BJUI BJU INTERNATIONAL
External validation of a nomogram predicting mortality in patients with adrenocortical carcinoma
Laurent Zini ** , Umberto Capitanio ** , Claudio Jeldres*, Giovanni Lughezzani ** , Maxine Sun*, Shahrokh F. Shariat*, Hendrik Isbarn$, Philippe Arjane1, Hugues Widmer1, Paul Perrotte1, Markus GraefenS, Francesco Montorsi+ and Pierre I. Karakiewicz*1
*Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center, Montreal, QC, Canada, and Departments of Urology, +Lille University Hospital, Lille, France, 1University of Montreal, Montreal, Canada, *Vita- Salute San Raffaele, Milan, Italy and$Martini Clinic, Hamburg, Germany Accepted for publication 18 March 2009
Study Type - Prognosis (retrospective cohort) Level of Evidence 2b
OBJECTIVE
To develop nomograms predicting cancer- specific and all-cause mortality in patients managed with either surgery or no surgery for adrenocortical carcinoma (ACC).
PATIENTS AND METHODS
The models were developed in 205 patients with ACC and externally validated using 207 other patients with ACC, identified in the 1973-2004 Surveillance, Epidemiology and End Results database. The predictors comprised age, gender, race, stage and
surgery status. Nomograms based on Cox regression model-derived coefficients were used for predicting the cancer-specific and all-cause mortality, and were tested using area under the receiver operating characteristics (ROC) curve.
RESULTS
In cancer-specific analyses, the median survival of patients within the development cohort was 26 months, vs 71 months in the external validation cohort (P < 0.001). In overall survival analyses, the median values were 21 vs 32 months for, respectively, the development and the external validation cohort (P < 0.001). Three variables (age, stage and surgical status) were included in the nomograms predicting cancer-specific
and all-cause mortality. In the external validation cohort, the nomograms achieved between 72 and 80% accuracy for prediction of cancer-specific or all-cause mortality at 1-5 years after either surgery or diagnosis of ACC for non-surgical patients.
CONCLUSION
Our models are the first standardized and individualized prognostic tools for patients with ACC. Their accuracy was confirmed within a large external population-based cohort of patients with ACC.
KEYWORDS
adrenocortical carcinoma, nomogram, cancer-specific mortality, all-cause mortality, natural history
INTRODUCTION
Adrenocortical carcinoma (ACC) is a rare solid tumour with an estimated incidence of 0.5-2.0 per million; its prognosis is extremely poor [1-5]. On average, patient survival rarely exceeds 2-3 years [1-4] but some patients might enjoy long-term remissions that last up to ≥10 years [6-8]. The heterogeneity of the outcome makes it difficult to provide an estimate of the prognosis after resection of primary ACC. Moreover, in the light of available adjuvant chemotherapy, an accurate prognostic system could help to identify patients at greater risk of death, as such
patients could benefit from early systemic therapy [9]. We explored the possibility of developing a prognostic model aimed at predicting the mortality rate according to patient and tumour characteristics. For that purpose, we relied on the Surveillance Epidemiology and End Results (SEER) database, which represents one of the largest population-based cancer databases.
PATIENTS AND METHODS
Patients diagnosed with ACC between 1973 and 2004 were identified within nine SEER
registries [10], including the Atlanta, Detroit, San Francisco-Oakland, Seattle-Puget Sound metropolitan areas, and the states of Connecticut, Hawaii, Iowa, New Mexico, and Utah. Characteristics of the SEER population are comparable to the general USA population [10]. The ACC diagnostic code (International Classification of Diseases for Oncology, Second edition, ICD-O-2, C74.0 code) was used as the main inclusion criterion. Exclusions consisted of pathology other than ACC (1362, 64.8%), unknown disease stage (44, 2.1%) and unknown surgery status (265, 12.8%). Finally, we also excluded patients with local tumour destruction (14, 0.7%).
