This meta-analysis was reported in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines (15). It has been registered at PROSPERO (https://www.crd.york.ac.uk/PROSPERO/) under the ID no. CRD42025633357.

The relevant literature was searched using the following databases: PubMed (https://pubmed.ncbi.nlm.nih.gov/), Cochrane Library (https://www.cochranelibrary.com/), Web of Science (https://www.webofscience.com/), China National Knowledge Infrastructure (CNKI; https://www.cnki.net/), WanFang Data (https://www.wanfangdata.com.cn/), China Science and Technology Journal Database (VIP; http://www.cqvip.com/) and Sinomed (http://www.sinomed.ac.cn/). The objective was to identify RCTs and cohort studies, comparing tirofiban combined with oral antiplatelet drugs to oral antiplatelet drugs alone in treating patients with PIS. The comprehensive search strategy relied on the following search terms: (‘PIS’, ‘progressive cerebral infarction’ or ‘progressive stroke’ or ‘progressive cerebral thrombosis’ or ‘early neurological deterioration’) and (‘tirofiban’). Furthermore, the reference lists of eligible studies and related reviews were manually screened to identify other potentially relevant literature for inclusion. The search was conducted on May 5th, 2025, covering literature published up to May 5th, 2025 without any language restrictions, and strictly followed the PRISMA 2020 guidelines. The PRISMA flowchart (Fig. 1) details the study selection process.

The study's inclusion criteria were defined as follows: First, the study type was confined to prospective or retrospective cohort studies and RCTs. Observational studies were included despite their inherent risk of bias (e.g., selection bias, unmeasured confounding). The Newcastle-Ottawa Scale (NOS) was used to assess their quality, although residual bias might persist (16). These studies were designed to explore the application of tirofiban in combination with oral antiplatelet drugs in treating patients diagnosed with PIS, irrespective of whether blinding procedures were incorporated. Secondly, PIS is defined as an increase of ≥2 points from baseline in the National Institutes of Health Stroke Scale (NIHSS) score (17) within 72 h of stroke onset. The included studies used different definitions, but all met the core criterion of neurological deterioration within 72 h. The subjects of this study were carefully selected. The subjects were patients with PIS who had a persistent decline in neurological function after onset, an NIHSS score of no less than 2 and an onset time within 72 h. Thirdly, concerning the intervention measures, the control group was subjected to conventional treatment along with oral antiplatelet drug therapy. By contrast, the experimental group was further administered tirofiban in addition to the control group's treatment protocol. Lastly, the outcome measures were bifurcated into efficacy and safety measures. The efficacy measures encompassed the NIHSS score, modified Rankin scale (mRS) score (18), activities of daily living (ADL) scores (19), Barthel index (BI) scale scores (19), platelet aggregation rate (PAgR) and platelet adhesion rate (PAdR) (20,21). The safety measures comprised the incidence of intracranial hemorrhage, hemorrhage in other systems, mortality rate and serious adverse events. The exclusion criteria were single-arm trials, animal studies, conference abstracts, case reports and any reports from which data cannot be extracted based on published articles. Also excluded were studies focusing on non-PIS during the entire surgical process, those where the intervention did not involve the combination of tirofiban with oral antiplatelet drugs and studies in which each group had fewer than 25 participants.

A total of two authors (YY and YFJ) conducted the literature screening independently and then cross-verified the results. In the event of any disagreement, they consulted a third researcher until a consensus was ultimately reached. The data retrieved from the literature comprised the first author's name, publication year, study type, sample size, therapeutic outcome and safety outcome. The efficacy outcomes were as follows: i) 3-month mRS score of 0-2, ii) NIHSS score, iii) ADL score, iv) PAgR, v) PAdR and vi) effective rate. The following safety outcomes were evaluated: i) Symptomatic ICH (sICH), ii) other systemic hemorrhage, iii) mortality rate and iv) serious adverse events.

The quality of RCTs was appraised using the Cochrane Risk of Bias Tool (22). For observational studies, quality was assessed using the NOS, which assesses three domains: Participant selection (0 to 4 points), inter-group comparability (0 to 2 points) and outcome assessment (0 to 3 points), with a maximum score of 9. It is important to note that no NOS score threshold was pre-set as an inclusion criterion; all observational studies that met the initial inclusion criteria (e.g., study design, population, interventions) were included in the meta-analysis, regardless of their potential NOS score. After assessing quality, it was found that the six included observational studies all had NOS scores between 7 and 9.

