Central non-enhancement on CT and histologic necrosis in renal masses

Multicenter reader analysis

Author

Lu Mao

Published

November 17, 2026

Statistical analysis

Categorical variables were summarized as counts and percentages, and continuous variables as medians and interquartile ranges. Following the multireader, multicase approach described by Obuchowski and Bullen (2022), we analyzed each radiologist’s assessment and included random intercepts for site, reader within site, and mass within site. We estimated the sensitivity and specificity of central non-enhancement (CNE) for histologic necrosis using a logistic mixed-effects model, as in the multicenter renal-mass study by Schieda et al. (2022). Linear mixed-effects models related the volumetric CNE-to-mass ratio and subjective CNE percentage to histologic percent necrosis. We adjusted separately for the same reader’s longest tumor diameter and tumor volume. We assessed reader agreement using Cohen’s κ for CNE presence and single-reader absolute-agreement intraclass correlation coefficients (ICCs) for CNE extent. Secondary analyses examined WHO/ISUP grade, AJCC stage, and rhabdoid and sarcomatoid features. We report 95% confidence intervals (CIs). P-values less than 0.05 were considered statistically significant. Analyses were performed using R 4.6.2.

Results

Study cohort

The six institutions contributed 281 surgically resected cT1b or cT2 masses in 281 patients. Both local radiologists assessed each mass. Histologic necrosis status was known for 280 masses, including 114 with necrosis. A numeric percentage was available for 269 masses. Reports documenting necrosis without a percentage contributed to the binary analysis. A missing pathology result was excluded from the corresponding analysis.

Table 1: Cohort characteristics by institution
Characteristic Overall
N = 2811
UTSW
N = 511
MGB
N = 251
NYU
N = 431
Stanford
N = 531
Michigan
N = 591
UW
N = 501
Age, years 65 (56, 71) 67 (58, 74) 61 (54, 67) 64 (51, 71) 65 (56, 74) 63 (56, 70) 66 (53, 70)
Sex






    Female 89 (31.7%) 13 (25.5%) 6 (24.0%) 13 (30.2%) 27 (50.9%) 20 (33.9%) 10 (20.0%)
    Male 192 (68.3%) 38 (74.5%) 19 (76.0%) 30 (69.8%) 26 (49.1%) 39 (66.1%) 40 (80.0%)
Clinical T stage






    T1b 188 (66.9%) 32 (62.7%) 12 (48.0%) 25 (58.1%) 28 (52.8%) 50 (84.7%) 41 (82.0%)
    T2 93 (33.1%) 19 (37.3%) 13 (52.0%) 18 (41.9%) 25 (47.2%) 9 (15.3%) 9 (18.0%)
WHO/ISUP grade






    1 6 (2.6%) 1 (2.3%) 2 (8.7%) 2 (6.3%) 0 (0.0%) 0 (0.0%) 1 (2.2%)
    2 95 (41.1%) 22 (50.0%) 7 (30.4%) 21 (65.6%) 16 (44.4%) 10 (19.6%) 19 (42.2%)
    3 88 (38.1%) 14 (31.8%) 8 (34.8%) 7 (21.9%) 12 (33.3%) 33 (64.7%) 14 (31.1%)
    4 42 (18.2%) 7 (15.9%) 6 (26.1%) 2 (6.3%) 8 (22.2%) 8 (15.7%) 11 (24.4%)
    Unavailable 50 7 2 11 17 8 5
AJCC stage






    I 129 (46.1%) 24 (47.1%) 9 (36.0%) 12 (27.9%) 15 (28.8%) 40 (67.8%) 29 (58.0%)
    II 28 (10.0%) 6 (11.8%) 6 (24.0%) 3 (7.0%) 7 (13.5%) 6 (10.2%) 0 (0.0%)
    III 111 (39.6%) 19 (37.3%) 7 (28.0%) 28 (65.1%) 24 (46.2%) 13 (22.0%) 20 (40.0%)
    IV 12 (4.3%) 2 (3.9%) 3 (12.0%) 0 (0.0%) 6 (11.5%) 0 (0.0%) 1 (2.0%)
    Unavailable 1 0 0 0 1 0 0
Histologic necrosis






