Hip Fracture Risk among Hemodialysis-Dependent Patients Prescribed Opioids and Gabapentinoids

Significance Statement

Among the general population, use of opioids have generated concern regarding their effect on fall risk and bone metabolism. Also, association of opioids or gabapentinoid use with fractures has been described in the general population. Patients with ESKD on hemodialysis are at high risk for falls and retain unique bone pathology related to renal osteodystrophy; consequently, they are about four times more likely than individuals in the general population to experience a hip fracture. In a case-control study involving 4912 patients who are dependent on hemodialysis and experience a first-time hip fracture and 49,120 controls, the authors reported an association between hip fractures and opioid use, but not gabapentinoid use. Increasing cumulative opioid exposure conferred a stepwise increase in hip fracture risk. These findings highlight potential detriment with opioid use in this high-risk subpopulation.

Abstract

Background

Despite opioids’ known association with hip fracture risk in the general population, they are commonly prescribed to patients with ESKD. Whether use of opioids or gabapentinoids (also used to treat pain in patients with ESKD) contributes to hip fracture risk in patients with ESKD on hemodialysis remains unknown.

Methods

In a case-control study nested within the US Renal Data System, we identified all hip fracture events recorded among patients dependent on hemodialysis from January 2009 through September 2015. Eligible cases were risk-set matched on index date with ten eligible controls. We required >1 year of Medicare Parts A and B coverage and >3 years of part D coverage to study cumulative longer-term exposure. To examine new, short-term exposure, we selected individuals with >18 months of Part D coverage and no prior opioid or gabapentinoid use between 18 and 7 months before index. We used conditional logistic regression to estimate unadjusted and multivariable-adjusted odds ratios (ORs) and 95% confidence intervals (95% CI).

Results

For the longer-term analyses, we identified 4912 first-time hip fracture cases and 49,120 controls. Opioid use was associated with increased hip fracture risk (adjusted OR, 1.39; 95% CI, 1.26 to 1.53). Subgroups of low, moderate, and high use yielded adjusted ORs of 1.33 (95% CI, 1.20 to 1.47), 1.53 (95% CI, 1.36 to 1.72), and 1.66 (95% CI, 1.45 to 1.90), respectively. The association with hip fractures was also elevated with new, short-term use (adjusted OR, 1.38; 95% CI, 1.25 to 1.52). There were no associations between gabapentinoid use and hip fracture.

Conclusions

Among patients dependent on hemodialysis in the United States, both short-term and longer-term use of opioid analgesics were associated with hip fracture events.

Hip fractures are debilitating events of significant consequence in patients with ESKD. In patients who undergo maintenance hemodialysis (HD), the risk of hip fractures is at least four times that of the general population,^1,2 and as would be expected, the associated cost and mortality risk are also much higher.^3,4 The elevated risk is typically attributed to mineral and bone disorders, dysautonomia, acidosis, cachexia, and inflammation. However, patients with ESKD take almost 20 pills per day,⁵ and commonly prescribed medications may also increase their risk of hip fractures.

The experience of HD is fraught with uncomfortable symptoms, pain being chief among them. As much as half of patients dependent on HD report difficulties with pain.⁶ From generalizable aches to more specific neuropathies to intermittent vascular claudication, pain is prevalent among patients with kidney disease.⁶ Additionally, the necessity of obtaining vascular access and the abrupt process of HD not only generates new discomforts, but likely exacerbates ongoing episodes of pain.^7,8 Opioid use has recently moved into focus as the fastest growing form of substance abuse.⁹ Between 2006 and 2008, almost 40% of patients dependent on HD in the United States with Medicare Parts A, B, and D as their primary payer received at least one opioid prescription quarterly.¹⁰ Although addressing pain is a very important aspect of care, the risks involved with commonly prescribed medications for pain should be understood.

In the general population, several observational studies and meta-analyses have demonstrated an association between hip fractures and opioid use.^11,12 Sedation, cognitive impairment, and dizziness are all potential medication side effects that may increase fall risk. Impaired endogenous sex steroid production has also been described as reducing both bone mineral density and muscle mass.^13,14 Opioid receptors are present in substantial quantities in osteoblasts, suggesting a role for modifying bone metabolism.^15,16 Thus, understanding the risk that opioid use may present to patients dependent on HD, who already retain a markedly elevated risk for hip fractures and experience a distinct bone pathology that differs from osteoporosis, is worthy of further exploration. We aim to uncover any potential association that may exist between opioid use and hip fracture events among patients with ESKD dependent on HD.

