hide

library(reportROC)
library(kableExtra)
tar_load(site_perf_75)
tar_load(oper_indic_perf)
tar_load(eutirads_perf)

gbarplot <- function(.df, dep_var=malignant, by_var, title="") {
  
  dep_var <- enquo(dep_var)
  
  by_var <- enquo(by_var)
  
  .df %>% 
    group_by(!! by_var) %>%
    summarise(avg = round(mean(!! dep_var),2 )) %>% 
    ggplot(aes(x=!! by_var, y=avg)) + 
    geom_bar(stat="identity") +
    geom_text(aes(label=avg), vjust = 1.5, color = "white") +
    labs(x=as_label(by_var), y=paste0("mean of `",as_label(dep_var), "`"))
}

Ocena związku ze zmienną zależną: `malignant`

opis zmiennej malignant, zob. tutaj. NIFTP traktowana jako malignant

hide

df %>%
  select(
    malignant 
  ) %>%
  tbl_summary(
    label = list(malignant ~ "Malignant diagnosis")
  )  %>%
    modify_caption("**Table. Histological main diagnosis category**")

Table 1: **Table. Histological main diagnosis category**
Characteristic	N = 20,762¹
Malignant diagnosis	7,907 (38%)
¹ n (%)

Płeć

hide

  df %>% 
    gbarplot(by_var = sex)

Wiek

hide

df %>% 
  gbarplot(by_var=age_cat)

wynik BAC

hide

df %>%
  gbarplot(by_var=beth_class)

TIRADS

hide

df %>%
  gbarplot(by_var=eutirads)

Overall EU-TIRADS diagnosic performance analysis

Predictor: EU-TIRADS class

hide

t <-table(df$malignant, df$eutirads)
dimnames(t) <- list( malignant=c("False","True"), `EU-TIRADS`=c("TR1", "TR2", "TR3", "TR4", "TR5"))
t

         EU-TIRADS
malignant  TR1  TR2  TR3  TR4  TR5
    False 1136 2169 3337 4572 1641
    True    93  243  796 2259 4516

hide

reportROC(gold=df$malignant, predictor = as.integer(df$eutirads))

 Cutoff   AUC AUC.SE AUC.low AUC.up     P   ACC ACC.low ACC.up   SEN
  4.500 0.778  0.003   0.772  0.784 0.000 0.758   0.758  0.758 0.571
 SEN.low SEN.up   SPE SPE.low SPE.up   PLR PLR.low PLR.up   NLR
   0.560  0.582 0.872   0.867  0.878 4.474   4.260  4.699 0.492
 NLR.low NLR.up   PPV PPV.low PPV.up   NPV NPV.low NPV.up   PPA
   0.479  0.505 0.733   0.722  0.745 0.768   0.761  0.775 0.571
 PPA.low PPA.up   NPA NPA.low NPA.up   TPA TPA.low TPA.up KAPPA
   0.560  0.582 0.872   0.867  0.878 0.758   0.752  0.763 0.463
 KAPPA.low KAPPA.up
     0.451    0.476

Performance metrics for each EU-TIRADS threshold

hide

eutirads_perf %>%
          select(
            everything(),
            -"P",
            -"AUC.SE",
            -starts_with(c("PPA", "NPA","TPA", "KAPPA"))
          ) %>%
          select(
            -ends_with(c(".up",".low"))
          ) %>% kable() %>% kable_styling()

threshold	crit_benign	crit_malign	AUC	ACC	SEN	SPE	PLR	NLR	PPV	NPV
2	TR1-1	TR2-5	0.538	0.431	0.988	0.088	1.084	0.133	0.400	0.924
3	TR1-2	TR3-5	0.607	0.524	0.958	0.257	1.289	0.165	0.442	0.908
4	TR1-3	TR4-5	0.687	0.646	0.857	0.517	1.773	0.277	0.522	0.854
5	TR1-4	TR5-5	0.722	0.758	0.571	0.872	4.474	0.492	0.733	0.768

Oznaczenia: crit_benign / crit_malign – kryterium łagodności / złośliwości, AUC = Area Under Curve, ACC = accuracy, SEN = sensitivity, SPE = specificity, PLR = positive likelihood ratio, NLR = negative likelihood ratio, PPV = positive predictive value, NPV = negative predictive value.

