A tailored machine learning approach for mortality prediction in severe COVID-19 treated with glucocorticoids.

The impact of severe COVID-19 pneumonia on healthcare systems highlighted the need for accurate predictions to improve patient outcomes. Despite the established efficacy of glucocorticoids (GCs), variable patient responses are observed, and the existing clinical scores are limited in predicting non-responders. We propose a machine learning (ML) based approach to predict mortality in COVID-19 pneumonia treated with GCs.This is an ML-driven retrospective analysis involving 825 patients. We leveraged XGBoost to select the most appropriate features from the initial 52, including clinical and laboratory data. Six different ML techniques were compared. Shapley additive explanation (SHAP) values were used to describe the influence of each feature on classification. Internal validation was performed.Nine key predictors of death were identified: increasing C-reactive protein (CRP), decreasing arterial partial pressure of oxygen to fraction of inspired oxygen ratio (PaO2/FiO2), age, coronary artery disease, invasive mechanical ventilation, acute renal failure, chronic heart failure, PaO2/FiO2 earliest value, and body mass index. Random forest achieved the highest test area under the receiver operating characteristic curve at 0.938 (95% CI 0.903-0.969). SHAP values highlighted age and PaO2/FiO2 improvement as the most influential features; the latter showed a higher impact than CRP reduction over time.The proposed ML algorithm effectively predicts the risk of hospital death in COVID-19 pneumonia patients undergoing GCs. This approach can be adapted to datasets measuring similar clinical variables..

The quality of care in type 1 and type 2 diabetes - A 2023 update of the AMD Annals initiative.

AIMS: An initiative of continuous monitoring of the quality of diabetes care, promoted by the Association of Medical Diabetologists, is in place in Italy since 2006 (AMD Annals). The initiative was effective in improving quality of care indicators, assessed periodically through standardized measures. Here, we show the 2023 AMD Annals data on type 2 (T2D) and type 1 (T1D) diabetes. METHODS: A network of over 1/3 of diabetes centers in Italy periodically extracts anonymous data from electronic medical records, using a standardized software. Process, treatment and outcome indicators, and a validated score of overall care, the Q-score, were evaluated. RESULTS: 296 centers provided data on 573,164 T2D (mean age 69.7 ± 11.2 years) and 42,611 T1D subjects (mean age 48.6 ± 16.9 years). A HbA1c value ≤ 7.0 % was documented in 56.3 % of patients with T2D and 35.9 % of those with T1D. Only 6.6 % of T2D patients and 3.5 % of those with T1D reached the composite outcome of HbA1c ≤ 7.0 % + LDL-C < 70 mg/dl + BP < 130/80 mmHg. Notably, only 2.8 % and 3.2 % of T2D and T1D patients, respectively, showed a Q score < 15, which correlates with an 80 % higher risk of incident CVD events compared to scores > 25. CONCLUSIONS: We documented an overall good quality of care in both T1D and T2D subjects. However, the failure to achieve the targets of the main risk factors, especially if combined, in a still too large proportion of patients testify the difficulty to apply the more and more stringent indications recommended by guidelines in the everyday clinical practice.

Quality of care and clinical inertia in the management of cardiovascular risk factors in patients with type 1 and type 2 diabetes: data from AMD annals.

BACKGROUND: Cardiovascular disease is the leading cause of morbidity and mortality among patients with diabetes, and for this reason, all guidelines for CV risk management provide the same targets in controlling traditional CV risk factors in patients with type 1 or type 2 diabetes at equal CV risk class. Aim of our study was to evaluate and compare CV risk management in patients with type 1 and type 2 diabetes included in AMD Annals Database paying particular attention to indicators of clinical inertia. METHODS: This was a multicenter, observational, retrospective study of AMD Annals Database during year 2022. Patients with diabetes were stratified on the basis of their cardiovascular risk, according to ESC-EASD guidelines. The proportion of patients not treated with lipid-lowering despite LDL cholesterol > to 100 mg/dl or the proportion of patients not treated with antihypertensive drug despite BP > 140/90 mmhg and proportion of patients with proteinuria not treated with angiotensin converting enzyme inhibitors or angiotensinogen receptor blockers (ACE/ARBs) were considered indicators of clinical inertia. The proportion of patients reaching at the same time HbA1c < 7% LDL < 70 mg/dl and BP < 130/80 mmhg were considered to have good multifactorial control. Overall quality of health care was evaluated by the Q-score. RESULTS: Using the inclusion criteria and stratifying patients by ESC/EASD Cardiovascular Risk categories, we included in the analysis 118.442 patients at High Cardiovascular risk and 416.246 patients at Very High Cardiovascular risk. The proportion of patients with good multifactorial risk factor control was extremely low in both T1D and T2D patients in each risk class. At equal risk class, the patients with T1D had lower proportion of subjects reaching HbA1c, LDL, or Blood Pressure targets. Indicators of clinical inertia were significantly higher compared with patients with T2D at equal risk class. Data regarding patients with albuminuria not treated with RAAS inhibitors were available only for those at Very High risk and showed that the proportion of patients not treated was again significantly higher in patients with T1DM. CONCLUSIONS: In conclusion, this study provides evidence of wide undertreatment of traditional cardiovascular risk factors among patients with diabetes included in AMD Annals Database. Undertreatment seems to be more pronounced in individuals with T1D compared to those with T2D and is frequently due to clinical inertia.

