This evidence-driven analysis was performed using the PRISMA 2020 guidelines (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) [8] (Supplementary Table 1) and was prospectively registered in the PROSPERO (CRD42023416670, https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=416670). We conducted the search using data from PubMed, Cochrane, Web of Science and Embase, up to April 2023. The included English language studies compared the efficacy and safety of CHLOR and HCTZ for the treatment of cardiovascular diseases. We reviewed databases using the following terms: “Chlorthalidone”, “Chlorphthalidolone”, “Phthalamudine”, “Oxodoline”, “Chlortalidone”, “Hygroton”, “Thalitone”, “Hydrochlorothiazide”, “HCTZ”, “Dichlothiazide”, “Dihydrochlorothiazide”, “HydroDIURIL”, “Oretic”, “Sectrazide”, “Esidrix”, “Esidrex”, “Hypothiazide”, “cardiovascular diseases”, “Cardiovascular Disease”, “Disease, Cardiovascular”, “Major Adverse Cardiac Events”, “Cardiac Events”, “Cardiac Event”, “Event, Cardiac”, “Adverse Cardiac Event”, “Adverse Cardiac Events”, “Cardiac Event, Adverse” and “Cardiac Events, Adverse ”. The comprehensive search strategy is shown in Supplementary Table 2. To ensure the completeness of our literature search, the lists of all eligible references were also examined manually. Both investigators systematically and independently evaluated each included study. Any discrepancies in the search process were addressed by mutual discussion and consent.

Inclusion criteria were defined as follows: (1) the study design must have been randomized, cohort or case-control; (2) the study enrolled adult participants with diagnosed cardiovascular diseases; (3) the study compared the effectiveness of CHLOR and HCTZ; (4) the study must have reported at least one of the following clinical outcomes: myocardial infarction (MI), heart failure (HF), non-cancer-related death, stroke, death from any cause, hypokalemia, hospitalization for acute kidney injury, or cardiovascular events; and (5) sufficient data was available to allow calculation of a weighted mean difference (WMD) or an odds ratio (OR). This study eliminated non-original articles such as editorial comments, letters, case reports, reviews, pediatric articles, and conference abstracts. Additionally, only published articles were included. Articles written in languages other than English were not considered for inclusion.

All data collection was carried out independently by two investigators, with any discrepancies resolved through discussion and final decisions made by a third investigator. The following data were extracted from the included article: publication year, first author, study period, original country, design of study, size of sample, participant demographics including age, gender and body mass index (BMI), as well as specific clinical outcomes such as glomerular filtration rate (GFR) 60 mL/min/1.73 m², MI, HF, non-cancer-related death, stroke, death from any cause, hypokalemia, hospitalization for acute kidney injury, and cardiovascular events. In cases where continuous variables were presented as median range or interquartile range, mean $\pm$ standard deviation was calculated using dependable mathematical methods [9, 10]. In instances where data were either lacking or not reported in the original studies, the corresponding author(s) were contacted to provide the data.

The quality of included articles were evaluated by The Newcastle–Ottawa Scale (NOS) [11], and those studies considered high quality received seven to nine points [12]. We assessed the quality of eligible randomized controlled trials (RCTs) based on the Cochrane Handbook for Systematic Reviews of Interventions 5.1.0, using 7 criteria. These included random sequence generation, allocation concealment, blinding, incomplete outcome data, selective reporting, and other sources of bias. Each criterion was assigned a low, high or unclear risk outcome and studies with more low-risk outcomes were considered superior [13]. Two researchers independently assessed study quality and evidence and resolved discrepancies via discussion.

Data analysis was performed by Review Manager 5.3 (Cochrane Collaboration, Oxford, UK). WMD and OR were used for continuous and dichotomous variables. All metrics were presented with 95% CI. Study heterogeneity was assessed using $\chi$² test (Cochran’s Q) and I² index [14]. Significant heterogeneity was determined by either $\chi$², p value 0.05 or I² $>$50%. A random-effects model was applied for combined WMD or OR estimations in cases where assignable heterogeneity was detected. Conversely, the fixed effect model was used for estimations when heterogeneity was not significant. To evaluate the effect of the studies with significant heterogeneity, one-way sensitivity analyses were performed. We evaluated publication bias visually using funnel plots in Review Manager 5.3 version (Cochrane Collaboration, Oxford, UK) and Egger’s regression tests were conducted using Stata 12.0 version (Stata Corp, College Station, TX, USA) for results that have been studied in 3 or more trials. A p value 0.05 for publication bias was considered statistically significant.

