Machine Learning Insights into Uric Acid Elevation with Thiazide Therapy Commencement and Intensification

doi:10.21203/rs.3.rs-3401602/v1

Download PDF

Research Article

Machine Learning Insights into Uric Acid Elevation with Thiazide Therapy Commencement and Intensification

https://doi.org/10.21203/rs.3.rs-3401602/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Serum uric acid, associated with cardiovascular conditions such as atherosclerotic heart disease and hypertension, can be elevated by thiazide or thiazide-like drugs (THZ), essential in hypertension management. Identifying clinical determinants affecting THZ-related uric acid elevation is critical.

Methods

In this retrospective cross-sectional study, we explored the clinical determinants influencing uric acid elevation related to THZ, focusing on patients where THZ was initiated or the dose escalated. A cohort of 143 patients was analyzed, collecting baseline and control uric acid levels, alongside basic biochemical studies and clinical data. Feature selection was conducted utilizing criteria based on mean squared error increase and enhancement in node purity. Four machine learning algorithms—Random Forest, Neural Network, Support Vector Machine, and Gradient Boosting regressions—were applied to pinpoint clinical influencers.

Results:

Significant features include uncontrolled diabetes, index eGFR level, absence of insulin, action of indapamide, and absence of statin treatment, with absence of SGLT2 inhibitors, low dose aspirin exposure, and older age also being noteworthy. Among the applied models, the Gradient Boosting regression model outperformed the others, exhibiting the lowest MAE, MSE, RMSE values, and the highest R2 value (0.779). While Random Forest and Neural Network regression models were able to fit the data adequately, the Support Vector Machine demonstrated inferior metrics.

Conclusions

Machine Learning Algorithms can precisely predict THZ-related uric acid changes, facilitating optimized therapy tailoring, minimizing unnecessary THZ abstinence, and guiding to prevent usage in cases where uric acid levels might reach undesirable levels.

Hypertension

Hyperuricemia

Machine Learning

Thiazides

Indapamide

HydrochlorothiazideUric Acid

Uric acid, a product of the metabolic breakdown of purine nucleotides, plays a crucial role in human health, with abnormal levels being associated with various medical conditions, including gout, kidney stones, and atherosclerotic heart disease [1]. High serum uric acid levels often coexist with hypertension, reflecting a causality dilemma and contributing to increased cardiovascular risk [2, 3]. This is likely due to the shared underlying mechanisms that lead to hypertension and cardiovascular diseases, and their association with elevated uric acid levels [4, 5]. The relationship between uric acid and hypertension is complex, with proposed mechanisms highlighting the roles of inflammation, oxidative stress, and arterial stiffness leading to atherosclerosis[6–9].

Thiazide or thiazide-like diuretics (THZ) are foundational antihypertensive medications, considered 'old but gold.' Their prominent status stems not only from their efficacy in managing blood pressure and reducing cardiovascular morbidity and mortality, as supported by meta-analyses, but also from their associated low rates of nonadherence[10–13]. However, while they are frequently prescribed for hypertension and associated with favorable cardiac outcomes, the initiation or escalation of these medications can notably impact uric acid levels, which are linked with unfavorable cardiac events, thus presenting a clinical conundrum [14, 15].

Although the relationship between uric acid levels and THZ is well-established, there remains a significant lack of detailed understanding of the clinical factors that dictate uric acid variations following the initiation or intensification of THZ. A detailed evaluation of these influential elements is crucial in promoting the best clinical practices, individualized treatment strategies, and improved patient care, particularly in groups with hypertension. Motivated by this, this paper aims to evaluate the crucial factors affecting changes in uric acid levels related to THZ, utilizing sophisticated machine learning methods to examine contributing and alleviating factors and to furnish insights into predicting the magnitude of change.

Design, setting and the stuy population

This study employed a subset of patients derived from the dataset of our earlier research, acting as the foundation for this study [16]. This cohort comprises cases from four distinct clinics, both secondary and tertiary care centers, spanning from October 1st, 2020, to October 30th, 2021. The dataset had undergone rigorous cleaning and mild imputation processes as outlined by our prior study's exclusion criteria. Subsequently, this research focused on individuals for whom THZ drugs were either initiated or escalated and involved those with available index and control uric acid levels. Patients who were treated or initiated with losartan were excluded due to its uric acid-lowering properties, alongside those administered Xanthine Oxidase inhibitors and those with acute kidney injuries, to eliminate their influences on THZ associated uric acid changes.

We conducted a power analysis to ascertain the minimum sample size needed for our study, focusing on the smallest expected difference between hydrochlorothiazide and indapamide. Based on a literature review, a randomized, double-blinded study revealed a mean difference of 0.2 mg/dl between the two drugs (hydrochlorothiazide arm: mean 8.3 ± 0.2 mg/dl; indapamide arm: mean 8.1 ± 0.2 mg/dl). With an alpha level of 0.01 (1% type 1 error) and a desired power of 0.80 (80%), we determined that our study would require at least 106 participants to detect this difference with adequate power.

Case Definition and Data Acquisition:

Cases were required to have undergone two visits: an index visit, where the intervention took place, and a control visit, occurring in four weeks, where blood studies were reassessed. The intervention or "the action" involved decisions related to the initiation or escalation of hydrochlorothiazide or indapamide. Hydrochlorothiazide is available only in combination pills, while indapamide can be obtained as both a single pill and in combination with perindopril. Chlortalidone, which is solely available in combination with atenolol, is rarely prescribed in Turkey and did not have any instances in our cohort. Therefore, in this study, the only thiazide-like diuretic was indapamide, and it was abbreviated along with hydrochlorothiazide as "THZ".

