Multiple imputation in a large-scale complex survey: a practical guide.

The Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium is a multisite, multimode, multiwave study of the quality and patterns of care delivered to population-based cohorts of newly diagnosed patients with lung and colorectal cancer. As is typical in observational studies, missing data are a serious concern for CanCORS, following complicated patterns that impose severe challenges to the consortium investigators. Despite the popularity of multiple imputation of missing data, its acceptance and application still lag in large-scale studies with complicated data sets such as CanCORS. We use sequential regression multiple imputation, implemented in public-available software, to deal with non-response in the CanCORS surveys and construct a centralised completed database that can be easily used by investigators from multiple sites. Our work illustrates the feasibility of multiple imputation in a large-scale multiobjective survey, showing its capacity to handle complex missing data. We present the implementation process in detail as an example for practitioners and discuss some of the challenging issues which need further research.

A semiparametric estimate of treatment effects with censored data.

A semiparametric estimate of an average regression effect with right-censored failure time data has recently been proposed under the Cox-type model where the regression effect beta(t) is allowed to vary with time. In this article, we derive a simple algebraic relationship between this average regression effect and a measurement of group differences in k-sample transformation models when the random error belongs to the G(rho) family of Harrington and Fleming (1982, Biometrika 69, 553-566), the latter being equivalent to the conditional regression effect in a gamma frailty model. The models considered here are suitable for the attenuating hazard ratios that often arise in practice. The results reveal an interesting connection among the above three classes of models as alternatives to the proportional hazards assumption and add to our understanding of the behavior of the partial likelihood estimate under nonproportional hazards. The algebraic relationship provides a simple estimator under the transformation model. We develop a variance estimator based on the empirical influence function that is much easier to compute than the previously suggested resampling methods. When there is truncation in the right tail of the failure times, we propose a method of bias correction to improve the coverage properties of the confidence intervals. The estimate, its estimated variance, and the bias correction term can all be calculated with minor modifications to standard software for proportional hazards regression.

Omnibus tests of the martingale assumption in the analysis of recurrent failure time data.

The Andersen-Gill multiplicative intensity (MI) model is well-suited to the analysis of recurrent failure time data. The fundamental assumption of the MI model is that the process Mi(t) for subjects i = 1, ..., n, defined to be the difference between a subject's counting process and compensator, i.e., Ni(t) - Ai(t); t > 0, is a martingale with respect to some filtration. We propose omnibus procedures for testing this assumption. The methods are based on transformations of the estimated martingale residual process Mi(t) a function of consistent estimates of the log-intensity ratios and the baseline cumulative hazard. Under a correctly specified model, the expected value of Mi(t) is approximately equal to zero with approximately uncorrelated increments. These properties are exploited in the proposed testing procedures. We examine the effects of censoring and covariate effects on the operating characteristics of the proposed methods via simulation. The procedures are most sensitive to the omission of a time-varying continuous covariate. We illustrate use of the methods in an analysis of data from a clinical trial involving patients with chronic granulatomous disease.

Sample size calculations for the two-sample problem using the multiplicative intensity model.

In this paper we propose formulae for calculating the expected number of events or, alternatively, the required trial duration, for clinical trials involving two treatment groups in which patients may potentially experience multiple events and the data will be analysed using a multiplicative intensity (MI) model. We use a partial likelihood-based approach and examine in detail two MI models: one that includes a binary treatment variable as the only covariate and a three-state Markov process model in which a binary time-varying covariate is added to the previous model. For the simpler model, our formula coincides with those derived by Cook using full likelihood methods. We present applications of the derived formulae to chronic granulomatous disease and breast cancer data sets.

Chemotherapy compared with autologous or allogeneic bone marrow transplantation in the management of acute myeloid leukemia in first remission.

