World J Surg DOI 10.1007/s00268-014-2524-0
Risk Estimator for Adrenal Tumor Functionality Claire Sadler • Melanie Goldfarb
Ó Socie´te´ Internationale de Chirurgie 2014
Abstract Background Adrenal lesions are a common imaging finding with a prevalence approaching 10 %. Although guidelines recommend dedicated laboratory tests to rule out tumor functionality, many patients never undergo this workup. This study investigates the use of demographic and clinical variables to create an easy scoring system for predicting adrenal tumor functionality (functional adrenal tumors, or FATs). Methods Altogether, 2,807 patients in the NSQIP 2005–2010 database underwent adrenalectomy as their principal operation and had a postoperative diagnosis consistent with an adrenal lesion/disorder. Patients were divided into two groups based on a postoperative diagnosis consistent with tumor functionality. Univariate and multivariate logistic regression analyses were performed to identify specific predictors of FATs and for Cushing’s, Conn’s, or pheochromocytoma. Results Overall, 13.2 % (n = 402) of adrenalectomies performed were for FATs. Patients with a FAT were younger (age \40, p \ 0.01), overweight (BMI [ 30 kg/ m2, p \ 0.01), hypertensive (p \ 0.001). They also had elevated white blood cells (WBC [ 11, p \ 0.001), serum creatinine (Cr [ 1.25 mg/dl, p \ 0.001), and sodium (Na [ 143 mmol/L, p \ 0.001). On multivariate regression, patients with these characteristics were 20.53 times Presented at the 84th Annual Pacific Coast Surgical Meeting, Kauai, Hawaii, February 17, 2013. C. Sadler M. Goldfarb (&) Division of Breast/Soft Tissue and Endocrine Surgery, Department of Surgery, University of Southern California Keck School of Medicine, 1510 San Pablo Street, Suite 412 k, Los Angeles, CA 90033, USA e-mail: [email protected]
(CI 15.79–25.27) more likely to have a FAT (model c-statistic 0.634, CI 0.605–0.663; Hosmer–Lemeshow test (H–L), p = 0.035). Patients who were younger (p \ 0.001), female (p \ 0.001), diabetic (p = 0.07), overweight (p = 0.027), with elevated WBCs (p \ 0.001) and lower Cr (p \ 0.001) were 63.62 times (CI 58.03–69.21) more likely to have Cushing’s (model c-statistic 0.685, CI 0.648–0.722; H–L p = 0.954). Conclusions After external validation, this risk estimator might be used to quantify the probability of tumor functionality in patients with incidental adrenal masses. Although predictive power may be limited, it helps identify patients at high risk for FATs that need more urgent referral to a specialist.
Introduction Incidentally discovered adrenal lesions, or adrenal incidentalomas (AIs), are an increasingly common finding. With the increasing use of computed tomography (CT) or magnetic resonance imaging (MRI) to investigate almost any abdominal complaint, the reported prevalence of these lesions now approaches 5 % and reaches 10 % in older individuals [1–7]. Although the majority of AIs are benign, many produce subclinically elevated amounts of hormones [8, 9]. Left untreated, these ‘‘subclinical’’ functional tumors, where patients produce excess levels of hormones without recognizable pathology, are associated with increased morbidity [4, 6, 10–12]. However, with the exception of overt pheochromocytomas, these functional tumors, unlike a primary adrenal malignancy or metastatic tumor, do not demonstrate any distinct radiographic findings. Therefore, current guidelines recommend that all
123
World J Surg
patients with an AI undergo a complete endocrine evaluation to rule out any tumor functionality [13–15]. Despite these published recommendations from both medical and surgical endocrine societies, many patients with AIs fail to receive an endocrinology referral or undergo the appropriate workup by a clinician with adrenal tumor experience. This is especially true when AIs are detected in an emergency room or acute setting and not uncommonly at large public hospitals. Many adrenal lesions are not even reported [3]. This may be due to other, more pressing concerns at the time, angst and therefore dismissal by the patient, or a patient not being made aware of the diagnosis. The purpose of this study is therefore to develop an easy clinical scoring system that can be used in emergency and primary care settings to identify patients at highest risk for a functional adrenal tumor (FAT).
Methods Data retrieval The general approach and methods of data collection for the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database are described elsewhere [16, 17]. The ACS-NSQIP collects data from trained surgical clinical reviewers and includes preoperative, operative, and postoperative variables for 30 days following the procedure. All cases in the ACS-NSQIP 2005–2010 participant use file that documented adrenalectomy as the principal operation (current procedural terminology codes 60540 and 60650) and had a postoperative diagnosis—using international classification of disease (ICD-9) codes—consistent with an adrenal lesion or disorder were identified (Table 1). A total of 2,807 subjects were included in the final statistical analysis. Patients were divided into two groups based on tumor functionality: Patients with a postoperative diagnosis of Conn’s syndrome, Cushing’s syndrome, or pheochromocytoma were categorized as having functional adrenal tumors (FATs). All other neoplasms were classified as nonfunctional. Some variables that were not analyzable using a binary code were regrouped and classified as follows based on published literature or standard reference ranges: diabetes (none versus oral medication/insulin), dyspnea (none versus any), anesthesia technique (general versus other), young age (B45 years), American Society of Anesthesiologists (ASA) class I/II (mild) and class III/IV (severe), elevated creatinine (Cr [ 1.25 mg/dl), elevated sodium (Na [ 143 mmol/ L), elevated white blood cell count (WBC [ 11 9 109/L), low hematocrit (Hct B 36.0 %), and low alkaline phosphatase (AP) (AP \ 150 IU/L). All wound infections were
123
