Diagnostic accuracy of IgG-specific versus polyspecific enzyme-linked immunoassays in heparin-induced thrombocytopenia: a systematic review and meta-analysis

Summary

Background

There are conflicting data on whether the IgG-specific or polyspecific antiplatelet factor 4/heparin (PF4/H) enzyme-linked immunosorbent assay (ELISA) is preferred for the laboratory diagnosis of heparin-induced thrombocytopenia (HIT).

Objectives

To directly compare diagnostic accuracy of IgG-specific versus polyspecific ELISA in HIT.

Patients/Methods

A systematic search yielded nine studies comprising 1948 patients with suspected HIT tested by both IgG-specific and polyspecific ELISAs and a reference standard against which the diagnostic accuracy of the ELISAs could be measured. Study quality was assessed by QUADAS-2 criteria.

Results

There was identical sensitivity for IgG-specific and polyspecific ELISAs (0.97; 95% confidence interval (CI), 0.95–0.99) and superior specificity of IgG-specific compared with polyspecific ELISA (0.87 [0.85–0.88] vs. 0.82 [0.80–0.84], respectively). Performance was similar in subgroups using the serotonin release assay and a single commercial ELISA manufacturer. The negative predictive values of IgG-specific and polyspecific ELISA were similarly high (0.99, [0.99–1.00], but the positive predictive value was superior with IgG-specific compared with polyspecific ELISA (0.56 [0.52–0.61] vs. 0.32 [0.28–0.35], respectively). The positive likelihood ratio (LR) was higher in IgG-specific than polyspecific ELISA, although negative LRs were similar. There was high risk of quality concerns in domains of index test and reference standard.

Conclusions

The superior diagnostic accuracy of IgG-specific ELISA reinforces the ISTH-SSC recommendation for standardization of laboratory testing for HIT. Likelihood ratios of individual assays may be used in combination with clinical scoring systems as part of an integrated diagnostic algorithm for HIT.

Introduction

Heparin-induced thrombocytopenia (HIT) is a prothrombotic complication of heparin therapy. This potentially life-threatening syndrome results from formation of platelet-activating antibodies against multi-molecular complexes of platelet factor 4 (PF4) and heparin (PF4/H). Enzyme-linked immunosorbent assays (ELISAs) that detect antibodies against PF4/H complexes are widely used in the laboratory diagnosis of HIT [1,2]. These assays have high sensitivity but limited specificity because they are unable to distinguish platelet-activating antibodies from their more numerous non-pathogenic (i.e. non-platelet-activating) counterparts. Although anti-PF4/H immunoglobulins may be of the IgG, IgA or IgM class [3], IgG antibodies are thought to have the predominant, if not sole, capacity for triggering platelet activation responsible for the clinical manifestations of HIT [4,5].

Early anti-PF4/H ELISAs were exclusively polyspecific in nature, detecting IgG, IgA and IgM antibodies. IgG-specific ELISA kits became commercially available in the late-2000s in the hopes of improving diagnostic specificity. Currently, both IgG-specific and polyspecific ELISAs are in use, and there is conflicting information on whether one should be preferred. The International Society on Thrombosis and Haemostasis (ISTH) Scientific and Standardization Committee (SSC) stated a preference for the IgG-specific ELISA because of its basis in the pathobiology of HIT and superior specificity demonstrated in some studies [6]. At variance with ISTH recommendations, a recent meta-analysis of the diagnostic accuracy of immunoassays for HIT [7] reported no significant difference in specificity between the IgG-specific and polyspecific ELISAs. In fact, the polyspecific ELISA with intermediate optical density (OD) cut-off (and not the IgG-specific ELISA) was found to be one of only five immunoassays with both high sensitivity (> 95%) and high specificity (> 90%).

We hypothesized that the unexpected results of this meta-analysis may have arisen as a consequence of pooling data from heterogeneous studies with different study populations and reference standards. To minimize the effects of study heterogeneity on estimates of diagnostic accuracy, we conducted a systematic review and meta-analysis of only studies in which the polyspecific and IgG-specific ELISAs were directly compared within the same study, against the same reference standard, and from blood samples of the same patient.

