Risk Factors for Pelvis Fracture in Older Persons

Abstract

From 1996 to 2001, the authors undertook a case-control study of 192 pelvis fracture cases (men and women) and 2,402 controls aged ≥45 years at five Kaiser Permanente medical centers in Northern California. Most information on potential risk factors was obtained by means of an interviewer-administered questionnaire. Number of fractures since age 45 years and a maternal history of hip fracture were associated with increased risks. Several factors thought to protect against loss of bone mass, including recent use of menopausal hormone therapy (adjusted odds ratio (OR) = 0.55, 95% confidence interval (CI): 0.33, 0.91) and high body mass index (weight (kg)/height (m)²; per 5-unit increase, adjusted OR = 0.65, 95% CI: 0.52, 0.81), were associated with decreased risks, while cigarette smoking (adjusted OR = 2.17, 95% CI: 1.34, 3.52) and hysterectomy (adjusted OR = 1.75, 95% CI: 1.15, 2.66) were associated with increased risks. Various conditions related to propensity to fall were associated with increased risks. Most indicators of frailty, including use of walking aids and needing help with or being unable to perform various activities of daily living, conferred increased risks. Thus, low bone mass, frailty, and probably a propensity to fall appear to be associated with increased risk of pelvis fracture.

Fractures of the pelvis, though relatively uncommon, are among the most disabling of fractures. By 85 years of age, approximately 2 percent of White females, 0.7 percent of Black females, and 0.5 percent of White males will experience a pelvis fracture (1). Patterns of age- and gender-specific incidence rates are similar to those for hip fracture, with a rapid increase in incidence rates starting at about age 55 years in females and 10 years later in males (2–7). Low bone mass (8) and low body weight (9) have been found to be strong predictors of pelvis fracture. Although severe trauma, especially from automobile accidents, is a major precipitating cause in persons younger than age 60 years, moderate trauma is the main immediate cause at older ages (5, 6, 10). Other than a high resting heart rate (11), little is known of other risk factors.

In view of this limited knowledge, the objective of this project was to identify risk factors for fracture of the pelvis. We hypothesized that risk factors for pelvis fracture, like those for other age-related fractures such as hip and proximal humerus fractures, would include attributes associated with low bone mass, a propensity to fall, and frailty (12–18). In this paper, we use the term “frailty” in a general sense to indicate reduced ability to perform activities of daily living, impaired physical functioning, limited mobility, and other indicators of suboptimal health. Previous publications from this study have described a probable protective effect of menopausal hormone therapy against pelvis fracture (19), the nature of falls that lead to pelvis fractures (20), and a possibly increased risk among left-handed people (21).

MATERIALS AND METHODS

This study of risk factors for pelvis fracture was part of a larger case-control study designed to identify risk factors for fracture at five sites (the distal forearm, the foot, the proximal humerus, the pelvis, and the shaft of the tibia or fibula) in people aged 45 years or older. Cases and controls were selected from five Kaiser Permanente medical centers (Hayward, Oakland, San Francisco, Santa Clara, and South San Francisco) in Northern California between October 1996 and May 2001. Details on the study design have previously been reported (19, 21, 22) and are presented only briefly here. The protocol was approved by the institutional review boards of the Kaiser Permanente Division of Research and Stanford University School of Medicine.

Cases

Pelvis fracture cases were identified each week from computerized radiology reports and medical records by a trained medical record abstractor. Fractures of the pelvis included all pelvic bones except the coccyx (i.e., the sacrum, ileum, acetabulum, pubis, and ischium). To be included in the study, fractures had to be confirmed by radiograph, bone scan, or magnetic resonance imaging. Persons with prior pelvis fractures after the age of 45 years were not included in these analyses. Pathologic fractures resulting from diseases such as Paget's disease and cancer were also excluded. For persons who had incurred fractures at multiple sites at the same time as the pelvis fracture, only those who had pelvis fracture listed first on the medical record, among the five fracture sites included in this study, were used as cases in these analyses. This was done in order to ensure that these analyses would be comparable to those of other papers from this study in which attributes of persons with fractures at specific sites were compared directly with each other. Seventy-five percent of eligible cases participated.

