Proton pump inhibitor use and risk of hip fracture in patients with type 2 diabetes

Type 2 diabetes mellitus (T2DM) is associated with a high rate of comorbidity, including osteoporosis and peptic ulcers. Proton pump inhibitors (PPIs) are a group of acid-suppressing drugs commonly used for treating peptic ulcers. However, observational studies have reported an association between PPI therapy and osteoporotic fractures. This study investigated the association between PPI use and hip fracture (HFx) among patients with T2DM. We conducted this population-based propensity-matched retrospective cohort study using the National Health Insurance Research Database in Taiwan. Patients newly diagnosed with T2DM between 2000 and 2008 were identified. After excluding those who previously used PPIs or suffered HFx, 398,885 patients were recruited (44,341 PPI users; 354,544 non-users). HFx risk data from 2000 to 2013 were collected to calculate the cumulative rate of HFx in these two groups. Sensitivity analyses were conducted to evaluate the effects of PPI dose. After propensity score matching of 1:4, 44,431 and 177,364 patients were assigned to the PPI user and non-user groups, respectively. PPI user group showed an increased risk of HFx with an adjusted hazard ratio of 1.41 (95% CI 1.29–1.54) without dose–response relationship. Thus, there is an increased risk of HFx in patients with T2DM receiving long-term PPI treatment.


Materials and methods
Data sources. In 1995, Taiwan launched a single-payer mandatory enrollment National Health Insurance (NHI) program. Taiwan's established NHRI has continued to maintain the NHIRD in Taiwan since 2002. At the end of 2018, the coverage rate was more than 99.9% of Taiwan's population 18 .
The NHIRD includes all claims data of the NHI program and makes these data available to scientists in Taiwan for research purposes. The NHIRD is an individual-level claim database comprising data on sex, date of birth, diagnostic codes, medical records of clinical visits, hospitalizations, diagnosis codes, prescriptions, procedures/ surgeries, and expenditures.
We conducted a population-based propensity-matched retrospective cohort study using data from the Longitudinal Cohort of Diabetes Patients (LHDB 2000), which contains random samples from 120,000 patients each calendar year with newly diagnosed diabetes mellitus enrolled from the 2000 Registry of NHI Beneficiaries (a total of 1.68 million enrollees from 2000 to 2013).

Identification of cases. Patients aged ≥ 18 years with newly diagnosed T2DM between 2000 and 2008 were
identified. Those who previously used PPIs [defined daily dose (DDD) ≥ 28] or suffered HFx prior to the index date were excluded. In total, 399,481 patients were eligible for analysis (Fig. 1).
During the longitudinal study period, we excluded fractures associated with HFx due to traffic accident. A final total of 398,885 patients were recruited, including 44,341 PPI users and 354,544 non-users.
Propensity score matching. Propensity score matching was performed to minimize selection bias by balancing baseline characteristics, including age and gender. The propensity score was calculated for each PPI user and non-user using all covariates by logistic regression model. The optimal ratio from the analysis of variable multiple pairing was 1:4.
Outcome and comorbidities. Primary outcome was defined as any new diagnosis of HFx (ICD-9-CM codes 820-820.9) with medical codes for internal fixation or hemiarthroplasty (ICD-9-CM codes 79.15, 79.35, and 81.52). Data on HFx during the period from 2000 to 2013 were collected to calculate the cumulative rate of HFx in the PPI and non-PPI groups. Patients in both the PPI and non-PPI groups were followed up until either the endpoint of study (December 31, 2013) or the occurrence of one of the following events: primary outcome, censoring due to loss to follow-up, or withdrawal from insurance. Hazard ratios (HRs) were calculated for risk of HFx according to PPI use or non-use. Sensitivity analyses were conducted to evaluate the dose effects of PPI treatment.
Statistical analysis. Data on HFx during the period from 2000 to 2013 were collected to evaluate the cumulative rate of HFx in the PPI and non-PPI groups. Student's t test was used to analyze the continuous variables and Chi-squared test was used for categorical variables. The cumulative incidences of HFx according to PPI use and cumulative dose were estimated using the Kaplan-Meier method, and differences between the cumulative incidence curves were compared using log-rank test. Cox proportional hazards models were used to calculate the HRs and 95% confidence intervals (CIs) of HFx in the PPI user and non-user groups in patients with T2DM over a 5-year period. HR was calculated for risk of HFx according to PPI use or non-use. Sensitivity analyses were conducted to evaluate the dose effects of PPIs. This study complied with the Helsinki Declaration. The data in this study were collected with the approval of the Institutional Review Board of Kaohsiung Medical University Hospital (KMUHIRB-EXEMPT(II)-20170017) after obtaining informed consent.

