Elevated Hip Fracture Risk in Type 1 Diabetic Patients: A Population-Based Cohort Study in Sweden

This study was approved by the regional ethics committee at the Karolinska Institutet.

With minor exceptions, in-hospital medical services in Sweden have been exclusively public, organized by the community. Since patients have been obliged to use the hospitals in their county of residence, in-hospital care registration is, in practice, population-based and referable to the county where the patient lives. The Swedish Inpatient Register was established by the National Board of Health and Welfare in 1964–1965, but most counties joined the registration later, the last one, in 1987, when the Register became nationwide. Each record in the register corresponds to one hospital admission and contains, in addition to the patient’s national registration number (a unique identifier assigned to all Swedish residents), the dates of admission and discharge, codes for all surgical procedures, and discharge diagnoses. In five counties, the Inpatient Register lacked national registration numbers during 1 or 2 years (1984–1986).

The ICD coding before the 10th revision introduced in 1997 did not allow us to separate type 1 from type 2 diabetes. Even after this date, some patients coded as having insulin-dependent diabetes actually had advanced type 2 diabetes that had developed into insulin dependency. Therefore, we used age <30 years at first hospitalization for diabetes (even if it preceded the start of cohort accrual) as an obligatory criterion. In an analysis confined to patients hospitalized in 1998, when the ICD-10 codes were used exclusively in Sweden and permitted differentiation between insulin-dependent and other diabetes, of 3,730 diabetic patients diagnosed younger than age 31, 3,542 had a diagnosis of insulin-dependent type. Thus, our age algorithm had a positive predictive value for insulin-dependent diabetes of 95%. Cohort accrual started on different dates in different counties but was always at least 2 years after the registration had attained full coverage without interruption in that county. The earliest starting date was 1 January 1975 (Uppsala and Gävleborg counties); the latest was 1 January 1989 (Kronoberg county).

We initially identified 25,221 records with a discharge diagnosis of type 1 diabetes. These records were further linked to the Register of Total Population, the Register of Domestic and Foreign Migration, and the Death Register. These linkages resulted in the exclusion of 171 records with national registration numbers that could not be found in any of the registers, i.e., without a link to any currently or previously existing person. Further excluded were 376 patients with date of emigration before the index hospitalization, 43 patients who died on the index hospitalization, and 26 patients who already had hip fractures before or on the index hospitalization. Hence, our final type 1 diabetes cohort included 24,605 patients.

Cohort members were followed from immediately after the index hospitalization for type 1 diabetes until occurrence of a first hospitalization for hip fracture, emigration, migration to a county without or with incomplete Inpatient Register coverage, death, or the end of follow-up (31 December 1998), whichever occurred first. The first hospitalization for hip fracture, if any, was identified through cross-linkage within the Inpatient Register. Complications of diabetes were identified in a similar way. Vital status and coverage by the Inpatient Register was ascertained by linkage to the registers of Death and Domestic and Foreign Migration.

By using the Kaplan-Meier method, we calculated lifetime cumulative probability of developing hip fracture before the age of 65 years. Standardized hospitalization ratios (SHRs; the ratios of the observed to the expected numbers of first hospitalizations for hip fractures) served as our measure of relative risk. To calculate the background hospitalization rates, we used the entire Inpatient Register and counted the number of first hospitalizations for each of the diagnoses of interest (femoral neck, pertrochanteric, and multiple site) and then for any hip fracture regardless of type in the general population by age (in 5-year groups), sex, and calendar period (every 2 years from 1975 to 1998). Stratum-specific hospitalization rates were computed by dividing the number of hip fractures by the corresponding number of general population at risk. The expected number of hip fracture hospitalizations in the cohorts was derived by multiplying the observed number of person-years in age, sex, and calendar period strata by the corresponding stratum-specific hospitalization rates. Ninety-five percent CIs were calculated by assuming that the number of observed events followed a Poisson distribution (13). We did not analyze multiple-site hip fractures since no such case was found in our cohort.

Additional analyses were stratified according to follow-up duration, and a χ² test for linear trend was used to evaluate the time-risk relationship (14). We further stratified by presence/absence of diabetes complications. Person-time experienced before the onset of complications was allocated to the complication-negative strata.