Controls

Controls were randomly selected from 5-year age groups within each gender from the membership lists of the above five Kaiser Permanente medical centers over the same period of time as the cases. The same control group was used for cases with fracture at each of the five sites included in the larger study from which this project was derived, and thus controls were not matched specifically to the pelvis fracture cases. Every 3 months, the members in each age group were randomly ordered, and the first 34 females and seven males were selected. All persons belonging to a minority group or of unknown race/ethnicity, along with 39 percent of the White females and 78 percent of the White males, were randomly chosen for inclusion in the study. Controls with a previous fracture of the pelvis at age 45 years or older were excluded from these analyses. Sixty-five percent of the eligible controls participated.

Information on potential risk factors

Most information on possible risk factors was obtained using a standardized questionnaire administered by trained interviewers in English or Spanish. For the first 3 years of the study, most of the interviews were conducted in person; after November 15, 2000, most interviews were conducted over the telephone in order to increase numbers and participation. There was no evidence of effect modification by mode of interview, as the associations between potential risk factors and pelvis fracture were similar in those interviewed in person (n = 117 cases, 1,002 controls) and those interviewed by telephone (n = 61 cases, 1,326 controls). Mode of interview was included as a covariate in the analyses. Exclusion of proxy respondents (14 cases and 74 controls) from analyses did not substantially change the odds ratios, so proxy data were retained. However, the number of proxy respondent cases was too low for stratification in the analysis.

The questionnaire covered the time before the fracture for cases and the time before the interview for controls. Topics covered in the questionnaire were: demographic characteristics; weight and height (used to calculate body mass index (weight in kilograms divided by height in meters squared)); handedness; family history of hip fracture; history of practitioner-diagnosed medical conditions (diabetes; angina, heart attack, or heart failure; stroke or blood clot in the brain; epilepsy, seizures, convulsions, or fits; kidney disease; cataracts; glaucoma; Parkinson's disease; arthritis; depression; cancer; hyperthyroidism; and hypothyroidism); self-reported foot problems; history of certain neuromuscular symptoms in the past year (numbness or weakness in the feet or legs; problems with balance or unsteadiness; limping; tremors or shakes; dizziness or lightheadedness; difficulty walking in dim light; and pain, numbness, burning, or tingling in the legs or feet when not walking); history of fractures; history of use of selected medications (thiazide diuretics, water pills, Tums (GlaxoSmithKline Consumer Healthcare, Pittsburgh, Pennsylvania) (calcium carbonate), other calcium supplements, multivitamins, melatonin, steroid pills, and seizure medications) at least once per week for at least 1 year; and recent use of medications to help with sleep, calm nerves, or lift mood. Detailed questions, taken from the Women's Health Initiative (23), were asked of women about use of menopausal hormone therapy for at least 3 consecutive months, age at the last menstrual period, and prior hysterectomy and/or oophorectomy. Respondents were queried about their overall health status in comparison with others of similar age.

Measurement of physical functioning involved asking about the respondent's ability to do or difficulty in doing heavy housework; walking up and down stairs; walking half a mile (0.8 km) without help; pulling or pushing objects; stooping, crouching, or kneeling; lifting 10 pounds (4.5 kg); extending an arm above shoulder level; and writing or handling small objects during the past month. Ability to perform activities of daily living in the past month was assessed by asking whether help was needed in using the telephone, getting groceries, getting to places outside of walking distance, preparing meals, doing chores around the house, taking medications, and handling finances. For each item, a respondent could answer “yes,” “no,” or “unable to do the activity.” A question on the use of walking aids covered canes, walkers, wheelchairs, artificial legs, and braces.

Leisure-time physical activity was assessed by questioning participants about the frequency and duration of walking/hiking, gardening, exercise classes, swimming, bicycling, tennis, calisthenics/weight training, social dancing, jogging, bowling, golfing, stretching exercise or yoga, tai chi, and heavy housework in the past year. Questions modified from the Physical Activity History questionnaire (24) were used. The data thus obtained were converted into metabolic equivalent-hours of exercise per month (25). A calcium-validated food frequency questionnaire (26), to which a few ethnic foods had been added, was used to assess dietary calcium intake. Data on cigarette smoking, alcohol consumption, history of falling, and the immediate cause and circumstances surrounding the fracture were also obtained.

Vision was assessed by asking participants whether they wore corrective lenses for activities other than reading, and, if so, whether they could see well enough to recognize a friend across the room. Hearing was assessed by means of two questions about use of a hearing aid. Detailed questions about falls were asked, as described by Keegan et al. (20).

