Introduction

Vertebral fractures are associated with increased mortality [1, 2, 3, 4], morbidity [5, 6, 7, 8] and financial burden [9, 10], and they are important predictors of future vertebral and hip fractures [11, 12, 13]. With an estimated prevalence of approximately 25% in both men and women [14, 15], vertebral fractures represent the most common of all osteoporotic fractures [16, 17, 18]. Seventy percent of vertebral fractures remain undiagnosed in clinical practice [19, 20, 21], yet sub-clinical fractures have been associated with back pain, functional limitations and loss of utility [1, 21] and may act as a clinical marker for subsequent fractures [11, 12, 13, 16].

In recent years, there has been a shift towards recognizing that the measurement of health-related quality of life (HRQL) provides a more complete representation of an individual’s experience with osteoporosis [22]. This is in contrast to previous studies that have focused solely on back-specific pain, activities of daily living (ADL) or general health [8, 21, 23, 24]. HRQL refers to a patient’s perceived physical and mental health over time and may be used by clinicians to better understand how a chronic illness interferes with a person’s day-to-day life [25]. In addition to generic HRQL tools such as the SF-36 [26], osteoporosis-specific HRQL tools have been developed and validated, including the osteoporosis quality of life questionnaire (OQLQ) and mini-OQLQ [27, 28, 29]. Disease-specific tools are more clinically sensitive and may be more responsive to detecting change [30]. The mini-OQLQ is a ten-item tool designed to be administered in a clinic setting [28].

The negative impact of vertebral fractures on HRQL have been well demonstrated [24, 31, 32, 33, 34, 35, 36, 37, 38], although the extent to which HRQL is impaired by factors other than the fracture itself is not well understood. The time after vertebral fracture [24] and number of subsequent fractures [31, 38, 39, 40] have been associated with HRQL. However, a wider range of patient and clinical factors and HRQL in patients has not been systematically examined. A better understanding of the full range of factors that may contribute to HRQL would provide a basis for implementing therapeutic strategies aimed at altering modifiable risk factors and subsequently improving HRQL post-vertebral fracture. We examined the independent contribution of clinical (co-morbidities, medications, fracture history and family disease history) and patient factors (demographics, exercise, education and living environment) to HRQL in vertebral fracture patients registered in the Canadian Database of Osteoporosis and Osteopenia (CANDOO).

Materials and methods

Study group

Patient records were selected from the CANDOO database, a prospective registry of approximately 10,000 patients seen in specialized tertiary care referral centers in eight sites across Alberta, Saskatchewan, Ontario and Quebec [41]. During the course of routine specialist care, a computerized record with over 400 data fields is generated for each patient. Data fields include demographic information, medications and adverse effects, female reproductive history, diet, exercise, fracture history, bone mineral density measurements and laboratory investigations. A self-administered quality of life questionnaire is completed by the patient at each clinical visit and reviewed by the specialist or nurse clinician.

The cohort for this study was comprised of all postmenopausal women registered in the CANDOO database between 1990 and 2002 who had sustained a prior vertebral fracture. Prior vertebral fractures were determined from the CANDOO questionnaire item, “Have you ever had any fractures?” Vertebral fractures may or may not have been confirmed by X-ray. Patients were included if they completed at least a portion of the HRQL questionnaire.

Quality of life questionnaire

HRQL was measured using the mini-OQLQ [28], an abbreviated tool derived from the original 30-item OQLQ [29]. Like the original questionnaire, the mini-OQLQ measures five domains: symptoms, physical functioning, emotional functioning, ADL and leisure. The shortened version has ten items, constructed from the two items with the highest impact in each of the five domains on the original OQLQ. The mini-OQLQ is self-administered, takes about 2–3 min to complete and was designed to facilitate questionnaire administration in a clinic setting [28]. Scores in each domain can vary from 2 to 14, although to standardize the values, scores are divided by the number of items used to generate the values. A score of 1 represents the worst possible function, and a score of 7 represents the best possible function. An increase or decrease of approximately 0.5 within each domain score is considered to be a clinically important difference in quality of life [42, 43]. The mini-OQLQ has previously been found to be a sensitive measure of HRQL; Adachi et al. [34] found patients with incident vertebral fracture had lower scores than patients without fractures on all five domains.

