Fractures After Stroke
Frequency, Types, and Associations
Background and Purpose— Stroke patients may have an increased risk of fractures because of weak bones or an increased risk of falling. Our goal was to estimate the frequency of fracture after stroke and to identify those at greatest risk.
Methods— This study incorporated 2 complementary strategies: a prospective, single-center, cohort study and an analysis of Scottish routine hospital discharge data.
Results— Eighty-eight fractures (30% hip) occurred in 2696 hospital-referred stroke patients. The proportions sustaining any fracture or hip fracture within 2 years were 4% and 1.1%, respectively, 1.4 (95% CI, 0.92 to 2.07) times the rate of hip fracture in the general population (ie, observed number divided by expected number or standardized morbidity ratio). Female sex, older age, low abbreviated mental test score, and prestroke dependence were associated with an increased hip fracture rate. Routine data identified 129 935 acute stroke patients admitted to Scottish hospitals. During 363 447 patient-years, 4528 patients had hip fractures, 2.0% had fractures by 1 year, and 10.6% had fractures by 10 years. This is 1.7 times the rate of hip fracture in the general population and 2.3 times that in patients with myocardial infarction. Older patients predictably had the highest rate of poststroke hip fractures but a lower standardized morbidity ratio than younger patients.
Conclusions— Fractures after stroke are probably frequent and serious enough to justify the development of preventive strategies, but the modest event rate would mean that randomized, controlled trials to test these strategies specifically in stroke patients would need to enroll thousands of patients.
A fracture may be catastrophic for a patient already disabled by stroke. In 1957, Peszczynski1 found that 28 of 150 patients (18.7%) with hip fractures had had prior hemiplegic stroke. Most fractures occurred on the affected side. He thought that mental confusion, postural instability, and accelerated osteoporosis in the hemiplegic limb might contribute to the risk of fractures, although he had few data to confirm these views.
Studies of various types have since reported associations between strokes and hip fractures. These include retrospective case-control studies comparing characteristics of hip fracture patients with age- and sex-matched control subjects,2–12⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓ prospective cohort studies of population samples followed up to identify subsequent hip fractures,13–16⇓⇓⇓ and cross-sectional studies in which questionnaires were sent to large numbers of people to ascertain how often stroke and hip fractures occur in the same individual.17,18⇓ These suggest, with only 1 exception,14 that stroke is a risk factor for subsequent hip fracture, with most studies reporting risk (or odds) ratios of between 1.5 and 4. This variation could be due to differences in the populations studied (age, sex, and ethnicity), the methods used, and chance because most included few patients with prior stroke.
Interventions to reduce falls or prevent weakening of bones might reduce the frequency of fracture after stroke. Before introducing such interventions after stroke or planning randomized trials to test them, we need to establish whether the fracture rate justifies such rigorous efforts and find data to predict their impact. In this study, we aimed to establish the absolute rate of fracture after stroke and compare that rate with the rate in the general population (ie, the standardized morbidity ratio); the bony sites affected, because this determines the impact of the fracture; the timing of fractures; and the factors identifying patients at greatest risk of fracture, which might help us focus any preventive treatment on those patients most likely to benefit.
Patients and Methods
We used 2 complementary research strategies.
We followed up stroke patients admitted to our urban teaching hospital between 1990 and 1998. Between 1994 and 1998, we also included all stroke patients referred to our neurovascular outpatient clinic. Their assessment included clinical details, Oxfordshire Community Stroke Project stroke classification,19 prestroke functional status on the modified Rankin Scale,20 conscious level (Glasgow Coma Scale21), motor function in the arm and leg (Medical Research Council grading22), and mental status [abbreviated Mental Test Score (AMT23)]. Patients with subarachnoid hemorrhage were excluded.
We followed up patients at 6 months and 1 and 2 years by postal or telephone questionnaire. We asked surviving patients or caregivers whether the patient had sustained a fracture. If so, we reviewed all available medical records relating to the fracture. In addition, to identify forgotten or unreported fractures, we reviewed the general practitioner records of all patients who died during follow-up and obtained routinely collected diagnostic data relating to any subsequent hospital stay in Scotland. Fractures resulting from neoplastic bony infiltration were excluded from the analyses.