The search and selection process is summarized in Fig. 1. In total, 1336 articles were identified from PubMed (n = 295), Embase (n = 653), Cochrane (n = 113), and Web of Science (n = 275). After removing duplicates, 925 titles and abstracts were screened, leading to 6 full-text articles involving 368,066 patients (36,999 CHLOR vs. 331,067 HCTZ) for pooled analysis. Among them, 5 were retrospective cohort studies, and 1 was a prospective, randomized study. Table 1 (Ref. [2, 15, 16, 17, 18, 19]) provides the characteristics and quality score of each study; 5 were identified as good quality, with a median (range) score of 7. Supplementary Figs. 1,2 demonstrate the risk of bias for 1 RCT study. Supplementary Table 3 presents details of the quality assessment for all the eligible studies.

Fig. 1.

Flowchart of the systematic search and selection process. HCTZ, hydrochlorothiazide; CHLOR, chlorthalidone.

The groups did not differ significantly in age (WMD 0.08; 95% CI–0.01, 0.17; p = 0.07), gender (male/total, OR 1.00; 95% CI 0.98, 1.03; p = 0.80), or BMI (WMD –0.04; 95% CI –0.17, 0.09; p = 0.51). However, they differed significantly in baseline GFR 60 mL/min/1.73 m² (WMD 1.15; 95% CI 0.88, 1.51; p = 0.30) (Table 2).

MI data were reported in 3 studies [2, 15, 16] for a total of 348,047 patients (31,457 CHLOR vs. 316,590 HCTZ). Pooled results showed that there was no significant difference between the two groups (OR 0.80; 95% CI 0.56, 1.14; p = 0.41), yet there was significant heterogeneity (I² = 83%, p = 0.03) (Fig. 2A). No visual (Fig. 3A) or statistical (Egger’s test, p = 0.802, Fig. 4A) publication bias was found. The exclusion of Dhalla 2013 data eliminated heterogeneity associated with MI (I² = 0%, p = 0.45), indicating that this article was the primary factor contributing to heterogeneity in the analysis.

Fig. 2.

Forest plots of perioperative outcomes. (A) Myocardial infarction. (B) Heart failure. (C) Cardiovascular event. (D) Hypokalemia. (E) Non-cancer-related death. (F) Stroke. (G) Death of any cause. (H) Hospitalization for acute kidney injury. CHLOR, chlorthalidone; HCTZ, hydrochlorothiazide; RCT, randomized controlled trial; M-H, Mantel-hanszel.

Fig. 3.

Funnel plots of perioperative outcomes. Funnel plots of (A) Myocardial infarction. (B) Heart failure. (C) Cardiovascular event. (D) Hypokalemia. (E) Non-cancer-related death. (F) Stroke. (G) Death of any cause. (H) Hospitalization for acute kidney injury. RCT, randomized controlled trial.

Fig. 4.

Egger plots of perioperative outcomes. (A) Myocardial infarction. (B) Heart failure. (C) Cardiovascular event. (D) Hypokalemia. (E) Non-cancer-related death. (F) Stroke. (G) Death of any cause.

The analysis included three studies [2, 15, 16] comprising 348,047 patients (31,457 CHLOR vs. 316,590 HCTZ), revealing a similar occurrence of HF events between the two groups (OR 0.86; 95% CI 0.64, 1.14; p = 0.05) (Fig. 2B), but with substantial heterogeneity (I² = 80%, p = 0.008). No visual (Fig. 3B) or statistical evidence of publication bias was found (Egger’s test, p = 0.93, Fig. 4B). Eliminating Dhalla 2013 data removed heterogeneity associated with heart failure (I² = 0%, p = 0.39), indicating that this study was the primary contributor to heterogeneity in the analysis.