grouped into one category. Laboratory cutoff values are based on previously identified cutoffs for abnormal values that can be seen in adrenal dysfunction [18, 19]. The BMI was calculated from height and weight variables, with ‘‘obese’’ defined as BMI [ 30 kg/m2. Missing laboratory values were not imputed. Statistical analysis Preoperative laboratory values, demographics, and comorbidities were compared between the FAT and nonfunctional groups. Univariate analysis was performed to identify specific predictors of FAT and for the specific FAT syndromes: Cushing’s, Conn’s, and pheochromocytoma. Categoric variables were analyzed using Pearson’s v2 test and continuous variables with the Mann–Whitney U test. A complete multivariate regression model was constructed for FAT and each FAT syndrome using only patients with complete data. We first entered all demographic and preoperative values into a forward logistic regression. A reduced model was then generated that included only variables that were statistically significant (p \ 0.05). Receiver operating characteristic (ROC) curves and the Hosmer–Lemeshow (H–L) test were used to assess model quality. Then c-statistics were abstracted from ROC curves to assess for discriminative ability of the model, with a higher c-statistic indicating a better ability of the model to distinguish functional and nonfunctional adrenal tumor groups. A higher H–L result indicated better calibration of the model, meaning that the predicted probabilities agree with the outcomes. A simple scoring system derived from the slope, or log-odds of the reduced model was created by multiplying each log-odds by 2 and then rounding to the Table 1 International statistical classification of diseases and healthrelated problems, 9th revision (ICD-9) codes utilized 194.0: Malignant neoplasm of adrenal gland
255.5: Other adrenal hypofunction
227.0: Benign neoplasm of adrenal gland 237.2 Neoplasm of uncertain behavior of adrenal gland
255.6: Medulloadrenal hyperfunction 255.8: Other specified disorders of adrenal glands
239.7: Neoplasm of uncertain nature of endocrine glands
255.9: Unspecified disorder of adrenal glands
255.0: Cushing’s syndrome
258.0: Polyglandular activity in multiple endocrine adenomatosis 276.8: Hypopotassemia
255.1: Hyperaldosteronism 255.2: Adrenogenital disorders
401.0: Malignant essential hypertension
255.3: Other corticoadrenal overactivity
401.9: Unspecified essential hypertension
255.4: Corticoadrenal insufficiency
759.1: Anomalies of adrenal gland
13.8 % (8)
19 % (11)
0
14.3 % (8)
20.7 % (12)
18 % (11)
78.7 % (48)
32.8 % (20)
33.3 % (20)**
** p \ 0.05 and * p \ 0.01 compared to the overall cohort
AP alkaline phosphatase, BMI body mass index, Cr creatinine, Hct hematocrit, Pheo pheochromocytoma, WBC white blood cell count
16.1 % (22) 16.6 % (29) NS 17.5 % (392) Hct \ 36 %
16.5 % (62)
35 % (48)* 5.2 % (9)* \0.001 10.7 % WBC [ 11 9 109/L
17.4 %*
6.5 % (6) 1.2 % (1) NS 3.6 % (48) :AP
3.5 % (8)
6.7 % (9)
20 % (27) 19.3 % (34)
26.7 % (48)* \0.001
0.004 19.5 % (73)*
Cr [ 1.25 mg/dl
17.8 % (67)*
13.7 % (305)
10.8 % (239)
Na [ 143 mmol/L
19 % (458) Diabetes on meds
Laboratory values
70.1 % (103)
27.9 % (41)* 19.7 % (37)
95.2 % (179)*
0.072
\0.001 70.5 % (1,695) Hypertension on meds
83.1 % (334)*
46.2 % (1,100) BMI [ 30 kg/m2
22.9 % (92)
61.2 % (90)* 54.5 % (102)** 0.004
30.1 % (724) Age \45 years
54 % (216)*
60 % (36) 84.4 % (124)*
51 % (75)* 29.8 % (56)
48.4 % (91)* NS
\0.001
63.5 % (1,523) Female sex
38.6 % (155)
Nonfunctional n = 2,405 (85.7 %)
64.1 % (257)
nearest whole digit number. The scoring system was compared to the logistic regression model using the ROC curves. All statistical analyses used two-sided tests and were conducted using SPSS software, version 20.0 (SPSS, Chicago, IL, USA).
Results
Parameter
Table 2 Demographics of functional versus nonfunctional tumors
Functional n = 402 (14.3 %)
p
Conn’s n = 188
Cushing’s n = 147
Pheo n = 61
World J Surg
Among 2807 patients who underwent adrenalectomy for a primary adrenal tumor, 402 (14.3 %) of the tumors were FATs. Patients with a FAT were more likely young (p \ 0.01), obese (p \ 0.01), and hypertensive (p \ 0.001) with elevated WBC count (p \ 0.001), serum Cr (p \ 0.001), and serum Na (p \ 0.001) (Table 2). A complete multiple regression model was created that included all significant or nearly significant variables (p \ 0.05) in the univariate analysis. In all, 2424 patients had sufficiently complete data to be included in this portion of analysis. Forward regression resulted in six variables for use in modeling (Table 3). The model demonstrated moderate discriminative ability but poor calibration: c-statistic 0.634, CI 0.605–0.663; H–L, p = 0.035). When BMI was removed from the model, calibration increased, resulting in a better model (c-statistic 0.631, CI 0.602–0.660; H–L, p = 0.157). Based on this logistic regression model, a simplified scoring system was created (Table 4). When a cutoff point of 4 was used, the model resulted in 91.4 % specificity and 16.3 % sensitivity for predicting FAT. An ROC curve for this model demonstrated good discriminative ability, with a c-statistic of 0.622 (CI 0.592–0.651). A cutoff value of 3 (of 6) produced a model with a sensitivity of 53.7 % and a specificity of 62.8 %, whereas a cutoff value of 5 (of 6) yielded a model with 2.2 % sensitivity and 99.4 % specificity. Compared to the overall cohort, patients with Cushing’s syndrome (n = 147) were more likely female (p \ 0.001), diabetic (p = 0.07), young (p \ 0.001), and obese (p = 0.027) with an elevated WBC count (p \ 0.001) and lower Cr (p \ 0.001) (Table 2). A complete multiple regression model was created including all significant or nearly significant variables (p \ 0.05) in the univariate analysis. Altogether, 2490 patients had sufficiently complete data to be included in this portion of analysis. Forward logistic regression resulted in a pared-down regression model that included age, diabetes, elevated Na, and elevated WBC (Table 5). The model displayed both good discrimination and calibration (c-statistic 0.70, CI 0.650–0.748; H–L, p = 0.787). In contrast, patients with Conn’s syndrome (n = 188) were more likely male (p \ 0.001) and hypertensive (p \ 0.001) with elevated Cr (p \ 0 0.001) (Table 1). A complete multiple regression model including all significant
123