To calculate the standardized morbidity ratio (observed number divided by expected number), we used an estimate of the mean age- and sex-specific incidence for fractured hip for Scotland from 1990 to 1996. The numerator for the incidence was derived from routinely collected hospital discharge data for hip fractures in Scotland provided by the Information and Statistics Division. We identified the patients with a hospital stay in each year in which at least 1 of the included Scottish Morbidity Records included a hip fracture code [International Classification of Disease (ICD)-9 820, ICD-10 S720-S722)] in the first (of 6 possible) diagnostic position. Estimates were very similar when we included the numbers of hip fractures based on diagnoses in any diagnostic position on the Scottish Morbidity Records but counted only the first occurrence in the study period. Age- and sex-specific incidence based on these numerators was calculated using population denominators derived from midyear estimates from the most relevant census.
We used linked Scottish data from Information and Statistics Division to estimate the risk of admission with a hip fracture (ICD-9 820, ICD-10 S720-S722) in patients with prior admission with a stroke (ICD-9 431*, 4329, 434*, 436*, 437*, and ICD-10 I161*, I6219, I6229, I63*, I64*, I670, I672, I675, I677, I678, I679) from April 1, 1981, until March 31, 1997. The follow-up period finished on March 31, 1998.
The hip fracture rate among patients with a previous stroke diagnosis was compared with that among patients first admitted with acute myocardial infarction (MI; ICD-9 410; ICD-10 I21, I22). We also calculated the number of hip fractures we would have expected in the stroke and MI groups if they had an incidence similar to that of the overall population (see strategy 1). For strategy 2, we used the mean hip fracture incidence for the period in which each patient was in the study to calculate the expected risk of hip fracture.
For both strategies, standardized morbidity ratios (observed divided by expected) were calculated by use of a subject-year approach,24 and 95% CIs were calculated with Confidence Interval Analysis.25 Other analyses were performed with S-PLUS for Windows.26 S-PLUS regression analysis and Cox proportional-hazards modeling were used to identify those factors independently associated with subsequent fractures.
Strategy 1: A Hospital-Based Cohort Study
We identified 2696 stroke patients (mean age, 68 years; 53% male). The patient characteristics are shown in Table 1. Eighty-one patients suffered 88 fractures during 3966 patient-years of follow-up. Their characteristics are also shown in Table 1; the side, site, and details of their fractures are given in Table 2. Sixty-nine percent of fractures affected the side of the body contralateral to any lateralized brain lesion.
The proportion of patients having any fracture within 2 years of a stroke was 4% (95% CI, 3 to 5), a rate of 22 per 1000 patient-years. The proportion of patients having hip fractures was 1.1% (95% CI, 0.7 to 2.0), a rate of 7 per 1000 patient-years. The overall standardized morbidity ratio for hip fractures was 1.41 (95% CI, 0.92 to 2.07). These data for each sex and age group are shown in Table 3. The ratio decreased with increasing age.
The associations between selected baseline variables and the risk of subsequent fractures are shown in Table 4. The variables identified as being independently associated with a greater rate of any fracture were female sex, older age, and an AMT <8 of 10, whereas having a prestroke modified Rankin score <3 was protective. Although based on rather few events, similar variables were independently associated with hip fracture—ie, older age, an AMT <8, and a prestroke modified Rankin score <3. Smoking history, alcohol consumption, prior MI, atrial fibrillation, peripheral vascular disease, diabetes, epilepsy, and side or pathological type (hemorrhagic versus ischemic) of brain lesion were not significantly associated (P>0.05) with fracture rate, although we could easily have missed associations because of the limited power of our study.
Strategy 2: National Hospital Discharge Data
During the study period, 129 935 people (mean age, 72.9 years; 44% male) were admitted to a Scottish hospital with an acute stroke. By March 31, 1998, 7999 had been admitted with a subsequent fracture, of which 4528 (59%) had hip fractures, a rate of 12.5 hip fractures per 1000 patient-years. The proportion with hip fracture was 2.0% (95% CI, 1.87 to 2.07) in the first year but fell in subsequent years, so that the cumulative proportion was 10.6% (95% CI, 10.23 to 10.99) by 10 years (Table 5).
Over the same period, 183 155 people (mean age, 66.8 years; 59% male) were admitted with an MI. Of these, 6552 were subsequently admitted with any fracture. Of these, 2699 (41%) were hip fractures, a rate of 3.2 hip fractures per 1000 patient-years. The proportion with hip fracture was 0.36% (95% CI, 0.33 to 0.39) in the first year and was similar in subsequent years, with a cumulative proportion of only 3.1 (95% CI, 2.98 to 3.26) over 10 years. The proportions with hip fracture after stroke and MI are compared in the Figure. Patients with stroke had a greater admission rate with a hip fracture than those with MI after adjustment for both age and sex (hazard ratio, 2.25; 95% CI, 1.77 to 1.90).