Each patient’s data included demographic information, comorbidities, current medications, laboratory measurements, and details about actions taken regarding antihypertensive classes. Specific conditions and diseases were defined based on established diagnoses and guidelines, and active cancer was identified based on recent diagnoses or treatments. Diabetes mellitus was defined as self-reported DM diagnoses or via anti-diabetic medication use. As a subgroup, uncontrolled DM was arbitrarily defined as patients with an A1c over 8%. Detailed definitions of specific diseases are avalible from our previous research [16].

Concerning renin-angiotensin-aldosterone system inhibitors (RAASi), β-blockers (βbl), or calcium channel blockers (CCb), a medication from these classes may have been newly initiated or might have already been in use. If in use, their administration could have been ceased, or their doses could have been adjusted—either increased, decreased, or maintained at the existing level. For the purpose of comparative analysis, the initiation of a drug from these classes was grouped with the same drug’s dose escalation and the cessation was grouped with dose reduction, due to their anticipated similar impacts on uric acid levels. In cases where a drug was not introduced to patients who were naïve to it, those instances were grouped with cases where the same drug dose remained unchanged.

For laboratory parameters, serum uric acid levels acquired at the index visit (pre-action) were labeled as 'index uric acid.' Those obtained at the control visit (post-action) were denoted as 'control uric acid.' The numeric target variable was represented as 'Δuric acid' and was calculated by subtracting index levels from control levels.

Descriptive statistics

Continuous variables are presented as mean ± standard deviation, as they are all normally distributed. The Student's t-test or ANOVA was used for analyzing differences between normally distributed continuous variables, with Levene’s test assessing the equality of variances for each variable calculated across groups. Categorical variables are represented as counts and percentages, analyzed using the chi-squared (χ2) or Fisher’s exact test as appropriate. All conventional statistical analyses were conducted utilizing corresponding functions and ‘psych’ package in R 4.2.2 programming language.

Feature selection:

Before developing the regression models, a feature selection process was conducted to eliminate unnecessary, irrelevant, and redundant features. This study employed a methodical approach, selecting eight features based on their linear relationship with the target variable, assessed using correlation coefficients, and on model-based feature importance scores. The importance of each feature was evaluated using random forest regression, focusing on the increase in mean squared error and improvement in node purity, to enhance the understanding of each feature's relevance. This approach allowed the inclusion of impactful features, omitting the less significant ones without compromising predictive accuracy. The feature selection was carried out using the randomForest() package in R 4.2.2.

Machine Learning Algorithms

Four machine learning algorithms were employed to predict THZ-associated uric acid changes. Given that the numeric target variable converts the prediction into a regression problem, we opted for Random Forest Regression (RFreg), Gradient Boosting Regression (GBreg), Support Vector Machine Regression (SVMreg), and Neural Network Regression (NNreg). For SVMreg, the Radial Basis Function (RBF) kernel was used. Both RFreg and GBreg were configured to create 200 decision trees, and NNreg was set with 250 neurons in hidden layers, utilizing a Rectified Linear Unit (ReLU) activation function and the Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS-B) optimization algorithm.

Machine learning algorithms were implemented using Orange 3.34.0 Software (University of Ljubljana, Slovenia), and each model’s performance was evaluated through the software’s Test and Score module using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and adjusted R². The former two quantify the average magnitude of the residuals or prediction errors; the third assesses the fit of the model, and the latter represents the average absolute difference between the observed actual outcomes and the predictions made by the model.

Patient selection

From the initial dataset, 174 patients were identified as having either initiated or escalated THZ. Among these, 14 lacked both index and control uric acid levels. Four had undergone treatment with or had started losartan treatment. Six had records of allopurinol, and seven experienced acute kidney injury, which was the principal area of investigation in our preceding study, leading to their exclusion (Fig. 1). A total of 143 patients were found to be eligible and included in the study.

Clinical data and Laboratory parameters

The cohort studied displayed a modest male dominance, constituting 65.7% (n = 94), and 52.4% were aged above 60, as shown in Table 1. Diabetes Mellitus (DM) was the prevalent comorbidity, appearing in 43.4% (n = 62) of cases, with metformin being the predominant medication, used by 35% (n = 50). Table 1 reveals that both initial and Δuric acid levels maintained uniformity across genders and age groups. Diabetes didn't appear to impact these levels; however, uncontrolled diabetes mellitus revealed a notably higher Δuric acid (1.17 ± 1.12; p = 0.003). Patients with coronary artery disease and chronic kidney disease presented higher initial uric acid levels (5.8 ± 1.4; p = 0.005 and 6.8 ± 1.12; p = 0.001, respectively) but the Δuric acid levels was consistent with the rest of the cohort.

