ELSEVIER
Surgery
journal homepage: www.elsevier.com/locate/surg
SURGERY
MOWEMBER 2018
$ -
==
Validated predictive model for treatment and prognosis of adrenocortical carcinoma
Check for updates
Samuel M. Zuber, MDa,b,*, Kristine Kuchta, MScc, Simon A. Holoubek, DOd, Amna Khokar, MDe, Tricia Moo-Young, MDa,b, Richard A. Prinz, MDa,b, David J. Winchester, MDf
ª Department of Surgery, NorthShore University Health System, Evanston, IL
b Department of Surgery, University of Chicago Medicine, Chicago, IL
Bioinformatics and Research Core, NorthShore University Health Evanston, IL
d Division of Endocrine Surgery, University of Wisconsin School of Medicine and Public Health, Madison, WI
e Department of Surgery, John H. Stroger Jr. Cook County Hospital, Chicago, IL Department of Surgery, City of Hope, Zion, IL
ARTICLE INFO
Article history: Accepted 17 August 2023 Available online 11 November 2023
ABSTRACT
Background: Adrenocortical carcinoma has a poor prognosis and multiple clinical, pathological, and treatment variables. Currently, we lack a prognostic and treatment calculator to determine the survival and efficacy of adjuvant chemoradiation. We aimed to validate a calculator to assess prognosis and treatment.
Methods: We searched the National Cancer Database to identify patients with adrenocortical carcinoma surgically treated from 2004 to 2020 and randomly allocated them into a training (80%) or validation set (20%). We analyzed the variables of age; sex; Charlson Comorbidity Index; insurance status; tumor size; pathologic tumor, node, and metastasis categories; surgical margins; and use of chemotherapy and radi- ation therapy. We used Cox regression prediction models and bootstrap coefficients to generate a mathe- matical model to predict 5- and 10-year overall survival. After using the area under the curve analysis to assess the model’s performance, we compared overall survival in the training and validation sets.
Results: Multivariable analysis of the 3,480 patients included in the study revealed that all variables were significant except sex (P < . 05) and incorporated into a mathematical model. The area under the curve for 5- and 10-year overall survival was 0.68 and 0.70, respectively, for the training set and 0.70 and 0.72, respectively, for the validation set. For the bootstrap coefficients, the 5- and 10-year overall survival was 6.4% and 4.1%, respectively, above the observed mean.
Conclusion: Our model predicts the overall survival of patients with adrenocortical carcinoma based on clinical, pathologic, and treatment variables and can assist in individualizing treatment.
@ 2023 Elsevier Inc. All rights reserved.
Introduction
Adrenocortical carcinoma (ACC) is a rare cancer with an inci- dence of 1 to 2 cases per million people globally.1,2 The prognosis for these patients remains poor, with an associated 5-year overall
survival (OS) rate ranging between 16% and 64% based on stage.2-6 The most common clinical presentation includes an incidental mass seen on cross-sectional imaging, and approximately 60% of patients exhibit clinical signs or symptoms associated with hor- mone excess.7-10 Surgical resection with an en bloc removal of the mass and clear surgical margins offers ACC patients the best overall prognosis. Presenting with surgically resectable disease is, how- ever, rare.6,11 Additionally, even with this approach, local recur- rence rates can be as high as 90%12,13 With such poor oncologic outcomes, adjuvant treatment and better treatment stratification are warranted.14,15
At present, adjuvant radiation has been indicated for patients with R1 or R2 margins after surgical resection and has been shown
Abstract accepted for oral presentation at Central Surgical Association, Cleve- land, OH, June 8-10.
* Reprint requests: Samuel M. Zuber, MD, Department of Surgery, NorthShore University Health System, 2650 Ridge Avenue, Walgreen Suite 2507, Evanston, IL 60201.
E-mail address: zuber.sam@gmail.com (S.M. Zuber);
Twitter: @samzuberMD
https://doi.org/10.1016/j.surg.2023.08.047
to demonstrate improved OS when compared to surgery alone.13,16 In addition, adjuvant radiation has been shown to reduce the risk of recurrence.17 In a multivariable analysis of 6 studies using a pooled analysis, Zhu et al found significantly favorable outcomes for adjuvant radiotherapy regarding local recurrence-free survival, with an odds ratio of 4.08.16 Systemic therapies are still relatively limited.13 The adrenolytic activity of mitotane targets the inner adrenal cortex and is currently the only drug approved by the Food and Drug Administration for the treatment of ACC.12,18 Although mitotane does not necessarily reduce the risk of recurrence, it has been shown to delay recurrence and prolong disease-free survival (DFS) and OS compared to surgery alone.12,15,19 In a prospective randomized controlled trial for patients with unresectable or advanced ACC, the First International Randomized Trial in Locally Advanced and Metastatic Adrenocortical Carcinoma Treatment group was able to demonstrate the efficacy of etoposide, doxoru- bicin, and cisplatin with mitotane. However, the difference in DFS was only several months, and there was no difference in OS.20 Furthermore, the sample sizes in these studies were small, with the number of patients ranging between 11 and 72, with the exception being the First International Randomized Trial in Locally Advanced and Metastatic Adrenocortical Carcinoma Treatment study, in which 304 patients underwent randomization.21-23
Prediction models and prognostic calculators have been created for ACC.24-29 The purpose of these models is to act as a guide for providers and patients regarding clinical decision-making for pa- tients with ACC. These models were derived from varied databases, including the Surveillance, Epidemiology, and End Results (SEER) database and the European Network for Study of Adrenal Tumors registry. Compared to the present study of the American College of Surgeons National Cancer Database (NCDB), studies using the SEER database included fewer patients and did not include co-morbidity data.3º Previous prognostic models seldomly incorporated adju- vant chemotherapy and radiation information in their develop- ment, and none included insurance status. To better facilitate individualized treatment and tailored discussions with patients, we aimed to study a relatively large dataset to create a prognostic calculator predicting OS based on clinical, socioeconomic, patho- logic, and treatment variables.
