Prognosis in node-negative primary breast cancer: a neural network analysis of risk profiles using routinely assessed factors

A Simulation Study to Compare the Predictive Performance of Survival Neural Networks with Cox Models for Clinical Trial Data

Background

Studies focusing on prediction models are widespread in medicine. There is a trend in applying machine learning (ML) by medical researchers and clinicians. Over the years, multiple ML algorithms have been adapted to censored data. However, the choice of methodology should be motivated by the real-life data and their complexity. Here, the predictive performance of ML techniques is compared with statistical models in a simple clinical setting (small/moderate sample size and small number of predictors) with Monte-Carlo simulations.

Methods

Synthetic data (250 or 1000 patients) were generated that closely resembled 5 prognostic factors preselected based on a European Osteosarcoma Intergroup study (MRC BO06/EORTC 80931). Comparison was performed between 2 partial logistic artificial neural networks (PLANNs) and Cox models for 20, 40, 61, and 80% censoring. Survival times were generated from a log-normal distribution. Models were contrasted in terms of the C-index, Brier score at 0-5 years, integrated Brier score (IBS) at 5 years, and miscalibration at 2 and 5 years (usually neglected). The endpoint of interest was overall survival.

Results

PLANNs original/extended were tuned based on the IBS at 5 years and the C-index, achieving a slightly better performance with the IBS. Comparison with Cox models showed that PLANNs can reach similar predictive performance on simulated data for most scenarios with respect to the C-index, Brier score, or IBS. However, Cox models were frequently less miscalibrated. Performance was robust in scenario data where censored patients were removed before 2 years or curtailing at 5 years was performed (on training data).

Conclusion

Survival neural networks reached a comparable predictive performance with Cox models but were generally less well calibrated. All in all, researchers should be aware of burdensome aspects of ML techniques such as data preprocessing, tuning of hyperparameters, and computational intensity that render them disadvantageous against conventional regression models in a simple clinical setting.

Properties of manual toothbrush that influence on plaque removal of interproximal surface in vitro

Background/purpose

Few papers were available on the interproximal cleaning efficiency by manual toothbrushes when used alone. The aim was to investigate the efficiency of commercially available toothbrushes on interproximal cleaning and determine the key properties that would make the differences.

Materials and methods

Artificial-teeth were coated with manicure type experimental dental plaque covering mainly the interproximal surface and fixed in the jaw model of a dental simulator. A modified scrubbing technique was employed to brush out the plaque conducted by one trained dentist using 26 different toothbrushes from the equal number of separate interproximal conditions. The rate of the plaque removal (%) was calculated by measuring the plaque free areas on the post-brush images.

Results

The data analysis using mixed effect modelling showed that stiffness, number of tufts and total length have effect on the rate of the plaque removable from the interproximal surfaces.

Conclusion

This study indicated consideration should be given to toothbrush properties to enhance plaque removal from the interproximal surfaces.

A Prediction Model for Breast Cancer Recurrence after Adjuvant Hormone Therapy

Hormone therapy with tamoxifen has long been the established adjuvant treatment for node-positive, estrogen–receptor-positive breast cancer in postmenopausal women. Since 30–40% of these patients fail to respond, reliable outcome prediction is necessary for successful treatment allocation. Using pathobiological variables (available in most clinical records: tumor size, nodal involvement, estrogen and progesterone receptor content) from 596 patients recruited at a comprehensive cancer center, we developed a prediction model which we validated in an independent cohort of 175 patients recruited at a general hospital. Calculated at 3 and 4 years of follow-up, the discrimination indices were 0.716 [confidence limits (CL) 0.641, 0.752] and 0.714 (CL 0.650, 0.750) for the training data, and 0.726 (CL 0.591, 0.769) and 0.677 (CL 0.580, 0.745) for the testing data. Waiting for more effective approaches from genomic and proteomic studies, a model based on consolidated pathobiological variables routinely assessed at relatively low costs may be considered as the reference for assessing the gain of new markers over traditional ones, thus substantially improving the conventional use of prognostic criteria.