World J Surg Table 3 Multivariate analysis for adrenal tumor functionality (n = 402) Factor
Odds ratio of being functional
Confidence interval
Age \45 years
1.71
1.34–2.19
\0.001
BMI [ 30 kg/m2
1.29
1.03–1.63
0.029
Hypertension on meds
2.00
1.47–2.72
\ 0.001
WBC [ 11 9 109/L
1.76
1.29–2.40
0.029
Cr [ 1.25 mg/dl
1.72
1.26–2.34
0.001
Na [ 143 mmol/L
1.49
1.10–2.00
0.009
Table 5 Multivariate analyses for Cushing’s syndrome (n = 133) Factor
Odds ratio of being Cushing’s
Confidence interval
p
Diabetes
1.75
1.15–2.66
\0.01
p
Age \45 years old
2.37
1.63–3.43
\0.001
Elevated Na [ 143 mmol/ dl
1.77
1.12–2.81
0.015
WBC [ 11 9 109
4.24
2.87–6.24
\0.001
Table 6 Multivariate analysis for Conn’s syndrome (n = 180) Factor
Odds ratio of being Conn’s
Confidence interval
p
Table 4 Scoring system for adrenal tumor functionality Factor
Points
Hypertension requiring medication
2
Age \45 years
1
Cr [ 1.25 mg/dl
1
Na [ 143 mmol/L
1
WBC [ 11 9 109/L Maximum score
1 6
or nearly significant variables (p \ 0.05) in the univariate analysis was created. In all, 2,574 patients had sufficiently complete data to be included in this portion of analysis. Forward logistic regression resulted in a model that included only hypertension and sex, although because elevated serum Cr had a significance of exactly p = 0.05 it was included in the final model as well (Table 6). This model showed moderate discriminative ability and excellent calibration (c-statistic 0.685, CI 0.648–0.722; H–L, p = 0.954). Patients with pheochromocytoma were less likely to be obese, but no variables were sufficiently significant on multivariate regression to create a predictive model (Table 2).
Discussion Adrenal incidentalomas are a common imaging finding. However, with benign-appearing lesions, ensuring that patients receive an appropriate workup and subsequent follow-up for tumor functionality remains a challenge. Many adrenal tumors are discovered in an emergency room setting or an independent radiology facility, and patients are not made aware of their AI, not referred to an endocrine specialist, or the adrenal lesion is not reported. As failure to treat a FAT can negatively affect patient morbidity, is it important that all patients with an AI undergo hormonal evaluation [10, 11]. The proposed simple, easy-to-use scoring system could help clinicians identify patients most
123
Male sex
1.60
1.16–2.18
Hypertension on meds Creatinine [1.25 mg/dl
6.38
3.23–12.59
\0.001
0.004
2.13
1.47–3.07
\0.001
at risk for a FAT and prioritize which patients require a more urgent endocrine referral. The proposed risk estimator uses a cutoff value of 4 (of 6) to produce a model with a sensitivity of 16.3 % and a specificity of 91.4 % for predicting a FAT. As a screening tool, higher specificity is desired with sensitivity sacrificed because clinical guidelines recommend that all patients with an AI should undergo a thorough workup for both functionality and malignancy. The proposed scoring system would thus identify patients most at risk for having a FAT. Ideally, this would prompt a treating clinician to ensure that a referral and/or follow-up is truly obtained once such a patient is identified. In this study, approximately 10 % of patients had a score of C4 and would have been considered to be at high risk for a FAT. In reality, 25 % of these patients had a FAT based on their diagnosis code, which is likely an underestimate of the true number of patients with subclinical FATs. This is due in part to underutilization of FAT diagnosis codes and the composition of patients enrolled in this study (all surgical patients). In clinical practice, the percentage of all AIs identified with a score of C4 would likely be even smaller, making heightened awareness and definitive referral a reasonable objective for any clinical setting. The 14.3 % incidence of FATs in the present study falls at the lower end of the range of FATs reported in the literature [8, 13, 20–22]. Because there is no specific diagnostic code for ‘‘subclinical’’ Cushing’s syndrome or ‘‘subclinical’’ pheochromocytoma, one explanation for this discrepancy may be that only patients with overt clinical signs and symptoms preoperatively were labeled with a
World J Surg
FAT diagnosis by a surgeon. Hence, those with subclinical FATs were not identified in the present study. If this were true, the addition of more FAT patients would likely produce a stronger model and more robust scoring system. Additionally, the breakdown of FATs in the present study was also somewhat different than that in the reported literature—most likely a result of the population being only surgical patients and the potential underreporting of subclinical disease. Conn’s syndrome was the most common FAT in the current study, comprising 46.7 % of the FATs, although it usually represents only 5–10 % of adrenal tumors [23]. Cushing’s syndrome, normally the most common diagnosis, making up 1–47 % of adrenal tumors in various series, only accounted for 33 % of FATs in the NSQIP data set [4, 8, 9, 23–26]. Lastly, pheochromocytomas made up 15.2 % of the FATs, whereast in some surgical series it is the most common and has been reported to have an overall prevalence of 1–23 % [24, 27–29]. There are several limitations to this study. Subclinical Cushing’s and subclinical pheochromocytoma have no specific diagnostic codes and are not always recognized as important diagnoses outside of the endocrine community. Also, these conditions have no readily identifiable clinical characteristics. Thus, some or all of these tumors may be classified as ‘‘nonfunctional,’’ which would lower the sensitivity for the FAT model. However, the study’s main limitation is patient composition of the NSQIP data set. Because the entire data set was comprised of surgical patients, the present study does not include patients who were appropriately evaluated and determined to have a small, nonfunctional adrenal tumor not requiring surgery, had never undergone an endocrine hormone workup and/or follow-up after documentation of an AI, or who displayed endocrine laboratory results consistent with a subclinical FAT but were not properly referred for surgery according to the most recent guidelines.