Methods
Institutional Review Board approval was not necessary for this study. We queried the American College of Surgeons National Cancer Database (2020; NCDB) from 2004 to 2020 to identify all adult patients with ACC. Patients were identified using Interna- tional Classification of Diseases code 8370. The inclusion criteria were ACC patients who had surgery, including those who had other tumors and were 18 years or older (Figure 1). We collected patient demographic, clinical, and surgical data, including age; sex; race; insurance status; Charlson-Deyo Comorbidity Index (CCI); tumor size; pathologic tumor (T), node (N), and metastasis (M) categories; lymphadenectomy; surgical approach; surgical margins;
| Included | Excluded | Inclusion/Exclusion Criteria |
|---|---|---|
| 177,188 | Patients in NCDB 2020 Other Endocrine Database | |
| 169,379 | ||
| 7,809 | Adrenal Gland site | |
| 2,904 | ||
| 4,905 | Malignant Behavior and Histology Code 8370 | |
| 1,425 | ||
| 3,480 | Surgical Patients |
neoadjuvant therapy; and adjuvant chemotherapy and radiation (Table I).
We randomly divided the patients into training (nt) and vali- dation (ny) sets with an 80%/20% allocation, which allowed for a robust training group while providing an adequate sample size for the validation group. We reported descriptive statistics, including the frequency as the percentage and mean with the SD, for both the nt and ny cohorts. We compared the cohorts using the x2 analysis and independent samples t-tests. For the nt cohort, we performed univariable analysis using Cox regression to identify associations between patient variables and OS and reported the results as haz- ard ratios with 95% CIs. We then performed a multivariable analysis of the nt cohort to identify patient and clinical factors for use in our clinical tool.
Statistical Analysis
We performed all statistical analyses using SAS version 9.4 (SAS Institute, Cary, NC) with 2-tailed tests. We used variables with P < . 05 on multivariable analysis to create a mathematical model to predict 5- and 10-year OS in the training set. We calculated the baseline 5- and 10-year OS rates for the Cox regression model using the PHREG procedure and the BASELINE function in SAS, with the baseline corresponding to estimated survival rates when all model covariates were set to the reference category. We obtained final model parameters using 100 bootstrap samples, which provided a sample size that attained a normal distribution of the coefficients, with the mean value chosen as our final coefficient. We used cali- bration plots, receiver operating characteristic curve analysis, and area under the curve (AUC) analysis to assess model performance (Figure 2), considering that the closer the AUC was to 1.0, the higher the model’s performance. We then applied the final model to the ny cohort using calibration plots, receiver operating curves, and AUC analysis to assess model performance. We created calibration plots by dividing patients into risk vigintiles for the nt cohort and risk deciles for the ny cohort and calculated observed and predicted 5- and 10-year survival for each cohort using Kaplan-Meier methods (Figure 2).
Results
Our query of the NCDB retrieved 177,188 patients, of whom we identified 3,480 surgical patients with tumors in the adrenal gland (histology code 8370) diagnosed between 2004 and 2020 for analysis (Figure 1). The mean age was 54 + 15 years, with 39% under 50 years and 62% female. Table I shows their complete descriptive statistics. Univariable analysis of the nt and ny sets revealed that all variables were significant except sex, race, adjuvant chemotherapy, and radiation (Table II). Multivariable analysis identified age; CCI; insurance status; tumor size; pathologic T, N, and M categories; surgical margins; adjuvant radiation; and adjuvant chemotherapy as significant variables. Based on these results, we incorporated these variables into the predictive model. As sex, lymphadenec- tomy, and neoadjuvant therapy were not significant, we did not incorporate them into the model.
We created a mathematical Cox regression model with the factors identified as significant on multivariable analysis to predict 5- and 10-year survival (YS; Table II). The AUC for the ny cohort for 5-YS was 0.68, and for 10-YS was 0.70 (Figure 2). When we repeated this procedure with the ny cohort, the AUC was 0.70 for 5-YS and 0.72 for 10-YS. The baseline 5-YS was 83.6%, and the baseline 10-YS rate was 75.6%. The coefficients that were generated for each var- iable are displayed in Table III. We internally validated the model by performing 100 bootstrap coefficients for the ny and ny cohorts (Tables IV and V). The 5-year OS predicted mean was 6.4% above the
| Variable | All patients | Training set | Validation set | P value |
|---|---|---|---|---|
| N (%) | N (%) | N (%) | - | |
| Total no. of patients | 3,480 | 2,784 | 696 | - |
| Age, y, mean + SD | 54 ± 15 | 54 ± 15 | 54 ±16 | .6482 |
| Age, y, n (%) | .8495 | |||
| ≤50 | 1368 (39.3) | 1086 (39.0) | 282 (40.5) | |
| 51-60 | 840 (24.1) | 678 (24.4) | 162 (23.3) | |
| 61-70 | 747 (21.5) | 602 (21.6) | 145 (20.8) | |
| >70 | 525 (15.1) | 418 (15.0) | 107 (15.4) | |
| Sex, n (%) | .7276 | |||
| Male | 1340 (38.5) | 1068 (38.4) | 272 (39.1) | |
| Female | 2140 (61.5) | 1716 (61.6) | 424 (60.9) | |
| Race, n (%) | .4580 | |||
| White | 2782 (79.9) | 2234 (80.2) | 548 (78.7) | |
| Black | 323 (9.3) | 253 (9.1) | 70 (10.1) | |
| Hispanic | 227 (6.5) | 174 (6.3) | 53 (7.6) | |
| Asian/Pacific islander | 92 (2.6) | 78 (2.8) | 14 (2.0) | |
| Other/unknown | 56 (1.6) | 45 (1.6) | 11 (1.6) | |
| Insurance, n (%) | .9717 | |||
| Private | 1944 (55.9) | 1557 (55.9) | 387 (55.6) | |
| Medicare | 966 (27.8) | 769 (27.6) | 197 (28.3) | |
