Thiazide Use and Fracture Risk: An updated Bayesian Meta-Analysis

The association between thiazide use and fracture risk is still controversial. We conducted an updated meta-analysis on the association between thiazide use and fracture risk. We systematically searched PubMed, Embase, and Cochrane library databases for all types of human studies, including observational and experimental studies that were published up until July 2019. We also manually searched the reference lists of relevant studies. The pooled relative risks (RRs) with 95% credible interval (CrI) were calculated using a Bayesian hierarchical random effect model. A total of 19 case-control (N = 496,568 subjects) and 21 cohort studies (N = 4,418,602 subjects) were included in this meta-analysis. The pooled RR for fractures associated with thiazide use was 0.87 (95% CrI: 0.70–0.99) in case-control and 0.95 (95% CrI: 0.85–1.08) in cohort studies. The probabilities that thiazide use reduces any fracture risk by more than 0% were 93% in case-control studies and 72% in cohort studies. Significant heterogeneity was found for both case-control (p < 0.001, I2 = 75%) and cohort studies (p < 0.001, I2 = 97.2%). Thiazide use was associated with reduced fracture risk in case-control studies, but not in cohort studies. The associations demonstrated in case-control studies might be driven by inherent biases, such as selection bias and recall bias. Thus, thiazide use may not be a protective factor for fractures.

Statistical analysis. We synthesized the data using both classical and Bayesian hierarchical random-effects models [24][25][26] . In classical meta-analysis, we used the DerSimonian-Laird method 27 to calculate the pooled risk ratio. In the Bayesian model, the risk ratios (RRs) for all the studies were converted into a logarithmic scale (denoted as φ i ). Each φ i was assumed to have a normal distribution with a true, but unknown effect size ( i θ ) and known within-study variance (δ i 2 ). The collection of θ i across the studies was assumed to have a normal distribution, with unknown mean (µ) and variance τ ( 2 ), where μ was the estimate of the overall log (RR), and τ 2 was a measure of variation between the studies. The prior information of τ 2 was assumed to be an inverse gamma distribution (0.001, 0.001). The prior function for μ was assumed equivocal prior; i.e., thiazide use does not affect fracture risk (µ = 0, variance = 10,000). We also examined the probability that thiazide use reduces fracture risk by more than 0%, 10%, and 20% (i.e., RR < 1.0, 0.9, 0.8). Heterogeneity of the included studies was assessed with Cochran's Q-statistic test, and inconsistency was quantified by I 2 statistic 28,29 . Funnel plots were generated to identify potential publication bias using Egger's test 30  Results characteristics of studies. We identified a total of 959 articles from different electronic databases and other sources. Of these, 633 duplicate articles and 181 irrelevance articles were excluded after reading the title or abstract. Finally, 19 case-control studies and 21 cohort studies were met for inclusion in this meta-analysis (Fig. 1). A majority (72.5%) of the included studies were considered as high quality based on NOS standards (Table 1). In the case-control studies, approximately 79% of the participants (Total sample size = 496,568) were female; the average participant www.nature.com/scientificreports www.nature.com/scientificreports/ age in the case-control studies was 72 years old. Approximately 63% of the subjects (Total sample size = 4,418,602) were female in cohort studies. The average participant age in the cohort studies was 73 years old. thiazide use and fracture risk in case-control studies. In the classical meta-analysis of case-control studies, we found a negative association between thiazide use and fracture risk (Risk ratio (RR): 0.87, 95% confidence interval (CI): 0.76-0.98). We observed moderate heterogeneity between studies (p < 0.001, I 2 = 75%; Fig. 2). In the Bayesian analysis, the pooled RR for fractures associated with thiazide use was 0.87 (95% credible interval (CrI) 0.70-0.99). The probabilities that thiazide use reduces fracture risk by more than 0%, 10%, and 20% were 93%, 66%, and 23%, respectively (Table 2). thiazide use and fracture risk in cohort studies. In the classical meta-analysis of cohort studies, there was no significant association between thiazide use and fracture risk (RR: 0.93, 95% CI: 0.83-1.05). The heterogeneity between studies was significant (p < 0.001, I 2 = 97.2%; Fig. 3). In the Bayesian analysis, the pooled RR for www.nature.com/scientificreports www.nature.com/scientificreports/ fractures associated with thiazide use was 0.95 (95% CrI: 0.85-1.08). The probabilities that thiazide use reduces fracture risk by more than 0%, 10%, and 20% were 72%, 23%, and 2%, respectively (Table 2). publication bias. The funnel plot of risk ratio versus standard error for the association between thiazide use and fracture risk was shown in Fig. 4. No significant publication bias was observed for both case-control studies (Egger's test: p = 0.65; Fig. 4a) and cohort studies (Egger's test: p = 0.52; Fig. 4b).

Discussion
This meta-analysis provides evidence to support that thiazide exposure is associated with a 13% reduction of fracture risk in case-control studies. However, while an inverse association was noted in cohort studies, it failed to reach statistical significance.
Our findings were partly comparable with the effect shown in the previous two meta-analyses reported by Wiens et al. 8 and Xiao et al. 9 ; both studies suggested that thiazide was associated with the reduction of any fracture risk by 14%. However, to the best of our knowledge, our meta-analysis is the first to distinguish a difference in the relationship between thiazide use and fracture risk by study design. We found that there is a null relationship between thiazide use and fracture risk in cohort studies. A recently published meta-analysis also suggested that the effect of thiazide use on fracture risk was weaker in cohort studies 9 . Although the results from the Bayesian meta-analysis were consistent with that generated from the classical meta-analysis approach, the Bayesian meta-analysis provides additional regarding the probabilities that thiazide use reduces fracture risk by certain percentages. Such information is useful for making clinically relevant decisions about the use of thiazides, and cannot be obtained using the traditional meta-analysis methodology.
The controversial relationship between thiazide diuretics and fractures involves conflicting mechanisms. On the one hand, thiazide could exert beneficial effects on the bone via decreasing urinary calcium excretion by 25-40% 31,32 . In addition, thiazides are associated with an increased level of metabolic alkalosis, which is an  www.nature.com/scientificreports www.nature.com/scientificreports/ inhibitor of bone resorption 33,34 . On the other hand, thiazides diuretics could induce hyponatremia, which has a negative impact on the metabolism and integrity of the bone 35,36 . In addition, thiazide induced-hyponatremia could have harmful neurological side effects, such as gait disturbances and imbalance, which leads to an increased risk of falls and fractures 37 .
This meta-analysis has several limitations. First, due to the absence of relevant experimental studies in humans, our meta-analysis included only observational studies. A meta-analysis based on observational studies cannot make causal inferences about thiazide use and fracture risk. Second, we observed considerable heterogeneity between individual studies, which might bias our results. Lastly, due to insufficient data from individual studies, we did not evaluate the effect of dose and duration of thiazide use on bone fractures.
In conclusion, this meta-analysis included 19 case-control and 21 cohort studies to examine the relationship between thiazide use and fracture risk. Our results suggest that thiazide use was associated with reduced fracture risk in case-control studies, but not in cohort studies. The associations demonstrated in case-control studies might be driven by inherent biases such as selection bias and recall bias. Thus, thiazide use may not be a protective factor for fractures. Randomized clinical trials are still warranted to confirm our findings.