Predictor: EU-TIRADS = 5

hide

t <- table(df$malignant, df$eutirads_5, deparse.level = 2) 
dimnames(t) <- list( malignant=c("False","True"), `EU-TIRADS`=c("TR1-4","TR5"))
t

         EU-TIRADS
malignant TR1-4   TR5
    False 11214  1641
    True   3391  4516

hide

reportROC(gold=df$malignant, predictor.binary = df$eutirads_5, plot = FALSE)

   AUC AUC.SE AUC.low AUC.up     P   ACC ACC.low ACC.up   SEN SEN.low
 0.722  0.004   0.713  0.730 0.000 0.758   0.758  0.758 0.571   0.560
 SEN.up   SPE SPE.low SPE.up   PLR PLR.low PLR.up   NLR NLR.low
  0.582 0.872   0.867  0.878 4.474   4.260  4.699 0.492   0.479
 NLR.up   PPV PPV.low PPV.up   NPV NPV.low NPV.up   PPA PPA.low
  0.505 0.733   0.722  0.745 0.768   0.761  0.775 0.571   0.560
 PPA.up   NPA NPA.low NPA.up   TPA TPA.low TPA.up KAPPA KAPPA.low
  0.582 0.872   0.867  0.878 0.758   0.752  0.763 0.463     0.451
 KAPPA.up
    0.476

Diagnostic Odds Ratio and 95% CI

hide

m <- glm(malignant ~ eutirads_5, data = df, family="binomial")
round(exp(cbind(m$coefficients,confint(m)))[2,],1)

        2.5 % 97.5 % 
   9.1    8.5    9.7

Clinical profile of True Positives vs False Positives

Nie martwimy się niską sensitivity dla Eutirads 5 i dużą ilością wyników FalseNeg Bo właśnie to sprzyja ograniczaniu overtreatment

I dobrze że tak właśnie wyszło, takiego narzędzia potrzebujemy (zakładając że działałoby tak samo w populacji osób z chorobami tarczycy - a nie tylko tych operowanych )

Może warto przyjrzeć się kim byli ci pacjenci FalseNeg? Może to w 99 procentach osoby z nowotworami low-risk? (edited)

hide

tpos_vs_tneg <- df |> 
  filter(malignant==1) |>  
  count( eutirads_5, clinical_risk)  |>
  group_by(eutirads_5) |> 
  mutate(
    group_n = sum(n), 
    group_percent = 100*n/group_n
  ) |>
  ungroup()

hide

tpos_vs_tneg |> 
  pivot_wider(id_cols = -c(group_percent), names_from = clinical_risk, values_from =n ) |>
  kable( 
    caption = "N = 7907 malignant tumors"
    ) |>
  kable_styling()

Table 2: N = 7907 malignant tumors
eutirads_5	group_n	high	intermediate	low
0	3391	155	1936	1300
1	4516	848	2361	1307

hide

tpos_vs_tneg |> 
  pivot_wider(id_cols = -c(n), names_from = clinical_risk, values_from = group_percent ) |>
  kable( 
    caption = "N = 7907 malignant tumors",
    digits = 1
    ) |>
  kable_styling()

Table 3: N = 7907 malignant tumors
eutirads_5	group_n	high	intermediate	low
0	3391	4.6	57.1	38.3
1	4516	18.8	52.3	28.9

czyli gdyby w naszej populacji badanej* wykorzystać EU-TIRADS 5 jako predyktor obecności nowotworu złośliwego, to nie wykryjemy 3391 / 7909 = 42.9% raków, z których 4.6% to raki wysokiego ryzyka

(populacja u nas = osoby dorosłe zakwalifikowane do operacji tarczycy)

Site-specific EU-TIRADS diagnosic performance analysis

excluded sites with < 75 operations

Overall performance

hide

ggplot(site_perf_75, aes(x=AUC)) + geom_histogram(bins=9)

hide

site_perf_75 %>%
  summarise( n_sites=n(), n_oper= sum(n),  auc_p50 = median(AUC), auc_min = min(AUC), auc_max = max(AUC))