BACKGROUND: In young adults with acute myeloid leukemia, intensive chemotherapy during the initial remission improves the long-term outcome, but the role of bone marrow transplantation is uncertain. We compared high-dose cytarabine with autologous or allogeneic marrow transplantation during the first remission of acute myeloid leukemia. METHODS: Previously untreated adolescents and adults 16 to 55 years of age who had acute myeloid leukemia received standard induction chemotherapy. After complete remission had been achieved, idarubicin (two days) and cytarabine (five days) were administered. Patients with histocompatible siblings were offered allogeneic marrow transplantation, whereas the remaining patients were randomly assigned to receive a single course of high-dose cytarabine or transplantation of autologous marrow treated with perfosfamide (4-hydroperoxycyclophosphamide). Oral busulfan and intravenous cyclophosphamide were used as preparative regimens for both allogeneic and autologous marrow transplantation. The end points were survival from the time of complete remission and disease-free survival. RESULTS: In an intention-to-treat analysis, we found no significant differences in disease-free survival among patients receiving high-dose chemotherapy, those undergoing autologous bone marrow transplantation, and those undergoing allogeneic marrow transplantation. The median follow-up was four years. Survival after complete remission was somewhat better after chemotherapy than after autologous marrow transplantation (P=0.05). There was a marginal advantage in terms of overall survival with chemotherapy as compared with allogeneic marrow transplantation (P=0.04). CONCLUSIONS: A postinduction course of high-dose cytarabine can provide equivalent disease-free survival and somewhat better overall survival than autologous marrow transplantation in adults with acute myeloid leukemia.

Ionized and total magnesium concentrations in blood from dogs with naturally acquired parvoviral enteritis.

OBJECTIVE: To determine whether pretreatment total and ionized blood magnesium concentrations were associated with outcome for dogs with parvoviral enteritis and whether ionized magnesium concentration was related to total magnesium concentration or other laboratory values. DESIGN: Prospective cohort study. ANIMALS: 61 healthy dogs and 72 dogs with parvoviral enteritis. PROCEDURE: Total, ionized, and pH-normalized ionized magnesium concentrations, ionized and pH-normalized ionized calcium concentrations, pH, sodium and potassium concentrations, and Hct were measured prior to treatment. chi 2 Analyses were used to test for associations between outcome and age and between outcome and treatment with antiendotoxin antibody. Pearson's correlation coefficients were calculated to determine whether ionized magnesium concentration was linearly associated with other laboratory values. RESULTS: Total and ionized magnesium concentrations were not significantly different between healthy dogs and dogs with parvoviral enteritis or between dogs surviving and those not surviving parvoviral enteritis. The only laboratory value strongly correlated with ionized magnesium concentration was pH-normalized ionized magnesium concentration. Of the factors tested, none were significantly associated with outcome, except that dogs 16 weeks old or less treated with antiendotoxin antibody were significantly more likely to die than were dogs 16 weeks old or less that were not treated with antiendotoxin antibody. CLINICAL IMPLICATIONS: Total and ionized blood magnesium concentrations cannot be used to consistently predict outcome for dogs with parvoviral enteritis. Antiendotoxin antibody should be used with caution in dogs 16 weeks old or less.

Comparison of melphalan and prednisone with vincristine, carmustine, melphalan, cyclophosphamide, and prednisone in the treatment of multiple myeloma: results of Eastern Cooperative Oncology Group Study E2479.

BACKGROUND: The Eastern Cooperative Oncology Group (ECOG) performed a Phase III comparison of melphalan and prednisone (MP) with vincristine, carmustine (BCNU), melphalan, cyclophosphamide, and prednisone (VBMCP) in an attempt to determine which of these regimens should be the standard treatment for multiple myeloma. METHODS: Four hundred seventy-nine previously untreated patients with multiple myeloma from 23 ECOG institutions were enrolled. Treatment, assigned by randomization, consisted of either 4-week cycles of MP or 5-week cycles of VBCMP. After 1 year of induction therapy, patients received MP or VBMCP maintenance therapy at 6- and 8- week intervals, respectively, until relapse. Patients who experienced treatment failure with MP were eligible for crossover therapy with VBMCP. RESULTS: Objective responses were obtained for 51% of patients receiving MP, as compared with 72% of patients receiving VBMCP (P < 0.001). Response duration was also longer with VBMCP (median, 18 months with MP vs. 24 months with VBMCP; P = 0.007). Overall survival was not significantly different between MP and VBMCP (P = 0.30). The 5-year survival for VBMCP was 26%, as compared with 19% for MP. VBMCP was associated with more nausea, peripheral nerve toxicity, alopecia, and neutropenia, but the infection rate was equal to that observed with MP. Both regimens were generally well tolerated. The main exception was that elderly patients who were confined to bed had a higher risk of death with VBMCP. The two regimens produced a similar incidence of late secondary myelodysplastic syndrome and acute leukemia. Crossover VBMCP for patients failing with MP was only minimally effective, with an objective response rate of 20% and median survival of 11 months after crossover. CONCLUSIONS: VBMCP is more effective than MP in producing and sustaining remission of multiple myeloma. It is associated with a marginal survival advantage and an apparently greater chance of surviving 5 years for patients who can tolerate moderately intensive combination chemotherapy. Cancer 1997;79:1561-7. 1997 American Cancer Society.