| Medicaid/other government | 339 (9.7) | 271 (9.7) | 68 (9.8) | |
| Unknown/uninsured | 231 (6.6) | 187 (6.7) | 44 (6.3) | |
| Charlson Comorbidity Index, n (%) | .8980 | |||
| 0 | 2593 (74.5) | 2074 (74.5) | 519 (74.6) | |
| 1 | 634 (18.2) | 505 (18.1) | 129 (18.5) | |
| ≥2 | 253 (7.3) | 205 (7.4) | 48 (6.9) | |
| Tumor size, cm, mean ± SD | 11.9 ±9.9 | 12.0 ± 9.9 | 11.4 ± 10.2 | .0089 |
| Tumor size, cm | .3300 | |||
| <5.0 | 409 (11.8) | 321 (11.5) | 88 (12.6) | |
| 5.0-9.9 | 1148 (33.0) | 903 (32.4) | 245 (35.2) | |
| ≥10.0 | 1770 (50.9) | 1435 (51.5) | 335 (48.1) | |
| Unknown | 153 (4.4) | 125 (4.5) | 28 (4.0) | |
| Pathologic T category, n (%) | .0040 | |||
| T1-2 | 1675 (48.1) | 1339 (48.1) | 336 (48.3) | |
| T3 | 958 (27.5) | 736 (26.4) | 222 (31.9) | |
| T4 | 531 (15.3) | 444 (15.9) | 87 (12.5) | |
| Unknown | 316 (9.1) | 265 (9.5) | 51 (7.3) | |
| Pathologic N category, n (%) | .8041 | |||
| N0 | 1010 (29.0) | 804 (28.9) | 206 (29.6) | |
| N1 | 236 (6.8) | 186 (6.7) | 50 (7.2) | |
| Unknown | 2234 (64.2) | 1794 (64.4) | 440 (63.2) | |
| Metastasis, n (%) | 584 (16.8) | 479 (17.2) | 105 (15.1) | .1808 |
| Lymphadenectomy, n (%) | 817 (23.5) | 658 (23.6) | 159 (22.8) | .6600 |
| Surgical approach, n (%) | .5586 | |||
| Open | 1361 (39.1) | 1102 (39.6) | 259 (37.2) | |
| Laparoscopic | 495 (14.2) | 398 (14.3) | 97 (13.9) | |
| Robotic | 219 (6.3) | 170 (6.1) | 49 (7.0) | |
| Unknown | 1405 (40.4) | 1114 (40.0) | 291 (41.8) | |
| Surgical margins, n (%) | .7463 | |||
| R0 | 2361 (67.8) | 1881 (67.6) | 480 (69.0) | |
| R1 | 658 (18.9) | 529 (19.0) | 129 (18.5) | |
| Unknown | 461 (13.2) | 374 (13.4) | 87 (12.5) | |
| Neoadjuvant therapy, n (%) | 86 (2.5) | 72 (2.6) | 14 (2.0) | .3824 |
| Adjuvant radiation, n (%) | 644 (18.5) | 516 (18.5) | 128 (18.4) | .9304 |
| Adjuvant chemotherapy, n (%) | 1247 (35.8) | 990 (35.6) | 257 (36.9) | .5018 |
T, tumor status; T1, tumor ≤5cm; T2, tumor >5 cm; T3, tumor any size growing into fat surrounding adrenal; T4, tumor any size growing into nearby organs; N, lymph node status; N0, no lymph node involvement; N1, positive lymph node involvement; R0, margin free of microscopic disease; R1, positive microscopic margin.
observed mean, and the 10-year OS predicted mean was 4.1% above the observed mean.
Table VI demonstrates the model’s functioning. Constant vari- ables in this table include White male patients with a CCI of 0 with private insurance who underwent an adrenalectomy with T1-2, NO, M0, and negative margins. Age, adjuvant chemotherapy, and radi- ation were varied to demonstrate the impact on 5-year OS. The white boxes indicate a 5% or less difference, light gray boxes a 5.01% to 7.5% difference, and dark gray boxes a >7.5% difference in 5-year OS.
Discussion
Adrenocortical carcinoma is rare and has a poor prognosis, with a 5-YS of 16% to 64% based on the stage.2-6 En bloc surgical resection with R0 margins is the gold standard for treatment. Adjuvant treatment includes mitotane and other regimens along with external beam radiation.12,13,16-19 Although the 5-year OS and DFS for ACC can be extended with these treatment modalities, recurrence rates are still high, and these therapies are not without
Training Set 5-Year Survival
Validation Set 5-Year Survival
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
Observed Probability
0.2
0.2
0.0
0.0
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
Training Set 10-Year Survival
Validation Set 10-Year Survival
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
0.0
0.2
0.4
0.6
0.8
1.0
0.0
0.2
0.4
0.6
0.8
1.0
Predicted Probability
risk.12,13,15,19 Because of the rarity of ACC, research into it is limited by the number of patients that can be included in a study sample. Prognostic calculators and prediction models are tools that can help individualize patient care. Calculators are used for a variety of cancers, including thyroid cancer, colon cancer, and pancreatic cancer.31-33 For ACC, several prognostic models have been devel- oped and published.24-29 The prognostic calculator reported by Kim et al was based on an analysis of 148 patients from 13 high- volume centers, whereas our calculator was based on an analysis of almost 3,500 patients from over 1,500 Commission on Cancer- accredited centers that report to the NCDB.25 The Kim et al retro- spective, multi-institutional study included tumor size and nodal and margin status to generate an OS nomogram, whereas we also included age, insurance status, CCI, presence of metastatic disease, and adjuvant chemotherapy and radiation in our model. In the SEER study, age and pathologic T category had a significant effect on OS in the univariate analysis but not in the multivariable analysis or the prediction model, which could be explained by its smaller sample size.
Li et al utilized the SEER database from 1973 to 2015 to analyze 788 patients with ACC and predict their OS and cancer-specific survival (CSS) with a nomogram.26 Prognostic variables for OS and CSS included age at diagnosis, year of diagnosis, histologic grade, historic stage (broken down into localized, regional, and distant disease), and chemotherapy. They developed nomograms to
predict OS and CSS at 1, 3, and 5 years and internally validated the models with logistic regression models and 200 bootstrap co- efficients. The C-index was 0.68 and 0.66 for predicting 3-year OS and 3-year CSS, respectively. Increasing age and metastatic disease predicted worse OS, consistent with our findings. Although insur- ance status is listed as a variable in the SEER database, it was not included in their study. In addition to a 5-year OS, our prediction model predicted a 10-year OS. Whereas Li et al found that patients who had undergone chemotherapy had worse OS, our study demonstrated improved OS with adjuvant chemotherapy.