The data analysis was conducted with Stata software (version 17.0; StataCorp. LP). The random-effects model was employed for categorical variables to calculate the pooled estimates of odds ratios (ORs), while the mean differences were computed for continuous variables. Heterogeneity was evaluated via the Cochran Q test and the I² test. When I² was <50% or P>0.1 (indicating low heterogeneity), the fixed-effects model was utilized. Continuous data were evaluated using the standardized mean difference (SMD) and its 95% CI, whereas binary data were analyzed with the OR and 95% confidence interval. P<0.05 was regarded as statistically significant. Publication bias was evaluated by visually examining the data and calculating the P-value with Egger's test and Begg's test. If publication risk of bias was detected, the nonparametric trim-and-fill method was further used to adjust for the potential effect of missing studies. Additionally, subgroup analyses were performed to explore the potential sources of heterogeneity across studies. Based on clinical relevance, subgroup analyses were pre-defined according to the antiplatelet treatment regimens (monotherapy and dual therapy), but comparisons between subgroups should be considered exploratory analyses. To explore the potential source of heterogeneity, subgroup analyses were further performed based on the duration of tirofiban administration (short course vs. long course). A short course was defined as tirofiban infusion ≤48 h and a long course as >48 h, according to the median duration reported in included studies.

To further validate the robustness of the primary study results and to account for possible biases in the mixed study design, sensitivity analyses were performed for only 13 RCTs. The statistical methods were consistent with the main analysis (using Stata 17.0 software; ORs were used for categorical outcomes and SMDs for continuous outcomes).

A total of 855 relevant entries were identified through database searches. After eliminating 456 redundant articles, excluding 353 studies upon reading the titles, abstracts and perusing the full texts, 46 full texts remained for detailed evaluation. Among these, 27 articles were further excluded (23 due to low quality and 4 due to unavailable data). Eventually, 19 articles were included (10,23-40), of which 12 were published in Chinese and 7 in English. Fig. 1 illustrates the detailed screening process.

A total of 19 studies were included in the meta-analysis. All articles were published between 2020 and 2024. A total of 3,667 patients were included, among which 1,859 were in the experimental group and 1,808 were in the control group. Table I presents the key characteristics of the individual studies included in the analysis.

Two authors (YY and YFJ) assessed the quality of 13 RCTs using the Cochrane risk of bias tool. All studies were randomised, with one study not describing a specific method of randomisation and 12 studies having specified it. All RCTs were at low risk of attrition bias. In addition, none of the 13 studies selectively reported their findings and only four studies had incomplete outcome measures. Regarding blinding bias of participants and investigators, 12 of the 13 RCTs had high or unclear risk for blinding and five studies were inadequately allocated for concealment (selection bias). Risk of bias maps were generated using Review Manager 5.3 (https://tech.cochrane.org/revman) and the specific results are shown in Fig. 2. The NOS scale was used to assess six observational studies (see Table SI for full assessment details). No study was excluded based on the NOS score; all six studies met the initial inclusion criteria (e.g., focusing on PIS and comparing ticagrelor combined with oral antiplatelet agents vs. oral antiplatelet agents alone) and were included in the analysis. Their NOS scores ranged from 7 to 8 (with a median of 8), demonstrating high scores in selection (all ≥3/4) and comparability (all ≥1/2), indicating a low risk of bias.

A total of 4 studies (10,23,24,37) reporting on the occurrence of mRS scores ranging from 0 to 2 within 3 months were comprehensively analyzed. Data were obtained from 2,032 patients, among whom 1,028 cases received tirofiban in combination with oral antiplatelet drugs, while 1,004 cases were treated only with oral antiplatelet drugs. Heterogeneity analysis showed I²=0% and P>0.1 (low heterogeneity), so the fixed-effects model was used. The pooled OR showed the experimental group had a higher 3-month mRS 0-2 rate [OR=1.31, 95% CI (1.10, 1.58), P=0.003] (Fig. 3A).

