(118 days)
The T2Candida Panel and T2Dx® Instrument is a qualitative T2 Magnetic Resonance (T2MR®) assay for the direct detection of Candida species in EDTA human whole blood specimens from patients with symptoms of, or medical conditions predisposing the patient to, invasive fungal infections. The T2Candida Panel identifies five species of Candida and categorizes them into the following three species groups:
- Candida albicans and/or Candida tropicalis,
- Candida parapsilosis
- Candida glabrata and/or Candida krusei
The T2Candida Panel does not distinguish between C. albicans and C. tropicalis. The T2Candida Panel does not distinguish between C. glabrata and C. krusei.
The T2Candida Panel is indicated for the presumptive diagnosis of candidemia. The T2Candida Panel is performed independent of blood culture. Concomitant blood cultures are necessary to recover organisms for susceptibility testing or further identification.
The T2Candida positive and negative External Controls are intended to be used as quality control samples with the T2Candida Panel when run on the T2Dx® instrument system. These controls are not intended for use with other assays or systems.
The T2Candida panel and T2Dx® Instrument is comprised of the T2Candida Panel performed on the T2Dx® Instrument. The T2Candida Panel is a qualitative molecular diagnostic assay that employs a whole blood compatible PCR amplification followed by T2 magnetic resonance (T2MR) detection. The T2Candida Panel is performed on the T2Dx Instrument which executes all steps after specimen loading. A K2 EDTA whole blood specimen is loaded onto the T2Candida Sample Inlet, which is then placed on the T2Candida Base along with the T2Candida Reagent Pack. The Reagent Pack contains the internal control, amplification reagent, enzyme and the probe-coupled superparamagnetic particles for each Candida target. Three milliliters of the blood specimen is transferred to the T2Dx® Instrument where lysis of the red blood cells, concentration and lysis of the Candida cells and amplification of the Candida DNA takes place. Amplification products are detected by T2MR detection using species-specific probes which are attached to the superparamagnetic particles. At the end of each assay, the T2Dx® Instrument uses a bleach solution to neutralize all liquids on the cartridge to mitigate the risk of amplicon contamination. The assay provides an identification of Candida albicans and/or Candida tropicalis, Candida parapsilosis, and Candida glabrata and/or Candida krusei. The test does not distinguish between C. albicans and C. tropicalis. The test does not distinguish between C. glabrata and C. krusei.
Acceptance Criteria and Device Performance Study for T2Candida Panel and T2Dx® Instrument
1. Table of Acceptance Criteria and Reported Device Performance
The provided document does not explicitly present a table of acceptance criteria in numerical values for clinical performance metrics (like minimum desired PPA/NPA). Instead, it describes various studies and their results, implying that these results were deemed acceptable for market clearance. For the purpose of this response, I will interpret "acceptance criteria" as the performance metrics that were reported and considered satisfactory by the FDA for the device to be classified.
| Performance Metric | Acceptance Criteria (Implied) | Reported Device Performance |
|---|---|---|
| Analytical Performance | ||
| Precision/Reproducibility | High agreement with expected results across sites, lots, and operators. | C. parapsilosis: 100% agreement (Low & High LoD) |
| C. glabrata: 97.2% (Low LoD), 98.1% (High LoD) | ||
| C. albicans: 95.4% (Low LoD), 99.1% (High LoD) | ||
| Negative: 100% agreement | ||
| Invalid Results: 1.6% (13/830) | ||
| Instrument Failures: 2.1% (17/830) | ||
| False Positives: 0.4% (3/830) | ||
| Limit of Detection (LoD) | 95% detection rate at stated LoD. | C. albicans: 2 CFU/mL |
| C. tropicalis: 1 CFU/mL | ||
| C. parapsilosis: 3 CFU/mL | ||
| C. glabrata: 2 CFU/mL | ||
| C. krusei: 1 CFU/mL | ||
| Invalid results: 2.7% (23/845) | ||
| False positives: 0.6% (5/845) | ||
| Single/Multi-species Spike Equiv. | Multi-spike results equivalent to single-spike results. | Results from multi-species spiked samples were equivalent to single-species spiked samples. Invalid Results: 1.8% (5/281) |
| Instrument failure: 1.4% (4/281) | ||
| False positive: 0.7% (2/281) | ||
| Analytical Sensitivity | High percentage of strains detected at 2-3X LoD. | C. albicans: 100% (15/15) |
| C. tropicalis: 93.3% (14/15) - remaining strain detected on retest. | ||
| C. krusei: 100% (15/15) | ||
| C. glabrata: 100% (15/15) | ||
| C. parapsilosis: 100% (15/15) | ||
| False positives: 0.9% (2/225) | ||
| Co-infection Studies | High detection rates for target Candida species in presence of other organisms. | Candida sp./Candida sp. (both 1-2X LoD): 95.2% (118/124) |
