The safety and efficacy of phosphodiesterase type 5 inhibitors in the treatment of diabetic erectile dysfunction: a systematic review and meta-analysis

Inclusion and exclusion criteria

Inclusion criteria: definite diagnosis of type 1 diabetes mellitus or type 2 diabetes mellitus according to the latest guidelines of the International Diabetes Federation (IDF) and the American Diabetes Association (ADA) (Zhang et al., 2025); the diagnostic criteria for ED were in accordance with the definition established by the National Institutes of Health (NIH) Consensus Development Conference on Impotence: the inability to achieve and/or maintain an erection sufficient for satisfactory sexual performance (NIH Consensus Conference, 1993). Included were patients with a definite diagnosis of diabetes mellitus and suffering from ED.

Exclusion criteria: absence of original data in the literature data retrieved, combination drug therapy, duplicate articles, animal testing, review articles.

Main outcomes

The International Index of Erectile Function (IIEF-5) (Jayaram et al., 2024) overall and related evaluation questionnaires were used as the primary efficacy evaluation index.

Search strategy

PubMed, Embase, Web of Science, Cochrane Library, China National Knowledge Infrastructure (CNKI), VIP, Wanfang and China Biology Medicine Disc (CBM) were searched. The date of database last search was December 31, 2024, and reading the retrieved articles and references. Databases were searched with OR and AND collocation by subject words and free words related to disease type (DED), intervention (PDE-5 inhibitors) and study design (RCT). The literature search was performed independently by Zhu Zexin and Xu Jian. If there were any differences in the included literature, the two sides reached a consensus through discussion. If no agreement can be reached, the final decision will be made by Nan Zheng. Articles that met the inclusion criteria were individually evaluated for literature quality, including random allocation method, whether the allocation method was concealed, whether blinding was used, whether the outcome data were complete, and whether the results were selectively reported. Other sources of bias, etc., and the Cochrane literature quality assessment tool was used to evaluate the quality of statistical data and evidence. The Cochrane Collaboration’s tool was used to assess the bias in seven dimensions, including random sequence generation, allocation concealment, blinding, etc. Six of the 10 studies were rated as low risk of bias, and the other four studies were rated as medium risk because of unclear allocation concealment, which did not significantly affect the stability of meta-analysis (Sterne et al., 2016).

Data strategy

Patient or Population, Intervention, Comparison or Control, and Outcome (PICOS) was used to easily retrieve the included studies (Amir-Behghadami & Janati, 2020) where Patient: diabetic erectile dysfunction; Intervention: PDE-5 inhibitors; Comparison: pre-treatment and post-treatment controls; Outcomes: International Index of Erectile Function (IIEF-5) Overall and Associated Evaluations and Adverse Reactions; Study design: RCT (Tantry et al., 2021). The acquisition process is obtained separately by two judges, and if there is any dispute, the third judge will rule. Finally, the above results are obtained. For missing data, we first tried to contact the original authors to obtain them, and if they were not available, we decided whether to use them or not on a case-by-case basis.

Data analysis

Review Manager 5.3 and STATA18 software were used for statistical analyses. Firstly, the included trials were analyzed for heterogeneity; if there was no heterogeneity, a fixed effect model was used; if there was heterogeneity, a random effect model was carried out, and in the process, the reasons for heterogeneity were sought, and subgroup analyses and non-linearities were explored for the efficacy of the PDE-5 inhibitors in relation to the age of the patients. Test of efficacy effect sizes: the ratio (RR) was used to express the dichotomous variables, and 95% CIs were given for all the evaluated indicators. Then, according to the probability of Z value to the statistic P-value, P < 0.05 suggests that the combined statistic of multiple studies is statistically significant.

Reporting bias assessment

Prior to data synthesis and to minimize bias in the reporting section, registration for systematic evaluation and Meta-analysis was performed at PROSPERO (CRD42025637725). Institutional Review Board approval was not required because the relevant data collected for this study came from previously published articles and databases.

Risk of bias assessment

In the risk of bias assessment of 10 studies such as Carson et al. (2005), the risk of bias in aspects such as random sequence generation is at a low level, but other possibilities cannot be ruled out. In terms of allocation concealment, some studies showed an unclear risk of bias. However, in terms of blinding participants and implementors, some studies have high risk of bias, which suggests that more attention should be paid to blinded design in subsequent studies. The risk of bias assessment is shown in Figs. 2 and 3.