Temporal trends in the starting of insulin therapy in type 2 diabetes in Italy: data from the AMD Annals initiative.

AIMS: Opportunities and needs for starting insulin therapy in Type 2 diabetes (T2D) have changed overtime. We evaluated clinical characteristics of T2D subjects undergoing the first insulin prescription during a 15-year-observation period in the large cohort of the AMD Annals Initiative in Italy. METHODS: Data on clinical and laboratory variables, complications and concomitant therapies and the effects on glucose control after 12 months were evaluated in T2D patients starting basal insulin as add-on to oral/non-insulin injectable agents, and in those starting fast-acting in add-on to basal insulin therapy in three 5-year periods (2005-2019). RESULTS: We evaluated data from 171.688 T2D subjects who intensified therapy with basal insulin and 137.225 T2D patients who started fast-acting insulin. Overall, intensification with insulin occurred progressively earlier over time in subjects with shorter disease duration. Moreover, the percentage of subjects with HbA1c levels > 8% at the time of basal insulin initiation progressively decreased. The same trend was observed for fast-acting formulations. Clinical characteristics of subjects starting insulin did not change in the three study-periods, although all major risk factors improved overtime. After 12 months from the starting of basal or fast-acting insulin therapy, mean HbA1c levels decreased in all the three investigated time-periods, although mean HbA1c levels remained above the recommended target. CONCLUSIONS: In this large cohort of T2D subjects, a progressively earlier start of insulin treatment was observed during a long observation period, suggesting a more proactive prescriptive approach. However, after 12 months from insulin prescription, in many patients, HbA1c levels were still out-of-target.

The AMD ANNALS: A continuous initiative for the improvement of type 2 diabetes care.

AIMS: Since 2006, the Italian AMD (Associations of Medical Diabetologists) Annals Initiative promoted a continuous monitoring of the quality of diabetes care, that was effective in improving process, treatment and outcome indicators through a periodic assessment of standardized measures. Here, we show the 2022 AMD Annals data on type 2 diabetes (T2D). METHODS: A network involving ∼1/3 of diabetes centers in Italy periodically extracts anonymous data from electronic clinical records, by a standardized software. Process, treatment and outcome indicators, and a validated score of overall care, the Q-score, were evaluated. RESULTS: 295 centers provided the annual sample of 502,747 T2D patients. Overall, HbA1c value ≤7.0% was documented in 54.6% of patients, blood pressure <130/80 mmHg in 23.0%, and LDL-cholesterol levels <70 mg/dl in 34.3%, but only 5.2% were at- target for all the risk factors. As for innovative drugs, 29.0% of patients were on SGLT2-i, and 27.5% on GLP1-RAs. In particular, 59.7% were treated with either GLP1-RAs or SGLT2-i among those with established cardiovascular disease (CVD), 26.6% and 49.3% with SGLT2-i among those with impaired renal function and heart failure, respectively. Notably, only 3.2% of T2D patients showed a Q score <15, which correlates with a 80% higher risk of incident CVD events compared to scores >25. CONCLUSIONS: The 2022 AMD Annals data show an improvement in the use of innovative drugs and in the overall quality of T2D care in everyday clinical practice. However, additional efforts are needed to reach the recommended targets for HbA1c and major CVD risk factors.

