The ethics committee of the Saiseikai Kure Hospital, where participants were recruited, approved this prospective cohort study on patients with vertebral fracture (approval no., 127). This study was conducted in accordance with STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement. After provision of a detailed description of the protocol, written informed consent was obtained from all patients. To protect patient confidentiality, identifying information, including patient name, date of birth, and address, was not collected. The study was conducted in accordance with the Declaration of Helsinki.

The study setting was a hospital. Between March 2018 and October 2019, recruitment, follow-up, and data collection were performed. The potential participants were recruited by therapists in the rehabilitation department.

All patients were Japanese and met the following inclusion criteria: (1) osteoporotic vertebral fracture (grade ≥1 in the semiquantitative [SQ] method) treated with conservative therapy, (2) ability to walk independently before fracture, and (3) absence of another or second vertebral fracture during hospitalization. Patients were excluded based on the following criteria: (1) complications during hospitalization and (2) severe dementia (defined as a Revised Hasegawa’s Dementia Scale [HDS-R] score ≤9). HDS-R scores of 21–30, 15–20, 10–14, and ≤9 indicated normal, dementia doubt, light to moderate dementia, and advanced dementia, respectively [10].

Based on the presence or absence of vertebral collapse progression between admission and ≥4 weeks after admission, patients were divided into two groups: if progression of collapse was detected, patients were assigned to the collapsed group, and if progression of collapse was not detected, patients were assigned to the non-collapsed group. In the comparison of computed tomography (CT) or X-ray images obtained at admission and ≥4 weeks after admission, vertebral collapse was determined. The definitions of collapse were as follows: in the case of X-ray, grade assessed by the SQ method increased by at least 1; in the case of CT, collapse was determined by the physician.

Rehabilitation programs were determined by the therapists according to orders from physicians and were started the next day after admission. Until a corset was made, interventions were performed on the bed on prevention of disuse syndrome. After completion of the corset (>2 weeks), patients performed rehabilitation programs in the rehabilitation room [11]. Rehabilitation programs included functional exercises (neuromuscular facilitation, joint range of motion, and muscle strengthening exercises), ADL exercise, and psychological support for patients and their families. Basically, the frequency and duration of intervention were 5 days per week and 30 minutes to 1 hour per day, respectively. These interventions were implemented until discharge. Discharge from the hospital was set to be the patient’s original life, approximately 2 months after hospitalization. Nonsteroidal anti-inflammatory drugs were habitually used for pain control.

The following basic information was collected: age, sex, body mass index (BMI), duration of bed rest after admission, presence or absence of assistance needed to walk independently in the hospital, dwelling place (home or institution), length of hospital stay, and cognitive function assessed by HDS-R.

Medical information was as follows: vertebral fracture information (location of fracture, number of fractures, thoracolumbar spine fracture, magnetic resonance imaging [MRI] intensity ratio, degree of fracture, vertebral fractures with posterior wall, presence or absence of neurologic symptoms), previous medical history (chronic obstructive pulmonary disease [COPD], cardiovascular disease, cerebrovascular disease, cancer, hypertension, diabetes, kidney failure, and vertebral fracture), presence or absence of pharmacotherapy (parathyroid hormone, anti-receptor activator of nuclear factor-κB (NF-κB) ligand antibody, bisphosphonate, and selective estrogen receptor modulator), bone metabolism marker (total P1NP, 25-hydroxy vitamin D [25(OH) vitamin D], tartrate-resistant acid phosphatase 5b [TRACP-5b]), and Geriatric Nutritional Risk Index (GNRI). The location of fracture was recorded with detailed spinal fracture sites. The number of fractures was recorded for one or two vertebrae. MRI intensity ratio was calculated as follows: intensity change height/vertebral body height×100 [12]. To determine the degree of vertebral fracture, the SQ method was used, which assesses the degree of vertebral fracture based on visual inspection without measurement of vertebral heights [13]. The grading was as follows (Fig. 1): normal (grade 0); mild deformity (grade 1: 20%–25% reduction in anterior, middle, and/or posterior height and 10%–20% reduction in area); moderate deformity (grade 2: 25%–40% reduction in height and 20%–40% reduction in area); and severe deformity (grade 3: ≥40% reduction in height and area). Grade ≥1 was considered “fractured”, and grade 0 was considered “not fractured”. Previous studies reported that both validity and reliability of the SQ method that assessed the degree of osteoporotic vertebral fracture were high [14,15]. Medical history was recorded if patients had diseases requiring hospital treatments (outpatient or inpatient). Moreover, medical history that was significantly associated with vertebral collapse was investigated in more detail.

Physical functions included pain assessed by Pain Catastrophizing Scale (PCS), Central Sensitization Inventory (CSI), Visual Analog Scale (VAS), knee extension muscle strength, and movement control during one leg standing. We hypothesized that physical functions are related to vertebral collapse. Therefore, pain, muscle strength, and balance ability are assessed as universal physical functions. Pain-related catastrophizing was assessed using the Japanese version of PCS [16], comprised of 13 items describing different thoughts and feelings associated with pain. Patients answered questions according to the following five categories: 0, not at all; 1, slight degree; 2, moderate degree; 3, great degree; and 4, all the time. The total score was used for the analyses. Higher PCS scores indicate greater pain catastrophizing. Central sensitization was assessed using the Japanese version of the CSI [17], which comprised of two parts: A and B. Part A is a questionnaire including 25 self-reported items, and patients answer questions according to the following five categories: 0, never; 1, rarely; 2, sometimes; 3, often; and 4, always. Part B determines characteristic diagnosis history and was not used in this study. Higher scores indicate more severe CS. Pain intensities during rest and activity were assessed by 100-mm VAS. No pain was scored 0.