Conclusions After external validation, the proposed risk estimator could serve as a quick, easy screening mechanism for patients with a newly discovered AI. Although its predictive power may be limited, it can help identify patients at high risk for FATs who need more urgent referral to an endocrine specialist.
Conflicts of interest Disclosure
None.
None.
References 1. Chidiac RM, Aron DC (1997) Incidentalomas: a disease of modern technology. Endocrinol Metab Clin North Am 26:233–253 2. Kloos RT, Korobkin M, Thompson NW et al (1997) Incidentally discovered adrenal masses. Cancer Treat Res 89:263–292 3. Hammarstedt L, Muth A, Wa¨ngberg B et al (2010) Adrenal lesion frequency: a prospective, cross-sectional CT study in a defined region, including systematic re-evaluation. Acta Radiol 51:1149–1156 4. Androulakis II, Kaltsas G, Piaditis G et al (2011) The clinical significance of adrenal incidentalomas. Eur J Clin Invest 41:552–560 5. Muth A, Hammarstedt L, Hellstro¯m M et al (2011) Cohort study of patients with adrenal lesions discovered incidentally. Br J Surg 98:1383–1391 6. Vassiliadi DA, Tsagarakis S (2011) Endocrine incidentalomas: challenges imposed by incidentally discovered lesions. Nat Rev Endocrinol 7:668–680 7. Davenport C, Liew A, Doherty B et al (2011) The prevalence of adrenal incidentaloma in routine clinical practice. Endocrine 40:80–83 8. Ng VW, Ma RC, So WY et al (2010) Evaluation of functional and malignant adrenal incidentalomas. Arch Intern Med 170:2017–2020 9. Barzon L, Sonino N, Fallo F et al (2003) Prevalence and natural history of adrenal incidentalomas. Eur J Endocrinol 149:273–285 10. Rossi R, Tauchmanova L, Luciano A et al (2000) Subclinical Cushing’s syndrome in patients with adrenal incidentaloma: clinical and biochemical features. J Clin Endocrinol Metab 85:1440–1448 11. Tauchmanova` L, Rossi R, Biondi B et al (2002) Patients with subclinical Cushing’s syndrome due to adrenal adenoma have increased cardiovascular risk. J Clin Endocrinol Metab 87:4872–4878 12. Angeli A, Osella G, Reimondo G et al (2002) Adrenal incidentalomas and subclinical Cushing’s syndrome: is there evidence for glucocorticoid-induced osteoporosis? Front Horm Res 30:73–85 13. Terzolo M, Stigliano A, Chiodini I et al (2011) AME position statement on adrenal incidentaloma. Eur J Endocrinol 164:851–870 14. Zeiger MA, Thompson GB, Duh QY et al (2009) American association of clinical endocrinologists and american association of endocrine surgeons medical guidelines for the management of adrenal incidentalomas: executive summary of recommendations. Endocr Pract 15:450–453 15. Anonymous (2002) NIH state-of-the-science statement on management of the clinically inapparent adrenal mass (‘‘incidentaloma’’). NIH Consens State Sci Statements 19(2):1–25 16. Khuri SF (2005) The NSQIP: a new frontier in surgery. Surgery 138:837–843 17. Khuri SF, Daley J, Henderson W et al (1995) The national veterans administration surgical risk study: risk adjustment for the comparative assessment of the quality of surgical care. J Am Coll Surg 180:519–531 18. Gregoire JR (1994) Adjustment of the osmostat in primary aldosteronism. Mayo Clin Proc 69:1108–1110 19. Reincke M, Rump LC, Quinkler M et al (2009) Risk factors associated with a low glomerular filtration rate in primary aldosteronism. J Clin Endocrinol Metab 94:869–875 20. Mantero F, Arnaldi G (2000) Management approaches to adrenal incidentalomas: a view from Ancona, Italy. Endocrinol Metab Clin North Am 29:107–125 ix
123
World J Surg 21. Barzon L, Fallo F, Sonino N et al (2002) Development of overt Cushing’s syndrome in patients with adrenal incidentaloma. Eur J Endocrinol 146:61–66 22. Cawood TJ, Hunt PJ, O’Shea D et al (2009) Recommended evaluation of adrenal incidentalomas is costly, has high falsepositive rates and confers a risk of fatal cancer that is similar to the risk of the adrenal lesion becoming malignant; time for a rethink? Eur J Endocrinol 161:513–527 23. Bittner JG, Brunt LM (2012) Evaluation and management of adrenal incidentaloma. J Surg Oncol 106:557–564 24. Grogan RH, Mitmaker E, Vriens MR et al (2010) Adrenal incidentaloma: does an adequate workup rule out surprises? Surgery 148:392–397 25. Piaditis GP, Kaltsas GA, Androulakis II et al (2009) High prevalence of autonomous cortisol and aldosterone secretion from adrenal adenomas. Clin Endocrinol (Oxf) 71:772–778
123
26. Terzolo M, Pia A, All A et al (2002) Adrenal incidentaloma: a new cause of the metabolic syndrome? J Clin Endocrinol Metab 87:998–1003 27. Kloos RT, Gross MD, Francis IR et al (1995) Incidentally discovered adrenal masses. Endocr Rev 16:460–484 28. Mantero F, Terzolo M, Arnaldi G et al (2000) A survey on adrenal incidentaloma in Italy: study group on adrenal tumors of the Italian society of endocrinology. J Clin Endocrinol Metab 85:637–644 29. Mansmann G, Lau J, Balk E et al (2004) The clinically inapparent adrenal mass: update in diagnosis and management. Endocr Rev 25:309–340
Risk Estimator for Adrenal Tumor Functionality Claire Sadler • Melanie Goldfarb
Ó Socie´te´ Internationale de Chirurgie 2014
