Geovariation in Fracture Risk among Patients Receiving Hemodialysis

Abstract

Background and objectives

Fractures are a major source of morbidity and mortality in patients receiving dialysis. We sought to determine whether rates of fractures and tendon ruptures vary geographically.

Design, setting, participants, & measurements

Data from the US Renal Data System were used to create four yearly cohorts, 2007–2010, including all eligible prevalent patients on hemodialysis in the United States on January 1 of each year. A secondary analysis comprising patients in a large dialysis organization conducted over the same period permitted inclusion of patient-level markers of mineral metabolism. Patients were grouped into 10 regions designated by the Centers for Medicare and Medicaid Services and divided by latitude into one of three bands: south, <35°; middle, 35° to <40°; and north, ≥40°. Poisson regression was used to calculate unadjusted and adjusted region–level rate ratios for events.

Results

Overall, 327,615 patients on hemodialysis were included. Mean (SD) age was 61.8 (15.0) years old, 52.7% were white, and 55.0% were men. During 716,962 person-years of follow-up, 44,014 fractures and tendon ruptures occurred, the latter being only 0.3% of overall events. Event rates ranged from 5.36 to 7.83 per 100 person-years, a 1.5-fold rate difference across regions. Unadjusted region–level rate ratios varied from 0.83 (95% confidence interval, 0.81 to 0.85) to 1.20 (95% confidence interval, 1.18 to 1.23), a 1.45-fold rate difference. After adjustment for a wide range of case mix variables, a 1.33-fold variation in rates remained. Rates were higher in north and middle bands than the south (north rate ratio, 1.18; 95% confidence interval, 1.13 to 1.23; middle rate ratio, 1.13; 95% confidence interval, 1.10 to 1.17). Latitude explained 11% of variation, independent of region. A complementary analysis of 87,013 patients from a large dialysis organization further adjusted for circulating mineral metabolic parameters and protein energy wasting yielded similar results.

Conclusions

Rates of fractures vary geographically in the United States dialysis population, even after adjustment for known patient characteristics. Latitude seems to contribute to this phenomenon, but additional analyses exploring whether other factors might influence variation are warranted.

Introduction

Bone fractures are a major source of morbidity and mortality in patients receiving maintenance dialysis (1–6). Fractures are substantially more common in patients on dialysis than in the general population (4,7–11), and they contribute to significant morbidity and health resource utilization (3). Furthermore, fractures in patients on dialysis are associated with mortality rates substantially higher than in the general population (12,13).

Although fractures of major bones are acute catastrophic events, they may be caused by longstanding physiologic derangements. Such derangements also likely contribute to tendon ruptures, which are relatively rare but serious events in patients on dialysis (14,15). Derangements include abnormalities directly linked with the mineral bone disorder of CKD, those associated with CKD (such as poor muscle strength or disturbances in body balance), the risk of orthostasis associated with large ultrafiltration volumes, and general risk factors (such as diabetes and osteoporosis) (1,2,16). Several factors have been associated with increased fracture risk in the dialysis population, including older age, female sex, longer dialysis duration, diabetes, and derangements in parathyroid hormone (PTH), among others (8–12,17–23).

However, gaps remain in understanding the etiology of fractures in the dialysis population. For example, no study has evaluated whether geographic factors, demographic factors (such as race/ethnicity), or local treatment practices, all of which may influence access to care and have been studied in relationship to other aspects of dialysis outcomes (24–30), are associated with fracture risk in the United States dialysis population. Whether rates of fracture and tendon rupture in the dialysis population vary by geographic region has not been specifically studied.

Therefore, we conducted a large retrospective study using a national cohort of Medicare patients on hemodialysis; a secondary analysis using patient-level data was conducted on a subset of patients receiving dialysis from a large dialysis organization (LDO). We hypothesized that there would be significant variation in fracture rates across the country. Improved understanding of fracture risk among these patients may help identify those who could benefit from targeted interventions.

