ORIGINAL ARTICLE
Development and internal validation of an adrenal cortical carcinoma prognostic score for predicting the risk of metastasis and local recurrence
Daniel Soares Freire*, Sheila Aparecida Coelho Siqueirat, Maria Cláudia Nogueira Zerbinit, Bernardo Léo Wajchenberg*, Maria Lúcia Corrêa-Giannella, Antônio Marmo Lucon§ and Maria Adelaide Albergaria Pereira*
*Serviço de Endocrinologia do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo, țDivisão de Anatomia Patológica do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo, ¿Laboratório de Endocrinologia Celular e Molecular (LIM-25) da Faculdade de Medicina da Universidade de São Paulo and §Disciplina de Urologia do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo, Sao Paulo, Brazil
Abstract
Objective To develop and internally validate a prognostic score to predict the risk of metastases or recurrence in patients with adrenal cortical carcinomas (ACC).
Design Clinical, laboratory and pathological data from 129 ACC patients, treated in a tertiary centre, were retrospectively reviewed. Results Using a multivariate binary logistic regression analysis, we developed a prognostic score with five covariates: a func- tional pattern other than isolated hyperandrogenism, a tumour size >7.5 cm, a primary tumour classified as T3/T4, the presence of microscopic venous invasion and a mitotic index >5/50 high- power fields. The prognostic score was calibrated according to the Hosmer-Lemeshow goodness-of-fit test (P = 0.9329) and exhibited excellent overall performance (Brier score = 0-0738). Finally, the discriminatory ability of the model, determined by the area under the ROC curve (AROc), was near perfect (AROC) 0.9611; 95% CI, 0-92676-0-99552). The prediction model was internally validated with 200 bootstrap resamples and achieved excellent performance for estimating the risk of metastasis and recurrence in eight additional patients with apparently localized disease at diagnosis.
Conclusion We developed and internally validated a prognostic score based on the clinicopathological data that are readily avail- able to any attending physician. Our model can be used to accu- rately estimate the risk of unfavourable outcomes in ACC patients. This score could be beneficial for both patient counsel- ling and the identification of patients in whom adjuvant mito- tane is justified.
(Received 12 December 2012; returned for revision 15 February 2013; accepted 17 February 2013)
Introduction
Adrenal cortical carcinomas (ACCs) are rare tumours associated with poor prognosis, even in patients presenting an apparently localized disease.14 The major and well-established determinants of patient outcome include tumour staging and complete resec- tion by an experienced surgeon; the latter represents the only potentially curative form of treatment for the disease. Currently, there are no prediction tools for estimating the risk of unfavour- able outcomes (metastasis or local recurrence) in patients with adrenal cortical carcinoma. Although the data from two studies applying gene expression arrays are promising in distinguishing between patients with good and poor prognosis,5,6 this molecu- lar approach is not yet reasonable for routine patient care.
Nomograms, which are largely employed in oncology, can reduce complex statistical prediction models to simpler numeri- cal estimates of the likelihood of an event, such as death or relapse. The use of graphical interfaces or web-based tools has increased the use of nomograms to complement or even replace conventional cancer staging systems, such as the tumour-node- metastasis (TNM) classification.7
Therefore, the purpose of this study was to construct and internally validate a nomogram for predicting the risk of metas- tasis and local recurrence in ACC patients who were followed up at a single institution.
Materials and methods
Patient population
Between November 1976 and December 2010, the clinical and anatomo-pathological data were retrospectively collected from
235 patients with adrenal cortical tumours. Of these patients, 96 had adrenal adenomas (Weiss score <3) and were not included in the analysis. Ten patients with a histologically confirmed adrenal cortical carcinoma (Weiss score ≥3) were excluded because they were followed up for less than 24 months without any evidence of metastasis or local recurrence. Our study was based on the remaining 129 patients with a diagnosis of ACC, 63 of whom had no metastases or local recurrences for a mini- mum period of 24 months (median follow-up, 117 months), and 66 of whom had a metastatic outcome (median follow-up, 20 months).
The performance of the prognostic score was also evaluated in eight additional patients with apparently localized disease who were not included in the main cohort due to insufficient follow-up (<24 months without evidence of metastases and/or recurrence) or because they were referred to our service after December 2010. At the end of follow-up, three patients remained free of disease (follow-up, 18-39 months), while five experienced metastases and/or local recurrence after 4-14 months.
Clinicopathological evaluation
The medical records of patients who were diagnosed with ACC and treated at the Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo were retrospectively reviewed. Information on patient gender, age at diagnosis, clini- cal presentation, the type and extent of surgery and the use of adjuvant therapy was recovered, as well as the results of the pathological review and laboratory tests.
