Vertical modelling: Analysis of competing risks data with missing causes of failure.

We propose vertical modelling as a natural approach to the problem of analysis of competing risks data when failure types are missing for some individuals. Under a natural missing-at-random assumption for these missing failure types, we use the observed data likelihood to estimate its parameters and show that the all-cause hazard and the relative hazards appearing in vertical modelling are indeed key quantities of this likelihood. This fact has practical implications in that it suggests vertical modelling as a simple and attractive method of analysis in competing risks with missing causes of failure; all individuals are used in estimating the all-cause hazard and only those with non-missing cause of failure for relative hazards. The relative hazards also appear in a multiple imputation approach to the same problem proposed by Lu and Tsiatis and in the EM algorithm. We compare the vertical modelling approach with the method of Goetghebeur and Ryan for a breast cancer data set, highlighting the different aspects they contribute to the data analysis.

Comparing and ranking hospitals based on outcome: results from The Netherlands Stroke Survey.

BACKGROUND: Measuring quality of care and ranking hospitals with outcome measures poses two major methodological challenges: case-mix adjustment and variation that exists by chance. AIM: To compare methods for comparing and ranking hospitals that considers these. METHODS: The Netherlands Stroke Survey was conducted in 10 hospitals in the Netherlands, between October 2002 and May 2003, with prospective and consecutive enrollment of patients with acute brain ischaemia. Poor outcome was defined as death or disability after 1 year (modified Rankin scale of > or =3). We calculated fixed and random hospital effects on poor outcome, unadjusted and adjusted for patient characteristics. We compared the hospitals using the expected rank, a novel statistical measure incorporating the magnitude and the uncertainty of differences in outcome. RESULTS: At 1 year after stroke, 268 of the total 505 patients (53%) had a poor outcome. There were substantial differences in outcome between hospitals in unadjusted analysis (chi(2) = 48, 9 df, P < 0.0001). Adjustment for 12 confounders led to halving of the chi(2) (chi(2) = 24). The same pattern was observed in random effects analysis. Estimated performance of individual hospitals changed considerably between unadjusted and adjusted analysis. Further changes were seen with random effect estimation, especially for smaller hospitals. Ordering by expected rank led to shrinkage of the original ranks of 1-10 towards the median rank of 5.5 and to a different order of the hospitals, compared to ranking based on fixed effects. CONCLUSION: In comparing and ranking hospitals, case-mix-adjusted random effect estimates and the expected ranks are more robust alternatives to traditional fixed effect estimates and simple rankings.

Score test for age at onset genetic linkage analysis in selected sibling pairs.

A new score statistic is derived, which uses information from registries (age-specific incidences) and family studies (sib-sib marginal correlation) to weight affected sibling pairs according to their age at onset. Age at onset of sibling pairs is modelled by a gamma frailty model. From this model we derive a bivariate survival function, which depends on the marginal survival and on the marginal correlation. The score statistic for linkage is a classical nonparametric linkage (NPL) statistic where the identical by descent sharing is weighted by a particular function of the age at onset data. Since the statistic is based on survival models, it can also be applied to discordant and healthy sibling pairs. Simulation studies show that the proposed method is robust and more powerful than standard NPL methods. As illustration we apply the new score statistic to data from a breast cancer study.

A polygenic model for integration of linkage and pathway information.

We introduce an approximate model for linkage curves which accommodates the polygenic structure of complex diseases and accounts for the simultaneous action of closely located genes. The model is extended so that information on biological pathways can be integrated. Using data on rheumatoid arthritis, we describe some of the many applications which the model allows: it can be used to test for residual linkage in the presence of already established loci, to derive a global test for linkage, to test for the relevance of a gene list in terms of linkage and to help in candidate gene prioritization by integration of gene-pathway annotation data.

[Early surgery or a wait-and-see policy in lumbosacral radicular syndrome: a randomized study]. / Vroeg opereren of langer afwachten voor het lumbosacraal radiculair syndroom: een gerandomiseerd onderzoek.