Materials and Methods

Data Source and Study Populations

We used the US Renal Data System (USRDS) ESRD data for our primary analysis; for the secondary analysis, we used data linked between a United States LDO and the USRDS ESRD data. The USRDS ESRD database consists of data from the ESRD Medical Evidence Report (form Centers for Medicare and Medicaid Services [CMS] -2728), the ESRD Death Notification (form CMS-2746), the United Network for Organ Sharing kidney transplant database, Medicare Part A institutional claims (inpatient, outpatient, skilled nursing facility, home health, and hospice), and Medicare Part B physician/supplier claims (inpatient, outpatient, and supplier). For the subset of patients receiving care from the LDO (LDO Medicare), we obtained information on laboratory values, medication use, dialysis treatments, date of first service in the system, and censoring date and reasons (death, transfer to a different system, or modality change) from the dataset generated by the dialysis provider (31).

We created four yearly cohorts (2007–2010) for the national Medicare population and four for the subgroup of patients who were dialyzing with the LDO. Each yearly cohort included all point prevalent patients who were receiving maintenance hemodialysis on January 1 of the year (that is, the index date) and met the following criteria as of that date: age 18 years or older, Medicare as the primary payer for both Parts A and B for at least 1 year, and receiving hemodialysis for at least 1 year. Patients with a previous kidney transplant were retained for the primary analysis but removed in a secondary analysis. Patients were followed from January 1 of the calendar year to the earliest date of death, kidney transplant, modality switch, loss of Medicare coverage, or 1 year.

Patient Characteristics

Patient characteristics were assessed as of the index date and included demographics (age, sex, and race [black, white, or other]), body mass index (BMI), cause of ESRD (diabetes, hypertension, GN, or other, including unknown and missing), dialysis duration, comorbid conditions, and newly created functional status score. We procured demographic, BMI, cause of ESRD, and dialysis duration information from the ESRD Medical Evidence Report and derived comorbid condition and functional status information using medical claims in the year preceding each yearly cohort (32). Comorbid conditions included diabetes, atherosclerotic heart disease, congestive heart failure, other cardiovascular diseases, cerebrovascular events (including transient ischemic attack), peripheral vascular disease, dysrhythmias, osteoporosis, and prior fracture. We derived comorbid conditions using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes on the basis of previously established methods (details are in Supplemental Material 1 and Supplemental Table 1). Elements included in the functional status score were use of home oxygen, walkers, or wheelchairs; presence of home health claims; diagnosis of depression; and diagnosis of visual impairment. Higher scores indicated worse functional status (details about derivation of the elements and calculation of the functional status score are in Supplemental Material 2 and Supplemental Table 2); the theoretical range of this score is from 0 to 10.

Assessment of Laboratory Values and Protein Energy Wasting

For the subset of patient dialyzing with the LDO, serum mineral metabolic parameters, namely PTH, calcium, and phosphorus, were collected from the LDO. Protein energy wasting (PEW) was characterized using serum albumin, creatinine, BUN, potassium, protein catabolic rate, and white blood cell count. Details on the construction of the PEW score and resultant groupings are described in Supplemental Material 3. The theoretical range of this score is from 0 to 28.

Markers of Geographic Variability

We evaluated several markers of geographic variability. First, patients were grouped into the ten official CMS–designated regions (33), except that region 8 was amalgamated with the adjacent region 10 (both being in the northwestern part of the United States) because of the comparatively small population sizes of these regions. A description of the regions is in Supplemental Material 4. Second, dialysis units were designated as rural or urban on the basis of the rural-urban commuting area codes (34). Urbanicity was characterized on the basis of the rural-urban commuting area value: 1–3 (most urban areas), 4–6, and 7–10 (most rural areas). Third, because ecologic studies have reported that fracture incidence varies by latitude in the general population (35–38), we arbitrarily divided the United States into three bands by latitude: south, <35°; middle, 35° to <40°; and north, ≥40°.