| Candida sp. (1-2X LoD)/Candida sp. (100 CFU/mL): 96.8% (244/252) | ||
| Candida sp. (1-2X LoD)/other genus (100 CFU/mL): 94.5% (189/200) | ||
| Invalid Results: 1.5% (24/1597) | ||
| False Positives: 0.3% (5/1597) | ||
| Analytical Specificity (Cross-reactivity) | No cross-reactivity with non-target organisms at clinically relevant concentrations. | 41 species showed no cross-reactivity. 30 species causing invalid results at high concentrations but no cross-reactivity at clinical concentrations. 5 species causing positive/invalid results at high concentrations but no cross-reactivity at clinical concentrations. Cross-reactive species: C. bracarensis, S. cerevisiae, C. metapsilosis, C. orthopsilosis. |
| Interfering Substances | No significant interference in T2 signal. | Interference observed with: Feraheme, Magnevist, EDTA, Ablavar, and Intralipid (simulating lipemia). |
| Invalid Results: 0.7% (12/1647) | ||
| False Positives: 0.3% (5/1647) | ||
| Cross-over/Contamination | Low false positive rate due to carryover. | 1.7% false positive rate (3/175 negative samples), all from samples spiked at 100 CFU/mL. |
| Clinical Performance | ||
| Clinical Sensitivity (Contrived) | High positive percent agreement (PPA), especially at concentrations ≥ LoD. | Overall PPA: A/T: 94.0%, P: 94.0%, K/G: 88.0% |
| PPA at ≥ LoD: A/T: 97.5% (C. albicans), 100% (C. tropicalis); P: 100% (C. parapsilosis); K/G: 97.4% (C. krusei), 100% (C. glabrata). | ||
| Overall NPA: 100% (negative samples) | ||
| False positive: 0.4% (1/250 spiked samples) | ||
| Clinical Specificity (Prospective) | High negative percent agreement (NPA). | NPA: A/T: 98.8%, P: 99.2%, K/G: 99.9% (Based on blood culture negative for Candida, from 1501 prospective samples). |
| Clinical Sensitivity: A/T: 50.0% (2/4), P: 100% (2/2), K/G: 100% (1/1) for blood culture positive samples. | ||
| The document notes a low sensitivity for C. albicans in the prospective study due to small sample size (4 positive samples). |
2. Sample sizes used for the test set and data provenance
-
Test Set (Clinical Study - Contrived Samples):
- Sample Size: 250 positive contrived samples (50 isolates for each of the five Candida targets, spiked at specific concentrations) + 50 negative unspiked blood specimens = 300 total samples.
- Data Provenance: The human blood specimens were collected from patients at three clinical sites in K2EDTA tubes. These specimens were then transported to T2 Biosystems for preparation of contrived samples. The prepared samples were de-identified and shipped to testing sites. The data is thus a mix of prospective (blood collection) and contrived (spiking) for the positive samples, and prospective for the negative samples. The isolates used for spiking were obtained from a reference laboratory and a clinical laboratory (US, implicitly).
-
Test Set (Clinical Study - Prospective Samples):
- Sample Size: 1501 blood specimens from adult patients.
- Data Provenance: Prospective. Specimens were collected at nine geographically diverse sites.
3. Number of experts used to establish the ground truth for the test set and their qualifications
The document does not explicitly state the number or qualifications of experts used to establish the ground truth for the test set.
- For the Contrived Samples: The "ground truth" (presence and concentration of Candida species) was established through controlled laboratory spiking (colony count testing) and sequencing of isolates. This is an analytical determination, not typically requiring expert consensus on diagnosis.
- For the Prospective Samples: The "ground truth" for candida presence/absence was primarily established by blood culture results, which is a standard laboratory reference method. The identification of isolates was confirmed by sequence analysis of the ITS2 region. While a clinical microbiologist or pathologist would interpret these results, the document does not specify a panel of experts for adjudication.
4. Adjudication method for the test set
- For the Contrived Samples: No formal adjudication method involving multiple human readers is described for the contrived samples. The ground truth was based on the known spiking composition and verified colony counts. Discrepancies (e.g., initial LoD not validated) led to retesting.
- For the Prospective Samples: The "gold standard" was blood culture. For false positive T2Candida results (where blood culture was negative), a chart and case history review was conducted. This review process itself can be seen as a form of clinical adjudication, attempting to find clinical evidence supporting or refuting the T2Candida result when the primary reference method (blood culture) disagreed. However, the document does not specify an "adjudication method" in the sense of a consensus panel (e.g., 2+1, 3+1).
