Statistics

All groups were tested before statistical analysis for normal distribution using Kolmogorov Smirnoff test. If data was normally distributed, a one-way ANOVA was performed; otherwise, the Kruskal-Wallis test was used. Statistical analyses were performed using SPSS (Version 29.0. Armonk, NY: IBM Corp., USA).

Choroidal metastases originating from 14 subgroups of primary tumors were analyzed. Overall survival was calculated from date of first diagnosis of choroidal metastasis until death. Survival distributions were generated using the Kaplan-Meier method and group differences in survival were assessed with the log rank test. Patients who were alive at the time of the last follow-up or were lost to follow-up were right-censored at the date of their last clinical examination. Comparisons were made by primary tumor type and age. For Kaplan-Meier and Cox regression analyses, primary tumors were categorized into three groups to ensure adequate sample sizes and reliable survival estimates:

Group 1 included all patients with primary lung cancer (metastatic non-small-cell and small-cell carcinoma). Group 2 included primary breast cancer (all metastatic mammary carcinomas). Group 3 included all other primary tumors gastrointestinal (colorectal and pancreatic), genitourinary (renal, prostate, bladder and gynecologic; each being metastatic), and head and neck/endocrine/melanoma primaries.

Cox Regression Analysis

To identify independent predictive factors associated with overall survival, a multivariate Cox proportional hazards regression model was performed. Continuous variables included age at the time of choroidal metastasis diagnosis, tumor volume and height, while categorical variables included number of other non-ocular metastasis elsewhere in the body (≤1 vs. > 1), tumor classification (lung, breast, or others), presence of subretinal fluid/retinal detachment (yes/no), systemic therapy prior to metastasis (yes/no), and order of diagnosis (eye first vs. primary tumor first). Univariate Cox regressions were conducted first to screen for potential predictors. In univariate Cox regression analyses, age at metastasis was analyzed as a continuous variable to assess the general effect of increasing age on survival. For multivariate modeling, age was dichotomized at the sample median (59.6 years) to facilitate clinical interpretation and maintain balanced group sizes. Variables with a significance level of p < 0.15 in univariate analyses or with strong clinical relevance were entered into the multivariate Cox model. Sex was excluded from the multivariate analysis due to its complete overlap with tumor classification (all breast cancer patients were female), which would otherwise introduce collinearity. The number of non-ocular metastases elsewhere in the body was tested in univariate analysis. Despite the small subgroup size (n = 13, eyes with 0 or 1 non-ocular metastasis elsewhere in the body), the variable was retained in the multivariate model due to its biological relevance as a marker of disease burden. Consequently, results should be interpreted with caution. Results were expressed as hazard ratios (HR) with 95% confidence intervals (CI), and a p-value less than 0.05 was considered statistically significant.

Kaplan Meier Survival depending on Primary Tumor Types

At the time of analysis, 48 of 70 patients (68.6%) had deceased. Event occurrence of death by primary site was 19/28 (67.9%) in lung cancer, 14/23 (60.9%) in breast cancer, and 15/19 (78.9%) in other tumors. The mean survival time was 115.3 weeks for lung cancer, 205.7 weeks for breast cancer, and 112.3 weeks for other primary tumors. Median survival was 52.7, 94.4, and 40.6 weeks. For the whole study group, the mean survival time was 158.2 weeks, and the median was 71.6 weeks. The mean observation period from the diagnosis of choroidal metastasis until death or last follow-up was mean 92.9 weeks (SD 120.18 weeks).

There was no statistically significant difference in overall survival between the three groups (p = 0.269). The estimated 1-year survival rate was 53.6% in lung cancer, 76.2% in breast cancer and 42.1% in the other primary tumor group. At 3-years, the estimated survival rate was approximately 9.5% for the lung cancer group, 38.3% for the breast cancer group, and 19.7% for the other tumor group. At 5-years, the estimated survival rate was approximately 9.5% for the lung cancer group, 25.5% for the breast cancer group, and 19.7% for the other tumor group. The estimated overall 1-year, 3-years and 5-years survival rates for the whole study cohort were approximately 56%, 24% and 18%. The Kaplan Meier survival curve can be seen in Figure 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Kaplan Meier survival curves according to primary tumor group: 1 (blue) lung cancer, 2 (red) breast cancer, 3 (green) other tumors.