Methods

Study identification

A literature search was performed of PubMed, EMBASE and The Cochrane Library databases from inception to 1 December 2015 using the following keywords: [(ELISA OR EIA OR enzyme linked immunosorbent assay OR enzyme immunoassay) AND (heparin induced thrombocytopenia OR HIT OR HITT OR heparin induced thrombocytopenia with thrombosis OR heparin associated thrombocytopenia.)] An additional review of the first 100 results using the above search terms in Google Scholar was performed, as was a manual search of reference lists of eligible studies. The search was restricted to English-language articles. References were screened for eligibility by title, abstract and full text as indicated by a single reviewer (HDH).

Study selection

Studies were eligible for inclusion if: (i) patients in whom there was clinical suspicion for HIT were enrolled (thus, serosurveillance studies were excluded); (ii) both polyspecific (IgG/A/M) and IgG-specific anti-PF4/H ELISAs were performed on aliquots from the same patient samples; (iii) a reference standard (which, at a minimum, included a functional HIT assay) was performed against which the performance of the polyspecific and IgG-specific ELISAs could be compared; and (iv) the study was published as a full-length manuscript. Studies were excluded if the numbers of true-positive (TP), false-positive (FP), true-negative (TN) and false-negative (FN) results for each assay could not be extracted or calculated.

Data extraction

Key characteristics of eligible studies were extracted by two independent reviewers (HDH and AMP). Data collected included author, year of publication, study design, single-versus multi-center design, country of study center (s), study population (e.g. medical, surgical, cardiovascular surgery, critically ill, etc.), median age, gender distribution, heparin agent (unfractionated heparin and/or low-molecular-weight heparin), reference standard, type of IgG-specific and polyspecific ELISA, and OD thresh-old. If available, pretest clinical probability of HIT (low, intermediate or high probability) was noted. The number of TP, FP, TN and FN results for both the polyspecific and IgG-specific assay were extracted or calculated from data provided in the manuscripts of eligible studies and underwent evaluation by a third reviewer (MB) to confirm the accuracy of calculations. Any discrepancies were discussed between the reviewers in an attempt to reach consensus. If a consensus could not be reached among the three reviewers, the discrepancy was brought to the principal investigator (AC) for final determination.

Data analysis

Primary outcome data for each ELISA included the number of TP, TN, FP and FN results as compared with the reference standard. Individual and pooled sensitivity and specificity values, positive and negative predictive values, and positive and negative likelihood ratios (LRs) were calculated.

We prespecified several subgroup and sensitivity analyses. Because the incidence of HIT varies by patient population (e.g. medical vs. surgical) [8–12], we planned subgroup analyses by patient population. Additionally, we planned subgroup analyses of patients with varying pretest clinical probabilities of HIT. We prespecified sensitivity analyses of studies that used a reference standard that included both a clinical assessment and washed platelet functional assay, in accordance with recommendations of the ISTH SSC [6]. Because different kits may vary in their operating characteristics, we also prespecified sensitivity analyses of studies using ELISA kits from the same manufacturer.

A positive immunoassay was defined by a threshold optical density (OD) value specified by each study’s authors. One study investigated two different OD thresholds, a standard and an ‘optimized’ threshold [13]. In this case, the standard threshold was used. Three studies (14–16) reported results from overlapping datasets. In this case, only the study with the largest number of evaluable patients was included [15].

Individual and pooled sensitivity and specificity values, positive and negative predictive values, and prevalence, and associated 95% confidence intervals were estimated using a fixed-effect model [17] within the full cohort and subgroups. Positive and negative LRs were generated from the pooled estimates of sensitivity and specificity. Stata13 software was used for statistical analyses (Stata-Corp. 2013; Stata Statistical Software: Release 13; Stata-Corp LP, College Station, TX, USA).

Quality assessment

The methodological quality of included studies was assessed by two independent reviewers (HDH and AMP) using QUA-DAS-2, a standardized tool that evaluates risk of bias and concerns regarding applicability to clinical practice in tests of diagnostic accuracy [18]. We developed signaling questions specific to our particular study (available in Data S1) addressing risk of bias and concerns regarding applicability in the four domains of patient selection, index test, reference standard, and flow and timing, as well as detailed decision criteria according to published guidelines.

