Lowering Serum Urate With Urate‐Lowering Therapy to Target and Incident Fracture Among People With Gout

Gout is associated with a higher risk of fracture; however, findings on the associations of hyperuricemia and urate‐lowering therapy (ULT) with the risk of fracture have been inconsistent. We examined whether lowering serum urate (SU) levels with ULT to a target level (i.e., <360 μmoles/liter) reduces the risk of fracture among individuals with gout.


INTRODUCTION
Fractures are a leading cause of morbidity and mortality worldwide (1). The formidable social and economic burden of fractures, especially in the elderly, make their prevention a major public health goal (2,3). Gout is the most common form of inflammatory arthritis and its prevalence and incidence have been increasing over time (4)(5)(6). Previous studies have found that gout is associated with a higher risk of fracture; however, the exact mechanism linking gout to fracture remains unclear (7,8).
Several studies have examined the association between serum urate (SU) levels (regardless of gout status) and the risk of fracture (9)(10)(11)(12)(13)(14)(15). Most of these studies evaluated the relation of levels of SU either continuously or categorically (e.g., quartiles No funding bodies had any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The Health Improvement Network (THIN) is a registered trademark of Cegedim in the UK and other countries. References made to the THIN database are intended to be descriptive of the data asset licensed by IQVIA.
Dr. Wei's work was supported by the Project Program of the National Clinical Research Center for Geriatric Disorders (award 2021LNJJ06), the Natural Science Foundation of Hunan Province (award 2022JJ20100), and the Technology Innovation Program of Hunan Province (award 2022RC1009). Dr. Zeng's work was supported by the National Natural Science Foundation of China (award 82072502), the Project Program of the National Clinical Research Center for Geriatric Disorders (award 2022LNJJ07), and the Science and Technology Innovation Program of Hunan Province (award 2022RC3075). Dr. Lei's work was supported by the National Natural Science Foundation of China (awards 81930071 and U21A20352). 1 or quintiles) to the risk of fracture, but the results were inconsistent (9)(10)(11)(12)(13)(14). One study assessed the association between hyperuricemia and the risk of fracture and reported that the risk of hip fracture was higher among men with hyperuricemia than those with normouricemia (15). A few studies also examined the relation of urate-lowering therapy (ULT) to the risk of fracture (16)(17)(18)(19); however, the results were inconsistent. In addition, none of these studies specifically examined the effect on the risk of fracture of lowering SU levels to target levels (i.e., <360 μmoles/liter) using ULT.
Randomized clinical trials have assessed the effect of treatto-target SU with ULT to below 360 μmoles/liter on the risk of recurrent gout flares, tophi, or radiographic joint damage for people with gout (20)(21)(22); however, these studies were unable to evaluate the effect on the risk of fractures of lowering SU levels to target levels using ULT, owing to the limited statistical power. Using a population-based electronic medical records database, we conducted a cohort study emulating analyses of a hypothetical target trial to examine the effect of lowering SU with ULT to the target level (i.e., <360 μmoles/liter) on the risk of fracture among people with gout.

PATIENTS AND METHODS
Data source. We used data from The Health Improvement Network (THIN, a subset of IQVIA Medical Research Data, provided by Cegedim), an electronic health records database from general practitioners (GPs) in the UK. The THIN consist of 17 million individuals in the UK. The computerized information includes sociodemographics, anthropometric characteristics, lifestyle factors, and details from visits to GPs (i.e., prescriptions, diagnoses from specialist referrals, hospital admissions, and results of laboratory tests). The Read code system is used to code specific diagnoses, whereas a dictionary based on the Multilex classification system is used to code drugs. The validity of THIN for use in clinical and epidemiologic research studies has been demonstrated previously (23,24). The scientific review committee for the THIN database and the institutional review board at Xiangya Hospital approved this study with a waiver of informed consent. This study followed the recommendations of the Strengthening the Reporting of Observational studies in Epidemiology initiative.