Statistical analyses

Due to missing data for some variables, a statistical modeling technique termed multiple imputation described by Rubin [44], and subsequently used in numerous longitudinal datasets, creates datasets that allow for the inclusion of subjects with missing data. Briefly, multiple imputation replaces each missing value with a set of plausible values that represent the uncertainty regarding the right value to impute. The multiple imputed datasets are then analyzed by using standard procedures for complete data and the results are combined from these analyses.

In this analysis, ten completed data sets were generated using the following baseline terms: past surgeries of the breast, stomach, bowel, hip or spine; medications including thiazides, sedatives, anticonvulsants, chemotherapeutics, immunosuppressants, NSAIDs, beta blockers, calcium channel blockers, corticosteroids, hormone replacement therapy and bisphosphonates (alendronate, etidronate or risedronate); calcium/vitamin D supplementation; co-morbidities including osteoarthritis or rheumatoid arthritis, atherosclerotic disease (cardiac, cerebral, chest pain or angina), cancer (breast, cervical, endometrial or colon), diabetes, kidney stones; migraines; hypertension; thyroid disease; epilepsy; family history of cardiovascular events, cancer and osteoporosis; non-vertebral fractures after age 20 (excluding fingers, and toes); time since last non-vertebral fracture; prevalent vertebral fractures; time since last vertebral fracture; falls in past year; age; baseline height and weight; smoking status; exercise (hours per week); alcohol (drinks per week); dietary calcium intake (mg/day); living arrangement (nursing home vs. community); employment status; educational level.

Separate multivariable regression analyses were performed for each mini-OQLQ domain using the above independent variables to determine regression coefficient estimates as well as 95% confidence intervals (CI). Interaction terms were included for age, smoking and exercise. Regression models were chosen by the r-square selection method using the Mallows’ Cp, statistic for final model selection. Statistical significance was defined as P <0.05. All analyses were preformed on a personal computer using a SAS/STAT (version 8.0; SAS Institute Inc., Cary, N.C.) software package.

Results

Patient characteristics and HRQL scores

Data were analyzed for 1,129 post-menopausal patients (mean age 67.2, SD 11.9 years) with vertebral fractures. Patient characteristics and scores on the mini-OQLQ domains are presented in Table 1. Mean scores on each of the five mini-OQLQ domains varied from 3.9 (SD 2.4; ADL) to 4.9 (SD 1.9; emotional functioning). The mean number of prevalent vertebral fractures per patient was 2.2 (SD 1.6) with a mean length of time since last vertebral fracture of 4.3 years (SD 9.0). Nearly 40% of the study group had a college or university education, 20% were still in the workforce, and fewer than 2% of participants lived in long-term care. Approximately 17% of participants were current smokers.

Table 1 Patient characteristics and mean mini-OQLQ scores
Full size table

About one-half of the patients completed the family history portion of the questionnaire ( n =600). Of patients with data, 54% had a family history of osteoporosis, 30% a history of cancer and 61% a family history of cardiovascular events. Medications and co-morbidities of the study sample are presented in Table 2.

Table 2 Medications and co-morbidities in study sample ( n =1,129)*
Full size table

Associations with HRQL

Regression coefficient estimates and final models for each mini-OQLQ domain are presented in Table 3.

Table 3 Adjusted regression coefficients and 95% confidence intervals for the mini-OQLQ domains
Full size table

Patient/lifestyle factors

Post-secondary education (college or university) was associated with improved HRQL scores across all domains by approximately 0.5 points. A family history of osteoporosis was also associated with improved HRQL across domains (\(\ifmmode\expandafter\hat\else\expandafter\^\fi{B}\) varied from 0.27 to 0.49). Women who were still in the work force had less severe symptoms \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= 0.57, \;\text{95\%} \;\text{CI:} \;0.29, \;0.85)\) and emotional function \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= 0.42, \;\text{95\%} \;\text{CI:} \;0.13, \;0.71)\) scores.

There was a positive association between exercise and HRQL, particularly on the symptom, emotional function and physical domains. However, it should be noted that six or more hours/week were necessary for even the upper CI bound to be clinically important. Significant interactions were found between exercise and smoking and exercise and corticosteroid use. On the leisure domain, exercise made a greater difference for smokers than for non-smokers. On the ADL domain, exercise had a greater impact for those using corticosteroids than for those not using corticosteroids.

