FDA 510(k) Search - By Innolitics (SaMD and AI Experts)

HemosIL HIT-Ab(PF4-H) is a qualitative, fully automated, latex enhanced immunoassay for the detection of anti-platelet factor 4/heparin (PF4/H) antibodies. The assay is for use in human 3.2% or 3.8% citrated plasma on the ACL TOP® Family of instruments in a laboratory setting.

The result provided by the assay should be interpreted as either positive based on the assay cut-off (1.0 U/ mL). The positive or negative result aids in determining the risk for heparin induced thrombocytopenia (HIT) when used in conjunction with other laboratory and clinical findings.

Anti-PF4/Heparin antibodies are commonly found in patients with HIT. For use in adult population suspected of HIT. Not for use in isolation to exclude HIT.

HemosIL HIT-Ab(PF4-H) Controls are for the Quality Control of the HemosIL HIT-Ab(PF4-H) assay as performed on the ACL TOP® Family of instruments.

For prescription use.

Device Description

The HemosIL HIT-Ab(PF4-H) kit is a latex particle enhanced immuno-turbidimetric assay to detect total Anti-PF4/Heparin (PF4/H) antibodies found in HIT patients. A monoclonal antibody that mimics human HIT antibodies is coated onto latex particles.
In the presence of PF4 from human platelets complexed to polyvinyl sulfonate (PVS), and the patient sample, a competitive agglutination reaction occurs.
The degree of agglutination is inversely proportional to the concentration of antibodies in the sample and is determined by measuring the decrease of transmitted light caused by the aggregates.
The HemosIL HIT- Ab(PF4-H) kit consists of:
Latex Reagent: Suspension of polystyrene latex particles coated with purified mouse monoclonal anti-PF4-Heparin in Tris buffer, containing bovine serum albumin, stabilizers and preservative.
Stabilizer: PBS buffer containing bovine serum albumin, stabilizers and preservative.
Complex: Solution of PF4-PVS complex (PF4 from human platelets complexed to PVS), in PBS buffer containing bovine serum albumin, stabilizers and preservative. Contains 0.02% Bronidox™ as a preservative.
Calibrator: Lyophilized solution of a monoclonal anti- PF4-Heparin antibody in Tris buffer containing bovine serum albumin, stabilizers and preservative.

The Low and High HIT-Ab(PF4-H) Controls are prepared by means of a dedicated process and contain different concentrations of humanized monoclonal anti-PF4-Heparin-human IgG.
Low HIT-Ab(PF4-H) Control: Control intended for the assessment of precision and accuracy of the assay at PF4/H antibody levels at or below the cut-off.
High HIT-Ab(PF4-H) Control: Control intended for the assessment of precision and accuracy of the assay at abnormal PF4/H antibody levels.
Use of both controls is recommended for a complete quality control program.

AI/ML Overview

Here's a breakdown of the acceptance criteria and the studies performed for the HemosIL HIT-Ab(PF4-H) device, based on the provided FDA 510(k) summary:

1. Table of Acceptance Criteria and Reported Device Performance

The document does not explicitly state "acceptance criteria" in a singular section with numerical targets for clinical performance. Instead, it presents various performance metrics derived from different studies. The conclusion states that the device is "substantially equivalent" to the predicate and "safe and effective."

Given this, I will infer "acceptance criteria" from the comparative studies presented and consider the reported performance to be the "device performance."