OBJECTIVE: To compare early surgery with expectative policy and later surgery if necessary in patients with sciatica that did not resolve within 6 weeks. DESIGN: Randomized multicentre clinical trial (ISRCTN 26872154). METHODS: Patients who had had severe sciatica for 6 to 12 weeks were randomized to early surgery or to prolonged conservative treatment with later surgery if necessary. The primary outcomes were the Roland Disability Questionnaire score, the visual-analogue scale for leg pain score, and the patient's report of their perceived recovery over the first year after randomization. Repeated measures analysis according to the intention-to-treat principle was used to analyse the outcome curves for both groups. RESULTS: A total of 283 patients were included and randomized. Of 141 patients assigned to undergo early surgery, 125 (89%) underwent microdiscectomy after a mean of 2.2 weeks. Of 142 patients assigned to conservative treatment, 55 (39%) still had to undergo surgical treatment after a mean of 18.7 weeks. There was no significant overall difference in disability scores during the first year (p = 0.13). Leg pain lessened more quickly in patients assigned to early surgery (p < 0.001). Patients assigned to early surgery also reported a faster rate of perceived recovery (hazard ratio (HR): 1.97; 95% CI: 1.72-2.22; p < 0.001). In both groups, however, the probability of perceived recovery after 1 year of follow-up was 95%. CONCLUSIONS: The 1-year outcomes were similar for patients assigned to early surgery and those assigned to extended conservative treatment with later surgery if necessary but the rates of reduction of leg pain and of perceived recovery were faster in those assigned to early surgery.

Score test for linkage in generalized linear models.

We derive a test for linkage in a Generalized Linear Mixed Model (GLMM) framework which provides a natural adjustment for marginal covariate effects. The method boils down to the score test of a quasi-likelihood derived from the GLMM, it is computationally inexpensive and can be applied to arbitrary pedigrees. In particular, for binary traits, relative pairs of different nature (affected and discordant) and individuals with different covariate values can be naturally combined in a single test. The model introduced could explain a number of situations usually described as gene by covariate interaction phenomena, and offers substantial gains in efficiency compared to methods classically used in those instances.

Azathioprine/methylprednisolone versus cyclophosphamide in proliferative lupus nephritis. A randomized controlled trial.

Until recently, intravenous cyclophosphamide pulses with oral corticosteroids were regarded standard therapy for proliferative lupus nephritis (LN). Azathioprine, a less toxic alternative, was never proven to be inferior. In the first Dutch lupus nephritis study (enrollment between 1995 and 2001), we randomized 87 proliferative LN patients to either cyclophosphamide pulses (750 mg/m(2), 13 pulses in 2 years) combined with oral prednisone (CY) or to azathioprine (2 mg/kg/day in 2 years) combined with intravenous pulses of methylprednisolone (3 x 3 pulses of 1000 mg) and oral prednisone (AZA). After a median follow-up of 5.7 years (interquartile range 4.1-7.2 years), doubling of serum creatinine was more frequent in the AZA group, although not statistically significant (relative risk (RR): 4.1, with 95% confidence interval (95% CI): 0.8-20.4). Relapses occurred more often in the AZA group (RR: 8.8, 95% CI: 1.5-31.8). Creatinine and proteinuria at last visit did not differ between the two treatment arms. Moreover, 88.4% of the patients in the AZA arm were still free of cyclophosphamide treatment. During the first 2 years, the frequency of remission was not different, but infections, especially herpes zoster virus infections (HZV) were more frequent in the AZA group. Parameters for ovarian function did not differ between the two groups. In conclusion, in this open-label randomized controlled trial, cyclophosphamide was superior to azathioprine with regard to renal relapses and HZV. At last follow-up, there were no differences in serum creatinine or proteinuria between the two groups. However, since our study lacked sufficient power, longer follow-up is needed to reveal putative differences.

Risk of colorectal and endometrial cancer for carriers of mutations of the hMLH1 and hMSH2 gene: correction for ascertainment.

