**Purpose.**:
To model regression rate of thickness and cross-sectional area of choroidal melanomas after ruthenium and iodine brachytherapy by shape and regression pattern (RP).

**Methods.**:
We enrolled 330 of 388 consecutive uveal melanomas from 2000 to 2008 and analyzed images from I^{3} System-ABD 10-MHz B-Scan at diagnosis and during a 3-year-long follow-up. We classified tumor shape by Collaborative Ocular Melanoma Study definitions and RP according to an earlier study. We plotted regression over time and compared thickness and cross-sectional area.

**Results.**:
The observed RP by thickness was classified as decrease (D) in 43%, stable (S) in 5%, increase (I) in 1%, and other in 40% of eyes; main subpatterns were decrease-stable (DS) in 16% and zigzag in 10% of eyes. The corresponding percentages by area were 42%, 3%, 1%, 45%, 16%, and 14%. Regression pattern was discordant in 34% of eyes for thickness versus area. Area reduced faster than thickness when shape was oval to dome or mushroom and if RP was D. Pooled patterns D, DS, decrease-increase (DI), and zigzag corresponded with progressive but increasingly less pronounced regression for 3 years (56% by thickness and 69% by area), 2 years (50% and 52%), 6 months (29% and 33%), and 6 months (19% and 23%), respectively. First-order exponential decay function fitted thickness and area regression for every shape and for patterns D, DS, DI, and zigzag.

**Conclusions.**:
Heterogeneity in RP, variation in shape, and tumor cross-sectional area as an alternative measure must be considered when tumor regression rate is used in outcome analysis.

^{1}and most medium-sized and small tumors are generally treated with radiation therapy. Radiotherapy initiates tumor regression that is routinely evaluated in clinical practice as change in tumor thickness measured by B-scan ultrasonography.

^{2}This is reasonable because the diameter of the tumor base does not change.

^{3}If the treatment is successful, the tumor flattens to where it arose.

^{3–7}iodine,

^{7–11}and ruthenium,

^{10,12–23}plaques, teletherapy with charged particles,

^{24–27}gamma knife,

^{17,28}and linear accelerator,

^{29}sometimes in combination with transpupillary thermotherapy,

^{8,19,21–23}has been addressed in many studies over three decades, often with contradicting results. In particular, although several studies have suggested that rapid initial or overall regression would be associated with higher mortality,

^{6,9,15,20,24}not all have confirmed such a difference.

^{3,14,18}Moreover, some studies have recently reported that uveal melanomas with monosomy 3 or class 2 gene expression profile, which would be expected to lead to metastasis, did not regress faster than disomy 3 and class 1 tumors regressed,

^{10,11,26}whereas others have in fact found a difference.

^{8,9}

^{6,9,15,20,24}might be a spurious finding, reflecting different case mix with regard to thickness, cell type, and other tumor characteristics.

^{11}Most studies agree that large uveal melanomas regress faster than smaller ones,

^{5,7,9,16,24,27}and some have found that those with epithelioid cells

^{14}and high metabolic activity

^{23}may also regress faster. Additional factors likely contribute to the contradicting results, however. With one exception,

^{7}data from all treated tumors have been pooled for analysis, which results in a curve with progressively decreasing tumor thickness. However, some studies have revealed that the regression pattern of an individual uveal melanoma after irradiation in fact varies widely, from more or less rapid decrease through no change to even increase in thickness.

^{7,10,11,18}Secondly, uveal melanomas come in a number of shapes, such as flat, crescent, oval, dome, mushroom (collar button), lobulated, and irregular

^{30,31}; and when they grow and regress, their shape may change. For these reasons, it is not obvious that measurement of pooled tumor thickness would accurately reflect the regression relative to tumor characteristics such as its genetic profile or propensity to spread.

^{32}Indeed, the current version of the Response Evaluation Criteria In Solid Tumors (RECIST), which the majority of clinical trials in oncology use, has now adopted a one-dimensional measurement.

^{33}However, in a number of patients the response to treatment may still be classified differently based on volumetric as opposed to one- or two-dimensional imaging.

^{34}

^{30,35–40}varies more widely than that of a typical solid tumor in other organs, tumor thickness is not necessarily the best surrogate measure for tumor volume, and especially for a change in tumor volume. For example, the base of a mushroom-shaped melanoma seems to regress to a different extent than the part that has broken through Bruch's membrane. Although volume measurements of uveal melanoma have been obtained from time to time,

^{38,41–44}none of these methods has become routine. Formulas for estimating tumor volume from two-dimensional measurements of uveal melanoma have proved inconsistent and largely unsatisfactory.