Trop-2 is a determinant of breast cancer survival

Trop-2 is a calcium signal transducer that drives tumor growth. Anti-Trop-2 antibodies with selective reactivity versus Trop-2 maturation stages allowed to identify two different pools of Trop-2, one localized in the cell membrane and one in the cytoplasm. Of note, membrane-localized/functional Trop-2 was found to be differentially associated with determinants of tumor aggressiveness and distinct breast cancer subgroups. These findings candidated Trop-2 states to having an impact on cancer progression. We tested this model in breast cancer. A large, consecutive human breast cancer case series (702 cases; 8 years median follow-up) was analyzed by immunohistochemistry with anti-Trop-2 antibodies with selective reactivity for cytoplasmic-retained versus functional, membrane-associated Trop-2. We show that membrane localization of Trop-2 is an unfavorable prognostic factor for overall survival (1+ versus 0 for all deaths: hazard ratio, 1.63; P = 0.04), whereas intracellular Trop-2 has a favorable impact on prognosis, with an adjusted hazard ratio for all deaths of 0.48 (high versus low; P = 0.003). A corresponding impact of intracellular Trop-2 was found on disease relapse (high versus low: hazard ratio, 0.51; P = 0.004). Altogether, we demonstrate that the Trop-2 activation states are critical determinants of tumor progression and are powerful indicators of breast cancer patients survival.

Time-dependent risk of developing distant metastasis in breast cancer patients according to treatment, age and tumour characteristics

BACKGROUND

Metastatic breast cancer is a severe condition without curative treatment. How relative and absolute risk of distant metastasis varies over time since diagnosis, as a function of treatment, age and tumour characteristics, has not been studied in detail.

METHODS

A total of 9514 women under the age of 75 when diagnosed with breast cancer in Stockholm and Gotland regions during 1990-2006 were followed up for metastasis (mean follow-up=5.7 years). Time-dependent development of distant metastasis was analysed using flexible parametric survival models and presented as hazard ratio (HR) and cumulative risk.

RESULTS

A total of 995 (10.4%) patients developed distant metastasis; the most common sites were skeleton (32.5%) and multiple sites (28.3%). Women younger than 50 years at diagnosis, with lymph node-positive, oestrogen receptor (ER)-negative, >20 mm tumours and treated only locally, had the highest risk of distant metastasis (0-5 years' cumulative risk =0.55; 95% confidence interval (CI): 0.47-0.64). Women older than 50 years at diagnosis, with ER-positive, lymph node-negative and ≤20-mm tumours, had the same and lowest cumulative risk of developing metastasis 0-5 and 5-10 years (cumulative risk=0.03; 95% CI: 0.02-0.04). In the period of 5-10 years after diagnosis, women with ER-positive, lymph node-positive and >20-mm tumours were at highest risk of distant recurrence. Women with ER-negative tumours showed a decline in risk during this period.

CONCLUSION

Our data show no support for discontinuation at 5 years of clinical follow-up in breast cancer patients and suggest further investigation on differential clinical follow-up for different subgroups of patients.

γ -H2AX: A Novel Prognostic Marker in a Prognosis Prediction Model of Patients with Early Operable Non-Small Cell Lung Cancer