BACKGROUND: Hereditary non-polyposis colorectal cancer (HNPCC) is caused by germline mutations of mismatch repair genes, usually in hMLH1 or hMSH2. All earlier studies on penetrance except one population based study were conducted in HNPCC families and did not correct for the way in which these families were ascertained. OBJECTIVE: To obtain estimates of the risk of colorectal cancer (CRC) and endometrial cancer (EC) for carriers of disease causing mutations of the hMSH2 and hMLH1 genes. METHODS: Families with known germline mutations of hMLH1 (n = 39) and hMSH2 (n = 45) were extracted from the Dutch HNPCC cancer registry. Ascertainment-corrected maximum likelihood estimation was carried out on a competing risks model for cancer of the colorectum and endometrium. RESULTS: Both loci were analysed jointly as there was no significant difference in risk (p = 0.08). At age 70, colorectal cancer risk for men was 26.7% (95% confidence interval, 12.6% to 51.0%) and for women, 22.4% (10.6% to 43.8%); the risk for endometrial cancer was 31.5% (11.1% to 70.3%). CONCLUSIONS: Current estimates of the CRC risk of mutations to the hMLH1 and hMSH2 locus should be replaced by considerably lower risks which account for the selection of the families.

Surgery groups differed in adverse outcome probabilities and can be used to adjust hospital comparisons.

BACKGROUND AND OBJECTIVE: In the Netherlands, all procedures in general surgery are categorized into 12 surgery groups by the Association of Surgeons of the Netherlands. The purpose of this study was to assess whether surgery groups differ in adverse outcome probabilities, to decide whether hospital comparisons on adverse outcomes should be adjusted for differences in surgery groups. METHODS: All surgical patients in one hospital discharged in 1997-1999 were included. Only the first operation during admission was included, with the assumption that successive operations were treatment of adverse outcomes. To avoid bias, only operations with procedures from the same surgery group were included. A total of 6,025 admissions were included and analyzed by a two-step multilevel analysis. RESULTS: Four surgery groups had fewer admissions with adverse outcomes than expected, and two groups had more. After adjustment for patient and operation characteristics, the remaining variance between surgery groups is still large. Similar results were found when differences in mortality were analyzed. CONCLUSION: Surgery group can therefore be used to adjust hospital comparisons for differences in surgical procedure mix.

Predictors of lung transplant survival in eurotransplant.

This study was undertaken to assess the influence of patient/donor and center factors on lung transplantation outcome. Outcomes of all consecutive first cadaveric lung transplants performed at 21 Eurotransplant centers in 1997-99 were analyzed. The risk-adjusted center effect on mortality was estimated. A Cox model was built including donor and recipient age and gender, primary disease, HLA mismatches, patient's residence, cold ischemic time, donor's cause of death, serum creatinine, type of lung transplant, respiratory support status, clinical condition and percentage predicted FEV1. The center effect was calculated (expressed as the standardized difference between the observed and expected survival rates), and empirical and full Bayes methods were applied to evaluate between-center differences. A total of 590 adults underwent lung transplantation. The primary disease (p=0.01), HLA-mismatches (p = 0.02), clinical condition(p < 0.0001) and the patient's respiratory support status (p = 0.05) were significantly associated with survival. After adjusting for case-mix, no between-center differences could be found. An in-depth empirical Bayes analysis showed the between-center variation to be zero. Similar results were obtained from the full Bayes analysis. Based on these data, there is no scientific basis to support a hypothesis of possible association between center volume and lung survival rates.

Mortality rates after heart transplantation: how to compare center-specific outcome data?