Results

Study selection

Results of the comprehensive literature search are illustrated in the accompanying PRISMA diagram (Fig. 1). The search strategy yielded a total of 882 articles: 590 articles from PubMed, 292 from EMBASE and 0 from Cochrane. One additional relevant manuscript was identified through bibliography search. No additional articles were identified after review of the top 100 results of a Google Scholar search. Of the total number of articles (883), 363 were duplicates, leaving 520 articles for review. After exclusion of 401 articles by title and abstract, 119 articles underwent full-text review. Of the full-text articles, 102 were excluded because IgG-specific and polyspecific ELISAs were not performed on the same samples (n = 81), patients without clinical suspicion of HIT were enrolled (n = 8), the article was a review or editorial and did not report primary data (n = 8), the study was published as an abstract only (n = 3), or a confirmatory functional assay was not performed (n = 2). Of the remaining 17 studies, six were excluded because of inability to extract relevant data (n = 4) or use of patient datasets overlapping with those of an included study (n = 2). Nine studies met eligibility criteria and were included in the final meta-analysis [13,15,19–25].

Fig. 1.

PRISMA diagram. Selection of studies included in systematic review and meta-analysis.

Study characteristics

The nine eligible studies collectively enrolled 1948 patients and were performed in seven countries. Characteristics of these studies are summarized in Table 1. Patient populations included a mixture of medical, surgical, cardiovascular surgery, hemodialysis and critically ill patients. The prevalence of HIT among included studies ranged from 0.04 to 0.40, with a weighted pooled prevalence of 0.08 (95% CI, 0.07–0.09). Among nine included studies, six utilized a serotonin-release assay (SRA), one used the heparin-induced platelet activation (HIPA) test and one used a functional flow cytometric assay as part of the reference standard. Seven of the nine included studies made use of washed platelet assays.

Table 1.

Characteristics of studies

References	Setting (n)	Population (%)	Heparin agent	Median age (years)	No. (%) female	Use of washed platelets?	Prevalence of HIT	Index test		Reference standard
								IgG-ELISA [cut-off OD]	Poly-ELISA [cut-off OD]	IgG-ELISA [cut-off OD]	Poly-ELISA [cut-off OD]
Warkentin 2013 [19]	Canada, Australia, New Zealand (409*)	Mixed medical/surgical from PROTECT trial	UFH or dalteparin	NR	NR	Yes	0.04	In-house (McMaster) [> 0.45]	Hologic Gen-Probe [> 0.4]	4T score ≥ 4	SRA ≥ 50% release
Cuker 2013 [20]	United States (58 samples polyspecific ELISA-positive	CV surgery (28), non-CV surgery (14), medical (15), other (1)	NR	64†	28 (48.2%)	No	0.36	Hologic Gen-Probe [> 0.4]	Hologic Gen-Probe [> 0.4]	Clinical suspicion of HIT. 4T score assigned retrospectively by chart review. Considered true HIT if 4T score ≥ 4.	SRA ≥ 50% release
Galea 2013 [21]	France (200)	Not specified	UFH or LMWH	NR	NR	Yes	0.11	Zymutest HIA [> 0.3]	Zymutest HIA [> 0.5]	Clinical suspicion of HIT. 4T score assigned but all clinical probabilities tested.	SRA ≥ 20% release
McFarland 2012 [22]	US (116)	CV surgery (41), other surgery (14), 0ther medical (45)	UFH (92), LMWH (15), both (6), NR (3)	63	54 (46.6%)	Yes	0.22	In-house ELISA [≥ 0.4]	In-house ELISA [≥ 0.4]	Clinical suspicion of HIT. 4T score assigned retrospectively.	SRA ≥ 75% release and inhibition with high-dose heparin
Van Hoecke 2012 [13]	Belgium (87)	CV surgery, hemodialysis, medical (critically ill and treated for thrombosis)	UFH or LMWH	NR	NR	No	0.09	Zymutest HIA IgG, calculated cut-off [0.24-0.39]	Zymutest HIA IgGAM, calculated cut-off [0.41-0.46]	Clinical suspicion of HIT; development of thrombocytopenia with exposure to heparin.	Functional flow cytometry (% platelet activation ≥ 2 times control)
Morel-Kopp 2010 [23]	Australia (107 samples from 97 individuals)	Medical (51), surgical (56); majority low risk by 4T score (75)	NR	69†	39 (40.2%)	Yes	0.11	Zymutest IgG [> 0.4]	Zymutest IgGAM [> 0.4]	Clinical suspicion of HIT. 4T score assigned, but all clinical probabilities tested.	SRA ≥ 20% release
Pouplard 2010 [24]	France (101)	Medical (32), surgical (42), CV surgical (27)	UFH or LMWH	NR	44 (43.6%)	Yes	0.40	Zymutest HIA IgG, [≥ 0.5]	Zymutest HIA IgGAM, [≥ 0.5]	Clinical suspicion of HIT	SRA ≥ 20% release
Bakchoul 2009 [25]	Germany (500)	Mixed medical and surgical	UFH or LMWH	NR	NR	Yes	0.07	In-house IgG-ELISA [> 0.40]	Poly-ELISA GTI [> 0.4]	4T score ≥ 4	HIPA (loss of turbidity in 30 min in at least two suspensions of 0.2 U mL−1 but not 100 U mL−1)
Warkentin‡ 2008 [15]	Canada (405)	Mixed medical, surgical, including ICU and CV surgery	NR	NR	NR	Yes	0.10	In-house IgG-ELISA [≥ 0.45]	Poly-ELISA GTI [≥ 0.4]	Clinical suspicion of HIT. 4T score assigned at time of enrollment, but all clinical probabilities tested.	SRA ≥ 50% release

