Patrizia Agati

A classification tree approach for pituitary adenomas

It is difficult to evaluate the recurrence and progression potential of pituitary adenomas at presentation. The World Health Organization classification of endocrine tumors suggests that invasion of the surrounding structures, size at presentation, an elevated mitotic index, a Ki-67 labeling index higher than 3%, and extensive p53 expression are indicators of aggressive behavior. Nevertheless, Ki-67 and p53 labeling index evaluation is subject to interobserver variability, and their cutoff values are controversial. In the present study, the prognostic value of Ki-67 and p53 protein labeling indices and their correlation with clinical and radiologic parameters were evaluated using digital image analysis in a series of 166 pituitary adenomas in patients having undergone a follow-up of at least 6 years to evaluate the impact on the recurrence and progression potential of pituitary adenomas. The data were analyzed using the receiver operating characteristic curve and classification and regression tree analysis. The results showed that, in the unstratified data set, the commonly used threshold of the Ki-67 index of 3% has a high specificity (89.5%) but a low sensitivity (53.8%). Unsatisfactory performance results were obtained by performing receiver operating characteristic curve analysis on the p53 labeling index. On the contrary, the classification and regression tree analysis-derived tree demonstrated that each pituitary adenoma subtype has specific prognostic factors. Specifically, the Ki-67 labeling index is a useful prognostic factor in nonfunctioning, adrenocorticotropin, and prolactin adenomas, but with different thresholds. In conclusion, our study emphasizes that the term pituitary adenomas includes different types of tumors, each one having specific prognostic factors.

The combined role of biomarkers and interim PET scan in prediction of treatment outcome in classical Hodgkin's lymphoma: a retrospective, European, multicentre cohort study

BACKGROUNDnEarly-interim fluorodeoxyglucose (FDG)-PET scan after two ABVD (doxorubicin, bleomycin, vinblastine, and dacarbazine) chemotherapy courses (PET-2) represents the most effective predictor of treatment outcome in classical Hodgkins lymphoma. We aimed to assess the predictive value of PET-2 combined with tissue biomarkers in neoplastic and microenvironmental cells for this disease.nnnMETHODSnWe enrolled 208 patients with classical Hodgkins lymphoma and treated with ABVD (training set), from Jan 1, 2002, to Dec 31, 2009, and validated the results in a fully matched independent cohort of 102 patients with classical Hodgkins lymphoma (validation set), enrolled from Jan 1, 2008, to Dec 31, 2012. The inclusion criteria for both the training and validation sets were: the availability of a representative formalin-fixed, paraffin-embedded tissue sample collected at diagnosis; treatment with ABVD with or without radiotherapy; baseline staging and interim restaging after two ABVD courses with FDG-PET; no treatment change based solely on interim PET result; and HIV-negative status. We used Cox multivariate analysis classification and regression tree (CART) to compare the predictive values of these markers with that of PET-2 and to assess the biomarkers ability to correctly classify patients whose outcome was incorrectly predicted by PET-2.nnnFINDINGSnIn multivariate analysis, PET-2 was the only factor able to predict both progression-free survival (hazard ratio [HR] 33·3 [95% CI 13·6-83·3]; p<0·0001) and overall survival (HR 31·3 [95% CI 3·7-58·9]; p=0·002). In the training set, no factor had a stronger adverse predictive value than a positive PET-2 scan and none was able to correctly reclassify PET-2 positive patients. In PET-2 negative patients, expression of CD68 (≥25%) and PD1 (diffuse or rosetting pattern) in microenvironmental cells, and STAT1 negativity in Hodgkin Reed Sternberg cells identified a subset of PET-2 negative patients with a 3 year progression-free survival significantly lower than that of the remaining PET-2 negative population (21 [64%] of 33 [95% CI 45·2-79·0] vs 130 [95%] of 137 [95% CI 89·4-97·7]; p<0·0001). These findings were reproduced in the validation set.nnnINTERPRETATIONnThe CART algorithm correctly predicted the response to treatment in more than a half of patients who had a relapse or disease progression despite a negative PET-2 scan, thus increasing the negative predictive value of PET-2. In keeping with preliminary results from interim PET response adapted clinical trials of patients with advanced Hodgkins lymphoma, there might be a non-negligible proportion of treatment failures in the interim PET negative group treated with standard ABVD.nnnFUNDINGnItalian Association for Cancer Research, Bologna Association against leukaemia, lymphoma and myeloma, and Bologna University.

Exploring unobserved heterogeneity in perinatal and neonatal mortality risks: The case of an Italian sharecropping community, 1900–39

Previous researchers have found that traditional determinants explain only a limited part of the variation in perinatal and infant mortality at the family level. In the study reported in this paper, we explored the factors that make the perinatal/neonatal death risk more heterogeneous across families. We estimated logistic regressions with cluster random effects at the maternal level, using data from the Italian village of Granarolo from 1900 to 1939. We estimated the effects of selected predictors on perinatal/neonatal mortality and unexplained inter-family variation. We found that non-rural skilled and lower-skilled workers experienced higher perinatal and neonatal mortality risks. Unexplained heterogeneity at the maternal level was lower for women living in sharecropper families than for those in landless labourer and non-rural worker families. Unexplained perinatal and neonatal mortality components were also due to socio-economic differences and were not necessarily related only to maternal biological features or shared genetic frailty.