Fractures and Tendon Ruptures as an Outcome

Fractures and tendon ruptures were identified using ICD-9-CM diagnosis codes and Healthcare Common Procedure Coding System procedure codes from inpatient and physician claims. We sought to study fractures of major bones that would likely impair mobility, require prolonged healing (and perhaps, surgical intervention), and affect quality of life in major ways. Thus, we included fractures of the pelvis/hip, femur, tibia and fibula, humerus, radius and ulna, and rib/sternum in our outcome; we did not include fractures of the fingers, toes, vertebra, or collarbone, because these tend to be less severe and are not associated with significant morbidity. Tendon ruptures, although rare, were also included, because high PTH levels have been associated with a weakened bone cortex at the tendon insertion site, leading to rupture (39). Details regarding fracture and tendon rupture definitions and associated codes are shown in Supplemental Material 5 and Supplemental Table 3. For each fracture site, only the first event in each cohort year was counted. Of note, because tendon ruptures accounted for <0.3% of all outcomes, we henceforth refer to our outcome as fractures.

Statistical Analyses

We used descriptive statistics (count [n] and percentage) to characterize the patient population. Fracture rates were calculated as observed number of first events divided by follow-up time per 100 person-years. Poisson regression models were used to calculate unadjusted and adjusted CMS region–level rate ratios and corresponding 95% confidence intervals (95% CIs) in both cohorts. The reference groups were the respective nationwide overall fracture rates in each cohort, meaning that rate ratios should cluster around unity if there is no substantial geographic variation. To determine whether and how geography was associated with fracture variation, models were constructed with successive layers of adjustment. The first adjustment was for age, race, sex, BMI, primary cause of ESRD, and dialysis duration; the second adjustment added comorbid conditions and functional status score, the third added urban-rural status, and the fourth added latitude. Because fracture rate changed slightly over time, cohort year (2007, 2008, etc.) was included in all models (including the initial unadjusted model). We characterized variation in fracture rates among regions using the difference between the largest and smallest rate ratios and the SD of the rate ratio (SD_RR), a marker of rate ratio variability. We examined the percentage reduction of the SD_RR over the sequential adjustments to assess the degree to which various adjustments accounted for variability in fracture rates. For the secondary analysis of the LDO Medicare cohort, we included a fifth adjustment for levels of PTH (<150, 150–600, or >600 pg/ml), corrected calcium (<8.4, 8.4–10.2, or >10.2 mg/dl), phosphorus (<3.5, 3.5–5.5, or >5.5 mg/dl), and PEW. The functional forms of calcium, phosphorus, and PTH were examined to determine whether treatment as a spline function would be informative, and they were modeled as splines in a sensitivity analysis. To determine the factors associated with fracture, we used the final respective fully adjusted models for both cohorts.

Finally, to help determine the robustness of estimates for the latitude effect, sensitivity analyses were conducted in both cohorts by removing the CMS region from the final models; this was done because the CMS region and latitude are highly correlated.

All analyses were conducted using SAS, version 9.2 (SAS Institute Inc., Cary, NC).

Results

Patient Characteristics

In total, 327,615 unique patients were included in the primary analysis of the national Medicare cohort. Most patients (66.5%) appeared in multiple yearly cohorts, such that the cumulative patient total for all four yearly cohorts was 807,654. In total, 87,013 unique patients were included in the secondary analysis of the LDO Medicare cohort; the cumulative patient total was 160,931. Characteristics for the first and last yearly cohorts for the national Medicare and LDO Medicare patients are shown in Table 1. Overall, characteristics across years and between the national Medicare and LDO Medicare cohorts were similar.

Table 1.