The laboratory tests used to determine the hormonal status were as follows: a 24-h urinary cortisol determination; an over- night dexamethasone suppression test (DST); a midnight salivary cortisol (MSC) test; and the determination of ACTH, DHEAS, androstenedione, testosterone and estradiol levels. Because of the retrospective nature of this study, not every patient had under- gone all of the above-mentioned tests; however, a functional tumour pattern was only defined when there was clear evidence of such a pattern from the clinical and laboratory test data. Accordingly, tumours were functionally classified as follows: (I) tumours secreting cortisol alone: elevated 24-h urinary cortisol or abnormal DST/MSC results together with suppressed ACTH and adrenal androgens; (II) tumours secreting androgens alone: elevated androgens for age and gender together with normal 24-h urinary cortisol, DST, MSC and ACTH; (III) tumours with mixed secretion (cortisol + androgens ± estradiol): concomi- tance of ACTH-independent hypercortisolism with elevated androgens or estradiol (in men); and (IV) nonfunctioning tumours. The patients with clinical and laboratorial diagnosis of hyperandrogenism in whom the tests revealed subclinical ACTH-independent Cushing’s syndrome were functionally classified in group III.
Slides were reviewed according to the Weiss score by a pathol- ogist who was experienced in adrenal diseases (SACS or MCNZ). Information about tumour size and extension was also retrieved. Briefly, tumour extension was assessed according to the TNM classification: T1 (tumours ≤5 cm restricted to the adrenal
gland), T2 (tumours >5 cm restricted to the adrenal gland), T3 (tumours invading the periadrenal fat, irrespective of tumour size) and T4 (tumours invading the renal vein, inferior vena cava or adjacent organs, irrespective of tumour size).
The research was approved by a local ethics committee. Due to the retrospective nature of the research and the absence of risks to the patients, the ethics committee did not judge neces- sary to use a written informed consent term.
Statistical analysis
The selected outcome was the occurrence of metastasis or local recurrence of ACC at any time during follow-up, including pre- sentation. The covariates studied were gender, age at diagnosis, functional pattern, tumour size, primary tumour extension, each individual criterion of the Weiss system and the total Weiss score.
To categorize the covariate ‘age at diagnosis’, we used a cut- off of 3 years, which is described in the scientific literature as being of prognostic significance.8 The covariate ‘tumour size’ was dichotomized because the relationship between the covariate and the outcome was not linear. The cut-off was the one with the highest sensitivity and specificity for the outcome (metastasis or local recurrence), established by a ROC curve analysis. The covariate ‘Weiss score’ was also categorized using the optimal cut-off determined by ROC curve analysis.
The covariate ‘primary tumour extension’ was created by cate- gorizing the patients with T1 and T2 tumours in one group (localized disease) and the patients with T3 and T4 tumours in another group (locally advanced disease).
The disease-specific survival of patients with androgen-only secreting ACCs has been reported to be superior compared with that of patients with other functional tumour patterns.1,8-11 Accordingly, in this study, the functional status was further cate- gorized as: (i) tumours secreting androgens alone or (ii) any other functional pattern (nonfunctioning tumours and those secreting cortisol alone or cortisol + androgens ± estradiol).
Initially, a univariate binary logistic regression analysis was conducted to test the association between each covariate and the independent variable. Next, the covariates with a P value <0-20, as determined from a univariate analysis, were tested for the presence of multicollinearity, which was defined by a Spearman coefficient >0-60 or a variance inflation factor >2.5. The covari- ates without significant multicollinearity were entered into a multivariate binary logistic regression analysis with a backward elimination procedure. A P value <0-05 was used as the criterion for retention of variables in the model. The two-sided 95% con- fidence intervals (95% CIs) for odds ratios (ORs) were based on the Wald statistic. The patients with one or more missing values were omitted from the multivariate regression analysis.
The overall performance, calibration and discriminatory abil- ity of the final multivariate binary logistic regression analysis were assessed by the Brier score (BS), the Hosmer-Lemeshow goodness-of-fit test and the area under the ROC curve (AROC), respectively.12
A weighted risk prognostic score was constructed using the coefficients of the multivariate binary logistic regression model,
which were converted into scores by multiplying them by 10 and rounding the values to the nearest whole number.13 For each patient, an individual score was calculated by totalling the scores of the present covariates.
After determining the prognostic score for each patient, a ROC curve analysis was performed to establish the best cut-off points for sensitivity, specificity and the likelihood ratios for the positive and negative results.
A predicted probability was calculated using the regression coefficients of the multivariate binary logistic regression analysis with the following equation:
Probability (metastasis or recurrence) 1 + e(Bo+B1+B2+·+Bn) e(Bo+ß1+2+ … +ßn)
Bootstrap techniques were used to internally validate the prognostic score and to reduce overfit bias. In bootstrap, ran- dom samples with replacement are drawn from the original data set. These resamples have the same size as the original cohort, but due to replacement, their composition is different. The pro- cess is repeated multiple times (in our study, 200 times), and the model derived from the original data set is tested in these bootstrap resamples. The average of the performance indexes (AROC and BS) is considered the bias-corrected estimate of how well the model would perform in the future.
A Mann-Whitney U-test was used for the comparison of ordinary variables between two independent groups.
The model was also tested in an independent group of patients, to infer how it would behave in clinical practice. For this, the prognostic scores and the probabilities of unfavourable outcomes (local recurrence or metastases) were calculated in eight patients with apparently localized disease at diagnosis. The results were then compared with the actual outcomes.