Cancer is a leading cause of death worldwide and the prognostic evaluation of cancer patients is of great importance in medical care. The use of artificial neural networks in prediction problems is well established in human medical literature. The aim of the current study was to assess the prognostic value of a series of clinical and molecular variables with the addition of γ -H2AX-a new DNA damage response marker-for the prediction of prognosis in patients with early operable non-small cell lung cancer by comparing the γ -H2AX-based artificial network prediction model with the corresponding LR one. Two prognostic models of 96 patients with 27 input variables were constructed by using the parameter-increasing method in order to compare the predictive accuracy of neural network and logistic regression models. The quality of the models was evaluated by an independent validation data set of 11 patients. Neural networks outperformed logistic regression in predicting the patient's outcome according to the experimental results. To assess the importance of the two factors p53 and γ -H2AX, models without these two variables were also constructed. JR and accuracy of these models were lower than those of the models using all input variables, suggesting that these biological markers are very important for optimal performance of the models. This study indicates that neural networks may represent a potentially more useful decision support tool than conventional statistical methods for predicting the outcome of patients with non-small cell lung cancer and that some molecular markers, such as γ -H2AX, enhance their predictive ability.

Assessing the effect of quantitative and qualitative predictors on gastric cancer individuals survival using hierarchical artificial neural network models

Background

There are numerous unanswered questions in the application of artificial neural network models for analysis of survival data. In most studies, independent variables have been studied as qualitative dichotomous variables, and results of using discrete and continuous quantitative, ordinal, or multinomial categorical predictive variables in these models are not well understood in comparison to conventional models.

Objectives

This study was designed and conducted to examine the application of these models in order to determine the survival of gastric cancer patients, in comparison to the Cox proportional hazards model.

Patients and Methods

We studied the postoperative survival of 330 gastric cancer patients who suffered surgery at a surgical unit of the Iran Cancer Institute over a five-year period. Covariates of age, gender, history of substance abuse, cancer site, type of pathology, presence of metastasis, stage, and number of complementary treatments were entered in the models, and survival probabilities were calculated at 6, 12, 18, 24, 36, 48, and 60 months using the Cox proportional hazards and neural network models. We estimated coefficients of the Cox model and the weights in the neural network (with 3, 5, and 7 nodes in the hidden layer) in the training group, and used them to derive predictions in the study group. Predictions with these two methods were compared with those of the Kaplan-Meier product limit estimator as the gold standard. Comparisons were performed with the Friedman and Kruskal-Wallis tests.

Results

Survival probabilities at different times were determined using the Cox proportional hazards and a neural network with three nodes in the hidden layer; the ratios of standard errors with these two methods to the Kaplan-Meier method were 1.1593 and 1.0071, respectively, revealed a significant difference between Cox and Kaplan-Meier (P < 0.05) and no significant difference between Cox and the neural network, and the neural network and the standard (Kaplan-Meier), as well as better accuracy for the neural network (with 3 nodes in the hidden layer). Probabilities of survival were calculated using three neural network models with 3, 5, and 7 nodes in the hidden layer, and it has been observed that none of the predictions was significantly different from results with the Kaplan-Meier method and they appeared more comparable towards the last months (fifth year). However, we observed better accuracy using the neural network with 5 nodes in the hidden layer. Using the Cox proportional hazards and a neural network with 3 nodes in the hidden layer, we found enhanced accuracy with the neural network model.

Conclusions

Neural networks can provide more accurate predictions for survival probabilities compared to the Cox proportional hazards mode, especially now that advances in computer sciences have eliminated limitations associated with complex computations. It is not recommended in order to adding too many hidden layer nodes because sample size related effects can reduce the accuracy. We recommend increasing the number of nodes to a point that increased accuracy continues (decrease in mean standard error), however increasing nodes should cease when a change in this trend is observed.

Prognostic interaction between expression of p53 and estrogen receptor in patients with node-negative breast cancer: results from IBCSG Trials VIII and IX

INTRODUCTION

The prognostic significance of p53 protein expression in early breast cancer remains uncertain, with some but not all studies finding an association with poorer outcomes. Estrogen receptor (ER) expression is both a positive prognostic marker and predictive of response to endocrine therapies. The relationship between these biomarkers is unknown.

METHODS

We constructed tissue microarrays (TMAs) from available pathological material from 1113 patients participating in two randomized clinical trials comparing endocrine therapy alone versus chemo-endocrine therapy in node-negative breast cancer. Expression of p53 defined as >10% positive nuclei was analyzed together with prior immunohistochemical assays of ER performed at central pathological review of whole tumor sections.