Methods

Study Design and Source Population

We conducted a case-control investigation, among patients dependent on HD aged ≥18 years, recorded in the US Renal Data System (USRDS) between January 1, 2006 and September 30, 2015, as previously reported.^17,18 To identify hip fracture cases, we captured all inpatient claims containing the International Classification of Diseases, Ninth Edition diagnosis codes 820.xx and 821.xx. We further specified that cases have a corresponding International Classification of Diseases, Ninth Edition surgical procedure code, either 78.55, 79.15, 81.52, 79.05, 79.25, or 81.40. We used a pooled strategy to efficiently select controls. More specifically, we randomly selected 10×n controls per n cases on each index date, without specifically linking to an individual case (M:N matching).^19,20 The index date was defined as the date of the hip fracture diagnosis. Lastly, the randomly selected controls for each particular index date could still be identified as cases at a later date. Eligible patients were required to have a completed Medical Evidence Report form in the versions coming into use in 1995 or later, that accurately documented demographic data and treatment start date. Additionally, we required at least 1 year (≥365 days) of Parts A and B coverage, at least 3 years (≥1095 days) of Part D with low-income subsidy coverage, and at least 90 days of in-center HD before the hip fracture date or corresponding index date for controls (Figure 1). We excluded patients with preemptive kidney transplantation or a recorded prior hip fracture. To examine short-term, new-use exposure, we conducted a separate analysis in which the requirement of Part D and low-income subsidy coverage was reduced to 18 months (≥547 days) before index date; however, only patients without filled opioid prescriptions in the preceding year (>6–18 months) before index date were eligible participants (Supplemental Figure 1).

Figure 1.

Based on selection criteria, 4912 hip fracture cases were M:N matched with 49,120 controls. Subjects studied were required to have 1 year of Parts A and B coverage, 3 years of Part D with low-income subsidy coverage, and no prior hip fracture or kidney transplant.

Exposure of Interest

We identified Part D claims for opioids, specifically for buprenorphine, codeine, fentanyl, hydrocodone, hydromorphone, meperidine, methadone, morphine, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol. Gabapentinoid use was classified using claims for gabapentin and pregabalin. By examining Part D claims for the proportion of days covered by filled prescriptions of gabapentinoids and opioids, we categorized degrees of medication use before the index date. “Any use” was defined as having at least one prescription filled in the 3 years before index date. “Moderate use” was reserved for patients with ≥20% and <80% of the proportion of days covered by filled prescriptions. We categorized “high use” as patients with ≥80% of the proportion of days covered. We further examined the association with hip fracture events per additional month of opioid or gabapentinoid use. The short-term, new-user analysis specifically looked at opioid exposure in the 6 months before index date among a separate set of cases and controls with no use in the year before index date (between the seventh and 18th month before index date).

Covariates

The required Medical Evidence Form provided information regarding age, sex, race (white, black, other), Hispanic ethnicity, body mass index (BMI), census division, and duration of dialysis-dependence before index date (vintage). With at least 1 year of Parts A and B coverage, we aimed to record comorbid conditions that could influence hip fracture risk. From the institutional claims files and physician/supplier claims files of USRDS (see Supplemental Table 1), we identified the presence of diagnosis codes that corresponded with hypertension, diabetes mellitus, coronary artery disease, cerebrovascular disease, peripheral vascular disease, arrhythmia, rheumatologic disorder, osteoporosis, depression, and tobacco use (see Supplemental Table 1 for specifications). Prior use of other potentially influential medications, such as bisphosphonates, corticosteroids, proton-pump inhibitors, and selective-serotonin reuptake inhibitors, were determined by any Part D claims filled in the 3 years before index date.