Characteristics of the patients in the two analytic cohorts

Characteristics	National Medicare				National LDO
	2007		2010		2007		2010
	n	%	n	%	n	%	n	%
Total	188,146	100	214,238	100	24,545	100	47,631	100
Age, yr
19–44	27,893	14.8	30,377	14.2	3493	14.2	6609	13.9
45–64	74,487	39.6	89,011	41.6	9753	39.7	19,947	41.9
65–74	45,335	24.1	50,907	23.8	6175	25.2	11,371	23.9
≥75	40,431	21.5	43,943	20.5	5124	20.9	9704	20.4
Sex
Men	102,716	54.6	118,300	55.2	13,320	54.3	26,306	55.2
Women	85,430	45.4	95,938	44.8	11,225	45.7	21,325	44.8
Race
White	99,235	52.7	112,340	52.4	12,997	53.0	24,106	50.6
Black	77,428	41.2	88,827	41.5	9937	40.5	20,436	42.9
Other	11,483	6.1	13,071	6.1	1611	6.6	3089	6.5
ESRD primary cause
Diabetes	81,813	43.5	94,287	44.0	11,118	45.3	21,585	45.3
Hypertension	54,778	29.1	62,268	29.1	7152	29.1	14,146	29.7
GN	20,538	10.9	22,067	10.3	2418	9.9	4502	9.5
Other/unknown/missing	31,017	16.5	35,616	16.6	3857	15.7	7398	15.5
BMI, kg/m2
<18	5805	3.1	5538	2.6	723	3.0	1216	2.6
18 to <25	60,446	32.1	62,954	29.4	7869	32.1	14,245	29.9
25 to <30	51,851	27.6	59,164	27.6	6882	28.0	13,334	28.0
30 to <35	32,061	17.0	38,646	18.0	4288	17.5	8626	18.1
35 to <40	16,830	9.0	21,456	10.0	2179	8.9	4556	9.6
≥40	14,893	7.9	20,750	9.7	1877	7.7	4491	9.4
Missing	6260	3.3	5730	2.7	727	3.0	1163	2.4
Dialysis duration, yr
1 to <3	69,441	36.9	70,125	32.7	9518	38.8	14,904	31.3
3 to <5	49,454	26.3	55,936	26.1	7135	29.1	13,108	27.5
≥5	69,251	36.8	88,177	41.2	7892	32.2	19,619	41.2
Comorbid conditions
Diabetes	118,032	62.7	139,622	65.2	15,381	62.7	30,700	64.5
ASHD	85,148	45.3	96,745	45.2	11,080	45.1	21,397	44.9
CHF	93,902	49.9	106,883	49.9	12,374	50.4	24,158	50.7
CVA/TIA	36,560	19.4	43,046	20.1	4783	19.5	9602	20.2
PVD	73,335	39.0	85,018	39.7	9816	40.0	19,365	40.7
Dysrhythmia	51,055	27.1	58,908	27.5	6558	26.7	13,300	27.9
Cardiac disease, other	56,951	30.3	71,543	33.4	7498	30.6	16,123	33.9
Prior fracture	7894	4.2	8629	4.0	991	4.0	1913	4.0
Osteoporosis	4986	2.7	5680	2.7	595	2.4	1161	2.4
Functional status score
0	85,999	45.7	93,969	43.9	10,932	44.5	20,159	42.3
1	52,285	27.8	59,235	27.7	7251	29.5	13,861	29.1
2	33,315	17.7	38,055	17.8	4291	17.5	8590	18.0
3	16,547	8.8	22,979	10.7	2071	8.4	5021	10.5
CMS region
1	6270	3.3	7094	3.3	241	1.0	1432	3.0
2	17,554	9.3	19,457	9.1	1772	7.2	2383	5.0
3	19,126	10.2	21,392	10.0	3041	12.4	6007	12.6
4	45,389	24.1	51,870	24.2	5101	20.8	12,132	25.5
5	32,431	17.2	35,899	16.8	4570	18.6	6832	14.3
6	28,250	15.0	33,206	15.5	4127	16.8	6869	14.4
7	6652	3.5	7184	3.4	787	3.2	1581	3.3
8/10	8090	4.3	9119	4.3	1212	4.9	1647	3.5
9	24,384	13.0	29,017	13.5	3694	15.1	8748	18.4
Urban-rural status
1 (RUCA 1–3)	154,065	81.9	176,014	82.2	21,049	85.8	40,378	84.8
2 (RUCA 4–6)	19,204	10.2	21,601	10.1	1666	6.8	4025	8.5
3 (RUCA 7–10)	14,877	7.9	16,623	7.8	1830	7.5	3228	6.8
Latitude
North	71,781	38.2	83,860	39.1	9483	38.6	19,759	41.5
Middle	54,150	28.8	61,511	28.7	8123	33.1	16,781	35.2
South	62,215	33.1	68,867	32.2	6939	28.3	11,091	23.3
PEW category
1					1377	5.6	2991	6.3
2					7855	32.0	16,574	34.8
3					13,608	55.4	25,393	53.3
4					1705	7.0	2673	5.6
Phosphorus, mg/dl
<3.5					2068	8.4	4241	8.9
3.5–5.5					13,348	54.4	27,556	57.9
>5.5					9129	37.2	15,834	33.2
Calcium, corrected, mg/dl
<8.4					2264	9.2	6327	13.3
8.4–10.2					20,685	84.3	39,922	83.8
>10.2					1596	6.5	1382	2.9
PTH, pg/ml
<150					3933	16.0	6896	14.5
150–600					17,478	71.2	35,120	73.7
>600					3134	12.8	5615	11.8