| TABLE 1 Descriptive data for the 205 patients in the development and 207 in the external validation cohorts, treated for ACC | ||||
|---|---|---|---|---|
| Variable | Development | Validation | ||
| No surgery | Surgery | No surgery | Surgery | |
| No. of patients | 49 | 156 | 45 | 162 |
| SEER registries | ||||
| San Francisco-Oakland | 18 (36.7) | 53 (34.0) | ||
| Hawaii | 1 (2.0) | 12 (7.7) | ||
| Iowa | 16 (32.7) | 53 (34.0) | ||
| New Mexico | 7 (14.3) | 11 (7.1) | ||
| Metropolitan Atlanta | 7 (14.3) | 27 (17.3) | ||
| Connecticut | 16 (35.6) | 40 (24.7) | ||
| Metropolitan Detroit | 10 (22.2) | 59 (36.4) | ||
| Seattle (Puget Sound) | 17 (37.8) | 43 (26.4) | ||
| Utah | 2 (4.4) | 20 (12.3) | ||
| Mean (median, range): age, years | 54.8 (58.0, 18-89) | 53.2 (53.0, 17-89) | 59.8 (63.0, 30-90) | 53.2 (55.0, 17-87) |
| Male | 30 (61.2) | 70 (44.9) | 17 (37.8) | 60 (37.0) |
| Female | 19 (38.2) | 86 (55.1) | 28 (62.2) | 102 (63.0) |
| Race | ||||
| Caucasian | 42 (85.7) | 133 (85.3) | 39 (86.7) | 148 (91.4) |
| Other | 7 (14.3) | 23 (14.7) | 6 (13.3) | 14 (8.6) |
| SEER stage | ||||
| localized | 4 (8.2) | 92 (59.0) | 4 (8.9) | 91 (56.2) |
| regional | 6 (12.2) | 32 (20.5) | 4 (8.9) | 49 (30.2) |
| distant | 39 (79.6) | 32 (20.5) | 37 (82.2) | 22 (13.6) |
| Mean (median, range): tumour size, cm | 10.5 (10.0, 4.0-19.0) | 12.0 (11.0, 3.0-25.0) | 9.8 (9.00, 1.9-20.0) | 11.8 (12.0, 1.0-25.0) |
| CSM at 2 years, % | 26.0 | 61.4 | 31.3 | 72.3 |
| ACM at 2 years, % | 4.5 | 57.5 | 13.7 | 64.7 |
These criteria resulted in 412 records, of which 94 patients (22.8%) had no surgery and the remaining 318 were treated with adrenalectomy (206, 50.0%), or radical locoregional resection that also included other organs (112, 27.2%). The stage of surgical and non-surgical patients was coded according to the SEER staging system: localized (confined to the adrenal gland), regional (extension into adjacent tissue or lymph node involvement) and distant (metastatic). For all surgery patients, malignant histology was confirmed with the ICD-O-3 SEER histological codes. The cause of death was defined according to the SEER specific cause of death (code 32020). Patients who did not die from ACC were considered as having died from other causes and their follow-up was censored at the time of death unrelated to ACC analyses of cancer- specific mortality (CSM). All-cause mortality (ACM) was used as an endpoint in overall mortality analyses.
For the purpose of model development and its validation, the 412 assessable patients were divided into two similarly sized cohorts,
according to the SEER registry. The development cohort originated from Atlanta and San Francisco-Oakland metropolitan areas, and the states of Hawaii, Iowa, New Mexico registries, and consisted of 205 patients. The validation cohort originated from Detroit and Seattle-Puget Sound metropolitan areas, and the states of Connecticut and Utah, and included 207 patients (Table 1).
The statistical significance of differences in mean and proportions was tested using, respectively, Student’s t-test and the chi- square test. Univariable and multivariable Cox regression models addressed time to CSM and ACM in the development cohort. The predictors consisted of the SEER stage, patient age, gender and race. The Cox model was stratified according to surgery status (surgery vs no surgery), as stage for stage, surgical patients had markedly better prognoses than patients who did not qualify for surgery. Separate models were developed for predicting CSM and ACM. Stepwise variable removal was then applied to the full model, according to the Akaike information criterion, with the intent of developing the
most accurate and parsimonious model [11-13].
Proportional-hazards assumptions were systematically verified for all proposed models, using the Grambsch-Therneau residual-based test [14]. Actuarial survival probabilities were estimated using the Kaplan-Meier method. Subsequently, the multivariable Cox regression coefficients of the most parsimonious model were used to generate nomograms predicting CSM and ACM probabilities at 1, 2, 3 and 5 years after surgery, or 1 and 2 years after diagnosis for non-surgical patients.