Ten percent of participants agreed to a slightly abbreviated interview that did not include questions on some or all of the following variables: dietary calcium intake, physical activity, cigarette smoking, and part of the medication history. Because of the reduction in sample size for these variables, analyses including these variables were based on somewhat lower numbers than other analyses.

Statistical analysis

Data were analyzed using SAS software, version 8.2 (SAS Institute, Inc., Cary, North Carolina). All odds ratios were adjusted using unconditional logistic regression for the sampling variables of 5-year age group, gender, and race/ethnicity as recorded in Kaiser Permanente records (White vs. Nonwhite or unknown) and whether the interview was conducted in person versus by telephone. Age in years and self-reported race/ethnicity (White, Native American, or other; Asian or Pacific Islander; Black; Hispanic) were also included to control more tightly for these variables. In addition, unconditional logistic regression was used in some analyses to control for several other variables associated with pelvis fracture in this study.

Approximate age-adjusted odds ratios for pelvis fracture were computed for females relative to males and for self-reported race/ethnicity relative to Whites, taking into account the ways in which controls were sampled. Because Kaiser Permanente-reported race/ethnicity did not entirely agree with self-reported race/ethnicity, especially for Hispanics (27), and because self-reported race/ethnicity is more generally used, estimates were made according to self-reported race/ethnicity. To estimate the odds of a control being in a specific racial/ethnic group, data from the 1999 Northern California Kaiser Permanente membership survey were used for the five medical centers included in this study. In the membership survey, a questionnaire was sent to a random sample of 40,000 Kaiser Permanente members aged 20 years or older. It included questions on race/ethnicity similar to those used in our questionnaire and had a response rate (61.5 percent for members aged 45 years or older) similar to that for the controls in our study (65 percent). We applied the distribution of self-reported race/ethnicity within each 5-year age group, gender, and facility in the membership survey to the control group in our study to obtain an estimate of what the control group distribution would have been if random sampling rather than weighted sampling had been used. Unconditional logistic regression controlling for 5-year age group was then used to compute the odds ratios that would have been obtained if random rather than weighted sampling had been used. We present the logistic regression-derived estimates obtained under these assumptions but do not provide confidence intervals because of the uncertainty of the various assumptions that had to be made in computing these odds ratios. To the extent that the relatively low numbers of pelvis fracture cases permitted, we assessed effect modification by gender, age, race/ethnicity, and selected other variables, both by visual inspection and by including cross-product terms in the logistic regression models. The power of the study to detect effect modification was limited, but little evidence of it was seen. Accordingly, all subgroups were combined in the analyses.

RESULTS

The study included 192 pelvis fracture cases and 2,402 controls (table 1). Females outnumbered males, especially in the older age groups. Twenty-one percent of cases and almost half of controls were from minority racial/ethnic groups, reflecting the oversampling of minority controls.

Taking into account the way in which controls were sampled (see Materials and Methods), the overall odds ratio for males relative to females was estimated to be 0.41, but this ratio was lower at older ages than at younger ages. Relative to White females, the odds ratio was estimated to be 0.56 for Asian females, 0.68 for Black females, and 0.72 for Hispanic females. Relative to White males, the odds ratio was 0.20 for Asian males, 0.37 for Black males, and 0.52 for Hispanic males.

The immediate cause of fracture for most cases was a fall (79.8 percent). This was followed by motor vehicle accidents (11.8 percent), spontaneous fractures (“bone just broke”) (2.8 percent), recreational accidents (2.3 percent), and miscellaneous other causes (3.3 percent). Because odds ratios were similar when analyses were undertaken with and without the inclusion of cases whose fractures were attributed to a motor vehicle accident, we included all cases in the data presented here in order to maximize statistical power.

Table 2 presents odds ratios for pelvis fracture. Odds ratios were first adjusted only for the design variables. They were then adjusted, in the multivariate analysis, for the design variables plus the other potential risk factors identified in this study, for determination of which factors were independently associated with pelvis fracture. With a few exceptions (noted below), the design-adjusted and multivariate-adjusted odds ratios were similar. Not included in the multivariate model results presented here are attributes not associated with pelvis fracture in this study and attributes of low frequency such as Parkinson's disease and left-handedness.