Abstract Background Adrenal lesions are a common imaging finding with a prevalence approaching 10 %. Although guidelines recommend dedicated laboratory tests to rule out tumor functionality, many patients never undergo this workup. This study investigates the use of demographic and clinical variables to create an easy scoring system for predicting adrenal tumor functionality (functional adrenal tumors, or FATs). Methods Altogether, 2,807 patients in the NSQIP 2005–2010 database underwent adrenalectomy as their principal operation and had a postoperative diagnosis consistent with an adrenal lesion/disorder. Patients were divided into two groups based on a postoperative diagnosis consistent with tumor functionality. Univariate and multivariate logistic regression analyses were performed to identify specific predictors of FATs and for Cushing’s, Conn’s, or pheochromocytoma. Results Overall, 13.2 % (n = 402) of adrenalectomies performed were for FATs. Patients with a FAT were younger (age \40, p \ 0.01), overweight (BMI [ 30 kg/ m2, p \ 0.01), hypertensive (p \ 0.001). They also had elevated white blood cells (WBC [ 11, p \ 0.001), serum creatinine (Cr [ 1.25 mg/dl, p \ 0.001), and sodium (Na [ 143 mmol/L, p \ 0.001). On multivariate regression, patients with these characteristics were 20.53 times Presented at the 84th Annual Pacific Coast Surgical Meeting, Kauai, Hawaii, February 17, 2013. C. Sadler M. Goldfarb (&) Division of Breast/Soft Tissue and Endocrine Surgery, Department of Surgery, University of Southern California Keck School of Medicine, 1510 San Pablo Street, Suite 412 k, Los Angeles, CA 90033, USA e-mail: [email protected]
(CI 15.79–25.27) more likely to have a FAT (model c-statistic 0.634, CI 0.605–0.663; Hosmer–Lemeshow test (H–L), p = 0.035). Patients who were younger (p \ 0.001), female (p \ 0.001), diabetic (p = 0.07), overweight (p = 0.027), with elevated WBCs (p \ 0.001) and lower Cr (p \ 0.001) were 63.62 times (CI 58.03–69.21) more likely to have Cushing’s (model c-statistic 0.685, CI 0.648–0.722; H–L p = 0.954). Conclusions After external validation, this risk estimator might be used to quantify the probability of tumor functionality in patients with incidental adrenal masses. Although predictive power may be limited, it helps identify patients at high risk for FATs that need more urgent referral to a specialist.
Introduction Incidentally discovered adrenal lesions, or adrenal incidentalomas (AIs), are an increasingly common finding. With the increasing use of computed tomography (CT) or magnetic resonance imaging (MRI) to investigate almost any abdominal complaint, the reported prevalence of these lesions now approaches 5 % and reaches 10 % in older individuals [1–7]. Although the majority of AIs are benign, many produce subclinically elevated amounts of hormones [8, 9]. Left untreated, these ‘‘subclinical’’ functional tumors, where patients produce excess levels of hormones without recognizable pathology, are associated with increased morbidity [4, 6, 10–12]. However, with the exception of overt pheochromocytomas, these functional tumors, unlike a primary adrenal malignancy or metastatic tumor, do not demonstrate any distinct radiographic findings. Therefore, current guidelines recommend that all
123
World J Surg
patients with an AI undergo a complete endocrine evaluation to rule out any tumor functionality [13–15]. Despite these published recommendations from both medical and surgical endocrine societies, many patients with AIs fail to receive an endocrinology referral or undergo the appropriate workup by a clinician with adrenal tumor experience. This is especially true when AIs are detected in an emergency room or acute setting and not uncommonly at large public hospitals. Many adrenal lesions are not even reported [3]. This may be due to other, more pressing concerns at the time, angst and therefore dismissal by the patient, or a patient not being made aware of the diagnosis. The purpose of this study is therefore to develop an easy clinical scoring system that can be used in emergency and primary care settings to identify patients at highest risk for a functional adrenal tumor (FAT).
Methods Data retrieval The general approach and methods of data collection for the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database are described elsewhere [16, 17]. The ACS-NSQIP collects data from trained surgical clinical reviewers and includes preoperative, operative, and postoperative variables for 30 days following the procedure. All cases in the ACS-NSQIP 2005–2010 participant use file that documented adrenalectomy as the principal operation (current procedural terminology codes 60540 and 60650) and had a postoperative diagnosis—using international classification of disease (ICD-9) codes—consistent with an adrenal lesion or disorder were identified (Table 1). A total of 2,807 subjects were included in the final statistical analysis. Patients were divided into two groups based on tumor functionality: Patients with a postoperative diagnosis of Conn’s syndrome, Cushing’s syndrome, or pheochromocytoma were categorized as having functional adrenal tumors (FATs). All other neoplasms were classified as nonfunctional. Some variables that were not analyzable using a binary code were regrouped and classified as follows based on published literature or standard reference ranges: diabetes (none versus oral medication/insulin), dyspnea (none versus any), anesthesia technique (general versus other), young age (B45 years), American Society of Anesthesiologists (ASA) class I/II (mild) and class III/IV (severe), elevated creatinine (Cr [ 1.25 mg/dl), elevated sodium (Na [ 143 mmol/ L), elevated white blood cell count (WBC [ 11 9 109/L), low hematocrit (Hct B 36.0 %), and low alkaline phosphatase (AP) (AP \ 150 IU/L). All wound infections were
123
