FST Grouped
Samuel Goldstein
6/10/2022
Table 1
Demographics by FST (with P-values)
| Demographics | Overall, N = 8541 | Fitzpatrick Skin Type (FST) | p-value2 | ||
|---|---|---|---|---|---|
| FST Type I N = 97 (11%)1 |
FST Type II N = 665 (78%)1 |
FST Type III-V N = 92 (11%)1 |
|||
| Patient Age | <0.001 | ||||
| Mean (SD) | 58 (14) | 64 (14) | 59 (13) | 50 (15) | |
| Median (IQR) | 60 (51, 68) | 66 (57, 73) | 60 (51, 68) | 50 (40, 61) | |
| Range | 10, 94 | 27, 94 | 10, 92 | 17, 90 | |
| Sex | 0.7 | ||||
| Male | 446 (52%) | 50 (52%) | 344 (52%) | 52 (57%) | |
| Female | 408 (48%) | 47 (48%) | 321 (48%) | 40 (43%) | |
| Melanocytosis | 0.002 | ||||
| No | 815 (97%) | 92 (99%) | 641 (97%) | 82 (91%) | |
| Yes | 27 (3.2%) | 1 (1.1%) | 18 (2.7%) | 8 (8.9%) | |
| Heterochromia | 0.068 | ||||
| No | 825 (98%) | 92 (99%) | 646 (98%) | 87 (97%) | |
| Yes | 17 (2.0%) | 1 (1.1%) | 13 (2.0%) | 3 (3.3%) | |
| Study Eye | 0.13 | ||||
| Right Eye | 457 (54%) | 55 (57%) | 359 (54%) | 43 (47%) | |
| Left Eye | 396 (46%) | 42 (43%) | 306 (46%) | 48 (53%) | |
| Visual Acuity | 0.060 | ||||
| 20/20-20/50 | 624 (73%) | 66 (68%) | 501 (75%) | 57 (62%) | |
| 20/60-20/200 | 147 (17%) | 21 (22%) | 104 (16%) | 22 (24%) | |
| 20/400-NLP | 83 (9.7%) | 10 (10%) | 60 (9.0%) | 13 (14%) | |
| 1 n (%) | |||||
| 2 Kruskal-Wallis rank sum test; Pearson's Chi-squared test; Fisher's exact test | |||||
Tables 2
A) Clinical Features by FST
| Clinical Features | Overall, N = 8541 | Fitzpatrick Skin Type (FST) | p-value2 | ||
|---|---|---|---|---|---|
| FST Type I N = 97 (11%)1 |
FST Type II N = 665 (78%)1 |
FST Type III-V N = 92 (11%)1 |
|||
| Distance to optic disc (mm) | 0.4 | ||||
| Mean (SD) | 4.6 (4.2) | 4.9 (4.0) | 4.5 (4.3) | 4.9 (4.5) | |
| Median (IQR) | 4.0 (1.0, 7.0) | 4.0 (2.0, 7.0) | 3.5 (1.0, 7.0) | 4.0 (1.0, 7.0) | |
| Range | 0.0, 20.0 | 0.0, 15.0 | 0.0, 20.0 | 0.0, 19.0 | |
| Unknown | 52 | 4 | 37 | 11 | |
| Distance to foveola (mm) | 0.3 | ||||
| Mean (SD) | 4.4 (4.2) | 4.7 (3.8) | 4.3 (4.3) | 4.3 (4.5) | |
| Median (IQR) | 3.0 (1.0, 6.0) | 4.0 (1.0, 7.0) | 3.0 (1.0, 6.0) | 3.0 (0.0, 6.0) | |
| Range | 0.0, 18.4 | 0.0, 15.0 | 0.0, 18.4 | 0.0, 17.0 | |
| Unknown | 53 | 4 | 38 | 11 | |
| Largest basal diameter (mm) | 0.003 | ||||
| Mean (SD) | 12.3 (4.4) | 13.2 (4.2) | 12.0 (4.3) | 13.3 (4.7) | |
| Median (IQR) | 12.0 (9.0, 16.0) | 14.0 (10.0, 16.0) | 12.0 (8.0, 16.0) | 14.0 (9.0, 17.0) | |
| Range | 1.0, 24.0 | 5.0, 24.0 | 1.0, 24.0 | 3.5, 22.0 | |