Study design and cohort definition. We included participants who were 40-89 years old, had gout between January 1, 2000 and October 31, 2021, and had at least 1 year of continuous enrollment with a GP prior to entering the study (Supplementary Figure 1, available on the Arthritis & Rheumatology website at http://onlinelibrary.wiley.com/doi/10.1002/art. 42504). The diagnosis of gout was based on the presence of at least 1 gout Read code (24)(25)(26). Of the diagnosed gout patients, we identified individuals for whom ULT was initiated (i.e., allopurinol, febuxostat, probenecid, benzbromarone, or sulphinpyrazone) based on the first record of a ULT prescription after the diagnosis of gout. The first ULT prescription date was assigned as each participant's index date. Persons were excluded if they had cancer or any fracture before the index date, or if they had missing values for body mass index (BMI), alcohol consumption, smoking status, socioeconomic deprivation index score, SU level, and estimated glomerular filtration rate (eGFR) before the index date.
Using observational data, we emulated analyses of a hypothetical target trial using a "cloning, censoring, and weighting" approach to evaluate the effect of "achieving the target SU level" (i.e., <360 μmoles/liter) on the risk of fracture in individuals initiating ULT (24,(27)(28)(29). We created a data set with 2 clones of each initiator at baseline and assigned each of the clones to one of the intervention arms (i.e., "achieving the target SU level" arm versus "not achieving the target SU level" arm). For example, we assigned 1 clone to the "achieving the target SU level" arm (i.e., achieving the target SU level during 1 year after the index date) and the other clone to the "not achieving the target SU level" arm (i.e., not achieving the target SU level during 1 year after the index date). "Cloning" makes 2 comparison groups compatible with their observed data at time 0 ( Figure 1). We allowed for a grace period of 1 year after ULT initiation for individuals to achieve the target SU level (21). If clones deviated from their assigned strategy during the first year of follow-up, they would be artificially censored. Specifically, clones assigned to the "achieving the target SU level" arm were censored if they did not achieve the target SU level at the end of the first year of follow-up, and clones assigned to the "not achieving the target SU level" arm were censored if they achieved the target SU level during the first year of follow-up. During the grace period, if an individual experienced an incident fracture or loss to follow-up or died before achieving the target SU level, that person was considered consistent with his/her assignment in both arms (or clones) and contributed the outcome to each of the assigned arms ( Figure 1). The key protocol components are shown in Supplementary Assessment of outcomes. The primary outcome was hip fracture (30,31). Secondary outcomes were composite fractures (fractures at any site) (27), major osteoporotic fractures (hip, vertebral, wrist, or humerus fracture), vertebral fractures, and nonvertebral fractures (32,33). We defined fracture outcomes using Read codes as per previously published studies (27,31,34). Positive predictive values were 91.0% for hip fracture and 88.1% for vertebral fracture (34).

Assessment of covariates.
Covariates prior to the index date were obtained from THIN. These included sociodemographic characteristics (age, sex, socioeconomic deprivation index score, and region), anthropometric characteristics (BMI), lifestyle factors (smoking status and alcohol consumption), SU level, gout duration, comorbidities (hypertension, venous thromboembolism, myocardial infarction, stroke, pneumonia or infection, hyperlipidemia, varicose veins, depression, chronic obstructive pulmonary disease, fall, osteoporosis, atrial fibrillation, osteoarthritis, diabetes, and chronic kidney disease) prior to the index date, as well as medication use (disease-modifying antirheumatic drugs, antihypertensive drugs, antidiabetic drugs, statin, anticoagulants, aspirin, nonsteroidal antiinflammatory drugs [NSAIDs], opioids, nitrates, colchicine, diuretics, systemic glucocorticoids, proton pump inhibitors, and antiosteoporosis drugs [i.e., bisphosphonates, denosumab, alendronate, ibandronate, risedronate, zoledronic acid, teriparatide, and hormone replacement therapy]) within 1 year before the index date. Serum creatinine level was obtained from the database before the index date. The eGFR was calculated from serum creatinine values using the Modification of Diet in Renal Disease formula (35). Finally, we calculated the number of visits to a GP and hospital admissions within 1 year before the index date.