Isometric knee extension muscle strength was measured using a handheld dynamometer (HHD; Sakai Medical Corp., Tokyo, Japan) [18]. Patients sat on a chair and kept their trunk straight and perpendicular using both hands on the bench beside the body. An HHD sensor was fixed to the anterior lower leg using a traction belt placed between the level of the lateral malleolus of the lower leg and the main pillar behind the chair. Using the HHD, isometric knee muscle extension strength was measured for 5 seconds at a flexion angle of 90°, and care was taken to keep the traction belt parallel to the floor. The measurements were repeated 3 times with intervals of at least 30 seconds between measurements. The corrected muscle strength was obtained using the following equation: muscle strength measured by HHD×lower leg length (nearly to the moment arm)/body weight. Lower leg length was measured as the distance between the lateral epicondyle of the femur and lateral malleolus, with the measurement the leg being optional.

Movement control during one leg standing was assessed according to the criteria described by Lenzlinger-Asprion et al. [19]. The reliability of these criteria had been confirmed by a previous study [20]. Criteria 1: The hip joint should remain stable in rotation, abduction, and extension. The pelvis and upper body should not move from their initial position. Criteria 2: The vertical axis of the leg should remain straight, and varus or valgus should not develop. The patella should be directed to the third metatarsal. Criteria 3: If intermittent support by the hand against the wall or contralateral foot on the floor was necessary, the component was considered incorrect. If additional support was needed throughout the entire exercise, the component was valued as >1 incorrect component. Evaluation was conducted using a 3-point Likert scale: 2 points, correct; 1 point, almost correct; and 0 point, incorrect/false. The leg measurement was optional.

Quality of life was assessed as a psychological function using the questionnaire for quality of life of elderly subjects with osteoporosis (EOQOL), which includes 49 questions [21]. The subdirectories of the EOQOL include pain, ADL, movement, physiological activity, duty, recreation, and well-being. Each question was scored as follows: 5 points, very good; 4 points, good; 3 points, usually; 2 points, bad; and 1 point, very bad. A higher score on the EOQOL was better.

Activity was defined as ADL and walking ability, and these were assessed using the Barthel Index (BI) [22]. BI assesses ADL performance (i.e., feeding oneself, bathing, dressing, grooming, and ability to move) on a scale from 0 to 100 (0, very dependent; 100, independent). Walking ability was scored 0–15 points at intervals of 5 points, where 0 points represent the worst and 15 points represent the best.

Basic medical information was assessed at hospital admission and discharge. Pain (PCS, CSI, VAS) was measured at hospital admission and 4 weeks after admission. Muscle strength and movement control during one leg standing were assessed 4 weeks after admission. ADL and walking ability were assessed before fractures and 4 weeks after admission.

To reduce the measurement bias, the author (T.U.) was not involved in participant enrollment or data collection. Patients were informed of the purpose of the study but not the hypothesis. To reduce selection bias, outcomes were selected based on previous studies.

First, characteristic data were compared using Mann-Whitney U-test, t-test, and chi-square test, as appropriate. Next, univariable logistic regression analysis was performed to explore the predictors of vertebral collapse at >4 weeks after admission. Finally, hierarchical logistic regression analysis was performed to explore the predictors of vertebral collapse at >4 weeks after admission. Further, hierarchical logistic regression analysis was conducted to explore the association between the desired outcome of vertebral collapse at >4 weeks after admission (scored 1) and vertebral non-collapse at >4 weeks after admission (scored 0). Hierarchical logistic regression analysis used two models: model 1, admission model using basic medical information and physical function at admission; model 2, prediction model using a data set of basic medical information and physical function at 4 weeks. Initially, age, sex, and BMI were forcibly input as confounding factors in block 1. Thereafter, other independent variables were input using a variable increasing method in block 2. The threshold of correlation coefficient between independent factors was set to 0.8, and a high correlation with the dependent variable was selected to account for multicollinearity. To comprehensively examine the extracted variables, the logit transformed value of the p-value obtained from the regression analysis was defined as the score. The score was provided as a summary of the effects of the extracted independent variables in the form of a linear expression, and this score can appropriately correspond to the probability value by inverse transformation of the logit transformed value. The score can be used to appropriately estimate the mutual association (tradeoff) between each extracted independent variable considering the main effect leading to collapse at ≥4 weeks. Further, to predict vertebral collapse at >4 weeks after admission, prior probability was calculated based on Bayes’ theorem to calculate the posterior probability for the score. All statistical analyses were conducted using IBM SPSS ver. 22.0 for Macintosh (IBM Corp., Armonk, NY, USA), and the significance level was set at a p-value <0.05.

Study size was calculated using MedCalc statistical software ver. 19.2 (MedCalc Software, Ostend, Belgium). The area under the receiver operating characteristic curve (AUROC) could distinguish between nonpredictive (AUROC <0.5), less predictive (0.5< AUROC <0.7), moderately predictive (0.7< AUROC <0.9), highly predictive (0.9< AUROC <1), and perfect prediction (AUROC=1) [23]. In this study, an AUROC value of 0.7 was set as criteria to indicate the superiority of statistical discrimination. As for the ratio between the two groups, Kanchiku et al. [7] reported that 57% of elderly Japanese patients with osteoporotic vertebral fractures (average age, 79 years) had collapse after admission. Based on this finding, we assumed that the ratio of the negative to positive patients (i.e., those with collapse at ≥4 weeks after admission) was 1 (50%) to 1 (50%). The alpha value was set at 0.05, and power was set at 0.8. The hypothesized AUROC and null hypothesis AUROC were set at 0.75 (indicating moderate power) and 0.5 (indicating no discriminating power), respectively. Consequently, 19 patients were required in the positive group and 19 in the negative group, with a total of 38 patients.