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. A multi-reader multi-case (MRMC) comparative effectiveness study involving human readers or AI assistance was not conducted or described. The T2Candida Panel and T2Dx® Instrument is a molecular diagnostic assay that provides a direct result for Candida species; it is not an image-based AI device that would typically involve human readers interpreting AI output.
6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done
Yes. The studies described for the T2Candida Panel and T2Dx® Instrument represent standalone (algorithm only) performance. The device fully automates the processing and analysis steps after manual specimen loading, and "the T2Dx® Instrument reports a positive or negative result for each detection channel." The performance metrics (sensitivity, specificity, PPA, NPA, precision, LoD, etc.) are all indicative of the device's inherent capability to detect Candida species from whole blood without real-time human interpretation or modification of its direct output.
7. The type of ground truth used
- For Analytical Studies (Precision, LoD, Analytical Sensitivity, Co-infection, Analytical Specificity): The ground truth was established by controlled laboratory preparation of samples with known Candida species and concentrations, verified by colony count testing and sequence analysis of isolates. Negative controls were known to be free of Candida.
- For Clinical Studies (Contrived Samples): The ground truth was established by controlled laboratory spiking of blood specimens with specific, verified Candida isolates at known concentrations. Negative samples were unspiked blood.
- For Clinical Studies (Prospective Samples): The ground truth for the presence or absence of Candida infection was primarily based on blood culture results, which is a widely accepted laboratory reference standard for candidemia. For "false positives" by T2Candida where blood culture was negative, chart and case history review was used to provide further context.
8. The sample size for the training set
The document does not explicitly specify a "training set" in the context of machine learning model development. For molecular diagnostic assays like the T2Candida Panel, the development process typically involves extensive analytical studies to establish assay design, primer/probe specificity, cutoff values, and performance characteristics, rather than training a machine learning algorithm on a distinct dataset.
However, the assay cut-off values (Table 12) for the T2MR signal were established using a "limit of blank study" with negative blood specimens from 113 healthy and 39 unhealthy patients (total 152 samples), tested with two different reagent lots. This serves a similar purpose to training data for setting a threshold.
9. How the ground truth for the training set was established
As noted above, there isn't a "training set" in the traditional machine learning sense. However, the ground truth for establishing the assay cut-off (analogous to a threshold used in training) was:
- Negative blood specimens: Confirmed to be negative for Candida (implied by sourcing from healthy/unhealthy patients and used to determine the upper limit of T2 signal distribution for negative measurements).
- Positive internal control measurements: Used to determine the lower limit of T2 signal for positive internal controls. The internal control itself is a known DNA sequence amplified and detected within each sample.
§ 866.3960 Nucleic acid-based device for the amplification, detection, and identification of microbial pathogens directly from whole blood specimens.
(a)
Identification. A nucleic acid-based device for the amplification, detection, and identification of microbial pathogens directly from whole blood specimens is a qualitative in vitro device intended for the amplification, detection, and identification of microbial-associated nucleic acid sequences from patients with suspected bloodstream infections. This device is intended to aid in the diagnosis of bloodstream infection when used in conjunction with clinical signs and symptoms and other laboratory findings.(b)
Classification. Class II (special controls). The special controls for this device are:(1) Premarket notification submissions must include detailed device description documentation, including the device components, ancillary reagents required but not provided, and a detailed explanation of the methodology, including primer/probe sequence, design, and rationale for sequence selection.
(2) Premarket notification submissions must include detailed documentation from the following analytical and clinical performance studies: Analytical sensitivity (limit of detection), reactivity, inclusivity, precision, reproducibility, interference, cross reactivity, carryover, and cross contamination.
(3) Premarket notification submissions must include detailed documentation from a clinical study. The study, performed on a study population consistent with the intended use population, must compare the device performance to results obtained from well-accepted reference methods.
(4) Premarket notification submissions must include detailed documentation for device software, including, but not limited to, software applications and hardware-based devices that incorporate software.
(5) The device labeling must include limitations regarding the need for culture confirmation of negative specimens, as appropriate.
(6) A detailed explanation of the interpretation of results and acceptance criteria must be included in the device's 21 CFR 809.10(b)(9) compliant labeling.
(7) Premarket notification submissions must include details on an end user device training program that will be offered while marketing the device, as appropriate.
(8) As part of the risk management activities performed as part of your 21 CFR 820.30 design controls, you must document an appropriate end user device training program that will be offered as part of your efforts to mitigate the risk of failure to correctly operate the instrument.