The process of study selection according to the inclusion and exclusion criteria is shown in the PRISMA flow diagram (Fig. 1). A total of 1,625 references were screened, and 51 studies were identified as potentially relevant studies whose full texts were retrieved. After removing studies that did not meet the inclusion criteria, 19 studies [8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26] with 9411 patients were included in the data assessment.

PRISMA schematic diagram of the search strategy

The main characteristics of the included studies are shown in Supplementary Table 1. Table 1 shows the details of the outcomes. Although the authors did not report the ECMO type in 4 studies [14, 15, 19, 22], 1293 patients were supported by veno-venous extracorporeal membrane oxygenation (VV-ECMO), 6863 patients were supported by veno-arterial extracorporeal membrane oxygenation (VA-ECMO), and 2 patients were on other types of ECMO. Patient age ranged from 23 to 76 years. The average ECMO duration ranged from 4 to 17 days. Eighteen studies reported the incidence of HIT. Of these, 12 studies reported suspicions of HIT. All studies presented their diagnostic mode. All studies performed immunoassays, and 15 studies performed functional assays to confirm HIT. Diagnostic algorithms were described in 9 studies, and 11 studies presented the PLT count of HIT patients. Fifteen studies reported alternative anticoagulation therapy for confirmed HIT patients. Eleven studies reported the occurrence of thrombotic events in HIT patients. Seven studies reported the incidence of bleeding events in HIT patients. Fourteen studies reported the mortality of HIT patients.

Because all studies eligible for inclusion were retrospective studies, we evaluated study quality using the Newcastle‒Ottawa Quality Assessment Scale adapted for cross-sectional studies, which showed a high level of quality in 10 studies with a score greater than 7/10. The other 9 studies achieved a score of 5/10. The summary of the risk of bias is reported in Supplementary Table 2. Funnel plots for all the included outcomes are shown in Supplementary Figs. 3–6.

A total of 18 studies reported the number of patients with HIT during ECMO support [8,9,10,11,12,13,14,15,16,17,18,19,20, 22,23,24,25,26]. The lowest incidence of HIT was 0.4%, while the highest was 39.3% [10, 25]. According to the random-effects analysis, the pooled incidence of HIT on ECMO was 4.2% (95% CI: 2.7–5.6, I2 = 90.5%) (Fig. 2).

Forest plot showing meta-analysis of the incidence of HIT in ECMO-supported patients

The pooled incidence of HIT on ECMO was 4.2% (95% CI: 2.7–5.6, I2 = 90.5%). The black diamonds indicate individual estimates, and the black lines indicate individual 95% CIs. The gray squares represent the individual study weights. The vertical red dashed line indicates the pooled estimate. The vertical axis of the white diamond indicates a pooled estimate, whereas the horizontal axis indicates a pooled 95% CI.

In most institutions, HIT is suspected when the patient has a 4Ts score ≥ 4, a decrease in the PLT over 50%, or thrombosis after receiving heparin. Twelve of 19 studies presented the occurrence of suspected HIT [8, 10, 12,13,14,15, 17,18,19, 22, 23, 26]. The lowest incidence of suspected HIT is 0.7%, and the highest incidence of suspected HIT is 38.8% [10, 19]. There was severe heterogeneity (I2 = 98.2%). A random-effects model was used to analyze the data. As shown in Figs. 3 and 15.9% (95% CI: 9.0-22.8) of patients who were supported by ECMO were suspected of having HIT.

Forest plot showing the meta-analysis of the incidence of suspected HIT in ECMO-supported patients

The pooled incidence of HIT on ECMO was 15.9% (95% CI: 9.0-22.8, I2 = 98.2%). The black diamonds indicate individual estimates, and the black lines indicate individual 95% CIs. The gray squares represent the individual study weights. The vertical red dashed line indicates the pooled estimate. The vertical axis of the white diamond indicates a pooled estimate, whereas the horizontal axis indicates a pooled 95% CI.

Nine studies [8, 10, 13, 15, 17, 21,22,23, 26] described the diagnostic algorithms used to determine the likelihood of HIT in ECMO patients. The 4Ts score (score 0–3 = low HIT probability; score 4–5 = intermediate HIT probability; score 6–8 = high HIT probability) was used in all 9 studies. The 4T score of 11.2-30% of patients on ECMO was ≥ 4 [15, 17]. According to two studies [13, 21], 80-84.5% of confirmed HIT patients who received ECMO support had a 4Ts ≥ 4. The median 4Ts in HIT patients was 5 in 3 studies [8, 10, 26]. The HEP (HIT Expert Probability) score was used in two studies [15, 22], and 3.7% (5/134) and 20% (21/105) of ECMO patients were considered HEP positive. Only one study reported the outcome of LLL (Lilo-Le Louet score); 11.2% (15/134) of ECMO patients were LLL positive [15]. The details are shown in Table 1.

Immunoassays detect anti-heparin/platelet Factor 4 (PF4) antibodies, which are ordered assist in the diagnosis of HIT. At least one kind of immunoassay was performed in all studies. As shown in Table 1, ELISA was used as the immunoassay in the majority of studies. The cutoff of the ELISA optical density (OD) value was 0.4 in 7 studies [8, 11, 12, 15, 19, 21, 22], 0.5 in 2 studies [13, 17], and 1.0 in 1 study [14]. The mean/median ELISA OD value of HIT patients ranges from 1.080 to 2.10 [10, 19]. Hemosil AcuStar HIT-IgG was used in 3 studies [13, 20, 23], and the cutoff was 1.0 in Arachchillage et al.’s study [13].

Fifteen studies performed functional assays to confirm HIT. Among them, 12 studies [8, 10,11,12, 14, 15, 17,18,19, 21, 22, 26] used the serotonin release assay (SRA) as the confirmatory test for HIT. A heparin-induced platelet activation assay (HIPA) was used in 4 studies [10, 23, 24, 26], and a platelet aggregation test (PAT) was performed in 3 studies [10, 20, 26].

Functional assays were performed in 15 studies as the confirmatory test for HIT [8, 10,11,