Table 1

Index uric acid and Δuric acid levels according to demographic and clinical characteristics
	Total	Index Uric Acid	Sig.	Δuric acid	Sig.
Total Cohort	133 (100%)	5.2 ± 1.3		0.6 ± 0.87
Demographic Status
Age > 60	75 (52.4%)	5.13 ± 1.3	0.54	0.61 ± 0.82	0.87
Female	49 (34.3%)	5.12 ± 1.4	0.3	0.52 ± 0.7	0.17
Comorbidities
Diabetes Mellitus	62 (43.4%)	5.4 ± 1.4	0.07	0.64 ± 1.03	0.6
Uncontrolled Diabetes Mellitus	18 (12.6%)	5.1 ± 1.8	0.8	1.17 ± 1.12	0.003
Coronary Artery Disease	28 (19.6%)	5.8 ± 1.4	0.005	0.4 ± 0.8	0.16
Heart Failure	3 (2.1%)	6.45 ± 2.5	0.08	0.45 ± 0.6	0.7
Chronic Kidney Disease	7 (4.9%)	6.8 ± 1.4	0.001	0.7 ± 0.5	0.8
Obstructive Airway Disease	14 (9.8%)	5.1 ± 1.15	0.7	0.5 ± 0.9	0.83
Active Cancer	3 (2.1%)	5.7 ± 0.4	0.13	-0.06 ± 0.77	0.18
Antihypertensives Before Action
RAAS blockers	62 (53.4%)	5.3 ± 1.3	0.35	0.53 ± 0.9	0.43
Thiazide & Thiazide like Diuretics	19 (13.2%)	5.6 ± 1.6	0.08	0.4 ± 0.72	0.3
Ca-channel blockers	38 (26.6%)	5.02 ± 1.1	0.33	0.74 ± 1.06	0.3
Beta blockers	46 (32.2%)	5.5 ± 1.1	0.04	0.66 ± 0.9	0.54
Alpha blockers	3 (2.1%)	5.03 ± 1.3	0.8	0.63 ± 0.13	0.9
Loop diuretics	3 (2.1%)	6.7 ± 2.25	0.03	0.3 ± 0.6	0.5
Medication List Before Action
Insulin	19 (13.3%)	5.5 ± 1.5	0.27	0.16 ± 1	0.02
Metformin	50 (35%)	5.4 ± 1.4	0.3	0.55 ± 1	0.6
Sulphonylurea	13 (9.1%)	5.5 ± 1.2	0.3	0.3 ± 0.78	0.2
Dipeptidyl Peptidase-4 Inhibitors	9 (6.3%)	5.9 ± 1.5	0.05	0.42 ± 0.76	0.5
Thiazolidinediones	8 (5.6%)	6.4 ± 0.9	0.006	0.36 ± 0.91	0.46
Sodium glucose cotransporter 2	13 (9.1%)	5.5 ± 1	0.36	0.13 ± 0.7	0.04
Non-steroidal anti-inflammatory drugs	8 (5.6%)	5.9 ± 2.3	0.35	0.26 ± 1	0.35
Beta-2 agonists	9 (6.3%)	5.4 ± 1.2	0.51	0.12 ± 0.8	0.09
Proton pump inhibitors	29 (20.3%)	5.4 ± 1.1	0.22	0.3 ± 1.1	0.06
Antidepressants	7 (4.9%)	6.5 ± 1.6	0.005	1 ± 1.4	0.24
Salicylates	10 (7%)	5.6 ± 1.04	0.32	0.65 ± 1.32	0.86
Statin treatment	15 (10.5%)	5.4 ± 1.2	0.5	0.21 ± 0.58	0.02

RAASi was the leading antihypertensive treatment administered before any action, succeeded by beta-blockers. Initial levels were marginally elevated in patients treated with beta-blockers and notably elevated in those receiving loop diuretics (5.5 ± 1.1; p = 0.03 and 6.7 ± 2.25; p = 0.03, respectively). Continuing the list of medications examination, DPP-4, thiazolidinediones, and antidepressants were correlated with elevated index uric acid levels, whereas insulin, statins, and SGLT-2i were significantly connected to a lower Δuric acid (0.16 ± 1; p = 0.02, 0.21 ± 0.5; p = 0.02, and 0.13 ± 0.7 p = 0.04, respectively).

Table 2 illustrates the results of specific interventions across various groups and the entire cohort. There were no significant differences in either index or Δuric acid levels associated with any actions related to RAASi, CCb, and βbl. Similarly, the initiation and escalation of THZ revealed comparable initial and Δuric acid levels. However, the initiation of indapamid did result in a notably higher increase in uric acid. Several minor switches between hydrochlorothiazide and indapamid were observed. In one case, a change from indapamide (1.25 or 1.5mg) to hydrochlorothiazide (25mg) led to a decrease in uric acid level. In six instances, hydrochlorothiazide (12.5mg) was substituted with indapamide (2.5mg), resulting in a mean Δuric acid of 0.8 (± 0.5). A comparison between these cases and those without drug switches showed statistical significance (p = 0.05), revealing that, despite the minimal occurrences of such THZ drug switches (n = 7), the differences in effects were significant and warranted attention.

Table 2

Comparison of Initial and ΔUric Acid Levels Based on Actions, Stratified by Treatment Status
		Index Uric Acid	Sig.	Δuric acid	Sig.
In total cohort
Addition of THZ	124	5.1 ± 1.2	0.08	0.63 ± 0.9	0.3
Escalation of THZ	19	5.6 ± 1.6		0.41 ± 0.16
Addition or escalation of RAASi	111	5.07 ± 1.17	0.1	0.61 ± 0.83
RAASi dose was not changed or not added	30	5.6 ± 5.6		0.53 ± 1.06
Cessation or RAASi dose decrement	2	5.5 ± 0.4		0.7 ± 0.42
Addition or escalation of β-blockers	8	4.8 ± 0.9	0.7	0.84 ± 0.3	0.8
β-blocker dose was not changed or not added	131	5.2 ± 1.3		0.6 ± 0.5
Cessation or β-blocker dose decrement	4	5.4 ± 1.26		1 ± 0.34
Addition or escalation of CCb	23	5.5 ± 1.7	0.3	0.75 ± 0.7	0.4
CCb dose was not changed or not added	102	5.1 ± 1.2		0.54 ± 0.8
Cessation or CCb dose decrement	18	5.3 ± 0.1		0.76 ± 1.12
THZ naive group
Indapamid initiated	49	4.95 ± 1.21	0.22	0,85 ± 0,83	0.026
Hydrochlorothiazide initiated	75	5.22 ± 1.2		0,48 ± 0,9
Thiazide drug switches among treated group
Hydrochlorothiazide ◊ Indapamide	6	5.9 ± 2.35	0.87	0,8 ± 0,51	0.05
Indapamide ◊ Hydrochlorothiazide	1	6.1		-0.8
Dose escalation without drug change	12	5.52 ± 1.24		0.3 ± 0.7