https://doi.org/10.1371/journal.pone.0354598.g001

Kaplan Meier Survival in Age

When dichotomizing age at the group median (59.6 years), 32 patients were classified as younger and 38 as older than the median. At the time of analysis, 18/32 younger patients (56.3%) and 30/38 older patients (78.9%) had experienced the event (death). The mean survival time was 209.5 weeks (median 88.3 weeks) for patients younger than 59.6 years, and 118.7 weeks (median 41.9 weeks) for those aged 59.6 years or older. The estimated 1-year, 3-years and 5-years survival was approximately 70.5%, 33.2% and 26.5% in patients younger than 59.6 years and 46.2%, 16.8% and 12.6% in patients aged ≥59.6 years.

According to the Log-Rank test, the difference in survival between the two age groups was statistically significant (p = 0.037), indicating a longer survival in the younger group. The Kaplan Meier survival curve can be seen in Figure 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan Meier survival curves according to age group (median split at 59.6 years), blue: younger (<59.6years) red: older (≥59.6 years).

https://doi.org/10.1371/journal.pone.0354598.g002

Univariate Cox regression analysis

Univariate Cox regression analyses were conducted to identify variables associated with overall survival. Variables were selected based on clinical relevance and prior evidence from the literature. A trend toward reduced survival was observed with increasing age at metastasis (HR = 1.025; 95% CI 0.999–1.051; p = 0.056).

Patients with more than one non-ocular metastasis elsewhere in the body had an approximately 1.9-fold higher risk of death compared to those with a single or no metastasis elsewhere in the body (HR = 1.87; 95% CI 0.83–4.18; p = 0.129).

Both variables were therefore included in the multivariate model. Tumor classification (lung, breast, or others) was not significantly associated with survival (p = 0.276) but was included due to biological plausibility.

Sex was not a significant predictor (HR = 1.53; 95% CI 0.86–2.74; p = 0.148) and was excluded from the multivariate model due to collinearity with tumor type (all breast cancer patients were female). Other clinical and ocular factors including systemic therapy before metastasis, subretinal fluid or serous retinal detachment, tumor volume, tumor height, and timing of primary cancer diagnosis (choroidal metastasis vs. primary tumor) showed no significant association with survival and were not included for further analysis.

Multivariate Cox Regression Model

A multivariate Cox proportional hazards model was performed including primary tumor group (lung, breast, other), age at metastasis (2 groups: median split 59.6 years) and number of other non-ocular metastasis elsewhere in the body (≤1 vs. > 1).

The multivariate Cox proportional hazards regression model was statistically significant (p = 0.013), indicating that the included variables collectively improved the prediction of survival. Among the individual predictors, both age group (median split at 59.6 years) and number of non-ocular metastases elsewhere in the body were statistically significant. Patients aged ≥59.6 years had a 2.1-fold higher hazard of death compared with those younger than 59.6 years (HR = 2.10; 95% CI: 1.15–3.82; p = 0.016). Likewise, patients with more than one metastasis elsewhere in the body had a statistically significantly increased risk for mortality (HR = 2.53; 95% CI: 1.10–5.83; p = 0.029). The primary tumor classification (lung, breast, others) did only show a non-significant trend but did not reach statistical significance (p = 0.088).

Effect of age on patient survival

In our findings, age at choroidal metastasis diagnosis was lower in patients diagnosed with breast cancer by a mean of 8.77 years compared to lung cancer (p = 0.024). These results are consistent with previous reports by Shields et. al [6,22].