All significance probabilities (P values) presented are bilateral, and values <0.05 were considered to be statistically significant. The software R version 2.13.2 (The R Foundation for Statistical Computing, Vienna, Austria) was used in all steps of the con- struction of the nomogram and the validation using bootstrap techniques.
Results
Sixty-three patients without metastases or recurrent tumours and 66 patients with metastatic or recurrent tumours were eval- uated. Table 1 shows the results of the clinicopathological evalu- ation for the entire cohort and for the metastatic/recurrent and nonmetastatic/nonrecurrent subgroups.
The ROC curve analyses for tumour size and the Weiss score identified the optimum cut-offs for the detection of metastases and recurrent tumours as 7.5 cm and 6 points, respectively. These values were used to categorize the tumour size and Weiss score covariates.
Table 2 shows the results of the univariate binary logistic regres- sion (ORs with 95% CIs and P values) for each of the covariates.
The covariate ‘atypical mitotic figures’ exhibited multicollin- earity with the covariate ‘mitotic index >5/50 HPF’, as evidenced
by a Spearman coefficient of 0-611. Because the performance of the first covariate was inferior to that of the latter in the multi- variate analysis, it was excluded. The covariate ‘Weiss >6’ was excluded for having a variance inflation factor of 3.33.
One patient developed peritoneal carcinomatosis with implants 6 months after a laparoscopic surgery was performed for an 8-cm ACC. Because the surgery may have contributed to the clinical outcome,14,15 data from this patient were not used in the multivariate regression analysis.
The results of the multivariate binary logistic regression analy- sis and the weighted scores for each variable are shown in Table 3. The sample used for the regression comprised 95 indi- viduals, 40 of whom had the clinical outcome ‘occurrence of metastasis or local recurrence’.
The Brier score was 0-0738, indicating an excellent overall performance of the model. The Hosmer-Lemeshow goodness-of- fit test demonstrated that the model was calibrated, indicating that the odds predicted by the model accurately reflected the occurrence of events in real life (P =0.9329). Finally, the dis- criminatory ability of the model was near perfect (AROC, 0-9611; 95% CI, 0-92676-0-99552). As shown in Fig. 1, the maximum sum of sensitivity (90%) and specificity (90.9%) was found at a cut-off point of 77.
An internal validation of the model was performed using 200 bootstrap resamples. The AROC and Brier score were recalculated after bootstrap resampling, and the findings were similar to those obtained with the original sample (BS, 0-0825; AROC 0.9432).
As shown in Fig. 2, the threshold of 77 points consistently separated the patients with an unfavourable outcome from those with a disease-free survival >24 months. The patients with a score <57 points did not have metastases or local recurrences during the follow-up period.
The diagnostic accuracy of the prognostic score was excellent for both adults and children, as shown in Fig. 3.
Table 4 shows the clinicopathological features of the patients not belonging to the main cohort, in whom the performance of the prognostic score was assessed. All five patients with a score ≥ 77 points at diagnosis had unfavourable outcomes during fol- low-up, whereas all three patients with a score <57 points remained free of disease for 18, 34 and 39 months.
Discussion
We proposed, developed and internally validated using bootstrap techniques a clinicopathological prognostic score to predict the risk of unfavourable outcomes in patients with ACC. This score was developed from data retrospectively obtained from a single institution that has over 30 years of experience in the treatment of ACC. Our model was based on the clinical and pathological vari- ables that are readily available to any attending physician.
We chose ‘presence of metastases or tumour recurrence’ as the independent variable because the retrospective nature of the study precludes the use of time-to-event, as imaging re-evalua- tion was not the same for each patient during the 30-year follow-up.
| Absence of mets/ recurrence | Presence of mets/ recurrence | All patients | ||
|---|---|---|---|---|
| n | 63 | 66 | 129 | |
| Gender | Male | 13 | 19 | 32 |
| Female | 50 | 47 | 97 | |
| Age | Average | 20-3 | 34-3 | 27-5 |
| SD | 20-7 | 17.3 | 20-2 | |
| Median | 17 | 36 | 29 | |
| Range | 0-6-79 | 1-71 | 0-6-79 | |
| <15 y | 31 | 9 | 40 | |
| Side | Right | 29 | 32 | 61 |
| Left | 34 | 32 | 66 | |
| Bilateral | 0 | 2 | 2 | |
| Functional | I | 6 | 2 | 8 |
| pattern | II | 31 | 2 | 33 |
| III | 19 | 40 | 59 | |
| IV | 6 | 10 | 16 | |
| N/D | 1 | 12 | 13 | |
| Tumour size | Average | 6-8 | 12.9 | 9.9 |
| SD | 3.8 | 5.3 | 5.5 | |
| Median | 6-0 | 12.0 | 9.0 | |
| IVC/RV | Absent | 62 | 46 | 108 |
| thrombus | Present | 1 | 20 | 21 |
| T | T1 | 25 | 1 | 26 |
| T2 | 32 | 28 | 60 | |
| T3 | 4 | 11 | 15 | |