BACKGROUND: Studies of outcome in cardiac transplantation have focused primarily on identifying patient- and donor-related factors associated with patient mortality. Less consideration has been given to the impact of the transplant center. This study was undertaken to assess variability in heart transplantation outcome in Eurotransplant centers to provide a framework for auditing. METHODS AND RESULTS: In a 2-year period, 1,401 adult patients underwent heart transplantation in 45 centers. The 1-year patient survival rate was 76% (95% CI, 74%-78%) with a range of 0% to 100% at the center level. The risk-adjusted center effect on mortality was estimated by calculating a standardized difference between the observed number of deaths 1 year after transplantation and the expected number of deaths based on the case mix. By assessing within- and between-center variations with empirical Bayes (EB) methods, after adjustment for all registered prognostic factors, an improved estimate of the true center effect was obtained. Compared with the standard risk-adjusted center effect method, fewer outlying centers were identified with the EB method. CONCLUSION: EB methods, because they are known to incorporate more information from the data, enable a more precise and realistic portrayal of heart transplant centers' performances, compared with other risk-adjusted center effect methods. In the context of auditing procedures, EB methods should preferably be used for the identification of centers that deviate significantly from quality standards.

Cerebral microbleeds in CADASIL.

BACKGROUND: Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) is a hereditary arteriopathy leading to recurrent cerebral infarcts and dementia. Intracerebral hemorrhage (ICH) has been described sporadically in patients with CADASIL, suggesting that the affected arteries in CADASIL are not bleed-prone. However, the presence of cerebral microbleeds, which often remain undetected on conventional MRI, has not been determined in CADASIL. OBJECTIVE: To determine whether cerebral vessels in patients with CADASIL are prone to microbleeding. METHODS: T2*-weighted gradient echo MRI, which is highly sensitive for visualizing microbleeds, was performed in patients with CADASIL and their family members (n = 63). Known risk factors for ICH were determined for all individuals. On an exploratory basis, the presence of cerebral microbleeds was correlated with demographic variables, vascular risk factors, disease progression, ischemic MR lesions, and genotype. RESULTS: Cerebral microbleeds were present in 31% of symptomatic CADASIL mutation carriers, predominantly in the thalamus. Vascular risk factors such as hypertension did not account for the microbleeds in these patients. Factors associated with microbleeds were age (p = 0.008), Rankin disability score (p = 0.017), antiplatelet use (p = 0.025), number of lacunae on MRI (p = 0.009), and the Arg153Cys Notch3 mutation (p = 0.017). After correction for age, only the Arg153Cys mutation remained significantly associated with the presence of microbleeds. CONCLUSION: Patients with CADASIL have an age-related increased risk of intracerebral microbleeds. This implies that they may have an increased risk for ICH, which should be taken into account in CADASIL diagnosis and patient management.

Female donors influence transplant-related mortality and relapse incidence in male recipients of sibling blood and marrow transplants.

INTRODUCTION: The male H-Y antigen is recognised as a minor histocompatibility antigen with debatable relevance and has not been substantiated thus far in clinical solid organ transplantation. An increase in transplant-related mortality in male recipients of female stem cells has been recognised and attributed to graft vs host effect against H-Y; in contrast, decreased incidence of relapse, a graft vs leukaemia effect has not yet been described. METHODS: To detect potentially small but significant differences, we performed an analysis in a highly homogeneous group. We have analysed 782 patients, 438 males and 344 females, who were transplanted from HLA-identical female donors for CML in 1st chronic phase between 1989 and 1997. The risk of transplant related mortality (TRM) and relapse incidence (RI) were compared for male and female recipients over three successive time periods, zero to three months, three months to two years and two to five years post-transplant. Both groups were comparable for known risk factors prior to transplant. RESULTS: The cumulative risk of TRM at five years was significantly higher in males (42%, s.e.=3.6) than in females (27%, s.e.=2.8; P=0.02) with no overall effect on risk of relapse. Within the three specified time intervals post-transplantation, the number of events of TRM and RI in the two groups of recipients diverged over time. There was no difference within the first time interval (zero to three months) for both events. TRM was higher in male recipients as manifested from three months onwards over the whole observation period of five years (P=0.02). Risk of relapse was significantly reduced in male patients; this difference became manifest only beyond two years (P=0.01). CONCLUSION: These data confirm the importance of male gender for both, TRM and RI: male recipients of female blood or marrow grafts are at risk of enhanced procedure related mortality, eg GvHD, but benefit from reduced incidence of disease recurrence, hence a GvL effect. GvL in this setting needs more time than GvHD. These data give indirect but convincing evidence for a clinical relevance of H-Y as a minor histocompatibility antigen in humans.