The accuracy of the nomograms was quantified with the receiver operating characteristic-derived area under the curve at 1, 2, 3 and 5 years [12,13]. For each time point, we graphically explored the nomogram calibration. Specifically, the nomogram- predicted survival rates were compared with those observed in the external validation sample. All statistical tests were performed with S-Plus Professional and R statistical packages. Statistical significance was set at 0.05.
A
1.0
B
1.0
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
Validation cohort
0.5
Development cohort
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0
12
24
36
48
60
72
84
96 108 120
0.0
0
12
24
36
48
60
72
84
96 108 12
Number at risk 205 118 81
59
43
37
33
28
23
18
14
Number at risk 207 136 98
78
60
50
45
39
31
27
21
Time, Months
Time, Months
C
1.0
D
0.9
1.0
0.8
0.9
0.7
0.8
0.6
0.7
Adrenalectomy
0.5
HR 6.2
Adrenalectomy
0.6
HR 3.5
0.4
Log rank p<0.001
0.5
Log rank p<0.001,
0.3
0.4
No surgery
0.2
No surgery
0.3
0.1
0.2
0.0
0.1
Number at risk
0
12
24
36
48
60
72
84
0.0
96
108
120
Number at risk
0
12
24
36
48
60
72
84
96
108
20
Adrenalectomy 156
57
37
28
18
Adrenalectomy
92
74
57
48
43
37
No surgery 49
110 78
43
33
23
14
162 123
30
26
20
8
3
2
No surgery
45
13
6
4
3
2
2
2
1
1
1
Time, Months
Time, Months
E
1.0
F
0.9
1.0
0.8
0.9
HR 1.3
0.7
0.8
Log rank p=0.4
0.6
0.7
0.5
0.6
Localized
0.4
LHR 2.0
Localized
0.5
0.4
HR 3.5
Regional
0.3
HR 3.3
Log rank p=0.005
Regional
0.3
Log rank p<0.001
0.2
rank p<0.001
0.2
Distant
0.1
0.0
Distant
0.1
0.0
Number at risk 0
12
24
36
48
60
72
84
96
108
120
Number at risk 0
12
24
36
48
60
72
84
96
108
120
Localized
96
75
57
45
33
29
26
22
18
13
11
Localized
95
77
59
47
39
31
27
22
17
17
13
Regional
38
25
19
12
9
8
7
6
5
5
3
Regional
53
39
29
23
17
17
16
15
12
9
7
Distant
71
18
5
2
1
Distant
59
20
10
8
4
2
2
2
2
1
1
Time, Months
Time, Months
| Variable | Hazard ratio, P | ||
|---|---|---|---|
| Univariable | Full multivariable | Reduced multivariable | |
| CSM | |||
| Age, years | 1.01, 0.1 | 1.02, 0.004 | 1.02, 0.008 |
| Gender, (female vs male) | 0.88, 0.5 | 0.93, 0.7 | – – , |
| Race (other vs Caucasian) | 1.37, 0.2 | 1.12, 0.6 | – – , |
| SEER stage | |||
| Regional vs localized | 2.04, 0.007 | 1.90, 0.015 | 1.90, 0.02 |
| Distant vs localized | 7.93, <0.001 | 5.78, <0.001 | 5.72, <0.001 |
| Surgery status (surgery vs no surgery) | 0.15, <0.001 | 0.38, <0.001 | 0.36, <0.001 |
| ACM | |||
| Age, years | 1.02, 0.001 | 1.03, <0.001 | 1.03, <0.001 |
| Gender, (female vs male) | 0.77, 0.1 | 0.81, 0.2 | – – , |
| Race (other vs Caucasian) | 1.23, 0.4 | 1.06, 0.8 | – – , |
| SEER stage | |||
| Regional vs localized | 1.67, 0.03 | 1.44, 0.1 | 1.46, 0.1 |
| Distant vs localized | 6.21, <0.001 | 4.32, <0.001 | 4.14, <0.001 |
| Surgery status (surgery vs no surgery) | 0.15, <0.001 | 0.32, <0.001 | 0.30, <0.001 |
RESULTS
The characteristics of the 412 assessable patients are listed in Table 1; the characteristics of the surgical and non- surgical patients in the development and external validation cohort were similar for age, race, tumour size and stage (all P> 0.05). Conversely, differences were recorded in gender (P = 0.02) of the non-surgical patients in the development and external validation cohort. In cancer-specific analyses, the median survival of patients within the development cohort was 26 months, vs 71 months in the external validation cohort (P < 0.001). In overall survival analyses, the median values were 21 vs 32 months for, respectively, the development and the external validation cohort (P < 0.001).