Table 3 outlines the relationships between initial laboratory parameters, revealing a mild correlation between initial glucose levels and initial uric acid. Initial creatinine, urea, eGFR, serum K levels, and triglycerides also showed mild, yet significant, correlations with initial uric acid.

Table 3

Correlation between Initial Biochemistry Levels and Initial and Δ Uric Acid Levels
Laboratory parameters	Index Uric acid		Δuric acid
	Correlation coefficient	Sig.	Correlation coefficient	Sig.
Index Uric Acid	1		-0.12	0.17
Delta Uric acid	-0.12	0.17	1
Index serum glucose corrected Na	-0.1	0.2	0.1	0.2
Index serum K	0.2	0.01	0.09	0.26
Index albumin corrected Ca	0.87	0.46	0.02	0.86
Index creatinine	0.3	< 0.001	0.04	0.5
Index estimated GFR	-0.2	0.02	-0.13	0.13
Index urea	-0.28	0.001	-0.16	0.06
Index glucose	0.04	0.6	0.17	0.049
Index HbA1c (present in 99 patients)	-0.08	0.42	0.17	0.09
Index LDL (present in 135 patients)	0.022	0.8	0.028	0.75
Index triglycerides (present in 128 patients)	0.31	< 0.001	-0.11	0.23

Machine learning models

Every conceivable feature was assessed to determine its contribution to a random forest regression model (supplementary table 1). Based on the percentage increase in mean squared error, the top five most crucial features are uncontrolled diabetes, index eGFR level, absence of insulin treatment, initiation or escalation action of indapamid, and absence of statin treatment, among others, as depicted in Fig. 2.

Four ML algorithms were employed to ascertain the best fits to the dataset (Table 4). The GBreg model surpassed the others, exhibiting the lowest MAE, MSE, RMSE values, and the highest R² value (0.779). The least effective performance was noted in the SVM model (R² = 0.279). Random forest regression ranked third (R² = 0.494), while the NNreg approach was nearly aligned with the GBreg algorithm in evaluating each feature (R² = 0.57). This evaluation considered increases in mean squared error and improvements in node purity.

Table 4

Performance metrics for each machine learning algorithm
	Mean Squared Error	Root Mean Squared Error	Mean Absolute Error	R²
Random Forest Regression	0.39	0.625	0.48	0.494
Gradient Boosting Regression	0.171	0.413	0.28	0.779
Support Vector Machine Regression	0.557	0.746	0.556	0.279
Neural Network Regression	0.326	0.571	0.39	0.57

This study is designed to identify co-determinants that could potentially intensify or attenuate the elevation of uric acid when THZ is commenced or the dose is escalated in hypertensive patients. To the best of our knowledge, this represents the first study that utilizes ML algorithms to assess the magnitude of uric acid change associated with THZ and to delineate the contributing factors.

According the results of this study, three most important features affecting THZ related uric acid rise include having uncontrolled diabetes, lower eGFR and the existance of insulin at the time of “THZ action”. Having a lower eGFR level, indicative of chronic kidney disease, is logically coherent since uric acid is predominantly excreted through the kidneys. Consequently, uric acid clearance would be correlated with creatinine clearance [17, 18]. Also, individuals with moderately low eGFR would potentially be more susceptible to volume depletion. However, the relationship might not be linear, as those with advanced nephron loss might not respond to thiazides, making volume contraction less of a concern for this group and volume contraction become a less pronounced problem than increased volume-loads. However, this study does not include advanced chronic kidney disease, and any unresponsiveness is beyond its scope.

Our findings regarding insulin use and diabetes contribute to the ongoing debates surrounding insulin’s multifaceted effects on uric acid levels. At a preliminary glance, insulin seems to elevate renal uric acid reabsorption by enhancing the expression of renal urate transporter 1 (URAT1) and glucose transporter-9 (GLUT9), and by altering the levels of ATP-binding cassette subfamily G member 2 (ABCG2)[19]. Notably, while insulin clamp tests in healthy individuals did not elevate serum uric acid levels, they did result in a significant reduction in urinary uric acid excretion [20]. In this scenario, a marked correlation was observed between diminished fractional uric acid and sodium excretion, inducing a net sodium retention, a recognized outcome of insulin exposure [21]. On the flip side, poor glycemic control and insulin resistance, typical in type 2 diabetes mellitus, are linked with elevated serum uric acid levels and reduced uric acid excretion [22, 23]. A smaller intervention study revealed that both troglitazone treatment and low-energy diets reduced uric acid levels in hypertensive patients, with a more pronounced decline observed with troglitazone [24]. The reciprocal cause-effect relationship also exists as uric acid is known to impede insulin signalling and induce insulin resistance [25]. Accordingly, reducing uric acid levels with febuxostat and allopurinol notably ameliorated insulin resistance [26–29].

