Machine learning to predict overall short-term mortality in cutaneous melanoma.

BACKGROUND: Cutaneous malignant melanoma (CMM) ranks among the ten most frequent malignancies, clinicopathological staging being of key importance to predict prognosis. Artificial intelligence (AI) has been recently applied to develop prognostically reliable staging systems for CMM. This study aims to provide a useful machine learning based tool to predict the overall CMM short-term survival. METHODS: CMM records as collected at the Veneto Cancer Registry (RTV) and at the Veneto regional health service were considered. A univariate Cox regression validated the strength and direction of each independent variable with overall mortality. A range of machine learning models (Logistic Regression classifier, Support-Vector Machine, Random Forest, Gradient Boosting, and k-Nearest Neighbors) and a Deep Neural Network were then trained to predict the 3-years mortality probability. Five-fold cross-validation and Grid Search were performed to test the best data preprocessing procedures, features selection, and to optimize models hyperparameters. A final evaluation was carried out on a separate test set in terms of balanced accuracy, precision, recall and F1 score. The best model was deployed as online tool. RESULTS: The univariate analysis confirmed the significant prognostic value of TNM staging. Adjunctive clinicopathological variables not included in the AJCC 8th melanoma staging system, i.e., sex, tumor site, histotype, growth phase, and age, were significantly linked to overall survival. Among the models, the Neural Network and the Random Forest models featured the best prognostic performance, achieving a balanced accuracy of 91% and 88%, respectively. According to the Gini importance score, age, T and M stages, mitotic count, and ulceration appeared to be the variables with the greatest impact on survival prediction. CONCLUSIONS: Using data from patients with CMM, we developed an AI algorithm with high staging reliability, on top of which a web tool was implemented ( unipd.link/melanomaprediction ). Being essentially based on routinely recorded clinicopathological variables, it can already be implemented with minimal effort and further tested in the current clinical practice, an essential phase for validating the model's accuracy beyond the original research context.

Free fatty acid kinetics during long-term treatment with pioglitazone added to sulfonylurea or metformin in Type 2 diabetes.

BACKGROUND: Free fatty acids (FFAs) are linked to impaired insulin action, but their role in mediating long-term insulin sensitization during diabetes treatment is unclear. OBJECTIVES: To examine the effect of pioglitazone addition to existing therapy on FFA dynamics and insulin action. DESIGN: Two 2-year, randomized, parallel-group, double-blind, double-dummy, clinical trials. SETTING: One hundred and seventy-one centres in Europe, Australia and Canada. SUBJECTS: Male and female patients with Type 2 diabetes inadequately managed with metformin or sulfonylurea. INTERVENTIONS: Patients were randomized to pioglitazone (15-45 mg day(-1); n=319) or metformin (850-2550 mg day(-1); n=320) as add-on therapy to gliclazide or pioglitazone (n=317) versus gliclazide (80-320 mg day(-1); n=313) as add-on therapy to metformin. OUTCOME MEASURE: Plasma FFA profiles during oral glucose tolerance tests in selected centres before and during treatment (n=588). RESULTS: At Week 104, pioglitazone treatment decreased fasting FFAs by 0.08 mmol L(-1) when added to sulfonylurea and by 0.11 mmol L(-1) when added to metformin versus the respective sulfonylurea + metformin groups (0.03 mmol L(-1), P=0.05 and 0.04 mmol L(-1), P<0.05), and this was accompanied by significant improvements in fasting adipose tissue insulin sensitivity. Changes in postchallenge FFAs were similar between groups and not related to changes in liver transaminases, insulin action and secretion. However, the sensitivity of FFA to insulin was affected by treatment (P<0.001) and visit (P<0.05). Insulin sensitivity of FFA rose when pioglitazone was added to sulfonylurea (P<0.05), but decreased for gliclazide + metformin (P<0.05). CONCLUSION: Long-term improvements in adipose tissue insulin sensitivity and reduction in fasting FFAs with pioglitazone may help to reduce lipotoxicity in Type 2 diabetes.

A simulation-based comparison of techniques to correct for measurement error in matched case-control studies.

The presence of measurement errors affecting the covariates in regression models is a relevant topic in many scientific areas, as, for example, in epidemiology. An example is given by an epidemiological population-based matched case-control study on the aetiology of childhood malignancies, which is currently under completion in Italy. This study was aimed at evaluating the effects of childhood exposure to extremely low electromagnetic fields on the risk of disease occurrence by taking into account the possibility of erroneous measures of the exposure. Within this framework, we focus on the application of likelihood methods to correct for measurement error. This approach, which has received less attention in literature with respect to alternatives, is compared with commonly used methods such as regression calibration and SIMEX. The comparison is performed by simulation, under a broad range of measurement error structures.