| T4 | 1 | 24 | 25 | |
| Tx | 1 | 2 | 3 | |
| N | N0 | 63 | 52 | 115 |
| N1 | 0 | 14 | 14 | |
| M | M0 | 63 | 38 | 101 |
| M1 | 0 | 28 | 28 | |
| Resection | R0 | 58 | 25 | 83 |
| status | R1 | 1 | 1 | 2 |
| R2 | 0 | 23 | 23 | |
| Rx | 4 | 11 | 15 | |
| N/A | 0 | 6 | 6 | |
| Staging (ENSAT) | 1 | 25 | 1 | 26 |
| 2 | 32 | 17 | 49 | |
| 3 | 5 | 18 | 23 | |
| 4 | 0 | 28 | 28 | |
| N/A | 1 | 2 | 3 | |
| Weiss score | Mode | 4 | 7 | 5 |
| Range | 3-8 | 4-9 | 3-9 | |
| Führman nuclear | Absent | 8 | 2 | 10 |
| grade III/IV | Present | 48 | 44 | 92 |
| Mitotic | Absent | 34 | 14 | 48 |
| index >5/50 HPF | Present | 22 | 32 | 54 |
| Atypical mitotic figures | Absent | 35 | 13 | 48 |
| Present | 21 | 33 | 54 | |
| Clear cells in | Absent | 9 | 5 | 14 |
| ≤ 25% of the tumour | Present | 47 | 41 | 88 |
| Diffuse architecture | Absent | 5 | 8 | 13 |
| in >33% of the | Present | 51 | 38 | 89 |
| tumour | ||||
(continued)
| Absence of mets/ recurrence | Presence of mets/ recurrence | All patients | ||
|---|---|---|---|---|
| Confluent necrosis | Absent | 21 | 2 | 23 |
| Present | 35 | 44 | 79 | |
| Venous invasion | Absent | 42 | 15 | 57 |
| Present | 14 | 31 | 45 | |
| Sinusoidal invasion | Absent | 50 | 36 | 86 |
| Present | 6 | 10 | 16 | |
| Capsular invasion | Absent | 39 | 17 | 56 |
| Present | 17 | 29 | 46 | |
| Adjuvant | Yes | 15 | 15 | 30 |
| mitotane | No | 48 | 51 | 99 |
| Disease-free | Average | 131-7 | 16-5 | 86-3 |
| survival (mo) | SD | 89.7 | 19.5 | 90-5 |
| Median | 117 | 10 | 50-5 | |
| Disease-specific | Average | 131-7 | 32.1 | 80-8 |
| survival (mo) | SD | 89.7 | 48.3 | 87,1 |
| Median | 117 | 20 | 40 | |
| Survival status | NED | 62 | 6 | 68 |
| AWD | 0 | 6 | 6 | |
| DOD | 0 | 51 | 51 | |
| NCRM | 1 | 3 | 4 |
The tumour functionality was classified into four patterns: pattern I (cortisol alone); pattern II (androgens alone); pattern III (mixed secre- tion); pattern IV (nonfunctioning). n, number of subjects; mets, metasta- sis; SD, standard deviation; N/D, not done; N/A, not available; ENSAT, European Network for the Study of Adrenal Tumors; R0, negative resec- tion margins; R1, microscopic tumour infiltration; R2, macroscopic residual tumour; Rx, uncertain resection status; IVF, inferior vena cava; RV, renal vein; HPF, high-power fields; mo, months; NED, no evidence of disease; AWD, alive with disease; DOD, died of disease; NCRM, non- cancer-related mortality.
The covariates that remained statistically significant in the multivariate binary logistic regression analysis were a tumour with a functional status other than an isolated hyperandroge- nism, a tumour size >7.5 cm, the presence of locally invasive disease (T3/T4 tumours), the presence of microscopic venous invasion and a mitotic index >5/50 HPF. Most of these covari- ates have been shown to be associated with a poor prognosis in both adults2,16-18 and children.1 10,19,20
The functional status was the most important variable in determining the risk of metastasis or local recurrence. We observed that nonfunctioning tumours and those secreting only cortisol or with mixed secretion had an odds ratio of 45-17 for the occurrence of metastasis or relapse, using the androgen-only secreting tumours as a reference. The association of isolated hy- perandrogenism with a better prognosis was reported by us1,9 and others,8,10,11 both among children1,8-11 and adults.1,9 The International Consensus on Adrenal Carcinoma (Ann Arbor, 2003) mentioned isolated hyperandrogenism as a factor that is independently associated with a good prognosis.21 The exclusive secretion of androgens in some tumours possibly indicates that these tumours originate from the foetal adrenal zone, which
| Covariates | OR | 95% CI | P | |
|---|---|---|---|---|
| Size >7.5 cm | 14.998 | 6.143 | 36-615 | <0-001 |
| Weiss score >6 | 13.767 | 4.651 | 40-753 | <0-001 |
| Age >3 years | 12-400 | 4-019 | 38-257 | <0-001 |
| Male gender | 0-643 | 0-286 | 1.446 | 0-285 |
| Functional pattern I, III or IV | 25.988 | 5.815 | 116.140 | <0-001 |
| Primary tumour T3 or T4 | 13.759 | 4.872 | 38-857 | <0-001 |
| Führman nuclear grade III/IV | 2.444 | 0-610 | 9-800 | 0.207 |
| Mitotic index >5/50 HPF | 3.297 | 1-460 | 7.446 | 0-004 |
| Atypical mitotic figures | 3.929 | 1-719 | 8.980 | 0-001 |
| Clear cells in ≤ 25% of the tumour | 1-608 | 0-499 | 5-181 | 0-425 |
| Diffuse architecture in | 0-478 | 0-145 | 1.575 | 0-225 |
| >33% of the tumour | ||||
| Confluent necrosis | 13.500 | 2.964 | 61.486 | <0-001 |
| Venous invasion | 5.837 | 2-471 | 13-788 | <0-001 |
| Sinusoidal invasion | 2.252 | 0-751 | 6.751 | 0-147 |
| Capsular invasion | 3.111 | 1-386 | 6-983 | 0-006 |
The tumour functionality was classified into four patterns: pattern I (corti- sol alone); pattern II (androgens alone); pattern III (mixed secretion); pattern IV (nonfunctioning). Primary tumour extension was evaluated according to the TNM classification (see text). OR, odds ratio; 95% CI, 95% confidence interval; P, statistical significance; HPF, high-power fields.