Kong et al performed a retrospective analysis of surgical adult patients over 20 years with ACC to generate a prognostic nomo- gram predicting OS.27 They divided the 718 patients listed in the SEER database between 1988 and 2015 into a training set with 404 patients and an internal validation set with 318 patients. In contrast, our study created a training set with 2,784 patients and an internal validation set with 696 patients. The variables that they analyzed in the nomogram were age and TNM categories, which produced a better prediction model with the separation of TNM categories from the stage, which we also found. Kong et al then used their model to predict 1-, 3-, and 5-year OS with their training set AUC at 0.72 and their validation set AUC at 0.67 (variation of 0.5). For our calculator, the AUC variation between the training and validation sets was 0.2, suggesting ours may be more generalizable to an external population. The Kong et al model was then externally
| Variable | Univariable analysis | Multivariable analysis | ||
|---|---|---|---|---|
| HR (95% CI) | P value | HR (95% CI) | P value | |
| Age, y | ||||
| 51-60 vs ≤50 | 1.27 (1.13-1.44) | <. 0001 | 1.23 (1.08-1.39) | .0012 |
| 61-70 vs ≤50 | 1.52 (1.35-1.72) | < . 0001 | 1.47 (1.27-1.71) | < . 0001 |
| >70 vs ≤50 | 2.10 (1.84-2.39) | < . 0001 | 1.80 (1.50-2.15) | < . 0001 |
| Sex, female vs male | 1.01 (0.92-1.11) | .8173 | 0.97 (0.88-1.07) | .5289 |
| Race | ||||
| Black vs White | 0.95 (0.81-1.12) | .5567 | - | - |
| Hispanic vs White | 1.02 (0.85-1.24) | .8178 | - | - |
| Asian/Pacific Islander vs White | 0.85 (0.62-1.16) | .2965 | - | - |
| Other/unknown vs White | 1.42 (1.02-1.99) | .0375 | - | - |
| Insurance | ||||
| Medicare vs private | 1.58 (1.43-1.75) | < . 0001 | 1.18 (1.02-1.37) | .0243 |
| Medicaid/other government vs Private | 1.28 (1.09-1.51) | .0026 | 1.35 (1.14-1.59) | .0004 |
| Unknown/uninsured vs private | 1.25 (1.05-1.49) | .0142 | 1.08 (0.90-1.30) | .4087 |
| Charlson Comorbidity Index | ||||
| 1 vs 0 | 1.43 (1.28-1.60) | < . 0001 | 1.40 (1.24-1.57) | <. 0001 |
| ≥2 vs 0 | 1.57 (1.33-1.85) | <. 0001 | 1.37 (1.15-1.62) | .0003 |
| Tumor size | ||||
| 5.0-9.9 vs <5.0 cm | 1.24 (1.05-1.47) | .0106 | 1.28 (1.08-1.51) | .0047 |
| ≥10.0 vs <5.0cm | 1.45 (1.23-1.70) | < . 0001 | 1.29 (1.10-1.52) | .0020 |
| Unknown vs <5.0 | 1.52 (1.19-1.95) | .0008 | 1.26 (0.97-1.63) | .0812 |
| Pathologic T category | ||||
| T3 vs T1-2 | 1.83 (1.64-2.03) | <. 0001 | 1.54 (1.37-1.72) | < . 0001 |
| T4 vs T1-2 | 2.20 (1.95-2.49) | < . 0001 | 1.58 (1.38-1.81) | <. 0001 |
| Unknown vs T1-2 | 1.31 (1.06-1.62) | .0131 | 1.03 (0.83-1.28) | .7612 |
| Pathologic N category | ||||
| N1 vs N0 | 2.56 (2.15-3.04) | < . 0001 | 1.67 (1.39-2.01) | <. 0001 |
| Unknown vs N0 | 1.00 (0.90-1.11) | .9734 | 0.96 (0.84-1.10) | .5853 |
| Metastasis, yes vs no | 2.94 (2.64-3.27) | < . 0001 | 2.60 (2.31-2.93) | <. 0001 |
| Lymphadenectomy, yes vs no | 1.28 (1.15-1.42) | < . 0001 | 0.94 (0.79-1.10) | .4252 |
| Surgical approach | ||||
| Laparoscopic vs open | 0.91 (0.78-1.06) | .2095 | - | - |
| Robotic vs open | 0.87 (0.69-1.09) | .2195 | - | - |
| Unknown vs open | 1.09 (0.98-1.20) | .1082 | - | - |
| Surgical margins | ||||
| R1 vs R0 | 2.14 (1.92-2.39) | < . 0001 | 1.84 (1.64-2.07) | <. 0001 |
| Unknown vs R0 | 1.79 (1.57-2.03) | < . 0001 | 1.70 (1.49-1.94) | <. 0001 |
| Neoadjuvant therapy, yes vs no | 1.42 (1.07-1.89) | .0147 | 0.86 (0.63-1.16) | .3163 |
| Adjuvant Radiation, yes vs no | 0.92 (0.82-1.04) | .2038 | 0.81 (0.72-0.93) | .0017 |
| Adjuvant chemotherapy, yes vs no | 1.03 (0.94-1.14) | .5195 | 0.90 (0.81-1.00) | .0401 |
HR, hazard ratio; T, tumor status; T1, tumor ≤5cm; T2, tumor >5 cm; T3, tumor any size growing into fat surrounding adrenal; T4, tumor any size growing into nearby organs; N, lymph node status; N0, no lymph node involvement; N1, positive lymph node involvement; R0, margin free of microscopic disease; R1, positive microscopic margin.
validated to predict OS using data from 2 external databases with an AUC of 0.81 and 0.73. The median OS was 22 and 5 months for the low- and high-risk groups, respectively, when all 886 patients were plotted. Although Kong et al included externally validated data, they examined fewer patients and included fewer variables in their model than ours. Their model also only predicts a 5-year OS, whereas our prediction model predicts both 5- and 10-year OS.
The prediction model created by Ettaieb et al analyzed 160 pa- tients listed in the multi-institutional Dutch Adrenal Network be- tween 2004 and 2013.28 They included age, modified European Network for the Study of Adrenal Tumors score, and resection margin status as variables in the prediction model after demon- strating their significance on multivariable analysis. They then computed a risk score that they plotted into a Kaplan-Meier esti- mate to determine survival at 1, 2, and 5 years but not out to 10 years, as we did for our prediction model. In addition, our group was able to include additional variables with demonstrated sig- nificance on multivariable analysis, including comorbidity index and adjuvant treatment variables.