After stratification for the use of surgery vs no surgery, the surgical patients fared significantly better than the other group in both the development (hazard ratio 6.2, log rank P < 0.001) and external validation (hazard ratio 3.5, log rank P < 0.001) cohorts (Fig. 1A-F). In the development cohort, patients with regional ACC had worse survival than those with localized ACC (hazard ratio 2.0, log rank P = 0.005). This was not the case in the external validation cohort (hazard ratio 1.3, log rank P = 0.4) Patients with distant ACC were significantly more likely to die than patients with regional ACC, in both cohorts. Trends were closely similar when ACM was examined (Fig. 2A-F).
Table 2 lists the univariable and multivariable Cox ACC-specific survival models that were fitted in the development cohort. In the full multivariable model, age (P = 0.004), stage (P < 0.001) and surgery status (P < 0.001) were independent predictors; conversely, gender (P= 0.7) and race (P = 0.6) were not. After variable removal according to the Akaike information criterion, all three independent predictors (age, stage and surgery status) remained in the model. Similarly, in the multivariable analysis addressing ACM (Table 2), age, stage and surgery status were independent predictors and remained in the model after variable removal.
Finally, when we examined the CSM rates in all patients with regional ACC, surgically treated patients (318) had a median survival of 43 months, vs 3 months in 94 patients treated without surgery (P < 0.001). Due to important CSM and ACM differences
ZINI ET AL.
according to surgery status, the current nomograms were stratified according to this variable with the intent of providing the most accurate and specific prediction of prognosis. Figure 3A shows the nomogram predicting CSM, where age and stage of the disease represent highly informative variables. Stage was more informative than age; e.g. the presence of distant metastases contributes to 100 risk points, vs 73 risk points for an age of 90 years. Patients treated with surgery invariably had a lower risk of CSM than their non-surgical counterparts. For example, a 60- year-old surgical patient with regional ACC had a 25% and 43% risk of CSM at 1 and 2 years, vs 43% and 68% for a non-surgical patient.
In external validation, the cancer-specific nomogram achieved 79.6%, 75.2%, 73.7% and 71.8% accuracy at 1, 2, 3 and 5 years, respectively. There were minor departures from ideal predictions at 1, 2, 3 and 5 years (Fig. 4A-D); e.g. at 5 years, for patients with a low probability of CSM (0-20%), the nomogram underestimated the true mortality rate by up to 7%. Conversely at 5 years, in patients with a high risk of CSM (>50%), the nomogram overestimated the true risk of CSM by up to 9%.
Within the nomogram predicting ACM (Fig. 3B), age represented the most informative variable and exceeded the importance of stage. For example, the presence of distant metastases contributed to 75 risk points, vs 100 risk points for an age of 90 years. For example, a 60-year-old surgical patient with regional ACC had a 33% and 50% risk of ACM at 1 and 2 years, vs 59% and 83% for a non-surgical patient.
In external validation, the ACM nomogram achieved 78.3%, 76.2%, 72.2% and 74.9% accuracy at 1, 2, 3 and 5 years, respectively. There were minor departures from ideal predictions at 1, 2, 3 and 5 years (Fig. 4E-H). For example at 5 years, for patients with a low probability of ACM (0-20%), the nomogram underestimated the true mortality rate by up to 10%. Conversely at 5 years, in patients with a high risk of ACM (> 50%) the nomogram overestimated the true risk of CSM by up to 12%.
DISCUSSION
Our objective was to develop a model capable of accurately predicting the prognosis of
FIG. 2. Kaplan-Meier survival curves of all-cause survival in the development cohort (A,C,E) and in the external validation cohort (B,D,F); data were stratified according to treatment type (adrenalectomy vs no surgery, C, D) and stage (E, F).
