AI-Based Analysis of Port Catheter Tip Position

Mean absolute error (MAE), in millimeters, will be calculated as the arithmetic mean of the absolute differences between the catheter advancement length estimated by the locked AI-based automated A+B+Delta method and the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The analysis population will include participants with both a valid automated output and an evaluable reference measurement. Lower values indicate smaller estimation error. A 95% confidence interval will be estimated using 10,000 participant-level bias-corrected and accelerated bootstrap resamples.

Median absolute error (MedAE), in millimeters, will be calculated as the median of the absolute differences between the catheter advancement length estimated by the locked AI-based automated A+B+Delta method and the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The analysis population will include participants with both a valid automated output and an evaluable reference measurement. Lower values indicate smaller estimation error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

Root mean squared error (RMSE), in millimeters, will be calculated as the square root of the arithmetic mean of the squared differences between the catheter advancement length estimated by the locked AI-based automated A+B+Delta method and the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The analysis population will include participants with both a valid automated output and an evaluable reference measurement. Lower values indicate smaller overall estimation error, and larger individual errors have greater influence because differences are squared before averaging. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

Mean signed error, in millimeters, will be calculated as the arithmetic mean of the catheter advancement length estimated by the locked AI-based automated A+B+Delta method minus the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The analysis population will include participants with both a valid automated output and an evaluable reference measurement. Positive values indicate that the automated method estimates a longer advancement length than the reference, negative values indicate a shorter estimate, and zero indicates no average directional error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

For each participant, absolute error will be calculated for both the locked AI-based automated A+B+Delta method and the manual A+B method relative to the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The paired difference, in millimeters, will be calculated as automated absolute error minus manual absolute error, and its arithmetic mean will be reported. The analysis population will include participants with an evaluable reference measurement and valid estimates from both methods. Negative values indicate a smaller absolute error for the automated method, positive values indicate a smaller absolute error for the manual method, and zero indicates no average paired difference. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

The proportion of evaluable participants whose absolute error is 10 mm or less will be calculated for the locked AI-based automated A+B+Delta method. Absolute error will be defined as the absolute difference between the automated estimate and the reference catheter advancement length derived from the first eligible post-implantation noncontrast chest computed tomography examination. The denominator will include participants with both a valid automated output and an evaluable reference measurement, and the result will be reported as a percentage with a 95% Wilson score confidence interval. A higher percentage indicates that a larger proportion of estimates meet this prespecified pragmatic accuracy criterion. The 10 mm threshold is an operational technical accuracy criterion and not a clinical safety boundary.

For each participant, the locked AI-based automated A+B+Delta estimate and the manual A+B estimate will each be classified by whether the absolute error from the reference catheter advancement length is 10 mm or less. The paired proportion difference, in percentage points, will be calculated as the automated qualifying proportion minus the manual qualifying proportion. The analysis population will include participants with an evaluable reference measurement and valid estimates from both methods. Positive values indicate a higher qualifying proportion for the automated method, and negative values indicate a higher qualifying proportion for the manual method. A 95% confidence interval will be calculated using Newcombe's method 10; an exact two-sided McNemar test will be used as an auxiliary analysis. The 10 mm threshold is a prespecified pragmatic accuracy criterion and not a clinical safety boundary.

Technical success will be defined as the generation of a valid numerical output by the AI-based automated A+B+Delta method after the method has been locked for prospective evaluation. The numerator will be the number of participants enrolled in the prospective cohort for whom the method generates a valid automated output, and the denominator will be all participants enrolled in the prospective cohort. A valid output requires the prespecified image-scale, anatomical-landmark, projection, image-quality, and uncertainty criteria to be satisfied. The result will be reported as a percentage with a 95% Wilson score confidence interval. Technical success will be reported separately from measurement accuracy among participants with evaluable reference data. A higher percentage indicates greater technical output availability and does not represent clinical effectiveness or safety.

Intraobserver agreement of the computed tomography (CT)-derived catheter tip-to-cavoatrial junction distance (dBT) will be assessed using a two-way random-effects, absolute-agreement, single-measure intraclass correlation coefficient [ICC(A,1)]. The same radiologist will repeat the dBT measurement in a prespecified random subset of 60 cases after a 4-week washout period, using newly assigned identifiers and reshuffled images while blinded to the initial measurement. Only cases evaluable at both readings will be analyzed. Higher ICC values indicate greater relative agreement. The estimate will be reported with a 95% confidence interval.

