Brief description of the matrix risk model
In the model of Vastesaeger et al. [3], RRP was defined as a threshold change in modified Sharp/van der Heijde score (SHS) of > or = 5 U/year. The developed and validated matrix risk model enables to determined RRP without actual radiographs based on three simple variables. In this model, the 28 swollen joint count (SJC), RF and CRP levels were used as trichotomous variables. These three variables weighed equally.
Sample size calculation and patient selection process
We applied the precision-based sample size calculation with the following formula to calculate the minimum sample size: π (1-π)/e2, where π is the expected proportion, e is the required size of standard error. Precision is defined as the ± range around the estimated proportion, and was calculated as ±1.96 × e [19]. Calculating with π = 0.2 as the expected proportion and ± 0.02 as the required precision, the minimum required sample size was 1537. Assuming as the estimated proportion of patients with incomplete data will be 15%, the adjusted minimum sample size calculated was 1537/0.85, which equals 1808 patients. This was the number of patients required for the estimation with the specified precision of ±0.02 (data not shown).
A multi-stage sampling method was applied to ensure equal probability of selection of target patients. In order to obtain a population-based sample, each of the 20 regional rheumatology centers evenly distributed throughout Hungary (1st-stage sampling units) were invited to participate in the study. The number of cases per center (n) was allocated according to the number of treated RA patients in each center at the time of the start of the study. Investigators then collected data on patients up to the allocated sample size (n), thus ensuring sampling probability proportional to size of the 1st-stage sampling units (i.e., the 20 rheumatology centers).
Random selection of patients (2nd-stage sampling units) was then performed to avoid any selection bias. The list of all currently treated RA patients (sampling frame) was created in each center. The allocated number of patients was selected from the sampling frame with simple random sampling (with tables of random numbers) in order to each patient having the same chance of being chosen [20].
Based on these calculations, 1843 patients were consecutively chosen for data analysis (Fig. 1). The only inclusion criterion was the diagnosis of RA. There were no exclusion criteria except for age ≤ 16 years (definition of juvenile arthritis).
Data capture process
Patient data from the last visit that occurred right before the date of patient selection were collected. For those already on biologic therapy, the last data before initiation of biologics were retrospectively recorded. Thus, only data obtained from biologic-naïve patients were evaluated. We used clinical data obtained from hospital records and also assess radiographs at baseline for the presence or absence of erosions.
Standardized electronic spreadsheed (Microsoft Excel) was used to capture the data. The following data were collected based on hospital records:
-
age
-
gender
-
duration of RA
-
history of DMARD use (MTX and others; current/past/never use)
-
MTX response (responder/non-responder)
-
SJC, CRP and RF levels (for calculation of the matrix-based RRP risk)
-
anti-citrullinated protein antibody (ACPA) status (positive/negative)
-
DAS28 activity score
-
presence of baseline erosions on radiographs at baseline (yes/no)
-
cigarette smoking (current/past/never)
For determining RRP, we used the traditional three variables, SJC, RF and RF, as determined by Vastesaeger et al. [3]. However, we added a few binary variables described above in order to look for further denominators.
Objectives
The primary objective was to estimate the prevalence of high (≥40%) risk of RRP in a community-based sample of RA patients, naïve to biologic treatment presenting at rheumatology departments. Active disease was defined as DAS28 ≥ 5.1, which is the threshold for the use of biologics in Hungary.
The secondary objectives were:
-
to assess the difference in the prevalence of high RRP risk in RA patients classified MTX-non-responders versus MTX-responders (MTX non-responders are patients with DAS28 > 5.1 despite MTX treatment for at least 6 months in stable doses);
-
to assess the multivariate association of patient characteristics with high RRP risk (independent of the parameters used in matrix model).
Statistical analysis
The calculation of sample size is described above. MS Excel was used to record, summarize and clean the data. Statistical analyses were performed with IBM SPSS 20 program. Continuous variables were described by mean and standard deviation, the distribution was described with number of cases and percentage. Distribution was analyzed with Kolmogorov-Smirnov test. Between group difference was analyzed with Mann-Whitney test and Chi2-test. Independent predictive factors were identified applying univariable and multivariable regression analyses. We considered correlations to be significant in case of a p-value less than 0.05.