Gender Differences in Mortality After Hip Fracture: The Role of Infection†‡
The authors have no conflict of interest.
Portions of this report were presented orally at the World Congress on Osteoporosis 2000, Chicago, IL, June 15–18, 2000 at a plenary session of the 23rd Annual Scientific Meeting of the American Society for Bone and Mineral Research, Phoenix, Arizona, October 12–16, 2001, and as a plenary poster at the 22nd Annual Scientific Meeting of the American Society for Bone and Mineral Research, Toronto, Ontario, Canada, September 22–26, 2000.
Possible explanations for the observed gender difference in mortality after hip fracture were examined in a cohort of 804 men and women. Mortality during 2 years after fracture was identified from death certificates. Men were twice as likely as women to die, and deaths caused by pneumonia/influenza and septicemia showed the greatest increase.
Introduction: Men are more likely to die after hip fracture than women. Gender differences in predisposing factors and causes of death have not been systematically studied.
Materials and Methods: Participants (173 men and 631 women) in the Baltimore Hip Studies cohort enrolled in 1990 and 1991, at the time of hospitalization for hip fracture, were followed longitudinally for 2 years. Cause-specific mortality 1 and 2 years after hip fracture, identified from death certificates, was compared by gender and to population rates.
Results and Conclusions: Men were twice as likely as women to die during the first and second years after hip fracture (odds ratio [OR], 2.28; 95% CI, 1.47, 3.54 and OR, 2.21; 95% CI, 1.48, 3.31). Prefracture medical comorbidity, type of fracture, type of surgical procedure, and postoperative complications did not explain the observed difference. Greatest increases in mortality, relative to the general population, were seen for septicemia (relative risk [RR], 87.9; 95% CI, 16.5, 175 at 1 year and RR, 32.0; 95% CI, 7.99, 127 at 2 years) and pneumonia (RR, 23.8; 95% CI, 12.8, 44.2 at 1 year and RR, 10.4; 95% CI, 3.35, 32.2 at 2 years). The magnitude of increase in deaths caused by infection was greater for men than for women in both years. Mortality rates for men and women were similar if deaths caused by infection were excluded (3.46 [1.79, 6.67] and 2.47 [1.63, 3.72] at 1 year and 0.96 [0.48, 1.91] and 1.26 [0.80, 1.98] at 2 years). Deaths related to infections (pneumonia, influenza, and septicemia) seem to be largely responsible for the observed gender difference. In conclusion, an increased rate of death from infection and a gender difference in rates persists for at least 2 years after the fracture.
ALTHOUGH OSTEOPOROSIS IS often considered to be a disease of older white women, 9–15 million men in the United States are estimated to have either low bone mass (“osteopenia”) or osteoporosis,(1) and men sustain 20–30% of all hip fractures.(2,3) Secular increases in hip fracture have been observed in men since early in the 20th century and have been most rapid in the last three decades.(3,4) In conjunction with anticipated continuing increases in life expectancy, this trend leads to the projection that, by 2050, approximately 200,000 and 1.8 million hip fractures will occur annually among men in the United States and worldwide, respectively.(3,5,6)
The most serious of the consequences of hip fracture is increased mortality. In women, mortality during the year after fracture has been estimated to increase by 12–20%.(7) Although much of the increase occurs during the first year after the fracture, the effect may persist for several years.(8) Both short- and long-term mortality in men are about 2-fold higher than those seen in women.(9–19) The reasons for the gender difference in mortality are unclear, although the general belief seems to have been that men are sicker and frailer than women at the time of fracture.(20) When all hip fracture patients are considered collectively, both pre-existing comorbidity and poor functional status before the fracture seem to affect mortality.(14,21–23)
Gender- and cause-specific mortality after hip fracture has not been extensively investigated.(13,20,24–26) Gender differences may result from differences in pre-existing illness, from different proportional increases in overall mortality, or from differential increases in certain causes of death. Whether hip fracture is associated with disproportionate increases in some of these and whether men and women show similar patterns has not been systematically studied.
In this report, we compared the prefracture status of men and women who had fractured and assessed their postfracture mortality. Specifically, we asked whether, after controlling for prefracture status, there is a gender difference in overall and cause-specific mortality within the first 2 years after fracture. However, a gender difference in mortality after hip fracture may have nothing to do with the fracture itself but may merely reflect the gender difference in mortality in the population. Consequently, we also compared mortality among men and women in this study to expected age-specific mortality, based on vital statistics, to assess the extent to which observed mortality after hip fracture differs from population rates.
