| OBJECTIVEPCCs are catecholamine-secreting neoplasms of the sympathochromaffin system. They usually arise from the adrenal medulla, although about one-tenth of tumors may arise from extra-adrenal chromaffin tissue (extra-adrenal PCCs or paragangliomas). PCCs may be malignant, as documented by the presence of lymph node, bone, or visceral metastasis either at first operation or at recurrence. Extra-adrenal PCCs have the higher incidence of malignance than those arising from the adrenal gland. Tumor recurrence may occur months or years after the initial operation. Identifying primary malignant PCCs remains essential for establishing optimal treatment and following-up. However the distinction of malignant from benign PCCs has been difficult although its histological diagnosis is relatively easy.Some researchers attempted to find approaches in the following areas: tumor cell necrosis extensive local invasion, vascular invasion, the significance of angiogenesis, the amount of mitotic figures, tumor molecular marker, flow cytometric DNA, and gene expression abnormality et al. Some features were found to be associated with the malignance of PCCs. However, to date, it is commonly accepted that no single feature is diagnostic for malignant PCCs without documented metastatic diseaseIt would seem reasonable that the combined and simultaneous use of several different variables, all of them independent predictors of malignancy, would offer better diagnostic performance than a single variable.Lester D.R. Thompson developed PASS system to predict the prognosis of patients with PCCs in 2002. In the present investigation, ROC curve analysis was first introduced to evaluate the diagnostic performance of the PASS in a group of 130 patients from our institution. And then we develop a logistic model encompassing the use of 15 variables to discriminate between benign and malignant PCCs, and compared its diagnostic performance with PASS using ROC curve analysis. MATERIAL AND METHODSBetween 1965 and 2005 all 796 consecutive patients were diagnosed as "PCC," "PCC with malignant features," and "malignant PCC" at our institution. The classi- fication of "malignant" was mainly based on the clinical behavior (recurrence or metastasis); The PCCs without malignant clinic behavior during the follow-up of at least 10 years were regarded as an assumed benign classification in this investigation. As a result, only 130 patients who had complete clinic material, adequate follow-up information, and clear eosin-stained slides qualified for this study.In all 130 cases with malignant or benign PCCs data had been collected retrospectively. 15 observations were recorded as follow: gender; age; tumor size (greatest dimension in centimeters); capsular invasion; vascular invasion (defined as either gross or microscopic involvement of veins or lymphatic channels); periadrenal adipose tissue invasion; necrosis (individual cell, focal [identified in the center of large nests], and confluent or diffuse type); cell nests (small, "zellballen-type" nests, to large nests [a "large nest" was defined as three to four times the size of a "zellballen" or the normal size of the medullary paraganglia nests] or diffuse growth); cellularity (divided into low [greater amount of area accounted for by a few cells with ample cytoplasm], moderate [intermediate between low and high], and high [greater amount of area accounted for by many cells with a high nuclear-tocytoplasmic ratio); cellular monotony; spindle cell (when present, the percent of tumor area spindled); nuclear pleomorphism (mild, moderate, profound); nuclear hyperchromasia, mitotic figures (number of mitoses per 10 high power fields [HPF] [magnification at×40 with a×10 objective lens using an Olympus BX40 microscope]); atypical mitotic figure (defined by abnormal chromosome spread, tripolar or quadripolar forms, circular forms, or indescribably bizarre). This study series spanned nearly half a century. It was not reasonable to adopt the original pathological report. So personnel at the pathology unit at our institution were requested to review all specimens again.1. The score of each case in PASS was obtained using the method of Thompson. L. D. ROC curve analysis was used to evaluate the diagnostic performance of the PASS.2. A binary logistic regession model was constructed to discriminate between malignant and benign PCCs. All these variables entered in the logistic regression analysis in a backward stepwise way. Continuous variables, such as age and tumor volume, were transformed into categorical variables by grouping them. Each categorical variable was classified into two groups, negative and positive group, coded as 0 and 1 respectively. The probability of malignant PCCs was estimated by applying the following mathematical formula: 1/ (1+exp[-Z]), where Z was the linear predictor in the model.3. To compare the diagnostic performance of the logistic model, each individual variable, and PASS using ROC curve analysis.4. The diagnostic performance of the logistic model including three selected variables which had comparative high OR was evaluated by ROC curve analysis, and was compared with that of the logistic model including all variables.All statistical analyses were performed using SAS software (SAS Institute Inc. Cary, North Carolina, USA). P<0.05 was considered statistically significant.RESULT1. In ROC analysis, PASS had the area under ROC curve of 0.899 (95% CI, 0.844 to 0.954).2. 