Two years follow-up of relapsing eosinophilic pneumonia with concomitant severe asthma successfully treated with benralizumab: A case report and brief review of the literature.

Relapsing eosinophilic pneumonia and severe eosinophilic asthma are rare and disabling diseases, which share common inflammatory backgrounds and often require long-term systemic steroid therapy. Benralizumab is a humanized antibody targeting IL-5 receptor that reduces corticosteroid dependence and flares up in severe eosinophilic asthma on long term. In this case report, successful treatment of eosinophilic pneumonia and severe eosinophilic asthma with benralizumab is described after a 2-year follow up, showing the promising results of this therapy for eosinophilic pneumonia management.

Automated particle recognition for engine soot nanoparticles.

A pre-trained convolution neural network based on residual error functions (ResNet) was applied to the classification of soot and non-soot carbon nanoparticles in TEM images. Two depths of ResNet, one 18 layers deep and the other 50 layers deep, were trained using training-validation sets of increasing size (containing 100, 400 and 1400 images) and were assessed using an independent test set of 200 images. Network training was optimised in terms of mini-batch size, learning rate and training length. In all tests, ResNet18 and ResNet50 had statistically similar performances, though ResNet18 required only 25-35% of the training time of ResNet50. Training using the 100-, 400- and 1400-image training-validation sets led to classification accuracies of 84%, 88% and 95%, respectively. ResNet18 and ResNet50 were also compared for their ability to categorise soot and non-soot nanoparticles via a fivefold cross-validation experiment using the entire set of 800 images of soot and 800 images of non-soot. Cross-validation was repeated 3 times with different training durations. For all cross-validation experiments, classification accuracy exceeded 91%, with no statistical differences between any of the network trainings. The most efficient network was ResNet18 trained for 5 epochs, which reached 91.2% classification after only 84 s of training on 1600 images. Use of ResNet for classification of 1000 images, the amount suggested for reliable characterisation of soot sample, requires <4 s, compared with >30 min for a skilled operator classifying images manually. Use of convolution neural networks for classification of soot and non-soot nanoparticles in TEM images is highly promising, particularly when manually classified data sets have already been established.

Acute rapamycin rescues the hyperexcitable phenotype of accumbal medium spiny neurons in the valproic acid rat model of autism spectrum disorder.

We previously demonstrated that prenatal exposure to valproic acid (VPA), an environmental model of autism spectrum disorder (ASD), leads to a hyperexcitable phenotype associated with downregulation of inward-rectifying potassium currents in nucleus accumbens (NAc) medium spiny neurons (MSNs) of adolescent rats. Aberrant mTOR pathway function has been associated with autistic-like phenotypes in multiple animal models, including gestational exposure to VPA. The purpose of this work was to probe the involvement of the mTOR pathway in VPA-induced alterations of striatal excitability. Adolescent male Wistar rats prenatally exposed to VPA were treated acutely with the mTOR inhibitor rapamycin and used for behavioral tests, ex vivo brain slice electrophysiology, single-neuron morphometric analysis, synaptic protein quantification and gene expression analysis in the NAc. We report that postnatal rapamycin ameliorates the social deficit and reverts the abnormal excitability, but not the inward-rectifying potassium current defect, of accumbal MSNs. Synaptic transmission and neuronal morphology were largely unaffected by prenatal VPA exposure or postnatal rapamycin treatment. Transcriptome analysis revealed extensive deregulation of genes implied in neurodevelopmental disorders and ionic mechanisms exerted by prenatal VPA, which was partially reverted by postnatal rapamycin. The results of this work support the existence of antagonistic interaction between mTOR and VPA-induced pathways on social behavior, neurophysiological phenotype and gene expression profile, thus prompting further investigation of the mTOR pathway in the quest for specific therapeutic targets in ASD.

Moderate ethanol drinking is sufficient to alter Ventral Tegmental Area dopamine neurons activity via functional and structural remodeling of GABAergic transmission.