mainly secretes DHEA and DHEAS, what could result in a dif- ferent biological behaviour.22 Alternatively, cortisol-secreting tumours (including those with mixed secretion) may exhibit a more aggressive behaviour possibly due to the immunosuppres- sive effect of cortisol, as reported previously by Abiven et al.23
The second most important covariate was tumour size. The threshold of 7.5 cm was defined by the ROC curve analysis as the one that maximizes the sensitivity and specificity for the occurrence of metastasis or local recurrence of ACCs. Tumour size is known to be correlated with prognosis, both in children and adults. Ribeiro et al.24 studied 40 children and found that a tumour volume >200 cm3 was the main covariate associated with the occurrence of metastases. Considering a spherical tumour shape, this finding corresponds to a tumour diameter of approximately 7-25 cm (V = 4/3*Tr3). Wieneke et al.19 found
100
80
Optimum Cutoff: >77
Sp., 90%; Sn., 90-9%
LR+, 9.90; LR-, 0-11
Sensitivity (%)
60
40
20
0
0
20
40
60
80
100
1-Specificity (%)
that tumours larger than 10-5 cm were associated with a poor prognosis in children. In adults, the tumour size cut-offs associ- ated with a poor prognosis ranged from 6-5 to 12 cm.2,16,25,26 According to the data from the German ACC registry, a tumour diameter of 8 cm has been used to identify carcinomas with a greater risk of recurrence.2 27,28
The local extent of disease also had a prognostic significance in our study. To avoid multicollinearity with the covariate tumour size, we separated the patients with localized disease (T1 + T2) from those with locally invasive disease (T3 + T4). This approach allows for the analysis of tumour invasiveness irrespective of tumour size. In paediatric patients with ACC, invasion into periadrenal adipose tissue (T3) and the inferior
| Covariate | b | SE | P | OR | 95% CI | Points | ||
|---|---|---|---|---|---|---|---|---|
| Size | >7.5 cm | 2-784 | 0-849 | 0-001 | 16-18 | 3-06 | 85-50 | 28 |
| ≤7.5 cm | 1.0 (Ref.) | 0 | ||||||
| Primary tumour | T3 or T4 | 2.617 | 0.917 | 0-004 | 13-70 | 2.27 | 82.60 | 26 |
| T1 or T2 | 1.0 (Ref.) | 0 | ||||||
| Mitotic index | >5/50 HPF | 1-878 | 0-828 | 0-023 | 6.54 | 1.29 | 33.20 | 19 |
| ≤ 5/50 HPF | 1.0 (Ref.) | 0 | ||||||
| Venous invasion | Present | 1.993 | 0-880 | 0-023 | 7.34 | 1-31 | 41.20 | 20 |
| Absent | 1-0 (Ref.) | 0 | ||||||
| Functional pattern | Any other | 3-810 | 1.329 | 0-004 | 45.17 | 3.34 | 612.00 | 38 |
| Androgens only | 1-0 (Ref.) | 0 |
ß, regression coefficients; SE, standard error; P, statistical significance; OR, odds ratio; 95% CI, 95% confidence interval; points, score calculated for each covariate from the regression coefficients (see text); Ref., reference; HPF, high-power fields. The regression coefficient of the intercept was -7-7747.
120
100
Predicted probability (%)
80
60
40
20
, Absence of mets/recurrence
O Presence of mets/recurrence
o
-20
0
40
60
80
100
120
140
160
20
Prognostic score (points)
vena cava (T4) was associated with a malignant behaviour;19 moreover, the presence of inferior vena cava invasion was one of the three remaining covariates with a significant prognostic value in the multivariate model, in addition to necrosis and mitotic index.19 Data from the Sloan-Kettering Cancer Center, published in 2002 by Stojadinovic et al.,2 confirm these findings in a population comprising children and adults; namely, inva- sion into adjacent organs (T4) was negatively associated with prognosis in the univariate (P = 0-004) and multivariate analyses (P = 0-03). According to the German ACC Registry, which is the largest cohort of patients with ACC with complete clinical annotation, comprising more than 400 patients, the presence of a venous tumour thrombus in the renal vein or the vena cava (T4), the invasion of adjacent organs (T4) or the infiltration of surrounding adipose tissue (T3) were independently associated with decreased disease-specific survival.4
Finally, two microscopic covariates (mitotic index >5/50 HPFs and presence of venous invasion) were negatively correlated with
prognosis. Several studies have reported a higher mitotic rate to be associated with a reduced specific survival.2,3,16,17,19,26,29 The presence of more than 5 mitoses per 50 HPFs is already known to be negatively associated with survival; this correlation becomes more evident with a higher mitotic rate, particularly in paediatric patients, in whom a cut-off of 15 mitoses per 20 CGA (equivalent to 37.5 mitoses/50 HPFs) has been reported.19 Wieneke et al.19 demonstrated that venous invasion has a prognostic significance in paediatric ACC patients.