15 variables were observed and entered in the logistic regression analysis, and 9 variables were retained in this model. High cellularity had the largest OR (92.652), followed by spindle cell (>10% of tumor volume), atypical mitotic figure, periadrenal adipose tissue invasion, mitotic figures (> 3/10 HPF), cellular monotony, capsular invasion, vascular invasion, and central or confluent tumor necrosis. The constant parameter had the OR of 0.001. The linear predictor in the model was as follows: Z=-7.44+ 4.529 (high cellularity) + 4.048 spindle cell (>10% of tumor volume) + 3.838 (atypical mitotic figure) + 3.409 (periadrenal adipose tissue invasion)+ 3.231 (mitotic figures [> 3/10 HPF])+ 1.953 (cellular monotony)+ 1.764 (capsular invasion)+1.674 (vascular invasion) -1.933 (central or confluent tumor necrosis). The logistic model had the best area under ROC curve (0.983; 95% confidence interval [CI], 0.967 to 0.998), being significantly higher than high cellularity (0.863; 95% CI 0.798-0.929), spindle cell (>10% of tumor volume) (0.714; 95% CI 0.617-0.811), atypical mitotic figure (0.622; 95% CI 0.519-0.726), periadrenal adipose tissue invasion (0.659; 95% CI 0.556-0.762), mitotic figures (> 3/10 HPF) (0.622; 95% CI 0.519-0.726), cellular monotony (0.686; 95% CI 0.584-0.788), capsular invasion (0.665; 95% CI 0.566-0.765), vascular invasion (0.748; 95% CI 0.656-0.840), central or confluent tumor necrosis (0.621; 95% CI 0.517-726).3. The diagnostic performance of PASS was improved using this logistic model. The area under the ROC increased form 0.899 (95% CI, 0.844 to 0.954) to 0.983 (95% CI, 0.967 to 0.998).4. High cellularity, spindle cell (>10% of tumor volume) and atypical mitotic figure were selectd to built a logistic model. This model had the area under the ROC curve of 0.927 ( 95% confidence interval, 0.883 to 0.971).COMMENTThe prevalence of PCCs ranges from 1.2 to 13% of all benign and malignant primary adrenal gland neoplasms; the incidence of malignant PCCs varies in different series from 8% to 12.5% of all PCCs. The distinction of benign and malignant PCCs remains a great diagnostic difficulty.PASS by Lester D.R. Thompson may improve the situation, and then we think it is worth evaluating its diagnostic performance.ROC curves have been widely accepted as the standard method for describing and comparing the accuracy of medical diagnostic tests. The accuracy of a diagnostic test is characterized by the sensitivity (detection of disease when disease is truly present) and specificity (recognition of disease absence when the disease of interest is truly absent). A ROC curve displays the sensitivity of a diagnostic test over all possible false-positive rates. Two or more diagnostic tests can be compared at any or all false-positive rates. Both components of accuracy, i.e., sensitivity and specificity can also be incorporated into a single measure of accuracy such as the area under the ROC. In this clinical series, PASS worked well in differentiating between benign and malignant PCCs, having an area under ROC curve of xFor getting better diagnostic performance, we next, mainly based on variables from PASS, developed a logistic model. Logistic regression is a commonly accepted statistical tool. The rationale for using these approaches is the possibility of using, simultaneously and combined, a series of features of tumor in the way of more accurate mathematical formula to predict benignity or malignancy of PCCs. In this investigation, Of the 15 variables entered in the backward stepwise logistic regression analysis, 9 were retained in the model. The variables scored a high weighted scale in PASS had, In general, high OR in this logistic model, which validated PASS further. ROC curves were used to test diagnostic performance. The result showed that the area under the ROC curve of this logistic model was greater than that of each variable, confirming the hypothesis proposed as above. Moreover, the diagnostic performance of PASS was improved using this logistic model, the area under the ROC increasing form 0.899 (95% CI, 0.844 to 0.954) to 0.983 (95% CI, 0.967 to 0.998).As a whole, this logistic model is in agreement with PASS, but inconsistence could also be found. One variable (central or confluent tumor necrosis) weighted two point scores in PASS had lower OR in logistic model, and three variables (large nest, profound nuclear pleomorphism, and nuchlear hyperchromasia) in PASS were not included in the logistic model. The inconsistence should be partly induced by the characteristic of logistic regression. In this kind of statistic analysis, if one variable contained repetitive information already included in other, its OR may have decreased, even be removed from the model.Even with the same logistic analysis, nothing could assure complete consistent results from independent groups. No logistic models have been introduced to predict malignant PCCs before, but some prospective evaluation of logistic regression models for the diagnosis of other tumor have once yielded inconsistent results.Consequently, how to improve the reproducibility of the model by different and independent groups is an important question. It should be taken into account that the definitions of the variables might be interpreted in different ways by independent investigators when applying logistic models or PASS system. That makes it essential to use reproducible and clearly defined variables. Therefore, such protean or less reproducibility variables as hypertension, levels of catecholamines in the blood or urine have not been included, although they are most common and obvious features in PCC patients. Increasing number of cases can also improve the reproducibility of the model, but PCCs, especially malignant PCCs, have a low incidence. So accumulation of PCC cases with complete data is continuous work.On the other hand, an ideal model should be as simple as possible, regarding calculating, the PASS may have an advantage, but logistic model can present possibility of malignant PCCs in the way of percentage, and it should be more easy to interpret this possibility than PASS. As matter of fact, the logistic model was only a simple mathematical formula. Once all variables used in the model are obtained, it is easy to get the probability of malignant pheochromocytoma with only a computer or even a simple hand calculator. Decreasing the number of variables in this model can make it simpler, but its performance can be affected. The ROC analysis showed that the performance of the simplified logistic model was also practical.CONCLUSIONSWe proposed a logistic model for predicting malignancy in PCCs for the first time. The model was significantly more accurate than PASS in diagnostic performance. Among all variables, high cellularity, spindle cell (>10% of tumor volume), and atypical mitotic figure were the strongest predictors. The estimation of malignant PCCs may be useful for the stratification of patients for treatment, such as additional surgery, adjuvant therapies, or close clinical follow-up. However, we still emphasize that this is only a preliminary study. A validation of our model in another independent series including more cases is needed to draw a definitive picture. |
PDF Full Text Request