Test methods for MRI safety of AIMD systems has been defined through ISO/TS 10974¹. Specific to the assessments of RF lead heating and RF unintended stimulation, the ISO/TS 10974 defines methods for validating transfer functions (TFs) to within known uncertainty bounds. Since there is uncertainty with any of the RF measurements, duplication of a measurement or measurement set is unlikely.

AIMD systems are typical comprised of an electrical wire (lead) that terminates in one side in biologic tissue, and the other into a device with capacitive coupling to electronics. The leads of these devices should be expected to have the same RF characteristics (heating and voltage) when attached to variants of the devices (pacemaker, ICD or CRT-D) due to similar device input impedance and its corresponding linear scaling of RF voltage. A rigorous method to demonstrate equivalence of two systems in the presence of the uncertainty is yet to be defined. In absence of a rigorous criteria for determining equivalence, a complete retest of the new systems may be requested by regulators.

The Concordance Correlation Coefficient² (CCC) statistically determines data reproducibility. CCC measures how far observations deviate from the 45^o line between two sets of measurements. It consists of a measurement of precisions through the Pearson correlation coefficient (ρ)multiplied by a measure of accuracy (C_b)

The measurement of precision, here the Pearson correlation coefficient , evaluates how far the observations deviate from the best-fit linear line. The measure of accuracy (C_b), evaluates how far the best-fit line deviates from the concordance line. C_b consists of the scale shift (the ratio of two standard deviations) and location shift (the squared difference in means relative to the product of the standard deviations). The equations for these are shown in figure 1.

Acceptance criteria values to discriminate equivalence from non-equivalence for CCC, ρ , C_b were generated for over uncertainty values that ranged from 1 to 50%. These values are simulated across 2000 examples. CCC_acceptance, ρ_acceptance , C_{b, acceptance} was chosen that captured the 95% results of these 2000 simulated examples of CCC, ρ, C_b.

The flowchart in figure 1 is used to determine equivalence.

We evaluated if this method is an appropriate by evaluating 3 different paired datasets, and following the chart. The datatsets are one pair known to be equivalent, one pair known to be equivalent with scaling, and one pair known to be different.

Equivalence Pair - To demonstrate that the test can show equivalence two versions of SJM Quartet™ leads (1456Q and 1458Q) that differ only in electrode placement along the lead of the 4 electrodes (D1, M2, M3, P4) were tested as the equivalent pair. Uncertainty was 19%. As the CCC_comparison >CCC_acceptance, this pair confirm the method can appropriately determine equivalence. Figure 2 shows the test criteria from the 2000 examples, and the results for this dataset

Equivalence with Scaling - To demonstrate the test can show scaling is required to show system equivalence, a single SJM Quartet™ lead was tested with a Quadra Allure™ CRT-P device and a Quadra Assura™ CRT-D device with a significant feedthrough capacitor difference. Uncertainty was 19%. Figure 3 contains a graph of the paired data from this test. Figure 4 shows the test criteria and results.

As ρ_comparison<ρ_acceptance but C_b,comparison > C_{b, acceptance}, Equivalence with scaling was determined. After scaling took place, CCC_comparison > C_acceptance.

Not Equivalent - To test a pair that does not correlate, an ICD lead was inserted into two different generations of ICD. The dataset from the same pathways were tested on the RV ring electrode and ICD, with the voltages from SVC electrode on ICD. As figure 5 shows, the data does is within uncertainty bounds but does not statistically correlate. The CCC and scaling criteria are not met, although most points are within the measurement uncertainty cone.

The CCC method has been previously described as a robust discriminator of whether a new assay/test can reproduce the result of a ‘gold standard’. We have extended the method by pre-defining an acceptance criteria based on the uncertainty in the baseline dataset. The method establishes a rigorous criteria for determining equivalence, and should eliminate the need for a complete retest for system variations. Moreover, the method showed better discrimination than comparison of points within the uncertatiny cone. The ability of this method to assess TF validation by testing the model predictions against the validation measurements will be investigated next.