Elhassan and the European Network for the Study of Adrenal Tumors (ENSAT), a multicenter group, conducted a retrospective analysis of 942 patients who had undergone adrenalectomy for
ACC.29 They examined the variables of ENSAT stage, grading, resection status, age, and tumor or hormone-related symptoms and generated a scoring system after dividing the patients into 4 groups. Their endpoints were progression-free survival and DSS. They demonstrated that the S-GRAS scoring system was superior for predicting progression-free survival and DSS compared to the ENSAT stage or Ki-67 index alone. Nevertheless, their scoring system did not differentiate between each variable calculated in the final score, unlike our calculator, which generates a cumu- lative risk score from each variable’s coefficient that is related to its hazard ratio. They did not find age to have any significance in their univariate analysis, whereas increasing age in our model predicted worse OS, and they included information on tumor grade and functionality as 2 variables that our model does not include.
The findings of our larger study offer guidance for the use of adjuvant therapy, which impacts 5- and 10-year OS and our anal- ysis included insurance status as an independent predictor of risk. Patients with private insurance have better OS than those with Medicare or Medicaid. While acknowledging that assessing socio- economic status is a sensitive topic that cannot be reflected in a single variable, we found it interesting that our findings regarding
| Variable | Coefficient* | SE | HR (95% CI) |
|---|---|---|---|
| Age, y | |||
| ≤50 | 0 | - | Reference |
| 51-60 | 0.14127 | 0.07005 | 1.15 (1.00-1.32) |
| 61-70 | 0.49386 | 0.07158 | 1.64 (1.42-1.89) |
| >70 | 0.68699 | 0.07818 | 1.99 (1.71-2.32) |
| Insurance status | |||
| Private | 0 | - | Reference |
| Medicare | 0.15836 | 0.08376 | 1.17 (0.99-1.38) |
| Medicaid/other government | 0.24990 | 0.09577 | 1.28 (1.06-1.55) |
| Unknown/uninsured | 0.12609 | 0.10554 | 1.13 (0.92-1.40) |
| Charlson Comorbidity Index | |||
| 0 | 0 | - | Reference |
| 1 | 0.38663 | 0.06566 | 1.47 (1.29-1.67) |
| ≥2 | 0.27987 | 0.09774 | 1.32 (1.09-1.60) |
| Tumor size, cm | |||
| <5.0 cm | 0 | - | Reference |
| 5.0-9.9 cm | 0.29020 | 0.09871 | 1.34 (1.10-1.62) |
| ≥10.0 cm | 0.36339 | 0.09465 | 1.44 (1.19-1.73) |
| Unknown | 0.29416 | 0.14455 | 1.34 (1.01-1.78) |
| Pathologic T category | |||
| T1-2 | 0 | - | Reference |
| T3 | 0.39315 | 0.06496 | 1.48 (1.30-1.68) |
| T4 | 0.41696 | 0.07532 | 1.52 (1.31-1.76) |
| Unknown | 0.01642 | 0.12004 | 1.02 (0.80-1.29) |
| Pathologic N category | |||
| N0 | 0 | - | Reference |
| N1 | 0.54469 | 0.10157 | 1.72 (1.41-2.10) |
| Unknown | -0.00272 | 0.06081 | 1.00 (0.89-1.12) |
| Metastasis | |||
| No | 0 | - | Reference |
| Yes | 0.96163 | 0.06588 | 2.62 (2.30-2.98) |
| Surgical margins | |||
| R0 | 0 | - | Reference |
| R1 | 0.55881 | 0.06761 | 1.75 (1.53-2.00) |
| Unknown | 0.51189 | 0.0746 | 1.67 (1.44-1.93) |
| Adjuvant radiation | |||
| No | 0.20049 | 0.074 | 1.22 (1.06-1.41) |
| Yes | 0 | - | Reference |
| Adjuvant chemotherapy | |||
| No | 0.08362 | 0.0583 | 1.09 (0.97-1.22) |
| Yes | 0 | - | Reference |
SE, standard error; HR, hazard ratio; T, tumor status; T1, tumor ≤5cm; T2, tumor >5cm; T3, tumor any size growing into fat surrounding adrenal; T4, tumor any size growing into nearby organs; N, lymph node status; N0, no lymph node involvement; N1, positive lymph node involvement; R0, margin free of microscopic disease; R1, positive microscopic margin. * Baseline at reference categories: 5-year survival = 83.0540%, 10-year survival = 74.8396%.
its significance corroborate those of Holoubek et al in their assessment of the treatment differences for ACC by insurance status.34
Compared to a nomogram, our calculator allows for a more specific calculation for an individual. Nomograms have a set number of predictions that they can provide. On the other hand, our calculator allows for a more continuous distribution and can pro- vide patients with personalized prognostic information. For example, consider a 65-year-old with private insurance and no co- morbidities with a 9.5 cm T2N0M0 ACC treated surgically with R0 margins. With surgery alone, the patient’s 5-year OS would be 61%, but with additional adjuvant chemotherapy and radiation would increase to 69%. The same patient with positive margins would have a 5-year OS of 42%, which adjuvant chemotherapy and radi- ation therapy would increase to 52%.
Study limitations
Our prognostic calculator using the robust NCDB database has several limitations. Our NCDB analysis was retrospective and only
accounted for OS and not cancer-specific survival, whereas the SEER database provides data for both. The NCDB also does not collect data on tumor functionality and grading using the Ki-67 index, the addition of which would have further enhanced our calculator. Boffa et al stress that one of the key distinctions between the NCDB and SEER is that the NCDB collects data from hospitals with Commission on Cancer accreditation, whereas the SEER database is based on geography and aims to reflect a more diverse population.30 Moreover, the NCDB does not contain the pathologic data of a sizeable portion of patients such that 4% of tumors are of unknown size, 9% are missing T-category data, 64% are missing N-category data, and 13% are of unknown margin status (Table I). As more data continue to be captured by the NCDB and other data- bases, we hope to adjust our calculator model to incorporate these changes and provide the most detailed prediction model possible for ACC.