In this context, effective insulin utilization and better glycemic control could potentially lower serum uric acid levels by enhancing insulin sensitivity and possibly altering the renal management of uric acid. With precision, it's crucial to note that these discussions pertain to the dynamics of uric acid when either a diabetic or a healthy individual is exposed to exogenous insulin. However, our study explores insulin's influence on the uric acid retaining effect of THZ. Whether it serves as a marker of heightened diabetes control or due to the associated reduction in glucotoxicity with improved insulin sensitivity, insulin seems to alleviate the uric acid retention associated with THZ. Indeed, the expanding corpus of knowledge indicates that THZ can induce hyperuricemia not only through their direct impact on organic anion transporters but also due to the ensuing volume depletion they cause [30–32]. Volume contraction induces H⁺ ion secretion due to increased sodium sodium–hydrogen exchanger-3 (NHE3) activity, and consequently, and subsequently, the increased cellular pH propels urate uptake through organin anion transporter-4 (OAT4) due to the amplified urate/hydroxyl ion exchange [31]. Corroborating this assertion, it appears that when volume depletion associated with diuretic use is balanced with adequate volume repletion, the occurrence of uric acid retention is averted [33]. Exogenous insulin use, hypothetically, may counteract the volume depletion associated with THZ by promoting tubular sodium reabsorption, hence indirectly reducing the rise in uric acid linked to volume depletion.

Another striking finding involves the ameliorative impact of pre-existing SGLT2i treatment, prior to the “THZ action,” on the elevation of uric acid. Sodium-glucose co-transporter-2 inhibitors (SGLT2i) are relatively new antidiabetic medications that have managed to capture the attention of physicians regularly treating diabetes, as they offer enhanced survival benefits in patients with high cardiovascular risk and heart failure [34, 35]. Moreover, they have earned a reputation as efficacious nephroprotective agents, effectively treating albuminuria [36]. The beneficial profile of SGLT2i extends further; evidenced by meta-analytical data revealing that the administration of SGLT2i induces a uricosuric effect, denoting a reduction in uric acid concentrations, ranging between 0.6 to 0.7 mg/dl [37]. These agents possess mild diuretic and antihypertensive properties; one study exploring the metabolic impacts of transitioning from low dose THZ to SGLT2i confirmed no elevation in blood pressure while effectively decreasing uric acid levels [38]. The increase in uricosuria and corresponding decrease in circulating uric acid attributed to SGLT2 inhibitors are likely due to the inhibition of GLUT9b activity, a transporter located on the apical membrane of renal tubular cells responsible for the transport of both uric acid and d-glucose [39]. SGLT2 inhibitors also inhibit the activity of URAT1 channels, a primary route for THZ to enter the lumen of the proximal tubule and one of the channels facilitating urate retention [40, 41]. This intersection is accountable for the ameliorative impact of SGLT2 inhibitors on the uric acid rise related to THZ, a process with many subtleties.

Our research has also illustrated the mitigating impact of statins on the uric acid elevation associated with THZ. A moderate uric acid-reducing effect has been evidenced for atorvastatin, while no such impact was observed for rosuvastatin, and pitavastatin may actually elevate serum uric acid levels[42–44]. Given the diversity within the class of statins, it is unfeasible to conduct post hoc analysis in our cohort to discern which specific statin was associated with the least rise in uric acid. One hypothesis suggests that the diminished serum uric acid levels may be attributed to the augmentation of renal blood flow, stemming from the enhanced endothelial function conferred by atorvastatin[42]. Thus, patients treated with atorvastatin might experience slightly superior renal perfusion and a correspondingly lesser elevation in uric acid levels.

Another subtle yet intriguing observation is that low doses of salicylates (such as aspirin), commercially available in 81-100mg, appeared to slightly intensify the magnitude of uric acid elevation in our study. This outcome is anticipated, considering salicylates are recognized to exhibit a bimodal effect on uric acid metabolism[45–47]. High doses of salicylate inhibit URAT1 in a manner akin to losartan and probenecid, leading to uricosuria [48]. Conversely, the utilization of low-dose salicylates, the only form observed in our cohort, inhibits URAT1, resembling the action of pyrazinamide [49]. The heightened URAT1 activity might synergize with THZ-associated uric acid retention, contributing to an elevated concentration of uric acid.

Lastly, our results reveals a more pronounced increase in uric acid levels with indapamide compared to hydrochlorothiazide and regression analyses have significantly underscored its contribution to the uric acid slope. This is noteworthy since a recently published hypertension guideline (2023 European Society of Hypertension) have not emphasized the differences in uric acid elevation between these medications[13]. Current literature provides limited comparisons of uric acid elevations between thiazide and thiazide-like drugs. However, a well-conducted, albeit older, randomized controlled study did illustrate a minor but significant difference [50].