MATERIALS AND METHODS
The study population consisted of 804 community-dwelling men and women, aged 65 years and older, who were admitted to one of eight metropolitan Baltimore hospitals with acute hip fractures from 1 January 1990 through 15 June 1991. The sample represented approximately 50% of all incident hip fractures in Baltimore during this time period. A more detailed description of the population and methods for enrolling patients has previously been published.(27) All study protocols were approved by Institutional Review Boards of the University of Maryland and participating hospitals, and all study participants or proxies provided informed consent.
Patients or proxies were evaluated during hospitalization (usually within 5–10 days of surgery) and at 2, 6, 12, 18, and 24 months after the fracture, using structured questionnaires and performance measures. Proxy respondents were used when patients were unable to provide responses (Mini-Mental State Examination score ≤ 16); proxies were most often spouses or caregivers of the patients. The reliability of proxy responses for key functional status measures has been demonstrated previously in a subset of the study population.(28) The in-hospital interview was used to collect information about the patient's physical and functional status before the fracture and to determine the patient's current cognitive and affective condition. Medical chart abstraction was used to collect information relating to the hospital course and to comorbid conditions. Follow-up interviews were conducted at the patient's residence.
Vital status was ascertained through follow-up interviews until 2 years after fracture. In the case of death of the participant, interviews were conducted with next of kin. All deaths were verified by death certificates. Date and cause of death information were obtained from death certificates. Ascertainment of vital status at 24 months was 99.7% complete. The 10 most common causes of death among elderly men and women were included in the analysis. They are (with the corresponding International Classification of Diseases, 1975, ninth revision, or ICD-9, codes): diseases of the heart (390–398, 402, 404–429), malignant neoplasms (140–208), cerebrovascular disease (430–438), chronic obstructive pulmonary diseases (490–496), pneumonia and influenza (480–487), diabetes mellitus (250), accidents and adverse events (E800-E807, E826-E949), renal disease including nephritis, nephrosis, and nephrotic syndrome (580–589), Alzheimer's disease (331.0), and septicemia (038).(29) Assignment of cause of death was based on a predetermined algorithm developed by an expert panel of physicians and researchers with prior experience in analysis of mortality. Initial determinations were made by one investigator (LEW), with review and adjudication of assigned cause of death by members of the panel.
Independent variables that were investigated included demographic characteristics, prefracture health and functional status, hospital treatment and course, and discharge disposition. Demographic characteristics included age, race, and marital status.
Prefracture functional status (during the 2 weeks before the fracture) was assessed for both physical and instrumental activities of daily living (PADL and IADL, respectively). PADL functioning was assessed by questions regarding 16 tasks related to both global and lower extremity activities. For each task, respondents indicated whether they carried out the activity independently, with assistance (human, equipment, or both) or, if they did not perform the task, whether this was for a health- or a non-health-related reason. Items included were walking 10 ft or across a room; walking one block on a level sidewalk; climbing five stairs; getting into a car; getting into and out of bed; rising from an armless chair; putting on a shirt or blouse; buttoning a shirt or blouse; putting on pants; putting socks and shoes on both feet; getting into and out of the bath or shower; taking a shower, bath, or sponge bath; getting on and off the toilet; grooming self (brushing hair, teeth); and reaching for an item on the ground. Performance of seven IADLs was assessed using a modified version of the Older Americans Resources and Services Instrument.(30) Items included were using the telephone; getting to places out of walking distance; shopping for groceries or clothes; preparing meals; housecleaning; handling money; and taking medication. For each task, response options included performing the function independently, needing partial assistance, or being completely unable to do so. The Center for Epidemiologic Studies Depression Scale (CES-D) was used to assess depression; a score of 16 or higher is indicative of depression.
Medical chart review was used to identify a history of any of 29 conditions, including anemia, angina, arrhythmias, arthritis, cancer, cirrhosis/alcohol abuse, congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD)/emphysema/asthma/bronchitis, deep venous thrombosis (DVT), dementia/organic brain syndrome/Alzheimer's disease/senility, diabetes (DM), dizziness/balance problems, drug abuse, epilepsy/seizures, falls, gastrointestinal (GI) disorders, previous hip fracture, hypertension, mental retardation, missing limbs, myocardial infarction (MI), obesity, osteoporosis, Paget's disease of bone, Parkinson's disease, peripheral vascular disease, rheumatoid arthritis (RA)/systemic lupus erythematosus (SLE)/scleroderma, stroke, transient ischemic attack (TIA), and ulcer. Prevalence of each of the conditions among men and women was examined individually, and the number of recorded conditions was summed for aggregate analysis.