| Thickness at DFS (mm) | 0.029 | ||||
| Mean (SD) | 5.8 (3.3) | 6.0 (3.5) | 5.6 (3.2) | 6.7 (3.8) | |
| Median (IQR) | 4.9 (3.0, 8.0) | 5.1 (3.1, 8.1) | 4.7 (3.0, 7.8) | 5.8 (3.0, 9.9) | |
| Range | 0.7, 20.4 | 1.0, 20.4 | 1.0, 19.8 | 0.7, 16.0 | |
| Thickness at DLS (mm) | <0.001 | ||||
| Mean (SD) | 3.23 (2.03) | 3.25 (1.86) | 3.09 (1.95) | 4.23 (2.47) | |
| Median (IQR) | 2.60 (2.00, 3.90) | 2.60 (1.90, 3.95) | 2.50 (1.90, 3.60) | 3.30 (2.40, 5.40) | |
| Range | 0.00, 19.80 | 1.00, 12.50 | 0.00, 19.80 | 0.70, 12.00 | |
| Unknown | 62 | 6 | 49 | 7 | |
| Change in thickness (DLS-DFS) | 0.2 | ||||
| Mean (SD) | -2.23 (2.27) | -2.46 (2.25) | -2.22 (2.24) | -2.09 (2.49) | |
| Median (IQR) | -1.50 (-3.50, -0.60) | -1.70 (-4.05, -0.75) | -1.60 (-3.40, -0.60) | -1.10 (-3.70, -0.40) | |
| Range | -13.50, 5.10 | -8.30, 2.10 | -13.50, 5.10 | -10.80, 2.90 | |
| Unknown | 62 | 6 | 49 | 7 | |
| Percent change in thickness (DLS-DFS/DFS) | 0.005 | ||||
| Mean (SD) | -0.35 (0.28) | -0.37 (0.27) | -0.35 (0.28) | -0.27 (0.26) | |
| Median (IQR) | -0.37 (-0.54, -0.18) | -0.41 (-0.57, -0.22) | -0.37 (-0.54, -0.18) | -0.24 (-0.45, -0.12) | |
| Range | -1.00, 2.12 | -0.76, 0.81 | -1.00, 2.12 | -0.79, 0.55 | |
| Unknown | 62 | 6 | 49 | 7 | |
| Tumor epicenter | 0.2 | ||||
| Choroid | 772 (91%) | 92 (95%) | 603 (91%) | 77 (84%) | |
| Ciliary Body | 64 (7.5%) | 4 (4.1%) | 49 (7.4%) | 11 (12%) | |
| Iris | 16 (1.9%) | 1 (1.0%) | 11 (1.7%) | 4 (4.3%) | |
| 1 n (%) | |||||
| 2 Kruskal-Wallis rank sum test; Fisher's exact test | |||||
Table 3
Clinical Treatment/Management by FST
| Clinical Treatment/Management | Overall, N = 8541 | Fitzpatrick Skin Type (FST) | p-value2 | ||
|---|---|---|---|---|---|
| FST Type I N = 97 (11%)1 |
FST Type II N = 665 (78%)1 |
FST Type III-V N = 92 (11%)1 |
|||
| Time from DFS to date of treatment (mons) | 0.5 | ||||
| Mean (SD) | 1.62 (9.11) | 2.15 (9.79) | 1.71 (9.61) | 0.38 (0.76) | |
| Median (IQR) | 0.10 (0.10, 0.33) | 0.10 (0.10, 0.33) | 0.10 (0.10, 0.33) | 0.10 (0.10, 0.33) | |
| Range | -28.43, 122.07 | -0.23, 68.70 | -28.43, 122.07 | -0.23, 5.57 | |
| Treatment | 0.3 | ||||
| Plaque | 657 (77%) | 66 (68%) | 518 (78%) | 73 (79%) | |
| Plaque and TTT | 131 (15%) | 25 (26%) | 93 (14%) | 13 (14%) | |
| Plaque and PDT | 5 (0.6%) | 2 (2.1%) | 3 (0.5%) | 0 (0%) | |
| Enucleation | 55 (6.4%) | 4 (4.1%) | 45 (6.8%) | 6 (6.5%) | |
| PLSU | 2 (0.2%) | 0 (0%) | 2 (0.3%) | 0 (0%) | |
| Stereotactic Radiation | 1 (0.1%) | 0 (0%) | 1 (0.2%) | 0 (0%) | |