The number of previous fractures since age 45 years and having a mother who had suffered a hip fracture, both of which are thought to have some association with low bone mass, were associated with increased risks (table 2). Among other variables related to low bone mass, recent use of menopausal hormone therapy, high body mass index, use of Tums at least once per week for at least 1 year, and ever having used thiazide diuretics for at least 1 year were associated with decreased risks, although the odds ratios for Tums and thiazide diuretics did not quite reach statistical significance and were considerably attenuated in the multivariate analysis. Cigarette smoking and having had a hysterectomy with or without removal of the ovaries were associated with increased risks. Amount of dietary calcium did not appear to affect risk.

A history of falls in the past year was not associated with an increased risk of pelvis fracture. Conditions thought to be associated with falling, including epilepsy or use of seizure medication, Parkinson's disease, possibly a history of a stroke, wearing corrective lenses (except for reading only), and frequent use of a hearing aid, tended to be associated with increased risks, although some of these associations were not statistically significant. Taller persons, who have farther to fall before hitting the ground, had a somewhat elevated risk. Being left-handed was possibly associated with an increased risk.

Many indicators of frailty, some of which reached statistical significance and some of which did not, were associated with an increased risk—including use of walking aids, needing help with or being unable to perform various activities of daily living, and, less strikingly, needing help with or being unable to perform some of the physical functions queried. Being physically inactive during leisure time in the past year was associated with, at most, a slightly increased risk of pelvis fracture, while self-reported poor or fair health was not associated with an increased risk. The multivariate model included summary variables for use of walking aids and needing help with or being unable to perform activities of daily living. Because the associations with difficulty performing physical functions, leisure-time physical activity, and self-reported health status were relatively weak, these variables were not included in the multivariate model. Finally, a few self-reported diseases or neuromuscular symptoms showed statistically significant or almost significant associations with pelvis fracture: Odds ratios were elevated for tremors or shakes in the past year and an overactive thyroid gland, and odds ratios were reduced for a limp in the past year; pain, numbness, burning, or tingling in the legs or feet in the past year when not walking; and a history of angina. Foot problems and other neuromuscular symptoms, medications, and diseases queried were not associated with risk of pelvis fracture.

DISCUSSION

Previous studies (2–7) have shown that the pattern of age- and sex-specific incidence rates of pelvis fracture is typical of other “frailty” fractures such as fractures of the hip and proximal humerus. These studies have shown that incidence rates increase steeply with age after approximately age 45–50 years and rise in males later than in females. In other reports (1, 3), as well as in the present study, higher risks for pelvis fracture were found in Whites than in Blacks. To our knowledge, no previous study has reported risks in Asians or Hispanics, but our findings of lower risks of pelvis fracture in these groups than in Whites are consistent with previous reports for hip fracture (28, 29).

The present study further suggests that other risk factors for pelvis fracture are generally similar to risk factors for fractures of the hip (12–15) and proximal humerus (16–18). Although some variation occurs in the magnitude and statistical significance of associations from one study to another, persons at high risk of fracture at all three of these sites tend to have low bone mass, to be at above-average risk of falling, and to be frail.

In the present study, we found associations between risk of pelvis fracture and number of previous fractures since age 45 years and a maternal history of hip fracture. Both of these variables are risk factors for fracture at various sites, probably in part because of their association with low bone mass and also because of their associations with other risk factors for fracture (30). Although bone mass was not measured in this study, Seeley et al. (8) reported a strong association between low bone mineral density and risk of pelvis fracture in the prospective Study of Osteoporotic Fractures. In addition, being a current smoker and having had a hysterectomy were associated with an increased risk, while recent use of menopausal hormone therapy, a high body mass index, and possibly use of Tums and thiazide diuretics were associated with lower risks. Use of menopausal hormone therapy is known to protect against loss of bone mass and fracture at most sites, particularly if use is recent (19, 31, 32). An association between low body weight and risk of pelvis fracture was also reported by Margolis et al. (9). High body weight (or body mass index) probably affords protection because it is associated with greater bone strength from higher endogenous estrogen and testosterone concentrations and from increased mechanical stress (31), and also because of greater cushioning in the event of a fall (33, 34). The mechanism for a potential protective effect of thiazide diuretics is not certain (31), but it is believed to be related to lower urinary calcium excretion. An increased risk from cigarette smoking is consistent with findings that smoking is associated with lower endogenous estrogen concentrations and lower bone mass (35). Why hysterectomy would be a risk factor, regardless of oophorectomy status, is unclear.