^{35,38,40,42,44}

^{43}

^{3}System-ABD B-Scan (Innovative Imaging, Inc., Sacramento, CA, USA) and who were treated with primary brachytherapy between September 14, 2000, and June 30, 2008, at the Ocular Oncology Service, Helsinki University Central Hospital, Finland. We excluded patients with iris and primary ciliary body melanomas, which were measured using other equipment, and tumor eyes that had silicone oil tamponade at the time of diagnosis.

^{3}System-ABD scan was introduced in the Ocular Oncology Service in September 2000. One of two ocular oncologists measured the thickness of the tumor from the inner surface of the sclera to tumor apex and the largest basal diameter (LBD) using a gain of 60 dB. Thickness was measured from two meridians, along the LBD and perpendicular to it. Representative digitized scans were prospectively stored at the time of each diagnostic and follow-up visit.

^{30}It provided more detail than the one in the COMS Manual of Procedures

^{36,37}that illustrated dome, mushroom, peaked, and lobulated as the most common shapes, although the Study Forms additionally listed the shapes flat, irregular, and other. No textual guidance was provided as to how to apply this classification. In the ancillary histopathological study, crescent, oval, and oval with extension were introduced; peaked was dropped; and irregular included also tumors that could not otherwise be classified.

^{30}An illustration was provided as guidance for each shape. Using this system, intra- and intergrader agreement were 90% and 93%, respectively.

^{30}

*n*-sided polygon. We used the millimeter scale on the ultrasound scans to calibrate the software so as to obtain all results in square millimeters.

**Figure 1**

**Figure 1**

^{45}and very thick tumors a reduced apical dose.

^{1}In 158 single ruthenium plaque treatments, the median delivered dose to apex and base were 102 Gy (range, 62–194) and 361 Gy (range, 192–1212), respectively; and in 163 single plaque iodine treatments the median delivered apical dose was 80 Gy (range, 49–164). The corresponding median delivered dose rates to tumor apex were 112 cGy/h (range, 22–400) and 68 cGy/h (range, 14–350) for ruthenium and iodine, respectively. The corresponding median treatment durations were 95 hours (range, 28–407) and 119 hours (range, 27–467), respectively (in five instances, the treatment time exceeded 14 days).

*P*value < 0.05 was taken to indicate statistical significance.

*P*values.

^{46}to assess agreement between the prospective I

^{3}System-ABD B-Scan and the retrospective Olympus DP-10 software (ODS) thickness measurement.

^{47}We assessed the appropriateness of linear fit using the goodness-of-fit statistic and runs test, and by comparing the regression line visually with a corresponding locally weighted scatterplot smoothing (lowess) curve.

^{48}

^{7}into four main patterns: D (decrease; progressive decrease in thickness by at least 15% after brachytherapy), S (stable; less than 15% change in thickness), I (increase; progressive increase in thickness by at least 15%), and other. We further subdivided other into five subtypes: DS (D followed by S), DI (D followed by I), ID (I followed by D), SD (S followed by D), and zigzag (alternating measurements with little evidence of a trend). We excluded from this analysis 18 patients who had only two acceptable follow-up measurements and 19 patients who had been followed for less than 6 months.

^{20}defined as the mean decrease at ODS measurement 3 months after brachytherapy, and we defined initial cross-sectional area regression rate (IARR) accordingly.

^{16,20,24}Goodness-of-fit statistic and runs test were used to evaluate model fit.

^{49}91 (28%) of the 330 choroidal melanomas represented size category T1 or small; 107 (32%) were T2 or medium-sized; 95 (29%) were T3 or large; and 37 (11%) were T4 or very large (Supplementary Table S1). Eighty-nine (27%) were TNM stage I, 166 (50%) stage II, 70 (21%) stage III, and 5 (2%) stage IV (metastases at the time of diagnosis).

^{2}(range, 5–294) as measured with ODS.

^{2}increase in cross-sectional area (flat-to-crescent shape, 9.6 mm

^{2}).