^*One patient not tested by polyspecific ELISA because of insufficient sample.

^†Mean value.

^‡Six patients underwent IgG-specific but not polyspecific testing because of insufficient sample. HIT, heparin-induced thrombocytopenia; NR, not reported; UFH, unfractionated heparin; OD, optical density; SRA, serotonin-release assay; CV, cardiovascular; LMWH, low-molecular- weight heparin; ICU, intensive care unit.

Among the polyspecific ELISAs, six utilized a cut-off OD of ≥ 0.4 and two a cut-off OD of ≥ 0.5 [21,24]. Among the IgG-specific ELISAs, one study utilized a cut-off OD of ≥ 0.3 [21], four utilized ≥ 0.4 [20,22,23,25], two utilized ≥ 0.45 [15,19] and one utilized ≥ 0.5 [24]. In the study by van Hoecke et al. [13], a calculated cut-off OD developed by the manufacturer was used for both IgG-specific and polyspecific ELISA. This ‘calculated cut-off’ was defined as percentage of the OD of the positive control minus OD of the blank. The reported ranges of calculated OD cut-offs were 0.24–0.39 for IgG-specific and 0.41–0.46 for polyspecific ELISA.

For one study [23], two different IgG (Zymutest and GTI-PF4-IgG) and polyspecific (Zymutest and Asserachrom^® IgGAM) ELISA pairs were analyzed per patient sample. In this case, data from the Zymutest IgG and Zymutest polyspecific ELISA comparison were included with exclusion of results from the alternate assays (GTI-PF4-IgG and Asserachrom^® IgGAM) to minimize heterogeneity and maintain independency of observations. Although all studies included patients in whom there was ‘clinical suspicion of HIT’, only three [19,20,25] of the nine studies included a clinical assessment (e.g. 4T score ≥ 4) as part of the reference standard in accordance with ISTH-SSC recommendations.

Study quality

Quality assessment as judged by the QUADAS-2 tool highlighted several significant limitations of included studies in both risk of bias and concerns regarding applicability to clinical practice. Three of nine studies had a high risk of bias with regards to patient selection because of non-consecutive patient sampling [20,22,24]. In the study by Cuker et al. [20], only patients in whom there was clinical suspicion of HIT who had a positive polyspecific ELISA result subsequently underwent testing by IgG-specific ELISA and SRA. In the study by McFarland et al. [22], patients were selected for inclusion based upon their ELISA OD result. Pouplard et al. [24] described patient sampling as ‘non-consecutive’.