On the use of Cox regression in the presence of an irregularly observed time-dependent covariate.

We consider the joint modelling of longitudinal and event time data. The longitudinal data are irregularly collected and the event times are subject to right censoring. Most methods described in the literature are quite complex and do not belong to the standard statistical tools. We propose a more practical approach using Cox regression with time-dependent covariates. Since the longitudinal data are observed irregularly, we have to account for differences in observation frequency between individual patients. Therefore, the time elapsed since last observation (TEL) is added to the model. TEL and its interaction with the time-dependent covariate show a strong effect on the hazard. The latter indicates that older recordings have less impact than recent recordings. Pros and cons of this methodology are discussed and a simulation study is performed to study the effect of TEL on the hazard. The fitted Cox model serves as a starting point for the prediction of future patient's events. Our method is applied to a study on chronic myeloid leukaemia (CML) with longitudinal white blood cell counts (WBC) as time-dependent covariate and patient's death as event.

Validation, calibration, revision and combination of prognostic survival models.

The problem of assessing the validity and value of prognostic survival models presented in the literature for a particular population for which some data has been collected is discussed. Methods are sketched to perform validation through 'calibration', that is by embedding the literature model in a larger calibration model. This general approach is exemplified for x-year survival probabilities, Cox regression and general non-proportional hazards models. Some comments are made on basic structural changes to the model, described as 'revision'. Finally, general methods are discussed to combine models from different sources. The methods are illustrated with a model for non-Hodgkin's lymphoma validated on a Dutch data set.

Asthma in adult patients presenting with symptoms of acute bronchitis in general practice.

OBJECTIVES: To investigate the association between asthma and acute bronchitis in adults and to ascertain criteria to help general practitioners (GPs) differentiate between acute bronchitis and asthma. DESIGN: Descriptive study. SETTING: Primary health care centre in The Netherlands. PATIENTS: 192 patients, aged 18-75 years, not known to have asthma or other pulmonary diseases, attending their GP with cough persisting for at least 2 weeks. METHODS: Patients were considered to have asthma or chronic obstructive pulmonary disease on the basis of a questionnaire on respiratory symptoms, spirometry and methacholine challenge testing. They were classified as having acute bronchitis if, according to international guidelines, coughing was more frequent than normal for at least 2 weeks, but no more than 4 weeks. Furthermore, either expectoration of purulent sputum for a maximum of 2 weeks and/or rhonchi as assessed by auscultation had to be present. By means of logistic regression symptoms, signs and peak expiratory flow measurements were sought to predict which patients with acute bronchitis actually had asthma. RESULTS: Of the 80 (41.7%) subjects with symptoms of acute bronchitis, 29 (36.9%) were patients with asthma. Within the acute bronchitis group, female sex and symptoms of current reported wheeze, reported episodes of attacks of dyspnoea over the last year and symptoms elicited by allergens were of help in identifying patients who actually had asthma. CONCLUSIONS: A considerable proportion of the patients presenting with acute bronchitis are actually patients with asthma. The absence or presence of a few symptoms and female sex may help to differentiate between these disorders.

Estimation of individual genetic effects from binary observations on relatives applied to a family history of respiratory illnesses and chronic lung disease of newborns.

This paper considers methods for estimating the relationship between a binary response Y and the genetic effects responsible for a second binary trait Z. The responses Y are observed only for target individuals, and the responses Z are observed only for the relatives of these targets. The analysis consists of two parts. The first part concerns the analysis of the family data Z and the second part estimates the relation between the genetic effects and Y. For the family data, a generalized linear mixed model with a logit link and Gaussian genetic (random) effects is used. Estimates of the variances of the genetic effects are obtained by using a pseudo-profile log-likelihood method. Estimation of the log likelihood involves averaging over n-dimensional normal distributions, which is done by importance sampling. The methods used in the second part are straightforward. The methods are applied to a data set containing chronic lung disease (CLDN) responses of newborns and asthma (AS), allergy (AL), chronic bronchitis (CB) and eczema (EC) responses observed for the relatives of these newborns. The clinical question is whether genetic effects of AS, AL, CB, and EC have an effect on the risk for CLDN. It can be concluded that for AS, AL, CB, and EC, the influence of genetic effects is significant. However, these genetic predispositions have no significant effect on CLDN.