    Absent 166 (59.3%) 38 (74.5%) 1 (4.0%) 29 (67.4%) 29 (54.7%) 38 (65.5%) 31 (62.0%)
    Present 114 (40.7%) 13 (25.5%) 24 (96.0%) 14 (32.6%) 24 (45.3%) 20 (34.5%) 19 (38.0%)
    Unavailable 1 0 0 0 0 1 0
Rhabdoid features






    Absent 233 (92.1%) 42 (89.4%) 23 (92.0%) 41 (95.3%) 26 (89.7%) 56 (94.9%) 45 (90.0%)
    Present 20 (7.9%) 5 (10.6%) 2 (8.0%) 2 (4.7%) 3 (10.3%) 3 (5.1%) 5 (10.0%)
    Unavailable 28 4 0 0 24 0 0
Sarcomatoid features






    Absent 252 (93.7%) 46 (93.9%) 24 (96.0%) 42 (97.7%) 36 (83.7%) 56 (94.9%) 48 (96.0%)
    Present 17 (6.3%) 3 (6.1%) 1 (4.0%) 1 (2.3%) 7 (16.3%) 3 (5.1%) 2 (4.0%)
    Unavailable 12 2 0 0 10 0 0
1 Median (Q1, Q3); n (%)

Detection of histologic necrosis

Readers called CNE in 202/227 readings of masses with histologic necrosis and 232/330 readings of masses without histologic necrosis. Table 2 gives the model-based estimates.

Table 2: Detection of histologic necrosis by an individual reader. Estimates and 95% confidence intervals.
Adjustment Masses Readings Sensitivity, % False-positive rate, % Specificity, % Difference, pp
Unadjusted 280 557 87.5 (76.4, 96.2) 73.9 (59.7, 86.7) 26.1 (13.3, 40.3) 13.6 (4.5, 22.0)
Diameter 280 556 84.2 (71.0, 92.0) 76.9 (63.4, 86.4) 23.1 (13.6, 36.6) 7.3 (-1.3, 15.9)
Volume 280 556 83.8 (70.1, 91.9) 77.1 (63.2, 86.8) 22.9 (13.2, 36.8) 6.7 (-1.7, 15.1)

The difference is sensitivity minus the false-positive rate, in percentage points (pp). Unadjusted intervals are from a parametric bootstrap. Size-adjusted intervals use the observed information and delta method after standardizing both probabilities to the same distribution of reader-measured tumor size.

Figure 1: Estimated probability of a positive CNE call by histologic necrosis status. Bars show 95% confidence intervals from the unadjusted mixed model.
Table 3: Odds ratio for a positive CNE call per 10-percentage-point increase in histologic necrosis. Estimates and model-based Wald confidence intervals are conditional on the model random effects.
Adjustment Masses Readings OR (95% CI)
Unadjusted 269 535 1.19 (0.64, 2.21)
Diameter 269 534 0.56 (0.30, 1.07)
Volume 269 534 0.54 (0.29, 1.01)

A sensitivity analysis standardized the within-site call rates to the observed distribution of masses across the six sites. The estimated probability of a positive CNE call was 87.2% in masses with histologic necrosis and 73.4% in masses without it.

Extent of CNE and histologic necrosis

The volumetric ratio and subjective assessment are both expressed on a 0–100% scale. Each coefficient is the difference in the individual CT measurement, in percentage points, for the stated histologic contrast. Confirmed absence of CNE contributes zero to the all-readings analysis.