Among stroke patients, female sex (hazard ratio, 1.90; 95% CI, 1.78 to 2.03) and increasing age (hazard ratio, 1.05 per year; 95% CI, 1.04 to 1.05) were associated with an increased hip fracture rate. Among those with a first admission as a result of stroke or MI, the overall standardized morbidity ratios were 1.68 (95% CI, 1.64 to 1.73) and 0.85 (95% 0.82 to 0.88), respectively. The age- and sex-specific ratios for patients with stroke are shown in Table 3 and, as in strategy 1, decreased with increasing age.
By combining 2 research strategies, each with its own strengths and weaknesses, we hoped to obtain more reliable information concerning the epidemiology of poststroke fractures.
The advantage of strategy 1 was that we had detailed and accurate baseline data and prospective follow-up, could identify fractures that did not lead to hospital admission, and could examine several variables that might be associated with fracture rate. In our univariate analyses, increasing age, female sex, prestroke dependency (modified Rankin score >2), an AMT <8, and inability to sit, stand, or walk were associated with a significantly greater risk of fractures (Table 4). Some, ie, increasing age and female sex, are well-recognized risk factors for both osteoporosis and fractures. One might hypothesize that prestroke dependency might be associated with prior immobility and poor nutrition with consequent osteoporosis or osteomalacia, or it might simply be a marker of an increased propensity to falls. Mental confusion and poor truncal control (as indicated by patient’s inability to sit, stand, or walk) are both established risk factors for falling. Our multivariate analyses showed that increasing age, female sex, prestroke Rankin score >2, and an AMT<8 are associated with higher fracture rates. The other factors identified in the univariate analysis dropped out of the model, presumably because of their strong correlations with these 4 factors.
Only 1 previous study27 has used a similar research strategy, although the follow-up in that study was not prospective but relied on local orthopedic department records and did not report baseline factors associated with increased fracture rates. They identified 120 patients with 154 fractures (70 hip fractures, or 45%) among 1139 hospital-admitted stroke patients (mean age, 73 years; 55% male) during 4132 patient-years in northern Sweden, equivalent to rates of 36 per 1000 patient-years for any fracture and 17 per 1000 patient-years for hip fractures. The proportions with any fracture were 4% in the first year, 15% by year 5, and 24% by year 10. The age-specific morbidity ratios for hip fracture were 3.8 for those <69 years of age, 3.0 for those 70 to 79 years of age, and 2.1 for those >80 years of age. Our total fracture rate (22 per 1000 patient-years) was lower than that in the Swedish study,27 and our hip fracture rate (7 per 1000 patient-years) was much lower than that using strategy 2 (12.5 per 1000 patient-years) and that in the previous study (17 per 1000 patient-years).27 Such differences might be attributable to environmental factors (eg, icier conditions) or differences in activity levels, rehabilitation schemes, or application of fracture prevention techniques. However, differences in case selection and fracture detection could easily explain the lower fracture rates in our patients. For instance, our hospital referred patients’ mean age was only 68 years compared with 73 years in unselected cohorts and the Swedish study.27 A younger age would be expected to be associated with a lower frequency of fractures given that we and others have shown that the frequency of poststroke fractures rises with age. In addition, 45% of our patients were not admitted (because our neurovascular outpatient clinic serves a much larger population than our inpatient stroke service) and probably had milder strokes with a lower risk of fracture. Of course, our low estimates of hip fracture rate might have been due to underreporting of hip fractures. This seems unlikely given our prospective follow-up and checks using routine hospital discharge data. Additionally, in our study, hip fractures formed a smaller proportion of fractures (30%) than in the Swedish study (45%). It seems unlikely that we would have missed hip fractures, which usually lead to hospital admission, but identified a greater proportion of less serious fractures, which are less likely to lead to hospital admission. We were much more likely to have missed rib and vertebral fractures, which would increase the apparent proportion of hip fractures. In both our study and that of Ramnemark et al,27 despite several thousand patient-years of follow-up, our estimates of hip fracture rate and that relative to the rate in the general population were relatively imprecise, and some of the differences in frequency and type of fracture could be due to chance. Both studies identified that the morbidity ratio fell with increasing age, although this was more marked in our study.