grouped into one category. Laboratory cutoff values are based on previously identified cutoffs for abnormal values that can be seen in adrenal dysfunction [18, 19]. The BMI was calculated from height and weight variables, with ‘‘obese’’ defined as BMI [ 30 kg/m2. Missing laboratory values were not imputed. Statistical analysis Preoperative laboratory values, demographics, and comorbidities were compared between the FAT and nonfunctional groups. Univariate analysis was performed to identify specific predictors of FAT and for the specific FAT syndromes: Cushing’s, Conn’s, and pheochromocytoma. Categoric variables were analyzed using Pearson’s v2 test and continuous variables with the Mann–Whitney U test. A complete multivariate regression model was constructed for FAT and each FAT syndrome using only patients with complete data. We first entered all demographic and preoperative values into a forward logistic regression. A reduced model was then generated that included only variables that were statistically significant (p \ 0.05). Receiver operating characteristic (ROC) curves and the Hosmer–Lemeshow (H–L) test were used to assess model quality. Then c-statistics were abstracted from ROC curves to assess for discriminative ability of the model, with a higher c-statistic indicating a better ability of the model to distinguish functional and nonfunctional adrenal tumor groups. A higher H–L result indicated better calibration of the model, meaning that the predicted probabilities agree with the outcomes. A simple scoring system derived from the slope, or log-odds of the reduced model was created by multiplying each log-odds by 2 and then rounding to the Table 1 International statistical classification of diseases and healthrelated problems, 9th revision (ICD-9) codes utilized 194.0: Malignant neoplasm of adrenal gland
255.5: Other adrenal hypofunction
227.0: Benign neoplasm of adrenal gland 237.2 Neoplasm of uncertain behavior of adrenal gland
255.6: Medulloadrenal hyperfunction 255.8: Other specified disorders of adrenal glands
239.7: Neoplasm of uncertain nature of endocrine glands
255.9: Unspecified disorder of adrenal glands
255.0: Cushing’s syndrome
258.0: Polyglandular activity in multiple endocrine adenomatosis 276.8: Hypopotassemia
255.1: Hyperaldosteronism 255.2: Adrenogenital disorders
401.0: Malignant essential hypertension
255.3: Other corticoadrenal overactivity
401.9: Unspecified essential hypertension
255.4: Corticoadrenal insufficiency
759.1: Anomalies of adrenal gland
13.8 % (8)
19 % (11)
0
14.3 % (8)
20.7 % (12)
18 % (11)
78.7 % (48)
32.8 % (20)
33.3 % (20)**
** p \ 0.05 and * p \ 0.01 compared to the overall cohort
AP alkaline phosphatase, BMI body mass index, Cr creatinine, Hct hematocrit, Pheo pheochromocytoma, WBC white blood cell count
16.1 % (22) 16.6 % (29) NS 17.5 % (392) Hct \ 36 %
16.5 % (62)
35 % (48)* 5.2 % (9)* \0.001 10.7 % WBC [ 11 9 109/L
17.4 %*
6.5 % (6) 1.2 % (1) NS 3.6 % (48) :AP
3.5 % (8)
6.7 % (9)
20 % (27) 19.3 % (34)
26.7 % (48)* \0.001
0.004 19.5 % (73)*
Cr [ 1.25 mg/dl
17.8 % (67)*
13.7 % (305)
10.8 % (239)
Na [ 143 mmol/L
19 % (458) Diabetes on meds
Laboratory values
70.1 % (103)
27.9 % (41)* 19.7 % (37)
95.2 % (179)*
0.072
\0.001 70.5 % (1,695) Hypertension on meds
83.1 % (334)*
46.2 % (1,100) BMI [ 30 kg/m2
22.9 % (92)
61.2 % (90)* 54.5 % (102)** 0.004
30.1 % (724) Age \45 years
54 % (216)*
60 % (36) 84.4 % (124)*
51 % (75)* 29.8 % (56)
48.4 % (91)* NS
\0.001
63.5 % (1,523) Female sex
38.6 % (155)
Nonfunctional n = 2,405 (85.7 %)
64.1 % (257)
nearest whole digit number. The scoring system was compared to the logistic regression model using the ROC curves. All statistical analyses used two-sided tests and were conducted using SPSS software, version 20.0 (SPSS, Chicago, IL, USA).
Results
Parameter
Table 2 Demographics of functional versus nonfunctional tumors
Functional n = 402 (14.3 %)
p
Conn’s n = 188
Cushing’s n = 147
Pheo n = 61
World J Surg
Among 2807 patients who underwent adrenalectomy for a primary adrenal tumor, 402 (14.3 %) of the tumors were FATs. Patients with a FAT were more likely young (p \ 0.01), obese (p \ 0.01), and hypertensive (p \ 0.001) with elevated WBC count (p \ 0.001), serum Cr (p \ 0.001), and serum Na (p \ 0.001) (Table 2). A complete multiple regression model was created that included all significant or nearly significant variables (p \ 0.05) in the univariate analysis. In all, 2424 patients had sufficiently complete data to be included in this portion of analysis. Forward regression resulted in six variables for use in modeling (Table 3). The model demonstrated moderate discriminative ability but poor calibration: c-statistic 0.634, CI 0.605–0.663; H–L, p = 0.035). When BMI was removed from the model, calibration increased, resulting in a better model (c-statistic 0.631, CI 0.602–0.660; H–L, p = 0.157). Based on this logistic regression model, a simplified scoring system was created (Table 4). When a cutoff point of 4 was used, the model resulted in 91.4 % specificity and 16.3 % sensitivity for predicting FAT. An ROC curve for this model demonstrated good discriminative ability, with a c-statistic of 0.622 (CI 0.592–0.651). A cutoff value of 3 (of 6) produced a model with a sensitivity of 53.7 % and a specificity of 62.8 %, whereas a cutoff value of 5 (of 6) yielded a model with 2.2 % sensitivity and 99.4 % specificity. Compared to the overall cohort, patients with Cushing’s syndrome (n = 147) were more likely female (p \ 0.001), diabetic (p = 0.07), young (p \ 0.001), and obese (p = 0.027) with an elevated WBC count (p \ 0.001) and lower Cr (p \ 0.001) (Table 2). A complete multiple regression model was created including all significant or nearly significant variables (p \ 0.05) in the univariate analysis. Altogether, 2490 patients had sufficiently complete data to be included in this portion of analysis. Forward logistic regression resulted in a pared-down regression model that included age, diabetes, elevated Na, and elevated WBC (Table 5). The model displayed both good discrimination and calibration (c-statistic 0.70, CI 0.650–0.748; H–L, p = 0.787). In contrast, patients with Conn’s syndrome (n = 188) were more likely male (p \ 0.001) and hypertensive (p \ 0.001) with elevated Cr (p \ 0 0.001) (Table 1). A complete multiple regression model including all significant
123
World J Surg Table 3 Multivariate analysis for adrenal tumor functionality (n = 402) Factor
Odds ratio of being functional
Confidence interval
Age \45 years
1.71
1.34–2.19
\0.001
BMI [ 30 kg/m2
1.29
1.03–1.63
0.029
Hypertension on meds
2.00
1.47–2.72
\ 0.001
WBC [ 11 9 109/L
1.76
1.29–2.40
0.029
Cr [ 1.25 mg/dl
1.72
1.26–2.34
0.001