Earlier studies have shown a major involvement of Ventral Tegmental Area (VTA) dopamine (DA) neurons in mediating the rewarding effects of ethanol (EtOH). Much less is known on the role of this system in mediating the transition from moderate to excessive drinking and abuse. Here we sought to explore the hypothesis that early stage drinking in rodents, resembling recreational EtOH use in humans, is sufficient to dysregulate VTA DA transmission thus increasing the propensity to use over time. To this purpose, midbrain slice recordings in mice previously exposed to an escalating (3, 6 and 12%) 18-day voluntary EtOH drinking paradigm was used. By recording from DA and γ-aminobutyric acid (GABA) VTA neurons in midbrain slices, we found that moderate EtOH drinking leads to a significant suppression of the spontaneous activity of VTA DA neurons, while increasing their response to acute EtOH application. We also found that chronic EtOH leads to the enhancement of GABA input frequency onto a subset of DA neurons. Structurally, chronic EtOH induced a significant increase in the number of GABA axonal boutons contacting DA neurons, suggesting deep rewiring of the GABA network. This scenario is consistent with a downmodulation of the reward DA system induced by moderate EtOH drinking, a neurochemical state defined as "hypodopaminergic" and previously associated with advanced stages of drug use in humans. In this context, increased sensitivity of DA neurons towards acute EtOH may represent the neurophysiological correlate of increased unitary rewarding value, possibly driving progression to addiction.

Phase II study of eribulin in combination with gemcitabine for the treatment of patients with locally advanced or metastatic triple negative breast cancer (ERIGE trial). Clinical and pharmacogenetic results on behalf of the Gruppo Oncologico Italiano di Ricerca Clinica (GOIRC).

BACKGROUND: The combination of a microtubule inhibitor (eribulin) with a nucleoside analog (gemcitabine) may synergistically induce tumor cell death, particularly in triple negative breast cancer (TNBC) characterized by high cell proliferation, aggressive behavior, and chemo-resistance. PATIENTS AND METHODS: This is an open-label, multicenter phase II study evaluating the combination of eribulin (0.88 mg/m2) plus gemcitabine (1000 mg/m2) on days 1 and 8 of a 21-day cycle as either first- or second-line treatment of locally advanced or metastatic TNBC. The primary endpoint was the objective response for evaluable patients. A prospective, molecular correlative study was carried out to assess the role of germinal BRCA pathogenic variants and single nucleotide polymorphisms (SNPs) in predicting efficacy and toxicity of the combination regimen. RESULTS: From July 2013 to September 2016, 83 evaluable patients were enrolled. They received a median number of six cycles of treatment. An overall response rate (ORR) of 37.3% (31 patients) was observed, with a complete response rate of 2.4% and a partial response rate of 34.9%; the clinical benefit rate was 48.8%. With a median follow-up of 28.8 months, the median response duration was 6.6 months, the median progression-free survival (PFS) was 5.1 months, and the median overall survival (OS) was 14.5 months. The most common grade 3-4 adverse events were aminotransferase elevation (in 25% of the patients) and neutropenia (in 23.8%). Women with BRCA1/2 pathogenic variants were associated with worse ORR, PFS, and OS than BRCA1/2 wild-type carriers. CYP3A4 and FGD4 SNPs were associated with increased risk of liver toxicity. Three different SNPs in CDA∗2, RRM1, and CYP2C8 genes were significantly associated with poorer OS. CONCLUSIONS: The combination of eribulin and gemcitabine showed promising activity and a moderate toxicity profile in metastatic TNBC. BRCA status and pharmacogenetics tests may help identify patients with high probability of response with negligible toxicity. EUDRACT NUMBER: 2012-003505-10.

Metformin plus chemotherapy versus chemotherapy alone in the first-line treatment of HER2-negative metastatic breast cancer. The MYME randomized, phase 2 clinical trial.