Comparative evaluation of simple insulin sensitivity methods based on the oral glucose tolerance test.

AIMS/HYPOTHESIS: We compared five surrogate insulin sensitivity (IS) methods against the euglycaemic-hyperinsulinaemic clamp. These methods were the homeostasis model assessment (HOMA) and four methods based on the OGTT (OGIS, MCRest, ISIcomp, SIORAL). METHODS: We compared these IS methods against the clamp (0.28 nmol.min(-1).m(-2) insulin infusion) M value in 147 women (58-61 years; BMI 19-38 kg/m2; 116 NGT, 25 IFG/IGT, six type 2 diabetic), by evaluating the correlation coefficient with M. We also tested the ability to reproduce the relationships between IS and typical IS correlates (BMI, fasting insulin, insulin to glucose OGTT area ratio and fasting, 2 h and mean glucose) by means of the "discrepancy index" D, in which (1) D=0 if the correlation between IS and the variable of interest is as with the clamp, (2) D is smaller than 0 if the correlation is overestimated, and (3) D is greater than 0 if underestimated. RESULTS: All IS methods correlated with M (r=0.57-0.83, p<0.0001); for MCRest the relationship was markedly curvilinear. All IS measures correlated with the considered variables (r=0.29-0.94, p<0.0005); however, no method had D approximately 0 for all variables. The best surrogates of M were OGIS (one D not =0) and MCRest (two D not =0); the other methods either under- or overestimated the degree of correlation (three or more D not =0), in particular with fasting insulin (HOMA: D=-57%; ISIcomp: D=-36%) and BMI (HOMA: D=-14%; ISIcomp: D=-14%; SiORAL: D=-11%). CONCLUSIONS/INTERPRETATION: All IS methods were correlated with M. OGIS and MCRest were preferable to the other methods and in particular to HOMA for reproducing relationships with the independent variables.

Novel erythromycin derivatives with aryl groups tethered to the C-6 position are potent protein synthesis inhibitors and active against multidrug-resistant respiratory pathogens.

A novel series of erythromycin derivatives has been discovered with potent activity against key respiratory pathogens, including those resistant to erythromycin. These compounds are characterized by having an aryl group tethered to the C-6 position of the erythronolide skeleton. Extensive structural modification of the C-6 moiety led to the discovery of several promising compounds with potent activity against both mef- and erm-mediated resistant Streptoccoccus pneumoniae. Preliminary mechanistic studies indicated that the new macrolides are potent protein synthesis inhibitors, which interact with methylated ribosomes isolated from resistant organisms. In experimental animal models, these compounds exhibited excellent in vivo efficacy and balanced pharmacokinetic profiles.

ErbB2 assay in breast cancer: possibly improved clinical information using a quantitative method.

ErbB2/neu protein (p185) expression was evaluated by ELISA in 115 breast cancer specimens. Distribution was subdivided in quartiles and showed a distinct behaviour in comparison with both clinico-biological parameters and clinical outcome. In particular, intermediate concentration groups showed a significantly better disease-free survival than the low and high concentration groups (p = 0.02). We classified the patients as "low risk" (64 samples with p185 concentrations between 2150 and 30000 U/mg of proteins) and "high risk" on the basis of the results of the multivariate analysis. The p185 grouped as described showed a significant relationship with the disease free survival in multivariate analysis. Although the data must be considered as preliminary, they suggest the possibility of identifying more appropriately the high risk patients through the biochemical determination of p185.

Quantitative chemiluminescent immunoassay of p53: prognostic significance in 220 node-negative breast cancer tissue.

The prognostic and predictive role of p53 overexpression in breast cancer samples is usually investigated by using molecular biology or immunohistochemical methods. However, the results are to date controversial, and this is in part due to the methodological pitfalls of both the methods. To study the possibility of overcoming, at least in part, these problems we evaluated a commercially available chemiluminescent immunoassay with which the p53 concentrations of 220 specimens from node negative breast cancer were determined. The assay showed good analytical performance and found detectable levels in 84.7% of cases (median 0.22 ng/mg of proteins, range 0-50 ng/mg of proteins). p53 has been found inversely correlated with estrogen receptors and directly correlated with cathepsin D. The prognostic role of p53 was evaluated in two different ways: a) two previous studies (Borg et al 1995, DeWitte et al. 1996) using the same method found almost 30% of samples had significantly shorter DFS and OS. We subdivided our cases in order to identify the same positivity rate and to verify if the previous cathegorizations were effective also in our patient series. We confirmed the independent association with DFS (p = 0.006) and OS (p = 0.0005); b) considering that any categorization of quantitative parameters could cause a loss of clinical information, we also evaluated p53 as a continuous variable. Multivariate analysis showed a significant quantitative relationship between p53 and both disease free (p = 0.026) and overall survival (p = 0.02).