Statistical analysis. We created a dataset with 2 copies of each individual initiating ULT at baseline. Each individual was assigned to both the "achieving the target SU level" and the "not achieving the target SU level" arms. We divided the follow-up time into 5, 1-year time blocks starting from ULT initiation. Replicates assigned to the "achieving the target SU level" arm were artificially censored 1 year after ULT initiation if they did not achieve the target SU level. Replicates assigned to the "not achieving the target SU level" arm were artificially censored if they achieved the target SU level before developing fracture or at any time within 1 year after ULT initiation. Because artificial censoring may lead to potential selection bias, we used inverse probability weighting (IPW) to account for censoring (28). The denominator of the IPW was the probability that a replicate adhered to his/her assigned arm (i.e., uncensored) using the logistic regression which consisted of the baseline covariates described above (see the "Assessment of covariates" section) and the time-varying covariates (i.e., BMI, eGFR, lifestyle factors, comorbidities, medication use, and healthcare utilization) between the index date and the date of artificial censoring. Participants were followed up until the first occurrence of the following events: incident fracture, death, disenrollment from a GP participating in THIN, 5 years of followup, or the end of the study (October 31, 2021). We compared the risk of fracture between 2 weighted comparison groups using a pooled logistic regression model including an indicator for "achieving the target SU level" and adjusting for the year of follow-up (linear and quadratic term), baseline confounders, and time-varying confounders in the weighted population (36,37). The odds ratio generated from this model approximated the hazard ratio (HR) because the outcome is rare (37). We used a robust standard error (SE) to compute 95% confidence intervals (95% CIs) for HR estimates. We estimated the absolute risk Figure 1. Study design of a hypothetical randomized controlled trial ("target trial") on which we modeled our observational data analysis (A), and a diagram of the cloning and censoring process in 4 hypothetical patients (B). Not achieving the target serum urate (SU) level was defined as an SU level ≥360 μmoles/liter within 1 year after index date, and achieving the target SU level was defined as an SU level <360 μmoles/liter within 1 year after index date.* = index date (date of urate-lowering therapy initiation); # = follow end (date of incident hip fracture, death, disenrollment from a general practice participating in THIN, 5 years of follow-up, or the end of the study, whichever occurred first);^= grace period (participants were given 1 year to achieve the target SU level after initiating treatment with urate-lowering therapy). difference of fracture over 5 years by fitting the pooled logistic models with product terms between the "achieving the target SU level" indicator and the year of follow-up variables. The models' predicted values were then used to estimate the risk of fracture from baseline (36). The risk curves were standardized to the baseline variables (38). We used a nonparametric bootstrap analysis with 20 samples to compute the 95% CI for absolute estimates.
We performed several sensitivity analyses to assess the robustness of the study findings. First, we calculated the E value to quantitatively evaluate the minimum residual confounding effect that would nullify an association observed in the primary analyses (39). Second, we performed an analysis among individuals who were enrolled in THIN for at least 1 year and who developed gout during the follow-up (i.e., individuals with incident gout). Third, we performed an analysis in participants whose gout diagnosis was defined by Read code plus receiving medication for gout (i.e., colchicine or NSAIDs). This definition had a positive predictive value of 90% in the General Practice Research Database (40), in which 60% of participants overlap with THIN. Fourth, we performed an analysis exclusively in participants initiating allopurinol to achieve the target SU level. Fifth, we performed an analysis among participants who received antiinflammatory treatments (i.e., glucocorticoids, colchicine, or NSAIDs) during the 1 year before the index date. Sixth, to evaluate whether there is an SU concentration-dependent relation between achieving the target SU level and the risk of hip fracture, we performed an analysis by creating 3 replicates for each initiator at baseline and assigning the 3 replicates to one of the following intervention arms: achieving the target SU level of <300 μmoles/liter (i.e., recommended in the previous British Society for Rheumatology/British Health Professionals in Rheumatology guideline) (41), achieving the target SU level of 300-360 μmoles/liter, and not achieving the target SU level during 1 year after the index date, and comparing the risk of fracture among the 3 comparison groups using the same approach. Finally, we examined the relation of achieving the target SU levels with ULT to the risk of traumatic injury, a control outcome for which we expected a null association. All analyses were conducted using SAS software version 9.4 (SAS Institute), and 2-sided P values less than or equal to 0.05 were considered statistically significant for all tests.
Ethical approval. This study received approval from the medical ethics committee at Xiangya Hospital (project no. 2018091077), with the waiver of informed consent. This study was approved by the THIN scientific review committee (project no. 21SRC003_A1). This work uses deidentified data provided by patients as a part of their routine primary care.

RESULTS
We identified 73,206 participants who met the inclusion criteria and initiated ULT during the study period. Of them, we excluded 44,652 ULT initiators who had cancer, any fracture prior to the index date, or missing values for BMI, alcohol consumption, smoking status, socioeconomic deprivation index score, SU level, and eGFR. The final study cohort consisted of 28,554 participants (Supplementary Figure 1, http://onlinelibrary.wiley.com/doi/10. 1002/art.42504). The mean age was 65.3 years, and 23.7% of the participants were women. The mean BMI was 30.3 kg/m 2 and the mean SU level was 510.3 μmoles/liter. The baseline characteristics of the remaining participants are shown in Table 1.
