Comparing the predictive value of neural network models to logistic regression models on the risk of death for small-cell lung cancer patients

Cancer is one of the leading causes of mortality in the developed world, and prognostic assessment of cancer patients is indispensable in medical care. Medical researchers are accustomed to using regression models to predict patient outcomes. Neural networks have been proposed as an alternative with great potential. Nonetheless, empirical evidence remains lacking to support the application of this technique as the appropriate method to investigate cancer prognosis. Utilizing data on patients from two National Cancer Institute of Canada clinical trials, we compared predictive accuracy of neural network models and logistic regression models on risk of death of limited-stage small-cell lung cancer patients. Our results suggest that neural network and logistic regression models have similar predictive accuracy. The distributions of individual predicted probabilities are very similar. On occasion, however, the prediction pairs are quite different, suggesting that they do not always give the same interpretations of the same variables.     

Prognostic significance of micrometastasis in non-small-cell lung cancer

Accurate staging of non-small-cell lung cancer (NSCLC) determines prognosis and facilitates decisions regarding treatment options. Unfortunately, even after an apparently complete resection in patients with stage I disease, the recurrence rates range from 25% to 50%, and overall survival is not encouraging. One possible reason for this may be that those patients with a poor outcome actually have more extensive disease, with occult locoregional and/or distant metastasis than originally identified by routine pathologic staging techniques. There is now a sizable body of literature on the detection and possible prognostic role of occult disease in lung cancer. The majority of these studies are based on immunohistochemical analysis of lymph nodes and/or bone marrow, but a handful of studies use molecular approaches. The purpose of this review is to summarize and critique the current literature on occult tumor cell spread to lymph nodes and bone marrow in patients with NSCLC. Based on this literature, we believe that the prognostic significance of bone marrow micrometastasis remains unclear. However, the majority of studies indicate that occult lymph node disease is associated with a poor outcome. Thus, our ability to detect individual tumor cells could result in more accurate staging of NSCLC in patients and would potentially lead to the development of novel therapies, as well as influence decisions regarding the use of appropriate multimodality treatment strategies, the choice of surgical technique, and extent of dissection. As data accumulate, the presence or absence of occult nodal involvement should probably be considered at the next revision of the staging system for NSCLC.     

Prognostic factors in stage III non-small-cell lung cancer

PURPOSE: The aim of this study is to present an analysis investigating the association of patient characteristics with overall survival (OS) in individuals with stage III non-small-cell lung cancer (NSCLC). PATIENTS AND METHODS: Ten prognostic factors were analyzed in 203 patients with NSCLC who were enrolled in a phase III trial conducted by the Hoosier Oncology Group and US Oncology between 2002 and 2006. Eligible patients had untreated stage III NSCLC, forced expiratory volume in one second (FEV(1)) > or = 1 liter, baseline performance status of 0/1, and weight loss < 5% in 3 months preceding the trial. Univariate analysis, Cox proportional hazards regression, and parametric accelerated failure time models were performed to identify the factors that affected survival duration. Variables analyzed included age (< 70 years vs. > or = 70 years), sex, ethnicity, body mass index, performance status (0 vs. 1), FEV(1) (> 2 L vs. 1-2 L), smoking status (current vs. never/former), hemoglobin (Hb) level, use of positron emission tomography scan in staging, and stage (IIIA vs. IIIB). RESULTS: Median follow-up was 25.6 months and the median OS was 21.2 months. The univariate analysis showed that Hb levels > or = 12 were associated with an improved survival (P = .033). The multivariable parametric accelerated failure time model demonstrated the association of FEV(1) > 2 L (P = .014), and higher pretreatment Hb values (P = .007) as independent prognostic factors for OS. Similarly in the Cox regression, survival was influenced by Hb and FEV(1) > 2 L. CONCLUSION: This analysis suggests that FEV(1) > 2 L and higher pretreatment Hb values are associated with improved OS in patients with stage III NSCLC. These factors can be useful in predicting for more favorable outcomes in patients with stage III NSCLC and provide additional information when designing future studies.     