To isolate the independent effects of explanatory variables, we estimated relative effects on the standardized hospitalization ratios using a multivariate Poisson regression method with the expected number as the offset, assuming multiplicative effects between outcome and explanatory variables. The Pearson’s χ² test was used to check the degree of fit of the model (14).

Both mean and median age at entry was 20.7 years. The cohort members were followed for an average of 9.9 years, yielding 242,428 accumulated person-years at risk. Of this person-time, 30,377 person-years occurred during ages 40 through 49, and 5,221 above the age of 50. The highest attained age during follow-up in the cohort was 64.5 years. During follow-up, a total of 121 incident cases of hip fracture were recorded (51 among women, 70 among men). Anatomic site of the fracture was the femoral neck in 64 (53%) and the pertrochanteric area in 57 (47%) patients. The mean age at diagnosis of first hip fractures was 43.1 years for women and 41.3 years for men (Table 1).

Figure 1 shows Kaplan-Meier estimates of the cumulative risk of hip fracture among patients with type 1 diabetes stratified by sex. The incidence of hip fractures before age 30 was almost negligible but increased quickly thereafter. The cumulative probability of getting a hip fracture was similar in both sexes before age 40, but thereafter, it increased more rapidly among men. At the age of 65, the cumulative probability for both sexes combined was 6.6%.

Among men, the SHR estimating relative risk for hip fracture relative to the age-, sex-, and calendar period–matched general population was 7.6 (95% CI 5.9–9.6). The excess risks were evident across different durations of follow-up but without a clear trend. An analysis stratified by attained age showed much higher relative risk for hip fracture among men who had reached age 40 years or older than among those who were younger. Analogously, men born before 1950 had a much higher excess risk than those born in 1950 or later. Men ever hospitalized with microvascular complications, such as ophthalmic or nephropathic complications, neuropathy, or cardiovascular complications, had a substantially higher relative risk for hip fracture as a whole or divided into subtypes than men without documented diabetes complications (Table 2).

An increased relative risk for hip fractures was also observed among women with type 1 diabetes (SHR 9.8 [95% CI 7.3–12.9]). The excess increased with follow-up duration (P = 0.02 for trend). Similar to men with type 1 diabetes, women had much higher relative risk for hip fracture when they were >40 years old at follow-up or born before 1950. Remarkably greater relative risk elevations were also observed among women with type 1 diabetes ever hospitalized with microvascular, neurological, or cardiovascular complications than among women with no records of such complications (Table 3).

There was no obvious difference of SHRs for hip fracture between men and women, either in the univariate or multivariate analysis. After controlling for effects of the other explanatory variables, SHRs over 10 years of follow-up were close to that with follow-up <5 years. Type 1 diabetic patients had an over fourfold relative risk when their attained ages at follow-up were ≥40 years, whereas no obvious difference of SHRs was noted between patients born in 1950 or later and those born before 1950. Patients with diabetes complications had generally higher relative risks, particularly among those with nephropathic or neuropathic complications (2.0 [95%CI 1.0–3.8] and 2.4 [1.3–4.6], respectively) (Table 4). The Pearson’s χ² statistic for the multivariate model was 500.5 with 330 degrees of freedom, which indicated modest overdispersion of the model. A scale parameter, the square root of the Pearson’s χ² divided by the degrees of freedom, was thus used to correct standard errors.

We observed unequivocally increased hip fracture risks among both men and women who had been hospitalized for type 1 diabetes. Presence of diabetic microvascular (ophthalmic or nephropathic), neurological, or cardiovascular complications indicated excess risks that ranged from 17- to 42-fold.

The putative mechanism nearest at hand is an impaired bone quality due to the lower bone mineral density observed among type 1 diabetic patients (1–3). Moreover, a link between microvascular complications of type 1 diabetes and long-term bone loss has been reported (15), notably between neuropathy and decreased bone mineral density of the femoral neck (16).