Smoking was associated with a substantially decreased HRQL across domains, and most markedly affected leisure \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.84}, \;\text{95\%} \;\text{CI:}\;{-1.40}, \;-0.32)\) and ADL \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.62}, \;\text{95\%} \;\text{CI:}\;{-1.11}, \;-0.13).\) Weight was negatively associated with HRQL on the symptom domain \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.01}, \;\text{95\%} \;\text{CI:}\;{-0.02}, \;-0.004);\) an extra 15 kg of weight, for example, resulted in a –0.15 decrease. The number of falls (since the last visit) was associated with decreased scores; however, approximately three or more falls were required for even the lower CI bound to reach clinical importance.

Living in long-term care (LTC) was associated with the most marked negative effect on HRQL scores across domains, decreasing scores by –1.84 (95% CI: −3.0, −0.7), −1.80 (95% CI: −3.02, −0.56), −1.16 (95% CI: −2.12, −0.21) and −0.95 (95% CI: −2.05, 0.14) on the ADL, leisure, emotional function and physical domains, respectively. The symptom domain was not affected by living in LTC.

Medications

Thiazide therapy was associated with a substantially improved HRQL on every domain except emotional function, \(\ifmmode\expandafter\hat\else\expandafter\^\fi{B}\) varied from 0.51 to 0.70. On the other hand, calcium channel blockers had a clinically important, but not statistically significant negative association on the emotional function, \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.42}, \;\text{95\%} \;\text{CI:}\;{-1.06}, \;-0.23))\) and leisure domains, \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.62}, \;\text{95\%} \;\text{CI:}\;{-1.45}, \;-0.58)).\) Anticonvulsant use had strong adverse associations on the physical, ADL and leisure domains, decreasing HRQL scores by greater than −1.0. Sedatives were consistently associated with reduced HRQL across domains by approximately −0.50. Chemotherapy reduced emotional function and ADL scores; NSAIDs were associated with reduced symptom scores, and those prescribed corticosteroids had lower ADL scores.

Co-morbidities/medical history

The number of prevalent vertebral fractures was not significantly related to HRQL scores; however, a decreasing pattern was noted for the symptom, emotional function and physical domains. Time since vertebral fracture was not related to HRQL. The number of non-vertebral fractures was associated with decreased HRQL on the symptoms \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.11}, \;\text{95\%} \;\text{CI:}\;{-0.18}, \;-0.04)\) and emotional function \((\ifmmode\expandafter\hat\else\expandafter\^\fi{B}= -{0.26}, \;\text{95\%} \;\text{CI:}\;{-0.33}, \;-0.18).\) Two or more fractures were associated with a clinically important change on emotional function, and four or more on the symptom domain. Time since non-vertebral fracture significantly improved ADL scores, and a positive pattern was noted for the physical and leisure domains.

A past surgery of the hip or spine had a notable impact on decreasing HRQL across all domains except emotional function. In particular, a −1.41 (95% CI: −1.94, −0.88) decrease on ADL was noted.

Atherosclerotic disease and hypertension produced a consistent reduction in HRQL across several domains, reducing scores by approximately −0.2 to −0.5 points. Diabetes was associated with a negative impact on physical and leisure domains, and migraines had a notable impact on ADL.

Discussion

This study examined a large osteoporosis patient database that contains information on a wide-range of clinical and patient factors collected during routine specialist visits. The results demonstrate that, in women with vertebral fractures, several factors were independent determinants of HRQL, as measured by the mini-OQLQ, and this may assist clinicians in caring for patients in a holistic manner.

Approximately 75% of patients with a clinical vertebral fracture will experience chronic pain [23, 45, 46], causing emotional distress and further limiting an individuals ability to work and engage in recreational and social activities [28]. Although we did not compare our vertebral fracture cohort with controls, overall there was a moderate level of impairment across all HRQL domains—mean HRQL scores varied from 3.9 (SD 2.4) on ADL to 4.9 (SD 1.9) on the leisure domain, out of a possible score of 7 (highest level of function).