Na [ 143 mmol/L
1.49
1.10–2.00
0.009
Table 5 Multivariate analyses for Cushing’s syndrome (n = 133) Factor
Odds ratio of being Cushing’s
Confidence interval
p
Diabetes
1.75
1.15–2.66
\0.01
p
Age \45 years old
2.37
1.63–3.43
\0.001
Elevated Na [ 143 mmol/ dl
1.77
1.12–2.81
0.015
WBC [ 11 9 109
4.24
2.87–6.24
\0.001
Table 6 Multivariate analysis for Conn’s syndrome (n = 180) Factor
Odds ratio of being Conn’s
Confidence interval
p
Table 4 Scoring system for adrenal tumor functionality Factor
Points
Hypertension requiring medication
2
Age \45 years
1
Cr [ 1.25 mg/dl
1
Na [ 143 mmol/L
1
WBC [ 11 9 109/L Maximum score
1 6
or nearly significant variables (p \ 0.05) in the univariate analysis was created. In all, 2,574 patients had sufficiently complete data to be included in this portion of analysis. Forward logistic regression resulted in a model that included only hypertension and sex, although because elevated serum Cr had a significance of exactly p = 0.05 it was included in the final model as well (Table 6). This model showed moderate discriminative ability and excellent calibration (c-statistic 0.685, CI 0.648–0.722; H–L, p = 0.954). Patients with pheochromocytoma were less likely to be obese, but no variables were sufficiently significant on multivariate regression to create a predictive model (Table 2).
Discussion Adrenal incidentalomas are a common imaging finding. However, with benign-appearing lesions, ensuring that patients receive an appropriate workup and subsequent follow-up for tumor functionality remains a challenge. Many adrenal tumors are discovered in an emergency room setting or an independent radiology facility, and patients are not made aware of their AI, not referred to an endocrine specialist, or the adrenal lesion is not reported. As failure to treat a FAT can negatively affect patient morbidity, is it important that all patients with an AI undergo hormonal evaluation [10, 11]. The proposed simple, easy-to-use scoring system could help clinicians identify patients most
123
Male sex
1.60
1.16–2.18
Hypertension on meds Creatinine [1.25 mg/dl
6.38
3.23–12.59
\0.001
0.004
2.13
1.47–3.07
\0.001
at risk for a FAT and prioritize which patients require a more urgent endocrine referral. The proposed risk estimator uses a cutoff value of 4 (of 6) to produce a model with a sensitivity of 16.3 % and a specificity of 91.4 % for predicting a FAT. As a screening tool, higher specificity is desired with sensitivity sacrificed because clinical guidelines recommend that all patients with an AI should undergo a thorough workup for both functionality and malignancy. The proposed scoring system would thus identify patients most at risk for having a FAT. Ideally, this would prompt a treating clinician to ensure that a referral and/or follow-up is truly obtained once such a patient is identified. In this study, approximately 10 % of patients had a score of C4 and would have been considered to be at high risk for a FAT. In reality, 25 % of these patients had a FAT based on their diagnosis code, which is likely an underestimate of the true number of patients with subclinical FATs. This is due in part to underutilization of FAT diagnosis codes and the composition of patients enrolled in this study (all surgical patients). In clinical practice, the percentage of all AIs identified with a score of C4 would likely be even smaller, making heightened awareness and definitive referral a reasonable objective for any clinical setting. The 14.3 % incidence of FATs in the present study falls at the lower end of the range of FATs reported in the literature [8, 13, 20–22]. Because there is no specific diagnostic code for ‘‘subclinical’’ Cushing’s syndrome or ‘‘subclinical’’ pheochromocytoma, one explanation for this discrepancy may be that only patients with overt clinical signs and symptoms preoperatively were labeled with a
World J Surg
FAT diagnosis by a surgeon. Hence, those with subclinical FATs were not identified in the present study. If this were true, the addition of more FAT patients would likely produce a stronger model and more robust scoring system. Additionally, the breakdown of FATs in the present study was also somewhat different than that in the reported literature—most likely a result of the population being only surgical patients and the potential underreporting of subclinical disease. Conn’s syndrome was the most common FAT in the current study, comprising 46.7 % of the FATs, although it usually represents only 5–10 % of adrenal tumors [23]. Cushing’s syndrome, normally the most common diagnosis, making up 1–47 % of adrenal tumors in various series, only accounted for 33 % of FATs in the NSQIP data set [4, 8, 9, 23–26]. Lastly, pheochromocytomas made up 15.2 % of the FATs, whereast in some surgical series it is the most common and has been reported to have an overall prevalence of 1–23 % [24, 27–29]. There are several limitations to this study. Subclinical Cushing’s and subclinical pheochromocytoma have no specific diagnostic codes and are not always recognized as important diagnoses outside of the endocrine community. Also, these conditions have no readily identifiable clinical characteristics. Thus, some or all of these tumors may be classified as ‘‘nonfunctional,’’ which would lower the sensitivity for the FAT model. However, the study’s main limitation is patient composition of the NSQIP data set. Because the entire data set was comprised of surgical patients, the present study does not include patients who were appropriately evaluated and determined to have a small, nonfunctional adrenal tumor not requiring surgery, had never undergone an endocrine hormone workup and/or follow-up after documentation of an AI, or who displayed endocrine laboratory results consistent with a subclinical FAT but were not properly referred for surgery according to the most recent guidelines.