Effectiveness of PDE-5 inhibitors

• 1.

  Heterogeneity test:

• The 10 articles in this study were tested for heterogeneity with I² = 73% and P = 0.0001 < 0.001 for Q-test, suggesting that the heterogeneity between the documents selected for this study was statistically significant, and further examination of the Labbe Graph and Galbraith Plot indicates that there is a strong likelihood that one or more of the articles had heterogeneity as shown in Figs. 4 and 5.

• The graphical analysis above concludes that there is moderate heterogeneity in the literature in this study, which can be combined using random effects and exploring the reasons for the heterogeneity.

• 2.

  Random effects combined effect size:

• The random effect combined RR was selected, which finally yielded RR = 2.91 (1.95–4.34), implying that the treatment with PDE-5 inhibitors was 2.91 times more effective than the placebo, which was statistically significant (Z = 5.25, P < 0.0001 < 0.05), suggesting that PDE-5 inhibitors was effective in treating diabetic erectile dysfunction. The forest plot of the effective rate combined effect size is shown in Fig. 6.

• 3.

  Sensitivity analysis:

• The stability of the combined effect size was assessed by excluding individual studies sequentially through sensitivity analyses. The results showed that the impact of individual studies on the overall combined effect size estimates was small, and the effect size estimates and confidence intervals did not change much after the exclusion of individual studies, indicating that the results of this meta-analysis had good stability.

• 4.

  Meta-regression to find the cause of heterogeneity:

• The study was divided into five groups according to treatment modality: tadalafil group, sildenafil group, avanafil group, vardenafil group, mirodenafil group, and meta-regression was performed with the group variable as a covariate, and the results of meta-regression suggested that the tadalafil group P = 0.027 < 0.05, and the outcome suggesting treatment modality as a source of generating heterogeneity, followed by relevant subgroup meta-analysis according to treatment modality group.

• 5.

  Subgroup studies

• Based on subgroup analyses:

• 5.1 Heterogeneity in the tadalafil group (I² = 0%, P = 0.74 > 0.1) was not statistically significant, then fixed effects were chosen to combine effect sizes, yielding RR = 1.59 (1.27 to 2.0). Suggesting that tadalafil was 1.59 times more effective than placebo and statistically significant (z = 4.04, P < 0.0001 < 0.05).

• 5.2 Heterogeneity in the vardenafil group (I² = 0%, P = 0.65 > 0.1) was not statistically significant, then fixed effects were chosen to combine effect sizes, yielding RR = 4.05 (2.84 to 5.78). Suggesting that vardenafil was 4.05 times more effective than placebo and statistically significant (z = 7.71, P < 0.00001 < 0.05).

• 5.3 Heterogeneity in the mirodenafil group (I² = 0%, P = 0.94 > 0.1) was not statistically significant, then fixed effects were chosen to combine the effect sizes, yielding RR = 3.53 (1.63 to 7.63). Suggesting that mirodenafil was 3.53 times more effective than placebo and statistically significant (z = 3.20, P = 0.001 < 0.05).

• 5.4 Due to the small sample sizes included in the sildenafil and avanafil groups, there are sample size limitations, and conducting subgroup analyses may result in less stable and reliable results, increasing the risk of false positive or false negative results. Despite the limitations, the above results show that the overall analysis is robust and trustworthy despite the sample size limitation, revealing that PDE-5 inhibitors is effective in treating diabetic erectile dysfunction while also suggesting that there may be differences between this different class of drugs. The forest plot of the subgroup analysis is shown in Fig. 7.

• 6.

  Bias test:

• The bias test was performed separately according to subgroups, and a funnel plot was drawn, as shown in Fig. 8: In the picture, 1 for tadalafil, 2 for sildenafil, 3 for avanafil, 4 for vardenafil, 5 for mirodenafil.

The symmetry test of the above funnel plot is performed, and the results are as follows:

No publication bias in the tadalafil group (P = 0.960 > 0.05)

No publication bias in the vardenafil group (P = 0.635 > 0.05)

Several other groups could not be assessed for bias testing due to small sample sizes. Still, the overall test of P = 0.727 > 0.05 suggests that the outcomes were reliable and there was no publication bias.