PURPOSE: To investigate the efficacy of metformin (M) plus chemotherapy versus chemotherapy alone in metastatic breast cancer (MBC). METHODS: Non-diabetic women with HER2-negative MBC were randomized to receive non-pegylated liposomal doxorubicin (NPLD) 60 mg/m2 + cyclophosphamide (C) 600 mg/m2 × 8 cycles Q21 days plus M 2000 mg/day (arm A) versus NPLD/C (arm B). The primary endpoint was progression-free survival (PFS). RESULTS: One-hundred-twenty-two patients were evaluable for PFS. At a median follow-up of 39.6 months (interquartile range [IQR] 24.6-50.7 months), 112 PFS events and 71 deaths have been registered. Median PFS was 9.4 months (95% CI 7.8-10.4) in arm A and 9.9 (95% CI 7.4-11.5) in arm B (P = 0.651). In patients with HOMA index < 2.5, median PFS was 10.4 months (95% CI 9.6-11.7) versus 8.5 (95% CI 5.8-9.7) in those with HOMA index ≥ 2.5 (P = 0.034). Grade 3/4 neutropenia was the most common toxicity, occurring in 54.4% of arm A patients and 72.3% of the arm B group (P = 0.019). M induced diarrhea (G2) was observed in 8.8% of patients in Arm A. The effect of M was similar in patients with HOMA index < 2.5 and ≥ 2.5, for PFS and OS. CONCLUSIONS: The MYME trial failed to provide evidence in support of an anticancer activity of M in combination with first line CT in MBC. A significantly shorter PFS was observed in insulin-resistant patients (HOMA ≥ 2.5). Noteworthy, M had a significant effect on CT induced severe neutropenia. Further development of M in combination with CT in the setting of MBC is not warranted.

Prenatal profile of Pallister-Killian syndrome: Retrospective analysis of 114 pregnancies, literature review and approach to prenatal diagnosis.

Pallister-Killian syndrome (PKS) is a tissue limited mosaic disorder, characterized by variable degrees of neurodevelopmental delay and intellectual disability, typical craniofacial findings, skin pigmentation anomalies and multiple congenital malformations. The wide phenotypic spectrum of PKS in conjunction with the mosaic distribution of the i(12p) makes PKS an underdiagnosed disorder. Recognition of prenatal findings that should raise a suspicion of PKS is complicated by the fragmentation of data currently available in the literature and challenges in diagnosing a mosaic diagnosis on prenatal testing. Ultrasound anomalies, especially congenital diaphragmatic hernia, congenital heart defects, and rhizomelic limb shortening, have been related to PKS, but they are singularly not specific and are not present in all affected fetuses. We have combined prenatal data from 86 previously published reports and from our cohort of 114 PKS probands (retrospectively reviewed). Summarizing this data we have defined a prenatal growth profile and identified markers of perinatal outcome which collectively provide guidelines for early recognition of the distinctive prenatal profile and consideration of a diagnosis of PKS as well as for management and genetic counseling.

Progress towards a methodology for high throughput 3D reconstruction of soot nanoparticles via electron tomography.

The aim of this work is to make progress towards the development of 3D reconstruction as a legitimate alternative to traditional 2D characterization of soot. Time constraints are the greatest opposition to its implementation, as currently reconstruction of a single soot particle takes around 5-6 h to complete. As such, the accuracy and detail gains are currently insufficient to challenge 2D characterization of a representative sample (e.g. 200 particles). This work is a consideration of the optimization of the steps included within the computational reconstruction and manual segmentation of soot particles. Our optimal process reduced the time required by over 70% in comparison to a typical procedure, whilst producing models with no appreciable decrease in quality.

Circulation of a Simbu Serogroup Virus, Causing Schmallenberg Virus-Like Clinical Signs in Northern Jordan.

Schmallenberg virus (SBV)-like clinical cases of abortions in northern Jordan in early 2013, together with the emergence of SBV in Europe in 2011, its rapid spread within the following years and the detection of this virus in Turkey, raised questions about the distribution of SBV or related orthobunyaviruses. To evaluate the occurrence of SBV or related members of the Simbu serogroup of orthobunyaviruses in Jordan, bulk milk (cattle) and serum samples (cattle, sheep and goat) collected in northern Jordan in 2013 were first tested by commercially available SBV antibody ELISAs. Indeed, 3 of 47 bulk milk samples and 57 of 115 serum samples provided positive results, but SBV specificity of the ELISA results could not be confirmed by virus neutralization assays. Instead, subsequent cross-neutralization tests were able to further investigate the specificity of these antibodies. Here, a significant inhibition of Aino virus was observed. Thus, the causative agent was most likely a Simbu serogroup virus closely related to Aino virus. Consequently, these results confirm that members of this group of virus are not only present in Europe, Africa or Australia, but also in the Middle East.

Sympathetic activity and early mobilization in patients in intensive and intermediate care with severe brain injuries: a preliminary prospective randomized study.