We acknowledge the strengths and limitations of the study. The major strength of the study is the fact that this is the first study to date to incorporate multiple ML regression algorithms to determine THZ-related uric acid change and comparing their performance. This study also provides preliminary data for rational and improved THZ utilization. Having data from multiple clinics with diverse backgrounds (i.e., secondary and tertiary care; internal medicine and cardiology clinics) allows the sample to reflect the real-world population. The primary limitations of this study are rooted in its retrospective structure. Beyond the inherent and recognized limitations of retrospective studies, such as selection and recall biases, this study also missed some critical variables, including body-mass index, obesity, the presence of MAFLD, and data on gout flares following the intensification or initiation of THZ. Comparisons involving switches between molecules lacked power due to the minimal sample size in which switches occurred. One of the oher constraints is the variance in the measurement of laboratory parameters, stemming from the use of different biochemistry tools and analyzers across diverse clinics and laboratories. Nonetheless, this limitation is less likely to impact the parameters of uric acid change since both index and control uric acid levels would be affected congruently and the substraction operation would minimize it. Additionally, two notable features—insulin and SGLT2i—are available commercially to only diabetic patients, who do not represent our entire study population, prompting considerations of potential bias. However, the potential detrimental influence of this bias, may not be substantial enough to overshadow our results, given that other antidiabetic treatments don’t have an impacth on uric acid level changes.

In conclusion, this paper illuminates the complex landscape of determinants affecting uric acid levels post initiation or escalation of Thiazide diuretics, utilizing machine learning as a powerful tool to dissect and understand these relationships. The results are anticipated to have far-reaching implications in the realms of clinical pharmacology and medicine, fostering a more nuanced and individualized approach to hypertension and hyperuricemia management.

CONFLICT OF INTEREST

All authors declare no conflict of interest regarding the study

FUNDING

No funding was received for this study

Ethical approval

To maintain participant confidentiality, each patient included in the study was allocated an anonymous identification number, and all collected data were securely stored as encrypted files on a dedicated computer within the institution. Given the retrospective nature of this study, processing of the data did not necessitate informed consent. The research adhered to the principles stipulated in the Declaration of Helsinki and received approval from the Non-Interventional Clinical Research Ethics Committee at Hacettepe University (evaluation date: July 5, 2022, project number: GO 22/734, decision number: GO 2022/12–29).

Consent to participate

Due to the retrospective design of this study, consent was not necessary to include patients

Consent for publication

All authors consented to the publication of this manuscript.

Competing interests

The authors declare no competing interests.

Contributions

Murat Özdede and Alper Turan Güven conceptualized the study. Murat Özdede, Alper Tuna Güven, Yusuf Ziya Şener collected the data. Murat Özdede analyzed the data and wrote the main manuscript. All authors reviewed the manuscript and contributed. In the preparation of this manuscript, OpenAI’s ChatGPT was consulted for advice on structuring the abstract and refining the language