| No Follow up after FNAB | 2 (0.2%) | 0 (0%) | 2 (0.3%) | 0 (0%) | |
| Fine Needle Aspiration Biopsy (FNAB) | >0.9 | ||||
| No | 2 (0.2%) | 0 (0%) | 2 (0.3%) | 0 (0%) | |
| Yes | 852 (100%) | 97 (100%) | 663 (100%) | 92 (100%) | |
| Route of FNAB | |||||
| Not done | 4 (0.5%) | 1 (1.0%) | 3 (0.5%) | 0 (0%) | |
| Trans-scleral | 230 (27%) | 32 (33%) | 167 (25%) | 31 (34%) | |
| Pars Plana | 544 (64%) | 59 (61%) | 434 (66%) | 51 (57%) | |
| Clear Cornea | 22 (2.6%) | 1 (1.0%) | 18 (2.7%) | 3 (3.3%) | |
| Enucleation | 48 (5.7%) | 3 (3.1%) | 40 (6.0%) | 5 (5.6%) | |
| Cells Visible | 0.8 | ||||
| No | 153 (19%) | 18 (20%) | 120 (19%) | 15 (17%) | |
| Yes | 653 (81%) | 70 (80%) | 509 (81%) | 74 (83%) | |
| 1 n (%) | |||||
| 2 Kruskal-Wallis rank sum test; Fisher's exact test; Pearson's Chi-squared test | |||||
Table 4
Outcomes by FST
| Outcomes | Overall, N = 8541 | Fitzpatrick Skin Type (FST) | p-value2 | ||
|---|---|---|---|---|---|
| FST Type I N = 97 (11%)1 |
FST Type II N = 665 (78%)1 |
FST Type III-V N = 92 (11%)1 |
|||
| Follow-up period (mons) | 0.3 | ||||
| Mean (SD) | 43 (35) | 39 (30) | 44 (36) | 41 (37) | |
| Median (IQR) | 33 (17, 63) | 34 (14, 58) | 34 (18, 64) | 28 (15, 54) | |
| Range | 0, 210 | 0, 115 | 0, 210 | 0, 154 | |
| Unknown | 3 | 0 | 1 | 2 | |
| Local Recurrence | 0.3 | ||||
| No | 828 (97%) | 94 (97%) | 645 (97%) | 89 (98%) | |
| Yes | 25 (2.9%) | 3 (3.1%) | 20 (3.0%) | 2 (2.2%) | |
| Vision Loss | 0.6 | ||||
| No | 374 (44%) | 39 (41%) | 291 (44%) | 44 (48%) | |
| Yes | 474 (56%) | 57 (59%) | 370 (56%) | 47 (52%) | |
| Death | 0.003 | ||||
| No | 808 (95%) | 85 (88%) | 636 (96%) | 87 (96%) | |
| Yes | 45 (5.3%) | 12 (12%) | 29 (4.4%) | 4 (4.4%) | |
| Death from malignant melanoma | 0.006 | ||||
| No | 35 (78%) | 9 (69%) | 22 (79%) | 4 (100%) | |
| Yes | 10 (22%) | 4 (31%) | 6 (21%) | 0 (0%) | |
| 1 n (%) | |||||
| 2 Kruskal-Wallis rank sum test; Fisher's exact test | |||||
Tables 5 (A, B and C)
A) Univariate Analysis
| Outcomes | Overall, N = 8541 | Fitzpatrick Skin Type (FST) | p-value2 | ||
|---|---|---|---|---|---|
| FST Type I N = 97 (11%)1 |
FST Type II N = 665 (78%)1 |
FST Type III-V N = 92 (11%)1 |
|||
| Metastasis | 0.019 | ||||
| No | 708 (83%) | 72 (74%) | 559 (84%) | 77 (85%) | |
| Yes | 145 (17%) | 25 (26%) | 106 (16%) | 14 (15%) | |
| Location of Metastasis | 0.023 | ||||
| Liver metastasis | 113 (78%) | 20 (80%) | 83 (78%) | 10 (71%) | |
