Two-by-two cross-over study to evaluate agreement between versions of a quantitative coronary analysis system (QAngio XA)

Abstract

The current version (ver. 7.3) of the popular quantitative coronary analysis system QAngio XA (Medis Medical Imaging System BV, Leiden, the Netherlands) is widely used without evaluating the agreement between the current and older versions in relation to a change of algorithms. The purpose of this study was to assess the equivalence of averages between QAngio XA versions 7.3 and 6.0. Based on the calculated sample size, angiographic images of 100 patients who underwent percutaneous coronary intervention of a single target lesion were randomly selected from two published studies (OUCH-TL: 154 lesions; OUCH-PRO: 160 lesions). The primary endpoint was the minimum lumen diameter (MLD), and the secondary endpoints were the reference diameter (RefD) and length of the stenotic lesion (LL). Two independent analysts measured the same frame using both previous and current versions of QAngio XA. Version-order for each lesion was randomly determined per coronary locations targeted. Data were analysed by using a mixed model that includes random lesion effects and fixed rater effects and reading-order effects. A Bland–Altman plot of parameters showed no large differences between the versions. Differences in parameters were estimated by the mixed model, and the 95% confidence interval of the MLD, RefD, and LL estimates was from −0.045 to −0.0001 mm, from −0.040 to 0.006 mm, and from −1.08 to 0.46 mm, respectively, compared with the predefined non-inferiority margin of ±0.2 mm. Measurements of MLD and RefD using QAngio XA showed no major systematic differences between versions.

Appendix: sample size calculation

We determined our sample size by using a random-intercept analysis-of-variance model to ascertain that the difference between the QAngio XA versions was negligible. The assumed model was

$${{Z}_{ij}}=\mu +{{\alpha }_{i}}+{{\beta }_{j}}+{{\varepsilon }_{ij}},$$

$${{\alpha }_{i}}\tilde{\ }N\left( 0,\sigma _{\alpha }^{2} \right)~~i.i.d.,{{\varepsilon }_{ij}}\tilde{\ }N\left( 0,\sigma _{\varepsilon }^{2} \right)~~i.i.d.,$$

where ${{Z}_{ij}}={{Y}_{ij2}}-{{Y}_{ij1}}$ is the difference between measurements of lesion $i=1,\,2,\,\ldots \,,\,n$ read by two analysts $~j=1,\,2$ with version 6.0 (${{Y}_{ij1}}$) and version 7.3 (${{Y}_{ij2}}$), respectively. We considered lesion effects ${{\alpha }_{i}}$ as random effects and rater effects ${{\beta }_{j}}$ as fixed effects for the measurement difference. In the model, the contrast $~\mu +\frac{1}{2}\left( {{\beta }_{1}}+{{\beta }_{2}} \right)~$ represents the mean difference of the measured values between versions 6.0 and 7.3. We considered the two versions of QAngio XA to be equivalent if the 95% confidence interval (CI) of the contrast lay within a ±0.2 mm margin.

To calculate the sample size required to maintain a power of 0.9 in order to detect equivalence, we took the following steps [27]: (1) fitting the aforementioned model to the measured MLD values as ${{Y}_{ij,\text{ver}}}$ from the data obtained in the previous Vampire study [13], which enrolled patients receiving PCI in real-world clinical practice; (2) using $\hat{\sigma }_{\alpha }^{2}$ and $\hat{\sigma }_{\varepsilon }^{2}$ (fitted values) to calculate the sample size under $\mu =0$ and ${{\beta }_{j}}=0$ as though there were no fixed effects; (3) calculating the final sample size by inflating the number in step 2 by (1 + ICC) [27], where ICC = $\hat{\sigma }_{\alpha }^{2}/\hat{\sigma }_{\alpha }^{2}+\hat{\sigma }_{\varepsilon }^{2}$ is an estimated intraclass correlation coefficient of ${{Z}_{ij}}$. Step 1 gave the estimates ICC = 0.29, and SAS/GLMPOWER requires 56 values of $\text{ }\!\!~\!\!\text{ }{{Z}_{ij}}$ in step 2. The resulting size was 73; that is, n = 37 patients. To ensure that the smallest subgroup (left anterior descending coronary artery lesions) provided meaningful results, at least 93 patients in total would be required in the study. Here, 100 patients were randomly sampled, presupposing more stringent conditions than those of the previous study.