Table 4: Difference in CT extent (percentage points) and 95% confidence interval, by surgical pathology.
CT measurement Pathology contrast Masses Readings Unadjusted Diameter Volume
Volumetric CNE-to-mass ratio Per 10 percentage points 269 535 2.8 (1.9, 3.8) 1.8 (0.7, 2.8) [269 masses, 534 reads] 1.7 (0.7, 2.5) [269 masses, 534 reads]
Volumetric CNE-to-mass ratio Present vs absent 280 557 6.2 (2.6, 9.6) 2.5 (-1.0, 5.8) [280 masses, 556 reads] 2.4 (-1.2, 6.0) [280 masses, 556 reads]
Subjective CNE percentage Per 10 percentage points 269 535 3.3 (2.0, 4.7) 1.4 (0.1, 2.8) [269 masses, 534 reads] 1.2 (-0.2, 2.5) [269 masses, 534 reads]
Subjective CNE percentage Present vs absent 280 557 10.0 (5.7, 14.6) 3.7 (-1.1, 8.2) [280 masses, 556 reads] 3.4 (-1.4, 7.8) [280 masses, 556 reads]
Table 5: Extent among individual readings with CNE detected and a recorded measurement. Coefficients are CT percentage points per 10-percentage-point increase in histologic necrosis.
CT measurement Masses Readings Unadjusted Diameter Volume
Volumetric CNE-to-mass ratio 226 414 3.2 (2.1, 4.3) 2.4 (1.2, 3.5) 2.3 (1.2, 3.5)
Subjective CNE percentage 225 413 3.4 (2.1, 4.9) 2.3 (0.9, 3.8) 2.1 (0.6, 3.5)

A sensitivity analysis resampled whole masses within each site and accounted for each participating radiologist separately. Both unadjusted associations remained positive, with 95% CIs above zero. After adjustment for tumor diameter, the estimated increases per 10 percentage points of histologic necrosis were +1.63 (95% CI, -0.06 to 3.31) percentage points for the volumetric CNE-to-mass ratio and +1.12 (95% CI, -0.88 to 3.04) for the subjective CNE percentage. Both intervals included zero.

Reader agreement

The two readers agreed on the presence of CNE in 240 of 278 masses (86.3%). The overall within-site κ was 0.57 (0.45, 0.67) (95% CI).

Table 6: Agreement between the two local readers on the presence of CNE.
Site Paired masses Agreement, n (%) Cohen’s κ (95% CI)
UTSW 51 43 (84.3%) 0.51 (0.20, 0.78)
MGB 25 25 (100.0%) NE
NYU 43 39 (90.7%) 0.81 (0.63, 0.95)
Stanford 53 45 (84.9%) 0.00 (CI NE)
Michigan 59 50 (84.7%) 0.64 (0.44, 0.83)
UW 47 38 (80.9%) 0.47 (0.16, 0.73)
Overall (within site) 278 240 (86.3%) 0.57 (0.45, 0.67)

The overall κ uses site-specific reader marginals to calculate chance agreement. Confidence intervals resample paired masses within site, conditional on the participating readers. NE denotes an inestimable coefficient or interval when one or both readers used a single category.

Table 7: Single-reader absolute-agreement ICC for CNE extent among masses with two recorded measurements.
CT measurement Paired masses Within-site ICC (95% CI)
Volumetric CNE-to-mass ratio 198 0.78 (0.72, 0.83)
Subjective CNE percentage 199 0.77 (0.70, 0.82)

The ICC describes agreement between readers at the same site. Site-specific mean levels were fitted separately. Confidence intervals account for variation among masses, readers, and repeated measurements through a parametric bootstrap.

Figure 2: Agreement on CNE extent among masses with two recorded measurements. Each point represents one mass. The diagonal marks equal readings.

Reader measurements and histologic necrosis

Figure 3: Volumetric CNE-to-mass ratio versus histologic percent necrosis. Each panel shows one institution. A horizontal line joins the two readings of the same mass.
Figure 4: Subjective CNE percentage versus histologic percent necrosis. Each panel shows one institution. A horizontal line joins the two readings of the same mass.

Reader-specific correlations

Table 8: Reader-specific Spearman correlations of each CT measure with histologic percent necrosis. N is the number of evaluable masses.
Site Reader Ratio, N Ratio, ρ Subjective %, N Subjective %, ρ
UTSW Reader 1 51 0.26 51 0.23
UTSW Reader 2 51 0.20 51 0.26
MGB Reader 1 25 0.56 25 0.33
MGB Reader 2 25 0.53 25 0.50
NYU Reader 1 43 0.22 43 0.21
NYU Reader 2 43 0.26 43 0.33
Stanford Reader 1 44 0.25 44 0.20
Stanford Reader 2 44 0.17 44 0.25
Michigan Reader 1 56 -0.04 56 -0.01
Michigan Reader 2 56 -0.09 56 -0.10
UW Reader 1 47 0.34 47 0.38
UW Reader 2 50 0.52 50 0.46

Tumor size and the volumetric CNE-to-mass ratio

Figure 5 shows the volumetric CNE-to-mass ratio against tumor diameter and volume for each reading.