Conclusions After external validation, the proposed risk estimator could serve as a quick, easy screening mechanism for patients with a newly discovered AI. Although its predictive power may be limited, it can help identify patients at high risk for FATs who need more urgent referral to an endocrine specialist.
Conflicts of interest Disclosure
None.
None.
References 1. Chidiac RM, Aron DC (1997) Incidentalomas: a disease of modern technology. Endocrinol Metab Clin North Am 26:233–253 2. Kloos RT, Korobkin M, Thompson NW et al (1997) Incidentally discovered adrenal masses. Cancer Treat Res 89:263–292 3. Hammarstedt L, Muth A, Wa¨ngberg B et al (2010) Adrenal lesion frequency: a prospective, cross-sectional CT study in a defined region, including systematic re-evaluation. Acta Radiol 51:1149–1156 4. Androulakis II, Kaltsas G, Piaditis G et al (2011) The clinical significance of adrenal incidentalomas. Eur J Clin Invest 41:552–560 5. Muth A, Hammarstedt L, Hellstro¯m M et al (2011) Cohort study of patients with adrenal lesions discovered incidentally. Br J Surg 98:1383–1391 6. Vassiliadi DA, Tsagarakis S (2011) Endocrine incidentalomas: challenges imposed by incidentally discovered lesions. Nat Rev Endocrinol 7:668–680 7. Davenport C, Liew A, Doherty B et al (2011) The prevalence of adrenal incidentaloma in routine clinical practice. Endocrine 40:80–83 8. Ng VW, Ma RC, So WY et al (2010) Evaluation of functional and malignant adrenal incidentalomas. Arch Intern Med 170:2017–2020 9. Barzon L, Sonino N, Fallo F et al (2003) Prevalence and natural history of adrenal incidentalomas. Eur J Endocrinol 149:273–285 10. Rossi R, Tauchmanova L, Luciano A et al (2000) Subclinical Cushing’s syndrome in patients with adrenal incidentaloma: clinical and biochemical features. J Clin Endocrinol Metab 85:1440–1448 11. Tauchmanova` L, Rossi R, Biondi B et al (2002) Patients with subclinical Cushing’s syndrome due to adrenal adenoma have increased cardiovascular risk. J Clin Endocrinol Metab 87:4872–4878 12. Angeli A, Osella G, Reimondo G et al (2002) Adrenal incidentalomas and subclinical Cushing’s syndrome: is there evidence for glucocorticoid-induced osteoporosis? Front Horm Res 30:73–85 13. Terzolo M, Stigliano A, Chiodini I et al (2011) AME position statement on adrenal incidentaloma. Eur J Endocrinol 164:851–870 14. Zeiger MA, Thompson GB, Duh QY et al (2009) American association of clinical endocrinologists and american association of endocrine surgeons medical guidelines for the management of adrenal incidentalomas: executive summary of recommendations. Endocr Pract 15:450–453 15. Anonymous (2002) NIH state-of-the-science statement on management of the clinically inapparent adrenal mass (‘‘incidentaloma’’). NIH Consens State Sci Statements 19(2):1–25 16. Khuri SF (2005) The NSQIP: a new frontier in surgery. Surgery 138:837–843 17. Khuri SF, Daley J, Henderson W et al (1995) The national veterans administration surgical risk study: risk adjustment for the comparative assessment of the quality of surgical care. J Am Coll Surg 180:519–531 18. Gregoire JR (1994) Adjustment of the osmostat in primary aldosteronism. Mayo Clin Proc 69:1108–1110 19. Reincke M, Rump LC, Quinkler M et al (2009) Risk factors associated with a low glomerular filtration rate in primary aldosteronism. J Clin Endocrinol Metab 94:869–875 20. Mantero F, Arnaldi G (2000) Management approaches to adrenal incidentalomas: a view from Ancona, Italy. Endocrinol Metab Clin North Am 29:107–125 ix
123
World J Surg 21. Barzon L, Fallo F, Sonino N et al (2002) Development of overt Cushing’s syndrome in patients with adrenal incidentaloma. Eur J Endocrinol 146:61–66 22. Cawood TJ, Hunt PJ, O’Shea D et al (2009) Recommended evaluation of adrenal incidentalomas is costly, has high falsepositive rates and confers a risk of fatal cancer that is similar to the risk of the adrenal lesion becoming malignant; time for a rethink? Eur J Endocrinol 161:513–527 23. Bittner JG, Brunt LM (2012) Evaluation and management of adrenal incidentaloma. J Surg Oncol 106:557–564 24. Grogan RH, Mitmaker E, Vriens MR et al (2010) Adrenal incidentaloma: does an adequate workup rule out surprises? Surgery 148:392–397 25. Piaditis GP, Kaltsas GA, Androulakis II et al (2009) High prevalence of autonomous cortisol and aldosterone secretion from adrenal adenomas. Clin Endocrinol (Oxf) 71:772–778
123
26. Terzolo M, Pia A, All A et al (2002) Adrenal incidentaloma: a new cause of the metabolic syndrome? J Clin Endocrinol Metab 87:998–1003 27. Kloos RT, Gross MD, Francis IR et al (1995) Incidentally discovered adrenal masses. Endocr Rev 16:460–484 28. Mantero F, Terzolo M, Arnaldi G et al (2000) A survey on adrenal incidentaloma in Italy: study group on adrenal tumors of the Italian society of endocrinology. J Clin Endocrinol Metab 85:637–644 29. Mansmann G, Lau J, Balk E et al (2004) The clinically inapparent adrenal mass: update in diagnosis and management. Endocr Rev 25:309–340