In the present study, the discriminatory capacity of the prog- nostic score was similar between the two age groups. Moreover, in the multivariate binary logistic regression analysis, the covari- ate ‘age >3 years’ did not have statistical significance, indicating that the other covariates had a more robust association with outcome in this population.
The observations on the prognostic significance of the studied covariates are not new; however, our approach allowed us to study the relative contribution of each variable in the final risk of an unfavourable outcome. The patients with a score ≥77 points had an approximately 10-fold increased risk of metastatic outcome compared with patients with scores below this thresh- old. The patients with less than 57 points did not experience any metastasis or local recurrence for a minimum follow-up of 24 months.
Metastases following an apparently complete resection of ACC occur in 60-80% of cases,2 suggesting the need for an effective adjuvant therapy. Available adjuvant strategies include mitotane monotherapy,30,31 mitotane plus streptozotocin32 and adrenal bed irradiation.33 Until the results of a prospective study in progress become available,34 most experts agree that some ACC patients may benefit from the adjuvant use of mitotane following radical surgery. 27,28
On the one hand, adjuvant therapy appears to reduce the risk of unfavourable events (recurrence, metastasis or death) related to adrenal carcinoma. On the other hand, it is associ- ated with significant toxicity. Thus, adjuvant adrenal bed irra- diation and/or chemotherapy are commonly reserved for patients in whom a potential benefit outweighs the associated toxicity. To make this decision, variables such as the extent of
140
Sn. = 83.3%
Sp. = 96-4% Ac. = 94-1%
Sn. = 91-2% Sp. = 81.5% Ac. = 86.9%
120
Prognostic score (points)
100
80
Sn. = 90-0%
77 points
Sp. = 90-9%
Ac. = 90-5%
60
40
20
Absence of mets/recurrence
Presence of mets/recurrence
0
<15 years
≥15 years
| Age | Weiss score | Tumour size (cm) | Primary tumour | Mitotic index | Venous invasion | Functional pattern | Total score | DFS (mo) | DSS (mo) | Final status | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 3 y | 8 | 13.5 | T2 | >5 | Present | III | 105 | 12 | 20 | DOD |
| 2 | 65 y | 4 | 7.0 | T2 | ≤5 | Absent | III | 38 | 39 | 39 | NED |
| 3 | 22 y | 6 | 11-0 | T2 | >5 | Absent | IV | 85 | 14 | 28 | AWD |
| 4 | 66 y | 8 | 20-0 | T3 | >5 | Present | I | 131 | 7 | 11 | AWD |
| 5 | 64 y | 8 | 13.0 | T3 | >5 | Present | II | 93 | 4 | 14 | DOD |
| 6 | 45 y | 3 | 4.0 | T1 | ≤5 | Absent | III | 38 | 18 | 18 | NED |
| 7 | 11 mo | 6 | 4.2 | T1 | >5 | Absent | III | 57 | 34 | 34 | NED |
| 8 | 27 y | 7 | 7.5 | T2 | >5 | Present | III | 77 | 6 | 28 | DOD |
The tumour functionality was classified into four patterns: pattern I (cortisol alone); pattern II (androgens alone); pattern III (mixed secretion); pattern IV (nonfunctioning); DFS, disease-free survival; DSS, disease-specific survival; mo, months; y, years; DOD, died of disease; NED, no evidence of disease; AWD, alive with disease. Mitotic index was calculated in 50 high-power fields.
surgical resection are taken into consideration, in addition to the following tumour characteristics that have a proven corre- lation with a higher likelihood of unfavourable outcome: a Ki- 67 index >10 or 20% (or a mitotic index >5/50 HPF if Ki-67 labelling is not available), a tumour size greater than 8 cm, microscopic venous invasion or invasion of the tumour capsule and the presence of tumour invasion extension into the infe- rior vena cava. 27,28
The use of adjuvant mitotane in our institution is made on an individual basis; thus, the presence of a selection bias makes our study inconclusive regarding the efficacy of adjuvant mito- tane. Indeed, the prognostic score of the patients selected to receive mitotane was significantly higher than that of the patients who were not selected (data not shown). This finding suggests that the features that supported the decision to select patients for mitotane adjuvant therapy could be objectively quantified with the score.
