Setting and study population
The Canadian Multi-centre Osteoporosis Study (CaMos) is a population-based study primarily designed to delineate the impact and prevention of osteoporosis in Canada . A random age-, sex-, and region-specific community-dwelling sample of 9423 adults (71% women), aged 25 years and older, able to communicate in English, French or Chinese was recruited in 1995/6 across seven provinces and nine urban and rural cities . Over 60% were followed prospectively for 20 years (until 2016), and examined comprehensively every 5 years using questionnaires and in-person clinical assessments, bone mineral density (BMD) tests and radiographs . We conducted a longitudinal analysis of the currently accessible 10-year data for a cohort of women and men aged 50 years and older, regardless of their history of fracture. Ethics approval was granted through the Research Ethics Board of academic institutions associated with each CaMos centre.
Main exposure and covariates
Prevalent clinical fractures were reported at baseline and incident fractures were reported at annual follow-ups and confirmed by structured interviews via telephone or in-person . The questionnaires included information related to fracture site, fracture number, circumstance, treatment, and radiographs or medical reports. Incident fractures were defined as any new low-trauma fracture excluding head, toe and finger fractures. Our main analyses categorized low-trauma fractures by location, which included hip, clinical vertebral, and non-hip non-vertebral (NHNV) fractures (e.g., leg not hip, pelvis, rib, shoulder, scapula, upper arm, wrist/forearm, and hand). We also examined the effects of prevalent (prior to baseline) and new single or multiple fractures where multiple fractures represented more than one event of any clinical low-trauma fracture (same or different type). Incident fractures were analyzed as time-varying predictors accounting for the presence of a new event in two time periods, between years 1 and 5 and years 6 and 10. Individuals without fractures represented the reference group.
All analyses were adjusted for baseline age and time of frailty assessment. We categorized age into three groups: 50–<65[reference], 65–<80 and ≥80 years; the time variable represented the duration of follow-up at which frailty was measured: baseline [reference, 1996], years 5 and 10 (2006). We explored the effects of sex and baseline body mass index (BMI, in kg/m2) on changes in frailty. BMI was categorized as: underweight (<18.49), normal weight (18.5–24.99, reference), overweight (25.0–29.9), obese class I/II (30.0–39.9), and pathologically obese-class III (≥ 40.0) . Our models also included: 1) socio-demographic factors: ethnicity (Caucasian vs. other), education (university or higher degrees vs. no university), employment history (employed full time or part time [reference], retired, homemaker, unemployed), and living arrangement (living alone: yes/no); 2) anthropometrics and lifestyle: excessive weight loss (> 10 pounds), physical activity related to strenuous, vigorous or moderate exercise reported in kilocals/week (changes in frailty analyzed per 1000 kilocals/week, equivalent to 3–6 METs [14, 15]), sedentary lifestyle (hours/day), smoking (never, past and current[reference]), and daily alcohol consumption (≥3 drinks, 1–<3, >0 to <1, and none); 3) bone health: femoral neck BMD T-score, history of falls (past month: yes/no), bed rest (immobilization: yes/no), antiresorptive therapy (baseline: yes/no), and total daily calcium and vitamin D intakes from food and supplements (changes in frailty analyzed per 1200 mg/day and 800 IU/day, respectively); 4) health-related quality of life (HRQL) measured by the physical and mental health subscales of the Medical Outcomes Trust SF-36 Health Survey (changes in frailty analyzed per a 5-point change in SF-36 scores); 5) cognitive status measured by the Mini Mental State Examination (MMSE), assessed in participants aged 65+ (changes in frailty analyzed per a 3.72-point change in MMSE scores). All analyses were also adjusted for the use of antiresorptives at year 10 due to substantial changes in their availability over time (1996: etidronate; 2006: etidronate, alendronate, clodronate, risedronate, pamidronate, zoledronate).
Frailty was measured by the 30-item CaMos Frailty Index (CFI). The construction and validation of the CFI is described in detail elsewhere . In brief, it was developed in the CaMos cohort aged 25 to 103 years (N = 9423) using a cumulative deficits framework . It included 30 variables related to a wide range of deficits in biologic systems (e.g., signs, symptoms, disease states and disabilities) that accumulated but did not saturate quickly with age, and had <5% of missing data . Thus, it included the following comorbidities: osteoarthritis, rheumatoid arthritis, thyroid disease, breast cancer, uterine/prostate cancer, inflammatory bowel disease, hypertension, heart disease (e.g., heart attack), stroke, thrombophlebitis, neuromuscular disease, diabetes type 1 or type 2, and kidney disease. It also included variables related to: general health, change in general health, feelings of having energy and tiredness, as well as deficits in: vision, hearing, walking, dexterity, cognition, pain, daily work, social activities, and limitations in: in moderate activities (e.g., moving table, vacuuming, golf, bowling), lifting or carrying groceries, climbing a flight of stairs, bending, kneeling, stooping, bathing or dressing (Additional file 1). Total CFI scores ranged from 0 to 1, with higher values indicating greater frailty; the upper limit was 0.66 and the mean rate of deficit accumulation per year of age was 0.04 (i.e., a minimal clinically important difference) .
In descriptive analyses, categorical variables were expressed as percentages and continuous data by means and standard deviations (SD). Generalized estimating equations models with an autoregressive correlation structure were used to analyze repeated measurements and associations between rates of change in frailty over time and predictors. Regression estimates generated in unadjusted and adjusted analyses indicated increases or decreases in the mean CFI score per 5 years (i.e., rates of change in frailty) for a unit change in a predictor (e.g., each new fracture). All models were adjusted for a statistically significant age-time interaction (p < 0.0001) indicating differences in changes in frailty for different age groups. We also confirmed the modifying effects of sex (p = 0.0005), prevalent hip (p = 0.006) and clinical vertebral fractures (p = 0.03), and developed the following five models: one for each sex for the whole sample (women, n = 5566; men, n = 2187), one for each sex for the sample without prior fractures (women, n = 4348; men, n = 1814), and one for the participants with prior fractures (n = 1574) as the rate of change was not significantly different between women and men with prior fractures (p = 0.62). We examined bias due to missing data or attrition using multiple imputations and worst-case scenarios. In the worst-case scenarios, for participants dropping out at years 5 and 10, we imputed the highest CFI value estimated for hip fractures (i.e., CFI = 0.268) or the highest upper limit reported in the literature (CFI = 0.70) . Statistical significance was set at an alpha-level of 0.05. Analyses were performed using SAS 9.4 (SAS Institute, Inc., NC). Additional results are presented in Additional file 2.