Regarding the index test, one study carried a low risk of bias because it stated that results of ELISA testing were reported without knowledge of results of the reference standard [23]. The remaining studies made no mention of blinding to reference standard results and therefore carried uncertain risk. In regards to the reference standard, four studies carried high risk of bias because the reference standard either did not use washed platelets [13,20] or did not clearly include clinical criteria [15,23]. Additionally, in the study by McFarland et al. [22], the authors acknowledged that 4T scores were assigned retrospectively, and both treating clinicians and researchers had knowledge of laboratory assay results at the time of scoring. In the remaining studies, exact clinical assessment was unclear or there was no mention of whether reference standard results were interpreted without knowledge of the index test results; in these cases, risk of bias was classified as uncertain. Regarding flow and timing, there were risks of bias identified in studies where it was presumed but not unequivocally stated that all patients received the same reference standard (in particular, the same clinical scoring system) [25] or where not all patients were included in the analysis [15,22,23].

All studies carried a low concern for applicability in patient selection, as inclusion criteria mandated clinical suspicion for HIT in all tested subjects. All studies carried a high concern for clinical applicability in regards to the reference standard, as the functional laboratory assay utilized in each study was developed by and performed within individual institutions and not commercially available for widespread use. Concern for applicability of index tests was considered low for the five studies utilizing commercially available ELISAs [13,20,21,23,24] and high for the remaining four studies that utilized in-house ELISAs [15,19,22,25]. Additional details of study quality assessment scoring can be found in Table 2.

Table 2.

Study quality assessment by QUADAS-2 criteria

	Risk of bias				Applicability concerns
References	Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
Warkentin* 2013 [19]	Low	Unclear	Low	Low	Low	High	High
Cuker 2013 [20]	High	Unclear	High	Low	Low	Low	High
Galea 2013 [21]	Low	Unclear	Unclear	Low	Low	Low	High
McFarland 2012 [22]	High	Unclear	High	High	Low	High	High
Van Hoecke 2012 [13]	Low	Unclear	High	Low	Low	Low	High
Morel-Kopp 2010 [23]	Low	Low	High	High	Low	Low	High
Pouplard 2010 [24]	High	Unclear	Low	Low	Low	Low	High
Bakchoul 2009 [25]	Low	Unclear	Unclear	Low	Low	High	High
Warkentin 2008 [15]	Low	Unclear	High	High	Low	High	High

^*Determined from data reported in original article, the PROTECT trial, and discussions with Dr. Theodore Warkentin.

Primary analysis: diagnostic accuracy of IgG versus polyspecific ELISA

Pooled data from all nine studies demonstrated identical sensitivity for the IgG-specific and polyspecific ELISAs (0.97; 95% confidence interval (CI), 0.95–0.99) and superior specificity of the IgG-specific compared with the polyspecific ELISA (0.87 [0.85–0.88] vs. 0.82 [0.80-0.84]) at standard cut-offs. Individual and pooled sensitivity and specificity data for the full cohort are shown in Table 3 and presented as Forest plots in Fig. 2. The negative predictive values (NPVs) for both IgG-specific and polyspecific immunoassays were equally high (0.99 [0.99–1.00] for both). The positive predictive value (PPV) for IgG-specific ELISA (0.56 [0.52–0.61]) was superior to that of polyspecific ELISA (0.32 [0.28–0.35]). Positive LR was higher for the IgG-specific compared with the polyspecific ELISA (7.38 vs. 5.07) and negative LRs were equivalent for both assays (Table 4).

Table 3.

Sensitivity and specificity of IgG-specific and polyspecific ELISA

	Sensitivity (95% CI)		Specificity (95% CI)
Subgroup (n)	IgG-specific	Polyspecific	IgG-specific	Polyspecific
Full cohort (9)	0.97 (0.97–0.99)	0.97 (0.97–0.99)	0.87 (0.85–0.88)	0.82 (0.80–0.84)
Serotonin-release assay as confirmatory assay (7)	0.97 (0.94–0.99)	0.97 (0.94–0.99)	0.87 (0.85–0.88)	0.82 (0.80–0.84)
Zymutest IgG and polyspecific commercial assay (4)	0.96 (0.92–1.00)	0.96 (0.92–1.00)	0.91 (0.88–0.94)	0.86 (0.82–0.89)
Other IgG and polyspecific commercial assay or in-house assay (4)	0.98 (0.95–1.00)	0.98 (0.95–1.00)	0.85 (0.83–0.87)	0.79 (0.77–0.81)

Fig. 2.