Determination of ErbB2 protein in breast cancer tissues by different methods. Relationships with other biological parameters.

Four different methods to measure in parallel the erbB2 protein expression (p185neu) were evaluated in order to: a) compare two enzyme immunoassays with the immunohistochemical assays (IHC) and western blotting (WB) and b) extrapolate eventual relationships between erbB2 and biological parameters. Tissue samples from 248 patients with primary breast cancer were consecutively assayed. We used two different cut-off levels for WB, ELISA, and EIA, defined as follows: 1) the highest level of expression of non malignant tissue was chosen as the discriminant threshold between 'low' and 'elevated' samples: 2) the elevated group was further subdivided into two subgroups: 'intermediate' and 'high', according to their median value. According to the first cut-off, the results were considered 'elevated' in about 52% of cases with the three biochemical methods, while using the second cut-off the percentage lowered to about 26%. Considering this cut-off, the concordance rates between the paired biochemical methods ranged between: 78.4% (WB vs EIA), 93% (ELISA vs EIA), and 82.6% (ELISA vs WB). The comparison between biochemical and immunohistochemical methods gave these concordance rates: 82% (WB vs IHC), 90.5% (ELISA vs IHC), and 85.5% (EIA vs. IHC). According to the first cut off level, 27.5% of tumor samples showed IHC detectable p185 levels, in agreement with other immunohistochemical studies. The relationship between high erbB2 and estrogen and progesterone receptors showed an inverse association. No relationship was found between erbB2 and axillary lymph node positivity or tumor size. In short, the results of the four methods seem generally well correlated; nevertheless, it appears that different methodological approaches of measuring p185neu are not completely equivalent, and there is a need for an authoritative standardization and quality control for clinical applications.

Epidermal growth-factor receptors and erbb2 protein expression in esophageal cancer and normal mucosa.

The epidermal growth factor receptor (EGFr) and the erbB2 protein (p185) concentrations were assessed in 31 esophageal cancer specimens and in the corresponding normal mucosa, in order to investigate the possible links with the main clinical and pathological parameters. Detectable and high affinity EGFr was found in 27/31 cancer and in 18/31 normal tissue samples. EGFr concentrations were not significantly different between cancer and normal tissue, although a trend toward higher levels in cancer was found. No relationships were found with histologic type, tumor bulk, lymph node status, pathologic stage, ploidy and type of surgical resection. A significant negative correlation between EGFr levels and overall survival was found. Detectable levels of p185 were found in all the tissues examined, but the expression was higher in adenocarcinoma than in squamous cell carcinoma samples. The EGFr role in malignant transformation still has to be established, but the determination could be of clinical use. As for p185, its role in the onset of esophageal cancer could be confined to the subgroup of the adenocarcinomas.

Comparison between western blotting, immunohistochemical and ELISA assay for p185neu quantitation in breast cancer specimens.

The protein encoded by erbB2 oncogene was evaluated by three different methodological approaches (Western blotting, ELISA and immunohistochemistry) in 147 breast cancer specimens. A highly significant correlation was found between Western blotting and ELISA results (p < 0.001). The data from ELISA and Western blotting were categorized as negatives (-), low-overexpressing (+) and high-overexpressing (++). Immunohistochemical results were classified as (-) or (+) according to the absence or the presence of specific staining. Overall positive cases (+ and ++) were 42.2% by Western blotting and 51% by ELISA, while (++) cases were 23.8% and 25.2% respectively. Histochemical staining was found in 29.8% of cases. The two by two evaluation of the assays showed a close association (chi2, p < 0.0001). In particular, the comparison between both ELISA and Western blotting with immunohistochemistry showed concordance rates of 78.9% for ELISA and 83.1% for Western blotting considering the + and ++ cases as a single group. When only the ++ cases were considered as positive, the overall agreement rises to 93.3% and 89.1% respectively. From these preliminary data we conclude that p185 values obtained with the three evaluated methods are possibly superimposable. Nevertheless, the biochemical methods seem to identify an intermediate p185 expression group, whose clinical meaning should be investigated.