LDO, large dialysis organization; BMI, body mass index; ASHD, atherosclerotic heart disease; CHF, congestive heart failure; CVA/TIA, cerebrovascular accident/transient ischemic attack; PVD, peripheral vascular disease; CMS, Centers for Medicare and Medicaid Services; RUCA, rural-urban commuting area (in which lower numbers are assigned to more urban areas); PEW, protein energy wasting; PTH, parathyroid hormone.

Rates of Fracture

In the national Medicare cohort, 44,014 fractures occurred in 716,962 person-years, and in the LDO Medicare cohort, 8767 fractures occurred in 143,426 person-years. Distribution of the anatomic site of fracture for both cohorts is shown in Supplemental Table 4. It varied little between cohorts. Hip fractures accounted for approximately 35% of fractures in both cohorts over all study years. The overall fracture rate was 6.14 (95% CI, 6.08 to 6.20) per 100 person-years in the national Medicare cohort and 6.11 (95% CI, 5.98 to 6.24) per 100 person-years in the LDO Medicare cohort. Rates varied by region in both cohorts: observed fracture rates ranged from 5.36 to 7.83 in the national Medicare cohort, a 1.46-fold rate difference (Figure 1). Rates ranged from 5.27 to 8.00 in the LDO Medicare cohort, a 1.52-fold rate difference (not shown).

Figure 1.

Observed fracture rate in the national Medicare cohort by region compared with the overall national rate. Error bars represent 95% confidence intervals.

Geographic Variation in Fracture

Unadjusted and fully adjusted CMS region–level rate ratios for fracture in the national Medicare cohort are shown in Figure 2 (the reference was the respective overall fracture rate in the cohort). For the national Medicare cohort, the unadjusted rate ratio for fracture varied from 0.83 (95% CI, 0.81 to 0.85) to 1.20 (95% CI, 1.18 to 1.23), a 1.45-fold rate difference; the SD_RR was 0.16. After adjustment for basic demographics, BMI, cause of ESRD, and dialysis duration, the SD_RR decreased to 0.13, indicating that 18.8% of fracture variation was explained by this adjustment. Additional adjustment for comorbid conditions and functional status had minimal effect on the variation in fracture rates (SD_RR = 0.12), and this remained unchanged after additional adjustment for urban-rural status. Finally, adjustment for latitude reduced the SD_RR to 0.10 (a 12.5% reduction in fracture variation), resulting in a 1.33-fold overall variation in rate ratios across regions in the final adjusted model.