Category	Specific Metric	Implied Acceptance Criteria (Inferred)	Reported Device Performance	Study
Precision	%CV for Low HIT-Ab(PF4-H) Control (Total)	(Not explicitly stated, but generally below 10-15% for controls is considered good for immunoassay)	Reagent Lot 1: 9.9%
Reagent Lot 2: 9.3%
Reagent Lot 3: 9.9%
Low Control Lot 1: 9.9%
Low Control Lot 2: 9.2%
Low Control Lot 3: 8.9%	Multi-Reagent Lot Precision, Multi-Control Lot Precision
	%CV for High HIT-Ab(PF4-H) Control (Total)	(Not explicitly stated, but generally below 10-15% for controls is considered good for immunoassay)	Reagent Lot 1: 3.3%
Reagent Lot 2: 3.9%
Reagent Lot 3: 4.0%
High Control Lot 1: 3.3%
High Control Lot 2: 3.0%
High Control Lot 3: 2.6%	Multi-Reagent Lot Precision, Multi-Control Lot Precision
	%CV for Patient Samples (Total)	(Not explicitly stated, but generally below 10-15% for clinical samples is considered good)	Ranging from 3.8% to 13.5% across various plasma samples and reagent lots	Multi-Reagent Lot Precision, Unadulterated Patient Plasma Precision
Reproducibility (Total %CV)	Low HIT-Ab(PF4-H) Control	(Not explicitly stated, but generally acceptable for multi-site)	Reagent Lot 1: 15.3%
Reagent Lot 2: 7.4%
Reagent Lot 3: 11.0%
Pooled 3-Site Data (Control Lot): 15.3%	Multi-Reagent Lot Site-to-Site Reproducibility, Multi-Control Lot Reproducibility
	High HIT-Ab(PF4-H) Control	(Not explicitly stated, but generally acceptable for multi-site)	Reagent Lot 1: 6.1%
Reagent Lot 2: 5.2%
Reagent Lot 3: 5.5%
Pooled 3-Site Data (Control Lot): 6.1%	Multi-Reagent Lot Site-to-Site Reproducibility, Multi-Control Lot Reproducibility
Instrument Equivalency	Correlation Coefficient (r) between ACL TOP 700 and 500 CTS	(Typically > 0.95 for good correlation)	0.9947	Instrument Model Equivalency
	Correlation Coefficient (r) between ACL TOP 700 and 300 CTS	(Typically > 0.95 for good correlation)	0.9981	Instrument Model Equivalency
Linearity Range	Auto Rerun off	Performance to support claims	0.6 – 5.7 U/mL	Linearity and Test Ranges
	Auto Rerun on	Performance to support claims	2.1 – 16.0 U/mL	Linearity and Test Ranges
Method Comparison vs. Predicate (Internal)	PPA	(Generally aims for high agreement, e.g., >80%)	84% (68/81)	Method Comparison (Internal)
	NPA	(Generally aims for high agreement, e.g., >70-80%)	78% (29/37)	Method Comparison (Internal)
	Total Percent Agreement	(Generally aims for high agreement, e.g., >80%)	82% (97/118)	Method Comparison (Internal)
Method Comparison vs. Predicate (Multicenter)	PPA	(Generally aims for high agreement, e.g., >70-80%)	75% (67/89)	Multicenter Method Comparison
	NPA	(Generally aims for high agreement, e.g., >90%)	93% (504/543)	Multicenter Method Comparison
	Total Percent Agreement	(Generally aims for high agreement, e.g., >80%)	90% (571/632)	Multicenter Method Comparison
Method Comparison vs. SRA (HemosIL HIT-Ab(PF4-H))	PPV	(Aids in determining risk, specific threshold not given as primary criterion)	37% (33/89)	Multicenter Method Comparison
	NPV	(Aims for high negative predictive power)	87% (389/448)	Multicenter Method Comparison
Method Comparison vs. SRA (Predicate - Asserachrom HPIA)	PPV	(For comparison to newly developed device)	46% (31/68)	Multicenter Method Comparison
	NPV	(For comparison to newly developed device)	87% (408/469)	Multicenter Method Comparison
Method Comparison vs. ASH 2013 Guidelines	PPA	(Aims for high agreement with clinical guidelines)	89% (17/19)	HemosIL HIT-Ab(PF4-H) Assay vs. 2013 American Society of Hematology (ASH) Guidelines
	NPA	(Aims for high agreement with clinical guidelines)	86% (446/518)	HemosIL HIT-Ab(PF4-H) Assay vs. 2013 American Society of Hematology (ASH) Guidelines
	NPV	(Aims for high negative predictive value with clinical guidelines)	100% (446/448)	HemosIL HIT-Ab(PF4-H) Assay vs. 2013 American Society of Hematology (ASH) Guidelines

