Translation

The translation and cross‑cultural adaptation process began after obtaining permission from the Mapi Research Trust, and was performed according to the Mapi’s linguistic validation manual and good clinical practice²⁵ in 3 steps including forward translation, backward translation, and patient testing. The English version of the FiRST was translated into Polish by 2 specialists who are native Polish speakers fluent in English, a rheumatologist and a clinical psychologist. After reaching a consensus within the research team, the Polish version was backtranslated by a bilingual (Polish and English) translator who was independent of the research team and did not have access to the original questionnaire. The backtranslated version was compared with the source instrument, showing no conceptual discrepancies between both versions. Pilot testing (cognitive interviews) of the Polish version with 10 axSpA patients (with and without FM) confirmed clarity, and no further adaptations of the translation were required. The Polish version of the FiRST was proofread and the results were reported to the Mapi Research Trust. The final Polish version of FiRST (Supplementary material, Questionnaire) was then administered to all recruited patients in a paper form. While there is no consensus on the required sample size to validate the psychometric properties of questionnaires,^26,27 Terwee et al²⁸ suggest a sample size of 7 times the number of items in the questionnaire and a minimum of 100 subjects. In our study, we therefore established a minimum sample size of 100 patients.

Validation study

The FiRST reliability was assessed using 2 methods: internal consistency and absolute stability. Internal consistency of the Polish version of the FiRST was analyzed by calculating the Cronbach α coefficient. One week after the first administration, test‑retest reliability of the questionnaire was assessed in a randomly selected group of patients via the online questionnaire. This stage followed the consensus‑based standards for the selection of health measurement instrument recommendations for test‑retest assessment.³¹ Reliability of each item was assessed by calculating the Cohen κ coefficient, and the global score was assessed by calculating the Spearman correlation coefficient.

The factor structure of the FiRST was examined using the exploratory factor analysis, that is, principal component analysis with the Varimax (orthogonal) rotation and maximum likelihood procedure. The sampling adequacy was detected with the Kaiser–Meyer–Olkin test and the Bartlet test of sphericity. Two criteria for the factor detection were used: eigenvalue (>1) and scree plot.

Several methods of validity estimation were employed. For the criterion validity, we used the 2016 ACR FM diagnostic criteria as the gold standard for diagnosing FM. Receiver operator characteristic (ROC) analysis was performed. The AUC and the Youden index (to identify the most relevant cutoff value) were calculated. The sensitivity, specificity, and accuracy (diagnostic agreement percent) values of the FiRST were calculated for each of the 6 items, and for the global score using the obtained cutoff value. Additionally, positive predictive value (PPV) and negative predictive value (NPV) were calculated from FM prevalence in our study and obtained specificity and sensitivity of the global score. The diagnostic agreement between the diagnoses based on the 2016 ACR criteria and the FiRST global score (using the obtained cutoff point) were assessed by calculating the Cohen κ coefficient.

We tested the construct validity by assessing various clinical outcomes, including FM indicators (widespread pain index [WPI], symptom severity scale [SSS]), pain (numeric rating scale [NRS]), anxiety (Generalized Anxiety Disorder 7]), depression (Patient Health Questionnaire‑9 [PHQ9]), fatigue (Multidimensional Fatigue Scale [MFI Total]), sleep disorders (Jenkins Sleep Scale [JSS]), axSpA disease activity (BASDAI, Ankylosing Spondylitis Disease Activity Score [ASDAS]), and disability [BASFI]). The correlation coefficients between each clinical outcome and the FiRST were calculated, and the differences between patients with and without a FM diagnosis based on the 2016 ACR FM diagnostic criteria and the FiRST test were measured.

Statistical analysis

Compliance of the data with normal distribution was assessed using the Shapiro–Wilk test. Variables with normal distribution were presented as mean (SD), those without normal distribution as median and min‑max range, and qualitative variables as number and percentage. In a subgroup analysis, when a variable did not follow the normal distribution in any of the examined subgroups, it was presented as median (min‑max range). The correlation was determined using the Spearman rank correlation coefficient for non‑normally distributed variables. Significance of the observed differences between the groups was assessed using the t test for the variables with normal distribution, the Mann–Whitney test for the variables with non‑normal distribution, and the χ² test for categorical variables (or the Fisher exact test for expected frequencies below 5). Statistical analysis was performed using Statistica 13.3 software (StatSoft Polska, Kraków, Poland). A P value below 0.05 was deemed significant.

Patient characteristics

The study involved 174 patients with axSpA (95 with AS, 54 with nrSpA, and 25 with PsA). Among all patients, 65 (37.4%) were diagnosed with FM according to the 2016 ACR diagnostic criteria. Patient characteristics are presented in Table 1. There were significant differences between the patients with and without FM. The first group included more women (64.6% vs 45%; P = 0.01), the patients were older (median, 44 vs 39 years [min‑max, 18–73 vs 19–74]; P = 0.02), less frequent HLA‑B27–positive (63.1% vs 81.4%; P = 0.008), and had higher BASDAI (median, 6.5 vs 4.2 [min‑max, 2.7–10 vs 0–9.6]; P <⁠0.001), ASDAS (mean [SD], 3.37 vs 2.62 [1.1 vs 1.3]; P <⁠0.001), BASFI (median, 6.7 vs 2.9 [min‑max, 0.5–9.8 vs 0–8.9]; P <⁠0.001), MFI total (mean [SD], 73.4 vs 55.6 [12.4 vs 16.4]; P <⁠0.001), JSS (median, 10 vs 4.5 [min‑max, 0–20 vs 0–20]; P <⁠0.001), and PHQ9 (median, 10 vs 6.5 [min‑max, 0–23 vs 0–16]; P <⁠0.001) scores. Importantly, the groups did not differ in the treatment they received (mostly nonsteroid anti‑inflammatory drugs, with only a minority on biologic / targeted synthetic disease‑modifying antirheumatic drugs [b/tsDMARDs]).