*r*

^{2}statistic was 0.82 for the flat-to-crescent shape and ranged from 0.90 to 0.93 for the other three shape groups, indicating that linear regression on tumor thickness captured 82% to 93% of total variation in tumor cross-sectional area (Supplementary Fig. S1). Runs test indicated no significant deviation from linearity for any shape (

*P*> 0.05 for all; Supplementary Fig. S1). The lowess plot deviated from the regression line for flat-to-crescent shapes when tumor thickness exceeded 3.5 mm because of two outliers that could alternatively have been classified as dome, and for oval-to-dome shapes that showed more variation when thickness exceeded 10 mm (Supplementary Fig. S1).

*P*= 0.0014) because of lower initial slope due to flat-to-crescent tumors and presence of outliers among tumors of other shapes thicker than 10 mm.

**Table 1**

*P*= 0.0008 Kruskal-Wallis test; and 36%, 58%, and 56% analyzed by cross-sectional area;

*P*= 0.0009). Of the other two regression patterns observed in over 25 patients, DS did not show a significant trend (

*P*= 0.83 and

*P*= 0.13, respectively), whereas decreasing proportions showed a zigzag pattern (

*P*= 0.074 and

*P*= 0.0010, respectively).

**Table 2**

*P*= 0.0001 and

*P*= 0.0002 nonparametric test for trend, respectively). The DS pattern was generally the second most common one but did not show any frequency trend. The zigzag pattern was observed mainly in the flat-to-crescent and oval-to-dome shape groups, with a decreasing trend toward mushroom and lobulated ones (

*P*= 0.017 and

*P*= 0.0010, respectively).

*P*= 0.0001, Kruskal-Wallis test). According to the COMS shape group, ITRR (median, 1.4% vs. 4.3% vs. 8.1%) and IARR (2.7% vs. 5.9% vs. 9.9%) increased significantly from flat to crescent shaped to oval and dome shaped to mushroom tumors (both

*P*< 0.0001 nonparametric test for trend).

**Table 3**

*P*< 0.0001 Wilcoxon signed-ranks test; and 32% vs. 38% at 1 year,

*P*< 0.0001). This was true irrespective of the status of Bruch's membrane or shape group (Table 3). With the exception of indeterminate Bruch's membrane, the difference ranged from 4 to 9 percentage points, which amounted to a 16% to 24% difference. By 3 years, the difference ranged from −3 to 7 percentage points (−6% to 17%), and was no longer significant for tumors without broken Bruch's membrane or those that were flat to crescent or lobulated in shape. Thereafter, the data were biased by partial referral of patients to follow-up elsewhere (Fig. 2A). Analysis was thus based on the first 3 years.

**Figure 2**

**Figure 2**

*P*= 0.0001 Kruskal-Wallis test; and 55% and 61% at 3 years, both

*P*= 0.0001). Compared with the flat-to-crescent group (22% and 26% at 1 year, and 36% and 34% at 3 years), the oval-to-dome group (33% and 41% at 1 year,

*P*= 0.0022 and

*P*= 0.0011; and 42% vs. 49% at 3 years;

*P*= 0.14 and

*P*= 0.048) and the mushroom shape (41% and 50% at 1 year; both

*P*= 0.0001; and 67% and 74% at 3 years; both

*P*= 0.0001) had a faster rate of reduction (Figs. 2E–G). The regression of lobulated tumors as a group was inconsistent over time with increasing measures between 6 and 18 months without recurrences of individual tumors (Fig. 2H), but their rate of reduction was also faster than that of flat-to-crescent tumors (Table 3;

*P*= 0.0034 and

*P*= 0.0006 at 1 year; and

*P*= 0.0042 and

*P*= 0.0032 at 3 years).

*P*< 0.0001 Wilcoxon signed-ranks test), and was significantly smaller than ITRR for the S pattern (

*P*= 0.0061; Table 3).

*P*< 0.0001; and 56% vs. 69% at 3 years,

*P*= 0.0002). A significant difference was found at 1 year for SD and ID patterns (

*P*= 0.0002 vs.

*P*= 0.036, respectively) and a minor trend at 3 years for SD and DI patterns (Table 3), but no certain difference was observed for any pattern other than D.

**Figure 3**

**Figure 3**

*P*> 0.05 for all; Table 4).

**Table 4**

^{9,16,20,24,27}was applicable to most, but not all, regression patterns. These main findings should be taken into account when ocular oncologists plan to evaluate any biological differences among uveal melanomas based on their regression after radiotherapy. In particular, our results lend support to multivariate modeling based on exponential decay,

^{9}but also show that such modeling must take into account that not all tumors fit this pattern, and that tumor area may be an equally relevant endpoint.