2. Sample Sizes Used for Test Sets and Data Provenance

Multi-Reagent Lot Precision:
- Sample Size: 2 controls (low and high) and 5 patient plasma pools. Each tested in duplicate, twice a day for 20 days (80 replicates per level per lot). Total of 7 levels * 80 replicates = 560 data points per reagent lot. With 3 reagent lots, this is 1680 data points for the precision study itself.
- Data Provenance: Not explicitly stated for patient pools, but controls are internal. Patient plasma samples 1-5's origin is not specified but are described as "prepared at different levels to span the assay range", implying they are manufactured or manipulated samples.
- Nature: Prospective (testing conducted for the study).
Multi-Reagent Lot Site-to-Site Reproducibility:
- Sample Size: 2 controls (low and high), 3 patient pools, and 1 manufactured material (Plasma Sample 4). Each tested in triplicate, twice a day for 5 days (30 replicates per level). Total of 6 levels * 30 replicates = 180 data points per site per reagent lot. With 3 sites and 3 reagent lots, this is 180 * 3 * 3 = 1620 data points.
- Data Provenance: Not explicitly stated for patient pools or manufactured material, but they are "patient pools (2 positive)" and "a manufactured material containing a citrated plasma sample spiked with monoclonal anti-PF4-Heparin antibody."
- Nature: Prospective (testing conducted for the study).
Multi-Control Lot Precision:
- Sample Size: 2 control levels (low and high) from 3 different control lots. Each tested in duplicate, twice a day for 20 days (80 replicates per level per lot). Total of 6 levels * 80 replicates = 480 data points.
- Data Provenance: Controls are manufactured by Instrumentation Laboratory (IL).
- Nature: Prospective (testing conducted for the study).
Multi-Control Lot Reproducibility:
- Sample Size: 2 control levels (low and high). Each tested in triplicate, twice a day for 5 days (30 replicates per level). Total of 2 levels * 30 replicates = 60 data points per site. With 3 sites, 180 data points.
- Data Provenance: Controls are manufactured by Instrumentation Laboratory (IL).
- Nature: Prospective (testing conducted for the study).
Multi-Calibrator Lot Precision:
- Sample Size: 3 different calibrator lots. Each tested in duplicate, twice a day for 20 days (80 replicates per lot). Total of 3 lots * 80 replicates = 240 data points.
- Data Provenance: Calibrators are kit components from the manufacturer.
- Nature: Prospective (testing conducted for the study).
Unadulterated Patient Plasma Precision:
- Sample Size: 2 patient plasma samples. Each tested in duplicate, twice a day for 10 days (40 replicates per sample). Total of 2 samples * 40 replicates = 80 data points.
- Data Provenance: Collected at a hospital in the United States.
- Nature: Retrospective (though tested prospectively for precision study, samples were previously collected).
Instrument Model Equivalency:
- Sample Size: 51 citrated plasma samples.
- Data Provenance: Individual clinical patients suspected of HIT.
- Nature: Retrospective (samples previously collected, tested on multiple instruments).
Linearity and Test Ranges:
- Sample Size: Not explicitly stated as a number of patient samples, but 10 concentrations across 0.6-5.7 U/mL and a third extremely high positive sample for 2.1-16.0 U/mL. Two positive patient samples were used to prepare serial dilutions.
- Data Provenance: Patient samples spiked into donor plasma from a blood bank.
- Nature: Prospective (dilution series prepared and tested).
Heparin Sensitivity:
- Sample Size: 126 citrated plasma samples.
- Data Provenance: Non-HIT suspected, heparin treated patients (UFH or LMWH).
- Nature: Retrospective (samples previously collected).
Antiphospholipid Syndrome (APS):
- Sample Size: 40 citrated plasma samples.
- Data Provenance: Patients diagnosed with Antiphospholipid Syndrome (APS).
- Nature: Retrospective (samples previously collected).
Reference Interval – Healthy Donors:
- Sample Size: 131 citrated plasma samples.
- Data Provenance: Apparently healthy individuals.
- Nature: Retrospective (samples previously collected).
Reference Interval – Heparin Exposed, HIT Suspected Patients (HIT Negative):
- Sample Size: 122 citrated plasma samples.
- Data Provenance: HIT-suspected, but negative by a commercially available ELISA.
- Nature: Retrospective (samples previously collected).
Cut-Off Determination:
- Sample Size: 63 citrated plasma samples (31 SRA positive, 32 SRA negative).
- Data Provenance: Hospitalized patients exposed to heparin with clinical signs of HIT.
- Nature: Retrospective (samples previously collected and tested by SRA).
Method Comparison (Internal):
- Sample Size: 118 frozen citrated patient plasma samples.
- Data Provenance: Patients referred for HIT testing from a hospital in the US and a hospital in France.
- Nature: Retrospective (samples previously collected, tested against predicate).
Multicenter Method Comparison (against predicate and SRA):
- Sample Size: 632 patient samples initially (N=537 for SRA comparison after removing invalid/indeterminate SRA results).
- Data Provenance: From patients exposed to heparin who showed HIT related symptoms, collected at three (3) hospitals.
- Nature: Multicenter, Retrospective (samples previously collected, tested against predicate and SRA).
HemosIL HIT-Ab(PF4-H) Assay vs. 2013 American Society of Hematology (ASH) Guidelines:
- Sample Size: 537 patient samples (subset of the multicenter study).
- Data Provenance: From patients exposed to heparin who showed HIT related symptoms, collected at three (3) hospitals.
- Nature: Retrospective (analysis of previously collected data against clinical guidelines).