RESULTS

ER was present (i.e. >1% positive tumor cell nuclei) in 80.1% (880/1092). p53 expression was significantly more frequent when ER was absent, 125/212 (59%) than when ER was present, 171/880 (19%), p <0.0001. A significant qualitative interaction was observed such that p53 expression was associated with better disease-free survival (DFS) and overall survival (OS) among patients whose tumors did not express ER, but worse DFS and OS among patients whose tumors expressed ER. The interaction remained significant after allowance for pathologic variables, and treatment. Similar effects were seen when luminal and non-luminal intrinsic subtypes were compared.

CONCLUSIONS

Interpretation of the prognostic significance of p53 expression requires knowledge of concurrent expression of ER. The reason for the interaction between p53 and ER is unknown but may reflect qualitatively different p53 mutations underlying the p53 expression in tumors with or without ER expression.

TRIAL REGISTRATION

Current Controlled Trials ACTRN12607000037404 (Trial VIII) and ACTRN12607000029493 (Trial IX).

Comparison of Artificial Neural Network with Logistic Regression as Classification Models for Variable Selection for Prediction of Breast Cancer Patient Outcomes

The aim of this study was to compare multilayer perceptron neural networks (NNs) with standard logistic regression (LR) to identify key covariates impacting on mortality from cancer causes, disease-free survival (DFS), and disease recurrence using Area Under Receiver-Operating Characteristics (AUROC) in breast cancer patients. From 1996 to 2004, 2,535 patients diagnosed with primary breast cancer entered into the study at a single French centre, where they received standard treatment. For specific mortality as well as DFS analysis, the ROC curves were greater with the NN models compared to LR model with better sensitivity and specificity. Four predictive factors were retained by both approaches for mortality: clinical size stage, Scarff Bloom Richardson grade, number of invaded nodes, and progesterone receptor. The results enhanced the relevance of the use of NN models in predictive analysis in oncology, which appeared to be more accurate in prediction in this French breast cancer cohort.

Partial logistic artificial neural network for competing risks regularized with automatic relevance determination

Time-to-event analysis is important in a wide range of applications from clinical prognosis to risk modeling for credit scoring and insurance. In risk modeling, it is sometimes required to make a simultaneous assessment of the hazard arising from two or more mutually exclusive factors. This paper applies to an existing neural network model for competing risks (PLANNCR), a Bayesian regularization with the standard approximation of the evidence to implement automatic relevance determination (PLANNCR-ARD). The theoretical framework for the model is described and its application is illustrated with reference to local and distal recurrence of breast cancer, using the data set of Veronesi (1995).

Assessment of gastric cancer survival: using an artificial hierarchical neural network

