Patient Characteristics: Of 1923 patients in the database, we found 273 patients with metastatic cancer had no systemic treatment prior to imaging/biopsy, and also had 2-[18F]FDG PET/CT within 6 months prior to the tissue biopsy (Table 1). All patients had metastatic (stage IV) disease at the time of blood draw. Because patients with early-stage cancer were not included in the study, the cancer stage was not a variable. Based on the TMB values and prior precedent for cut-off values[13], patients were categorized into three groups: TMB = 0-1 mutations/mb (N = 39 patients, 14%), TMB = 2-11 mutations/mbp (N = 174, 64%), and TMB ≥ 12 mutations/mb (N = 60, 22%); the mean TMB in each group was 0, 5, and 37 mutations/mbp. There was no difference in gender distribution between groups. Patients in the TMB = 2-11 mutations/mb group were slightly younger on average than those in the two other TMB groups (63.9 vs. 68.5 or 68.3 years, p = 0.03). organ distribution showed statistically significant differences between TMB categories in different cancer types, with higher TMB in melanoma and lung cancer than in breast, gastrointestinal or “other” cancers (Table 1). The most notable pattern was seen with melanoma, which expectedly had the highest percentage of patients among cancer types in the TMB ≥12 mutations/mb category (60% vs. 7%–30% in other cancer types)[12, 14].
Table 1. Patient characteristics in three quantiled TMB groups (N = 273 patients)
|
|
TMB (0–1 mutations/mb)
N = 39
|
TMB (2–11 mutations/mb)
N = 174
|
TMB (≥12 mutations/mb)
N = 60
|
p-value
|
TMB (Mean ± SD)
Median
|
0
0
|
5.2 ± 2.5
5
|
37.4 ± 47.6
20
|
Not applicable
|
SUVmax (Mean ± SD)
Median (range)
|
4.5 ± 3.9
3.9 (0-16.4)
|
8.4 ± 7.8
6.8 (0-74.0)
|
11.2 ± 8.5
9.2 (1.6-49.6)
|
p < 0.0001*
|
Age at time of biopsy (years) (Mean ± SD)
Median (range)
|
68.5 ± 12.5
71 (34-91)
|
63.9 ± 13.7
65 (23-96)
|
68.3 ± 12.8
70 (28-89)
|
p = 0.03
|
Women (N (%))
Men (N (%))
|
22 (14%)
17 (14%)
|
104 (67%)
70 (60%)
|
30 (19%)
30 (26%)
|
p = 0.4
|
Melanoma (N = 15)
Lung cancer (N = 61)
Gastrointestinal (N = 36)
Breast (N = 43)
Other (N = 118)*
|
2 (13%)
5 (8%)
3 (9%)
7 (16%)
22 (19%)
|
4 (27%)
38 (62%)
29 (80%)
33 (77%)
70 (59%)
|
9 (60%)
18 (30%)
4 (11%)
3 (7%)
26 (22%)
|
p < 0.001
|
Abbreviations: SD = standard deviation; SUV = Standardized Uptake Value; TMB = Tumor Mutational Burden.
*Other cancers consisted of head and neck, adrenal, bladder, ovary, uterus, prostate, musculoskeletal, and hematologic malignancies, and cancers of unknown primary.
SUVmax correlates with TMB: Median SUVmax was 3.9, 6.8 and 9.2 for the 0-1, 2-11 and >12 mutations/mb groups (p < 0.0001). Raw diagnostics showed that the TMB is the only variable with statistically significant relationship with SUVmax (p<0.003) (Table 2). However, due to the highly skewed distributions of both TMB and SUV, shifted log transformations were used for analysis because statistical model diagnostics indicated that both SUVmax and TMB should be analyzed on the log scale (Table 3). Post hoc analysis showed significant differences between SUVmax in all groups: mean increase in shifted SUVmax values for TMB ≥12 mutations/mb category vs. TMB 2-11 category was 38.6% with 95% CI= [11.7%,71.9%] (p<0.003); mean increase in shifted SUVmax values for TMB ≥12 category vs. TMB 0-1 category was 145% with 95% CI= [82.2%, 229.5%] (p<0.001), and for TMB 2-11 category vs. TMB 0-1 category was 76.8%, with 95% CI = [37.0%,128.2%], (p<0.001) (Figure 1). Linear correlation between all shifted-log TMB and shifted-log SUVmax had Pearson correlation coefficient r=0.34, (p<0.001) (Figure 2). Among different cancer types, breast cancer patients showed linear correlation between shifted-log TMB and shifted-log SUVmax with Pearson correlation coefficient r=0.40, (p=0.008). This linear correlation coefficient for the lung cancer patients was r=0.43, (p=0.001), and for the other cancer patients was r= 0.37 (p<0.001). For the melanoma and gastrointestinal patients this relationship was not statistically significant, perhaps due to smaller number of patients in these two groups (N= 15, and 36, respectively).