Epidermal growth factor receptor in breast cancer. Comparison with non malignant breast tissue.

Epidermal growth factor receptors were measured using a radioligand binding assay in membrane preparations from 67 cancer and 25 non-malignant tissues. The binding characteristics of EGFr were similar in tumour and normal breast membranes. The concentrations were significantly higher in non-malignant tissue than in cancer. EGFr concentrations were directly correlated with steroid receptors in non-malignant tissue, whereas in cancer an inverse correlation between EGFr and steroid receptors was found.

Immunoenzymatic assay of erbB2 protein in cancer and non-malignant breast tissue. Relationships with clinical and biochemical parameters.

The erbB2-encoded protein p185 was determined in 130 breast cancer specimens and in 29 non-malignant breast tissues using a recently available ELISA method. The assay showed good characteristics of precision and accuracy. In the non-malignant tissue p185 concentrations were normally distributed and directly correlated with estrogen receptors (ER) and progesterone receptors (PR). In cancer tissue p185 showed higher concentrations than in non-malignant tissue. No relationships were found between p185 and the main clinical and pathological parameters, except that a direct association with nuclear grade was found. We have categorized the breast cancer samples according to p185 concentrations as p185-negative (concentrations lower than the higher non-malignant tissue) and p185-positive. Our data suggested a different behaviour of p185 in samples with low or high p185 concentrations. Indeed, in p185-negative samples the concentrations were directly correlated with both ER and PR. Conversely, p185-positive samples were directly associated with node status and pT, and inversely associated with ER and PR.

Epidermal growth-factor receptor and erbb2 protein in colorectal tissue - comparison between cancer and normal mucosa.

Epidermal growth factor receptor (EGFr) and p185neu protein were measured in 55 samples of carcinoma and 55 of normal colorectal mucosa from the same patient, using a ligand binding assay and an ELISA method respectively. The binding characteristics of EGFr were similar in cancer and normal tissue. The concentrations of both EGFr and p185 showed gaussian distribution and were not significantly different between normal and cancer tissue, although a trend toward higher levels of EGFr in normal mucosa was found. Moreover, no significative variations were found in the ratios between cancer and normal tissue after desaturation of the EGFr. No correlations were found between EGFr and p185 and the main clinopathological parameters.

Epidermal growth factor receptor in lung malignancies. Comparison between cancer and normal tissue.

Epidermal growth factor receptors (EGFr) were measured using a radioligand binding assay, in membrane preparations from 51 human non-small cell lung cancers and in normal tissue of the same patients. The binding characteristics of EGFr were similar in tumour and normal lung membranes (range of dissociation constant of high affinity sites: 0.1-0.6 nM). However, the concentrations in tumours (median, 16.4 fmol mg-1 of protein; range, 1.5-176) were significantly higher than in normal tissues (median, 7.4 fmol mg-1 of protein; range, 1.9-13.4). The receptor levels in normal tissue were normally distributed. It was therefore possible to define a normal/pathologic cut-off level (12.9 fmol mg-1 of protein). In 57% of cases EGFr in cancer was higher than the cut-off. No relationships were found between receptor concentrations and positivity rates of EGFr and histology, stage, lymph node positivity and pT. A trend for a direct relation between receptor positivity and grading was found.

Radioligand binding assay of epidermal growth factor receptor: causes of variability and standardization of the assay.

Although experimental evidence indicates a probable role of epidermal growth factor receptor (EGFr) in clinical oncology, no standardized method for its determination has been yet described, and discrepant results have been reported in clinical studies. In standardizing a radioligand binding assay for EGFr, we evaluated the causes of variability in each step of the assay. Entrapment of EGFr in the nuclear fraction and contamination of the crude membrane fraction by cytosol protein were eliminated through preliminary purification steps. Both Scatchard and Rosenthal analysis of the saturation reaction of the membrane fraction with a wide range of concentrations of 125I-labeled EGF revealed a double class of binding sites. Study of the saturation reaction showed a partial exchange of 125I-labeled EGF with endogenous EGF within 20 h. The present method--incubation of partly purified membrane fraction with 125I-labeled EGF, 0.5 nmol/L, with and without 100-fold excess of cold EGF, for 20 h at 26 degrees C, followed by centrifugation at 5000 x g for 30 min to separate membrane-bound 125I-labeled EGF--shows good sensitivity, precision, and accuracy; is reasonably simple; and may be suitable for routine clinical use.