In conclusion, we validated an interactive prognostic calculator for ACC. Our model reliably predicts 5- and 10-year OS based on clinicopathologic and treatment variables and can be used to help guide and individualize therapy.
| Vigintile | N | Risk score | 5-year survival | 10-year survival | ||||
|---|---|---|---|---|---|---|---|---|
| Predicted | Observed | Predicted | Observed | |||||
| Mean ± SD | Min | Max | Mean | Mean | Mean | Mean | ||
| 1 | 135 | 0.34 ± 0.11 | 0.00 | 0.48 | 77.68% | 78.80% | 67.41% | 72.35% |
| 2 | 150 | 0.54 ± 0.03 | 0.48 | 0.56 | 73.49% | 75.61% | 61.79% | 67.16% |
| 3 | 132 | 0.62 ± 0.02 | 0.56 | 0.65 | 71.62% | 71.76% | 59.36% | 53.19% |
| 4 | 144 | 0.72 ± 0.03 | 0.65 | 0.76 | 69.17% | 63.99% | 56.22% | 46.89% |
| 5 | 136 | 0.81 ± 0.02 | 0.76 | 0.86 | 66.79% | 57.70% | 53.23% | 45.45% |
| 6 | 138 | 0.91 ± 0.03 | 0.86 | 0.96 | 64.02% | 41.44% | 49.82% | 33.35% |
| 7 | 139 | 0.99 ± 0.02 | 0.96 | 1.04 | 61.67% | 56.07% | 46.99% | 47.35% |
| 8 | 139 | 1.06 ± 0.02 | 1.04 | 1.09 | 59.62% | 56.60% | 44.57% | 36.37% |
| 9 | 140 | 1.13 ± 0.02 | 1.09 | 1.17 | 57.40% | 50.67% | 42.01% | 38.66% |
| 10 | 139 | 1.20 ± 0.02 | 1.17 | 1.24 | 55.25% | 51.68% | 39.57% | 31.59% |
| 11 | 139 | 1.28 ± 0.03 | 1.24 | 1.32 | 52.62% | 53.87% | 36.67% | 35.55% |
| 12 | 139 | 1.35 ± 0.02 | 1.32 | 1.40 | 50.12% | 52.38% | 33.99% | 22.52% |
| 13 | 140 | 1.42 ± 0.02 | 1.40 | 1.46 | 47.58% | 43.86% | 31.34% | 23.88% |
| 14 | 139 | 1.51 ± 0.02 | 1.47 | 1.54 | 44.61% | 35.18% | 28.33% | 22.44% |
| 15 | 139 | 1.58 ±0.02 | 1.54 | 1.63 | 41.77% | 34.31% | 25.57% | 18.42% |
| 16 | 139 | 1.67 ± 0.02 | 1.63 | 1.72 | 38.53% | 32.70% | 22.54% | 26.57% |
| 17 | 140 | 1.77 ± 0.03 | 1.72 | 1.82 | 34.95% | 28.59% | 19.36% | 17.37% |
| 18 | 139 | 1.91 ± 0.05 | 1.83 | 2.00 | 29.98% | 22.70% | 15.25% | 10.63% |
| 19 | 140 | 2.10 ± 0.06 | 2.00 | 2.21 | 23.17% | 20.84% | 10.21% | 14.04% |
| 20 | 138 | 2.46 ± 0.20 | 2.21 | 3.12 | 12.90% | 9.93% | 4.32% | 4.83% |
Min, minimum; Max, maximum.
| Decile | N | Risk score | 5-year survival | 10-year survival | ||||
|---|---|---|---|---|---|---|---|---|
| Predicted | Observed | Predicted | Observed | |||||
| Mean ± SD | Min | Max | Mean | Mean | Mean | Mean | ||
| 1 | 69 | 0.40 ± 0.12 | 0.08 | 0.56 | 76.40% | 66.02% | 65.70% | 49.63% |
| 2 | 69 | 0.63 ± 0.06 | 0.56 | 0.73 | 71.37% | 74.97% | 59.04% | 66.53% |
| 3 | 71 | 0.84 ± 0.05 | 0.74 | 0.94 | 66.13% | 64.68% | 52.41% | 52.55% |
| 4 | 69 | 1.01 ± 0.04 | 0.95 | 1.09 | 61.23% | 44.25% | 46.47% | 36.87% |
| 5 | 69 | 1.15 ± 0.04 | 1.09 | 1.21 | 56.69% | 56.39% | 41.20% | 42.83% |
| 6 | 71 | 1.28 ± 0.04 | 1.21 | 1.35 | 52.55% | 46.48% | 36.61% | 30.03% |
| 7 | 69 | 1.44 ± 0.05 | 1.35 | 1.54 | 46.79% | 34.98% | 30.54% | 28.06% |
| 8 | 70 | 1.64 ± 0.05 | 1.55 | 1.73 | 39.75% | 31.26% | 23.68% | 16.23% |
| 9 | 70 | 1.87 ± 0.09 | 1.75 | 2.07 | 31.27% | 21.91% | 16.33% | 14.09% |
| 10 | 69 | 2.37 ± 0.25 | 2.07 | 3.09 | 15.46% | 13.19% | 5.82% | 0.00% |
Min, minimum; Max, maximum.
| Adjuvant chemo-XRT+ | Age, y* | |||||||
|---|---|---|---|---|---|---|---|---|
| ≤50 | 51-60 | 61-70 | >70 | |||||
| No | Yes | No | Yes | No | Yes | No | Yes | |
| Tumor size (Cm) | ||||||||
| <5.0 | 78.9 | 83.6 | 75.7 | 81.0 | 68.8 | 75.3 | 64.8 | 72.0 |
| 5.0-9.9 | 73.1 | 78.9 | 69.2 | 75.7 | 60.9 | 68.7 | 56.2 | 64.5 |
| ≥10.0 | 71.4 | 77.5 | 67.3 | 74.1 | 58.6 | 66.8 | 53.9 | 62.6 |
Chemo-XRT, chemotherapy and radiation therapy.
* Prediction for a male with private insurance; no comorbidities; and T1-2, NO, MO, and RO tumor who had un- dergone adjuvant chemotherapy and radiation therapy.
White boxes indicate <5% difference in overall survival between chemo-XRT and no chemo-XRT, light gray boxes between >5% and ≤7.5%, and dark gray boxes >7.5% and ≤10%.
Funding/Support
This research did not receive any specific funding from any agencies in the public, commercial, or not-for-profit areas.
Conflict of interest/Disclosure
The authors have no conflicts of interests or disclosures to report.