Our findings indicate that the prognostic score can accurately predict outcome and therefore could be used to guide decision- making regarding adjuvant mitotane. Patients with a score ≥77 points are at a high risk of metastasis and, for these patients, mitotane use would be justified. In contrast, patients with less than 57 points would not benefit from an adjuvant mitotane treatment because their prognosis is already good. For patients with a prognostic score between 57 and 77, the decision to use adjuvant mitotane should be made on a case-by-case basis.
There are some limitations to this study: first, the histological covariates used in the analysis were the ones present in the Weiss system, and we did not evaluate other microscopic fea- tures, such as the Van Slooten index (VSI). However, most of the VSI criteria are somehow represented within the Weiss sys- tem: ‘abnormal nucleoli’, ‘nuclear hyperchromasia’ and ‘nuclear atypia’ are contemplated within the Weiss criterion ‘high nuclear grade’; Van Slooten’s ‘structural changes’ is represented by Weiss’ ‘diffuse architecture’ and the ‘regressive changes’ (haem- orrhage, fibrosis and calcification) of the VSI are related to necrosis. Moreover, the only microscopic covariates that remained statistically significant in the multivariate model were venous invasion and high mitotic index, both present in the Weiss system as the VSI.
Second, the analysis was performed using retrospective data, and due to the rarity of the disease, it was not possible to vali- date the prognostic score in an independent large population. However, the bootstrap techniques, which have been widely used to validate models for cancer prognosis while they cannot be externally validated,7 confirmed the robustness of the model (the AROC decreased from 0-9611 to 0.9432 after bootstrapping). Despite being an insufficient sample for external validation of the model, the findings we obtained in the subgroup of eight patients who were not part of the main cohort are promising and accurately reflect the results of the bootstrap validation. Finally, because restaging methods changed significantly in the last 35 years, the possibility of lead-time bias precluded the use of a time-to-event response variable such as disease-free survival.
Conclusion
In conclusion, we have developed a new, individualized prog- nostic model that predicts the risk of local recurrence or metas- tases in patients with ACC. The proposed risk prediction model has the potential to improve the management of ACC patients by allowing physicians to more precisely identify patients at high risk of metastases who might be candidates for adjuvant mito- tane.
Acknowledgements
We thank Prof. Berenice Bilharinho de Mendonça and Prof. Maria Cândida Barisson Villares Fragoso for providing clinical information of the patients treated by them and Mr. Frederico Moreira for statistical support.
Conflict of interest
The authors declare that there is no conflict of interest.
Funding
This research did not receive any specific grant from any fund- ing agency in the public, commercial or not-for-profit sector.
References
1 Wajchenberg, B.L., Albergaria Pereira, M.A., Medonca, B.B. et al. (2000) Adrenocortical carcinoma: clinical and laboratory obser- vations. Cancer, 88, 711-736.
2 Stojadinovic, A., Ghossein, R.A., Hoos, A. et al. (2002) Adreno- cortical carcinoma: clinical, morphologic, and molecular charac- terization. Journal of Clinical Oncology, 20, 941-950.
3 Miller, B.S., Gauger, P.G., Hammer, G.D. et al. (2010) Proposal for modification of the ENSAT staging system for adrenocortical carcinoma using tumor grade. Langenbeck’s Archives of Surgery, 395, 955-961.
4 Fassnacht, M., Johanssen, S., Quinkler, M. et al. (2009) Limited prognostic value of the 2004 International Union Against Cancer staging classification for adrenocortical carcinoma: proposal for a Revised TNM Classification. Cancer, 115, 243-250.
5 de Reynies, A., Assie, G., Rickman, D.S. et al. (2009) Gene expression profiling reveals a new classification of adrenocortical tumors and identifies molecular predictors of malignancy and survival. Journal of Clinical Oncology, 27, 1108-1115.
6 Giordano, T.J., Kuick, R., Else, T. et al. (2009) Molecular classification and prognostication of adrenocortical tumors by transcriptome profiling. Clinical Cancer Research, 15, 668-676.
7 Iasonos, A., Schrag, D., Raj, G.V. et al. (2008) How to build and interpret a nomogram for cancer prognosis. Journal of Clinical Oncology, 26, 1364-1370.
8 Michalkiewicz, E., Sandrini, R., Figueiredo, B. et al. (2004) Clini- cal and outcome characteristics of children with adrenocortical tumors: a report from the International Pediatric Adrenocortical Tumor Registry. Journal of Clinical Oncology, 22, 838-845.
9 Mendonca, B.B., Lucon, A.M., Menezes, C.A., et al. (1995) Clini- cal, hormonal and pathological findings in a comparative study of adrenocortical neoplasms in childhood and adulthood. Journal of Urology, 154, 2004-2009.
10 Pereira, R.M., Michalkiewicz, E., Sandrini, F. et al. (2004) Child- hood adrenocortical tumors. Arquivos Brasileiros De Endocrinolo- gia e Metabologia, 48, 651-658.
11 Sandrini, R., Ribeiro, R.C. & DeLacerda, L. (1997) Childhood adrenocortical tumors. Journal of Clinical Endocrinology and Metabolism, 82, 2027-2031.