Prediction of survival in patients with esophageal carcinoma using artificial neural networks

BACKGROUND: Accurate estimation of outcome in patients with malignant disease is an important component of the clinical decision-making process. To create a comprehensive prognostic model for esophageal carcinoma, artificial neural networks (ANNs) were applied to the analysis of a range of patient-related and tumor-related variables. METHODS: Clinical and pathologic data were collected from 418 patients with esophageal carcinoma who underwent resection with curative intent. A data base that included 199 variables was constructed. Using ANN-based sensitivity analysis, the optimal combination of variables was determined to allow creation of a survival prediction model. The accuracy (area under the receiver operator characteristic curve [AUR]) of this ANN model subsequently was compared with the accuracy of the conventional statistical technique: linear discriminant analysis (LDA). RESULTS: The optimal ANN models for predicting outcomes at 1 year and 5 years consisted of 65 variables (AUR = 0.883) and 60 variables (AUR = 0.884), respectively. These filtered, optimal data sets were significantly more accurate (P < 0.0001) than the original data set of 199 variables. The majority of ANN models demonstrated improved accuracy compared with corresponding LDA models for 1-year and 5-year survival predictions. Furthermore, ANN models based on the optimal data set were superior predictors of survival compared with a model based solely on TNM staging criteria (P < 0.0001). CONCLUSIONS: ANNs can be used to construct a highly accurate prognostic model for patients with esophageal carcinoma. Sensitivity analysis based on ANNs is a powerful tool for seeking optimal data sets.     

环氧合酶-2和表皮生长因子受体在非小细胞肺癌的表达及意义

目的 :探讨环氧合酶 2 (COX 2 )和表皮生长因子受体 (EGFR)在非小细胞肺癌 (NSCLC)中的表达与意义。方法 :应用免疫组织化学染色EnVision法检测 6 0例根治术后NSCLC肿瘤组织中COX 2和EGFR的表达。应用 χ2 检验和COX回归分析等比较COX 2和EGFR在不同分化级别的肺鳞癌和腺癌中表达上的差异及其与术后患者生存期之间的关系。结果 :77%肺腺癌表达COX 2 ,表达率明显高于肺鳞癌的 37% (χ2 =9.774 ,P <0 .0 1) ;NSCLC中COX 2和EGFR的表达与患者年龄、性别、肿瘤分化级别、原发灶大小、淋巴结转移范围和 p TNM分期等因素无关。COX 2表达“ ”和“ ”和EGFR表达“ ”与“ ”患者的中位生存期分别为 30个月和 16个月与 34个月和 15个月 ,均明显低于各自不表达患者的 4 5个月 (P <0 .0 5 )。COX多因素回归分析示COX 2表达和p TNM分期是影响该组NSCLC患者预后的两个独立因素。在肺鳞癌 ,EGFR也与预后不良有关。 结论 :COX 2和EGFR在NSCLC有表达 ,表达者预后较差。     

Survival prediction using artificial neural networks in patients with uterine cervical cancer treated by radiation therapy alone

Background: We evaluated the usefulness of artificial neural networks (ANNs) for survival prediction in patients with uterine cervical cancer treated by radiotherapy. Methods: We used data from 134 patients with uterine cervical cancer treated by combined external and high-dose-rate remote afterloading intracavitary radiotherapy between 1978 and 1993. The ANNs were trained using the data from 67 randomly selected patients. Using the trained ANNs, we predicted the 5-year survival in the remaining 67 patients, and compared it with the known 5-year survival. The performance of the ANNs was evaluated using a receiver operating characteristic (ROC) curve and was compared using the area under the ROC curve (Az). Results: When fundamental factors, such as age, performance status, hemoglobin, total protein, International Federation of Gynecology and Obstetrics (FIGO) stage, and histological type were used as inputs in the ANNs, Az was 0.5483 ± 0.0145 (mean ± SD). When the histological grading of radiation effect determined by periodic biopsy examination was used in addition to the fundamental factors, Az was highest (0.7782 ± 0.0105). When the cytological grading of radiation effect by the periodic smear was used in addition to the fundamental factors, Az was 0.5523 ± 0.0135, which was not significantly different from that when only the fundamental factors were used. Conclusion: ANNs allow us to evaluate the importance of prognostic factors, and make it possible to predict the survival of each patient. Using ANNs, the combination of histological grading of radiation effect determined by periodic biopsy examination, in addition to the fundamental factors, is the most effective for prediction of survival in patients with uterine cervical cancer. Received: November 8, 2001 / Accepted: June 7, 2002 Acknowledgments The authors are grateful to Masaji Takahashi, M.D. (Professor Emeritus, Kyoto University), Masahiro Hiraoka, M.D. (Professor, Department of Therapeutic Radiology and Oncology, Kyoto University), and Masaaki Kataoka, M.D. (Shikoku Cancer Center Hospital), for their valuable discussions and advice in carrying out this study. This work was presented, in part, at the 86th RSNA meeting in Chicago, USA, November 26 to December 1, 2000. Correspondence to:T. Ochi     

Predicting two-year quality of life after breast cancer surgery using artificial neural network and linear regression models