References
1. Kerkhofs TM, Verhoeven RH, Van der Zwan JM. Adrenocortical carcinoma: a population-based study on incidence and survival in the Netherlands since 1993. Eur J Cancer. 2013;49:2579-2586.
2. Kebebew E, Reiff E, Duh Q-Y, Clark OH, McMillan A. Extent of disease at pre- sentation and outcome for adrenocortical carcinoma: have we made progress? World J Surg. 2006;30:872-878.
3. Sturgeon C, Shen WT, Clark OH, Duh QY, Kebebew E. Risk assessment in 457 adrenal cortical carcinomas: how much does tumor size predict the likelihood of malignancy? J Am Coll Surg. 2006;202:423-430.
4. Fassnacht M, Johanssen S, Fenske W, et al. Improved survival in patients with stage II adrenocortical carcinoma followed up prospectively by specialized centers. J Clin Endocrinol Metab. 2010;95:4925-4932.
5. American Cancer Society. Survival rates for adrenal cancer. https://www. cancer.org/cancer/adrenal-cancer/detection-diagnosis-staging/survival-by- stage.html. Accessed 5 January 2023.
6. Bilimoria KY, Shen WT, Elaraj D, et al. Adrenocortical carcinoma in the United States. Cancer. 2008;113:3130-3136.
7. Allolio B, Fassnacht M. Clinical review: adrenocortical carcinoma: clinical up- date. J Clin Endocrinol Metab. 2006;91:2027-2037.
8. Fassnacht M, Allolio B. Clinical management of adrenocortical carcinoma. Best Pract Res Clin Endocrinol Metab. 2009;23:273-289.
9. Luton JP, Cerdas S, Billaud L, et al. Clinical features of adrenocortical carcinoma, prognostic factors, and the effect of mitotane therapy. N Engl J Med. 1990;322: 1195-1201.
10. Schteingart DE, Giordano TJ, Benitez RS, et al. Overexpression of CXC chemo- kines by an adrenocortical carcinoma: a novel clinical syndrome. J Clin Endo- crinol Metab. 2001;86:3968-3974.
11. Yip L, Duh Q-Y, Wachtel H, et al. American Association of Endocrine Surgeons Guidelines for Adrenalectomy: Executive Summary. JAMA Surgery. 2022;157: 870-877.
12. Else T, Kim AC, Sabolch A, et al. Adrenocortical carcinoma. Endocr Rev. 2014;35: 282-326.
13. Hallanger-Johnson JE. Systemic therapy for adrenocortical carcinoma: a review. AME Med J. 2020;5.
14. Else T, Williams AR, Sabolch A, Jolly S, Miller BS, Hammer GD. Adjuvant therapies and patient and tumor characteristics associated with survival of adult patients with adrenocortical carcinoma. J Clin Endocrinol Metab. 2014;99:455-461.
15. Terzolo M, Angeli A, Fassnacht M, et al. Adjuvant mitotane treatment for adrenocortical carcinoma. N Engl J Med. 2007;356:2372-2380.
16. Zhu J, Zheng Z, Shen J, Lian X, Miao Z, Shen J, et al. Efficacy of adjuvant radiotherapy for treatment of adrenocortical carcinoma: a retrospective study and an updated meta-analysis. Radiat Oncol. 2020;15:118.
17. Polat B, Fassnacht M, Pfreundner L, et al. Radiotherapy in adrenocortical car- cinoma. Cancer. 2009;115:2816-2823.
18. Schteingart DE, Doherty GM, Gauger PG, et al. Management of patients with adrenal cancer: recommendations of an international consensus conference. Endocr Relat Cancer. 2005;12:667-680.
19. Grubbs EG, Callender GG, Xing Y, et al. Recurrence of adrenal cortical carci- noma following resection: surgery alone can achieve results equal to surgery plus mitotane. Ann Surg Oncol. 2010;17:263-270.
20. Fassnacht M, Terzolo M, Allolio B, et al. Combination chemotherapy in advanced adrenocortical carcinoma. N Engl J Med. 2012;366:2189-2197.
21. van Slooten H, van Oosterom AT. CAP (cyclophosphamide, doxorubicin, and cisplatin) regimen in adrenal cortical carcinoma. Cancer Treat Rep. 1983;67: 377-379.
22. Williamson SK, Lew D, Miller GJ, Balcerzak SP, Baker LH, Crawford ED. Phase II evaluation of cisplatin and etoposide followed by mitotane at disease progression in patients with locally advanced or metastatic adrenocorti- cal carcinoma: a Southwest Oncology Group Study. Cancer. 2000;88: 1159-1165.
23. Berruti A, Terzolo M, Sperone P, et al. Etoposide, doxorubicin and cisplatin plus mitotane in the treatment of advanced adrenocortical carcinoma: a large prospective phase II trial. Endocr Relat Cancer. 2005;12:657-666.
24. Zini L, Capitanio U, Jeldres C, et al. External validation of a nomogram pre- dicting mortality in patients with adrenocortical carcinoma. BJU International. 2009;104:1661-1667.
25. Kim Y, Margonis GA, Prescott JD, et al. Nomograms to predict recurrence-free and overall survival after curative resection of adrenocortical carcinoma. JAMA Surg. 2016;151:365-373.
26. Li Y, Bian X, Ouyang J, Wei S, He M, Luo Z. Nomograms to predict overall survival and cancer-specific survival in patients with adrenocortical carcinoma. Cancer Manag Res. 2018;10:6949-6959.
27. Kong J, Zheng J, Cai J, et al. A nomogram for individualized estimation of sur- vival among adult patients with adrenocortical carcinoma after surgery: a retrospective analysis and multicenter validation study. Cancer Commun (Lond). 2019;39:80.
28. Ettaieb MHT, van Kuijk SMJ, de Wit-Pastoors A, et al. Development and internal validation of a multivariable prediction model for adrenocortical-carcinoma- specific mortality. Cancers (Basel). 2020;12:2720.
29. Elhassan YS, Altieri B, Berhane S, et al. S-GRAS score for prognostic classifica- tion of adrenocortical carcinoma: an international, multicenter ENSAT study. Eur J Endocrinol. 2021;186:25-36.
30. Boffa DJ, Rosen JE, Mallin K, et al. Using the National Cancer Database for outcomes research: a review. JAMA Oncol. 2017;3:1722-1728.
