J.C. van Houwelingen

Ridge estimators in logistic regression

SUMMARY In this paper it is shown how ridge estimators can be used in logistic regression to improve the parameter estimates and to diminish the error made by further predictions. Different ways to choose the unknown ridge parameter are discussed. The main attention focuses on ridge parameters obtained by cross-validation. Three different ways to define the prediction error are considered: classification error, squared error and minus log-likelihood. The use of ridge regression is illustrated by developing a prognostic index for the two-year survival probability of patients with ovarian cancer as a function of their deoxyribonucleic acid (DNA) histogram. In this example, the number of covariates is large compared with the number of observations and modelling without restrictions on the parameters leads to overfitting. Defining a restriction on the parameters, such that neighbouring intervals in the DNA histogram differ only slightly in their influence on the survival, yields ridge-type parameter estimates with reasonable values which can be clinically interpreted. Furthermore the model can predict new observations more accurately.

Prevention of Cisplatin Neurotoxicity with an ACTH(4–9) Analogue in Patients with Ovarian Cancer

Abstract In a randomized, double-blind, placebo-controlled study, we assessed the efficacy of an ACTH(4–9) analogue, Org 2766, in the prevention of cisplatin neuropathy in 55 women with ovarian cancer. The analogue was given subcutaneously in a dose of 0.25 mg (low dose) or 1 mg (high dose) per square meter of body-surface area before and after treatment with cisplatin and cyclophosphamide (75 and 750 mg per square meter every three weeks). The threshold of vibration perception was used as the principal measure of neurotoxicity. After four cycles of chemotherapy, the mean (±SEM) threshold value for vibration perception in the placebo group increased from 0.67±0.12 to 1.61±0.43 μm of skin displacement (P<0.0001). In the high-dose treatment group, there was no increase in the threshold value after four cycles (from 0.54±0.12 to 0.50±0.06 μm). After six cycles of chemotherapy, the threshold value was 5.87±1.97 μm in the placebo group (more than an eightfold increase from base line), as compared with 0.88±0.1...

Cancer risks in BRCA2 families: estimates for sites other than breast and ovary

Background: In BRCA2 mutation carriers, increased risks have been reported for several cancer sites besides breast and ovary. As most of the families included in earlier reports were selected on the basis of multiple breast/ovarian cancer cases, it is possible that risk estimates may differ in mutation carriers with a less striking family history. Methods: In the Netherlands, 139 BRCA2 families with 66 different pathogenic mutations were included in a nationwide study. To avoid testing bias, we chose not to estimate risk in typed carriers, but rather in male and female family members with a 50% prior probability of being a carrier (n = 1811). The relative risk (RR) for each cancer site with the exception of breast and ovarian cancer was determined by comparing observed numbers with those expected, based on Dutch cancer incidence rates. Results: We observed an excess risk for four cancer sites: pancreas (RR 5.9; 95% confidence interval (CI) 3.2 to 10.0), prostate (2.5; 1.6 to 3.8), bone (14.4; 2.9 to 42.1) and pharynx (7.3; 2.0 to 18.6). A small increase was observed for cancer of the digestive tract (1.5; 1.1 to 1.9). Histological verification was available for 46% of the tumours. Nearly all increased risks reached statistical significance for men only. Cancer risks tended to be higher for people before the age of 65 years. Moreover, families with mutations outside the previously defined ovarian cancer cluster region tended to have a higher cancer risk. Conclusions: We found that BRCA2 carriers are at increased risk for cancers of the prostate and pancreas, and possibly bone and pharynx. Larger databases with extended follow up are needed to provide insight into mutation specific risks of selected carriers in BRCA2 families.

Decreased interleukin-10 and increased interleukin-12p40 mRNA are associated with disease activity and characterize different disease stages in multiple sclerosis.

It has been shown that proinflammatory and antiinflammatory cytokines correlate with disease activity in multiple sclerosis (MS). To establish whether such correlations depend on the disease stage, we assessed in a longitudinal fashion the expression of interleukin (IL)‐12 (p40 and p35), tumor necrosis factor‐α, interferon‐γ, and IL‐10 mRNA by competitive polymerase chain reaction in unstimulated peripheral blood mononuclear cells of relapsing–remitting (RR) and secondary progressive (SP) MS patients, in relation to monthly clinical and magnetic resonance imaging monitoring. MS patients had increased levels of IL‐12p40 and decreased levels of IL‐10 mRNA compared with controls; this difference was most pronounced in SP patients. Both RR and SP patients had increased levels of IL‐12p40 mRNA compared with controls during the development of active lesions. Moreover, in RR MS an increase was found before relapse. IL‐12p35 mRNA was decreased in both groups, and in relation to disease activity it showed a pattern different from IL‐12p40 mRNA. In RR MS, IL‐10 mRNA was low 4 weeks before magnetic resonance imaging activity and 6 weeks before relapse; a significant increase to normal levels was noted when active lesions became apparent. In contrast, SP patients showed low IL‐10 mRNA levels constitutively, suggesting that IL‐10 plays an important role in the control of disease progression. Ann Neurol 1999;45:695–703

A Goodness-of-Fit Test for Binary Regression Models, Based on Smoothing Methods

A new global test statistic for models with continuous covariates and binary response is introduced. The test statistic is based on nonparametric kernel methods. Explicit expressions are given for the mean and variance of the test statistic. Asymptotic properties are considered and approximate corrections due to parameter estimation are presented. Properties of the test statistic are studied by simulation. The goodness-of-fit method is illustrated on data from a Dutch follow-up study on preterm infants. Recommendations for practitioners are given.