A
1.0
B
1.0r
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
0.5
0.5
0.4
Development cohort
0.4
Validation cohort
0.3
0.3
0.2
0.2
0.1
0.1
0.0
Number at risk 205 118
0
12
24
36
48
60
72
84
96
108
120
0.0
0
12
24
36
48
60
72
84
96
108
120
81
59
43
37
33
28
23
18
14
Number at risk
71
136
98
78
60
50
45
39
31
27
21
Time, Months
Time, Months
C
1.0
D
0.9
1.0
0.8
0.9
0.7
0.8
0.6
0.7
0.5
HR 6.2
0.6
HR 3.8
Adrenalectomy
0.4
Adrenalectomy
0.5
0.3
Log rank p<0.001
0.4
0.3
Log rank p<0.001
0.2
0.1
0.2
0.0
No
ırgery
0.1
No surgery
Number at risk
0
12
24
36
48
60
72
84
96
108
120
Number at risk
0.0
0
12
24
36
48
60
72
84
96
108
120
Adrenalectomy
No surgery
156 110 78
57
43
37
33
28
23
18
14
Adrenalectomy
162
123
92
74
57
48
43
37
30
No surgery
26
20
49
8
3
2
45 13
6
4
3
2
2
2
1
1
1
Time, Months
Time, Months
E
F
1.0
1.0
0.9
0.9
0.8
0.8
0.7
HR 1.2
0.7
Log rank p=0.4
0.6
0.6
0.5
HR 1.6
0.5
Localized
0.4
HR 3.2
og rank p=0.005.
Localized
0.4
0.3
og rank p<0.001-
Regional
0.3
HR 3.3
Regional
0.2
0.2
0.1
Distant
0.1
Log rank p<0.001
Distant
0.0
0.0
Number at risk 0
12
24
36
48
60
72
84
96
108
120
Number at risk 0
12
24
36
48
60
72
84
96
108
120
Localized
96
75
57
45
33
29
26
22
18
13
11
Localized
95
77
59
47
39
31
27
22
17
17
13
Regional
38
25
19
12
9
8
7
6
5
5
3
Regional
53
39
29
23
17
17
16
15
12
9
7
Distant
71
18
5
2
1
Distant
59
20
10
8
4
2
2
2
2
1
1
Time, Months
Time, Months
A
Points
0 10 20
30
40
50
60 70
80
90
100
Age
10 20 30 40 50 60 70 80 90 Regional
Stage
Localized
Distant
Total points
0
20 40 60 80 100 120 140 160 180 200
CSM at 1 year
No surgery patients Surgery patients
0.14 0.2
0.3
.4 0.5
.70.
0.8
0.9
0.1 0.14 0.2
0.3
0.4 0.50.60
0.
.8
CSM at 2 years
No surgery patients Surgery patients
0.3
0.4 0.50.
60.7
70.8 .8
0.9
0.98
0.99
0.14 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Applicable to surgery patients only:
CSM at 3 years
0.2
0.3 0.4 0.50.60.7 0.8 0.9
0.98
CSM at 5 years
0.3
0.4
0.50
.60.7
0.8
0.9
0.98
0.99
B
Points
0
10 20 30
40
50
60
70
80
90
Age
10
20
30
40
50
60
70
80
90
Stage
Regional
Localized
Distant
Total points
Mortality at 1 year
0
20
40
60
80
00 120 140 160 180 200
No surgery patients
0.2
0.3 0.4 0.50.60.70.8 0.9
0.98
Surgery patients
Mortality at 2 years
0.1 0.14 0.2
0.3 0.4 0.5 0.6
0. 7 0.8
0.9
No surgery patients Surgery patients
0.4 0.5 0.60.70.8 0.9 0.98 0.99 0.999
0.14 0.2
0.3 0.4 0.50.6 0.70.8 0.
