In accordance with the Health Insurance Portability and Accountability Act (HIPAA), our UM master database, which includes 1807 patients diagnosed with UM from 1984 to 2022 treated with LDR-EPBT, was utilized for analysis. This study was conducted with approval from the Institutional Review Board (IRB) of the University of Tennessee Health Science Center (ID: 16-0 4640). All tenets of the Declaration of Helsinki were followed in conduction of the study. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. Strict inclusion criteria encompassed only patients with tumors arising within the choroid or ciliary body. Iris melanomas were excluded. Only patients with documented B-scan ultrasound measurements at both the time of diagnosis and at one-year follow-up were included. Follow-up interval was calculated from the date of plaque placement to the date of most recent visit or death. All patients with a documented visit less than 1 year from study cessation were presumed to be alive unless otherwise notified. Patient survival status was verified with the Social Security Death Index (SSDI). Patients not identified as deceased within the SSDI were presumed to be alive and were censored on the date of the most recent follow-up. Incidence of metastatic disease was confirmed through longitudinal review of the institutional medical record, including medical oncology and hospital records when available.

B-scan ultrasonography was used to determine tumor dimensions ( Figure 1 ). The maximal tumor apex height and bidimensional basal diameters, both longitudinally (LD) ( Figure 1 A ) and transversally (TD) ( Figure 1 B ), were recorded. Volume was approximated as an ellipsoid using the following formula:

Representative B-scan ultrasonography image of a choroidal melanoma illustrating tumor dimension measurements. The images show the maximal tumor apex height (vertical caliper) and the basal diameters (horizontal calipers) in both longitudinal (A) and transverse (B) sections centered around the 1 ^st clock hour. The overlayed plots on (A) and (B) represent the A-scan echogenic profile of each image. (C) Fundus photo of the uveal melanoma seen in the B-scans provided. (D) A diagram showcasing the dimensions used to estimate tumor volume by ellipsoid modeling.

Percentage change was calculated to assess the rate of tumor shrinkage over time before and after brachytherapy. Percentage change was selected over absolute volume change (mm³/year) to represent tumor shrinkage rate, as it more accurately reflected shrinkage by preserving the rate of change relative to initial tumor size. Percentage change was calculated with the following formula:

Tumors were stratified based on the COMS guidelines into small (apex ≤ 3.0 mm and LBD diameter ≤ 16.0 mm), medium (>3.0 mm and ≤ 8.0 mm, LBD ≤ 16.0 mm); large (apex ≥ 8.0 or LBD > 16.0 mm and apex ≥ 2.0 mm) tumors, and by cAJCC cancer stages cT1-cT4. ^,

All statistical analyses were done utilizing the lifelines, matplotlib, and pandas libraries of Python v3.11. To explore the prognostic value of tumor volume regression, all possible one-year tumor shrinkage thresholds were systematically evaluated using log-rank testing in 0.1% increments across the range of − 100% to 100%, with each threshold tested by dichotomizing the cohort into patients above vs below the threshold and comparing OS. Analysis was split into 2 arms, −100% to 50% and 50% to 100%, to determine both lower and upper thresholds. Guardrails were placed to ensure each group, above and below the candidate threshold, comprised at least 10% of the total cohort to ensure statistical validity. Log-rank p -values were only plotted once both arms of the comparison contained at least 10% of the cohort. The maximally selected log-rank statistic was identified to provide the greatest separation in overall survival (OS). In conjunction, locally weighted scatterplot smoothing (LOWESS) was applied to visualize the relationship between tumor regression and survival probability. The scatterplot was constructed by grouping one-year tumor regression values into 1% bins across the 0% to 100% range. 10-year mortality calculated within each bin and plotted at the bin midpoint. The smoothing span was set to 0.12 a prior after pilot sensitivity checks. Robustness was confirmed by evaluating spans in the range of 0.10 to 0.15 in 0.01 increments, which yielded concordant inflection points. Inflection points were identified to inform evaluation of potential thresholds. Statistical significance of thresholds was confirmed by log-rank testing. After regression groups were defined based on threshold criteria, Cox proportional-hazards modeling was used to estimate hazard ratios (HRs) and 95% CIs for pairwise comparisons among regression groups and relative to the overall cohort.

Tumor volume characteristics at baseline and one-year post-EPBT were summarized using descriptive statistics. Only 2-sided p -values >.05 were considered as statistically significant. Box-and-whisker plots were used to illustrate the median, inter-quartile range (IQR), and 1.5 × IQR limits of the data.

Kaplan-Meier (KM) survival analysis was used to cf OS between patients in each regression range: delayed, intermediate, and accelerated. Survival curves were generated for the full study cohort and stratified based on COMS size groups and AJCC cancer stages. Log-rank tests were used to assess significance between strata. Survival probabilities and corresponding 95% CIs were calculated at 1, 3, 5, and 10 years (Supplemental Table 1). Corresponding event tables were generated for each KM survival analysis plot to illustrate the trends of patients at-risk, censored or deceased stratified by clinical subgroups (Supplemental Table 2 ). These tables highlight the distribution of patients contributing to the KM survival analysis with at-risk representing living patients, censored patients representing those lost to follow up, withdrew, or did not experience death by the end of the study period, and deceased representing those who passed away.