Although no association between number of falls in the past year and risk of pelvis fracture was found in this study, associations were seen for factors that put one at high risk of falling, such as having epilepsy or having used seizure medications (36), possibly having Parkinson's disease (36–38), poor vision (39), and hearing loss (40). A history of falling has been reported to be a risk factor for hip fracture (12) and proximal humerus fracture (17). To our knowledge, no other study has reported on propensity to fall as a risk factor for pelvis fracture, but given that most pelvis fractures in older persons are the direct result of a fall, as reported in this and other studies (5, 6), it stands to reason that the greater the likelihood of falling, the higher the risk of pelvis fracture. Taller people are believed to be at greater risk of hip fracture because they are falling from a greater height (41, 42), and the same explanation may apply to pelvis fracture. The possible association of pelvis fracture with left-handedness is consistent with previous reports of associations between left-handedness and distal forearm fracture (43), and between left-handedness and fractures in general and proximal humerus fractures in particular (previous report from the present study (21)). Possible mechanisms are that certain components of the environment favor right-handed people, thus putting left-handed people at greater risk of falling, and that left-handedness is a developmental abnormality such that left-handed people are innately at increased risk for a variety of problems, including falling (21, 43).

Likewise, no other study has considered indicators of frailty (other than age) as risk factors for pelvis fracture. Frail people with impaired gait and balance, less muscle strength, and slower reaction time are probably more likely to fall and are less likely to be able to break a fall with an outstretched hand (20, 41), so the association with frailty is highly plausible. In the present study, risk was elevated in persons who used walking aids, persons who needed help with activities of daily living, and to some extent persons who had difficulty performing or were unable to perform various physical functions. No association, however, was seen with self-reported health status. Positive associations were found for self-reported tremors or shakes and an overactive thyroid gland, and negative associations were found with people who reported limping; pain, numbness, burning, or tingling in their legs or feet when not walking; and angina. Cummings et al. (12) found hyperthyroidism to be a risk factor for hip fracture and suggested as possible mechanisms reduced bone strength, impaired neuromuscular function, and reduced muscle strength. We have no explanations for the negative associations with angina, with limping, and with pain, numbness, burning, or tingling in the legs or feet in the past year when not walking. The questions on limping and problems in the legs or feet in the past year were identical to those used in a previous case-control study of hip fracture (13), and no associations, either positive or negative, were found in that study. These associations would have to be evaluated in other studies before it was decided whether they are real.

Why, then, do some frail osteoporotic people who fall break their hip, some their proximal humerus, and others their pelvis? We hypothesize that the site of a fracture depends, at least in part, on how a person falls and on what body part she/he lands when falling. People who fracture their proximal humerus tend to fall to the side and land on their shoulder (20). People who fracture their hip tend to fall sideways and land on their hip or thigh (34, 41, 42, 44, 45). In a previous publication from the present study, Keegan et al. (20) found that falling in any direction relative to forward was associated with an elevated risk of pelvis fracture, probably reflecting the relatively wide area in which the pelvis is situated in the lower trunk of the body. More research is needed on why some frail people fall in one direction and some in another.

The various limitations of this study should be noted. Although the 192 cases make this, to our knowledge, the largest epidemiologic study of pelvis fracture reported to date, the number of cases was not high enough to provide much power to examine subgroups or to detect small associations. The study was of the case-control design and was therefore dependent on the ability of the participants to remember and report past events. Undoubtedly some misclassification occurred, although most of the error was probably nondifferential and would therefore have tended to dilute the associations. Response rates were only fair. Finally, many possible risk factors were queried, thereby increasing the likelihood that some associations would occur by chance. However, given that this was the first study with the objective of identifying a variety of risk factors for pelvis fracture, it made sense to cast a broad net and to provide leads for other studies to evaluate further.

In conclusion, if confirmed by others, our results suggest that frail people with low bone mass and probably a propensity for falling are at elevated risk of pelvis fracture.

This study was supported by research grant R01 AR42421 and training grant T32 AR07588 from the National Institute of Arthritis and Musculoskeletal and Skin Diseases.

The authors thank Beverly Peters and Luisa Hamilton for project management, Michael Sorel for computing and database management, and Carolyn Salazar for medical record abstraction.

Conflict of interest: none declared.

References