Forest plots of sensitivity and specificity of IgG-specific and polyspecific antiplatelet factor 4/heparin enzyme-linked immunoassays for eligible studies.

Table 4.

Operating characteristics of IgG-specific and polyspecific ELISA

	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	PLR*	NLR*
IgG-specific	0.97 (0.97–0.99)	0.87 (0.85–0.88)	0.56 (0.52–0.61)	0.99 (0.99–1.00)	7.38	0.03
Polyspecific	0.97 (0.97–0.99)	0.82 (0.80–0.84)	0.32 (0.28–0.35)	0.99 (0.99–1.00)	5.07	0.03

CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value; PLR, positive likelihood ratio; NLR, negative likelihood ratio.

^*Calculated as point estimates from sensitivity and specificity.

Sensitivity and subgroup analyses

Seven studies utilized an SRA as the functional assay included in their reference standard. As in the overall cohort, sensitivities of the IgG-specific and polyspecific ELISA tests were identical (0.97 [0.94–0.99]) in these seven studies, and specificity of the IgG ELISA was superior to that of the polyspecific ELISA (0.87 [0.85–0.88] vs. 0.82 [0.80–0.84]). The heterogeneity of the SRAs used in this subgroup should be noted: six studies used washed platelets [15,19,21–24] and one used platelet-rich plasma [20]. Additionally, three different positivity thresholds were used: ≥ 20% [20,21,23,24], ≥ 50% [15,19] and ≥ 75% serotonin release [22].

Four studies utilized the Zymutest IgG-specific and polyspecific ELISAs [13,21,23,24]. The remaining studies used either in-house immunoassays [22], an alternate commercial ELISA (Hologic-GTI) [20] or a combination of in-house and commercial assays [15,19,25]. Among the studies that used the Zymutest assays, sensitivity was identical for IgG-specific and polyspecific ELISAs (0.96 [0.92–1.00]) and was similar to the sensitivity observed in the full cohort. Also consistent with the full cohort, the Zymutest IgG-specific ELISA exhibited greater specificity compared with the Zymutest polyspecific assay, although the confidence intervals overlapped (0.91 [0.88–0.94] vs. 0.86 [0.82–0.89]).

Of the nine eligible studies, only two [19,25] used a reference standard that included both clinical criteria and a washed platelet functional assay in accordance with ISTH-SSC recommendations. Only one study [19] stratified a subset of SRA-positive patients (n = 17) by 4T score. Therefore, the prespecified sub-analyses by either alignment with ISTH-SSC testing recommendations or by pretest clinical probability score could not be performed.

Discussion

Key findings

IgG-specific and polyspecific ELISAs demonstrated equally high sensitivity across various reference standards and subgroups; specificity of the polyspecific assay, however, was consistently lower than that of the IgG-specific ELISA. The IgG-specific ELISA also conferred a higher PPV than the polyspecific ELISA, although both modalities retained high NPV. Sensitivity analyses among studies utilizing an SRA as part of the reference standard showed similar results. These findings imply that the IgG-specific ELISA has superior diagnostic accuracy compared with the polyspecific ELISA at standard OD thresholds.

Strengths and limitations

The greater specificity of the IgG-specific ELISA is consistent with the notion that IgG antibodies carry the primary, if not sole, potential for pathogenicity in HIT and supports the ISTH-SSC recommendation to use IgG-specific antibody testing in preference to a polyspecific assay. However, our results are inconsistent with a recent meta-analysis by Nagler and colleagues [7], which did not demonstrate an appreciable difference in specificity between the polyspecific and IgG-specific ELISAs. A potential explanation for this discrepancy may be the incorporation of very different studies into the polyspecific and polyspecific and IgG-specific ELISA subgroups. The disparate assays, study populations, study designs and reference standards among groups the analysis could have contributed to misleading results. For example, in one study of patients in the medical intensive care unit, the point estimate of specificity for the polyspecific ELISA was 93.3%, higher than that observed in most studies [26]. This study did not include an IgG-specific ELISA and therefore contributed data to the pooled estimates of the diagnostic accuracy of the polyspecific but not the IgG-specific ELISA [7]. To minimize the potential effects of study heterogeneity, we limited our analysis to studies in which the polyspecific and IgG-specific ELISAs were directly compared with one another using the same patient samples and reference standard.

