made for body mass index and also steroid therapy (as assessed in EPIC-Norfolk). In the second stage we used simple linear regression to examine the influence of the following disease related factors assessed at baseline on the ultrasound parameters : swollen and tender joint count, HAQ, rheumatoid factor, and whether the subject satisfied the 1987 ACR criteria at baseline (both the traditional four of seven criteria and the decision tree approach 34 and the disease duration as determined from the time of symptom onset until the ultrasound scan, after adjustment for age and sex 34. For the purposes of analysis, the joint counts were categorised into tertiles, the lowest tertile representing those with the least number of involved joints. The results were expressed as β coefficients and 95% confidence intervals (CI). Having determined the disease related predictors of the heel ultrasound parameters, we then used fixed effect linear regression to determine whether or not these explained differences between those with IP and the non IP controls. Statistical analysis was performed using STATA (Stata Corporation. Stata Statistical Software: Release 9.2, College Station, TX, USA, 2008).

In all, 139 subjects (52 men and 87 women) with IP and 278 age and sex matched controls were studied. The characteristics of these subjects are presented in Table 1. Women with IP had a slightly greater BMI (26.9 kg/m² vs 25.6 kg/m²) and (as expected) a higher prevalence of steroid use (17.2% vs 1.7%) than their matched controls. In men, there were no differences between those with and without IP with respect to body mass index or steroid use.

IP = inflammatory polyarthritis.

Disease related characteristics at baseline are presented in Table 2. The mean duration of disease from symptom onset to time of ultrasound was 4.8 years (standard deviation [SD] = 2.9) in men and 5.6 years (SD = 2.8) in women (mean duration of symptoms prior to baseline assessment in NOAR was 0.9 years (SD = 1.1) in men and 1.3 years (SD = 1.9) in women). In terms of joint involvement, men had a median of 1.0 (interquartile range [IQR] = 0.0,3.5) and women a median of 2.0 (IQR = 0.0, 5.0) joints that were both swollen and tender. The median HAQ score was 0.5 (IQR = 0.1, 1.1) in men and 0.8 (IQR = 0.4, 1.3) in women. 39% of men and 34% of women were rheumatoid factor positive and 42% and 60% respectively satisfied the ACR criteria for rheumatoid arthritis using the tree definition. Given the design (see methods) only 27 subjects had hand radiographs at baseline. Of these 75% of men and 56.5% of women had evidence of erosive disease.

^*Years between first symptom onset and ultrasound scan.

¹Assessed in those with baseline radiographs (n = 27).

Overall, mean SOS was 1621.8 m/s (SD = 43.4) and BUA was 76.3 dB/MHz (SD = 18.4) in those with IP (Table 3). Mean SOS was lower in those with IP than the controls for both men (1629.0 vs 1644.0 m/s) and women (1617.6 vs 1624.5 m/s) though the difference was statistically significant for men only. There was no difference in BUA between those with IP and their matched controls in men or women. There was no evidence of interaction between sex and case status, BMI or steroid use and in subsequent analysis the data for men and women were pooled. Using fixed effect linear regression, with the bone ultrasound parameter as the dependent variable and case status as the independent variable, SOS was lower among those with IP than the matched controls (difference −10.0; 95%CI −17.4, -2.6), a difference of 0.6%, see Table 4. This difference persisted after adjusting for body mass index and steroid use (difference −9.6; 95%CI −17.9, -1.3). BUA, however, was similar between those with IP and controls (difference −1.2; 95%CI −4.5, 2.1). When the analysis was restricted to those who fulfilled the ACR tree criteria at baseline the differences between those with IP and the controls became more marked for both SOS (difference = −17.3; 95% CI −28.8, -5.7) and for BUA, which now became statistically significant (difference = −7.2;95% CI −12.4, -2.1). Higher BMI at baseline was associated with a higher BUA (β coefficient 1.1; 95%CI 0.6, 1.6) independent of case status. BMI and steroid use were not independently associated with SOS, see Table 4.

IP = inflammatory polyarthritis.

^aunadjusted model.

^bmodel includes case/control status, BMI at baseline and steroid use.

*p < 0.05.

The influence of disease related factors on ultrasound parameters is presented in Table 5. After adjusting for age and sex, compared to those in the lowest tertile of swollen and tender joint counts, those in the mid and upper tertile had lower BUA though this was significant for the middle tertile only (difference = −6.4; 95%CI −12.7, -0.2). Compared to those who did not satisfy the ACR criteria (either list or tree definition) those that did satisfy the criteria had a lower BUA (difference = −8.5; 95%CI −13.7, -3.2 using the tree definition). With respect to SOS, compared to those in the lowest tertile of swollen joint count, those in the mid and upper tertiles had a lower SOS which was significant for the upper tertile (difference = −20.1; 95%CI −35.6, -4.6). There was evidence also of a significant linear trend for an increase in both the swollen joint count and swollen and tender joint count and lower SOS. Compared to those who did not satisfy the ACR criteria (using either list or tree definition), those that did also had a reduced SOS (difference = −18.7; 95%CI −32.1, -5.2 using the tree definition). Larsen score, presence of erosions (in those who had X-rays performed at baseline), disease duration (time between symptom onset and the ultrasound scan), rheumatoid factor and HAQ were not associated with the ultrasound parameters.

^aadjusted for age and sex; ^bp < 0.05 test for trend; *p < 0.05.

¹Years between first symptom onset and ultrasound scan.

²Assessed in those with baseline radiographs (n = 27).

We looked next at whether the swollen joint count as a marker of inflammation could explain the difference between cases and controls, assuming that the controls had no inflamed joints. After adjusting for BMI, steroid use and number of swollen joints the difference in SOS between those with and without IP became much weaker and non significant (difference = −3.9; 95%CI −14.6, 6.7).