0.98
Applicable to surgery patients only:
Mortality at 3 years
0.2
0.3 0.4 0.50.60.70.8 0.9
0.98
Mortality at 5 years
0.3 0.4 0.50.60.70.8
0.9
0.98
0.99
patients with ACC after either surgical or non- surgical management [15]. Our interest in this topic was triggered by clinical observations indicating that mortality is highly variable in this rare tumour [3,16]. ACC represents an aggressive cancer, but some patients can also
die from other causes. To address both endpoints, we developed the CSM and ACM nomogram to assist the clinician in defining individual prognoses in patients with ACC. Currently, few data are available to help with this task. We accomplished our goal and
100
.70
A
1.0}
E
1.0}
Actual Probability
0.8
Actual Probability
0.8
0.6
0.6
0.4
- Ideal
0.4
0.2
Nonparametric
-Ideal
0.2
Nonparametric
0.0
0.0
Predicted Probability at 1 year after ACC diagnosis
0.0
0.2
0.4
0.6
0.8
1.0
Predicted Probability at 1 year after ACC diagnosis
0.0
0.2
0.4
0.6
0.8
1.0
B
1.0}
F
Actual Probability
1.0
0.8
Actual Probability
0.8
0.6
0.6
0.4
- Ideal
0.4
- Ideal
0.2
Nonparametric
0.2
Nonparametric
0.0
0.0
0.0
0.0
0.2
0.4
0.6
0.8
Predicted Probability at 2 years after ACC diagnosis
0.2
0.4
0.6
0.8
1.0
Predicted Probability at 2 years after ACC diagnosis
1.0
C
1.0}
G
1.0
Actual Probability
0.8
Actual Probability
0.8
0.6
0.6
0.4
- Ideal
0.4
- Ideal
0.2
Nonparametric
0.2
Nonparametric
0.0
0.0
Predicted Probability at 3 years after ACC diagnosis
0.0
0.2
0.4
0.6
0.8
1.0
Predicted Probability at 3 years after ACC diagnosis
0.0
0.2
0.4
0.6
0.8
1.0
D
1.0
1.0}
Actual Probability
Actual Probability I
0.8
0.8
0.6
0.6
0.4
- Ideal
0.4
- Ideal
0.2
Nonparametric
0.2
Nonparametric
0.0
0.6
0.8
1.0
0.0
Predicted Probability at 5 years after ACC diagnosis
0.0
0.2
0.4
Predicted Probability at 5 years after ACC diagnosis
0.0
0.2
0.4
0.6
0.8
1.0
developed models that are 72-80% accurate in predicting CSM and ACM. These models rely on age, stage and surgery status. These three variables stem from five original variables that were considered for model inclusion.
The nomogram-predicted probabilities of CSM and ACM closely correlated with the observed rates. Nonetheless, there were some departures from ideal predictions (7-12%) for each of the prediction times (Fig. 4). Those departures need to be considered when the nomograms are applied in clinical practice.
To provide most specific predictions, we stratified the population according to surgery status. As very few of the non-surgical
patients survived for >2 years from the diagnosis, the nomogram can only predict for 1 and 2 years after the diagnosis in this patient group. Conversely, the predictions for surgery patients extend up to 5 years. Despite these important differences in CSM and ACM, the accuracy of the nomograms was very good (72-80%) for the CSM nomogram and for the ACM predictions. This proves the robustness of our models and confirms their accuracy even in populations with different characteristics.
The clinical implications of our nomograms include the ability to provide individualized estimates of CSM and ACM in patients with ACC. To the best of our knowledge, this is the
first model derived and validated for that specific purpose. However, it is likely that other databases might hold more detailed information, which could result in better accuracy and lesser departures from ideal predictions.
The nomogram-derived estimates of CSM might be used for adjusting the frequency and intensity of follow-up. Moreover, patients at greater risk of CSM might be offered adjuvant chemotherapy [17]. This treatment offers a survival benefit, albeit at higher cost to quality of life [17], and therefore, only high-risk patients should be exposed to such therapy.
Unfortunately, in the field of ACC there are no other prognostic models to which these nomograms could be compared. Therefore, their accuracy can only be assessed relative to bladder, kidney and prostate-cancer nomograms, where the area under the curve is generally 61-90% [18,19]. The strength of the current ACC nomograms relative to several existing bladder, prostate, and kidney cancer nomograms resides in their confirmed accuracy within an independent external validation cohort [18,20-26].
Although there are no tools capable of defining the prognosis of patients with ACC, two large-scale studies examined overall and cancer-specific survival in such patients exposed to surgery, radiation or other therapies. Paton et al. [3] assessed the overall survival in 602 patients with ACC registered within the SEER database. They showed the relative effect of localized vs regional vs metastatic stage on overall survival. Kebebew et al. [16] also examined a cohort of 725 patients with ACC, exposed either to surgery or no therapy, within the SEER database between 1973 and 2000. The authors found that tumour grade, stage and surgical resection status were the most important predictors of survival. We corroborate these findings for age, stage and surgical resection status, but we were unable to test the effect of grade as it was not available in 70% of the patients in the SEER database.