Statistical Analyses

We estimated the association between hip fracture case (versus control) status and prior medication use in both unadjusted and multivariable-adjusted conditional logistic regression models. We expressed these estimates as odds ratios (ORs) with corresponding 95% confidence intervals (95% CIs). Table 1 illustrates the patient characteristics and comorbid conditions that were incorporated into the multivariable analysis. After assessing the exposure of opioid use and gabapentinoid use among subgroups, we conducted interaction analyses with categories of age, sex, race, ethnicity, BMI, peripheral vascular disease, and cinacalcet use. Using the adjusted OR as an approximation of relative risk, we estimated the attributable portion to be OR/(1−OR). We then calculated a population attributable fraction by multiplying the proportion of cases with any opioid exposure by this approximation of the attributable portion. We then generated models, looking individually (and independently) at gabapentin, pregabalin, codeine, fentanyl, hydrocodone, oxycodone, tramadol, and uncommon opioids as a category. We also examined the association after excluding patients with both gabapentin and opioid exposure, and in a separate analysis, after excluding patients with osteoporosis. Lastly, we examined the association after incorporating cinacalcet use and orally available vitamin D receptor activator use (calcitriol, doxercalciferol, paricalcitol) into the model.

Table 1.

Characteristics of hip fracture cases and controls

Variable	Cases, n=4912, Mean±SD or n (%)	Controls, n=49,120, Mean±SD or n (%)
Age, yr	70.5±12.2	61.2±14.2
Median [IQR]	72 [63–80]	62 [52–72]
Sex
Female	2861 (58.3)	25,422 (51.8)
Male	2051 (41.8)	23,698 (48.2)
Ethnicity/race
Non-Hispanic black	1439 (29.3)	23,976 (48.8)
Non-Hispanic white	1939 (39.5)	12,316 (25.1)
Non-Hispanic other	412 (8.4)	3220 (6.6)
Hispanic white	1084 (22.1)	9119 (18.6)
Hispanic other	38 (0.8)	489 (1.0)
BMI, kg/m2; missing in n=959	28.2±7.5	30.2±8.7
Median [IQR]	26.7 [23.1–31.9]	28.4 [24.0–34.8]
Underweight	209 (4.3)	1616 (3.4)
Normal	1660 (34.3)	13,428 (27.9)
Overweight	1373 (28.4)	12,548 (26.1)
Obese	1234 (25.5)	14,324 (29.8)
Severely obese	362 (7.5)	6235 (13.0)
Dialysis vintage, yr	5.3±3.5	6.1±3.8
Median [IQR]	4.5 [2.7–7.1]	5.3 [3.5–8.1]
Census division
New England	150 (3.1)	1518 (3.1)
Middle Atlantic	503 (10.2)	5807 (11.8)
East North Central	633 (12.9)	6436 (13.1)
West North Central	257 (5.2)	2192 (4.7)
South Atlantic	1026 (20.9)	11,760 (23.9)
East South Central	380 (7.7)	4170 (8.5)
West South Central	828 (16.9)	8342 (17.0)
Mountain	274 (5.6)	1829 (3.7)
Pacific	861 (17.5)	7066 (14.4)
Comorbidities
Hypertension	4836 (98.5)	47,840 (97.4)
Diabetes mellitus	3764 (76.6)	32,275 (65.7)
Coronary artery disease	2854 (58.1)	22,200 (45.2)
Peripheral vascular disease	2073 (42.2)	15,629 (31.8)
Cerebrovascular disease	1326 (27.0)	9640 (19.6)
Heart failure	3188 (64.9)	26,300 (53.5)
Arrhythmia	1548 (31.5)	10,643 (21.7)
Osteoporosis	32 (0.7)	141 (0.3)
Rheumatological disorder	181 (3.7)	2030 (4.1)
Tobacco use	756 (15.4)	8190 (16.7)
Depression	1237 (25.2)	8694 (17.7)
History of steroid use	1481 (30.1)	14,677 (29.9)
History of bisphosphonate use	296 (6.0)	1211 (2.5)
History of proton-pump inhibitor use	3426 (69.8)	31,168 (63.5)
History of SSRI use	1814 (36.9)	13,632 (27.8)
Opioid use
Any opioid	4359 (88.7)	42,218 (86.0)
Codeine	937 (19.1)	9233 (18.8)
Hydrocodone	3429 (69.8)	32,489 (66.1)
Hydromorphone	251 (5.1)	2677 (5.5)
Oxycodone	1712 (34.9)	18,384 (37.4)
Tramadol	1568 (31.9)	13,841 (28.2)
Uncommon opioids	1454 (29.6)	12,410 (25.3)
Gabapentinoid use
Any gabapentinoid	1761 (35.9)	16,382 (33.4)
Gabapentin	1582 (32.2)	14,548 (29.6)
Pregabalin	420 (8.6)	4053 (8.3)
Primary cause of ESKD (missing in n<10)
Diabetes mellitus	2777 (56.5)	23,304 (47.5)
GN/vasculitis	260 (5.3)	5056 (10.3)
Interstitial/pyelonephritis	134 (2.7)	1236 (2.5)
Hypertensive/large vessel disease	1335 (27.2)	14,692 (29.9)
Cystic/hereditary/congenital	85 (1.7)	1308 (2.7)
Neoplasm/tumor	44 (0.9)	328 (0.7)
Miscellaneous	277 (5.6)	3191 (6.5)