Hospital course and treatment included type of fracture, surgical procedure, pre- and postoperative laboratory studies, anesthesiologist's risk assessment (ASA score), time to surgery, the occurrence of postoperative complications, and length of stay (LOS). Laboratory tests studied included complete blood count (CBC), sodium, potassium, chloride, blood urea nitrogen (BUN), glucose, uric acid, phosphate, calcium, total protein, albumin, creatinine, carbon dioxide, creatinine phosphokinase (CPK), alkaline phosphatase, alanine aminotransferase (ALT or SGPT), aspartate aminotransferase (AST or SGOT), chest X-ray (CXR), and electrocardiogram (ECG). The ASA score, assigned by the anesthesiologist before surgery, is an assessment of fitness for surgery and ranges from I (normal, healthy patient) to V (moribund patient, not expected to live 24 h, independently of surgery). Postoperative complications assessed included bowel obstruction/impaction, confusion, CHF, dislocation of hip, gastrointestinal (GI) bleeding, prolonged headache, MI, pneumonia, pressure ulcers, pulmonary edema, pulmonary embolism (PE), renal failure, reoperation of fractured hip, stroke, deep vein thrombosis (DVT), urinary tract infection (UTI), wound hematoma/bleeding, and wound infection. Each of these complications was examined individually, and events were summed for each patient. Length of stay, in days, was used both as a continuous variable and as a binary variable, dichotomized at the median.
For comparison of patient characteristics, laboratory values, and crude mortality rates, the statistical significance of gender differences in means was evaluated by t-tests and of differences in proportions by χ2 tests. A two-step process was used to evaluate the association of potential predictors and mortality. First, the association of each variable with mortality was evaluated, and variables that were significant at p ≤ 0.10 or were known risk factors for mortality in hip fracture patients were retained for use in subsequent modeling in adjusted logistic regression. Separate logistic regression models were developed to describe mortality at 12 and 24 months after fracture. All factors that had been identified in bivariable analysis were entered into the model and then deleted one at a time in order of least statistical significance to achieve a parsimonious model of associations with mortality. One-year age-specific probabilities of death were obtained from life tables,(31) and the age distribution of study participants was used to calculate expected mortality. Causes of death and cause-specific death rates per 100,000 for men and women ages 65 and older in the general population were obtained from reports of the National Center for Health Statistics.(29) Observed rates of death for each cause (per 100,000) within the study cohort were calculated to get a risk ratio as determined by the observed rate divided by the expected rate. 95% CIs around these ratios were then calculated.(32)
Table 1 shows the characteristics of the study population. The population included 804 fracture patients, 79% of whom were women. Men were significantly younger than women (79.5 versus 81.6 years, p < 0.001); ages ranged from 65 to 95 years for men and from 65 to 104 years for women. Men were significantly more likely to have comorbid conditions at the time of the fracture, with an average of 3.6 compared with 3.2 for women (p = 0.01). Men had significantly longer lengths of stay than women (17.2 versus 14.3 days, p = 0.002) and were significantly less likely to experience postoperative complications (1.4 versus 3.2, p < 0.001). There were no significant differences by gender in the ability to perform ADLs or IADLs before the fracture, type of fracture (intertrochanteric or subcapital), type of surgical repair, or time to surgery.
Table Table 1. Characteristics of Study Participants
Mortality after fracture, as shown in Table 2, was significantly different for men and women. Men were less likely to survive 12 and 24 months after fracture (p = 0.001). Compared with age-specific death rates in the general population, both men and women with hip fractures experienced an increased risk of death during both the first and second years after the fracture. Overall, men were 3.53 (95% CI, 2.27, 5.48) times more likely to die within 2 years after hip fracture than were men of the same age from the general population who had not sustained hip fracture; by comparison, women who had fractured were 2.43 (95% CI, 1.84, 3.22) times more likely to die during this period. Risk was greatest during the first year after fracture (for men, relative risk [RR] = 5.19; 95% CI, 2.77, 9.74; for women, RR = 3.24; 95% CI, 2.18, 4.82); however, increased risk seemed to persist through the second year (for men, RR = 1.31; 95% CI, 0.69, 2.47; for women, RR = 1.55; 95% CI, 1.01, 2.38). Even after adjusting for age, baseline comorbidity, ASA score, chest X-ray abnormalities, postoperative complications, and length of stay, men were significantly more likely to be dead at either time (p = 0.0002 and p = 0.0001, respectively; Table 3).
