A total of 1,646 members of Dutch AS patient societies received information about the study by mail, and 332 responded (Figure 1). Patients were eligible for the study if they 1) fulfilled the modified New York criteria for AS (12), 2) reported pain and stiffness or functional limitations for at least 3 months before entry, and 3) were able to stay away from home and work for 3 preplanned consecutive weeks. Exclusion criteria were inability or unwillingness to participate in weekly group physical therapy; pregnancy; claustrophobia; severe co-morbidity of heart, lung, liver, or kidneys; and a diagnosis of AS more than 20 years ago. Radiographs of the sacroiliac joints were checked for sacroiliitis according to the New York criteria.

Of the 332 responding patients, 111 were not eligible, 83 declined to cooperate, and 138 patients signed informed consent and were placed consecutively on an inclusion list (Figure 1). One week before the intervention, the first 120 patients on this list were randomized with a computer-generated random-number list (prepared by a rheumatologist not further involved in the study) into 3 groups of 40 patients each: group 1 (spa therapy, Austria), group 2 (spa therapy, Netherlands), and a control group (home). The remaining 18 patients were placed on a waiting list. After randomization but before the start of the intervention, 3 patients from group 1 and 5 from group 2 withdrew for various reasons. They were randomly replaced by patients from the waiting list. The remaining 10 patients from the waiting list were excluded from the study.

Patients allocated to the control group were offered spa therapy at the end of the study in order to prevent withdrawal from study and for ethical reasons.

Permission to conduct the study was obtained from the medical ethical committee of the University Hospital of Maastricht.

The intervention took place at 2 spa resorts: Bad Hofgastein, Austria (group 1), and Thermaalbad Arcen, The Netherlands (group 2), during 3 consecutive weeks in April 1999. Patients from group 1 traveled together to the spa resort by coach (12 hours) and stayed in the same hotel. Spa–exercise therapy was provided at the spa resort and at the so-called Gasteiner Heilstollen. Patients from group 2 arrived at the spa resort either individually or by coach and stayed in the same hotel. Spa–exercise therapy was provided at the spa resorts.

Patients in the intervention groups received spa–exercise therapy 5 days a week. Therapy programs were standardized for both spa resorts and were performed by trained physiotherapists not involved in the outcome assessment and analyses of the study (Table 1). Every morning patients started with 1 hour of physical exercises, followed by 30 minutes of walking, and postural correction therapy by lying supine on a bed (initially 14 minutes, but increasing daily by 2 minutes to a final period of 30 minutes a day). Every other afternoon, the patients in Bad Hofgastein visited the Gasteiner Heilstollen. The Gasteiner Heilstollen are former mine galleries with a climate characterized by temperatures from 38.0 to 41.5°C, a humidity of 70% to 98%, and concentrations of radioactive radon in the air. Patients traveled by train to one of the treatment areas situated 2 kilometers inside the mountain, where they rested (naked) in the supine position on a bed for 1 hour. After the stay in the Heilstollen, the patients rested for 30 minutes at room temperature. Instead of visiting the Gasteiner Heilstollen, the patients in Arcen received a similar thermal treatment by visiting the sauna and the thermal baths. Either the Heilstollen or the sauna was visited a total of 10 times within a period of 3 weeks. In the Heilstollen, the radon progeny activity was measured in Working Level using the Instant Radon Progeny Meter (IRPM, Type TN–IR-21, Thomson & Nielsen, Canada). For the total 10-hour stay, patients were exposed to a cumulative dose of 0.536 Working Level Month (WLM). The other afternoons were spent with 30 minutes of intensive hydrotherapy and 30 minutes bathing in thermal water, followed by 1 hour of sports. Individual therapies were not allowed. During the weekends, patients were permitted to visit the thermal baths but were instructed not to exercise.

The control patients stayed at home and continued their usual activities and drug treatment and participated in weekly group physical therapy. Weekly group physical therapy consisted of 1 hour of physical exercises, 1 hour of sports, and 1 hour of hydrotherapy. After the intervention period, all patients from the 3 groups engaged in weekly group physical therapy sessions. During the intervention and the followup periods, all patients continued their usual drug treatment but were allowed to decrease or increase the amount of antiinflammatory drugs based on present complaints during the study period. Another spa treatment was not permitted during the followup period.