This study is designed to assess the application of neural networks in comparison to the Kaplan-Meier and Cox proportional hazards model in the survival analysis. Three hundred thirty gastric cancer patients admitted to and surgically treated were assessed and their post-surgical survival was determined. The observed baseline survival was determined with the three methods of Kaplan-Meier product limit estimator, Cox and the neural network and results were compared. Then the binary independent variables were entered into the model. Data were randomly divided into two groups of 165 each to test the models and assess the reproducibility. The Chi-square test and the multiple logistic model were used to ensure the groups were similar and the data was divided randomly. To compare subgroups, we used the log-rank test. In the next step, the probability of survival in different periods was computed based on the training group data using the Cox proportional hazards and a neural network and estimating Cox coefficient values and neural network weights (with 3 nodes in hidden layer). Results were used for predictions in the test group data and these predictions were compared using the Kaplan-Meier product limit estimator as the gold standard. Friedman and Kruskal-Wallis tests were used for comparisons as well. All statistical analyses were performed using SPSS version 11.5, Matlab version 7.2, Statistica version 6.0 and S_PLUS 2000. The significance level was considered 5% (alpha = 0.05). The three methods used showed no significance difference in base survival probabilities. Overall, there was no significant difference among the survival probabilities or the trend of changes in survival probabilities calculated with the three methods, but the 4 year (48th month) and 4.5 year (54th month) survival rates were significantly different with Cox compared to standard and estimated probabilities in the neural network (p < 0.05). Kaplan-Meier and Cox showed almost similar results for the baseline survival probabilities, but results with the neural network were different: higher probabilities up to the 4th year, then comparable with the other two methods. Estimates from Cox proportional hazards and the neural network with three nodes in hidden layer were compared with the estimate from the Kaplan-Meier estimator as the gold standard. Neither comparison showed statistically significant differences. The standard error ratio of the two estimate groups by Cox and the neural network to Kaplan-Meier were no significant differences, it indicated that the neural network was more accurate. Although we do not suggest neural network methods to estimate the baseline survival probability, it seems these models is more accurately estimated as compared with the Cox proportional hazards, especially with today's advanced computer sciences that allow complex calculations. These methods are preferable because they lack the limitations of conventional models and obviate the need for unnecessary assumptions including those related to the proportionality of hazards and linearity.

Breast cancer predictions by neural networks analysis: a comparison with logistic regression

This paper presents an exploratory fixed time study to identify the most significant covariates as a precursor to a longitudinal study of specific mortality, disease free survival and disease recurrences. The data comprise consecutive patients diagnosed with primary breast cancer and entered into the study from 1996 at a single French clinical center, Centre Léon Bérard, based in Lyon, where they received standard treatment. The methodology was to compare and contrast multi-layer perceptron neural networks (NN) with logistic regression (LR), to identify key covariates and their interactions and to compare the selected variables with those routinely used in clinical severity of illness indices for breast cancer. The Logistic regression in this work was chosen as an accepted standard for prediction by biostatisticians in order to evaluate the neural network. Only covariates available at the time of diagnosis and immediately following surgery were used. We used for comparison classification performance indices: AUROC (AREA Under Receiver-Operating Characteristics) curves, sensitivity, specificity, accuracy and positive predictive value for the two following events of interest: Specific Mortality and Disease Free Survival.

Role of endocrine responsiveness and adjuvant therapy in very young women (below 35 years) with operable breast cancer and node negative disease

BACKGROUND

There is limited knowledge about prognosis, and treatment effects in young women with node-negative disease.

PATIENTS AND METHODS

We evaluated biological features, treatment recommendations and prognosis for 841 premenopausal patients with pT1-3, pN0 and M0, operated from 1997 to 2001.

RESULTS

Patients below 35 years (101, 12%) were more likely to have tumors > 2 cm (35.6% versus 24.2%, P = 0.002), grade 3 (48.5% versus 31.9%, P = 0.009) and with elevated Ki-67 expression (62.4% versus 50.7%, P = 0.002). At the multivariate analysis a statistically significant difference in disease-free survival (DFS, HR 4.44; 95% CI 2.53 to 7.78, P < 0.0001), risk of distant metastases (DDFS) (HR 3.23; 95% CI 1.32 to 7.94, P = 0.011) and overall survival (OS) (HR 2.89; 95% CI 1.06 to 7.87, P = 0.038) was observed for younger versus older patients and in the subgroup with endocrine responsive tumors (DFS, HR 5.17, 95% CI 2.72-9.83, P = < 0.0001; DDFS, 3.76, 95% CI 1.33-10.6, P = 0.013; OS, 4.71, 95% CI 1.09-20.4, P = 0.039 ).

CONCLUSIONS

Compared with less young, very young patients with endocrine responsive and node-negative breast cancer have a worse prognosis. Tailored treatments should be explored in this cohort of patients.