NHISS score. A total of 19 studies comparing the NHISS scores after treatment between the two groups were comprehensively analyzed. Among these, the data of 3 studies (23,29,39) did not follow a normal distribution and the specific arithmetic means and standard deviations were not provided explicitly in 4 studies (10,24,26,36). For the remaining 12 studies (25,27,28,30-35,37,38,40), 1,341 patients were included, with 672 placed in the experimental group and 669 in the control group. I²=80% and P<0.1 (high heterogeneity), so the random-effects model was used. The result demonstrated that the experimental group had a significantly lower NHISS score compared to the control group [SMD=-1.22, 95% CI (-1.49, -0.95), P<0.01]. Subgroup analysis was subsequently conducted based on different intervention measures, using tirofiban combined with oral single antiplatelet drugs or oral dual antiplatelet drugs. In the oral single antiplatelet drug subgroup, three studies involving 249 patients were incorporated, with 121 in the experimental group and 128 in the control group (28,31,37). Overall, the NHISS score of the experimental group was significantly lower than that of the control group [SMD=-0.99, 95% CI (-1.67, -0.31), P=0.004]. A total of nine studies, comprising 1,092 patients, were included in the oral dual antiplatelet drug subgroup (25,27,30,32-35,38,40). The experimental group comprised 551 participants, while the control group included 541 individuals. The analysis revealed that the NHISS score for the experimental group was significantly lower than that for the control group [SMD=-1.29, 95% CI (-1.58, -1.00), P<0.01]. The difference in the NHISS score between the experimental and control groups was more noticeable in the tirofiban combined with oral dual antiplatelet drug subgroup than in the subgroup using oral single antiplatelet drugs (Fig. 3B).

At the same time, an additional subgroup analysis included 12 studies for which data were available. In seven studies (n=859) in the short-course subgroup (27,28,32,33,37,38,40), NIHSS scores were significantly lower in the experimental group compared with the control group [SMD=-1.15, 95% CI (-1.51, -0.79), P<0.001]. The heterogeneity in this subgroup was high (I²=81.3%, P<0.001). In five studies (n=482) in the long-duration subgroup (25,30,31,34,35), NIHSS scores were also significantly lower in the trial group [SMD=-1.32, 95% CI (-1.80, -0.84), P<0.001], with similarly high heterogeneity (I²=82.4%, P<0.001). The overall effect remained significant [SMD=-1.22, 95% CI (-1.49, -0.95), P<0.001] under the random-effects model, with considerable heterogeneity observed across all studies (I²=80.0%, P<0.001). This suggests that tirofiban improved neurological function regardless of the duration of treatment, but there was no reduction in heterogeneity in either subgroup (Fig. 3C).

mRS score. A total of 6 studies (28,34,35,38-40) comparing the mRS scores after treatment were comprehensively analyzed. Data were retrieved from these studies, which involved 616 patients. Among them, 304 cases were included in the experimental group and 312 cases in the control group. Heterogeneity was significant (I²=87.5%, P<0.1), and a random-effects model was used. Overall, it was demonstrated that the mRS score of the experimental group was markedly lower than that of the control group [SMD=-1.10, 95% CI (-1.50, -0.53), P<0.01] (Fig. 3D). Notably, a higher mRS score indicates more severe symptoms.

ADL score. A total of 3 studies (25,28,31) comparing ADL scores were comprehensively analyzed. Data were obtainable from these studies. There were 244 patients in total, with 122 in the experimental group and 122 in the control group. The heterogeneity was significant (I²=78.9%, P<0.1) and a random-effects model was employed. Overall, the experimental group was shown to be superior to the control group in ADL scores [SMD=1.19, 95% CI (0.58, 1.80), P<0.01] (Fig. 4A).

BI index. A total of 3 studies (30,33,35) that reported on the BI index were included. The data retrievable from these studies involved a total of 557 patients, with 287 in the experimental group and 270 in the control group. A random-effects model was used due to significant heterogeneity (I²=78.9%, P<0.1). Overall, the SMD was significantly >0, which indicated that the BI index of the experimental group was superior to that of the control group [SMD=1.36, 95% Cl (0.92, 1.79), P<0.01] (Fig. 4B).