IQR, interquartile range; BMI, body mass index; SSRI, selective serotonin reuptake inhibitor.

Two variables included in the multivariable-adjusted analysis were incomplete: BMI (1.9%) and cause of ESKD (<0.01%). We assumed that the missing observations were not related to itself and that sufficient information from other variables existed to recover the distribution of BMI. Assuming data to be missing at random, we utilized multiple imputations using chained equations to generate ten imputed data sets that were subsequently analyzed.²¹ Imputations occurred under fully conditional specification where we used truncated regression to impute BMI and multinomial logistic regression for cause of ESKD. We ran a single multiple imputation model that included all variables from the adjusted analysis model (including the indicator for case/control) as well as the exposure variable with four categories (no use, proportion of days covered [PDC; sum of the number of days of drug supplied in prescriptions divided by the number of days of the interval] <20%, PDC ≥20% and <80%, and PDC≥80%). We estimated parameters and their SEMs after applying the analysis model to each imputed data set separately. These estimates and their SEMs were then combined using the rules of Rubin.²² We performed a complete case analysis as a sensitivity analysis. We used SAS software (version 9.4; SAS Institute) and StataMP (version 14; StataCorp) for analyses. An Institutional Review Board at Baylor College of Medicine approved this study (approval number H-36408).

Results

Patient Selection and Baseline Characteristics

Using the stated criteria, we identified 4912 first-time hip fracture cases between January 1, 2009 and September 30, 2015. Correspondingly, 49,120 control patients who satisfied the same inclusion requirements were M:N matched on index date (Figure 1). Cases were older than controls by an average of 9.3 years and had a more substantial burden of comorbid conditions. A larger proportion of non-Hispanic whites were case-compared with controls (39.5% versus 25.1%). Controls had a higher BMI, and a larger proportion of controls were obese and severely obese (Table 1).

Comparable proportions of cases and controls, 86.4% and 83.7%, respectively, had any exposure to opioids within the 3 years before index. Participants with a history of opioid use were younger and more obese, but they also had more comorbid conditions and a larger proportion of these individuals were women and non-Hispanic white (Supplemental Table 4). The proportions of cases with “high use” of opioids (11.0%) were also fairly equivalent to controls with “high use” (10.4%). Close to one third of controls and 35.9% of cases had any filled prescriptions for gabapentinoids in the 3 years before index date. Only 5.5% and 4.8% of cases and controls demonstrated “high use” of gabapentinoids.