Of 28,554 ULT initiators, 8,390 achieved the target SU level within 1 year after the index date. The mean of the final SU levels during the 5-year follow-up period was 311.2 μmoles/liter among individuals who achieved the target SU level within 1 year after the index date, and 454.4 μmoles/liter among individuals who did not achieve the target SU level within 1 year after the index date. Baseline SU level, eGFR, BMI, socioeconomic deprivation index, osteoarthritis, diabetes, prescription of statin, NSAIDs, nitrate, antiosteoporosis drugs, antihypertensive drugs, glucocorticoids, or aspirin were the most important predictors for adherence to "achieving the target SU level." Baseline SU level, eGFR, BMI, osteoarthritis, stroke, fall, prescription of statin, NSAIDs, opioids, nitrate, antiosteoporosis drugs, antihypertensive drugs, glucocorticoids, proton pump inhibitor, diuretics, or aspirin were the most important predictors for adherence to "not achieving the target SU level." The C statistics for predicting the adherence of "achieving the target SU level" and "not achieving the target SU level" were 0.73 and 0.76, respectively. The distribution of the estimated weights for adherence are shown in Supplementary  Figure 2 (http://onlinelibrary.wiley.com/doi/10.1002/art.42504). After IPW, baseline characteristics were well balanced between the 2 comparison groups, with all standardized mean differences <0.1 (Supplementary Table 2, http://onlinelibrary.wiley.com/doi/ 10.1002/art.42504). The mean follow-up time was 3.6 years for the "achieving the target SU level" arm and 3.5 years for the "not achieving the target SU level" arm.
As shown in Figure 2, the 5-year risk of hip fracture was lower for the "achieving the target SU level" arm (0.5%) compared with the "not achieving the target SU level" arm (0.8%). The 5-year risk difference of hip fracture for the "achieving the target SU level" arm compared with the "not achieving the target SU level" arm was -0.3% (95% CI -0.5%, -0.1%), and the HR was 0.66 (95% CI 0.46, 0.93) ( Table 2). The prevented fraction of achieving the target SU level with ULT for hip fracture was 37.5%, suggesting that of 8 hip fracture cases that occurred among 1,000 participants in the "not achieving the target SU levels" arm over 5 years, 3 cases can be prevented if participants reached the "target SU levels with the ULT". The E-value was 2.40 (95% CI 1.36, 3.77), indicating that the relation of potential residual confounder(s) to both "achieving the target SU level" and risk of hip fracture must be ≥2.40 to nullify the protective association between "achieving the target SU level" and risk of hip fracture observed in the primary analyses. Results from the sensitivity analyses conducted among participants with incident gout, among those whose gout diagnosis was defined by Read code plus receiving medication for gout, among those initiating treatment with allopurinol, and among those who received antiinflammatory treatment (i.e., glucocorticoids, colchicine, or NSAIDs) during 1 year before the index date also showed that "achieving the target SU level" was associated with a lower risk of hip fracture compared with "not achieving the target SU level," with HRs of 0.63 (95% CI 0.43, 0.92), 0.65 (95% CI 0.45, 0.95), 0.64 (95% CI 0.45, 0.92), and 0.67 (95% CI 0.47, 0.96), respectively (Table 2). Moreover, compared with the "not achieving the target SU level" arm, the protective effects for hip fracture were similar in the "achieving the target SU level of <300 μmoles/liter" arm and the "achieving the target SU level of 300-360 μmoles/liter"  Similar results were also observed for the effect of lowering SU level with ULT to the target levels on the secondary outcomes. As shown in Figure 3 and Table 3 for "achieving the target SU level" arm were all lower than those for the "not achieving the target SU level" arm.