Lanaspa MA, Ishimoto T, Li N, Cicerchi C, Orlicky DJ, Ruzycki P, et al (2013) Endogenous fructose production and metabolism in the liver contributes to the development of metabolic syndrome. Nature communications 4(1):2434.
Lanaspa MA, Andres-Hernando A, Kuwabara M (2020) Uric acid and hypertension. Hypertension Research 43(8):832–4. https://doi.org/10.1038/s41440-020-0481-6
Borghi C, Agnoletti D, Cicero AFG, Lurbe E, Virdis A (2022) Uric Acid and Hypertension: a Review of Evidence and Future Perspectives for the Management of Cardiovascular Risk. Hypertension 79(9):1927–36. https://www.ahajournals.org/doi/abs/10.1161/HYPERTENSIONAHA.122.17956
Borghi C, Rosei EA, Bardin T, Dawson J, Dominiczak A, Kielstein JT, et al (2015) Serum uric acid and the risk of cardiovascular and renal disease. Journal of hypertension 33(9):1729–41.
Kuwabara M, Hisatome I, Niwa K, Hara S, Roncal-Jimenez CA, Bjornstad P, et al (2018) Uric Acid Is a Strong Risk Marker for Developing Hypertension From Prehypertension. Hypertension 71(1):78–86. https://www.ahajournals.org/doi/abs/10.1161/HYPERTENSIONAHA.117.10370
Otani N, Toyoda S, Sakuma M, Hayashi K, Ouchi M, Fujita T, et al (2018) Effects of uric acid on vascular endothelial function from bedside to bench. Hypertension Research 41(11):923–31.
Cassano V, Crescibene D, Hribal ML, Pelaia C, Armentaro G, Magurno M, et al (2020) Uric Acid and Vascular Damage in Essential Hypertension: Role of Insulin Resistance. Nutrients 12(9):2509. https://www.mdpi.com/2072-6643/12/9/2509
Lanaspa MA, Sanchez-Lozada LG, Choi Y-J, Cicerchi C, Kanbay M, Roncal-Jimenez CA, et al (2012) Uric acid induces hepatic steatosis by generation of mitochondrial oxidative stress: potential role in fructose-dependent and-independent fatty liver. Journal of Biological Chemistry 287(48):40732–44.
Kuwabara M, Kanbay M, Hisatome I. Uric acid and hypertension because of arterial stiffness. Am Heart Assoc; 2018. p. 582–4.
Wei J, Galaviz KI, Kowalski AJ, Magee MJ, Haw JS, Narayan KMV, et al (2020) Comparison of Cardiovascular Events Among Users of Different Classes of Antihypertension Medications: A Systematic Review and Network Meta-analysis. JAMA Network Open 3(2):e1921618-e. https://doi.org/10.1001/jamanetworkopen.2019.21618
Chalmers J, Mourad J-J, Brzozowska-Villatte R, De Champvallins M, Mancia G (2023) Benefit of treatment based on indapamide mostly combined with perindopril on mortality and cardiovascular outcomes: a pooled analysis of four trials. Journal of Hypertension 41(3):508.
Olde Engberink RH, Frenkel WJ, van den Bogaard B, Brewster LM, Vogt L, van den Born B-JH (2015) Effects of thiazide-type and thiazide-like diuretics on cardiovascular events and mortality: systematic review and meta-analysis. Hypertension 65(5):1033–40.
Mancia G, Kreutz R, Brunström M, Burnier M, Grassi G, Januszewicz A, et al (2023) 2023 ESH Guidelines for the management of arterial hypertension The Task Force for the management of arterial hypertension of the European Society of Hypertension Endorsed by the International Society of Hypertension (ISH) and the European Renal Association (ERA). J Hypertens
Rahimi-Sakak F, Maroofi M, Rahmani J, Bellissimo N, Hekmatdoost A (2019) Serum uric acid and risk of cardiovascular mortality: a systematic review and dose-response meta-analysis of cohort studies of over a million participants. BMC Cardiovascular Disorders 19(1):218. https://doi.org/10.1186/s12872-019-1215-z
Raja R, Kavita F, Amreek F, Shah A, Sayeed KA, Sehar A (2019) Hyperuricemia Associated with Thiazide Diuretics in Hypertensive Adults. Cureus 11(8):e5457.
Güven AT, Özdede M, Şener YZ, Yıldırım AO, Altıntop SE, Yeşilyurt B, et al (2023) Evaluation of machine learning algorithms for renin-angiotensin-aldosterone system inhibitors associated renal adverse event prediction. Eur J Intern Med 114:74–83.
Guan H, Zheng Y, Zhou X, Xu Y, Fu C, Xiao J, et al (2020) Efficacy of different urinary uric acid indicators in patients with chronic kidney disease. BMC Nephrol 21(1):290.
Marangella M (2005) Uric acid elimination in the urine. Hyperuricemic syndromes: pathophysiology and therapy 147:132–48.
Toyoki D, Shibata S, Kuribayashi-Okuma E, Xu N, Ishizawa K, Hosoyamada M, et al (2017) Insulin stimulates uric acid reabsorption via regulating urate transporter 1 and ATP-binding cassette subfamily G member 2. American Journal of Physiology-Renal Physiology 313(3):F826-F34. https://journals.physiology.org/doi/abs/10.1152/ajprenal.00012.2017
Quinones Galvan A, Natali A, Baldi S, Frascerra S, Sanna G, Ciociaro D, et al (1995) Effect of insulin on uric acid excretion in humans. American Journal of Physiology-Endocrinology and Metabolism 268(1):E1-E5. https://doi.org/10.1152/ajpendo.1995.268.1.E1
Brands MW (2018) Role of Insulin-Mediated Antinatriuresis in Sodium Homeostasis and Hypertension. Hypertension 72(6):1255–62. https://www.ahajournals.org/doi/abs/10.1161/HYPERTENSIONAHA.118.11728
Vuorinen-Markkola H, Yki-Järvinen H (1994) Hyperuricemia and insulin resistance. The Journal of Clinical Endocrinology & Metabolism 78(1):25–9. https://doi.org/10.1210/jcem.78.1.8288709
Perez-Ruiz F, Aniel-Quiroga MA, Herrero-Beites AM, Chinchilla SP, Erauskin GG, Merriman T (2015) Renal clearance of uric acid is linked to insulin resistance and lower excretion of sodium in gout patients. Rheumatology International 35(9):1519–24. https://doi.org/10.1007/s00296-015-3242-0
Tsunoda S, Kamide K, Minami J, Kawano Y (2002) Decreases in serum uric acid by amelioration of insulin resistance in overweight hypertensive patients: effect of a low-energy diet and an insulin-sensitizing agent. American Journal of Hypertension 15(8):697–701. https://doi.org/10.1016/S0895-7061(02)02953-9
Zhu Y, Hu Y, Huang T, Zhang Y, Li Z, Luo C, et al (2014) High uric acid directly inhibits insulin signalling and induces insulin resistance. Biochemical and Biophysical Research Communications 447(4):707–14. https://www.sciencedirect.com/science/article/pii/S0006291X14007311