Accelerated partial breast radiotherapy (RT) strategies (3-D conformal external-beam RT (3-D CRT) and brachytherapy with balloon catheter) reduce time and transportation burdens of whole breast RT for breast cancer. Long-term clinical trial evidence is unavailable for accelerated modalities, but uncertainty might be acceptable for patients likely to receive suboptimal whole breast RT. The objective of this study is to assess the cost effectiveness of accelerated partial breast RT compared to on-time and delayed whole breast RT. The design used in this study is decision analytic Markov model. The data sources are published literature; and national/federal sources. The target population of this study is a hypothetical cohort of 60 years old women previously treated with breast-conserving surgery for node-negative, estrogen receptor-positive breast cancer with tumors <1 cm. The time horizon is 15 years, and the perspective is societal. The interventions are whole breast RT, 3-D CRT, and brachytherapy breast irradiation. The outcome measures are costs (2008 US$), quality-adjusted life years (QALYs), and incremental cost-effectiveness ratios. The base-case results were: 3-D CRT was the preferred strategy, costing on average $10,800 and yielding 11.21 QALYs. On-time whole breast RT costs $368,000/QALY compared to 3-D CRT, above the $100,000/QALY WTP threshold. 3-D CRT was also preferred over delayed whole breast RT. Brachytherapy was never preferred. Sensitivity analysis indicated that the results were sensitive to the rate of recurrence outside the initial tumor quadrant (“elsewhere failure”) in one-way analysis. Probabilistic sensitivity analysis indicated that results were sensitive to parameter uncertainty, and that the elsewhere-failure rate and treatment preferences may drive results. The limitation of this study is that efficacy estimates are derived from studies that may not fully represent the population modeled. As a conclusion, 3-D CRT was preferred to whole breast RT and for women likely to delay RT, indicating that 3-D CRT could be targeted more efficiently before randomized trial evidence.     

Therapy for non-small-cell lung cancer patients with brain metastasis Therapy for non-small-cell lung cancer patients with brain metastasis

Brain metastasis is a major cause of poor prognosis and high mortality for non-smal celllung cancer patients. The prognosis of non-smal-celllung cancer (NSCLC) patients with brain metastasis is general...     

Prognostic factors to predict survival in non-small-cell lung cancer with brain metastasis

Objective： The purpose of the study was to assess prognostic factors to predict overall survival （OS） and progres- sion-free survival （PFS） in non-small-cell lung cancer （NSCLC） with brain metastasis （BM）. Methods： From November 2011 to March 2013, the clinical data of 31 NSCLC cases with BM treated with multiple modalities including brain radiotherapy alone, systemic chemotherapy, whole brain radiotherapy （WBRT） combined with tyrosine kinase inhibitor （TKIs）. The efficacy and adverse reaction were evaluated after treatment. Results： In terms of intracranial lesions, the objective response rate （ORR） and the disease control rate （DCR） were 22.6% and 90.3%, respectively. As for systemic disease, ORR and DCR were 32.3% and 93.5%, respectively. The median time to progression-free survival （PFS） was 298 days （95% CI： 258.624-337.376 days）, whereas in the epidermal growth factor receptor （EGFR） mutation patients was 331 days. Patients who received EGFR-TKIs combined with brain radiation had better response rate （RR） than those only brain radiation. Univariate analysis showed that the EGFR-mutations could predictive factors for PFS, and not to other clinical pathological features. The most common toxicities were rash and diarrhea, but all were well-tolerated. Conclusion： EGFR-mutations is the independent prognostic factors affecting the survival rates of NSCLC patients with BM. Through the clinical observation, icotinib combined with WBRT may be effective on brain metastases in NSCLC patients, and toxicities are tolerable, which worth further study.     

Prognostic significance of osteopontin expression in early-stage non-small-cell lung cancer

Osteopontin (OPN) is a multifunctional protein, which has recently been shown to be linked to tumorigenesis, progression and metastasis in different malignancies. Since non-small-cell lung cancer (NSCLC)'s prognosis remains bad, with few predictors of outcome, the purpose of this study was to evaluate if OPN might be involved in NSCLC's biology and therefore represent a prognostic marker and a target for new therapeutic trials. Immunohistochemistry was used to detect OPN expression, evaluated as percentage of neoplastic cells with cytoplasmic immunoreactivity, in a wide cohort of patients with stage I NSCLC (136 cases). The median value of this series (20% of positive cells) was used as the cutoff value to distinguish tumours with low (<20%) from tumours with high (> or =20%) OPN expression. A statistically significant correlation between high levels of OPN and shorter overall (P = 0.034) and disease-free (P = 0.011) survival in our patients was shown. Our results support the hypothesis that high OPN expression is a significantly unfavourable prognostic factor for the survival of patients with stage I NSCLC. This conclusion has notable importance in terms of the biological characterization of early-stage tumours and therapeutic opportunities.     