Other studies using the QUALEFFO [20, 31] and OPAQ [38] (osteoporosis-specific HRQL tools) have found that multiple vertebral fractures have a marked cumulative effect in decreasing HRQL (i.e., scores lower with each fracture). A Canadian population-based study, using the SF-36, found multiple vertebral fractures had no additive negative effect on HRQL [35]. Our study found the number of prevalent vertebral fractures had a significant impact only on the emotional function domain, and several prior vertebral fractures were necessary for a clinically relevant effect. Although not significant, there was a negative pattern for the symptom and physical domains. Unlike the Canadian population-based study, most patients in the CANDOO dataset already had multiple vertebral fractures, so the lack of a large number of patients without multiple vertebral fractures may have made it difficult to identify this as a salient factor. The mean number of prevalent vertebral fractures in CANDOO participants was 2.2 (SD 1.6) with a mean time since last vertebral fracture of 4.3 years.

The benefits of using HRQL tools in daily clinical practice include: enhancing patient-physician communication, monitoring response to therapy and detecting physical or psychosocial problems that often go unnoticed [28, 34, 47]. The findings of this study have implications for clinicians regarding which variables to focus on during routine clinic visits. Encouraging patients to exercise, particularly patients who smoke or are taking corticosteroids, is an important consideration. Discussing osteoporosis in the context of family history may be an interesting strategy to assist patients with becoming more empowered and knowledgeable about the implications of managing the disease. Independent of other factors, patients with a high school or lower education had lower HRQL across domains, and it may be particularly important to consider strategies to enhance HRQL in this group. Patients with any of the following risk factors should also be carefully considered in the context of HRQL: past surgery of the hip or spine, extra weight, multiple falls, living in LTC and smoking. In terms of medical conditions, atherosclerotic disease, diabetes or hypertension should be flagged as possible markers for reduced HRQL. Thiazide use proved to be a therapy that may increase, rather than decrease, HRQL. In contrast, patients who are taking sedatives, anticonvulsants, chemotherapy, calcium channel blockers or corticosteroids should be monitored cautiously for deterioration to HRQL during the course of therapy. Anticonvulsant use in particular was associated with a strong negative effect.

Exercise was positively associated with HRQL; however, several hours a week was required to have clinical importance. We have previously established [48] that a home exercise program for women with vertebral fractures (3×60 min/week) improved HRQL over 12-months. Participants in that study were trained to complete stretching, strength training and aerobic activity. In the current study, we did not examine the type of exercise. Future studies should examine what role the type of exercise plays in HRQL for vertebral fracture patients.

Interestingly, a family history of osteoporosis improved HRQL across several domains. It is possible that patients who were aware of their family history were more educated about the disease and discussed similar signs or symptoms with relatives. Patients who are more educated about the disease likely feel a greater sense of empowerment and actively engage in osteoporosis self-management activities (i.e., knowledge empowers them to take care of their health and allows them to participate more fully in leisure, ADL and physical pursuits).

Our study and others [49] have found that hypertension has a negative effect on HRQL. Several large RCTs have examined HRQL in relation to hypertensive agents [50, 51, 52, 53]. In our study, thiazide use was strongly associated with improved HRQL across all domains except emotional function. Other studies examining diuretic therapy and HRQL have found that diuretics may either improve or have no deterioration in HRQL [50, 51]. In the Systolic Hypertension Elderly Program (SHEP) trial [51], there were no negative effects or major differences between the diuretic group and placebo, and the active treatment had a slight positive effect on cognitive, physical and leisure function. Similarly, we found no effect on measures related to emotional state [51]. Recent guidelines suggest diuretics are well tolerated and particularly recommended for elderly patients with systolic hypertension [50, 54, 55].

We found that calcium channel blockers were associated with clinically important reductions in HRQL on the emotional function and leisure domains. This finding makes sense given that the comparison group (those not on calcium channel blockers) would not have experienced side effects (related to calcium channel blockers) and were likely non-hypertensive. For hypertensive patients, calcium channel blockers may have a favorable impact on HRQL compared with placebo [50, 52, 53], although this finding is not equivocal. In the Systolic Hypertension in Europe Trial (SystEur), patients treated with a calcium channel blocker had some deterioration in HRQL compared with placebo, but were not substantially impaired [53].