# A tibble: 1 × 5
  n_sites n_oper auc_p50 auc_min auc_max
    <int>  <int>   <dbl>   <dbl>   <dbl>
1      37  18944   0.752   0.626   0.895

hide

ggplot(site_perf_75, aes(x=n, y=AUC)) + geom_point() + scale_x_continuous(trans = "log") + geom_smooth(method = "lm")

na tym wykresie każda kropka to 1 ośrodek na osi X liczba zgłoszonych operacji na osi Y wartość AUC ROC dla skali TIRADS

AUC ROC mierzy jak dobrze TIRADS w danym ośrodku sprawdza się w odróżnianiu nowotw. łagodnych i złośliwych

z takiego obrazka wynika, że

ośrodki różnią się znacznie pod względem AUC ROC (są takie ośrodki dla których AUC ROC < 0.7 ale też są też takie, dla których prawie 0.9) oraz pod względem wolumenu.
no i że jest jakaś korelacja pomiędzy wolumenem operacji a skutecznością TIRADS w identyfikacji nowotw złośliwych

EU-TIRADS performance by indication to surgery

hide

oper_indic_perf %>% 
  select(oper_indic, n, starts_with("AUC"))

# A tibble: 5 × 6
  oper_indic               n   AUC AUC.SE AUC.low AUC.up
  <fct>                <int> <dbl>  <dbl>   <dbl>  <dbl>
1 Excluding malignancy  8842 0.558  0.006   0.547  0.569
2 Malignancy            4235 0.674  0.019   0.636  0.712
3 Compression symptom   4318 0.553  0.014   0.527  0.58 
4 Thyreotoxicosis       2817 0.66   0.019   0.622  0.698
5 Other                  550 0.621  0.03    0.562  0.68

hide

ggplot(oper_indic_perf , aes(y=fct_rev(oper_indic), x=AUC)) + 
  geom_bar(stat="identity") + 
  geom_errorbar(aes(xmin=AUC.low, xmax=AUC.up), width=0.4) + 
  labs(x="EU-TIRADS AUC-ROC ± 95% CI", y="Main indication for surgery")

EU-TIRADS FNA indication

EU-TIRADS guidelines recommend FNAB for:

category 5, nodules ˃10, or ˂10 mm when suspicious lymph nodes are present;
category 4, nodules ˃15 mm;
category 3, nodules ˃20 mm.
Category 2 nodules are not indicated for FNAB

W dostępnych danych informacja o wymiarze guzka jest dostępna jedynie dla nowotworów złośliwych

Nie mamy informacji czy w przedoperacyjnej ocenie USG stwierdzano podejrzane węzły chłonne.

hide

df %>% 
  group_by(malignant) %>% 
  skim_without_charts(size_main)

Table 4: Data summary
Name	Piped data
Number of rows	20762
Number of columns	44
_______________________
Column type frequency:
numeric	1
________________________
Group variables	malignant

Variable type: numeric

skim_variable	malignant	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100
size_main	0	12839	0.00	16.44	19.96	0	5.75	8.5	20.25	80
size_main	1	245	0.97	15.87	14.13	0	7.00	12.0	20.00	150

hide

df %>% 
  #filter(!is.na(fna_indicated)) %>%
  count(malignant, fna_indicated) %>% 
  pivot_wider(names_from=malignant, values_from=n, values_fill = NA) %>% 
  kable() %>% 
  kable_styling()

fna_indicated	0	1
0	10	4145
1	6	3517
NA	12839	245

hide

#%>% 
#  add_header_above(c(" ", "malignant"=2))

Czyli:

wśród n=7907 pacjentów z nowotw. złośl. ocena wskazań do BAC wg algorytmu EU-TIRADS była możliwa u 7662 (dla 245 fna_indicated == NA, co wynika z braku wprowadzonej wielkości guza, zm. size_main)
U 4145 z tych 7662 osób (54.1) nie było wskazań do BAC.