Associations between Opioid Use and Hip Fractures

The unadjusted conditional OR of hip fracture status with any opioid use versus no use was 1.29 (95% CI, 1.18 to 1.41). The multivariable-adjusted analysis generated an OR of 1.39 (95% CI, 1.26 to 1.53). Considering hip fractures rare events, we used this OR as an approximation of relative risk and calculated a population attributable fraction of 24.9%. The unadjusted ORs for low (<20% of proportion of days covered), moderate (≥20% and <80% of proportion of days covered), and high (≥80% of proportion of days covered) opioid use were 1.24 (95% CI, 1.12 to 1.36), 1.43 (95% CI, 1.28 to 1.59), and 1.32 (95% CI, 1.17 to 1.50). The multivariable-adjusted ORs of hip fracture status for low, moderate, and high opioid use were 1.33 (95% CI, 1.20 to 1.47), 1.53 (95% CI, 1.36 to 1.72), and 1.66 (95% CI, 1.45 to 1.90) (Table 2). The association with hip fracture events remained when examining individual opioids in adjusted analyses (Table 3). When examined as a continuous variable, for each additional month of cumulative opioid exposure, the unadjusted OR was 1.003 (95% CI, 1.001 to 1.005) and the multivariable-adjusted OR was 1.006 (95% CI, 1.004 to 1.008). With the inclusion of calcimimetic (cinacalcet) and vitamin D receptor activator (calcitriol, paricalcitol, and doxercalciferol) use in the model, the adjusted OR with any, low, moderate, and high use remained unchanged (see Supplemental Table 3). Complete-case analysis also yielded almost identical estimations to our primary analysis (see Supplemental Table 3). After excluding patients with both opioid and gabapentin exposure, the findings also remained consistent (see Supplemental Table 3). The results were also unchanged with the exclusion of patients with a history of osteoporosis (see Supplemental Table 3). Finally, the ORs for any, low, moderate, and high opioid use after excluding patients with prior vertebral, upper extremity, or distal lower extremity fracture were 1.29 (95% CI, 1.16 to 1.45), 1.27 (95% CI, 1.14 to 1.42), 1.33 (95% CI, 1.16 to 1.52), and 1.50 (95% CI, 1.28 to 1.76), respectively. We did not identify any significant interactions with age, sex, ethnicity, vintage, peripheral vascular disease, BMI, or cinacalcet use, and the estimated associations among these subgroups were largely unchanged. However, we did note a nominally significant interaction with race (P=0.007). Among black participants, the multivariable-adjusted conditional OR of hip fracture status with any versus no opioid use resulted 1.15 (95% CI, 0.95 to 1.38). However, among nonblack participants, the multivariable-adjusted conditional OR was 1.49 (95% CI, 1.32 to 1.68). Finally, in the separate analysis examining the association with new opioid use in the 6 months before index date, we determined an unadjusted OR of 1.42 (95% CI, 1.30 to 1.56), which remained essentially unchanged after multivariable adjustment (OR, 1.38; 95% CI, 1.25 to 1.52).

Table 2.

Opioid and gabapentinoid use in hip fracture cases and controls and measures of association

Medication	Category of Exposure	Cases, n=4912, n (%)	Controls, n=49,120, n (%)	Unadjusted Odds Ratio (95% CI)	P Value	Adjusted Odds Ratio (95% CI)a	P Value
Opioids	No use	553 (11.3)	6902 (14.0)	1.00 (Referent)	—	1.00 (Referent)	—
	Any use	4359 (88.7)	42,218 (86.0)	1.29 (1.18 to 1.41)	<0.001	1.39 (1.26 to 1.53)	<0.001
	<20% PDC	2777 (56.5)	28,025 (57.0)	1.24 (1.12 to 1.36)	<0.001	1.33 (1.20 to 1.47)	<0.001
	≥20% to <80% PDC	1026 (20.9)	8953 (18.2)	1.43 (1.28 to 1.59)	<0.001	1.53 (1.36 to 1.72)	<0.001
	≥80% PDC	556 (11.3)	5240 (10.7)	1.32 (1.17 to 1.50)	<0.001	1.66 (1.45 to 1.90)	<0.001
Gabapentinoids	No use	3151 (64.2)	32,738 (66.7)	1.00 (Referent)	—	1.00 (Referent)	—
	Any use	1761 (35.9)	16,382 (33.4)	1.15 (1.08 to 1.23)	<0.001	1.05 (0.98 to 1.12)	0.19
	<20% PDC	793 (16.1)	7569 (15.4)	1.12 (1.03 to 1.22)	0.009	1.08 (0.99 to 1.18)	0.10
	≥20% to <80% PDC	696 (14.2)	6473 (13.2)	1.19 (1.09 to 1.30)	<0.001	1.06 (0.96 to 1.17)	0.3
	≥80% PDC	272 (5.5)	2340 (4.8)	1.15 (1.00 to 1.32)	0.05	0.92 (0.80 to 1.07)	0.3

95% CI, 95% confidence interval; PDC, proportion of days covered (sum of the number of days of drug supplied in prescriptions divided by the number of days of the interval; here: 1095 days).

^aAdjusting for demographics, body mass index, dialysis vintage, comorbidities, cause of ESKD, Census region, and other medication (steroid, selective serotonin reuptake inhibitor, proton-pump inhibitor, bisphosphonates) exposures.

Table 3.