Figure 5: Volumetric CNE-to-mass ratio by the same reader’s longest tumor diameter and total tumor volume. Both horizontal axes use logarithmic scales.

CNE and unfavorable pathologic features

Each one-grade increase in WHO/ISUP grade was associated with a 3.1-percentage-point increase in the volumetric CNE-to-mass ratio and a 6.2-percentage-point increase in subjective CNE percentage (Table 9).

Table 9: Reader-level CNE by pathologic feature. Estimates compare adjacent grades or stages, or feature present versus absent. Estimates are conditional odds ratios for CNE presence (Wald 95% CI) and percentage-point differences for CNE extent (bootstrap 95% CI).
Feature Assessment Masses Readings Estimate (95% CI)
WHO/ISUP grade Presence of CNE 231 459 7.98 (1.43, 44.72)
Volumetric ratio 231 459 3.1 (0.8, 5.1)
Subjective CNE percentage 231 459 6.2 (2.7, 9.7)
AJCC stage Presence of CNE 280 557 1.52 (0.54, 4.28)
Volumetric ratio 280 557 1.3 (-0.4, 3.0)
Subjective CNE percentage 280 557 2.2 (-0.2, 5.0)
Rhabdoid features Presence of CNE 253 503 47.38 (0.35, 6400.46)
Volumetric ratio 253 503 -0.5 (-6.5, 5.9)
Subjective CNE percentage 253 503 4.0 (-5.1, 13.1)
Sarcomatoid features Presence of CNE 269 535 71.31 (0.26, 19511.02)
Volumetric ratio 269 535 4.2 (-2.6, 11.4)
Subjective CNE percentage 269 535 6.8 (-3.7, 16.9)

Readers reported CNE in 36/40 (90%) readings of masses with rhabdoid features and 350/463 (76%) without them. The corresponding proportions for sarcomatoid features were 32/34 (94%) and 383/501 (76%).

Figure 6: Individual volumetric ratios by WHO/ISUP grade, AJCC stage, rhabdoid features, and sarcomatoid features. Points represent individual readings. Boxes show the median and interquartile range within each category.

Histologic necrosis and unfavorable pathologic features

Documented histologic necrosis was present in 17/20 (85.0%) of masses with rhabdoid features and 84/232 (36.2%) without them. The corresponding proportions for sarcomatoid features were 13/17 (76.5%) and 96/251 (38.2%) (Table 10).

Table 10: Site-adjusted differences in histologic necrosis per grade or stage higher, or feature present versus absent. Each mass contributes once. Differences and site-stratified bootstrap 95% CIs are in percentage points.
Feature Masses with status Difference in prevalence (95% CI) Masses with percent Difference in mean percent (95% CI)
WHO/ISUP grade 231 18.2 (11.1, 25.2) 222 3.8 (1.0, 6.7)
AJCC stage 279 8.2 (2.9, 13.1) 268 2.1 (0.2, 4.2)
Rhabdoid features 252 49.2 (30.9, 66.1) 249 6.3 (-0.8, 14.1)
Sarcomatoid features 268 38.6 (15.3, 60.7) 258 6.0 (-2.3, 16.5)

Summary

Among 281 resected renal masses from 6 centers, the sensitivity of an individual reader’s CNE assessment for histologic necrosis was 87.5% (95% CI, 76.4–96.2) and specificity was 26.1% (95% CI, 13.3–40.3). Readers agreed on CNE presence in 86.3% of evaluable pairs (κ = 0.57). The ICCs were 0.78 for the volumetric CNE-to-mass ratio and 0.77 for subjective CNE percentage. Before adjustment for tumor size, each 10-percentage-point increase in histologic necrosis corresponded to increases of 2.8 and 3.3 percentage points in these two CT measures, respectively.