| Lung metastasis | 4 (2.8%) | 0 (0%) | 4 (3.8%) | 0 (0%) | |
| Liver and lung metastasis | 25 (17%) | 4 (16%) | 19 (18%) | 2 (14%) | |
| Other metastasis | 3 (2.1%) | 1 (4.0%) | 0 (0%) | 2 (14%) | |
| 1 n (%) | |||||
| 2 Fisher's exact test | |||||
B) Univariate Binomial Regression
Interpretation: as the Fitzpatrick scale increases (i.e. as skin color becomes darker), the odds of metastasis decrease!| Metastasis | N | Event N | OR1 | 95% CI1 | p-value | q-value2 |
|---|---|---|---|---|---|---|
| Fitzpatrick Skin Type (FST) | 853 | 0.066 | 0.066 | |||
| FST Type I | 25 | — | — | |||
| FST Type II | 106 | 0.55 | 0.33, 0.91 | |||
| FST Type III-V | 14 | 0.52 | 0.25, 1.07 | |||
| 1 OR = Odds Ratio, CI = Confidence Interval | ||||||
| 2 False discovery rate correction for multiple testing | ||||||
| Metastasis | N | Event N1 | OR2 | 95% CI2 | p-value | q-value3 |
|---|---|---|---|---|---|---|
| Fitzpatrick Skin Type (FST) | 853 | 0.014 | 0.014 | |||
| FST Type I | 12 | — | — | |||
| FST Type II | 29 | 0.32 | 0.16, 0.68 | |||
| FST Type III-V | 4 | 0.33 | 0.09, 0.98 | |||
| 1 There were no instances of metastasis for FST types IV and V | ||||||
| 2 OR = Odds Ratio, CI = Confidence Interval | ||||||
| 3 False discovery rate correction for multiple testing | ||||||
C) Multivariate Regression
Adjustments will be made for: age, tumor largest diameter, and thickness
I will try both Poisson and Binomial distributions for the glm link function (not sure which one is more accurate). A binomial logistic regression was used in the Genetic Analysis of Uveal Melanoma paper from 2019.
Explanation of when to use Poisson versus Binomial from stackexchange:
"If your outcome is discrete, or more precisely, you are dealing with counts (how many times something happen in given time interval),then the most common choice of the distribution to start with is Poisson distribution. The problem with Poisson distribution is that it is rather inflexible in the fact that it assumes that mean is equal to variance, if this assumption is not met, you may consider using quasi-Poisson family, or negative binomial distribution (see also Definition of dispersion parameter for quasipoisson family).
If your outcome is binary (zeros and ones), proportions of “successes” and “failures” (values between 0 and 1), or their counts, you can use Binomial distribution, i.e. the logistic regression model. If there is more then two categories, you would use multinomial distribution in multinomial regression."
