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

In vitro diagnostic test for the quantitative determination of the HDL-cholesterol concentration in human serum and plasma on Roche/Hitachi cobas c systems.

A lipoprotein test system is a device intended to measure lipoprotein in serum and plasma. Lipoprotein measurements are used in the diagnosis and treatment of lipid disorders (such as diabetes mellitus), atherosclerosis, and various liver and renal diseases.

Device Description

The HDL-Cholesterol Gen.4 is a homogeneous enzymatic colorimetric test. Non-HDL lipoproteins such as LDL, VLDL and chylomicrons are combined with polyanions and a detergent forming a water-soluble complex. In this complex the enzymatic reaction of CHER and CHOD towards non-HDL lipoproteins is blocked. Finally only HDL-particles can react with CHER and CHOD. The concentration of HDL-cholesterol is determined enzymatically by CHER and CHOD.

AI/ML Overview

Here's a breakdown of the acceptance criteria and study details for the HDL-Cholesterol Gen.4 device, extracted from the provided document:

Device: HDL-Cholesterol Gen.4
Type: In vitro diagnostic test for the quantitative determination of HDL-cholesterol concentration in human serum and plasma on Roche/Hitachi cobas c systems.

Acceptance Criteria and Reported Device Performance

Acceptance Criteria Category	Specific Criteria (Implicitly met by study design and results)	Reported Device Performance
Precision	Low %CV for repeatability and intermediate precision across various HDL-C concentrations.	Repeatability:

PreciControl ClinChem Multi 1 (Mean 28.0 mg/dL): SD 0.20 mg/dL, CV 0.7%
PreciControl ClinChem Multi 2 (Mean 68.1 mg/dL): SD 0.44 mg/dL, CV 0.6%
Human Serum 1 (Mean 9.48 mg/dL): SD 0.17 mg/dL, CV 1.8%
Human Serum 2 (Mean 40.5 mg/dL): SD 0.26 mg/dL, CV 0.7%
Human Serum 3 (Mean 59.4 mg/dL): SD 0.32 mg/dL, CV 0.5%
Human Serum 4 (Mean 79.4 mg/dL): SD 0.51 mg/dL, CV 0.6%
Human Serum 5 (Mean 141 mg/dL): SD 0.83 mg/dL, CV 0.6%
Intermediate Precision:
PreciControl ClinChem Multi 1 (Mean 28.4 mg/dL): SD 0.30 mg/dL, CV 1.1%
PreciControl ClinChem Multi 2 (Mean 66.4 mg/dL): SD 0.9 mg/dL, CV 1.4%
Human Serum 1 (Mean 9.48 mg/dL): SD 0.20 mg/dL, CV 2.2%
Human Serum 2 (Mean 40.7 mg/dL): SD 0.33 mg/dL, CV 0.8%
Human Serum 3 (Mean 59.4 mg/dL): SD 0.40 mg/dL, CV 0.7%
Human Serum 4 (Mean 79.4 mg/dL): SD 0.65 mg/dL, CV 0.8%
Human Serum 5 (Mean 141 mg/dL): SD 1.07 mg/dL, CV 0.8% |
| Analytical Sensitivity (LoB, LoD, LoQ) | LoB, LoD, and LoQ should be below the claimed measuring range. | LoB Observed: 0.00 mg/dL (Claim: 3.09 mg/dL)
LoD Observed: 0.50 mg/dL (Claim: 3.09 mg/dL)
LoQ Observed: 2.89 mg/dL (Claim: 3.09 mg/dL) |
| Linearity/Assay Reportable Range | Measurements should be linear across the claimed measuring range (3.09 to 150 mg/dL), with good correlation and low deviation. | Serum: Slope 1.020, Intercept -0.399, Correlation Coefficient (r2) 0.9992, Repeatability 1.5%
Plasma: Slope 1.022, Intercept -0.173, Correlation Coefficient (r2) 0.9929, Repeatability 0.8%
Claimed Measuring Range: 3.09 to 150 mg/dL (for both serum and plasma).
Nonlinearity did not deviate by more than 10%. |
| Endogenous Interferences | No significant interference (bias >10%) from common interferents like hemolysis, lipemia, icterus, and triglycerides at specified levels. | Hemolysis: No Interference up to 1200 H index
Lipemia: No Interference up to 2000 L index
Unconjugated Bilirubin: No Interference up to 60 I index
Conjugated Bilirubin: No Interference up to 60 I index
Triglycerides: No Interference up to 1200 mg/dL |
| Exogenous Interferences (Drugs) | No significant interference from a panel of common drugs at specified concentrations. | Results provided in Table 7 (specific details of "no interference" for each drug are implied by listing them in a "Test Concentrations Results" table within the interference section, assuming they met the 10% bias criterion for non-interference). |
| Method Comparison to Predicate | Strong correlation with a legally marketed predicate device, with acceptable bias at medical decision points. | Regression Analysis (HDL-Cholesterol Gen.4 vs. Predicate): y = 0.956x - 0.949, r = 0.995
Bias at medical decision points:
-6.7 % at 40.2 mg/dL
-6.0 % at 59.9 mg/dL |
| Matrix Comparison | Acceptable correlation between serum and various plasma anticoagulant types. | Serum vs. Serum Gel Separation: y = 0.99x - 0.33, r = 0.999
Serum vs. Li-heparin: y = 0.99x - 0.32, r = 1.000
Serum vs. K2-EDTA: y = 0.98x - 0.70, r = 0.999
Serum vs. K3-EDTA: y = 0.95x - 0.08, r = 0.999 |