Our observation that younger age is associated with prolonged survival is strongly supported by d’Abbadie et al., who identified age ≥ 50 years as a significant predictor of treatment failure and linked mortality risk directly to metastatic involvement in the liver or pleura/lungs [23]. Consistent with this, our data showed that patients with more than one metastasis elsewhere in the body had a statistically significantly increased risk for mortality. Interestingly in our data, we could show that younger adults (<59.6) had significantly longer survival rates which is in contrast to Shields et al [22]. Multiple factors may explain this discrepancy. First, Shields et al. utilized clinically established and predefined age categories, which are ideal for characterizing such an extensive population. Whereas we applied a data-driven cutoff, specifically optimized for our cohort. We observed survival trends that may reflect the specific characteristics of our patient population, complementing the broader findings reported in larger series. Second, their study spans 1974–2017, when systemic cancer therapies evolved considerably; younger patients in our more recent cohort likely benefited from modern treatments. Furthermore, younger patients typically present with fewer comorbidities, better functional status, and less extensive metastatic disease, which allows more intensive systemic therapeutic strategies [22–24]. Third, Shields et al [22]. reported substantial differences in primary tumor distributions across age groups (e.g., higher rates of lung cancer in older adults), but did not adjust for this in multivariate models. In our analysis, older age remained an independent predictor of mortality after adjustment for metastatic burden and primary tumor.

Together, these methodological and clinical differences may account for the contrasting age-related survival patterns between the two studies.

Effect of primary tumor type on patient survival

In the literature, breast cancer (37–47%) and lung cancer (21–50%) are reported as the most common primary sites [6,16–19]. Our findings align with previously published data, confirming that the majority of choroidal metastasis arises from breast (32.9%) and lung (40%) malignancies.

Survival after diagnosis of choroidal metastasis is poor and varies depending on the primary tumor type. In the literature the overall survival rates of patients with different primary tumors differ between publications. Our observed 3-year survival rates (9.5% in the lung cancer group, 38.3% in the breast cancer group, and 19.7% in the “other tumors” group) support the trend that lung cancer metastases are associated with a lower survival compared to breast cancer metastasis which is supported by the literature [5,19,25–27]. While univariate analysis in Kaplan Meier survival curve did not reach statistical significance for primary tumor types (Kaplan Meier p = 0.269), the multivariable Cox model (p = 0.013) – which adjusts for age, non-ocular metastasis elsewhere in the body, and primary tumor type – showed a non-statistically significant trend (p = 0.088). We interpret this cautiously, as our sample size and number of events within each primary tumor subgroup were limited, and the study is likely underpowered to detect modest differences in survival.

One important limitation of this study is the relatively small sample size of 70 patients. While the primary tumor origins were stratified into three main groups—breast, lung, and others—less common primary sites were combined into a heterogeneous ‘other’ category. This ‘other primary tumors’ category includes a wide spectrum of different tumor origins (12 different rare malignancies) that exhibit varying metastatic behaviors and survival patterns [19]. Due to the rarity of these entities, which represents a consecutive cohort from the largest tertiary referral center in Austria, separate statistical analysis for each rare tumor type was not feasible. While grouping them into a single category is a limitation regarding specific prognostic forecasting, omitting these patients would have introduced a severe selection bias and impaired the real-world representation of our data. Grouping patients as ‘Others’, as it has been published in this context previously by Shields et al [19,22] a feasible way to address rare real-world data in ocular oncology. Furthermore, the overall limited sample size and the subsequent formation of even smaller subgroups for categorical variables (e.g., n = 13 for patients with ≤1 non-ocular metastasis) mathematically resulted in wide confidence intervals in our multivariate analysis; therefore, these specific hazard ratios should be interpreted with appropriate clinical caution. However, it is worth noting that the analysis was conducted retrospectively over a period from 2013 to 2025 in one specialized center, which may lead to a more homogeneous and targeted patient selection. Another limitation was the retrospective character of this study, which might introduce potential selection and information bias. Moreover, exact volumetric ultrasound data could not be included in the analysis, as historical clinical documentation primarily focused on tumor height rather than full three-dimensional measurements.

Although the presence of multiple extra-ocular metastases was a significant predictor of mortality, their specific anatomical locations were not separately analyzed to prevent statistical fragmentation and loss of power. Furthermore, our study is limited by the lack of molecular and histological subtyping of the primary malignancies (hormone-receptor or HER2 status in breast cancer, and specific oncogenic mutations in lung cancer). Given the retrospective design and a long inclusion period beginning in 2013, alongside the fact that many patients were referred from external institutions across different federal states, centralized access to primary tumor pathology reports or tissue samples was unavailable. While we acknowledge that molecular subtypes are critical confounders for overall systemic survival, our analysis focuses on easily accessible clinical and ophthalmic parameters that can guide the ocular oncologist at the time of presentation.