PAgR. A total of 7 studies (26,30-35) involving the measurement of PAgR were comprehensively analyzed. Data can be obtained from these studies, with 976 patients in total, 501 in the experimental groups and 475 in the control groups. Significant heterogeneity was observed (I²=92.4%, P<0.001) and a random-effects model was applied. It shows that the PAgR in the experimental group was significantly lower compared with that in the control group [SMD=-1.19, 95% CI (-1.71, -0.67), P<0.01]. Further subgroup analysis was based on different intervention measures. In the single antiplatelet drug subgroup, two studies included 193 patients (101 in experimental group and 92 in control group) (26,31), and the PAgR of the experimental group was lower [SMD=-0.92, 95% CI (-1.29, -0.55), P<0.01]. In the oral dual antiplatelet drug subgroup, 5 studies included 783 patients (400 in experimental group and 383 in control group) (30,32-35), and the PAgR of the experimental group was lower [SMD=-1.31, 95% CI (-2.00, -0.61), P<0.01]. Obviously, in terms of the PAgR results, the experimental group had a more significant decrease in PAgR than the control group in the subgroup using a combination of tirofiban and oral dual antiplatelet drugs (Fig. 4C).

Another subgroup analysis included 7 studies with available data. In the short-term subgroup with 3 studies (n=564) (26,32,33), the PAgT in the experimental group was significantly reduced compared to the control group [SMD=-1.78, 95% CI (-2.46, -1.09), P<0.001]. This short-term subgroup exhibited substantial heterogeneity (I²=90.6%, P<0.001). In the long-term subgroup with 4 studies (n=412) (30,31,34,35), the experimental group also showed a significant reduction in PAgR [SMD=-0.71, 95% CI (-0.96, -0.47), P<0.001]. In contrast, this long-term subgroup showed low heterogeneity (I²=33.0%, P=0.215). The overall effect was significant [SMD=-1.19, 95% CI (-1.71, -0.67), P<0.001] under the random-effects model. However, heterogeneity was observed across all studies (I²=92.4%, P<0.001). This indicates that the inhibitory effect of tirofiban on platelet aggregation was more pronounced in the short-term subgroup, while the heterogeneity in the long-term subgroup significantly decreased (Fig. 4D).

PAdR. Data were obtainable from 7 studies (26,30-35), encompassing a total of 976 patients, among whom 501 were allocated to the experimental group and 475 to the control group. A random-effects model was used owing to significant heterogeneity (I²=83.0%, P<0.1). Overall, it was demonstrated that the experimental group had a lower PAdR compared to the control group [SMD=-1.00, 95% CI (-1.34, -0.66), P<0.01]. Subgroup analysis was carried out based on distinct intervention modalities, specifically, tirofiban in combination with oral single antiplatelet or oral dual antiplatelet drugs. In the oral single antiplatelet drug subgroup, two studies (26,31) were incorporated, comprising 193 patients in total. Of these, 101 participants were allocated to the experimental group and 92 to the control group. It was ascertained that the experimental group had a lower PAdR than the control group [SMD=-0.76, 95% CI (-1.05, -0.47), P<0.01]. In the oral dual antiplatelet drug subgroup, five studies (30,32-35) were included, involving 783 patients, with 400 in the experimental group and 383 in the control group. The results indicated that the experimental group had a lower PAdR than the control group [SMD=-1.10, 95% CI (-1.56, -0.64); P<0.01]. Evidently, for the PAdR results, the experimental group in the subgroup using tirofiban and oral dual antiplatelet drugs had a significantly more significant PAdR reduction than the control group. Furthermore, this reduction was more pronounced than the subgroup using oral single antiplatelet drugs (Fig. 5A).

Effective rate. A total of 8 studies (25,27,29,31,32,34,35,38) regarding the determination of the effective rate were comprehensively analyzed. Data were obtainable from these studies, encompassing 776 patients, among whom 388 were in the experimental group and 388 in the control group. Heterogeneity was negligible (I²=0.0%, P=0.744), and thus a fixed-effects model was used. Overall, it was found that the experimental group had an advantage over the control group [OR=3.49, 95% CI (2.28, 5.37), P<0.01] (Fig. 5B).