Exploratory phase III study of paclitaxel and cisplatin versus paclitaxel and carboplatin in advanced ovarian cancer.

PURPOSE: To determine the side effects and feasibility of cisplatin and carboplatin each in combination with paclitaxel as front-line therapy in advanced epithelial ovarian cancer. PATIENTS AND METHODS: Patients were randomly allocated to receive paclitaxel 175 mg/m(2) intravenously as a 3-hour infusion followed by either cisplatin 75 mg/m(2) or carboplatin (area under the plasma concentration-time curve of 5), both on day 1. The schedule was repeated every 3 weeks for at least six cycles. Women allocated to paclitaxel-cisplatin were admitted to the hospital, whereas the carboplatin regimen was administered to outpatients. RESULTS: A total of 208 eligible patients were randomized. Both regimens could be delivered in an optimal dose and without significant delay. Paclitaxel-carboplatin produced significantly less nausea and vomiting (P: <.01) and less peripheral neurotoxicity (P: =.04) but more granulocytopenia and thrombocytopenia (P: <.01). The overall response rate in 132 patients with measurable disease was 64% (84 of 132 patients), and in patients with elevated CA 125 levels at start, it was 74% (132 of 178 patients). With a median follow-up time of 37 months, the median progression-free survival time of all patients was 16 months and the median overall survival time was 31 months. The small number of patients entered onto the study caused wide confidence intervals (CIs) around the hazards ratio for progression-free survival of paclitaxel-carboplatin compared with paclitaxel-cisplatin (hazards ratio, 1.07; 95% CI, 0.78 to 1.48) and did not allow conclusions about efficacy. CONCLUSION: Paclitaxel-carboplatin is a feasible regimen for outpatients with ovarian cancer and has a better toxicity profile than paclitaxel-cisplatin.

Permanent detrimental effect of nonimmunological factors on long-term renal graft survival: a parsimonious model of time-dependency.

PURPOSE: We attempted to model and test the pattern of effects of prognostic factors on renal graft survival during the posttransplantation time course. PATIENTS AND METHODS: Patients who received a cadaveric kidney-only transplant between January 1990 and December 1995 in Eurotransplant, who received cyclosporine as induction therapy, and who had a complete follow-up at the time of analysis were included in the study (n= 10614). An index summarizing all covariate information was calculated and used for modeling the time-dependent effects with relation to graft failure. RESULTS: The immunological factors (HLA mismatch and % panel-reactive antibody) were seen to have a slowly decreasing negative effect on renal graft survival. The cold ischemic trauma (>24 hr) exerted a permanent detrimental effect on the grafts. The use of organs obtained from old donors was associated with a constant higher risk of graft loss. CONCLUSIONS: An analysis of determinants of human allograft dysfunction should also study the interaction between the effects and time. Nonimmunological factors had a constant detrimental effect on graft failure, whereas the impact of the immunological factors--although remaining important for late graft loss--very slowly decreased. In the context of marginal transplants, clustering of unfavorable factors should be avoided to prevent late graft losses.

The (in)validity of sensitivity and specificity.

This paper is a legacy of the first author, and after her untimely death reconstructed by the second author as a tribute to Irene Guggenmoos's contribution to biostatistics. It discusses two different views on diagnostic testing: the classical view in which sensitivity and specificity of a diagnostic test are considered universal constants, and the more statistical point of view that focuses on predictive values. The differences between the two paradigms are outlined and practical examples are discussed to show that the familiar concepts of sensitivity and specificity must be handled with care and not used indiscriminately.