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications of Those Experts

For studies comparing to a predicate device (Asserachrom HPIA Test Kit), the predicate itself served as the comparator, and its established performance and cut-offs formed the "ground truth" for comparison. The predicate device's ground truth would have been established during its own clearance.
For studies comparing to the Serotonin Release Assay (SRA), the SRA results were considered the "reference device" and "ground truth." The SRA is a functional assay for HIT antibodies and is often considered the gold standard, although the document doesn't specify how the SRA results themselves were "ground-truthed" (e.g., if multiple SRA runs were done, or expert interpretation). No specific number of experts or their qualifications for interpreting SRA results is mentioned.
For the Cut-Off Determination study, SRA results (31 SRA positive and 32 SRA negative) from hospitalized patients were used as the ground truth. Again, no specific information on experts for SRA.
For the ASH Guidelines comparison, the 2013 ASH diagnostic algorithm, which incorporates clinical probability (4T score), ELISA results, and SRA results, served as the "ground truth" or reference framework. This is a clinical guideline, not directly "expert-established ground truth" for individual cases in this context, but rather an established diagnostic framework.
For other analytical studies (precision, reproducibility, linearity, etc.), the ground truth is based on the inherent properties of the samples and controls used, not dependent on expert interpretation.

4. Adjudication Method for the Test Set

The document does not specify an adjudication method for establishing ground truth for the clinical samples. For studies comparing to the SRA or the predicate, it appears the results of these reference methods were taken directly without a separate adjudication process by additional experts.
For the ASH guideline comparison, the algorithm itself incorporates different types of evidence.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study Was Done, If So, What Was the Effect Size of How Much Human Readers Improve with AI vs Without AI Assistance

No, an MRMC comparative effectiveness study was not done.
This device, HemosIL HIT-Ab(PF4-H), is an in-vitro diagnostic (IVD) immunoassay, which performs automated detection of antibodies. It is not an AI-assisted diagnostic imaging or interpretation device that would involve human "readers" or benefit from "AI assistance" in the way described for an MRMC study. Its output is a quantitative (U/mL) and qualitative (positive/negative) result based on an assay cut-off.

6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) Was Done

Yes, the primary performance studies are standalone algorithm-only performance. The HemosIL HIT-Ab(PF4-H) assay is a "fully automated, latex enhanced immunoassay" performed on the ACL TOP Family of instruments. The results generated by the instrument are quantitative (U/mL) and then interpreted as positive or negative based on a fixed cut-off (1.0 U/mL). All the precision, reproducibility, linearity, and method comparison studies described are evaluating the performance of this automated assay directly, without human interpretation of raw data beyond the standard laboratory operation of running the assay and reporting the final result.

7. The Type of Ground Truth Used

Predicate Device Performance: The primary ground truth for demonstrating substantial equivalence was the performance of the Asserachrom HPIA Test Kit, a legally marketed predicate ELISA device.
Serotonin Release Assay (SRA): For clinical validity, the SRA, a functional assay for HIT antibodies, was used as a reference device and
considered the functional ground truth.
Clinical Guidelines: The 2013 American Society of Hematology (ASH) guidelines (incorporating 4T score, ELISA, and SRA) were used as a comprehensive clinical framework for comparison.
Internal Controls and Materials: For analytical performance studies (precision, linearity), the ground truth was established by the known concentrations/compositions of manufacturer-prepared controls, calibrators, and spiked plasma pools.

8. The Sample Size for the Training Set

The document does not explicitly delineate separate "training" and "test" sets in the context of machine learning or AI development, as this is an IVD immunoassay.
Instead, development and optimization of the assay (analogous to training in ML) would involve internal R&D studies. The summary primarily focuses on the validation studies (analogous to testing).
The "Cut-Off Determination" study describes using 63 samples (31 SRA positive, 32 SRA negative) and performing ROC analysis to determine the clinical cut-off of 1.0 U/mL. While this is not a "training set" in the AI sense, these samples were used to establish a critical parameter for the device's interpretation.

9. How the Ground Truth for the Training Set Was Established

As noted above, there isn't a "training set" in the AI sense.
For the "Cut-Off Determination" study, which used samples to establish the 1.0 U/mL cut-off:
- Ground Truth was established using Serotonin Release Assay (SRA) results. These SRA results classified the 63 samples as positive or negative for HIT antibodies.
- These samples were from "hospitalized patients who had been exposed to heparin, and who displayed clinical signs consistent with Heparin Induced Thrombocytopenia (HIT)," and were "tested by the hospital with the Serotonin Release Assay (SRA)." This implies the SRA results were clinical results from a standard reference method.

Summary

Regulation Number and Section

§ 864.7695 Platelet factor 4 radioimmunoassay.

(a)
Identification. A platelet factor 4 radioimmunoassay is a device used to measure the level of platelet factor 4, a protein released during platelet activation by radioimmunoassay. This device measures platelet activiation, which may indicate a coagulation disorder, such as myocardial infarction or coronary artery disease.(b)
Classification. Class II (performance standards).