Figure 2.

Unadjusted and fully adjusted region–level rate ratios for fracture in the national Medicare cohort compared with the overall national rate. Error bars represent 95% confidence intervals.

We conducted similar analyses in the LDO Medicare cohort, except that adjustments for PEW and laboratory values (corrected calcium, phosphorus, and PTH) were made in the final model. The first adjustment for demographics, ESRD cause, and dialysis duration decreased the SD_RR from 0.16 to 0.14, explaining 16.0% of fracture variation. Adjustment for comorbid conditions and functional status had no effect on the SD_RR (<0.01% of variation), whereas urban-rural status modestly decreased it to 0.13 (5.3% of variation). Latitude adjustment reduced the SD_RR to 0.11 (11.3% of variation). Final adjustment for PEW and laboratory values explained 4.4% of variation, with the SD_RR decreasing to 0.10. As with the national Medicare sample, there was a 1.33-fold overall variation in rate ratios across regions in the final adjusted model. Replacement of categorical values with splined continuous values for calcium, phosphorus, and PTH did not materially change the results (data not shown).

On the basis of the fully adjusted models, factors associated with higher fracture risk in the national Medicare cohort were age ≥65 years, female sex, white race, prior fracture, history of osteoporosis, lower BMI, worse functional status, longer time on dialysis, comorbid conditions, and diabetes as the primary cause of ESRD (Table 2); results were similar for the LDO Medicare cohort (data not shown). Latitude was associated with fracture in the national Medicare cohort. Fracture rates were higher in the north and middle areas of the country compared with in the south (north rate ratio, 1.18; 95% CI, 1.13 to 1.23; middle rate ratio, 1.13; 95% CI, 1.10 to 1.17). In the LDO Medicare cohort (not shown), the effects of latitude were not as pronounced (north rate ratio, 1.10; 95% CI, 0.95 to 1.21; middle rate ratio, 0.99; 95% CI, 0.96 to 1.07). In contrast, degree of urbanicity was not associated with fracture risk in either cohort.

Table 2.

Factors associated with fracture in the national Medicare cohort

Variable	RR (95% CI)	P Value
Geographic factors
CMS region
1	0.94 (0.90 to 0.99)	0.02
2	0.85 (0.82 to 0.88)	<0.001
3	1.04 (1.01 to 1.07)	<0.01
4	1.11 (1.08 to 1.15)	<0.001
5	1.13 (1.10 to 1.16)	<0.001
6	0.95 (0.92 to 0.98)	0.002
7	1.10 (1.06 to 1.15)	<0.001
8/10	1.03 (0.99 to 1.07)	0.14
9	0.88 (0.86 to 0.91)	<0.001
Latitude
South	1 (reference)
Middle	1.13 (1.10 to 1.17)	<0.001
North	1.18 (1.13 to 1.23)	<0.001
Urban-rural status
1 (RUCA 1–3)	1 (reference)
2 (RUCA 4–6)	1.01 (0.98 to 1.04)	0.70
3 (RUCA 7–10)	1.02 (0.99 to 1.05)	0.30
Demographic factors
Age, yr
45–64	1 (reference)
19–44	0.66 (0.64 to 0.69)	<0.001
65–74	1.26 (1.23 to 1.29)	<0.001
≥75	1.65 (1.61 to 1.69)	<0.001
Race
White	1 (reference)
Black	0.49 (0.48 to 0.50)	<0.001
Other	0.73 (0.70 to 0.76)	<0.001
Sex
Men	1 (reference)
Women	1.35 (1.32 to 1.38)	<0.001
Kidney disease factors
Cause of ESRD
Diabetes	1 (reference)
Hypertension	0.90 (0.88 to 0.93)	<0.001
GN	0.85 (0.81 to 0.88)	<0.001
Other	0.94 (0.91 to 0.97)	<0.001
Dialysis duration, yr
<3	1 (reference)
3 to <5	1.11 (1.08 to 1.13)	<0.001
≥5	1.17 (1.14 to 1.19)	<0.001
Year
2007	1 (reference)
2008	1.00 (0.98 to 1.03)	0.80
2009	0.94 (0.92 to 0.97)	<0.001
2010	0.94 (0.91 to 0.96)	<0.001
Comorbidity factors
CHF	1.09 (1.07 to 1.12)	<0.001
CVA/TIA	1.09 (1.07 to 1.12)	<0.001
ASHD	1.05 (1.02 to 1.07)	<0.001
PVD	1.05 (91.03 to 1.07)	<0.001
Dysrhythmia	1.13 (1.11 to 1.16)	<0.001
Other cardiac	1.08 (1.06 to 1.10)	<0.001
Diabetes	1.13 (1.10 to 1.16)	<0.001
Osteoporosis	1.37 (1.32 to 1.43)	<0.001
Prior fracture	1.82 (1.76 to 1.87)	<0.001
Nutritional/frailty factors
BMI, kg/m2
0 to <18	1.01 (0.96 to 1.07)	0.70
18 to <25	1 (reference)
25 to <30	0.85 (0.83 to 0.87)	<0.001
30 to <35	0.80 (0.78 to 0.83)	<0.001
35 to <40	0.76 (0.73 to 0.78)	<0.001
≥40	0.70 (0.67 to 0.73)	<0.001
Missing	0.95 (0.89 to 1.01)	0.10
Functional status score
0	1 (reference)
1	1.21 (1.18 to 1.24)	<0.001
2	1.44 (1.40 to 1.48)	<0.001
3	1.67 (91.62 to 1.72)	<0.001