Another mechanism for the increased risk could be diabetes-related nonskeletal risk factors, such as a propensity for falls (17), conceivably mediated through impaired proprioception, balance, and gait due to neuropathy, visual impairment from diabetic retinopathy and cataracts (18), or frequent nocturia (19).

To date, there has been remarkably little data about the influence of diabetes on hip fracture risk among men. With 12,551 type 1 diabetic men, our cohort provided a unique chance to explore this association. The observed number of first hip fractures in our cohort exceeds that in all previous studies including men. Other strengths of our study include the cohort design, the population-based sample, the use of medical information rather than reliance on self-reports of exposure and outcome, and the virtually complete follow-up in view of the essentially obligatory need for hospital care after hip fracture and the absence of hospitals not covered by the Inpatient Register. Despite the many advantages of our study, we should point out its limitations. First, information on potential confounding factors such as weight change, smoking, and body mass/stature was unavailable for adjustment. It is, however, unlikely that any confounding from these risk factors could produce relative risk elevations of the magnitude observed in our study. Second, our study relied solely on age (≤30 years) when defining type 1 diabetes. Some patients with type 2 diabetes may have been misclassified as having type 1 diabetes, but their contribution is probably no more than 5% of the entire type 1 diabetes cohort. For some patients, the onset of diabetes may have preceded the start of inpatient registration in the county of residence (and thus the start of follow-up) by several years. Therefore, there is a hypothetical risk that we might have missed some hip fracture outcomes. Hip fracture incidence before age 40 was very low, though. Finally, our data on fracture incidence and expected rates were derived from the Inpatient Register, where some misclassification and misregistration of this outcome, indeed, occurs. Since the specificity of hip fracture diagnoses is high (20), only underascertainment requires serious consideration. However, hip fractures are virtually always treated on an inpatient basis and should therefore appear in the Inpatient Register. In a study from 1965 to 1983, underreporting of hip fractures was found to occur in <2% of cases (21). Besides, any underascertainment of hip fractures is likely to be due to technical errors and is thus probably nondifferential. Such underascertainment in follow-up studies will not affect the rate ratio (22).

There is no simple answer to the question of whether our restriction to hospitalized patients may have affected the observed relationship. Because of the often dramatic onset and the need for careful evaluation, we believe that most pediatric patients with type 1 diabetes during the studied period were hospitalized at least once. Thus, no important selection bias is expected. On the other hand, our cohort included a relatively large proportion of patients who were around 20 years of age or older at index hospitalization. Many of these patients may have had an earlier diabetes onset with an initial hospitalization that occurred before inpatient registration was started in their respective counties of residence. These patients might have been rehospitalized because of an atypically severe course of the disease, and the possible concentration of such patients in the cohort may have somewhat inflated the observed risks.

Although causality cannot be proven in this observational study, the strength of the association, the support provided from some, albeit not all, previous studies, the dose-response relationship reflected by the increasing relative risk with follow-up time and severity of the disease, and the biological plausibility based on bone density measurements in diabetic patients make us inclined to conclude that the observed association is likely to be causal. We were unable to study the risk of other fractures simply because few other fractures lead to obligatory hospital admissions. If there is a real choice between in- and outpatient treatment (the latter is not recorded in the Swedish Inpatient Register), the concomitant presence of diabetes might shift the balance toward hospital admission, leading to ascertainment bias. Prospective studies of the risk of all kinds of fractures are, however, clearly warranted, and we hypothesize that the susceptibility for fracture is not only confined to the hip.

Given that as many as 1 in 15 patients with type 1 diabetes may sustain a hip fracture before the age of 65, development of methods for primary prevention should be put high on the agenda. Although candidate treatments should ideally be evaluated in randomized intervention trials, our data provide some hints regarding the importance of tight metabolic control; the covariation with other diabetes complications and the considerably higher relative risk among patients born before 1950, who may have had a considerable part of their disease trajectory before the importance of meticulous metabolic control was clearly demonstrated, suggest that such control may also be a cornerstone in the prevention of diabetic hip fractures.

This study was partly funded by grants from K.I. Fonder (to W.Y. and O.N.), the Family S. Persson Foundation, and the Swedish Diabetes Foundation (to K.B.).