Strategy 2 had the advantage of identifying very large numbers of stroke (and MI) patients with prolonged follow-up so that any estimates of fracture rate and standardized morbidity ratios were more precise. The disadvantages are that an unknown proportion of diagnoses (stroke, MI, and fracture) were wrong because of miscoding and other errors and that it included only hospital-admitted cases. Previous studies have indicated some inaccuracies in routine coding of hip fractures in Scotland.28,29⇓ One estimated a false-positive error rate of 6%,28 and the other estimated that routine coding in 1 region may underestimate the number of fractures by 40%.29 However, both of these studies were based on data from the early 1980s when little use was made of routine data, and there have been rigorous efforts to improve the accuracy of coding in the subsequent 2 decades. Another limitation is the paucity of information about individual patient characteristics; thus, we could not examine their effects on fracture rate.
One other study from Sweden30 used national hospital discharge data to detect admissions with hip fracture in a cohort of 273 288 stroke patients (mean age, 73.5 years; 50% male) over a mean follow-up period of 2.5 years. This demonstrated an overall hip fracture rate of 19.8 per 1000 patient-years. Their rates were higher in women and highest in the first year after stroke. Although the hip fracture rate increased with age, their morbidity ratios fell with increasing age but remained greater than unity at all ages. Our study found similar patterns but lower fracture rates. Contrary to our findings, their morbidity ratio was higher in women than men.
It is intriguing to speculate why we have found lower poststroke fracture rates in Scotland than in Sweden. Interestingly, on the basis of routine hospital data, the age-standardized (based on the 1992 Scottish population) incidence of hip fracture in the general population in Sweden was 96.8 per 100 000 men and 262.9 per 100 000 women compared with 59.6 and 190.5 per 100 000, respectively, in Scotland. It is unclear whether differences in hip fracture rates, either in stroke patients or in the general population, result from differences in study methodology (hospitalization rates and coding practice), genetic factors, risk factor prevalence (body mass index, physical activity, diet, or sunlight exposure), environmental factors (severe winters), or treatments (rehabilitation, drugs). Despite their differences, all the studies have confirmed that the hip fracture rate after stroke was highest in women, in older patients, and in the first year. They also showed that the standardized morbidity ratio fell with increasing age. We have also shown that the hip fracture rate among patients with MI is not only lower than that after stroke but also lower than in the general population. One explanation might be that although obesity protects against hip fracture, it is associated with a greater risk of MI.
What might be the implications for developing prevention strategies? Should we be intervening to reduce our stroke patient’s fracture risk? Given the impact of hip fracture and its frequency after stroke, it seems reasonable to attempt to prevent this complication. Because the hip fracture rate after stroke is modest (6% to 15% over the first 5 years), randomized trials to establish that interventions are effective and worthwhile are likely to have to recruit several thousand patients and follow them up for years. Smaller numbers would be required if entry criteria focused recruitment on higher-risk patients—eg, elderly women—or all fractures. However, the hip fracture rate after stroke compared with that in the general population is highest in younger patients (<69 years of age) and decreases with age. If the standardized morbidity ratio reflects the strength of the association between stroke and hip fracture (in a similar way to an odds or risk ratio) in case-control or cohort studies, prior stroke may be a more powerful risk factor in younger patients. The reason might be that stroke-specific mechanisms (ie, increased risk of falls and hemi-osteoporosis) add relatively little to the already high background risk in older patients who are prone to falls (even without a stroke) and have generalized osteoporosis. Therefore, stroke may particularly increase the risk of hip fractures in younger patients who are at lower absolute risk. Any differential effect of treatments on stroke patients compared with other high-risk groups is likely to be in those for whom the morbidity ratio is highest and stroke-specific mechanisms (eg, hemi-osteoporosis) are most important.
Stroke patients are at increased risk of hip fracture, and it seems reasonable to take steps to minimize this complication. We should consider introducing measures to prevent falls,31 using hip protectors in those who fall frequently,32 and using drugs to treat osteoporosis and osteomalacia.33–35⇓⇓ The cost-effectiveness of applying preventative strategies routinely to all stroke patients needs to be established.
This study was funded by Chest Heart and Stroke Scotland. We would like to thank Stephanie Lewis for her invaluable statistical advice; Charles Warlow for helpful comments on the draft manuscripts; Dr Anna Ramnemark for sending us background information; all those who have contributed to the Lothian Stroke Register; our patients who have completed follow-up questionnaires; and Steve Kendrick, Rod Muir, and Jillian Campbell at the Information and Statistics Division for providing us with original data.
