Mortality After Osteoporotic Fractures: What Proportion Is Caused by Fracture and Is Preventable?

Abstract

Our knowledge of the association between low-trauma fractures and mortality has evolved considerably over the past half century. An early spike of mortality during the first several months after hip fracture was first established in the 1960s and was recognized as more likely for males and those with more preexisting comorbid conditions.1-4 Clinical vertebral fractures were first noted to be associated with higher than expected mortality rates in the early 1990s,5 and subsequently Center and colleagues demonstrated in the Dubbo observational cohort study that other major osteoporotic fractures (pelvis, proximal humerus, distal femur, proximal tibia, and multiple ribs) were also associated with higher than expected mortality rates, especially in men.6 Other fractures such as those at the distal forearm were not shown in earlier studies to be associated with excess mortality.5, 6 However, a subsequent analysis of the Dubbo cohort showed excess mortality associated with distal forearm and other minor low-trauma fractures among men and women aged 75 years and older,7 and excess mortality after distal forearm fractures has been shown among men in the Manitoba Bone Density Database cohort.8 The latter study also showed the risk of excess mortality after all fracture types (with the exception of distal forearm fractures among women) to be highest in the first year after fracture.8 However, there remains considerable debate and uncertainty as to what proportion of the excess mortality after low-trauma fractures is caused by the fracture itself (and is therefore preventable by preventing the fracture) and what proportion is caused by associated comorbid conditions and poor health status that predate the fracture. Kanis and colleagues postulated that the early spike of mortality during the first few months followed by a decreasing rate of mortality during the second 6 months after hip fracture indicated that the majority of these early deaths were caused by the fracture itself and that any excess deaths beyond this time point would likely be due to preexisting poor health status related to comorbid conditions.9 Two observational cohort studies reported in this issue of JBMR, by Andrich and colleagues and by Tran and colleagues, shed additional light on the association of low-trauma fractures and mortality but leave many questions unanswered. Andrich and colleagues estimated the excess mortality associated with pelvis fractures over the 1-year period after fracture occurrence in a large (more than 1.3 million) population of insurees aged 60 years and older of a large health insurer in northwest Germany using Cox proportional hazards models.10 The population for these analyses was 5865 individuals with a first pelvis fracture and 193,159 age- and sex-matched controls without pelvis fracture. This study is unique compared with prior studies in that it adjusted the estimated hazard of mortality associated with pelvis fracture not only for preexisting comorbid conditions but also for pre-fracture need for care assistance due to functional limitations in basic activities of daily living (ADL) such as personal hygiene and mobility. This study confirmed prior studies11-13 that pelvis fracture is associated with excess mortality and showed that excess mortality was highest in the first 4 weeks after its occurrence. This hazard ratio for mortality gradually fell over time such that by 32 weeks of follow-up, there was no longer any excess mortality associated with the prior pelvis fracture, after adjustment for comorbid conditions, pre-fracture level of functional dependency and capacity for self-care, and other covariates. As has been shown for hip fracture, the excess mortality was greater in men than in women. Those who had more severe fractures and those who had to be hospitalized had particularly high excess mortality in the first few months after fracture. Even in this subset, however, excess mortality attributable to pelvis fracture was no longer evident 8 or more months after the fracture after multivariable adjustment. In contrast, those who could be treated as an outpatient did not have any excess mortality after multivariable adjustment. Importantly, the need for hospitalization after pelvis fracture was associated with prefracture level of functional dependency and compromised capacity for self-care. The study of Andrich and colleagues did not describe specific causes of death in the early period after pelvis fracture. Prior studies have examined cause of death primarily after hip fracture; although a minority of the early deaths are the result of postoperative complications such as delirium,14, 15 pneumonia,16 acute cardiovascular events,17, 18 and pulmonary embolism19 (Fig. 1, arrow a), the majority of deaths post-hip fracture are from the same common comorbid conditions that are the causes of death in the broader older population (Fig. 1, arrows b and c),20-23 especially cardiovascular conditions and pneumonia.21 However, pre-fracture comorbid conditions,24, 25 functional status,24, 25 and physical performance capability (such as quadriceps strength)7 have been shown to be independent predictors of early mortality after hip and other low-trauma fractures. It is also plausible that reduced functional capacity and a high burden of preexisting comorbid illness, as markers of poor physiologic reserves,26 are effect modifiers of the association of fracture and early subsequent mortality (Fig. 1, dotted arrow d). The physiologic stress of hip fracture may render these individuals more susceptible to complications of and demise from common comorbid conditions compared with individuals with more robust physiologic reserves. Three studies, including a recent meta-analysis across eight cohorts, have reported the association of hip fracture with early incident mortality to be indeed stronger among those with a higher burden of pre-fracture comorbid illness.16, 27, 28 The findings of Andrich and colleagues are also consistent with the hypothesis that pre-fracture reduced functional status may also be an effect modifier of the association between major osteoporotic fractures and mortality. This needs to be tested using prediction models for early post-fracture mortality that include interaction terms between fracture and pre-fracture measures of functional capacity and/or