Because of the small population sizes of regions 8 and 10, these regions were combined. RR, rate ratio; 95% CI, 95% confidence interval; CMS, Centers for Medicare and Medicaid Services; RUCA, rural-urban commuting area; CHF, congestive heart failure; CVA/TIA, cerebrovascular accident/transient ischemic attack; ASHD, atherosclerotic heart diseases; PVD peripheral vascular disease; BMI, body mass index.

Sensitivity Analyses

Removing the CMS region from the final models yielded similar results for latitude in the national Medicare cohort (north rate ratio, 1.20; 95% CI, 0.99 to 1.23; middle rate ratio, 1.02; 95% CI, 0.99 to 1.05); the pattern was similar in the LDO Medicare cohort (north rate ratio, 1.18; 95% CI, 0.97 to 1.25; middle rate ratio, 1.01; 95% CI, 0.97 to 1.06). Finally, patients with a history of previous transplant, who comprised 6.0% of the national Medicare cohort, were removed from the cohort, and the analysis was conducted again. A 1.32-fold geographic variation between regions remained; the rate ratio for the north was 1.18 (95% CI, 1.13 to 1.23), and the rate ratio for the middle was 1.14 (95% CI, 1.10 to 1.17).

Discussion

In this large population–based study of patients receiving maintenance hemodialysis, we observed geographic variation in the risk of fractures; <50% of this variation was explained by differences in case mix, markers of poor functional status or PEW, or other established risk factors. The latitude at which patients received dialysis explained a substantial amount of variation in fracture rates (approximately 11%), with patients at higher latitudes being at greater risk. These data provide some evidence that geographic location may affect a patient’s risk of fracture, independent of other established risk factors.

Because geographic variation in the risk of several other outcomes in patients on dialysis is recognized (24–30), the geographic variation in fracture risk is not surprising. New in this study is the finding that latitude is also predictive and independent of geographic location. The gradation of fracture risk from north (highest) to south occurred in both cohorts, suggesting that environmental factors may play a role. The mechanism by which environmental factors might influence fracture risk is unknown. One possibility is the confluence of sunlight exposure, latitude, and nutritional vitamin D [25(OH)D] levels. Latitude, a likely proxy for sunlight exposure, has been associated with fracture risk in ecologic studies of the general population (36,40,41). Individuals in more solar–rich environments generally have higher levels of 25(OH)D, which itself is associated with lower fracture risk (42). Although vitamin D status affects calcium absorption and bone health, it is uncertain whether deficiency of 25(OH)D contributes to risk of fracture in patients receiving maintenance dialysis, in whom perturbations in vitamin D metabolism are common.