31. Hay ID, Bergstralh EJ, Goellner JR, Ebersold JR, Grant CS. Predicting outcome in papillary thyroid carcinoma: development of a reliable prognostic scoring system in a cohort of 1779 patients surgically treated at one institution during 1940 through 1989. Surgery. 1993;114:1050-1058.
32. Chang GJ, Hu CY, Eng C, Skibber JM, Rodriguez-Bigas MA. Practical application of a calculator for conditional survival in colon cancer. J Clin Oncol. 2009;27: 5938-5943.
33. Katz MH, Hu CY, Fleming JB, Pisters PW, Lee JE, Chang GJ. Clinical calculator of conditional survival estimates for resected and unresected survivors of pancreatic cancer. Arch Surg. 2012;147:513-519.
34. Holoubek SA, MacKinney EC, Khokar AM, et al. Treatment differences for adrenocortical carcinoma by race and insurance status. J Surg Res. 2022;280: 169-178.
Discussion
Check for updates
Dr Priya Dedhia (Ohio State): Your team has filled a gap in the research that has been done on adrenocortical carcinoma and our ability to prognosticate the risk of occurrence. Adrenocortical car- cinoma is a heterogeneous disease, and the clinical presentation and disease progression can be quite variable. Because of that, it is very difficult for us to predict what is going to happen to the pa- tient. Being able to prognosticate the risk of recurrence is impor- tant. To that effect, the American Joint Committee on Cancer has developed a staging system, and that staging system involves the T-stage and all the things that Dr Zuber has in his nomogram, which includes the size, the local invasion, as well as the lymph node status, and the distant metastasis. In addition to that, other prog- nostic factors include the pathologic rate. So there have been no- mograms that try to help individualize the risk of recurrence, and most of those have really been based on the Surveillance, Epide- miology, and End Results (SEER) data and The Cancer Genome Atlas Program (TCGA), but TCGA is not as accessible to all of us because
we do not perform all the genomic analyses that the TCGA has performed. On the other hand, SEER does not provide much of the information that the NCDB provides, which is what Dr Zuber and his team’s data is based on. So, this really fills a gap for us and al- lows us to better stratify the risk of recurrence for patients. With that in mind, I wanted to ask you a couple of questions. First, for the NCDB database, does that include cortisol production for the grade? If it does, I am curious as to why you did not include those factors in your nomogram.
Dr Samuel Zuber: Unfortunately, that data is not available for the National Cancer Database (NCDB), and that is one of the things that I do like when you look at the ENSAT study that looks at the S-GRAS score is that they were able to include that. Being able to look at a database that maybe does include that could be benefi- cial. For what we have and if we wanted to maintain the robust- ness of what the NCDB provides for us, then we are still a little bit limited on that.
Dr Dedhia: I think that is an excellent point that the NCDB has such a larger dataset compared to a lot of the nomograms that have been used and that have more robust variables. Could you provide a little bit more detail? Some types of patients might be better suited with the nomogram from either SEER or from some of the smaller institutional nomograms where your data may not fit those pa- tients quite as well.
Dr Zuber: I think just being able to use just one prognostic calculator, you know, there are other nomograms that are out there, so I do think as a provider, if you have that information right there in front of you, having that discussion and being able to pull up our calculator, but then also being able to put the S-GRAS score and some of the other nomograms, you can facilitate that discussion a little bit further. I do not think it is just necessary that you must use one or the other. I think you can use both depending on what data is available to you and what data your pathologist is providing.
Dr Dedhia: I was really surprised to see that insurance plays a role in terms of the risk. I thought that it might make sense to see that in a univariate analysis, but in a multivariate analysis, I would have guessed that no insurance would potentially lead to diagnosis at a later stage and would, therefore, be taken out with the multivariate analysis. Why do you think that is, and is there a way that you think we, as surgeons, could follow up on that finding and help our patients who might have a different insurance status?
Dr Zuber: When we are talking about insurance status and so- cioeconomic status, that can be a sensitive topic to discuss with patients. That being said, I think being able to identify these pa- tients that are at risk based on their insurance status, being able to identify who they are, and to have the discussion not just with the patients but really in a multidisciplinary fashion, whether it is getting social work involved and patient coordinators involved, is going to help identify what limitations they may have both from the patient side and the provider side. There could be some implicit bias from our standpoint, but it is not just about the insurance status. There may be other factors involved as well that we are not aware of.
Dr Matthew Walsh (Cleveland, OH): I have a comment and a question. The comment is around the database NCDB, which is
overall survival and not disease-free survival. This points to the challenge that you are looking specifically for this cancer, but your comorbidity suggests that they may be dying from other causes and not the cancer, correct?
Dr Zuber: Correct. That is one of the limitations of the NCDB. That is why I am a proponent of being able to use not just ours but multiple different prognostic calculators and nomograms when you’re having this discussion with your patients. I think what this does is further reinforce in that patient what their overall prognosis is going to be.
Dr Walsh: And my question is around modifiable factors, which is what I am interested in. As a surgeon, it would seem to me that in this disease. One of those is the surgical margin. Can you use these data to predict who should get neoadjuvant treatment, such as radiation in the neoadjuvant setting, so you might improve that negative margin rate?
Dr Zuber: Yes, I think you can use the data that way. These are all surgical patients, so all patients were treated already with an adrenalectomy. The data does not necessarily include patients who did not have surgery. You can create that model as you have your discussion. If we do your resection and it is R0, this is what survival would be. If your resection is R1, this is what survival would be. Maybe that would help guide patients into deciding that neo- adjuvant treatment might be more beneficial than going straight to surgery.
Dr Eren Berber (Cleveland, OH): Did you look at whether the surgeon or institution experienced volume-affected outcomes? Did you find any benefit to adjuvant therapy for patients with small nodules or tumors that are removed with clear margins? Most of the time, these patients are monitored.
Dr Zuber: As far as facility volume, in this prognostic calculator, we did not want to factor that in. I know that the data does show that higher-volume centers and surgeons will have better out- comes, but when we are having this discussion, a lot of the time, patients have already had surgery. We are using the calculator to decide the next steps. Regarding your other question. We did not look specifically at patients who had small tumors or R0 resection who did receive adjuvant radiation chemotherapy.