Results
Patient characteristics. After propensity score matching of 1:4, 221,795 patients with T2DM were enrolled.
Among these, 44,431 and 177,364 patients were assigned to the PPI user and non-user groups, respectively. The matching procedure achieved a good balance of baseline characteristics between the two groups, without significant between-group differences related to distribution of gender and age category (Table 1). Furthermore, the prevalence of some comorbidities, including hypertension, stroke, asthma, COPD, myocardial infarction, chronic heart failure, depression, chronic renal failure, and rheumatoid arthritis, were higher in the PPI user group compared with those in the non-user group. Use of prescribed medications, including NSAIDs, corticosteroids, anticoagulants, diuretics, antipsychotic, thyroxine, statins, antihypertensive, and sedatives, was also significantly greater in the PPI user group compared with that in the non-user group (Supplementary Table 1).
Cox proportional hazards regression model. Cox  www.nature.com/scientificreports/ sion models showed a lower risk of fracture in males compared with that in females, with an adjusted HR of 0.65 (95% CI 0.60-0.70; P < 0.001). In addition to female gender, old age (≥ 70 years), low insurance premium [< 20,000 NTD (New Taiwan Dollar)/month], and comorbidity (CCI ≥ 3) were statistically significant independent risk factors for HFx (P < 0.001).
The results of the comorbidity-stratified and medication-use analysis of HFx risk showed that stroke and chronic renal failure patients with T2DM showed an increased risk of HFx, with adjusted HRs of 1.84 (95% CI 1.65-2.05, P < 0.001) and 2.01 (95% CI 1.58-2.57; P < 0.001), respectively. Hypertension, asthma, COPD, myocardial infarction, chronic heart failure, depression, and rheumatoid arthritis were not associated with increased risk of HFx after adjustment for other characteristics. Although a higher proportion of PPI users used other medications (Table 2), only use of anticoagulants and sedatives were associated with an increased risk of HFx after adjustment for characteristics (Supplementary Table 2). The adjusted HRs of HFx in patients with T2DM using anticoagulants and sedatives were 1.26 (95% CI 1.13-1.40; P < 0.001) and 1.33 (95% CI 1.19-1.48, P < 0.001), respectively. Patients who used statins had a decreased risk of HFx, with an adjusted HR of 0.61 (95% CI 0.52-0.70, P < 0.001).
Association between risk of hip fracture and PPI use. Kaplan-Meier curves revealed the time-toreadmission events for cumulative incidence of fracture between the PPI user and non-user groups after a 5-year follow-up. The PPI group showed a significantly higher incidence of HFx than did the non-PPI group (incident rate, 3.27 vs. 2.16, respectively; log-rank test, P < 0.001; Fig. 2A).
We also evaluated the relationship between PPI dose and HFx ( Table 3). The World Health Organization defines DDD as the assumed average maintenance dose per day for a drug used for its main indication in adults. The cumulative DDDs were estimated as the sum of the dispensed DDD of any PPI during 60 months. The PPI group was categorized into four subgroups: non-use (< 28 DDD), 28-65 DDD, 66-124 DDD, and > 125 DDD. The adjusted HRs of the 28-65 DDD, 66-124 DDD, and > 125 DDD subgroups were 1.46 (95% CI 1.27-1.68; P < 0.001), 1.39 (95% CI 1.21-1.60; P < 0.001), and 1.39 (95% CI 1.21-1.59, P < 0.001), respectively. Analysis of each event showed no significant differences in the dose-response relationship between PPI use and risk of HFx (log-rank test, P > 0.05; Fig. 2B).