Adverse reactions in PDE-5 inhibitors therapy

In PDE-5 inhibitors, different kinds of drug treatment have various degrees of adverse reactions such as dizziness, headache, eye blinking, facial flushing, etc., so the safety of the drug was evaluated by observing adverse reactions as a secondary indicator.

1. Heterogeneity test: 10 documents of this study, after the heterogeneity test, I² = 11% < 50%, P = 0.35 > 0.1 of the Q test, suggesting that there is no heterogeneity between the documents selected for this study (heterogeneity is not statistically significant), then the fixed effects were chosen to combine the effect sizes.

2. Fixed effects meta-analysis: The effect RR = 2.0 (1.53–2.61) for the 10 studies using fixed effects combined and statistically significant, Z = 5. 06, P < 0.00001 < 0.05, suggests that while applying PDE-5 inhibitors for the treatment of diabetic erectile dysfunction, the occurrence of related adverse drug reactions should be noted, as shown in Fig. 9.

3. Bias test: The bias test was conducted jointly by Revman software and STATA software. A funnel plot was drawn to examine whether there was publication bias in the 10 papers of this study, which resulted in a symmetrical funnel plot (Egger’s Test yielded P = 0.123 > 0.05), with no publication bias, suggesting that the conclusions of this study were accurate and reliable.

Efficacy of PDE-5 inhibitors in relation to age

The study was carried out to determine the correlation between the efficiency of PDE-5 inhibitors and age, and subgroup analyses were performed by age using Revman software. The age of 55 years was taken as the boundary and divided into two groups, age ≥ 55 years and age < 55 years and a forest plot of the combined effect size was drawn, as shown in Fig. 10. As for why 55 years old was chosen as the boundary, there are the following reasons. The first one is that middle-aged people (≥55 years old) have a longer course of diabetes and more significant vascular complications. The second point is that existing studies (Deyoung et al., 2012) often stratify and analyze the therapeutic effect of ED at this age, which has clinical reference value.

In the age group 55 years and above, the RR was 2.73, 95% CI [1.97–3.80] with a p-value of less than 0.00001, which clearly indicates that PDE-5 inhibitors was more effective than placebo in this age group. The heterogeneity test showed I² = 54%, P = 0.06, and although there was some heterogeneity, the P-value was close to 0.05. In the age group under 55 years, the RR was 2.61, with a 95% CI of [2.10–3.23], and the heterogeneity test of I² = 87%, with a P < 0.00001, suggesting that the PDE-5 inhibitors efficacy rate was significantly higher than that of the placebo group in this age group as well. Comparing the RR values of the two age groups, the group aged 55 years and above was 2.73, and the group aged under 55 years was 2.61, which were relatively close to each other. The RR values of both groups were more significant than 1, the confidence intervals did not include 1, and the P values were all less than 0.00001, which indicated that the PDE-5 inhibitors had a significant efficacy relative to the placebo in their respective age groups. Still, it wasn’t easy to judge which group had a superior effect only from the subgroup analyses.

For this reason, we further explored the non-linear relationship between age and efficacy using STATA software. The results revealed that age had a non-linear effect on PDE-5 inhibitors efficacy. The coefficient of the primary term of age was 0.985 (P = 0.021), implying that the efficacy (logRR) increased significantly with each 1-year increase in age, while the coefficient of the squared term of age was - 0.0094 (P = 0.024), suggesting that there was a decreasing effect of age on the efficacy, which showed an inverted U-shape curve, as shown in Fig. 11. According to the formula of the inflection point of the quadratic regression model, -β₁/2β₂ = −0.985/2 * (−0.0094) ≈ 52.3 (years), which yields an inflection point age of about 52.3 years, around which the efficacy is optimal, and after which the efficacy decreases with age. The adjusted R² amounted to 91.19%, age and its squared term together explained 91% of the heterogeneity, the remaining heterogeneity I²_res = 25.93%, the model was effective in reducing the heterogeneity, and the overall F-test of the model P = 0.0367 indicated that the non-linear effect of age on the efficacy was statistically significant.