Table 1 shows the demographic and clinical characteristics of our patients with AN (n = 69) and EDNOS (n = 6) [see Additional file 1-T1]. No significant differences between completers (n = 51) and dropouts (n = 24) were found in the type of eating disorder, age of onset, duration of illness, age, or BMI at admission. When subjects with EDNOS were excluded, patients with restricting AN tended to be more likely to complete the inpatient treatment than binge eating/purging AN patients, but without significance (p = 0.08 by a chi-square test). Significant differences between completers and dropouts in body weight and BMI were seen at discharge (p < 0.0001). The duration of hospital stay was longer for completers than for dropouts (p < 0.0001).
A MANOVA revealed significant between group differences in the 13 validity and clinical scales of the MMPI (Wilks' lambda = 0.691, F(13,61) = 2.098, p = 0.027). Dropouts were significantly higher than completers on the Hs, Pd, Pa, Pt, Sc and Ma scales of the MMPI by t-tests with Bonferroni correction [p < 0.003 (0.05/13)] (Table 2) [see Additional file 1-T2]. MANOVA also revealed significant between group differences in the 13 Wiggins content scales (Wilks' lambda = 0.711, F(13,61) = 1.904, p = 0.047). Dropouts were significantly higher than completers on 10 of the 13 Wiggins contents scales by t-tests with Bonferroni correction [p < 0.004 (0.05/13)] (Table 3) [see Additional file 1-T3]. Significant between group differences were also found in the 11 Lachar-Wrobel critical items (Wilks' lambda = 0.457, F(11,62) = 2.573, p = 0.009). The dropouts had a higher score than completers for 9 of the 11 items by t-tests with Bonferroni correction [p < 0.004 (0.05/11)] (Table 4) [see Additional file 1-T4].
Logistic regression analysis with forward stepwise selection revealed that Sc, hypomania, deviant thinking and experience, and antisocial attitude were factors predicting dropout (Table 5) [see Additional file 1-T5]. Table 6 includes the subscales of Sc and shows that the dropouts had higher scores than completers for all subscales, social and emotional alienation, object loss, lack of ego mastery, and sensorimotor dissociation [see Additional file 1-T6].
Table 7 shows the reasons for dropping out. About 80% of all dropouts from inpatient treatment left during the middle phase treatment [see Additional file 1-T7]. There was no difference in the distribution of patients by phase between patients who left on their own initiative (group P) and patients whose treatment was terminated by their therapists (group T). Dropouts in the early phase (n = 4) had higher scores on many MMPI scales than patients who dropped out in the other phases (n = 20), but with no significant difference (data not shown). The reason for the lack of significance may have been that the sample size of patients in the early phase was too small. The direct causes of termination are as follows: reduced motivation for treatment (54%), including four patients (17%) who left the hospital ward or building without permission; shoplifting or theft of food (21%); suicide attempt or self-injury, mood instability, and violation of a rule of the hospital ward (each 8%). A significant difference in the direct cause of dropping out was found between group P and group T (χ2 = 15.2, p < 0.01). A reduction of motivation was found more often in group P than in group T. Patients who left on their own initiative had a significantly higher family problems (FAM) score on the MMPI than those who were discharged by their therapist (64.8 ± 11.4 vs. 54.3 ± 11.8, p = 0.033, 95%CI= -20.13~-0.87) or completers (64.8 ± 11.4 vs. 48.6 ± 11.4, p < 0.001, 95%CI= -23.80~-8.65) by LSD procedure following MANOVA (Wilks' lambda = 0.559, F(26,120) = 1.601, p < 0.05) and ANOVA (F(2,72) = 9.378) with a Bonferroni correction [p < 0.006, (0.05/8)] (data not shown in Table).