Liu P, Wang H, Zhang F, Chen Y, Wang D, Wang Y (2015) The effects of allopurinol on the carotid intima-media thickness in patients with type 2 diabetes and asymptomatic hyperuricemia: a three-year randomized parallel-controlled study. Internal medicine 54(17):2129–37.
Takir M, Kostek O, Ozkok A, Elcioglu OC, Bakan A, Erek A, et al (2015) Lowering uric acid with allopurinol improves insulin resistance and systemic inflammation in asymptomatic hyperuricemia. Journal of investigative medicine 63(8):924–9.
Meng J, Li Y, Yuan X, Lu Y (2017) Effects of febuxostat on insulin resistance and expression of high-sensitivity C-reactive protein in patients with primary gout. Rheumatology International 37(2):299–303. https://doi.org/10.1007/s00296-016-3612-2
Moriya CT, Satoh H, Watada H (2018) Febuxostat improves insulin resistance in the skeletal muscle in vitro and in vivo. Diabetes 67(Supplement_1)
Sica D, Schoolwerth A. Renal handling of Organic Anions and Cations: Excretion of Uric Acid. The Kidney 6th ed., Brenner BM. Saunders, Philadelphia, pp680-700; 2000.
Palmer BF, editor Metabolic complications associated with use of diuretics. Seminars in nephrology; 2011: Elsevier.
Ben Salem C, Slim R, Fathallah N, Hmouda H (2016) Drug-induced hyperuricaemia and gout. Rheumatology 56(5):679–88. https://doi.org/10.1093/rheumatology/kew293
Steele TH (1969) Evidence for altered renal urate reabsorption during changes in volume of the extracellular fluid. The Journal of laboratory and clinical medicine 74(2):288–99.
McMurray JJV, Solomon SD, Inzucchi SE, Køber L, Kosiborod MN, Martinez FA, et al (2019) Dapagliflozin in Patients with Heart Failure and Reduced Ejection Fraction. New England Journal of Medicine 381(21):1995–2008. https://www.nejm.org/doi/full/10.1056/NEJMoa1911303
Verma S, Mazer CD, Fitchett D, Inzucchi SE, Pfarr E, George JT, et al (2018) Empagliflozin reduces cardiovascular events, mortality and renal events in participants with type 2 diabetes after coronary artery bypass graft surgery: subanalysis of the EMPA-REG OUTCOME® randomised trial. Diabetologia 61(8):1712–23.
Piperidou A, Sarafidis P, Boutou A, Thomopoulos C, Loutradis C, Alexandrou ME, et al (2019) The effect of SGLT-2 inhibitors on albuminuria and proteinuria in diabetes mellitus: a systematic review and meta-analysis of randomized controlled trials. J Hypertens 37(7):1334–43.
Xin Y, Guo Y, Li Y, Ma Y, Li L, Jiang H (2019) Effects of sodium glucose cotransporter-2 inhibitors on serum uric acid in type 2 diabetes mellitus: A systematic review with an indirect comparison meta-analysis. Saudi J Biol Sci 26(2):421–6.
Kimura T, Sanada J, Shimoda M, Hirukawa H, Fushimi Y, Nishioka M, et al (2018) Switching from low-dose thiazide diuretics to sodium-glucose cotransporter 2 inhibitor improves various metabolic parameters without affecting blood pressure in patients with type 2 diabetes and hypertension. J Diabetes Investig 9(4):875–81.
Lytvyn Y, Škrtić M, Yang GK, Yip PM, Perkins BA, Cherney DZ (2015) Glycosuria-mediated urinary uric acid excretion in patients with uncomplicated type 1 diabetes mellitus. Am J Physiol Renal Physiol 308(2):F77-83.
Jutabha P, Anzai N, Wempe M, Wakui S, Endou H, Sakurai H (2011) Apical voltage-driven urate efflux transporter NPT4 in renal proximal tubule. Nucleosides, Nucleotides and Nucleic Acids 30(12):1302–11.
Kim G-H, Jun J-B (2022) Altered Serum Uric Acid Levels in Kidney Disorders. Life 12(11):1891. https://www.mdpi.com/2075-1729/12/11/1891
Athyros VG, Elisaf M, Papageorgiou AA, Symeonidis AN, Pehlivanidis AN, Bouloukos VI, et al (2004) Effect of statins versus untreated dyslipidemia on serum uric acid levels in patients with coronary heart disease: a subgroup analysis of the greek atorvastatin and coronary-heart-disease evaluation (GREACE) study. American Journal of Kidney Diseases 43(4):589–99. https://www.sciencedirect.com/science/article/pii/S0272638604000022
Milionis HJ, Kakafika AI, Tsouli SG, Athyros VG, Bairaktari ET, Seferiadis KI, et al (2004) Effects of statin treatment on uric acid homeostasis in patients with primary hyperlipidemia. American Heart Journal 148(4):635–40. https://www.sciencedirect.com/science/article/pii/S0002870304001930
Ogata N, Fujimori S, Oka Y, Kaneko K (2010) Effects of Three Strong Statins (Atorvastatin, Pitavastatin, and Rosuvastatin) on Serum Uric Acid Levels in Dyslipidemic Patients. Nucleosides, Nucleotides & Nucleic Acids 29(4–6):321–4. https://doi.org/10.1080/15257771003741323
Caspi D, Lubart E, Graff E, Habot B, Yaron M, Segal R (2000) The effect of mini-dose aspirin on renal function and uric acid handling in elderly patients. Arthritis Rheum 43(1):103–8.
Segal R, Lubart E, Leibovitz A, Iaina A, Caspi D (2006) Renal effects of low dose aspirin in elderly patients. Isr Med Assoc J 8(10):679–82.
Zhang Y, Neogi T, Chen C, Chaisson C, Hunter DJ, Choi H (2014) Low-dose aspirin use and recurrent gout attacks. Ann Rheum Dis 73(2):385–90.
Enomoto A, Kimura H, Chairoungdua A, Shigeta Y, Jutabha P, Ho Cha S, et al (2002) Molecular identification of a renal urate–anion exchanger that regulates blood urate levels. Nature 417(6887):447–52.
Roch-Ramel F, Guisan B, Diezi J (1997) Effects of uricosuric and antiuricosuric agents on urate transport in human brush-border membrane vesicles. Journal of Pharmacology and Experimental Therapeutics 280(2):839–45.
Elliott WJ, Weber RR, Murphy MB (1991) A double-blind, randomized, placebo-controlled comparison of the metabolic effects of low-dose hydrochlorothiazide and indapamide. J Clin Pharmacol 31(8):751–7.

No competing interests reported.

SupplementaryTable1.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Machine Learning Insights into Uric Acid Elevation with Thiazide Therapy Commencement and Intensification

Status:

Version 1

Abstract

Figures

Introduction

Patients and Methods

Design, setting and the stuy population

Case Definition and Data Acquisition:

Descriptive statistics

Feature selection:

Machine Learning Algorithms

RESULTS

Patient selection

Clinical data and Laboratory parameters

Machine learning models

DISCUSSION

CONCLUSION

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1