Table Table 2. Observed and Expected Mortality (Numbers of Deaths) Among Men and Women After Hip Fracture
Table Table 3. Risk Factors for Mortality After Hip Fracture; Multivariable Logistic Regression, With Odds Simultaneously Adjusted for all Covariates Shown
Table 4 shows rate ratios for the most common age-specific causes of death among men and women aged 65 and older, comparing rates for the general population to observed rates among persons who have sustained hip fracture. Fewer than 5% of death certificates listed hip fracture as an immediate, underlying, or contributory cause of death (data not shown). Although mortality is higher for each cause among men who have sustained hip fractures, the extent of increased mortality varies, with greatest and most persistent increases seen for infectious causes, including pneumonia and influenza (RR, 23.8; 95% CI, 12.8, 44.2 at year 1 and 10.3; 95% CI, 3.35, 32.2 at year 2) and septicemia (RR, 87.9; 95% CI, 16.5, 175 at year 1 and 32.0; 95% CI, 7.99, 128 at year 2). Among women, although increases in deaths from infection were greatest, the magnitude was somewhat smaller than seen in men: RR 10.4 (3.35, 32.2) and 4.00 (3.46, 4.62) for pneumonia and influenza and 32.0 (7.99, 128) and 13.3 (10.2, 17.5) for septicemia. As shown in Table 4, excluding deaths caused by pneumonia, influenza, and septicemia markedly attenuated the mortality difference between men and women.
Table Table 4. Rate Ratio of Observed to Expected Deaths Among Men and Women After Hip Fracture (95% CI)
In this series of hip fractures among community-dwelling elders, men were approximately twice as likely to die as women; by 2 years after fracture, 42% of men had died compared with 23% of women. This observation confirms previous reports of a greater risk of mortality experienced by men after hip fracture.(11–17,19) Increased likelihood of death was seen, although men were younger and reported similar functional status to women before the fracture. During the first year after fracture, mortality caused by each of the most common causes of death was significantly increased for men; women experienced significant increases in all categories except cancer and diabetes. The magnitude of increase for each cause was generally larger for men than women and ranged from 3- to 88-fold. By the second year after fracture, increases were much smaller, and in many cases, not significantly different from men in the general population. This suggests that men were in more precarious health at the time of fracture; therefore, their limited physiologic reserves were overstretched by the fracture experience, and they succumbed to pre-existing illness. By the second year, it could be hypothesized, as a consequence of death of those “at risk” individuals, that the remaining men were much more similar to men in the general population. Deaths caused by infection, however, remain substantially increased.
Although somewhat surprising, the younger age of men who sustain hip fractures has been consistently identified.(9,14,33–35) The mortality disadvantage persists for at least 2 years after the fracture event, a finding that is again consistent with many previous reports.(14–19,36–38) In addition, functional recovery across several domains of physical and instrumental function was slightly but not significantly poorer in men (WG Hawkes, LE Wehren, JR Idebel, DL Orwig, J Magaziner, unpublished data, 2003), although in an earlier study of functional recovery, men were significantly more likely than women to regain walking ability by 12 months after fracture.(39) Comorbidity has consistently been shown to predict mortality after hip fracture.(14,22,36) It has been suggested that greater prefracture comorbidity among men may be largely responsible for the observed gender differences,(36,40) and the men in our study did, on average, have more comorbid disease burden than did the women. However, inclusion of comorbidity as a predictor of mortality had no material effect on the observed gender difference, and only at very high numbers of conditions did it show an independent effect on overall mortality at either 1 or 2 years after fracture. No individual comorbid illness, in separate analyses, affected the mortality any differently. This is consistent with observations from Medicare discharge data, in which the number of comorbid diseases listed had no effect on differences in mortality by race and gender.(14)
Our findings in relation to cause-specific mortality are new and suggestive of underlying processes that may be amenable to intervention. Although early mortality after hip fracture is associated with infections, including pneumonia and septicemia,(9) the persistence of these as causes of death throughout the entire 2-year follow-up period raises the possibility of a long-term or permanent effect of the fracture event on immune function and resistance to infection. Immune competence declines with aging, whether considered as T-cell, B-cell, cytokine, accessory cell, or natural killer cell activity.(41–45) This so-called “immunosenescence” is believed to contribute to the increased incidence of infection and cancer in the elderly.(45) Interrelationships among bone, hematopoiesis, and the immune system have been hypothesized, but research is still preliminary.(46) Another possible explanation is that exposure to pathogens will be greater in hip fracture patients than in the general population. To the extent that men and women are differentially exposed to different pathogens, this too could explain the observed gender difference. This would be the case if, for example, rates of re-hospitalization were different for men and women, if different types and rates of instrumentation were involved (e.g., catheterization), or if men and women had different durations of residence in rehabilitation or nursing home environments.