The mean signed difference (second reading minus first reading), in millimeters, will be calculated for the computed tomography-derived catheter tip-to-cavoatrial junction distance in the prespecified random subset of 60 cases measured twice by the same radiologist after a 4-week washout period. The radiologist will use newly assigned identifiers and reshuffled images and will be blinded to the initial measurement. Only cases evaluable at both readings will be analyzed. Positive values indicate a larger measurement at the second reading, negative values indicate a smaller measurement, and values closer to zero indicate less average systematic difference. A 95% confidence interval will be reported.

Within-subject standard deviation (wSD), in millimeters, will quantify within-observer variation in the computed tomography-derived catheter tip-to-cavoatrial junction distance. In the prespecified random subset of 60 cases, the same radiologist will repeat the measurement after a 4-week washout period using newly assigned identifiers and reshuffled images while blinded to the initial measurement. Only cases evaluable at both readings will be analyzed. For these cases, wSD will be calculated as the standard deviation of the paired differences (second reading minus first reading) divided by the square root of 2. Lower values indicate less within-observer measurement variation. A chi-square-based 95% confidence interval will be reported.

Bland-Altman limits of agreement (LoA), in millimeters, will be calculated for the CT-derived catheter tip-to-cavoatrial junction distance in a prespecified random subset of 60 cases measured twice by the same radiologist after a 4-week washout period. Paired differences will be defined as the second reading minus the first. The mean paired difference and lower and upper LoA, calculated as the mean difference minus and plus 1.96 times the standard deviation of paired differences, will be reported. Only cases evaluable at both readings will be analyzed. The radiologist will use newly assigned identifiers and reshuffled images and will be blinded to the initial measurement. A 95% confidence interval for the mean difference will be calculated using the t distribution; 95% confidence intervals for the lower and upper LoA will be obtained using an exact two-sided tolerance-factor method. Narrower limits indicate less within-observer measurement variation.

The repeatability coefficient (RC), in millimeters, will quantify within-observer repeatability of the computed tomography-derived catheter tip-to-cavoatrial junction distance in the prespecified random subset of 60 cases measured twice by the same radiologist after a 4-week washout period. Paired differences will be defined as the second reading minus the first reading. The RC will be calculated as 1.96 times the standard deviation of the paired differences. Only cases evaluable at both readings will be analyzed. The radiologist will use newly assigned identifiers and reshuffled images and will be blinded to the initial measurement. Lower RC values indicate less within-observer measurement variation. A chi-square-based 95% confidence interval will be reported.

For participants with a valid automated output and an evaluable CT-derived reference measurement, the paired difference, in millimeters, will be defined as the catheter advancement length estimated by the locked AI-based automated A+B+Delta method minus the CT-derived reference catheter advancement length. The mean difference and the lower and upper 95% limits of agreement will be calculated as the mean difference minus and plus 1.96 times the standard deviation of the paired differences. A 95% confidence interval for the mean difference will be calculated using the t distribution; 95% confidence intervals for the lower and upper limits will be obtained using an exact two-sided tolerance-factor method. Positive differences indicate longer automated estimates than the reference, and negative differences indicate shorter automated estimates. The analysis describes agreement and does not establish clinical interchangeability or safety.

Mean absolute error (MAE), in millimeters, will be calculated as the arithmetic mean of the absolute differences between the catheter advancement length estimated by the prespecified manual A+B measurement method and the CT-derived reference catheter advancement length. The analysis population will include participants with both a valid manual A+B estimate and an evaluable CT-derived reference measurement. Lower values indicate smaller estimation error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

Median absolute error (MedAE), in millimeters, will be calculated as the median of the absolute differences between the catheter advancement length estimated by the prespecified manual A+B measurement method and the CT-derived reference catheter advancement length. The analysis population will include participants with both a valid manual A+B estimate and an evaluable CT-derived reference measurement. Lower values indicate smaller estimation error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

