Can an increase in autoantibody levels predict arthritis in arthralgia patients?

Rheumatology key message Repeating autoantibody levels in seropositive arthralgia patients adds little value in the prediction of future arthritis. Sir, retrospective studies have shown that ACPA and IgM RF often precede symptom onset of RA by many years and the closer the individual comes to clinical arthritis the more positivity for either one or the combination is found, together with increasing levels [1, 2]. To our knowledge, no individual prediction models using autoantibody levels over time have as yet been developed. In a prospective study we investigated whether a change in ACPA and RF levels over time within individuals improves prediction of future arthritis in seropositive arthralgia patients. A total of 983 antibody measurements were obtained from 263 consecutive seropositive arthralgia patients from an earlier described cohort [3], with a median of 4 per person (range 2–8). The measurements were a median of 12 months apart (range 3–60). Follow-up was at least 6 months and autoantibodies were measured at least twice and with the same test. Cohort visits were performed semi-annually the first year and yearly thereafter for 5 years or until arthritis occurred. Among those who developed clinical arthritis [N = 69 (26%)], 50% did so within 22 months and 87% of them fulfilled the 2010 ACR/EULAR classification criteria for RA (see supplementary Table S1, available at Rheumatology online). Written informed consent was obtained and the study was approved by the medical ethics committee of the Slotervaart Hospital, Amsterdam. On the individual level, the patterns of autoantibody levels over time differed widely, regardless of the outcome. For instance, 43 patients changed status for ACPA and/or RF positivity during the study period. We performed multivariable Cox proportional hazards modelling including baseline ACPA and RF and relevant clinical covariates to analyse their association with arthritis development (Table 1). The hazard ratio (HR) for baseline ACPA in this model was 2.6 (95% CI 1.9, 3.6; P < 0.001). The cumulative/dynamic area under the curve (AUC) was 0.809 within the first 3 years [4]. This prediction horizon was chosen to be able to compare it with the AUCs of three joint models [using time-updated ACPA and RF values in the first 2 years as predictors for the outcome at 5 years computed with R package JM (R Project for Statistical Computing, Vienna, Austria)] (Table 1) [5]. Joint models combine a linear mixed effects model of the development of ACPA and RF over time with a Cox proportional hazards survival model for the event hazard. The individual estimated slopes of ACPA and RF were included in the models to investigate whether they gave additional information over the values at different time points. The first univariate model included either time-updated ACPA or RF as predictor variables and controlled for their own baseline value. The HR for ACPA alone in the joint model was 2.3 (95% CI 1.8, 3.1; P < 0.001; AUC 0.739). This should be interpreted as follows: time-updated ACPA levels (adjusted for ACPA at baseline) were associated with a 2.3-fold higher rate of developing arthritis over time. The second bivariate model included time-updated ACPA or RF while controlling for the other autoantibody at baseline and the third full model included ACPA or RF while controlling for the other autoantibody at baseline and relevant clinical variables. The full joint model showed an HR of 2.8 (95% CI 2.1, 3.9; P < 0.001; AUC 0.836) for ACPA. In the models the ACPA and RF slopes did not show a significant association with the outcome arthritis and RF was not significantly related to arthritis development. Comparison of the AUC of the Cox model and full joint model was used to evaluate the added value (+0.027) of using repeated measurements of antibodies for individual prediction of the outcome. Table 1 Cox proportional hazard and joint models of ACPA and RF levels The multivariable model also contains clinical parameters (not shown here, including baseline alcohol use, smoking, age, gender, positive family history for RA, visual analogue scale pain, morning stiffness ≥1 h, duration of joint pain before inclusion, intermittent joint pain yes or no, tender joint count of 53 joints). ACPA and RF are log10 transformed in all models (including the multivariable Cox model). Table 1 Cox proportional hazard and joint models of ACPA and RF levels The multivariable model also contains clinical parameters (not shown here, including baseline alcohol use, smoking, age, gender, positive family history for RA, visual analogue scale pain, morning stiffness ≥1 h, duration of joint pain before inclusion, intermittent joint pain yes or no, tender joint count of 53 joints). ACPA and RF are log10 transformed in all models (including the multivariable Cox model). These data appear to contradict the findings of studies that showed an increase in autoantibody levels in the final years, as well as an ACPA repertoire maturation, before the diagnosis of RA [1, 2, 6, 7]. The present results are more in line with results of studies of RA patients shortly after the diagnosis, in which relatively stable levels of autoantibodies were observed on the group level [8]. This may be caused by the fact that the population of the present study was seen at a late stage of disease development in which the time window for detection of a change in autoantibody levels was too small or the levels had already reached the plateau typical of early RA [6]. The observation of increasing autoantibody levels 1–3 years before clinical arthritis in blood donor cohorts of preclinical RA suggests that this phenomenon might be useful for the prediction of future arthritis in autoantibody-positive individuals [1, 2]. However, we have shown that although higher time-updated ACPA values are significantly associated with the development of arthritis, they had only limited additional predictive value over baseline measurement of ACPA levels in this preselected cohort of seropositive arthralgia patients (difference in AUC of +0.027 and no additional predictive value of the ACPA slopes in the joint model). Neither RF levels at baseline nor over time were significantly associated with arthritis development in the Cox and joint models. All authors contributed substantially to the conception and design of the study and interpretation of the data. M.v.B.T., L.v.d.S. and D.v.S. made substantial contributions to the study conception and design. M.v.B.T., L.v.d.S., M.d.K. and D.v.S. contributed substantial to acquisition of the data. M.v.B.T., M.S., R.G. and D.v.S. analysed and interpreted the data. M.v.B.T., M.S., R.G. and D.v.S. prepared the first draft of the manuscript. All authors contributed to revision and approval of the final version. Funding: This work was financially supported by the Dutch Arthritis Association (grant number 0801034). Disclosure statement: No conflicts of interest present. Supplementary data are available at Rheumatology online.