Two articles [16, 18] reported Cardiovascular Event data in 317,373 patients (17,253 CHLOR vs. 300,120 HCTZ). There was no considerable difference in the occurrence rate between the two groups (OR 1.85; 95% CI 0.58, 6.44; p = 0.34) with no noticeable heterogeneity (I² = 99%, p 0.00001) (Fig. 2C). Funnel plots indicated a minor publication bias (Fig. 3C), conversely, the Egger’s test failed to detect any statistically significant bias (Fig. 4C).

Not all articles describe the definition of hypokalemia, but those that do agree that hypokalemia is defined when the blood potassium is 3.5 mmol/L. Hypokalemia analysis was performed in 3 studies [2, 15, 18] with 56,118 patients (20,076 CHLOR vs. 36,042 HCTZ), detecting significantly higher hypokalemia in the CHLOR group (OR 2.10; 95% CI 1.15, 3.84; p = 0.01) and heterogeneity with statistical significance (I² = 97%, p 0.00001) (Fig. 2D). The funnel plot showed mild publication bias (Fig. 3D), but there was no statistical significance in the Egger’s test (p = 0.644) (Fig. 4D).

The analysis included two studies [2, 15] comprising 43,394 patients (17,140 CHLOR vs. 26,256 HCTZ), revealing a similar rate of non-cancer related death between the two groups (OR 1.02; 95% CI 0.56, 1.85; p = 0.45), and meaningful heterogeneity (I² = 68%, p = 0.08) (Fig. 2E). Both the funnel plot (Fig. 3E) and Egger’s test (Fig. 4E) detected publication bias.

Stroke data were reported in three articles [2, 15, 16] involving 348,047 patients (31,457 CHLOR vs. 316,590 HCTZ), and no marked difference was demonstrated within the groups (OR 0.94; 95% CI 0.80, 1.10; p = 0.45), and no meaningful heterogeneity (I² = 26%, p = 0.26) (Fig. 2F). Funnel plots showed no publication bias (Fig. 3F), and no statistical significance was observed in the Egger’s test (p = 0.96) (Fig. 4F).

Data of death from any cause were available in 2 studies [2, 18] totaling 26,245 patients (9692 CHLOR vs. 16,553 HCTZ). Pooled analysis revealed no significant difference in the CHLOR group in comparison to the HCTZ group (OR 1.95; 95% CI 0.52, 7.28; p = 0.32) with significant heterogeneity (I² = 100%, p 0.00001) (Fig. 2G). No visual (Fig. 3G) or statistical (Fig. 4G) evidence of publication bias was detected.

Three studies [2, 15, 18] involving 30,959 patients (12,097 CHLOR vs. 18,862 HCTZ) were included in the analysis. Pooled results demonstrated that the rate of hospitalization for acute kidney injury showed that the two groups had similar results (OR 1.38; 95% CI 0.40, 4.78; p = 0. 61), and significant heterogeneity was observed (I² = 99%, p 0.00001) (Fig. 2H). Visual (Fig. 3H) or statistical (Egger’s test), evidence of publication bias were detected. Upon excluding the data reported by Edwards 2021, the heterogeneity observed with regards to hospitalization for acute kidney injury was reduced (I² = 93%, p 0.00001), suggesting that this study was the principal origin of heterogeneity in the analysis.

This study conducted a sensitivity analysis using hospitalization rates for MI, HF, and acute kidney injury, which displayed considerable heterogeneity. By systematically excluding studies, it was found that the heterogeneity for MI (Fig. 5A) and acute kidney injury hospitalization rates (Fig. 5B) did not significantly decrease, indicating stability in the results. However, upon the exclusion of the paper by Dhalla IA, there was a significant reduction in heterogeneity (I² = 0%, p = 0.39), demonstrating that this particular study was significantly heterogeneous compared to other studies regarding the HF endpoint (Fig. 5C). Nevertheless, the meta-analysis results for the HF endpoint remained consistent, with no statistical difference before and after the exclusion of this study.

Fig. 5.

Sensitivity analysis. Sensitivity analysis of (A) Myocardial infarction, (B) Heart failure, (C) Hospitalization for acute kidney injury.