Specific opioid use in hip fracture cases and controls and measures of association

Opioid Exposure (any use)	Cases, n=4912, n (%)	Controls, n=49,120, n (%)	Unadjusted Odds Ratio (95% CI)	P Value	Adjusted Odds Ratio (95% CI)a	P Value
Codeine	937 (19.1)	9233 (18.8)	1.02 (0.95 to 1.10)	0.63	1.07 (0.99 to 1.16)	0.08
Fentanyl	324 (6.6)	2137 (4.4)	1.55 (1.38 to 1.75)	<0.001	1.39 (1.22 to 1.58)	<0.001
Hydrocodone	3429 (69.8)	32,489 (66.1)	1.18 (1.11 to 1.26)	<0.001	1.16 (1.09 to 1.24)	<0.001
Hydromorphone	251 (5.1)	2677 (5.5)	0.93 (0.82 to 1.07)	0.51	1.24 (1.08 to 1.43)	0.003
Oxycodone	1712 (34.9)	18,384 (37.4)	0.89 (0.84 to 0.95)	<0.001	1.14 (1.06 to 1.22)	<0.001
Tramadol	1568 (31.9)	13,841 (28.2)	1.20 (1.12 to 1.27)	<0.001	1.16 (1.09 to 1.24)	<0.001
Uncommon opioidsb	1454 (29.6)	12,410 (25.3)	1.26 (1.18 to 1.35)	<0.001	1.27 (1.18 to 1.36)	<0.001

95% CI, 95% confidence interval.

^aAdjusting for demographics, body mass index, dialysis vintage, comorbidities, cause of ESKD, Census region, and other medication (steroid, selective serotonin reuptake inhibitor, proton-pump inhibitor, and bisphosphonates) exposures.

^bUncommon opioids include buprenorphine, meperidine, methadone, oxymorphone, pentazocine, propoxyphene, and tapentadol.

Associations between Gabapentinoid Use and Hip Fractures

The unadjusted conditional OR of hip fracture status with any versus no gabapentinoid use was 1.15 (95% CI, 1.08 to 1.23). The unadjusted ORs for low, moderate, and high gabapentinoid use were 1.12 (95% CI, 1.03 to 1.22), 1.19 (95% CI, 1.09 to 1.30), and 1.15 (95% CI, 1.00 to 1.32), respectively. The multivariable adjustment resulted in null associations with all categories of gabapentinoid use. When looking specifically at pregabalin or gabapentin use, these associations remained null.

Discussion

In this study of hip fractures in the United States HD population, we noted that >80% of patients that had experienced a hip fracture were exposed to opioids in the 3 years before their fracture. Furthermore, about 10% of patients were labeled high users, defined as having opioid prescriptions covering >80% of days in the 3-year period before their fracture. When formally comparing this exposure to that of risk-set matched controls, and adjusting for a number of established hip fracture risk factors and other potential confounders, we found that cases were 39% more likely to have any opioid exposure than controls. If opioid use were eliminated entirely, we estimate that hip fracture incidence would be reduced by up to 25%. This association was further corroborated by a monotonic dose-response relationship as indicated by the increasing risk associated with higher proportion of days covered by opioid pills. Further support is noted by the significant association between hip fracture events and a continuous measure, months of cumulative exposure. A series of sensitivity analyses corroborated these findings.

Among the general population, the association between opioid use and fractures, particularly hip fractures, is well described. In a meta-analysis of eight cohort studies, Teng et al.¹² found that patients with opioid exposure had an 88% increased risk of fracture and 100% increased risk of hip fracture. Another meta-analysis conducted by Ping et al.¹¹ determined that patients with opioid use were 1.54 times more likely to experience a hip fracture. In the high-risk subset of patients dependent on HD, one other study has examined the risk of fractures with opioid use. Ishida et al.²³ found similar results in that patients with opioid use had a 44% increased risk of developing any fracture. Similarly, they did not find a significant fracture risk with pregabalin use and only noted that the highest dose of gabapentin use (>300 mg) was associated with fractures (hazard ratio, 1.38; 95% CI, 1.18 to 1.61).²⁴ We do not dispute that exposure to gabapentinoids above recommended doses for patients dependent on dialysis may affect falls and fracture risk. However, fracture risk can vary substantially by location and patient profile, and the increased fracture risk could have been attributed to distal extremities. Thus, without specifically examining hip fractures, we could not assume that the findings from the general population or Ishida et al. were applicable^23,24. Given the event-rate of hip fractures, we suspect that the 1-year of person-time studied by Ishida et al. may not have been adequate to assess risk site–specific fractures.^23,24 Our study also has the advantage of specifically looking at both the short-term risk of new users as well as the risk associated with chronic exposure. Additionally, and importantly, we were able to report on the risk associated with specific medications.