The limitations of our study are the relatively small sample size (412), although ACC is infrequent. Therefore, sample sizes used for analyses of other disease models, such as prostate cancer, cannot be expected in ACC [27]. Our data also suffer from lack of detail; e.g. grade and mitotic index might provide
prognostic information in ACC. Unfortunately, neither was available in the SEER database [28]. Despite this limitation, the model was 72-80% accurate. Comorbidity data were also not included in the SEER database, which did not allow us to adjust for this important cofactor. However, ACC represents a very aggressive malignancy and therefore most patients die from ACC instead of dying with ACC. In our development cohort at 1, 2, 3 and 5 years, 11.3%, 11.3%, 15.1% and 16.9% of patients died from other causes.
Notably, CSM was significantly higher in the development cohort. The selection of development and external validation cohorts was made randomly, according to the SEER registry. Consequently, it is positive that there are important regional differences in presentation or management between various SEER registries. These registry-specific differences warrant more detailed assessment in future reports.
The current nomogram represents the only tool of its kind for patients with ACC, but its value has not been confirmed as an indicator for the need to administer adjuvant therapy. Consequently, the tool should be used to predict CSM and ACM in patients with ACC, without having a pre-established threshold for adjuvant chemotherapy in patients with a regional diagnosis. The indicator for adjuvant chemotherapy needs to be derived from the comparative trial that was reported by Terzolo et al. [29]
Finally, the prognostic variables within the nomogram predictor axes do not include a ‘surgery vs no surgery’ stratification. Nonetheless, the nomogram relies heavily on the use of surgery vs non-surgical management, as its stratification variables for ACM and CSM probability axes consist of surgery vs no surgery.
In conclusion, despite its inherent limitations, the current nomogram represents the first externally validated model with proven accuracy that can be applied to patients with ACC. Consequently, we strongly advocate its use in patients with ACC, and strongly encourage a future search for a more accurate prognostic model.
ACKNOWLEDGEMENTS
Pierre I. Karakiewicz is partially supported by the University of Montreal Health Center
Urology Associates, Fonds de la Recherche en Sante du Quebec, the University of Montreal Department Of Surgery and the University of Montreal Health Center (CHUM) Foundation. Laurent Zini is partially supported by the Association Francaise de Recherche sur le Cancer, the Fondation de France- Federation Nationale des Centres de Lutte Contre le Cancer, the Association Francaise d’Urologie and the Ministere Francais des Affaires Etrangeres et Europeennes (Bourse Lavoisier).
CONFLICT OF INTEREST
None declared.
REFERENCES
1 Fujita T, Iwamura M, Yanagisawa N et al. Prognostic impact of perirenal fat or adrenal gland involvement in patients with pT3b renal cell carcinoma. Urology 2007; 69: 839-42
2 Fulmer BR. Diagnosis and management of adrenal cortical carcinoma. Curr Urol Report 2007; 8: 77-82
3 Paton BL, Novitsky YW, Zerey M et al. Outcomes of adrenal cortical carcinoma in the United States. Surgery 2006; 140: 914-20
4 Sidhu S, Sywak M, Robinson B, Delbridge L. Adrenocortical cancer: recent clinical and molecular advances. Curr Opin Oncol 2004; 16: 13-8
5 Scott HW Jr, Abumrad NN, Orth DN. Tumors of the adrenal cortex and Cushing’s syndrome. Ann Surg 1985; 201: 586-94
6 Hara I, Sakamoto Y, Kanomata N et al. Long-term survival after bilateral adrenalectomy for metachronous adrenocortical cancer. Int J Uro/ 2004; 11: 1127-9
7 Vassilopoulou-Sellin R, Schultz PN. Adrenocortical carcinoma. Clinical outcome at the end of the 20th century. Cancer 2001; 92: 1113-21
8 Weitz J, Blumgart LH, Fong Y et al. Partial hepatectomy for metastases from noncolorectal, nonneuroendocrine carcinoma. Ann Surg 2005; 241: 269-76