Mortality has been associated with specific immune dysfunction, including alterations in the ratio of CD8 to CD4 cells and poor T-cell proliferation,(47) and low total lymphocyte counts have been shown to predict post-hip fracture mortality,(48) as well as general mortality in men.(49) The hip fracture event may serve to accelerate the decline in immune function in susceptible persons, so that they remain at increased risk of death from infections. In our relatively small series of cases, infection-related deaths seem to explain much of the observed gender difference in mortality. The mechanism(s) responsible for this phenomenon are not clear, although others have also found that men have an increased risk of major infections after surgery.(50–52) This effect seems to be at least partially independent of testosterone levels.(51,53)
Although age-specific mortality rates among persons age 65 and older are higher for men than for women,(31) the fracture event is also associated with greater increases in mortality among men than women. This effect has been shown in the Medicare population in the United States(36) as well as in Scandinavian populations,(16,54) and in this study has been observed to persist for at least 2 years. Among elderly women, the time course and magnitude of the attributable risk of hip fracture for mortality depend on both baseline comorbidity and baseline functional capacity.(8) Among women with pre-existing functional impairment or medical comorbidity, there is a marked increase in short-term mortality, but this effect is largely dissipated by approximately 6 months after the fracture event. However, among women with no or minimal baseline impairment or comorbidity, the excess persists for up to 5 years.(8) This observation suggests that comparisons that use age and gender “matching” may be inadequate to completely assess the effects of hip fracture on mortality. Appropriate evaluation of the effect of hip fracture on mortality in men would, therefore, necessitate comparison with men who are equally frail or ill but who have not fractured. This type of comparison could help to separate effects of physiologic functional decline from those attributable to the acute traumatic event of the fracture and its consequences. Thus, although the differences in cause-specific mortality that we observed after hip fracture are suggestive, a more accurate interpretation of the role of hip fracture would require comparison with deaths among equally frail men who had not fractured.
This study has other important limitations. Because it is a case series, the results may not be representative or generalizable to other populations of hip fracture patients. Because comorbid illness was identified by review of the medical chart for the hip fracture hospitalization, severity of illness could not be quantified, and a weighted index could not be used. However, use of individual diseases in modeling, rather than the number of diseases present, did not materially affect the results. Because of the relatively small number of deaths, estimates of cause-specific mortality are necessarily imprecise, with wide CIs. Comparisons to general population experience are imperfect: contemporaneously accrued nonfracture controls of similar age, gender, and health status would permit more precise estimation of the effects of the hip fracture on mortality. Death certificates provide little detail about disease processes leading to death, such as organisms responsible for fatal infections, and hospital or other records relating to the final illness were not available. Serum specimens were not obtained from participants; therefore, immune function could not be assessed.
Within the limits of the study design, we have found a marked gender difference in mortality for at least 2 years after hip fracture that is independent of age, prefracture medical condition, and prefracture functional capacity. Because of the anticipated increase in hip fracture incidence among men in the coming decades, more detailed investigation of this population to identify potentially modifiable factors that affect fracture outcomes is warranted. Specifically, because postoperative events and infections seem to play important roles in mortality, it may be possible to develop targeted interventions to reduce mortality. The role of osteoporosis and modifiable aspects of this disease process, the effects of immunological and environmental factors, and interactions of these factors with other health conditions (including the risk of falling) also require further elucidation.
The authors would like to acknowledge the participation of the following hospitals in the Baltimore Hip Studies: Franklin Square Hospital, Greater Baltimore Medical Center, Northwest Medical Center, Saint Agnes Hospital, Saint Joseph Hospital, Sinai Hospital of Baltimore, Union Memorial Hospital, and the University of Maryland Medical System. This research was supported by National Institutes of Health Grants R01 AG06322, R01 HD0073, and R37 AG09901.