C.1) Poisson
Interpretation: FST becomes less of a protective factor for metastasis as adjustments for age, largest basal diameter, and thickness DLS are made.
| Characteristic | Metastasis ~ FST | Metastasis ~ FST + Age | Metastasis ~ FST + Age + Largest Basal Diameter | Metastasis ~ FST + Age + Largest Basal Diameter + Thickness DLS | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| IRR1 | 95% CI1 | p-value2 | IRR1 | 95% CI1 | p-value2 | IRR1 | 95% CI1 | p-value2 | IRR1 | 95% CI1 | p-value2 | |
| Fitzpatrick Skin Type (FST) | 0.72 | 0.52, 0.99 | 0.046 | 0.78 | 0.56, 1.10 | 0.2 | 0.77 | 0.56, 1.05 | 0.10 | 0.75 | 0.54, 1.05 | 0.095 |
| Age | 1.01 | 1.00, 1.03 | 0.036 | 1.01 | 0.99, 1.02 | 0.3 | 1.01 | 1.00, 1.02 | 0.13 | |||
| Largest Basal Diameter (mm) | 1.18 | 1.14, 1.23 | <0.001 | 1.24 | 1.17, 1.30 | <0.001 | ||||||
| Thickness DLS (mm) | 0.95 | 0.88, 1.03 | 0.2 | |||||||||
| 1 IRR = Incidence Rate Ratio, CI = Confidence Interval | ||||||||||||
| 2 Statistically significant P-values (<0.05) emboldened | ||||||||||||
C.2) Binomial
Interpretation: Similar trends to the Poisson regressions. FST become less of a protective factor for metastasis when adjustments for age, largest basal diameter, and thickness DLS are made.
| Characteristic | Metastasis ~ FST | Metastasis ~ FST + Age | Metastasis ~ FST + Age + Largest Basal Diameter | Metastasis ~ FST + Age + Largest Basal Diameter + Thickness DLS | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OR1 | 95% CI1 | p-value2 | OR1 | 95% CI1 | p-value2 | OR1 | 95% CI1 | p-value2 | OR1 | 95% CI1 | p-value2 | |
| Fitzpatrick Skin Type (FST) | 0.67 | 0.46, 0.96 | 0.030 | 0.74 | 0.51, 1.08 | 0.12 | 0.69 | 0.47, 0.99 | 0.047 | 0.67 | 0.45, 0.99 | 0.048 |
| Age | 1.02 | 1.00, 1.03 | 0.021 | 1.01 | 1.00, 1.02 | 0.2 | 1.01 | 1.00, 1.03 | 0.070 | |||
| Largest Basal Diameter (mm) | 1.25 | 1.19, 1.31 | <0.001 | 1.31 | 1.23, 1.40 | <0.001 | ||||||
| Thickness DLS (mm) | 0.94 | 0.85, 1.04 | 0.2 | |||||||||
| 1 OR = Odds Ratio, CI = Confidence Interval | ||||||||||||
| 2 Statistically significant P-values (<0.05) emboldened | ||||||||||||
Survival Analysis
Table 6
| Characteristic | Metastasis | Death | ||||
|---|---|---|---|---|---|---|
| HR1 | 95% CI1 | p-value | HR1 | 95% CI1 | p-value | |
| FST | 0.72 | 0.52, 0.99 | 0.045 | 0.91 | 0.47, 1.78 | 0.8 |
| Age | 1.03 | 1.01, 1.04 | <0.001 | 1.04 | 1.01, 1.07 | 0.019 |
| Largest Basal Diameter | 1.03 | 0.97, 1.10 | 0.3 | 0.95 | 0.84, 1.09 | 0.5 |
| Thickness DLS (mm) | 1.19 | 1.07, 1.32 | <0.001 | 1.29 | 1.03, 1.62 | 0.027 |
| 1 HR = Hazard Ratio, CI = Confidence Interval | ||||||
Risk of and Survival Probability of Metastasis
Risk of and Survival Probability of Death
Log-Rank Tests and Cox Models for Metastasis
#Log-rank test: difference between the survival curves for metastasis
survival::survdiff(Surv(time_till_mets_new, mets_new) ~ fst_grouped, data=data)## Call:
