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

The Albumin BCP2 assay is an automated clinical chemistry assay. The Albumin BCP2 procedure is based on the binding of bromocresol purple specifically with human albumin to produce a colored complex. The absorbance of the complex at 604 nm is directly proportional to the albumin concentration in the sample.

Methodology: Colorimetric (Bromocresol Purple)

AI/ML Overview

This document is a 510(k) premarket notification for a new in vitro diagnostic device, the Albumin BCP2 assay, which measures albumin in human serum or plasma. It seeks to prove substantial equivalence to a predicate device, Albumin BCP.

Here's an analysis of the acceptance criteria and study that proves the device meets them, based on the provided text:

1. Table of Acceptance Criteria and Reported Device Performance:

The document doesn't explicitly present a formal "acceptance criteria" table with pass/fail thresholds for each performance characteristic. Instead, it describes various studies conducted and reports the results, implying that meeting standard analytical performance metrics demonstrates acceptable performance and substantial equivalence.

However, we can infer the acceptance criteria from the reported performance and the context of typical FDA 510(k) submissions for in vitro diagnostic assays. The studies are designed to demonstrate the new device performs comparably to established standards and the predicate device.

Performance Characteristic	Inferred Acceptance Criteria (General)	Reported Device Performance (Albumin BCP2)
Reportable Interval (Range)	Clinically relevant and accurate measurement across a broad range commensurate with predicate/clinical needs.	Analytical Measuring Interval: 0.3 – 9.0 g/dL. Extended Measuring Interval: 9.0 – 22.4 g/dL. Reportable Interval: 0.3 – 22.4 g/dL.
Within-Laboratory Precision	Low Coefficients of Variation (CV%) and Standard Deviations (SD) across different levels, indicating consistent and reproducible results. Specific numeric thresholds would be defined internally by Abbott based on regulatory/clinical expectations (e.g., %CV 0.95 or 0.98) and acceptable agreement (slope near 1.0, intercept near 0) with the predicate device, demonstrating substantial equivalence.	Correlation Coefficient = 1.00; Intercept = -0.20; Slope = 1.00; Concentration Range = 0.6 - 9.6 g/dL (Serum, n=127).
Tube Type Suitability	Acceptable performance across various specified blood collection tube types.	Deemed acceptable for use with Serum tubes, Serum separator tubes, Dipotassium EDTA tubes, Lithium heparin tubes, Lithium heparin separator tubes, and Sodium heparin tubes.
Dilution Verification	Acceptable agreement between automated and manual dilution methods (e.g., % difference within a specified tolerance like ±5-10%).	% difference values for automated vs. manual dilution ranged from -2.9% to -1.5%, demonstrating acceptable performance.

2. Sample Size Used for the Test Set and Data Provenance:

The document describes test methods rather than specific "test sets" in the context of an AI/ML algorithm that might have a dedicated validation dataset. Instead, for an in-vitro diagnostic device, studies are conducted across various analytical performance characteristics.

Precision Study: "2 controls and 3 human serum panels were tested in duplicate, twice per day on 20 days on 3 reagent lot/calibrator lot/instrument combinations." This means 80 data points for each control/panel (2 tests/day * 20 days * 2 replicates).
Accuracy Study: "2 lots of the Albumin BCP2 reagent, 2 lots of the Consolidated Chemistry Calibrator, and 1 instrument." (No specific sample count for patient samples, but implied to be sufficient for bias estimation against a reference material).
Lower Limits of Measurement: "n ≥ 60 replicates of zero-analyte samples" for LoB, and "n ≥ 60 replicates of low-analyte level samples" for LoD and LoQ.
Linearity: No specific sample size mentioned, but typically involves preparing multiple dilutions.
Interference Study: "Each substance was tested at 2 levels of the analyte (approximately 3.5 g/dL and 5.0 g/dL)."
Method Comparison: 127 serum samples.
Tube Type: "Samples were collected from a minimum of 40 donors".
Dilution Verification: 5 human serum samples prepared by spiking.