12 Steyerberg, E.W., Harrell, F.E. Jr, Borsboom, G.J. et al. (2001) Internal validation of predictive models: efficiency of some pro- cedures for logistic regression analysis. Journal of Clinical Epide- miology, 54, 774-781.
13 Moons, K.G., Harrell, F.E. & Steyerberg, E.W. (2002) Should scoring rules be based on odds ratios or regression coefficients? Journal of Clinical Epidemiology, 55, 1054-1055.
14 Miller, B.S., Ammori, J.B., Gauger, P.G. et al. (2010) Laparo- scopic resection is inappropriate in patients with known or suspected adrenocortical carcinoma. World Journal of Surgery, 34, 1380-1385.
15 Leboulleux, S., Deandreis, D., Al Ghuzlan, A. et al. (2010) Adrenocortical carcinoma: is the surgical approach a risk factor of peritoneal carcinomatosis? European Journal of Endocrinology, 162, 1147-1153.
16 Harrison, L.E., Gaudin, P.B. & Brennan, M.F. (1999) Pathologic features of prognostic significance for adrenocortical carcinoma after curative resection. Archives of Surgery, 134, 181-185.
17 Evans, H.L. & Vassilopoulou-Sellin, R. (1996) Adrenal cortical neoplasms. A study of 56 cases. American Journal of Clinical Pathology, 105, 76-86.
18 Tang, C.K. & Gray, G.F. (1975) Adrenocortical neoplasms. Prog- nosis and morphology. Urology, 5, 691-695.
19 Wieneke, J.A., Thompson, L.D. & Heffess, C.S. (2003) Adrenal cortical neoplasms in the pediatric population: a clinicopatho- logic and immunophenotypic analysis of 83 patients. American Journal of Surgical Pathology, 27, 867-881.
20 Dehner, L.P. & Hill, D.A. (2009) Adrenal cortical neoplasms in children: why so many carcinomas and yet so many survivors? Pediatric and Developmental Pathology, 12, 284-291.
21 Schteingart, D.E., Doherty, G.M., Gauger, P.G. et al. (2005) Management of patients with adrenal cancer: recommendations of an international consensus conference. Endocrine-Related Can- cer, 12, 667-680.
22 Keegan, C.E. & Hammer, G.D. (2002) Recent insights into organogenesis of the adrenal cortex. Trends in Endocrinology and Metabolism, 13, 200-208.
23 Abiven, G., Coste, J., Groussin, L. et al. (2006) Clinical and bio- logical features in the prognosis of adrenocortical cancer: poor outcome of cortisol-secreting tumors in a series of 202 consecu- tive patients. Journal of Clinical Endocrinology and Metabolism, 91, 2650-2655.
24 Ribeiro, R.C., Sandrini Neto, R.S., Schell, M.J. et al. (1990) Adrenocortical carcinoma in children: a study of 40 cases. Jour- nal of Clinical Oncology, 8, 67-74.
25 Aubert, S., Wacrenier, A., Leroy, X. et al. (2002) Weiss system revisited: a clinicopathologic and immunohistochemical study of 49 adrenocortical tumors. American Journal of Surgical Pathology, 26, 1612-1619.
26 Weiss, L.M., Medeiros, L.J. & Vickery, A.L. Jr (1989) Pathologic features of prognostic significance in adrenocortical carcinoma. American Journal of Surgical Pathology, 13, 202-206.
27 Fassnacht, M. & Allolio, B. (2010) What is the best approach to an apparently nonmetastatic adrenocortical carcinoma? Clinical Endocrinology (Oxf), 73, 561-565.
28 Fassnacht, M., Libe, R., Kroiss, M. et al. (2011) Adrenocortical carcinoma: a clinician’s update. Nature Reviews. Endocrinology, 7, 323-335.
29 Klein, J.D., Turner, C.G., Gray, F.L. et al. (2011) Adrenal cortical tumors in children: factors associated with poor outcome. Jour- nal of Pediatric Surgery, 46, 1201-1207.
30 Grubbs, E.G., Callender, G.G., Xing, Y. et al. (2010) Recurrence of adrenal cortical carcinoma following resection: surgery alone can achieve results equal to surgery plus mitotane. Annals of Surgical Oncology, 17, 263-270.
31 Terzolo, M., Angeli, A., Fassnacht, M. et al. (2007) Adjuvant mitotane treatment for adrenocortical carcinoma. New England Journal of Medicine, 356, 2372-2380.
32 Khan, T.S., Imam, H., Juhlin, C. et al. (2000) Streptozocin and o, p’DDD in the treatment of adrenocortical cancer patients: long-term survival in its adjuvant use. Annals of Oncology, 11, 1281-1287.
33 Fassnacht, M., Hahner, S., Polat, B. et al. (2006) Efficacy of adju- vant radiotherapy of the tumor bed on local recurrence of adre- nocortical carcinoma. Journal of Clinical Endocrinology and Metabolism, 91, 4501-4504.
34 Terzolo, M., Fassnacht, M., Berruti, A. et al. (2008) Efficacy of Adjuvant Mitotane Treatment in Prolonging Recurrence-free Survival in Patients With Adrenocortical Carcinoma at Low-intermediate Risk of Recurrence. Available at : http://www. In clinicaltrials.org.