A total of 12 studies (10,23,24,32-40) focusing on sICH were comprehensively analyzed. Data were obtainable from these studies, including a total of 3,086 patients, with 1,574 in the experimental group and 1,512 in the control group. No significant heterogeneity was found (I²=0.0%, P>0.1), and a fixed-effects model was used. Overall, no significant difference was found in the incidence of intracranial hemorrhage between the two groups [OR=1.77, 95% CI (0.86, 3.64), P=0.118] (Fig. 5C).

Other intracranial hemorrhage. A total of 14 studies (23,25,27-29,31-36,38-40) concerning other intracranial hemorrhages were comprehensively analyzed. Data were obtained from these studies, comprising 1,835 patients, 920 in the experimental groups and 915 in the control groups. No significant heterogeneity was detected (I²=0.0%, P>0.1), and a fixed-effects model was employed. Overall, no significant difference in the incidence of cerebral hemorrhage was observed between the two groups [OR=1.07, 95% CI (0.74, 1.54), P=0.734] (Fig. 5D).

Mortality rate. A total of 5 studies (23,24,36-38) regarding the mortality rate were comprehensively analyzed. Data were obtainable from these studies, which included 1,068 patients, among whom 543 were in the experimental group and 525 in the control group. Given the presence of moderate heterogeneity (I²=45.3%), a random-effects model was used for the meta-analysis. Overall, the two groups showed no statistically significant difference in mortality rates [OR=0.40, 95% CI (0.07, 2.35), P=0.098] (Fig. 6A).

Serious adverse events. A total of 3 studies (24,35,37), involving 649 patients (335 in the experimental group and 314 in the control group), were included in the meta-analysis of serious adverse events. Under the random-effects model, the analysis showed no heterogeneity among the studies (I²=0%, P=1.00), indicating that pooling their results was appropriate. The pooled analysis demonstrated that there was no statistically significant difference in the incidence of serious adverse events between the two groups [OR=1.10, 95% CI (0.33, 3.64), P=0.88 for effect] (Fig. 6B).

To assess the robustness of the results of the analyses of the present study for possible bias due to study design heterogeneity, pre-specified sensitivity analyses were performed, limited to RCTs. Functional recovery (90mRS 0-2) showed the same effect size (OR=1.31), there was a 95% CI overlap (1.10-1.58 vs. 1.06-1.59) and the I² increased from 0 to 2.6%. The NIHSS showed a 15% increase in effect size (SMD=-1.22 to -1.41) and the I² decreased from 80 to 67%. The safety profile remained stable, with no substantial change in the risk of sICH (OR=1.773 vs. 1.524) or mortality (OR=0.40). It is important to note that secondary outcomes including mRS, ADL and BI indices remained consistent in the direction of the effect, although only RCTs had wider CIs. When limited to RCT evidence, the persistence of treatment effects confirms the reliability of the primary conclusions of the present study (Table SII). The robustness of the main study results was verified. At the same time, future RCTs using standardized regimens (e.g., unified tirofiban dosing regimens) may reduce residual heterogeneity.

Publication bias was assessed using Egger's test and Begg's test, which showed no significant evidence of publication bias (P=0.861) (Fig. 7A), which was also confirmed by Begg's test (P=0.373) (Fig. 7B), indicating that the pooled effect size was not significantly affected by publication bias. Visual examination of the funnel plot (Fig. 7C) revealed a slight asymmetry, but in combination with statistical tests, this asymmetry is more likely to be due to random variation in a small-sample study than to systematic bias. Overall, the study was at low risk of publication bias and the reliability of the results was somewhat supported.

Inadequate allocation concealment and unblinding can lead to bias in assessing treatment effects. However, there are three pieces of evidence supporting the reliability of the primary conclusion: First, the objective safety outcomes remain unaffected; they rely on imaging or laboratory confirmation, and are less susceptible to detection bias. Second, the results from different study designs are consistent: The forest plots (Fig. 3) show that the effect direction of all efficacy outcomes (e.g., NIHSS, mRS, ADL) is consistent in both RCTs and observational studies. Third, the sensitivity analysis confirms that excluding studies with a high risk of bias did not result in substantial changes to the effect estimates [e.g., after removing three non-blinded RCTs, the mean difference in NHISS changed from -1.22 to -1.09, as indicated by the sensitivity analysis (data not shown)]. Therefore, although the extent of functional improvement may be moderately overestimated, the treatment effect shows a positive trend, with statistical significance and safety thoroughly validated.