Nearly all eligible studies used a cut-off of 0.4–0.5 OD units (Table 1). Although cut-offs in this range are consistent with manufacturer-recommended thresholds for commercial kits and are in wide clinical use, multiple studies suggest that both ‘intermediate’ (0.8–1.4) and ‘high’ (≥ 1.4) OD cut-offs may improve diagnostic accuracy [7,15,25–27]. We were not able to assess the diagnostic accuracy of the polyspecific and IgG-specific ELISAs at higher cut-offs because these thresholds were not investigated in all but one of the eligible studies [13]. We cannot exclude the possibility that a somewhat higher cut-off for the polyspecific ELISA would yield operating characteristics similar to those of the IgG-specific ELISA.

We included secondary pooled point estimates of predictive values and likelihood ratios because these data are often sought by clinicians weighing the performance merits of a diagnostic test [28]. However, we acknowledge these parameters are not only challenging to assess and interpret in a meta-analysis, but their validity and reporting are controversial [29]. Predictive values depend on disease prevalence [30,31], which differed among included studies. Pooled estimates of predictive value should therefore be interpreted with caution. Furthermore, meta-analysis of LRs is mathematically complex. Often when meta-analysis of LRs is performed it does not accurately represent this test characteristic [29]. Thus, some experts suggest exclusion of LRs in meta-analysis of diagnostic accuracy. An alternative is to derive LRs from pooled sensitivity and specificity estimates. We chose this second approach in our study [29].

The quality of included studies varied and was often difficult to assess because of incomplete reporting of methodology. Inclusion of exclusively published studies and English-language studies raises the possibility of publication bias and language bias, although the impact, if any, is expected to be small in meta-analyses of diagnostic accuracy [32,33].

Implications for clinical practice

Test characteristics reported here can be used as part of a diagnostic algorithm to aid in clinical decision-making. The LRs generated for both the IgG-specific and polyspecific ELISAs at low-threshold ODs (Table 4) can be combined with clinical estimates of the pretest probability of disease using tools such as the 4Ts score, HIT Expert Probability (HEP) score or other standardized means of pretest probability assessment to estimate the post-test probability of disease [34–37].

For clinical laboratories establishing immunoassay testing for HIT, this study offers direct comparison of test performance to inform choice of assay. For coagulation laboratories experienced in using either IgG-specific or polyspecific ELISA, these results may either reinforce confidence or highlight limitations in local testing practices. The superior diagnostic accuracy of IgG-specific ELISA demonstrated in this study suggests that IgG-specific ELISA, compared with polyspecific ELISA, should reduce the number of false-positive results in initial screening for HIT without compromising sensitivity. These results are aligned with the ISTH-SSC recommendations in favor of IgG-specific testing.

Addendum

H. D. Husseinzadeh designed the study protocol, conducted the literature search, reviewed eligible studies, abstracted data, conducted statistical analysis, and wrote the manuscript. P. Gimotty assisted with statistical analysis, and revised the manuscript. A. M. Pishko reviewed eligible studies, abstracted data, and revised the manuscript. M. Buckley confirmed extracted data values included in meta-analysis. T. E. Warkentin provided critical data and revised the manuscript. A. Cuker developed the study concept, designed the study protocol and wrote the manuscript.

Supplementary Material

Supplement 1

Data S1. QUADAS-2 quality assessment questions.

Essentials.

• Immunoassay specificity varies in heparin-induced thrombocytopenia (HIT) testing.

• This meta-analysis examined 9 studies that tested samples by both IgG and polyspecific methods.

• IgG-specific assays confer superior diagnostic accuracy compared with polyspecific assays.

• These results further support recommendations in favor of IgG-specific testing.