Root mean squared error (RMSE), in millimeters, will be calculated as the square root of the arithmetic mean of the squared differences between the catheter advancement length estimated by the prespecified manual A+B measurement method and the CT-derived reference catheter advancement length. The analysis population will include participants with both a valid manual A+B estimate and an evaluable CT-derived reference measurement. Lower values indicate smaller overall estimation error, and larger individual errors have greater influence because differences are squared before averaging. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

Mean signed error, in millimeters, will be calculated as the arithmetic mean of the catheter advancement length estimated by the prespecified manual A+B measurement method minus the CT-derived reference catheter advancement length. The analysis population will include participants with both a valid manual A+B estimate and an evaluable CT-derived reference measurement. Positive values indicate that the manual method estimates a longer advancement length than the reference, negative values indicate a shorter estimate, and zero indicates no average directional error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

For participants with a valid manual A+B estimate and an evaluable CT-derived reference measurement, the paired difference, in millimeters, will be defined as the catheter advancement length estimated by the prespecified manual A+B method minus the CT-derived reference catheter advancement length. The mean difference and the lower and upper 95% limits of agreement will be calculated as the mean difference minus and plus 1.96 times the standard deviation of the paired differences. A 95% confidence interval for the mean difference will be calculated using the t distribution; 95% confidence intervals for the lower and upper limits will be obtained using an exact two-sided tolerance-factor method. Positive differences indicate longer manual estimates than the reference, and negative differences indicate shorter manual estimates. The analysis describes agreement and does not establish clinical interchangeability or safety.

In the prespecified sensitivity subset of participants whose first eligible post-implantation noncontrast chest computed tomography examination is obtained within 24 hours after port implantation and before any documented catheter manipulation, mean absolute error (MAE), in millimeters, will be calculated as the arithmetic mean of the absolute differences between the catheter advancement length estimated by the locked AI-based automated A+B+Delta method and the CT-derived reference catheter advancement length. Only participants with a valid automated output and an evaluable CT-derived reference measurement will be included. Lower values indicate smaller estimation error. A 95% confidence interval will be estimated using participant-level bootstrap resampling.

In the prespecified sensitivity subset of participants whose first eligible post-implantation noncontrast chest computed tomography examination is obtained within 24 hours after port implantation and before any documented catheter manipulation, the proportion of evaluable participants whose absolute error is 10 mm or less will be calculated for the locked AI-based automated A+B+Delta method. Absolute error will be defined as the absolute difference between the automated estimate and the CT-derived reference catheter advancement length. The denominator will include participants with both a valid automated output and an evaluable CT-derived reference measurement. The result will be reported as a percentage with a 95% Wilson score confidence interval. The 10 mm threshold is a prespecified pragmatic technical accuracy criterion and not a clinical safety boundary.

In this prespecified exploratory analysis, the association between the interval from port implantation to the first eligible post-implantation noncontrast chest computed tomography examination and the signed error of the locked AI-based automated A+B+Delta method will be assessed among participants with a valid automated output and an evaluable CT-derived reference measurement. Signed error, in millimeters, will be defined as the automated catheter advancement length estimate minus the CT-derived reference length. The interval, recorded in hours, will be transformed as log2(1 + hours). A robust Huber regression slope with a two-sided 95% confidence interval will be reported; Spearman's rank correlation coefficient with a two-sided 95% confidence interval will be an auxiliary analysis. Raw and Holm-adjusted Spearman p-values will be reported. The slope is expressed in millimeters per log2(1 + hours); positive values indicate increasingly positive signed error with longer intervals.

In this prespecified exploratory analysis, the association between the interval from port implantation to the first eligible post-implantation noncontrast chest computed tomography examination and the absolute error of the locked AI-based automated A+B+Delta method will be assessed among participants with a valid automated output and an evaluable CT-derived reference measurement. Absolute error, in millimeters, will be defined as the absolute difference between the automated catheter advancement length estimate and the CT-derived reference length. The interval, recorded in hours, will be transformed as log2(1 + hours). A robust Huber regression slope with a two-sided 95% confidence interval will be reported; Spearman's rank correlation coefficient with a two-sided 95% confidence interval will be an auxiliary analysis. Raw and Holm-adjusted p-values will be reported for the Spearman test. Slope is in millimeters per log2(1 + hours); positive values indicate increasing absolute error.