A total of 1180 patients were selected based on inclusion criteria. For the entire cohort, the mean baseline tumor volume was 336.3 mm³ (SE: 11.0 mm³), which decreased to 129.5 mm³ (SE: 5.17 mm³) on average at 1 year following plaque brachytherapy. Median baseline tumor volume was 183.8 mm³ (IQR: 395.1 mm³), with a median decrease to 65.3 mm³ (IQR: 138.8 mm³) at 1 year. Most patients had measurable volume regression (94.4%, N = 1114) and, among those, the overall mean shrinkage was 58.2% (SE: 0.84%) and the median shrinkage was 56.5% (IQR: 42.7%). A total of 191 patients had 100% tumor regression in 1 year. Five-year Kaplan-Meier estimate of local control was 88.1% (95% CI, 85.4%- 90.3%) (Supplemental Figure 1A).

To evaluate the extent and variability of TVR following EPBT, we analyzed volumetric changes across clinical subgroups stratified by COMS size and AJCC T-stage ( Figure 2 ). Based on COMS criteria, the cohort included 330 small, 451 medium, and 398 large tumors. According to AJCC T-staging, there were 329 cT1, 357 cT2, 320 cT3, and 174 cT4 tumors. Among the COMS groups, small tumors demonstrated a mean shrinkage of 31.0% (SE: 10.24%), with a median of 45.1% and an IQR of 50.7%. Medium tumors shrank by 51.1% (SE: 1.45%), with a median of 50.6% and an IQR of 36.6%. Large tumors showed the greatest regression at 62.4% (SE: 1.59%), with a median of 66.6% and an IQR of 42.6% ( Figure 2 A and 2 C). Among tumors stratified by AJCC T-stage, cT1 tumors shrank by 30.9% (SE: 10.27%), with a median of 44.5% and an IQR of 51.7%. cT2 tumors shrank by 49.2% (SE: 1.71%), with a median of 48.3% and an IQR of 35.8%. cT3 tumors shrank by 59.0% (SE: 1.77%), with a median of 62.0% and an IQR of 36.5%. cT4 tumors demonstrated the highest degree of regression at 66.8% (SE: 2.07%), with a median of 69.1% and an IQR of 43.6% ( Figure 2 A and 2 D).

Tumor volume regression following episcleral plaque brachytherapy (EPBT) across clinical subgroups (A) Dot plot showing mean tumor volume shrinkage at one-year post-EPBT, stratified by COMS size (small, medium, large) and AJCC T-stage (cT1–cT4). Error bars represent 95% CIs. (B) Box-and-whisker plot comparing baseline and one-year post-treatment tumor volumes for the entire cohort, with (Pre) representing patients pre-EPBT and (Post) representing those post-EPBT. (C) Tumor volume before and after EPBT, stratified by COMS tumor size. (D) Tumor volume before and after EPBT, stratified by AJCC T-stage. Statistical comparisons were performed using paired tests; *** p <.001.

Box-and-whisker analysis confirmed significant tumor volume regression at one-year post-EBPT in the overall cohort ( Figure 2 B) . Examining patients stratified based on COMS size categories and AJCC T-stages, a trend emerged demonstrating a stepwise increase in percent shrinkage from small to large tumors ( Figure 2 C) and cT1 to cT4 tumors ( Figure 2 D ). Paired comparisons confirmed the statistical significance of these changes ( p <.001) across all stratified groups within the cohort, excluding cT1 vs cT2 tumors ( Figure 2 A ).

Maximal survival discrimination was determined via log-rank p -value sweep from − 100% to 50% to be a shrinkage percentage of 33.4%, at which point the p -value curve reached is nadir (log-rank p =.0204) ( Figure 3 A ). Upper range log-rank p -value sweep from 50% to 100% revealed 91.1% regression to have the maximum survival discrimination (log-rank p =.0003) ( Figure 3 B ). LOWESS analysis revealed notable inflection points in the relationship between 10-year mortality rate and 1-year tumor regression at 33.33% and 75% shrinkage ( Figure 3 C ). Informed by this analysis, the thresholds of 33.33% (log-rank p =.029) and 75% (log-rank p =.001) were selected a priori for clinical simplicity and prognostication as both represent exactly one-third and three-fourths reduction in tumor size respectively, while preserving statistical power across groups.

Only gold members can continue reading. Log In or Register to continue

Share this:

• X
• Facebook

Related

Related posts:

1. The Multimodal Imaging in Uveitis Virtual Special Issue
2. Unsuccessful Clinical Trials in Retina: Lessons Learned
3. Vitamin B Intake Is Associated With Lower Incidence of Open-Angle Glaucoma: The Rotterdam Study
4. Auto Machine Learning for Diabetic Retinopathy Screening: A Head-to-Head Multiplatform Comparison Against Human Graders and IDx-DR

Stay updated, free articles. Join our Telegram channel

Join