Artificial neural network for the joint modelling of discrete cause-specific hazards

OBJECTIVE

Artificial neural network (ANN) based regression methods have been introduced for modelling censored survival data to account for complex prognostic patterns. In the framework of ANN extensions of generalized linear models for survival data, PLANN is a partial logistic ANN, suitable for smoothed discrete hazard estimation as a function of time and covariates. An extension of PLANN for competing risks analysis (PLANNCR) is now proposed for discrete or grouped survival times, resorting to the multinomial likelihood.

METHODS AND MATERIALS

PLANNCR is built by assigning input nodes to the explanatory variables with the time interval treated as an ordinal variable. The logistic function is used as activation for the hidden nodes of the network, whereas the softmax, which corresponds to the canonical link of generalized linear models for polytomous regression, is adopted for multiple output nodes, to provide a smoothed estimation of discrete conditional event probabilities for each event. The Kullback-Leibler distance is used as error function for the target vectors, amounting to half of the deviance of a multinomial logistic regression model. PLANNCR can jointly model non-linear, non-proportional and non-additive effects on cause-specific hazards (CSHs). The degree of smoothing is modulated by the number of hidden nodes and penalization of the error function (weight decay). Model optimisation is achieved by quasi-Newton algorithms, while non-linear cross-validation (NCV) and the Network Information Criterion (NIC) were adopted for model selection. PLANNCR was applied to data on 1793 women with primary invasive breast cancer, histologically N-, who underwent surgery at the Milan Cancer Institute between 1981 and 1986.

RESULTS

Differential effects of covariates and time on the shape of the CSH for the three main failure causes, namely intra-breast tumor recurrences, distant metastases and contralateral breast cancer, have been enlightened.

CONCLUSIONS

PLANNCR can be suitably adopted in an exploratory framework for a thorough evaluation of the disease dynamics in the presence of competing risks.

Improvement of breast cancer relapse prediction in high risk intervals using artificial neural networks

The objective of this study is to compare the predictive accuracy of a neural network (NN) model versus the standard Cox proportional hazard model. Data about the 3811 patients included in this study were collected within the 'El Alamo' Project, the largest dataset on breast cancer (BC) in Spain. The best prognostic model generated by the NN contains as covariates age, tumour size, lymph node status, tumour grade and type of treatment. These same variables were considered as having prognostic significance within the Cox model analysis. Nevertheless, the predictions made by the NN were statistically significant more accurate than those from the Cox model (p < 0.0001). Seven different time intervals were also analyzed to find that the NN predictions were much more accurate than those from the Cox model in particular in the early intervals between 1-10 and 11-20 months, and in the later one considered from 61 months to maximum follow-up time (MFT). Interestingly, these intervals contain regions of high relapse risk that have been observed in different studies and that are also present in the analyzed dataset.

High-sensitivity and specificity of laser-induced autofluorescence spectra for detection of colorectal cancer with an artificial neural network

An artificial neural network (ANN) has been used in various clinical research for the prediction and classification of data in cancer disease. Previous research in this direction focused on the correlation between various input parameters such as age, antigen, and size of tumor growth. Recently, laser-induced autofluorescence (LIAF) techniques have been shown to be a useful noninvasive early diagnostic tool for various cancer diseases. We report on a successful application of ANN to in vitro LIAF spectra. We show that classification of tumor samples with ANN can be done with high sensitivity, specificity, and accuracy. Thus a combination of LIAF techniques and ANN can provide a robust method for clinical diagnosis.

Recent translational research: computational studies of breast cancer

The combination of mathematics – queen of sciences – and the general utility of computers has been used to make important inroads into insight-providing breast cancer research and clinical aids. These developments are in two broad areas. First, they provide useful prognostic guidelines for individual patients based on historic evidence. Second, by suggesting numeric tumor growth laws that are correlated to clinical parameters, they permit development of biologically relevant theories and comparison with patient data to help us understand complex biologic processes. These latter studies have produced many new ideas that are testable in clinical trials. In this review we discuss these developments from a clinical perspective, and ask whether and how they translate into useful tools for patient treatment.