## survival::survdiff(formula = Surv(time_till_mets_new, mets_new) ~
## fst_grouped, data = data)
##
## n=139, 715 observations deleted due to missingness.
##
## N Observed Expected (O-E)^2/E (O-E)^2/V
## fst_grouped=1 23 23 21.2 0.161 0.194
## fst_grouped=2 102 102 98.3 0.140 0.493
## fst_grouped=3 14 14 19.6 1.581 1.985
##
## Chisq= 2 on 2 degrees of freedom, p= 0.4# Traditional adjustment for confounders is to use a Cox model and estimate hazard ratio
# Unadjusted Cox model
survival::coxph(Surv(time_till_mets_new, mets_new) ~ fst_grouped, data = data)## Call:
## survival::coxph(formula = Surv(time_till_mets_new, mets_new) ~
## fst_grouped, data = data)
##
## coef exp(coef) se(coef) z p
## fst_grouped -0.1928 0.8247 0.1611 -1.196 0.232
##
## Likelihood ratio test=1.43 on 1 df, p=0.2318
## n= 139, number of events= 139
## (715 observations deleted due to missingness)# Adjusted Cox model
survival::coxph(Surv(time_till_mets_new, mets_new) ~ fst_grouped + age + largest_basal_diameter + thickness_dls, data = data)## Call:
## survival::coxph(formula = Surv(time_till_mets_new, mets_new) ~
## fst_grouped + age + largest_basal_diameter + thickness_dls,
## data = data)
##
## coef exp(coef) se(coef) z p
## fst_grouped -0.334225 0.715893 0.166843 -2.003 0.045153
## age 0.028326 1.028731 0.007826 3.619 0.000295
## largest_basal_diameter 0.031738 1.032247 0.030256 1.049 0.294191
## thickness_dls 0.173688 1.189684 0.051972 3.342 0.000832
##
## Likelihood ratio test=30.18 on 4 df, p=4.497e-06
## n= 123, number of events= 123
## (731 observations deleted due to missingness)Log-Rank Tests and Cox Models for Death
#Log-rank test: difference between the survival curves
survival::survdiff(Surv(time_till_death, death_new) ~ fst_grouped, data=data)## Call:
## survival::survdiff(formula = Surv(time_till_death, death_new) ~
## fst_grouped, data = data)
##
## n=42, 812 observations deleted due to missingness.
##
## N Observed Expected (O-E)^2/E (O-E)^2/V
## fst_grouped=1 11 10 9.43 0.0339 0.050
## fst_grouped=2 27 26 23.60 0.2434 0.666
## fst_grouped=3 4 4 6.96 1.2605 1.650
##
## Chisq= 1.7 on 2 degrees of freedom, p= 0.4# Traditional adjustment for confounders is to use a Cox model and estimate hazard ratio
# Unadjusted Cox model
survival::coxph(Surv(time_till_death, death_new) ~ fst_grouped, data = data)## Call:
## survival::coxph(formula = Surv(time_till_death, death_new) ~
## fst_grouped, data = data)
##
## coef exp(coef) se(coef) z p
## fst_grouped -0.2397 0.7869 0.2595 -0.924 0.356
##
## Likelihood ratio test=0.86 on 1 df, p=0.3551
## n= 42, number of events= 40
## (812 observations deleted due to missingness)# Adjusted Cox model
survival::coxph(Surv(time_till_death, death_new) ~ fst_grouped + age + largest_basal_diameter + thickness_dls, data = data)## Call:
## survival::coxph(formula = Surv(time_till_death, death_new) ~
## fst_grouped + age + largest_basal_diameter + thickness_dls,
## data = data)
##
## coef exp(coef) se(coef) z p
## fst_grouped -0.09265 0.91151 0.34191 -0.271 0.7864
## age 0.03626 1.03693 0.01547 2.344 0.0191
## largest_basal_diameter -0.04669 0.95438 0.06697 -0.697 0.4857
## thickness_dls 0.25716 1.29325 0.11608 2.215 0.0267
##
## Likelihood ratio test=7.75 on 4 df, p=0.1014
## n= 37, number of events= 36
## (817 observations deleted due to missingness)