Clinical data and TNM stage by justified/unjustified FNA

U ponad 50% pacjentów z histopatologicznie potwierdzonym nowotworem złośliwym tarczycy nie było wskazań do BAC wg algorytmu EU-TIRADS ??.

hide

df %>% 
  filter(
    malignant == 1,
#    !is.na(fna_indicated)  # usuń missing data
  ) %>%
  select(
    fna_indicated,
    age,
    sex,
    oper_indic,
    oper_type_th_aggr,
    oper_type_ln_aggr
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
   fna_indicated = fct_explicit_na(fna_indicated),

  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. Clinical by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 5: Table. Clinical by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
age	48 (37, 60)	49 (38, 60)	47 (36, 61)	51 (40, 65)
sex
Male	1,614 (20%)	702 (17%)	849 (24%)	63 (26%)
Female	6,293 (80%)	3,443 (83%)	2,668 (76%)	182 (74%)
oper_indic
Excluding malignancy	3,030 (38%)	1,661 (40%)	1,238 (35%)	131 (53%)
Malignancy	4,053 (51%)	1,885 (45%)	2,119 (60%)	49 (20%)
Compression symptom	462 (5.8%)	298 (7.2%)	118 (3.4%)	46 (19%)
Thyreotoxicosis	231 (2.9%)	195 (4.7%)	26 (0.7%)	10 (4.1%)
Other	131 (1.7%)	106 (2.6%)	16 (0.5%)	9 (3.7%)
oper_type_th_aggr
Lobectomy	3,022 (38%)	1,740 (42%)	1,182 (34%)	100 (41%)
Other	449 (5.7%)	264 (6.4%)	162 (4.6%)	23 (9.4%)
Thyroidectomy	4,436 (56%)	2,141 (52%)	2,173 (62%)	122 (50%)
oper_type_ln_aggr
None	4,022 (51%)	2,389 (58%)	1,469 (42%)	164 (67%)
CLND	2,641 (33%)	1,325 (32%)	1,260 (36%)	56 (23%)
CLND + LLND	946 (12%)	272 (6.6%)	663 (19%)	11 (4.5%)
Other	298 (3.8%)	159 (3.8%)	125 (3.6%)	14 (5.7%)
¹ Median (IQR); n (%)

hide

df %>% 
  filter(
    malignant == 1,
  #  !is.na(fna_indicated)  # usuń missing data
  ) %>%
  select(
    fna_indicated,
    tnm_t,
    tnm_n,
    tnm_m
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
   fna_indicated = fct_explicit_na(fna_indicated),
 #  tnm_t = fct_drop(tnm_t),    # nie wyświetlaj missing data
   
   tnm_t = fct_recode(tnm_t,

       "pT1a"="pT1a", 
       "pT1b"="pT1b", 
       "pT2"="pT2", 
       "pT3a"="pT3a", 
       "pT3b-pT4b"="pT3b", 
       "pT3b-pT4b"="pT4a", 
       "pT3b-pT4b"="pT4b", 
       "pTx"="pTx",
       "pTx"="(Missing)",
       "pTx"="pT0"
   ),
 
  tnm_t = fct_relevel(tnm_t, "pTx", after=Inf)

  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. TNM stage by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 6: Table. TNM stage by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
tnm_t
pT1a	3,502 (44%)	3,407 (82%)	42 (1.2%)	53 (22%)
pT1b	2,260 (29%)	607 (15%)	1,629 (46%)	24 (9.8%)
pT2	1,267 (16%)	57 (1.4%)	1,198 (34%)	12 (4.9%)
pT3a	425 (5.4%)	34 (0.8%)	383 (11%)	8 (3.3%)
pT3b-pT4b	301 (3.8%)	38 (0.9%)	258 (7.3%)	5 (2.0%)
pTx	152 (1.9%)	2 (<0.1%)	7 (0.2%)	143 (58%)
tnm_n
pN0	3,497 (44%)	2,076 (50%)	1,385 (39%)	36 (15%)
pN1a	1,414 (18%)	636 (15%)	768 (22%)	10 (4.1%)
pN1b	787 (10.0%)	197 (4.8%)	574 (16%)	16 (6.5%)
pNx	2,209 (28%)	1,236 (30%)	790 (22%)	183 (75%)
(Missing)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
tnm_m
pM0	6,535 (83%)	3,521 (85%)	2,967 (84%)	47 (19%)
pM1	96 (1.2%)	12 (0.3%)	80 (2.3%)	4 (1.6%)
pMx	1,276 (16%)	612 (15%)	470 (13%)	194 (79%)
(Missing)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
¹ n (%)

hide

df %>% 
  filter(
    malignant == 1,
  #  !is.na(fna_indicated)  # usuń missing data
  ) %>%
  select(
    fna_indicated,
    beth_class
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
   fna_indicated = fct_explicit_na(fna_indicated),
 