Table 2. Univariate analysis of relationship of variables to SUVmax in the raw scale
|
|
Unit increase in SUVmax compared to reference, univariate model (95% CI)
|
p-value univariate
|
TMB
|
0.06 (0.02, 0.09)
|
0.002*
|
Age (years)
|
0.04 (–0.03, 0.11)
|
0.3
|
Men (N = 117)
Women (N = 156)
|
–0.40 (–2.27, 1.47)
Reference
|
0.7
|
Melanoma (N = 15)
Lung cancer (N = 61)
Gastrointestinal (N = 36)
Breast (N = 43)
Other (N = 118)
|
1.97 (–2.73, 6.67)
2.20 (–1.02, 5.41)
Reference
–0.58 (–4.03, 2.88)
0.77 (–2.14, 3.68)
|
0.4
|
* Higher TMB was significantly correlated with increased SUVmax.
Table 3. Multivariate analysis of relationship of variables to SUVmax in the log scale*
|
|
Percent increase in (SUVmax+ 1) per unit increase in variable, univariate model (95% CI)*
|
p-valueunivariate
|
Percent increase in (SUVmax+ 1) per unit increase in variable, multivariate model (95% CI)*
|
p-value multivariate***
|
Log(TMB + 1)
|
27.9% (18.0%, 38.7%)**
|
<0.001
|
27.8% (17.8%, 38.7%)
|
<0.001****
|
Age (years)
|
0.2% (–0.4%, 0.9%)**
|
0.5
|
—
|
—
|
Men (N = 117)
Women (N = 156)
|
4.3% (–13.5%, 25.7%)
Reference
|
0.7
|
—
|
—
|
Melanoma
(N = 15)
Lung cancer
(N = 61)
Gastrointestinal
(N = 36)
Breast (N = 43)
Other (N = 118)
|
12.6% (–29.3%, 79.2%)
48.8% (8.3%, 104.5%)
Reference
3.3% (–26.6%, 45.5%)
13.6% (–14.8%, 51.6%)
|
0.07
|
–3.5% (–37.9%, 50.0%)
45.1% (7.5%, 95.8%)
Reference
13.3% (–18.0%, 56.6%)
17.1% (–10.7%, 53.7%)
|
0.08
|
* Statistical model diagnostics indicated that SUVmax and TMB should be analyzed on the log scale, due to the highly skewed distributions of both TMB and SUVmax. TMB and SUVmax values are analyzed as linear variables on shifted-log scale. CI= Confidence Interval.
**For every 1 unit increase in log(TMB + 1), there is a 27.9% increase in the predicted geometric mean. Similarly for every year increase in age, there is a 0.2% increase in the predicted geometric mean. See Methods for statistical analysis.
***Only variables with p-value ≤ 0.1 in univariate were tested in multivariate analysis.
****Higher log(TMB + 1) was significantly correlated with increased (SUVmax + 1).
Multivariate analysis of factors affecting SUVmax: We found that sex and age had no correlation with SUVmax. Cancer type relationship with SUVmax showed a statistically insignificant trend (with lung cancer having higher SUVmax, albeit not statistically significant) (multivariate p-value = 0.08). The only variable that correlated significantly with SUVmax, was TMB; Each unit increase in log(TMB + 1) resulted in a 27.8% increase in (SUVmax + 1) (multivariate p <0.001) (Table 3).