Study Details

Sample sizes used for the test set and the data provenance:
- Precision (Repeatability & Intermediate Precision): 5 human serum pools and 2 control samples were used. Tested for 21 days with 4 replicates/day. Data provenance is not explicitly stated as country of origin, but implied to be from a laboratory setting (likely in the US or Germany, given Roche's locations). The nature of the study (analyzing human serum pools and controls) suggests prospective data collection for evaluating device performance.
- Analytical Sensitivity (LoB, LoD, LoQ):
  - LoB: One analyte-free sample. Measured 10-fold in 6 runs over 3 days, per reagent lot (total 60 measurements per lot).
  - LoD: Five samples with low-analyte concentration. Measured two-fold in 6 runs over 3 days, per reagent lot (total 60 measurements per lot).
  - LoQ: A low-level sample set prepared by diluting 5 human serum samples. Tested in 5 replicates per sample on 5 days.
- Linearity/Assay Reportable Range: Dilution series prepared using 1 serum pool and 1 plasma pool. The dilution series contained 11 concentrations for serum and 15 concentrations for plasma.
- Endogenous Interferences: Two human serum pools spiked with HDL-Cholesterol and interfering substances.
- Exogenous Interferences (Drugs): Two human serum sample pools.
- Method Comparison to Predicate: 111 routine laboratory serum samples. Additionally, 4 samples spiked with high human serum HDL-Cholesterol and 1 sample diluted with 0.9% NaCl. Data provenance is not explicitly stated as country of origin, but implied to be from a laboratory setting. The use of "routine laboratory serum samples" suggests retrospective collection, though the spiking and dilution aspects are prospective for the study design.
- Matrix Comparison: 38 paired samples (serum and plasma) from single donors.
Number of experts used to establish the ground truth for the test set and the qualifications of those experts:
- For this type of in vitro diagnostic device (quantitative measurement of a biomarker), "ground truth" is typically established by reference methods or validated comparative methods, not by human expert consensus or clinical adjudication as would be seen in imaging diagnostics.
- The predicate device (Ultra N-geneous HDL Cholesterol Reagent, K021316) serves as a comparative "ground truth" for the method comparison study. The precision, sensitivity, linearity, and interference studies establish the intrinsic performance properties of the device against predefined analytical standards (e.g., CLSI guidelines).
- Therefore, the concept of "experts" as in "a radiologist with 10 years of experience" is not directly applicable here. The experts involved would be laboratory scientists, biochemists, and statisticians who designed and executed the studies according to CLSI guidelines and interpreted the analytical data.
Adjudication method (e.g. 2+1, 3+1, none) for the test set:
- Adjudication methods like 2+1 or 3+1 are typically used in clinical imaging studies where subjective interpretation is involved.
- For this in vitro diagnostic device, measurements are quantitative, and "adjudication" is done through statistical analysis and adherence to predefined acceptance criteria based on established analytical guidelines (e.g., CLSI EP5-A3 for precision, CLSI EP17-A2 for detection limits, CLSI EP6-A for linearity). There is no "human adjudication" process for individual results as there would be for image interpretation.
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 type of study is relevant for imaging devices where human readers interpret medical images, and AI might assist or replace them. The HDL-Cholesterol Gen.4 is an in vitro diagnostic device that quantifies a substance in a laboratory sample; it does not involve human "readers" interpreting results in a subjective or visual manner that AI would enhance.
If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:
- Yes, the performance evaluation reports are inherently demonstrating the "standalone" performance of the analytical algorithm/reagent system. The studies evaluate the device's ability to accurately and precisely measure HDL-cholesterol concentrations in samples, independent of further human interpretation beyond routine laboratory operation and quality control. The reported results (e.g., mean, SD, CV, regression equations) directly reflect the algorithm's performance.
The type of ground truth used (expert consensus, pathology, outcomes data, etc.):
- Analytical Ground Truth: For the precision, sensitivity, linearity, and interference studies, the "ground truth" is based on the known concentrations of analytes in prepared samples (e.g., spiked samples, diluted samples, control materials) or the absence of analytes (blank samples), and performance is evaluated against established analytical standards and acceptable deviations.
- Comparative Ground Truth: For the method comparison study, the predicate device (Ultra N-geneous HDL Cholesterol Reagent, K021316) served as the comparative "ground truth" or reference method for evaluating substantial equivalence. This is a common approach for 510(k) clearances.
- No pathology or outcomes data was used for establishing the ground truth for device performance in this submission, as it's an analytical performance study for an IVD.
The sample size for the training set:
- This document describes the pre-market notification (510(k)) studies for device validation, not the development or training of an AI algorithm. Therefore, there is no "training set" for an AI mentioned or implied in this submission. The device is a chemical reagent system, not an AI/ML independent medical device.
How the ground truth for the training set was established:
- As noted above, no "training set" for an AI algorithm is described in this submission. The ground truth for the analytical studies described (e.g., precision, linearity) is based on the preparation of samples with known concentrations or comparative analysis against a validated predicate device, as per standard laboratory practice and CLSI guidelines.

Summary

Regulation Number and Section

§ 862.1475 Lipoprotein test system.

(a)
Identification. A lipoprotein test system is a device intended to measure lipoprotein in serum and plasma. Lipoprotein measurements are used in the diagnosis and treatment of lipid disorders (such as diabetes mellitus), atherosclerosis, and various liver and renal diseases.(b)
Classification. Class I (general controls). The device is exempt from the premarket notification procedures in subpart E of part 807 of this chapter subject to § 862.9.