Prognostic value of genomic damage in non-small-cell lung cancer.

Genomic alterations have been analysed in 65 non-small-cell lung cancer (NSCLC) tissue samples by using the arbitrarily primed polymerase chain reaction (AP-PCR), which is a PCR-based genomic fingerprinting. We have shown that AP-PCR may be applied as a useful and feasible practical method for detection of the genomic alterations that accompany malignancy in NSCLC. Genomic changes detected by us consisted of: allelic losses or gains in anonymous DNA sequences, homozygously deleted DNA sequences and polymorphic DNA sequences. According to these genomic changes, lung tumours evaluated in the present study have been scored into three groups: low, moderate and high genomic damage tumours. The aim of this study was to investigate the effect of genomic damage on patient survival. Survival analysis was carried out in 51 NSCLC patients. Our results revealed that high genomic damage patients showed a poorer prognosis than those with low or moderate genomic damage (P = 0.038). Multivariate Cox regression analysis showed that patients with higher genomic alterations displayed an adjusted-by-stage risk ratio 4.26 times higher than the remaining patients (95% CI = 1.03-17.54). We can conclude that genomic damage has an independent prognostic value of poor clinical evolution in NSCLC.     

A comparison of the logistic regression and the cox proportional hazard models in retrospective studies on the prognosis of patients with castric cancer

To define the independent prognostic factors reducing survival time for gastric cancer, we compared the logistic regression and the Cox proportional hazard models applied to patients who underwent curative gastrec-tomy. All patients were evaluated after being followed for long fixed periods. Of 1,019, 269 (26.4%) died of tumor recurrence within a 5-year period and 36 (3.5%) died over 5 years after the original surgery. With regard to survival time, multivariate analyses using the Cox proportional hazard model in a stepwise manner adjusted for the sex, age, and 10 other factors, suggested that size of tumor (P<0.01, relative risk [rr] = 1.0962), degree of gastric wall invasion (P < 0.01, rr < 1.3520), and status of lymph node metastasis (P < 0.01, rr = 1.6572) were the most independent prognostic factors. As well as, using the stepwise logistic regression model, size of tumor, (P < 0.01, odds ratio [or] = 1.115), degree of gastric wall invasion (P < 0.01, or = 1.428), and status of lymph node metastasis (P < 0.01, or = 2.182) were also the most independent risk factors for recurrence within 5 years after surgery. Although regression coefficients are not all the same, these three factors proved significant in both multivariate analyses. This equation for risk factors for prognosis is approached when searching for an appropriate method of retrospective studies using multivariate analyses. © 1993 Wiley-Liss, Inc.     

Circadian function in patients with advanced non-small-cell lung cancer

This study aimed to evaluate whether patients with advanced non-small-cell lung cancer experience disrupted rest-activity daily rhythms, poor sleep quality, weakness, and maintain attributes that are linked to circadian function such as fatigue. This report describes the rest-activity patterns of 33 non-small-cell lung cancer patients who participated in a randomised clinical trial evaluating the benefits of melatonin. Data are reported on circadian function, health-related quality of life (QoL), subjective sleep quality, and anxiety/depression levels prior to randomisation and treatment. Actigraphy data, an objective measure of circadian function, demonstrated that patients' rest-activity circadian function differs significantly from control subjects. Our patients reported poor sleep quality and high levels of fatigue. Ferrans and Powers QoL Index instrument found a high level of dissatisfaction with health-related QoL. Data from the European Organization for Research and Treatment for Cancer reported poor capacity to fulfil the activities of daily living. Patients studied in the hospital during or near chemotherapy had significantly more abnormal circadian function than those studied in the ambulatory setting. Our data indicate that measurement of circadian sleep/activity dynamics should be accomplished in the outpatient/home setting for a minimum of 4-7 circadian cycles to assure that they are most representative of the patients' true condition. We conclude that the daily sleep/activity patterns of patients with advanced lung cancer are disturbed. These are accompanied by marked disruption of QoL and function. These data argue for investigating how much of this poor functioning and QoL are actually caused by this circadian disruption, and, whether behavioural, light-based, and or pharmacologic strategies to correct the circadian/sleep activity patterns can improve function and QoL.     