The lack of association among gabapentinoids suggests that the increased risk among opioid users likely extends beyond risk attributed to pain alone and may implicate pharmacodynamic properties of opioid receptor engagement. The most glaring mechanism of risk is likely the altered sensorium that can affect fall risk. Sedation and cognitive impairment are well known side effects of opioid use, and among the many adverse outcomes reported by Ishida et al.²³ altered mental status and falls were both 28% more likely with opioid use among patients on HD. Given our findings regarding new, short-term opioid use, some portion of risk likely stems from diminished cognition and attentiveness.

In addition to fall risk, opioid use affects bone density and architecture via both altered hormonal feedback²⁵ and direct modification of cellular metabolism. Three major receptors, μ, κ, and δ receptors, are frequently cited as the focus of action. However, ε-receptor engagement in the hypothalamus disturbs gonadotropin-releasing hormone secretion, leading to reduced pituitary gland secretion of luteinizing hormone and follicle-stimulating hormone, and consequently, dampened gonadal production of testosterone and estradiol. Increased prolactin production by μ, κ, and δ receptor agonism also contributes to the hypogonadal state.²⁶ Hypogonadism is already an all too common fate for patients with ESKD, but it is possible that opioid use could relatively exacerbate this condition. Other nongonadal, opioid-induced endocrinopathies are also of concern. Yet, these complicated feedback mechanisms are not completely understood as varied imbalances in hormone release result in conflicting effects on bone metabolism. Prolactin has also been described as stimulating bone resorption.²⁷ Opioid-induced μ-receptor agonism inhibits corticotropin-releasing hormone and downstream cortisol secretion, which results in decreased bone resorption. Thus, in patients with ESKD, the narrative of endocrine-mediated effect on bone metabolism remains muddied.

Opioid receptors are thought to be present on osteoblasts, and both endogenous²⁸ and exogenous opioid exposure affect the bone-building process. In a study of a human osteoblast-like cell line MG-63, decreased osteocalcin production was detected with morphine exposure.¹⁶ This finding has also been clinically observed with decreased osteocalcin levels among pregnant heroin users and their newborns.²⁹ In a cross-sectional analysis of 144 opioid-dependent men on opioid substitution therapy with either methadone, morphine, or buprenorphine and 35 of their age- and BMI-matched counterparts, those on opioid substitution therapy, regardless of type, had significantly lower bone mineral density of lumbar spine, femoral neck, and total hip.³⁰ Among patients maintained on methadone, few small studies have demonstrated lower bone mineral density compared with controls, and the findings appear to be most pronounced in men.^31–33 In a study of 38 men on methadone and 40 age- and weight-matched controls, Grey et al.³² found a 7%–14% decrease in bone mineral density, whereas no distinction was captured among women. In the Danish Osteoporosis Prevention Study, 32 perimenopausal women with opioid exposure were followed for 10 years, and compared with 1984 controls, no differences in the rate of change in bone mineral density were reported.³⁴ Nevertheless, we did not encounter any differences in our estimates that were dependent on sex.

Opioid exposure did not confer the same risk to black participants as it did to nonblack participants. As is the case with kidney transplant recipients using calcineurin inhibitors,³⁵ race-related distinctions in hepatic drug metabolism are not uncommon. The majority of opioid metabolism occurs through CYP3A4 enzymes, and lesser amounts are accounted for by CYP2D6 and UDP glucuronosyltransferases. Compared with white patients, patients of African ancestry have higher variability with regards to CYP2D6 expression. Despite this variability, allelic variants in CYP2D6 enzymes that result in rapid metabolism are more common among patients of African ancestry compared with white patients.³⁶ Additionally, the likelihood of being a poor metabolizer is lower among black Americans.³⁷ Codeine and hydrocodone metabolism lean significantly on CYP2D6 enzymes, and to a lesser degree, tramadol, oxycodone, and methadone require the same set of enzymes. Thus, the greater proportion of rapid metabolizers among black patients may result in relatively less opioid exposure and consequently, less adverse effects.