To determine how geographic variation is associated with fracture risk, we attempted case mix adjustment across a variety of novel domains. For example, we investigated functional status and PEW, both of which we found, not unexpectedly, to be associated with fracture. In contrast, the routinely measured mineral bone disorder of CKD biochemical parameters were not associated with fracture risk in our study, regardless of whether they were treated as categorical or continuous splined variables, consistent with prior epidemiologic studies evaluating fracture risk among patients receiving hemodialysis (8–11). This suggests that using PTH or any combination of biochemistries as a surrogate for bone histology is inherently problematic when a single laboratory measurement is used for each analyte, which was the case in our LDO Medicare cohort.

However, even after substantial case mix adjustment across many domains, more than one half of the geovariation in fracture risk remained unexplained. Possibly, factors such as bone architecture, bone quality, or other aspects of mineral metabolism are the underlying determinants of fracture risk in the hemodialysis population. Bone quality is challenging to assess without bone biopsy and histomorphometric evaluation. Although some circulating markers are associated with the degree of bone turnover and formation rates as well as with other aspects of bone quality (43), these are not routinely assessed in the clinical setting. Furthermore, these markers are imperfect proxies for bone status, with the gold standard remaining the invasive and complex procedure of bone biopsy. Knowledge of bone quality might, in theory, explain geovariation in fracture rates, but why bone quality would vary by region cannot be readily explained.

Our study has important limitations. Residual or unmeasured confounding is possible in any observational study; although we accounted for several important predictors of fracture, we lacked information on bone mineral density (BMD), an important clinical variable predictive of fracture risk in the general population. However, BMD is not commonly assessed in patients on maintenance dialysis, and the association of BMD and fracture in patients with renal failure may be weaker than in those with normal kidney function (44,45). Because of our claims-based approach, important clinical variables, such as smoking status, physical activity, and fall history, could not be ascertained. Additionally, although we used the commonly accepted CMS region designation for our analyses, this classification system may identify differing clinical practice patterns, thus blunting the ability to fully distinguish between different care patterns. Our classification of latitude into three categories was a reasonable approach, but the latitude cut points were arbitrarily selected. Finally, we did not have information on medical treatments, and medications used could possibly modify fracture risk. However, these weaknesses were counterbalanced by the relatively large sample sizes, the fact that two cohorts were examined, and the finding that the sensitivity analyses generally supported the results of the primary analyses.

In conclusion, geographic variation in risk of fracture seems to exist in the United States dialysis population, and latitude is a key contributor. However, even after extensive adjustment, geovariation remained, suggesting that other factors, perhaps environmental in nature, may contribute to the risk of fracture. Given the dire consequences of fracture in patients receiving maintenance dialysis, additional work examining the geovariation in and other determinants of fracture should be undertaken.

Disclosures

J.B.W., J.L., D.T.G., K.M.N., and A.J.C. are employed by the Chronic Disease Research Group (Minneapolis, MN), which receives research support from Amgen, Inc. (Thousand Oaks, CA). H.S.W., K.C., and B.D.B. are employed by Amgen, Inc. J.L. has provided consultation to Daiichi Sankyo (Tokyo, Japan). D.T.G. has provided consultation to GlaxoSmithKline (Brentford, United Kingdom) and DaVita Clinical Research (Minneapolis, MN). A.J.C. has provided consultation to Amgen, Inc.; Relypsa (Redwood City, CA); DaVita Clinical Research; NxStage (Lawrence, MA); Keryx (Boston, MA); and ZS Pharma (Coppell, TX).