Discussion
This is the first population-based cohort study to evaluate the impact of PPI use on risk of HFx in patients newly diagnosed with T2DM. Predictors of HFx included well-known risk factors such as older age, female gender, and higher CCI score, which may reflect a more fragile medical state and higher risk of osteoporosis, as previously reported. We adjusted for multiple confounding factors to evaluate the effect of PPIs on risk of HFx in patients with T2DM. In patients with T2DM, PPI use was a significant and independent predictor of HFx, with an adjusted HR of 1.41. However, we did not observe a dose-response relationship between risk of HFx and dose of PPI used. This is in contrast to other studies [11][12][13][14][15] , which demonstrated an association between dosage of www.nature.com/scientificreports/ PPI and HFx was dose-dependent. This difference might be explained by the use of different statistical analysis methods or the susceptibility of T2DM patients to HFx 5 . Further research is needed to clarify the safety margin of PPI to avoid the risk of HFx. Previous studies reported an elevated risk of subsequent fracture development after PPI therapy. Several possible relationships between fracture and PPI use have been reported, including postmenopausal women 20,21 , long-term use of PPIs 11 , history of smoking 14 , and conditions such as stroke 22 , Alzheimer's disease 23 , and hemodialysis 24 . The present study revealed that stroke and chronic renal failure were associated with higher risk of HFx in patients with T2DM after adjustment. Our findings are supported by the findings of previous studies. Stroke is associated with an increased risk of falls and fractures 25 , and patients with chronic renal failure have an increased risk of fractures due to disordered mineral and bone metabolism 26 .
Patients with T2DM have an increased risk of fracture due to falls related to nervous and vascular disease 25,27 as well as changes in the microarchitecture that decrease bone strength and quality 28,29 . The effects of bone structure due to use of PPIs may be linked to malabsorption of calcium, hypergastrinemia, hypochlorhydria, hyperparathyroidism, and regulation of bone cells 15,30 . Previous studies have suggested that use of PPIs leads to increased gastric pH, and prolonged hypochlorhydria may reduce calcium ionization and affect intestinal absorption 15,31,32 . Low circulating calcium triggers secretion of parathyroid hormone (PTH) from the parathyroid gland to enhance bone resorption. PPI users had significantly higher PTH levels compared with that of non-users 33,34 . However, T2DM is also associated with impaired calcium metabolism, which may increase bone fragility 35 . Since PTH elevation is associated with abnormalities in glucose metabolism 4,36 , the etiology and link between PPI-induced HFx and T2DM should be examined. On the other hand, PPIs influence osteoblast and osteoclast activity, and regulate bone resorption 37 . Treatment using PPIs decreased gene expression of type I collagen, alkaline phosphatase, and bone morphogenetic protein 2 in Hematopoietic stem cells 38 . Moreover, T2DM is associated with impaired osteoblast differentiation and activity 39,40 . Since PPI users and patients with T2DM show increased risk of HFx, use of PPIs may lead to deterioration of the bone microstructure and strength, and increase the risk of fracture in patients with T2DM. Figure 3 summarizes the process of PPI-induced bone fracture and its association with T2DM.
Although our analysis was adjusted for comorbidity-stratified analysis and medication use, it is possible that these adjustments did not incorporate all the health conditions influencing PPI use. Consistent with previous studies, stroke was a risk factor for falls 41 and was associated with increased risk of HFx. Elderly patients with diabetes were more likely to use sedatives, which was also associated with an increased risk of fall 3 . Furthermore, long-term users of anticoagulants often have a history of cerebrovascular accident, which may increase their likelihood of HFx due to an unsteady gait (Supplementary Table 2).
The present study has several limitations. First, the NHI database was not designed for academic research; hence, miscoding of diagnoses may have occurred. However, the coding error could be compensated for using  www.nature.com/scientificreports/ medication codes for DM control and procedure codes for HFx (internal fixation or hemiarthroplasty). Second, the NHI database does not include patients' functional status, compliance, personal habits (e.g., smoking, alcohol use), severity of comorbidities, nutritional status, biochemical data (e.g., hemoglobin A 1C ), time to surgery, and quality of postoperative care. Thus, it was not possible to show a dose response related to residual confounding or effect modification. Third, although we were unable to rule out the possibility that the dose-response relationship analyses may have been confounded by immortal time bias, our study revealed that use of PPIs was associated with an increased risk of HFx.
In conclusion, patients with T2DM using PPIs have an increased risk of HFx. Although unknown confounding factors may exist, these findings may provide a valuable basis for future prospective studies to investigate the relationship between PPI use and risk of HFx.