physical performance. More severe fractures plausibly cause greater acute physiologic stress, susceptibility to demise from early complications from and further reduction of functional capacity, and may have a greater impact on mortality. Hence, it is not surprising that Andrich and colleagues found that multiple pelvis fractures had a greater impact on mortality than less severe ones such as single pubic or ischial ramus fractures. This is consistent with findings from prior observational studies. Clinical vertebral fractures have been noted to cause excess mortality,29 but incident radiographic vertebral fractures (the majority of which do not come to clinical attention) are not associated with excess mortality after multivariable adjustment.30 Moreover, more serious non-hip non-vertebral fractures (at the pelvis, proximal humerus, proximal tibia, distal femur, and multiple ribs) are associated with higher excess mortality than less severe distal limb fractures.6, 7 Similar to hip fracture patients, pelvis fracture patients are older, have more comorbid conditions, and poorer health status than what is generally described for those who have distal limb fractures.31 Observational studies of hip fracture mortality have consistently found high excess mortality in the early period after hip fracture.8, 9, 32-35 However, after adjustment for comorbidity, and/or functional status, some studies report longer-term excess mortality after hip fracture33, 36, 37 and others do not.24, 38, 39 These inconsistent findings could be due to the fact that the time course of excess mortality after hip (and pelvis) fracture may be quite different for those with a higher burden of comorbidity or disability than those with better pre-fracture health status; this may not be apparent in studies that simply adjust for those covariates but becomes apparent with stratified analyses. Magaziner and colleagues noted that hip fracture patients with 3 or more comorbid conditions had very high short-term excess mortality compared with a control group of participants with 3 or more comorbid conditions in the Longitudinal Study on Aging, adjusted for age, education status, and ADL impairments.28 However, 4 years after hip fracture, their multivariable-adjusted mortality rate again matched that of the control group, perhaps because any residual effects of the fracture were dwarfed by a higher background mortality risk associated with their high comorbidity. In contrast, those with 2 or fewer comorbid conditions had a lower early spike of excess mortality but maintained a higher multivariable-adjusted mortality rate than controls throughout the entire follow-up period. The same pattern of excess mortality was found for the subset of hip fracture patients with 3 or more ADL impairments compared with those with 2 or fewer impaired ADLs; a higher pre-fracture level of disability was associated with short-term but not long-term excess mortality, whereas those with 2 or fewer pre-fracture ADL impairments had lower early post-fracture mortality but sustained excess mortality over the entire follow-up period. These findings are quite compatible with the Andrich study; pelvis fracture patients with more pre-fracture disability and poorer health status requiring hospitalization had high short-term excess mortality; those that could be treated as outpatients did not. The study of Tran and colleagues using data from the Canadian Multicentre Osteoporosis Study (CaMos) examined the association of mortality with low-trauma fractures at all skeletal sites in order to estimate not only the relative risk of long-term mortality (over a median follow-up time of 14 years) attributable to low-trauma fractures but also the impact of fractures on mortality at the population level.40 The authors hypothesized that although the relative risk of mortality associated with non-hip non-vertebral (NHNV) fractures (as an aggregate group) was likely to be lower compared to hip and clinical vertebral fractures, the mortality impact at the population level of NHNV may be as high as for hip and clinical vertebral fractures. This hypothesis was confirmed; three-quarters of all fractures occurred at NHNV skeletal sites, and the number of deaths in the entire population associated with NHNV fractures was higher than for hip or clinical vertebral fractures. The overall population impact of excess mortality associated with fractures was high; among women of all ages, 1 in 9 (11%) of all deaths were associated with prior fracture. However, excess longer-term mortality was not demonstrated in those under age 75 years, and there was lack of adequate power to estimate the association of NHNV with mortality in men. These results for NHNV fractures, by and large, differ from some others published previously. The Dubbo Osteoporosis Epidemiology Study (DOES) demonstrated excess age-adjusted mortality after NHNV fractures,7 but it is unclear if this excess mortality in DOES would still be evident after adjustment for comorbid conditions and pre-fracture functional status. Age-adjusted mortality rates in the Manitoba Bone Density Database cohort were not raised beyond 1 year for NHNV fractures even for those over age 70 years (except for men after humerus fractures)8 or in the Fracture Intervention Trial Population.41 These results need to be interpreted with caution for other reasons. First, these associations do not prove that these excess deaths were caused by the fractures and are preventable. Although the authors rigorously adjusted for confounders (including comorbid conditions, health status [measured by both the physical and mental components of the SF-36], bone mineral density, and demographic and lifestyle factors), there is likely to be residual confounding from both known and unknown confounders on the association between fracture and mortality (as is true of all observational studies). That said, residual confounding from both known and unknown confounders is likely to be present to a similar degree for the associations of mortality with hip, clinical vertebral, and NHNV fractures. Therefore, their conclusion that NHNV are associated with more mortality than hip or clinical vertebral fractures at the population level is sound. Moreover, it would seem unlikely that all of the excess deaths associated with fractures in this particular study are due to confounding, given the robust set of covariates the authors included in their analytic