Supportive care in patients with advanced non-small-cell lung cancer

The present study describes supportive care (SC) in patients with advanced non-small-cell lung cancer (NSCLC), evaluating whether it is affected by concomitant chemotherapy, patient's performance status (PS) and age. Data of patients enrolled in three randomised trials of first-line chemotherapy, conducted between 1996 and 2001, were pooled. The analysis was limited to the first three cycles of treatment. Supportive care data were available for 1185 out of 1312 (90%) enrolled patients. Gastrointestinal drugs (45.7%), corticosteroids (33.4%) and analgesics (23.8%) were the most frequently observed categories. The mean number of drugs per patient was 2.43; 538 patients (45.4%) assumed three or more supportive drugs. Vinorelbine does not produce substantial variations in the SC pattern, while cisplatin-based treatment requires an overall higher number of supportive drugs, with higher use of antiemetics (41 vs 27%) and antianaemics (10 vs 4%). Patients with worse PS are more exposed to corticosteroids (42 vs 30%). Elderly patients require drugs against concomitant diseases significantly more than adults (20 vs 7%) and are less frequently exposed to antiemetics (12 vs 27%). In conclusion, polypharmacotherapy is a relevant issue in patients with advanced NSCLC. Chemotherapy does not remarkably affect the pattern of SC, except for some drugs against side effects. Elderly patients assume more drugs for concomitant diseases and receive less antiemetics than adults.     

Undertreatment of elderly patients with non-small-cell lung cancer

The absolute number of patients with lung cancer is rising as a result of our aging population. Until recently, clinicians have been reluctant to aggressively treat elderly patients with non-small-cell lung cancer (NSCLC) because of a lack of supportive data and concern for potential toxicity. Recently, evidence has emerged that suggests that, similar to younger patients, healthy elderly patients can benefit from therapy in all stages of NSCLC. This review will discuss the findings that indicate that chronologic age alone should not be a barrier to appropriate treatment for NSCLC, but consideration should be given to more important prognostic factors such as comorbidities and performance status.     

Prognostic factors for metachronous contralateral breast cancer: A comparison of the linear Cox regression model and its artificial neural network extension

The purpose of the present study was toassess prognostic factors for metachronous contralateral recurrence ofbreast cancer (CBC). Two factors were of particularinterest, namely estrogen (ER) and progesterone (PgR) receptorsassayed with the biochemical method in primary tumortissue. Information was obtained from a prospective clinicaldatabase for 1763 axillary node-negative women who hadreceived curative surgery, mostly of the conservative type,and followed-up for a median of 82 months.The analysis was performed based on both astandard (linear) Cox model and an artificial neuralnetwork (ANN) extension of this model proposed byFaraggi and Simon [9]. Furthermore, to assess theprognostic importance of the factors considered, model predictiveability was computed.In agreement with already published studies, the resultsof our analysis confirmed the prognostic role ofage at surgery, histology, and primary tumor site,in that young patients ( 45 years) withtumors of lobular histology or located at inner/centralmammary quadrants were at greater risk of developingCBC. ER and PgR were also shown tohave a prognostic role. Their effect, however, wasnot simple in relation to the presence ofinteractions between ER and age, and between PgRand histology. In fact, ER appeared to playa protective role in young patients, whereas theopposite was true in older women. Higher levelsof PgR implied a greater hazard of CBCoccurrence in infiltrating duct carcinoma or tumors withan associated extensive intraductal component, and a lowerhazard in infiltrating lobular carcinoma or other histotypes.In spite of the above findings, the predictivevalue of both the standard and ANN Coxmodels was relatively low, thus suggesting an intrinsiclimitation of the prognostic variables considered, rather thantheir suboptimal modeling. Research for better prognostic variablesshould therefore continue.     

Prognostic factors in advanced nonseminomatous testicular cancer. A multivariate logistic regression analysis

In order to define prognostic factors for advanced stage of nonseminomatous germ cell tumors (NSGCT) of the testis, the authors reviewed 84 patients treated from 1978 through 1985. The survival rate was 51% at 3 years. Patients with elevated seric levels of human chorionic gonadotropin (HCG) and/or alpha-fetoprotein (AFP), or the presence of an abdominal mass had significantly worse survival. Only HCG and AFP levels retained their significance when multivariate Cox analysis was performed. The probability that a patient achieves a complete remission (CR) was assessed by a function of certain patient characteristics using a multivariate logistic regression analysis. The significant variables were a function of HCG and AFP values. Since both variables are related to the CR rate and survival the authors define the obtention of a CR as a unique outcome of interest. The probability of a CR greater than 70% adequately separates the patients into two prognostic subgroups. This model currently is being used to enrole NSGCT patients in a prospective modulated clinical trial according to these prognostic factors.