  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. Bethesda class by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 7: Table. Bethesda class by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
beth_class
Not performed	734 (9.3%)	472 (11%)	227 (6.5%)	35 (14%)
I	89 (1.1%)	54 (1.3%)	31 (0.9%)	4 (1.6%)
II	374 (4.7%)	256 (6.2%)	77 (2.2%)	41 (17%)
III	422 (5.3%)	235 (5.7%)	153 (4.4%)	34 (14%)
IV	1,834 (23%)	1,014 (24%)	754 (21%)	66 (27%)
V	1,043 (13%)	503 (12%)	502 (14%)	38 (16%)
VI	3,411 (43%)	1,611 (39%)	1,773 (50%)	27 (11%)
¹ n (%)

hide

df %>% 
  filter(
    malignant == 1,
  #  !is.na(fna_indicated)  # usuń missing data
  ) %>%
  select(
    fna_indicated,
    diagn_main
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
   fna_indicated = fct_explicit_na(fna_indicated),
    
   diagn_main = fct_infreq(diagn_main),
    diagn_main = fct_lump_prop(diagn_main, prop = 0.01)
  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. Histopathological type by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 8: Table. Histopathological type by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
diagn_main
Papillary cancer T-96 M-82603	6,766 (86%)	3,850 (93%)	2,836 (81%)	80 (33%)
Follicular cancer T-96 M-83303	395 (5.0%)	91 (2.2%)	294 (8.4%)	10 (4.1%)
Medullary cancer T-9605 M-85103	319 (4.0%)	154 (3.7%)	158 (4.5%)	7 (2.9%)
Hürtle cell (oxyphilic) carcinoma T-96 M-82903	183 (2.3%)	37 (0.9%)	144 (4.1%)	2 (0.8%)
Non-invasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTP)	133 (1.7%)	0 (0%)	1 (<0.1%)	132 (54%)
Other	111 (1.4%)	13 (0.3%)	84 (2.4%)	14 (5.7%)
¹ n (%)

hide

df %>% 
  filter(
    malignant == 1,
   # !is.na(fna_indicated)
  ) %>%
  select(
    fna_indicated,
    tnm_t1a
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
#    tnm_t1a = fct_relevel(tnm_t1a, "T1a", "T1b-T4b"),
   fna_indicated = fct_explicit_na(fna_indicated)
  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. Microcarcinomas by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 9: Table. Microcarcinomas by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
tnm_t1a	3,420 (45%)	3,325 (82%)	42 (1.2%)	53 (51%)
Unknown	308	90	77	141
¹ n (%)

hide

df %>% 
  filter(
    malignant == 1,
   # !is.na(fna_indicated)
  ) %>%
  select(
    fna_indicated,
    high_risk_profile,
    indeterminate_risk_profile,
    low_risk_profile
  ) %>%
  mutate(
    fna_indicated = factor(fna_indicated),
    fna_indicated = fct_recode(fna_indicated, `recommended` = "1", `not recommended`="0"),
#    tnm_t1a = fct_relevel(tnm_t1a, "T1a", "T1b-T4b"),
   fna_indicated = fct_explicit_na(fna_indicated)
  ) %>%
  tbl_summary(by=fna_indicated) %>% 
  add_overall() %>%
  modify_caption(paste0("**Table. thyroid cancer risk profile by EU-TIRADS FNA indications status<br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 10: Table. thyroid cancer risk profile by EU-TIRADS FNA indications status
(n=7907 malignant thyroid tumors)
Characteristic	Overall, N = 7,907¹	not recommended, N = 4,145¹	recommended, N = 3,517¹	(Missing), N = 245¹
high_risk_profile	1,003 (13%)	235 (5.7%)	750 (21%)	18 (7.3%)
indeterminate_risk_profile	4,297 (54%)	1,369 (33%)	2,743 (78%)	185 (76%)
low_risk_profile	2,607 (33%)	2,541 (61%)	24 (0.7%)	42 (17%)
¹ n (%)