^{24}or, more typically, a first-order exponential decay function.

^{9,16,20,24,27}We noted great variation among the remaining 57% of tumors, confirming an early report by Abramson et al.,

^{7}who found that no two uveal melanomas regressed exactly according to the same pattern after cobalt or iodine brachytherapy. Of their 82 tumors, 70% regressed progressively, 16% remained stable, 12% increased in size, and 2% showed other patterns. We classified 5% as stable and 1% as increasing in size. Conversely, we assigned 40% of melanomas to other patterns of regression. The disparate percentages reflect a finer analysis of patterns over time rather than underlying differences in the tumors treated or the isotopes used. In particular, we assigned many decreasing tumors to patterns DS and SD, and some increasing ones to patterns DI and ID. Published plots that display regression patterns of uveal melanoma case by case in Abramson et al.

^{7}and in two later studies

^{11,18}confirm the existence of these subpatterns and a much wider variation than can accurately be described using four categories.

^{20}found that the average half-life by first-order exponential decay of tumor thickness after ruthenium brachytherapy was 5.8 months with a plateau at 61%. In our series, the half-life was 5.1 months and the plateau was 59%, a very similar result. However, the half-life and plateau varied 4-fold by tumor shape, being approximately double the average for flat-to-crescent tumors and less than one-half for lobulated ones. The half-life of decay by cross-sectional area was 5.2 months, but the plateau was notably lower at 51%. Interestingly, half-lives of cross-sectional area by shape were much more similar and showed a less than 2-fold range, whereas the plateau varied even more widely than for tumor thickness.

^{16}and others

^{27}hypothesized that the plateau might correspond to a more radioresistant population of tumor cells. In that case, the four regression patterns with different plateaus may have biological significance.

^{16,27}In line with this hypothesis, Abramson et al.

^{7}noted an association between survival and regression pattern rather than regression rate. Importantly, the exponential decay function was not a reasonable fit for the S, SD, I, and ID patterns that together comprised 14% and 10% of uveal melanomas in our series based on thickness and cross-sectional area, respectively. These patterns may reflect other underlying biological differences.

^{6,8,10,11,23,24}For example, Kaiserman et al.

^{20}defined the ITRR as the average monthly regression in thickness during the first 3 months. The median was 3.9% per month for 147 eyes treated with ruthenium plaques as compared to 4.4% for our 330 eyes, a comparable result. However, the ITRR was four times as fast for oval-to-dome-shaped and more than six times as fast for mushroom-shaped and lobulated tumors than for flat-to-crescent ones, and showed even more variation by regression pattern. We observed wide variation in regression patterns after 3 months, indicating that later regression of uveal melanomas cannot be predicted from their ITRR or IARR. This may explain why some studies have found an association with survival based on some but not all time points.

^{6,24}

^{11}Typically, studies on regression of uveal melanomas have failed to report tumor shape

^{3,5–12,14–18,20,22–24,27}or have mentioned only the proportion of dome-shaped

^{29}(84% vs. 35% in our unselected series) or mushroom-shaped melanomas (16%, 22%, and 39% vs. 17% in our series).

^{21,26,29}Consequently, it is not possible to assess the extent of such bias. The same applies to the effect of regression patterns that have been reported only by Abramson et al.

^{7}

^{49,50}Ciliary body extension was present in 24% of eyes as compared to 25% in TNM data.

^{15,17,18}Such formulas often give disparate results because they do not take into account the variable shapes of these tumors.

^{51}

^{52,53}excluding tumors classified as small, shows reasonably similar percentages (Supplementary Table S2), although the proportion of medium-sized tumors was much smaller in the ancillary COMS study. We combined flat with crescent, oval with dome shapes, and lobulated with irregular shapes, which together with the mushroom shape made four main categories of reasonable size. We propose that this classification be adopted for future studies on tumor regression.

^{6,9,15,24}or absence

^{3,7,14,18}of an association with survival and the presence

^{8,9}or absence

^{10,11,26}of an association with unfavorable chromosomal and gene expression profiles. Secondly, although we found a linear relationship between the thickness and cross-sectional area of choroidal melanomas at diagnosis, disparate regression rates of these measures indicate that one cannot be substituted for the other. We continue to consider thickness an adequate measure to assess regression in clinical practice, but we cannot presume that thickness alone is an adequate measure for research purposes. We propose that studies on regression of uveal melanomas should use both surrogates until direct measurements of volume are practical. The main disadvantage of analyzing cross-sectional area is that it takes additional time and effort to determine.