Psychometric properties

To verify the factor structure of the FiRST, we performed the exploratory factor analysis. The basic assumptions of the factor analysis were met. Sampling adequacy measured by the Kaiser–Meyer–Olkin test was 0.667 and was satisfactory (recommended above 0.5). The value of the determinant of the correlation matrix (showing the strength of correlations between items) was 0.289 (recommended close to 0). Both analyses of the structure based on eigenvalues and scree plots showed that 2 factors could be extracted. Factor analysis with the Varimax rotation demonstrated that 2 factors explained 56.4% of the total variance. The first one, explaining 33.7% of the total variance, was created by items 1, 2, and 6 of the FiRST questionnaire (α coefficient equal to 0.69). The second, explaining 22.7% of the total variance, was created by items 3 and 4 of the FiRST questionnaire with α coefficient equal to 0.36 (Table 2). Item 5 of the questionnaire was not included in any of the isolated factors. The results suggest a more complex structure of the FiRST questionnaire, but due to a lack of sufficient clarity and poor internal consistency of the subscales, we decided to treat it as a unidimensional tool in further analysis.

The Polish version of the FiRST showed a borderline internal consistency with the Cronbach α coefficient of 0.644. As shown in Supplementary material, Table S1, removal of any item does not improve the α coefficient. The test‑retest reliability was assessed in 80 randomly selected axSpA patients after 1 week. The test‑retest reliability for each item varied with the Cohen κ coefficient between 0.48 to 0.76 (Supplementary material, Table S2), while the global score showed a good Spearman correlation coefficient of 0.75 (P <⁠0.001).

The ROC analysis showed the same global score cutoff point (≥5 points) for the Polish version of the FiRST as in the original instrument, with good performance of the translated questionnaire (AUC = 0.803; Figure 1). The accuracy of the obtained cutoff value of the global score was 75.3%, with better specificity (82.6%) than sensitivity (63.1%); PPV was 68.3% and NPV was 78.9%. The sensitivity, specificity, and accuracy for each of the 6 items are shown in Supplementary material, Table S3. The diagnostic agreement between the 2016 ACR criteria and the FiRST global score showed a fair Cohen κ coefficient of 0.46. The patients with false‑negative diagnosis did not differ from true‑positive patients, except for lower PHQ9 scores (Supplementary material, Table S4). However, false‑positive patients had higher BASDAI, BASFI, MFI Total, JSS, and PHQ9 scores than the true‑negative individuals (Supplementary material, Table S5).

The FiRST global score showed good convergence with the WPI (ρ = 0.6; P <⁠0.001), SSS (ρ = 0.62; P <⁠0.001), and NRS (ρ = 0.49; P <⁠0.001), as presented in Supplementary material, Table S6. However, we could not assess discriminant validity, as the patients with and without FM significantly differed in the disease activity, functional limitations, fatigue, depression, sleep disorders, and anxiety (based on both the 2016 ACR diagnostic criteria as shown in Table 1 and the FiRST as shown in Supplementary material, Table S5).

Strengths and limitations

Our study provides the first validated translation of the FiRST into Polish. The strength of the study is its extensive validation, taking into account internal consistency, absolute stability, criterion, and construct validity. Given the comparable psychometric properties of the Polish language version with other language versions of the FiRST, in our opinion, the Polish version may tentatively be used in everyday clinical practice in patients with axSpA. Our study’s both a strength and a limitation is the fact that it focuses on a narrow group of patients with axSpA. The study is the first reliable validation of the FiRST among patients with axSpA. Its results, when confirmed in axSpA patients using other language versions of the FiRST, may significantly limit the use of the FiRST in research studies. Still, a specific group of axSpA patients necessitates a cautious approach to generalizing the use of the Polish translation of the FiRST in other patient groups. Our results require confirmation in subsequent studies among patients with chronic widespread pain, as well as other inflammatory joint diseases, which we plan to perform. Until then, the Polish version of the FiRST should not be used in patients other than those with axSpA. Another limitation is the fact that the study involved axSpA patients with high disease activity, most of whom were not treated with b/tsDMARDs (most patients were awaiting qualification for b/tsDMARD treatment in the therapeutic program of the National Health Fund). It needs to be verified whether the performance of the FiRST is comparable among patients with low disease activity and / or in remission.

Conclusions

The FiRST questionnaire is a quick, easy‑to‑use screening tool that can be used in daily clinical practice. Given the good NPV of the Polish version of the instrument, the FiRST in the axSpA group can be used to screen for the exclusion of FM during everyday clinical practice. However, if FM is suspected based on the FiRST, because of lower PPV and observed significant differences between false‑positive patients and true‑negative patients, full assessment should be performed using the 2016 ACR diagnostic criteria to confirm FM diagnosis (Figure 2). Additionally, in our opinion, the use of the FiRST should be avoided in research studies in patients with axSpA until further studies validating its use are available.