Randomised Trial Comparing Two Combination Chemotherapy Regimens (HEXA-CAF VS CHAP-5) In Advanced Ovarian Carcinoma

186 patients with advanced epithelial ovarian carcinoma were treated with either a combination of hexamethylmelamine, cyclophosphamide, methotrexate, and 5-fluorouracil (Hexa-CAF) or cyclophosphamide and hexamethylmelamine alternating with doxorubicin and a 5-day course of cisplatin (CHAP-5). Treatment with CHAP-5 resulted in more complete remissions as determined by laparatomy or peritoneoscopy (p = 0.004), better overall response (p = 0.0001), and longer overall survival and progression-free survival (p less than 0.002). Therapy, histological grade, and Karnofsky index were reliable predictors of overall response, whereas therapy, FIGO-stage, and size of residual tumour before chemotherapy were independent predictors for complete remission and for prolonged survival. Peripheral neurotoxicity was a major problem in patients assigned to the CHAP-5-group and was likely to be due to the simultaneous administration of hexamethylmelamine and cisplatin. The CHAP-5 regimen is one of the most effective regimens for the initial treatment of ovarian cancer.

Logistic Regression for Correlated Binary Data

The modelling of correlated binary outcomes, in such a way that the marginal response probabilities are still logistic, is considered. Different association measures for the dependence between correlated observations are discussed. For paired correlated data the full likelihood can be evaluated; for an arbitrary number of correlated observations a pseudolikelihood approach to obtain parameter estimates is proposed. The results are illustrated on data from a Dutch follow‐up study on preterm infants.

Reversibility as a criterion for discriminating time series

Abstract We propose a test for the hypothesis that a time series is reversible. If reversibility can be rejected all static transformations of linear Gaussian random processes can be excluded as a model for the time series.

Incidence of neuropathy in 395 patients with ovarian cancer treated with or without cisplatin

In two consecutive trials, a total of 395 patients with ovarian cancer were treated with a combination of hexamethylmelamine, cyclophosphamide, methotrexate, and 5‐fluorouracil chemotherapeutic regimens including cisplatin or without this drug. With respect to neurotoxicity, 387 patients were fully eligible. The median follow‐up for survival was 45 months. Neurotoxicity in any grade of severity developed in 47% of the patients treated with a cisplatin‐containing regimen and in 25% of those treated with the non‐cisplatin‐containing regimen. The severity of neurotoxicity was much higher, however, in the cisplatin‐treated patients. Neurotoxicity‐free survival decreased below 50% at cumulative doses of cisplatin between 500 and 600 mg/m2. No additional effect of hexamethylmelamine on the incidence or severity of neurotoxicity could be demonstrated. In patients who survived for more than 5 years, the incidence of cisplatin neuropathy was 61%. Prognostic variables (age, International Federation of Gynecology and Obstetrics [FIGO] stage, performance status, and others) possibly associated with high‐risk subgroups could not be identified. The only consistent factor correlated with neurotoxicity was the total dose of cisplatin received.

A meta-analysis of prognostic factors in advanced ovarian cancer with median survival and overall survival (measured with the log (relative risk)) as main objectives

We performed a meta-analysis of 38 articles containing 66 treatment groups and 3443 patients in order to evaluate prognostic factors in advanced epithelial ovarian cancer. To evaluate overall survival we designed a method to summarize the overall survival curve into one single figure: the log (relative risk) (LRR). This is the first meta-analysis using overall survival (measured with the LRR) as an objective. We found that the main prognostic factors predicting an improved survival (measured with the LRR) are: chemotherapy including cisplatin as initial treatment, a residual tumour mass of less than 2 cm prior to therapy, FIGO stage II/III and a good performance status. In a multivariate model, the use of cisplatin and the residual tumour were found to be the only factors of prognostic relevance. No relation between median survival and the overall clinical response rate of all patients entered in the denominator, could be demonstrated. Undifferentiated tumours and patients treated with cisplatin regimens had higher response rates to treatment but younger patients and those with endometrioid histology were less likely to respond. A surgical complete remission was encountered more frequently among studies that included a high number of patients with small tumour masses prior to treatment. Trials using cisplatin included more patients with small tumour nodules in their patient material compared to studies not using this drug. The data illustrate the danger of comparing studies with each other. In the trials with a high percentage of patients with small tumour residuals in the study population more toxic deaths were seen. This probably reflects the fact that they had received more intensive treatment. The LRR correlated strongly with the median survival, response and the percentage of surgical complete remissions. We concluded that the introduction of the LRR can be a meaningful addition to the evaluation of the influence of prognostic factors on overall survival.