9 Terzolo M, Angeli A, Fassnacht M et al. Adjuvant mitotane treatment for adrenocortical carcinoma. N Engl J Med 2007; 356: 2372-80
10 Ries LAG, Melbert D, Krapcho M et al. SEER Cancer Statistics Review, 1975-
2004, National Cancer Institute. Bethesda, MD, Available at: http://seer.cancer.gov/ csr/1975-2004/, based on November 2006 SEER data submission, posted to the SEER web site. Accessed January 2007
11 Atkinson AC. A note on the generalized information criterion for choice a model. Biometrika 1980; 67: 413-8
2 Harrell FE Jr, Califf RM, Pryor DB, Lee 12 KL, Rosati RA. Evaluating the yield of medical tests. JAMA 1982; 247: 2543-6
13 Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15: 361-87
14 Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994; 81: 515-26
15 Capitanio U, Jeldres C, Shariat SF, Karakiewicz P. Clinicians are most familiar with nomograms and rate their clinical usefulness highest, look-up tables are second best. Eur Uro/ 2008; 54: 958- 9
16 Kebebew E, Reiff E, Duh QY, Clark OH, McMillan A. Extent of disease at presentation and outcome for adrenocortical carcinoma: have we made progress? World J Surg 2006; 30: 872-8
17 Schteingart DE. Adjuvant mitotane therapy of adrenal cancer - use and controversy. N Engl J Med 2007; 356: 2415-8
18 Karakiewicz PI, Briganti A, Chun FK et al. Multi-institutional validation of a new renal cancer-specific survival nomogram. J Clin Oncol 2007; 25: 1316- 22
19 Kattan MW, Potters L, Blasko JC et al. Pretreatment nomogram for predicting freedom from recurrence after permanent prostate brachytherapy in prostate cancer. Urology 2001; 58: 393-9
20 Chun FK, Briganti A, Jeldres C et al. Tumour volume and high grade tumour volume are the best predictors of pathologic stage and biochemical recurrence after radical prostatectomy. Eur J Cancer 2007; 43: 536-43
21 Briganti A, Chun FK, Salonia A et al. Validation of a nomogram predicting the probability of lymph node invasion among patients undergoing radical prostatectomy and an extended pelvic lymphadenectomy. Eur Uro/ 2006; 49: 1019-26
22 Gallina A, Jeldres C, Chun FK et al. Prediction of pathological stage is inaccurate in men with PSA values above 20 ng/ml. Eur Uro/ 2007; 52: 1374-80
23 Karakiewicz PI, Benayoun S, Kattan MW et al. Development and validation of a nomogram predicting the outcome of prostate biopsy based on patient age, digital rectal examination and serum prostate specific antigen. J Uro/ 2005; 173: 1930-4
24 Klotz L. Nomogram for predicting survival in men with clinically localized prostate cancer who do not undergo definitive therapy. Nat Clin Pract Urol 2008; 5: 362-3
25 Nakanishi H, Wang X, Ochiai A et al. A nomogram for predicting low-volume/ low-grade prostate cancer: a tool in
selecting patients for active surveillance. Cancer 2007; 110: 2441-7
26 Svatek R, Karakiewicz PI, Shulman M, Karam J, Perrotte P, Benaim E. Pre- treatment nomogram for disease-specific survival of patients with chemotherapy- naive androgen independent prostate cancer. Eur Urol 2006; 49: 666-74
27 Briganti A, Gallina A, Suardi N et al. A nomogram is more accurate than a regression tree in predicting lymph node invasion in prostate cancer. BJU Int 2008; 101: 556-60
28 Gallina A, Chun FK, Suardi N et al. Comparison of stage migration patterns between Europe and the USA: an analysis of 11 350 men treated with radical prostatectomy for prostate cancer. BJU Int 2008; 101: 1513-8
29 Terzolo M, Angeli A, Fassnacht M et al. Adjuvant mitotane treatment for adrenocortical carcinoma. N Engl J Med 2007; 356: 2372-80
Correspondence: Pierre I. Karakiewicz, Cancer Prognostics and Health Outcomes Unit, University of Montreal Health Center (CHUM), 1058 rue St-Denis, Montreal, Quebec, Canada, H2X 3J4.
e-mail: pierre.karakiewicz@umontreal.ca
Abbreviations: ACC, adrenocortical carcinoma; SEER, Surveillance Epidemiology and End Results; ICD-O, International Classification of Diseases for Oncology; CSM, cancer-specific mortality; ACM, all-cause mortality.