Even though only two percent of our sample lived in LTC (versus the community), LTC was the strongest independent determinant of HRQL. Our results demonstrate that individuals living in LTC had poorer HRQL, independent of several co-morbid conditions for which we controlled. Although many individuals require the extra medical care available in LTC, this finding reinforces the concept that keeping an individual in the community as long as possible may have a substantial impact on quality of life (although HRQL may also deteriorate prior to admission to LTC).

Smoking was repeatedly associated with poorer HRQL across domains, which is consistent with studies of HRQL in other disease groups [56, 57, 58] and in the general population [59]. Martinez et al. [59] found that even light-to-moderate smokers with a short smoking history have significantly impaired HRQL compared to never-smokers. Mitra et al. [56] found that those who quit smoking over time had significant improvements in HRQL, demonstrating the importance of encouraging patients to quit smoking.

Similar to other studies [60, 61], we found weight was also negatively associated with HRQL (symptom and physical domains). Heo et al. [61] found a J-shaped association between BMI and HRQL, and that the relationship was attenuated by joint pain and obesity-related co-morbidities. From a clinical standpoint, interventions aimed at weight loss have shown to improve HRQL over time [62, 63]. Fontaine et al. [63] found that a 13-week weight-loss intervention in mildly to moderately overweight persons improved HRQL immediately after the program and at 1-year follow-up.

Results from Cockerill et al. [20] suggest that it is really the second vertebral fracture that results in a marked reduction in HRQL. In that study, subjects with both recent and pre-existing vertebral fractures had the greatest impairment in HRQL, whereas cases with recent fracture and no preexisting deformity were not significantly different from controls. We were unable to confirm these results, as incident vertebral fractures were not examined in this analysis, and the majority of our patients already had two or more fractures. We have previously examined the effect of incident vertebral and non-vertebral fractures on HRQL using the CANDOO database [34]. In that study, CANDOO patients with incident vertebral fracture had significantly lower HRQL scores (adjusted for confounding factors) across all mini-OQLQ domains.

There was no consistent association between time since vertebral fractures and HRQL in our study. This finding is contrary to Begerow et al. [24], who found time after vertebral fracture had significant impact on HRQL, but is consistent with Hallberg et al. [39], who found vertebral fracture patients had below normal scores even 2 years after the last fracture. Adachi et al. [35] also found that the time since last vertebral fracture was not strongly related to HRQL. Although we were not able to prove this, it is possible that HRQL remains lower regardless of time from fracture.

Our results suggest that prior non-vertebral fractures may act as a marker for reduced HRQL (significant on symptom and emotional function domains); in particular, individuals with two or more non-vertebral fractures may be at greatest risk. Time since non-vertebral fracture was associated with HRQL on the ADL domain.

Strengths and limitations

CANDOO is a large, homogenous dataset collected from patients who are assessed in tertiary care settings. The mini-OQLQ tool is incorporated into the clinical visit, and patients are given an opportunity to consider the questionnaire properly. A major strength of this work was our ability to carefully control for a wide range of potential confounders in multivariable analyses. This is the first paper in the osteoporosis literature to examine other factors comprehensively, including modifiable ones, which are responsible for HRQL beyond just the vertebral fracture itself.

A few potential limitations should be noted. This was a cross-sectional study, and while we found a number of meaningful associations, we were not able to make inferences about causality. The CANDOO database provides a breadth of clinical and patient factors available for our analysis; however, it remains possible that there were other confounding factors not captured by our dataset. We were unable to look at social support, a factor that is often associated with HRQL. There was also missing information on certain variables; however, this was compensated for using a valid multiple imputation technique [44]. There is some possibility of recall bias as the CANDOO questionnaire and the mini-OQLQ rely on the factual recall of patients. Not all spinal fractures were confirmed by X-ray; however, any associations we found were likely weakened by including patients who did not truly have a spinal fracture. Due to lack of data on male patients, only female patients were included in this study, so our results should be used with caution when extrapolating to males. Our data were obtained from patients in specialty clinics (osteoporosis), and not necessarily representative of patients seen only in other settings; however, it seems unlikely that the actual relationship between risk factors and HRQL would vary between patients with or without a specialist.