In the absence of validated response criteria for AS, primary outcome measures were chosen on agreement by rheumatologists involved in the study during the design phase. Based on the preliminary core-set for physical therapy in AS developed by the Assessment in Ankylosing Spondylitis Working Group, measures that were clinically relevant and considered sensitive to change in the opinion of the investigators were included (13). All outcome measures were elicited by self-assessment questionnaires.

Primary outcomes were functional ability (measured with the Bath Ankylosing Spondylitis Functional Index [BASFI]) (14), patient's global well-being (measured on a 10-cm visual analog scale [VAS]), pain intensity (on a 10-cm VAS), and morning stiffness (in minutes). The BASFI contains 10 questions concerning activities of daily living and is scored on a 10-cm VAS, with anchors “easy” and “impossible” at either side. The mean of the items defines the final score. The scores on the BASFI, patient's global well-being, and pain intensity range from 0 (best) to 10 (worst).

Secondary outcomes were disease activity (measured with Bath Ankylosing Spondylitis Disease Activity Index [BASDAI]) (15), general health and function (measured with Health Assessment Questionnaire for Spondylarthropathies [HAQ-S]) (16), night pain (on a 10-cm VAS), quality of life (measured with Ankylosing Spondylitis Quality of Life questionnaire [ASQoL]) (17), and the intake of nonsteroidal antiinflammatory drugs (NSAIDs). The BASDAI consists of 6 questions answered on a VAS (15). The questions are related to fatigue, back pain, pain and/or swelling of the peripheral joints, localized tenderness, and duration and severity of morning stiffness. The 10-cm horizontal VAS scale has the labels “none” (0) and “very severe” (10) at either end in the first 5 questions, and “0 hours” (0) and “2 or more hours” (10) for the duration of morning stiffness. The mean of the 2 questions on morning stiffness counts as one variable. The total score is calculated by taking the mean of the 5 items, ranging from 0 (best) to 10 (worst). The HAQ-S consists of 8 subscales on health status and function and is extended with 5 additional AS-specific items on function, divided into 2 subscales (16). The questions are answered on Likert-formatted scales. The mean of all subscales defines the final score, ranging from 0 (best) to 3 (worst). The ASQoL is a new quality-of-life questionnaire specific to AS and was developed in cooperation with patients with AS (17). The ASQoL contains 18 yes/no questions. Scores range from 0 to 18, with lower scores implying a better quality of life. The exact amount of NSAIDs taken in the previous week was registered and recalculated into standard dosages, equipotent to diclofenac.

All questionnaires, except for the newly developed ASQoL, are widely used instruments in the research of AS and have been shown to be valid, reliable, and sensitive to changes after active intervention (14–23).

Spa–exercise therapy of 3 weeks duration took place from T_0w (start of spa therapy) to T_3w (3 weeks after the start of spa therapy). The questionnaires were completed at baseline (2 weeks before spa therapy, T_−2w) and at 4 weeks (T_4w), 16 weeks (T_16w), 28 weeks (T_28w), and 40 weeks (T_40w) after the start of spa–exercise therapy.

Data from the validation study of the BASFI were used to calculate the sample size (14). In a group of patients who were following an inpatient course of intensive physiotherapy, the BASFI changed 1.1 ± 2.3 (mean ± SD) (20% improvement compared with baseline) on a 0 to 10 scale after 3 weeks, which we considered a minimum clinically important difference. Based on the assumption that combined spa–exercise therapy would be at least as effective as intensive physiotherapy, and that the mean BASFI would remain constant in the control group (mean change 0.0 ± 1.0), we calculated a sample size of 41 patients per group to be sufficient (2-sided α = 0.05; β = 0.20) to detect a difference of at least 20% between one intervention group and the control group.

The analyses were based on intention-to-treat. All returned questionnaires were checked for possible missing answers. If necessary, additional answers were obtained by telephone or by mail. In the few cases in which data were still missing, the instructions of the authors of the questionnaires were followed. If such information was lacking, the mean of the nonmissing items of the patient or the group (in single-item questions) was filled in. Investigators not involved in the intervention of the groups performed the analyses.