BACKGROUND: Patients who experience severe brain injuries are at risk of secondary brain damage, because of delayed vasospasm and edema. Traditionally, many of these patients are kept on prolonged bed rest in order to maintain adequate cerebral blood flow, especially in the case of subarachnoid hemorrhage. On the other hand, prolonged bed rest carries important morbidity. There may be a clinical benefit in early mobilization and our hypothesis is that early gradual mobilization is safe in these patients. The aim of this study was to observe and quantify the changes in sympathetic activity, mainly related to stress, and blood pressure in gradual postural changes by the verticalization robot (Erigo®) and after training by a lower body ergometer (MOTOmed-letto®), after prolonged bed rest of minimum 7 days. METHODS: Thirty patients with severe neurological injuries were randomized into 3 groups with different protocols of mobilization: Standard, MOTOmed-letto® or Erigo® protocol. We measured plasma catecholamines, metanephrines and blood pressure before, during and after mobilization. RESULTS: Blood pressure does not show any significant difference between the 3 groups. The analysis of the catecholamines suggests a significant increase in catecholamine production during Standard mobilization with physiotherapists and with MOTOmed-letto® and no changes with Erigo®. CONCLUSIONS: This preliminary prospective randomized study shows that the mobilization of patients with severe brain injuries by means of Erigo® does not increase the production of catecholamines. It means that Erigo® is a well-tolerated method of mobilization and can be considered a safe system of early mobilization of these patients. Further studies are required to validate our conclusions. TRIAL REGISTRATION: The study was registered in the ISRCTN registry with the trial registration number ISRCTN56402432 . Date of registration: 08.03.2016. Retrospectively registered.

Sentinel lymph node surgery after neoadjuvant chemotherapy in patients with T2 to T4, N0 and N1 breast cancer.

BACKGROUND: Histological status of axillary lymph nodes is an important prognostic factor in patients receiving surgery for breast cancer (BC). Sentinel lymph node (SLN) biopsy (B) has rapidly replaced axillary lymph node dissection (ALND), and is now the standard of care for axillary staging in patients with clinically node-negative (N0) operable BC. The aim of this study is to compare pretreatment lymphoscintigraphy with a post primary systemic treatment (PST) scan in order to reduce the false-negative rates for SLNB. METHODS: In this single-institution study we considered 170 consecutive T2-4 N0-1 M0 BC patients treated with anthracycline-based PST. At the time of incisional biopsy, we performed sentinel lymphatic mapping. After PST, all patients repeated lymphoscintigraphy with the same methodology. During definitive surgery we performed further sentinel lymphatic mapping, SLNB and ALND. RESULTS: The SLN was removed in 158/170 patients giving an identification rate of 92.9% (95% confidence interval (CI) = 88.0-96.3%) and a false-negative rate of 14.0% (95% CI = 6.3-25.8%). SLNB revealed a sensitivity of 86.0% (95% CI = 74.2-93.7%), an accuracy of 94.9% (95% CI = 90.3-97.8%) and a negative predictive value of 92.7% (95% CI = 86.1-96.8%). CONCLUSION: Identification rate, sensitivity and accuracy are in accordance with other studies on SLNB after PST, even after clinically negative node conversion following PST. This study confirms that diagnostic biopsy and neoadjuvant chemotherapy maintain breast lymphatic drainage unaltered.

New biomarkers to predict the evolution of in situ breast cancers.

BACKGROUND: Genomic studies have shown that gene expression profiles are similar in in situ (CIS) and invasive breast cancers, suggesting that several biofunctional modifications of the transformation process occur before or during the development of CIS lesion. METHODS: We investigated 3 biomarkers in 44 patients with CIS: TG2 (transglutaminase 2), HJURP (Holliday junction recognition protein), and HIF-1α (hypoxia inducible factor-1 alpha). RESULTS: TG2 was more highly expressed than the other two markers and significantly more so in stromal than in tumor cells. HIF-1α evaluation showed a higher expression in both tumor and stromal cells in patients with relapsed G3 tumors, indicating a potential role of this marker in CIS evolution. A greater than sevenfold higher risk of relapse (P = 0.050) was observed in patients highly expressing HJURP in stroma and a tenfold higher recurrence risk (P = 0.026) was seen in those with a higher stromal HIF-1α expression. An important increase in risk accuracy (AUC 0.80) was obtained when HIF-1α and HJURP were evaluated together. CONCLUSIONS: Despite the limited number of relapsed patients, we formulated some hypotheses on the factors responsible for malignant evolution and recurrence which are now being tested in a large case series with a longer follow-up.