- Received October 9, 2001.
- Revision received November 20, 2001.
- Accepted November 26, 2001.
- ↵Cooper C, Barker DJP, Wickham C. Physical activity, muscle strength, and calcium intake in fracture of the proximal femur in Britain. BMJ. 1988; 297: 1443–1446.
- ↵Heidrich FE, Stergachis A, Gross KM. Diuretic drug use and the risk for hip fracture. Ann Intern Med. 1991; 115: 1–6.
- ↵Grisso JA, Kelsey JL, O’Brien LA, Miles CG, Sidney S, Maislin G, La Pann K, Moritz D, Peters B. Risk of hip fracture in men. Am J Epidemiol. 1997; 145: 786–793.
- ↵Kanis J, Johnell O, Gullberg B, Allander E, Elffors L, Ranstam J, Dequeker J, Dilsen G, Gennari C, Lopez Vaz A, Lyritis G, Mazzuoli G, Miravet L, Passeri M, Perez Cano R, Rapado A, Ribot C. Risk factors for hip fracture in men from Southern Europe: THE MEDOS Study. Osteoporos Int. 1999; 9: 45–54.
- ↵Farahmand BY, Persson P, Michaelsson K, Baron JA, Alberts A, Moradi T, Ljunghall S,. Physical activity and hip fracture: a population-based case-control study: Swedish Hip Fracture Study Group. Int J Epidemiol. 2000; 29: 308–314.
- ↵Guo Z, Wills P, Viitanen M, Fastbom J, Winblad B. Cognitive impairment, drug use, and the risk of hip fracture in persons over 75 years old: a community-based prospective study. Am J Epidemiol. 1998; 148: 887–892.
- ↵Meyer HE, Tverdal A, Falch JA. Risk factors for hip fracture in middle-aged Norwegian women and men. Am J Epidemiol. 1993; 137: 1203–1211.
- ↵Wolinsky FD, Fitzgerald JF. The risk of hip fracture among non institutionalized older adults. J Gerontol. 1994; 49: S165–S175.
- ↵Medical Research Council. Aids to the Examination of the Peripheral Nervous System. Grimsby, UK: Albert Gait Ltd, Castle Press; 1982: 1–60.
- ↵Hodkinson HM. Evaluation of a mental test score for assessment of mental impairment in the elderly. Age Ageing. 1972; 1: 233–238.
- ↵Armitage P, Berry G. The subject-years method.In: Statistical Methods in Medical Research. 2nd ed. Oxford, UK: Blackwell Scientific Publications; 1987: 438–439.
- ↵Gardner SB, Winter PD, Gardner MJ. Confidence Interval Analysis, version 1.0 BMJ. London, UK; 1989.
- ↵S-PLUS. Professional release. Mathsoft, Inc; 2000.
- ↵Currie AL, Reid DM, Brown N, Nuki G. An epidemiological study of fracture of the neck of femur. Health Bull. 1986; 44: 143–148.
- ↵Kanis J, Oden A, Johnell O. Acute and long-term increase in fracture risk after hospitalization for stroke. Stroke. 2001; 32: 702–706.
- ↵Gillespie LD, Gillespie WJ, Cumming R, Lamb SE, Rowe BH. Interventions for preventing falls in the elderly (Cochrane Review).In: The Cochrane Library, Issue 1. Oxford, UK: Update Software; 2001.
- ↵Parker MJ, Gillespie LD, Gillespie WJ. Hip protectors for preventing hip fractures in the elderly (Cochrane Review).In: The Cochrane Library, Issue 1. Oxford, UK: Update Software; 2001.
- ↵Gillespie WJ, Avenell A, Henry DA, O’Connell DL, Robertson J. Vitamin D and vitamin D analogues for preventing fractures associated with involutional and post menopausal osteoporosis (Cochrane Review).In: The Cochrane Library, Issue 1. Oxford, UK: Update Software; 2001.
- ↵Black DM, Cummings SR, Karpf DB, Cauley JA, Thompson DE, Nevitt MC, Bauer DC, Genant HK, Haskell WL, Marcus R, Ott SM, Torner JC, Quandt SA, Reiss TF, Ensrud KE. Randomised trial of effect of alendronate on the risk of fracture in women with existing vertebral fractures. Lancet. 1996; 348: 1535–1541.