The results of the primary outcomes were expressed in a pooled index of change (PIC) (24, 25). For each component of the PIC, the change compared with baseline per time period was calculated for each patient. To obtain a standardized change score for each group, the mean change per time period of each group was divided by the pooled SD_change at T_40w of that instrument. The PIC was calculated as the unweighted mean of the 4 standardized scores per time period. Change in morning stiffness did not appear to be normally distributed, and the change score was therefore logarithmically transformed before introducing it into the PIC.

In order to evaluate whether there was any overall effect of spa–exercise therapy compared with control therapy over time, analysis of variance for repeated measurements (general linear model) was performed, with the PIC as dependent variable and group allocation and time as factors. Post-hoc analyses for between-group differences were planned by doing Student's t-tests for unpaired observations. Non-normally distributed data, determined by probability plots (morning stiffness, pain at night, ASQoL, NSAIDs), were analyzed by Mann-Whitney U test. SPSS 10.0 software (Chicago, IL) was used for all analyses.

Analysis of variance with adjustment for within correlation (repeated measurements) showed a statistically significant time–effect (P < 0.001) and time-by-treatment interaction (P = 0.004), indicating that the 3 groups differed over time with respect to the course of the PIC. Post-hoc analyses demonstrated a significant time-by-treatment interaction for group 1 versus controls (P = 0.005) and group 2 versus controls (P = 0.022) but not for group 1 versus group 2 (P = 0.132). Multivariate analysis of the PIC with sex as covariate did not change the time–effect (P < 0.001) and only slightly changed the time-by-treatment interaction (P = 0.011).

Figure 2 shows the course of the PIC and individual variables over time. At T_4w, the mean (95% confidence interval [CI]) PIC was 0.56 (0.33–0.79) for group 1, 0.53 (0.32–0.74) for group 2, and 0.07 (−0.16, 0.30) for the control group. The mean differences (95% CI) between group 1 and controls (0.49 [0.16–0.82]) and between group 2 and controls (0.46 [0.15–0.78]) were statistically significant (P = 0.004 and P = 0.005, respectively).

At T_16w, the PIC was 0.72 (0.46–0.98) for group 1, 0.43 (0.18–0.68) for group 2, and 0.09 (−0.20, 0.38) for the control group. The difference between group 1 and controls (0.63 [0.23–1.02]) was statistically significant (P = 0.002). The difference between group 2 and controls (0.34 [−0.05, 0.73]) had lost statistical significance (P = 0.086).

At T_28w and T_40w the PIC gradually declined toward baseline values in the intervention groups. At T_28w the difference between group 1 and controls was still statistically significant (0.48 [0.11–0.86], P = 0.012), but at T_40w the difference had just lost statistical significance (0.29 [−0.02, 0.60], P = 0.062).

Compared with group 2, more improvement was observed in group 1 at T_16w, T_28w, and T_40w, but the differences between groups 1 and 2 were not statistically significant.

The course of the individual variables of the PIC is summarized in Table 3. A trend toward improvement in both intervention groups after spa–exercise therapy, with a gradual decrease toward baseline values at T_40w, was seen for all variables except morning stiffness. Maximum relative improvements (scores of group 1 or 2 minus those of controls) during followup (morning stiffness excluded) were 24% (group 1) and 19% (group 2) for BASFI, 24% (group 1) and 30% (group 2) for pain, and 33% (group 1) and 29% (group 2) for global well-being. At T_40w, group 1 still showed relative improvements of 12% in BASFI, 8% in pain, and 16% in global well-being.

The results of the secondary outcome variables are summarized in Table 4. Most secondary outcome variables significantly improved in both intervention groups compared with controls after spa–exercise therapy, and a trend similar to that found in the primary outcomes was seen. On average, the maximum relative improvements achieved were 54% (group 1) and 35% (group 2) at T_4w and 43% (group 1) and 16% (group 2) at T_16w. At T_40w group 1 still showed a relative improvement of more than 10% in the BASDAI, HAQ-S, and pain at night, and up to 29% in the ASQoL.