Few interactions exist between the various opioids and the medications involved in the treatment of the mineral bone disorder associated with CKD. Phosphate binders and vitamin D receptor agonists are largely free from any significant influence of opioids. Cinacalcet is a potent inhibitor of CYP2D6, and several opioids (mainly codeine and tramadol) are metabolized in some part through these enzymes. Patients with ESKD who are prescribed cinacalcet may then have even greater opioid exposure, inviting more adverse effects such as hip fractures. Yet, in our findings, we did not note any significant interaction with cinacalcet use. Our study was not powered to examine this interaction, and our study design does not examine whether cinacalcet use and opioid use overlapped.

In conducting an uncontrolled, observational study, we recognize that potential residual confounding remains a limitation, but we believe that the strength of the association and its relationship with increasing opioid exposure are noteworthy findings. Although we were unable to further specify doses of opioids used or detail temporal opioid exposure on the level of exact days of use, we did capture categorical (proportion of days covered) degrees of exposure and continuous (cumulative months) measures of exposure. Both assessments resulted in increasing risk with higher use. The requirement of Medicare coverage for a significant duration of time could potentially result in a selection bias related to survivorship. A higher incidence rate early after opioid exposure could could have gone uncaptured, and due to a potential depletion of susceptibles phenomenon, we could have underestimated the association. Our corroborated findings in the short-term analysis with a shorter period of required coverage somewhat diminishes this concern. By assessing comorbid conditions and other characteristics during the same period of assessment for medication use, it is possible that some characteristics could follow exposure and, thereby, be a consequence of exposure. Despite this limitation, which if anything, could bias toward the null, we still distinguished a significant association between opioid exposure and hip fractures. Also notable, our study does not include a complete representation of medical therapy directed at mineral bone disorders. Without access to intravenously administered drugs, we were only able to include oral vitamin D receptor activators and cinacalcet use. Lastly, mineral and bone disease biomarkers, namely, calcium and intact parathyroid hormone,³⁸ have known associations with hip fracture risk, and we did not have available laboratory data to incorporate into our models.

In a large study of patients dependent on HD in the United States, we found that patients who suffered hip fractures were more likely to have prior opioid exposure. Additionally, increasing cumulative exposure conferred a stepwise increase in hip fracture risk. Unfortunately, pain management is a critical aspect of care for patients with ESKD, but providers should still be aware of the risks that opioid use present. We recommend that opioid use should be reserved for patients with persistent pain after conservative measures have been implemented and neuropathic pain has been addressed. Even in those circumstances, when dealing with complex pain, opioid use should be part of a multimodal approach, and patients should be cautioned regarding the adverse effects associated with opioid use.

Disclosures

Dr. Navaneethan reports personal fees from Bayer, personal fees from Boehringer-Ingelheim, personal fees from REATA, personal fees from Tricida, and grants from Keryx, outside the submitted work. Dr. Niu reports personal fees from University of California Davis, and nonfinancial support from Xiangya Hospital, Central South University, outside the submitted work. Dr. Winkelmayer reports personal fees from Akebia, personal fees from AstraZeneca, personal fees from Bayer, personal fees from Merck, personal fees from Janssen, personal fees from Vifor FMC Renal Pharma (including Relypsa), outside the submitted work. All remaining authors have nothing to disclose.

Funding

Dr. Vangala was supported by a gift from Dr. and Mrs. Harold Selzman. Dr. Winkelmayer receives support through the endowed Gordon A. Cain Chair in Nephrology at Baylor College of Medicine. The work was also supported using facilities and resources of the Veterans Health Administration Health Services Research and Development Center for Innovations in Quality, Effectiveness and Safety through grant CIN13-413.

Supplemental Material

This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2019090904/-/DCSupplemental.

Supplemental Figure 1. Selection of hip fracture cases and controls with low-income subsidy coverage for 18 months.

Supplemental Table 1. Code algorithms used to identify outcomes and covariables.

Supplemental Table 2. Characteristics of hip fracture cases and controls in short-term analysis.

Supplemental Table 3. Comparison of primary analysis with analysis including vitamin D receptor activators/calcimimetics, complete-case analysis, and analysis excluding dual opioid and gabapentinoid exposure.