models. Third, lumping all NHNV fractures together obscures important differences between them; pelvis and proximal humerus fractures appear to have quite different associations with mortality than distal limb fractures. In the Andrich study, the excess mortality was highest in the first several weeks after pelvis fracture and no longer evident after 32 weeks, whereas the excess mortality in the study of Tran and colleagues after NHNV fractures (of which pelvis fractures were a small minority) remained steady and did not diminish over follow-up time. Older and more frail patients are both more likely to succumb early to the effects of a more serious fracture but also have higher pre-fracture background mortality rates. Hence, pelvis fracture patients who survive the early post-fracture period may not look much different from their older, more frail peers who did not fracture, where residual longer-term effects of the fracture are dwarfed by their high background mortality risk. Consistent with this, the causes of long-term death after fractures are similar to that found in the general older population.22 Hence, the finding of Tran and colleagues that even NHNV fractures may alter long-term mortality is at first glance puzzling. They hypothesize that long-term functional status changes caused by fractures even for individuals’ with more robust pre-fracture health may alter their subsequent health trajectories. A substantial proportion of those who do survive the early period after clinical vertebral42, 43 and especially hip38, 44 fractures do not recover their pre-fracture functional capacity. Less serious distal limb fractures may also unfavorably alter the long-term trajectory of functional capacity and health status (Fig. 1, arrow e). Edwards and colleagues showed that clinically significant declines in independent activities of daily living were more likely to occur in older women after a distal forearm fracture (over a mean follow-up time of 6.3 years) than in women who did not fracture after adjustment for age and pre-fracture comorbid conditions, neuromuscular function, and self-reported health status. This conceivably may lead to a negative cycle of reduced physical activity contributing to reductions of muscle strength, reduced appetite, and adverse changes in nutritional intake and status,45 and a downward trajectory over time that leads to incident frailty and disability. This further increases the risk of incident fractures7, 46-48 and other injurious falls (Fig. 1, arrow f) that can be a direct cause of mortality. Reduced muscle strength,49, 50 frailty,46-48 slower walk speed,51, 52 and disability53, 54 are associated with all-cause mortality; these individuals are more likely to develop fatal and non-fatal complications of surgical procedures55, 56 and to succumb to complications of atherosclerotic cardiovascular disease57-60 and cancer61, 62 that become more prevalent with advancing age. Hence, the altered functional status that can follow fractures may be both a mediator of the association of fracture with long-term mortality and an effect modifier of the association between chronic conditions (as they progress and become more common with advancing age) and mortality (Fig. 1, arrow h). If Tran and colleagues are correct, then some proportion of excess long-term mortality may be indirectly caused by fractures (through the pathways represented by arrows e-f-g, and e-h-c, Fig. 1). However, those aspects of the conceptual framework presented in Fig. 1 remain unproven. Further research studies are needed to better estimate what proportion of mortality occurs via these pathways. Large data sets with long follow-up and accurate ascertainment of both baseline comorbid conditions but also functional status physical performance capability and other covariates are required. The data sets will need to be particularly large if these pathways are to be estimated for different categories of fractures, as Tran and colleagues have done, and if analyses stratified for pre-fracture comorbidity and/or functional status are to be done. Many have proposed that this question may be addressed more directly by estimating the reduction in all-cause mortality within randomized controlled trials of pharmacologic fracture prevention agents versus placebo. However, each of these trials by themselves is not powered to test the hypothesis that mortality can be reduced through fracture prevention; for example, if (as estimated by Tran and colleagues among postmenopausal women) 11% of all mortality is associated with the fractures and 45% of incident fractures are prevented by treatment, then one can expect only a 5% reduction of mortality by preventing fractures. The 2-year post-hip fracture randomized trial of IV zoledronic acid versus placebo showed a significant 28% reduction of all-cause mortality,63 but post hoc analyses suggested that only 8% of the deaths were averted by preventing the fracture64 and that remaining proportion of the mortality reduction benefit occurred via other (currently unknown) mechanisms.65 Bolland and colleagues66 have used a meta-analysis of 10 pivotal randomized controlled trials of oral bisphosphonates,67-71 strontium ranelate,72, 73 iv zoledronic acid,63, 74 and denosumab75 to estimate the relative risk of all-cause mortality on pharmacologic fracture prevention medication to be 0.90, but the statistical significance of this was marginal. Investigating the causal pathways by which osteoporotic fractures may lead to preventable mortality is not simply an intellectually interesting exercise. Recent studies have documented persistent undertreatment of those with osteoporosis and recent osteoporotic fractures,76, 77 declining use of pharmacologic fracture prevention therapies78 and underappreciation by patients as to why a recent fracture signals that they are at high risk for subsequent fractures and other adverse events.79 It is important that we establish, as best as we can, the extent and mechanisms by which prevention of fractures may delay mortality. This will facilitate communicating cogent, convincing explanations (“connecting the dots”) to patients at high risk of fracture, their health care providers, health care delivery organizations, and health care payers regarding the short- and long-term adverse effects (including mortality) of fractures and why their prevention is important. The author have no conflicts of interest to disclose.