**M. Rashid**, None;

**J. Heikkonen**, None;

**T. Kivelä**, None

*. 2003; 110: 2223–2234.*

*Ophthalmology**. 2002; 109: 2197–2206.*

*Ophthalmology**. 1984; 91: 1716–1719.*

*Ophthalmology**. 1985; 5: 73–78.*

*Retina**. 1986; 224: 397–400.*

*Graefes Arch Clin Exp Ophthalmol**. 1987; 94: 1173–1177.*

*Ophthalmology**. 1990; 74: 359–362.*

*Br J Ophthalmol**. 2008; 28: 1289–1295.*

*Retina**. 2015; 22: 103–107.*

*Middle East Afr J Ophthalmol**. 2013; 7: 193–198.*

*Clin Ophthalmol**. 2014; 55: 556–559.*

*Invest Ophthalmol Vis Sci**. 1986; 70: 844–851.*

*Br J Ophthalmol**. 1987; 190: 17–20.*

*Klin Monatsbl Augenheilkd**. 1990; 4: 221–224.*

*Ophtalmologie**. 1990; 196: 6–10.*

*Klin Monatsbl Augenheilkd**. 2002; 109: 1137–1141.*

*Ophthalmology**. 2003; 217: 315–319.*

*Ophthalmologica*^{106}Ru.

*. 2003; 220: 853–860.*

*Klin Monatsbl Augenheilkd**. 2003; 87: 1370–1373.*

*Br J Ophthalmol**. 2004; 88: 892–895.*

*Br J Ophthalmol**. 2006; 244: 1575–1580.*

*Graefes Arch Clin Exp Ophthalmol**. 2013; 97: 653–658.*

*Br J Ophthalmol**. 2014; 34: 182–187.*

*Retina**. 1989; 96: 1566–1573.*

*Ophthalmology**. 1993; 100: 1555–1559, discussion 1560.*

*Ophthalmology**. 2012; 154: 227–232, e222.*

*Am J Ophthalmol**. 2014; 49: 60–65.*

*Can J Ophthalmol**. 1998; 82: 154–158.*

*Br J Ophthalmol**. 2011; 81: 1046–1048.*

*Int J Radiat Oncol Biol Phys**. 1997; 115: 474–477.*

*Arch Ophthalmol**. 1999; 18: 143–149.*

*J Ophthalmic Nurs Technol**. 1999; 91: 523–528.*

*J Natl Cancer Inst**. 2000; 92: 205–216.*

*J Natl Cancer Inst**. 2002; 225: 416–419.*

*Radiology**. 1980; 214: 139–146.*

*Albrecht Von Graefes Arch Klin Exp Ophthalmol**Collaborative Ocular Melanoma Study - Study Forms, May 31, 1991*.

*Baltimore, MD: Wilmer Ophthalmological Institute;*1991: 1–4.

*Collaborative Ocular Melanoma Study - Manual of Procedures, January 1995*.

*Baltimore, MD: Wilmer Ophthalmological Institute;*1995: 1–21.

*. 1993; 13: 69–74.*

*Retina**. 2003; 121: 68–72.*

*Arch Ophthalmol**. 2004; 18: 619–623.*

*Eye**. 1987; 15: 349–357.*

*Aust N Z J Ophthalmol**. 1988; 16: 59–66.*

*Aust N Z J Ophthalmol**. 1991; 19: 327–333.*

*Aust N Z J Ophthalmol**. 2001; 119: 1275–1282.*

*Arch Ophthalmol**. 2014; 34: 2110–2120.*

*Retina**. London: Chapman & Hall; 1991: 396–402.*

*Practical Statistics for Medical Research*. 1st ed*. Thousand Oaks: SAGE Publications; 1997: 85–111.*

*Applied Regression Analysis Linear Models, and Related Methods*. 1st ed*. 1979; 74: 829–836.*

*J Am Stat Assoc**. 2013; 31: 2825–2831.*

*J Clin Oncol**. 2003; 44: 4651–4659.*

*Invest Ophthalmol Vis Sci**. 2007; 114: 1212–1214.*

*Ophthalmology**. 2001; 119: 951–965.*

*Arch Ophthalmol**. 1998; 125: 767–778.*

*Am J Ophthalmol*