Collagen fibre and fibril ultrastructural arrangement of the superficial medial collateral ligament in the human knee

AbstractPurposeThe aim of the study was to investigate the collagen fibre ultrastructural arrangement and collagen fibril diameters in the superficial medial collateral ligament (sMCL) in the human knee. Considering sMCL’s distinctive functions at different angles of knee flexion, it was hypothesized a significant difference between the collagen fibril diameters of each portion of the sMCL.MethodsnFourteen sMCL from seven fresh males (by chance because of the availability) cadavers (median age 40xa0years, range 34–59xa0years) were harvested within 12xa0h of death. sMCLs were separated into two orders of regions for analysis. The first order (divisions) was anterior, central and posterior. Thereafter, each division was split into three regions (femoral, intermediate and tibial), generating nine portions. One sMCL from each cadaver was used for transmission electron microscopy (TEM) and morphometric analyses, whereas the contralateral sMCL was processed for light microscopy (LM) or scanning electron microscopy (SEM).ResultsLM and SEM analyses showed a complex tridimensional architecture, with the presence of wavy collagen fibres or crimps. TEM analysis showed significant differences in median collagen fibril diameter among portions inside the anterior, central and posterior division of the sMCL (pxa0<xa00.0001 within each division). Significant differences were also present among the median [interquartile range] collagen fibril diameters of anterior (39.4 [47.8–32.9]), central (38.5 [44.4–34.0]) and posterior (41.7 [52.2–35.4]) division (pxa0=xa00.0001); femoral (38.2 [45.0–32.7]), intermediate (40.3 [47.3–36.1]) and tibial (40.7 [55.0–32.2]) region (pxa0=xa00.0001).ConclusionsHuman sMCL showed a complex architecture that allows restraining different knee motions at different angles of knee flexion. The posterior division of sMCL accounted for the largest median collagen fibril diameter. The femoral region of sMCL accounted for the smallest median collagen fibril diameter. The presence of crimps in the medial collateral ligament, previously identified in the rat, was confirmed in humans (taking into consideration differences between these two species).

Predicting the evolution of a constrained network: a beta regression model

Social network analysis allows to map and measure relationships and flows (links) between people, groups, computers, URLs, or other connected knowledge entities (nodes). In this context, a relevant issue is the treatment of constrained scale-free networks such as the network of student transfers between degree courses offered by an University, that are strongly influenced by a number of institutional decisions. In the analysis of such a system, special attention has to be paid to identify current or future “critical points”, that is nodes characterized by a high number of outcoming or incoming links, on which to act in order to optimize the network. To predict the evolution of a constrained system over time in dependence of constraint modifications, a beta regression model is proposed, that fits links represented by quantities varying between 0 and 1. The algorithm was successfully applied to the network of student transfers within the University of Bologna: the link was defined by the out-transfer rate of the degree course (computed as the ratio of the number of out-transfers to the number of students enrolled) and the critical points of the system were defined by the courses characterized by a high out-transfer rate.

Donor/Recipient Delta Age: A Possible Risk for Arterial Stenosis in Renal Transplantation

Different arterial wall properties can significantly increase the risk of blood turbulent fluxes leading to complications such as atherosclerosis. Since the mechanical properties of arterial vessels are influenced by age, we investigated, in a retrospective study, the effects on renal artery stenosis of an age difference >15 years between donor and recipient in a cohort of 164 patients undergoing renal transplantation between 1981 and 1991. The age difference between donor and recipient was ≤15 years in 87 patients (53.0%) (Group A) and >15 years in 77 patients (47.0%) (Group B, p = ns). None of the Group A patients developed an anastomotic arterial stenosis, whereas 8/77 Group B patients (10.4%) had an anastomotic arterial stenosis (p < 0.001). This study shows that an age difference >15 years is significantly linked to the risk of developing arterial stenosis after renal transplantation. Indeed, different wall properties can significantly increase the risk of generation of blood turbulent fluxes and involve, in the arterial vessels, the development of complications such as atherosclerosis.

Sequential Combining of Expert Information Using Mathematica

In every real-world domain where reasoning under uncertainty is required, combining information from different sources (‘experts’) can be really a powerful tool to enhance accuracy and precision of the ‘final’ estimate of the unknown quantity. Bayesian paradigm offers a coherent perspective which can be used to address the problem, but an issue strictly related to information combining is how to perform an efficient process of sequential consulting: at each stage, the investigator can select the ‘best’ expert to be consulted and choose whether to stop or continue the consulting. The aim of this paper is to rephrase the Bayesian combining algorithm in a sequential context and use Mathematica to implement suitable selecting and stopping rules.

The Role of Classification Trees and Expert Knowledge in Building Bayesian Networks: A Case Study in Medicine

In clinical research an early and prompt detection of the risk class of a new patient may really play a crucial role in determining the effectiveness of the treatment and, consequently, achieving a satisfying prognosis of the patients chances. There exists a number of popular rule-based algorithms for classification, whose performances are very attractive whenever data of large number of patients are available. However, when datasets only include data of a few hundred patients, the most common approaches give unstable results and developing effective decision-support systems become scientifically challenging. Since rules can be derived from different models as well as expert knowledge resources, each of them having its advantages and weaknesses, this article suggests a “hybrid” approach to address the classification problem when the number of patients is too small to effectively use a single technique only. The hybrid strategy was applied to a case study and its predictive performance was compared with performances of each single approach: due to the seriousness of a misclassification of high-risk patients, special attention was paid on the specificity. The results show that the hybrid strategy outperforms each single strategy involved.