Data Provenance: The document doesn't explicitly state the country of origin for the human samples used in the studies. Given that Abbott Ireland Diagnostics Division submitted the application, the studies were likely conducted in a setting compliant with international standards, possibly in Ireland or the US given the FDA submission. The studies described are nonclinical laboratory studies, not human clinical trials. They are retrospective or prospective in the sense of laboratory-controlled experiments designed to evaluate performance characteristics.

3. Number of Experts Used to Establish Ground Truth and Qualifications:

This section is not applicable as this is an in-vitro diagnostic (IVD) device, specifically a clinical chemistry assay, not an AI/ML diagnostic software. The "ground truth" for an IVD device is established through:

Reference Methods: Using highly accurate and precise laboratory methods (e.g., mass spectrometry, enzymatic methods, or other established validated assays) or certified reference materials (like ERM-DA470k/IFCC for accuracy testing).
Standardization: Traceability to international standards (like IFCC).
Analytical Performance: Rigorous testing against defined analytical parameters (precision, linearity, limits of detection/quantitation).

There is no "ground truth" derived from human expert consensus for this type of device.

4. Adjudication Method for the Test Set:

This is not applicable for the same reasons as #3. Adjudication methods (like 2+1, 3+1) are common in AI/ML studies where human readers are establishing ground truth for image interpretation or similar tasks. For an IVD assay, performance is judged against analytical accuracy and precision, not human consensus on results.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:

This is not applicable. MRMC studies are specific to AI/ML software that assists human readers (e.g., radiologists, pathologists) in interpreting medical images or data. This device is a quantitative laboratory assay. There is no human "reader" assisted by this device in the same way. The device directly measures a biochemical analyte.

6. Standalone (Algorithm Only Without Human-in-the-Loop Performance):

This isn't an "algorithm" in the AI/ML sense but rather a chemical assay method run on an automated analyzer. The entire performance data presented (precision, accuracy, linearity, interference, method comparison) represents the "standalone" analytical performance of the assay without human intervention influencing the measurement itself. Human involvement is in the operation of the instrument, quality control, and interpretation of the results, but not in the measurement process being evaluated here.

7. Type of Ground Truth Used:

The ground truth for this chemical assay is primarily established by:

Standard Reference Materials (SRMs): For accuracy, the device's results are compared against ERM-DA470k/IFCC, which are certified reference materials with highly accurate assigned values.
Reference Measurement Procedures: Implied by the use of standard methods and CLSI (Clinical and Laboratory Standards Institute) guidance, which dictates how analytical performance (e.g., LoB, LoD, LoQ, linearity, precision) should be determined using robust statistical methods and replicates.
Predicate Device Comparison: For method comparison, the "ground truth" (or comparative truth) is the performance of the legally marketed predicate device (Albumin BCP) on patient samples.

8. Sample Size for the Training Set:

This concept of a "training set" is specific to AI/ML models. For an IVD assay, calibration and internal method development (which could be analogous to "training") would involve various reagent lots, calibrators, and QC materials provided by the manufacturer. The document does not specify a "training set size" in the AI/ML context because the development of a chemical assay follows different principles.

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

Again, this is not applicable in the AI/ML sense. For chemical assays, the establishment of the assay (analogous to "training") involves:

Reagent Formulation and Optimization: Developing the chemical reagents (Bromocresol Purple, buffers, etc.) to ensure proper reaction kinetics, stability, and specificity.
Calibrator Assignment: Assigning accurate values to calibrator materials used by the assay, often traceable to international standards (like ERM-DA470/IFCC).
Method Development on Platform: Optimizing the assay parameters (volumes, incubation times, temperatures, wavelength) on the specific ARCHITECT c System to achieve optimal performance.
Internal Validation: Initial testing during development to ensure the assay performs as expected before formal verification and validation studies are conducted for regulatory submission. This internal validation would use similar principles of analytical testing as described in Section 8 (e.g., accuracy, precision, linearity using reference materials and pooled human samples).

Summary

Regulation Number and Section

§ 862.1035 Albumin test system.

(a)
Identification. An albumin test system is a device intended to measure the albumin concentration in serum and plasma. Albumin measurements are used in the diagnosis and treatment of numerous diseases involving primarily the liver or kidneys.(b)
Classification. Class II.