DISCUSSION
In this large, population-based study that used a GP electronic health records database from the UK, lowering SU to target levels using ULT (i.e., <360 μmoles/liter) among people with gout was associated with a lower risk of hip fracture than not reaching target SU levels with ULT. Similar results were also observed for the risks of composite fracture, major osteoporotic fracture, vertebral fracture, and nonvertebral fracture. These findings suggest that a "treat-to-target SU level" with ULT may have a beneficial effect on reducing the risk of fracture among people with gout. Previous studies have investigated the association between ULT use (versus non-use) and risk of fracture; however, the results were inconsistent. One cohort study reported that allopurinol use was associated with a higher risk of major osteoporotic fractures or hip fracture in both the general population and in people with gout (17); whereas another cohort study including participants with gout found a lower risk of fracture in those treated with allopurinol users than in individuals not receiving treatment with allopurinol (19). Two other studies failed to show an association of ULT with either the risk of major osteoporotic fracture among people with gout (18) or the risk of hip fracture among older patients undergoing inpatient rehabilitation (16). These studies either included the prevalent allopurinol users in the exposure group (16), which may have led to potential selection bias, or included individuals with gout but who were not using ULT as a comparison group (17)(18)(19), which may have made the analyses susceptible to confounding by indication. In addition, none of these studies assessed the effect of using ULT to lower SU to a target level (i.e., <360 μmoles/liter) on the risk of fracture among people with gout. In the current study, we found that lowering SU to the target level (i.e., <360 μmoles/liter) within 1 year after the initiation of ULT among people with gout was significantly associated with a decreased risk of fracture, and the findings persisted in various sensitivity analyses.
Several biologic mechanisms have been proposed to explain the association between SU levels and the risk of fracture. First, studies have shown that SU may affect bone health through its impact on oxidative stress (42)(43)(44). When SU levels are hyperuricemic at supersaturated concentrations, such as at concentrations seen among people with gout (45), the antioxidant properties of SU could be overcome by its pro-oxidant effects, which can contribute to an inflammatory milieu, promote bone resorption, and inhibit bone formation (46)(47)(48), and ultimately contribute to the increased risk of fracture. Second, others have found that hyperuricemia could inhibit vitamin D activation by suppressing 1-α-hydroxylase, resulting in a lower 1,25-dihydroxyvitamin D level and higher parathyroid hormone level (49,50). As a result, hyperuricemia could affect bone remodeling through its effect on either vitamin D or parathyroid hormone levels or both (51). Although our study could not directly assess the effect of SU on the risk of fracture, our findings appear to suggest that a reduced risk of fracture with ULT among people with gout may result in part through the effect of ULT on lowering SU to the target levels.
Several strengths of our study merit comment. Using a realworld, population-based electronic database, we emulated a randomized controlled trial to compare the risk of fracture in people with gout who achieved the target SU level with those who did not achieve the target SU level with ULT. This causal analysis approach allows us to assess the role of hyperuricemia on the excess fracture risk in people with gout by minimizing both selection bias (i.e., initiators of ULT) and confounding by indication (i.e., all participants received ULT). Furthermore, the effects of "achieving the target SU level" with ULT were consistent across the different outcomes (i.e., hip fracture, composite fracture, major osteoporotic fracture, vertebral fracture, and nonvertebral fracture), indicating that our study findings are robust. Our study has some limitations. Although we used rigorous approaches to control for confounders by emulating an analysis of a randomized controlled trial, some covariates, such as disease severity, bone density or any frailty measurements, may not be well captured by the variables available in THIN; thus, we cannot rule out residual confounding. For example, individuals who experience more frailty and illness may be less likely to continue on "preventive medication" such as ULT, particularly for conditions that are not immediately fatal, or physicians may be reluctant to escalate the allopurinol dose for those people. Consequently, residual confounding could lead to a potentially biased protective effect of ULT on the risk of fracture. Nevertheless, the association between "achieving the target SU level" with ULT and the control outcome, the risk of traumatic injury, was null, lending the specificity to our findings. Second, ULT initiators who achieved target SU levels may have received better healthcare for their overall health needs and have taken ULT longer than their comparators; thus, we cannot separate the effect of quality of healthcare from that of lowering UA. However, when comparing healthcare utilization prior to ULT initiation among individuals who reached the target levels of ULT with those who did not reach the target levels of ULT, no difference was observed between two groups, suggesting the effect of healthcare utilization on the risk of fracture, if existed, may not completely explain the results found in our study. In addition, excluding participants who did not have the SU level before the index date may limit the generalizability of the current findings to a population who might receive less healthcare or experience less severe disease. In summary, in this population-based data, lowering SU levels with ULT to the guideline-based target level was associated with a lower risk of incident fracture in people with gout.