high_risk_profile = any of the following: pT3b or higher, pN1b, pM1

low_risk_profile = papillary thyroid cancer AND pT1a AND (pN0 or pNX) AND (pM0 or pMX)

indeterminate_risk_profile = neither high_risk_profile nor low_risk_profile

EU-TIRADS categories by tumor size categories

hide

df %>% 
  filter(malignant == 1) %>% 
  select(
    tnm_t1a,
    eutirads
  ) %>%
  mutate(
    tnm_t1a = factor(tnm_t1a),
    tnm_t1a = fct_recode(tnm_t1a, `T1a` = "1", `T1b-T4b`="0"),
    tnm_t1a = fct_relevel(tnm_t1a, "T1a", "T1b-T4b"),
    tnm_t1a = fct_explicit_na(tnm_t1a)
  )%>%
  tbl_summary(by=tnm_t1a,
                  label = list(eutirads ~ "EU-TIRADS category")) %>% 
  modify_caption(paste0("**Table. EU-TIRADS categories in T1a vs higher T   <br>(n=",sum(df$malignant)," malignant thyroid tumors)**"))

Table 11: Table. EU-TIRADS categories in T1a vs higher T
(n=7907 malignant thyroid tumors)
Characteristic	T1a, N = 3,420¹	T1b-T4b, N = 4,179¹	(Missing), N = 308¹
EU-TIRADS category
TR1	58 (1.7%)	29 (0.7%)	6 (1.9%)
TR2	143 (4.2%)	79 (1.9%)	21 (6.8%)
TR3	362 (11%)	367 (8.8%)	67 (22%)
TR4	934 (27%)	1,242 (30%)	83 (27%)
TR5	1,923 (56%)	2,462 (59%)	131 (43%)
¹ n (%)

EU-TIRADS performance vs microcarcinoma prevalence

Pojawiła się sugestia, że z performance TIRADS może zależeć od ilości mikroraków.

Zmienna tnm_t1a_prop to proporcja nowotw. złośliwych o wym. =< 1 cm w materiale danego ośrodka.

hide

site_perf_75 %>%
  summarise( 
    n_sites=n(), 
    n_oper= sum(n),  
    tnm_t1a_prop_p50 = median(tnm_t1a_prop), 
    tnm_t1a_prop_min = min(tnm_t1a_prop), 
    tnm_t1a_prop_max = max(tnm_t1a_prop)
  )

# A tibble: 1 × 5
  n_sites n_oper tnm_t1a_prop_p50 tnm_t1a_prop_min tnm_t1a_prop_max
    <int>  <int>            <dbl>            <dbl>            <dbl>
1      37  18944              0.4            0.132             0.75

hide

ggplot(site_perf_75, aes(x=tnm_t1a_prop, y=AUC)) + 
  geom_point() + 
  geom_smooth(method="lm", se=F) +
  ylab("EU-TIRADS AUC-ROC") + 
  xlab("Proportion of pT1a malignancies")

na tym wykresie każda kropka to 1 ośrodek, na osi X proporcja raków pT1a w danym ośrodku, na osi Y wartość AUC ROC dla skali TIRADS

EU-TIRADS performance evaluation

Ocena związku ze zmienną zależną: malignant

Płeć

Wiek

wynik BAC

TIRADS

Overall EU-TIRADS diagnosic performance analysis

Predictor: EU-TIRADS class

Performance metrics for each EU-TIRADS threshold

Predictor: EU-TIRADS = 5

Clinical profile of True Positives vs False Positives

Site-specific EU-TIRADS diagnosic performance analysis

Overall performance

EU-TIRADS performance by indication to surgery

EU-TIRADS FNA indication

Clinical data and TNM stage by justified/unjustified FNA

EU-TIRADS categories by tumor size categories

EU-TIRADS performance vs microcarcinoma prevalence

Ocena związku ze zmienną zależną: `malignant`