Effectiveness of extreme discordant sib pairs to detect oligogenic disease loci.

A method for the genomic screening of quantitative traits using extreme discordant sib pairs (EDSPs) has recently been described by Risch and Zhang [1995; 1996]. For many models relevant to common, genetically complex diseases, EDSPs are the most powerful siblings for detecting linkage. Thus, if such siblings can be identified and collected, powerful studies with reasonable genotyping budgets can be conducted. Using a subset of the GAW10 data, we have simulated a genomic screen using EDSPs. From the 4,780 total families in the first 20 replicates of 239 families, there were 100, 104, 155, 107, and 180 EDSP families for Q1, Q2, Q3, Q4, and Q5, respectively. EDSP data were analyzed for each trait using a modified version of MAPMAKER/SIBS capable of handling extreme discordant sib pairs. Four regions, one for Q1, one for Q2, and two for Q4, were able to exceed a threshold for linkage corresponding to a 0.001 pointwise significance level. In three cases, maximum lod score (MLS) peaks occurred within 3.8 cM of a major gene. In the fourth case, the MLS peak occurred 28.4 cM from a major gene. Omission of parents and an alternative definition of EDSP were also investigated.

The inversion of the exponential Radon transform for quantitative brain SPECT.

The mathematical derivation for the inversion of the exponential Radon transform is presented and the implementation of the inversion is detailed. The inversion can be verified and implemented by the readers for practical applications, such as for quantitative reconstruction of brain SPECT (single-photon emission computed tomography).

Revascularization for femoropopliteal disease. A decision and cost-effectiveness analysis.

OBJECTIVE: To evaluate the relative benefits and cost-effectiveness of revascularization for femoropopliteal disease using percutaneous transluminal angioplasty or bypass surgery. DESIGN: Decision analysis using a multistate transition simulation model (Markov process) and cost-effectiveness analysis from the perspective of the health care system. SETTING: Based on mortality, morbidity, patency, and cost data from a literature review. PATIENTS: Hypothetical cohort of patients with chronic femoropopliteal disease who desire revascularization. Subgroup analysis for patients defined by age, sex, indication, lesion type, and graft type. INTERVENTIONS: Percutaneous transluminal angioplasty, bypass surgery, and a strategies combining the two treatments. MAIN OUTCOME MEASURES: Five-year patency results, quality-adjusted life expectancy, lifetime costs, and incremental cost-effectiveness ratios. RESULTS: For 65-year-old men with disabling claudication and a femoropopliteal stenosis or occlusion and for 65-year-old men with chronic critical ischemia and a femoropopliteal stenosis, initial angioplasty increased quality-adjusted life expectancy by 2 to 13 months and resulted in decreased lifetime expenditures compared with bypass surgery. For patients with chronic critical ischemia and a femoropopliteal occlusion, initial bypass surgery increased quality-adjusted life expectancy by 1 to 4 months and resulted in decreased lifetime expenditures compared with angioplasty. Sensitivity analysis demonstrated that angioplasty would always be the preferred initial treatment if the angioplasty 5-year patency rate exceeds 30%. CONCLUSION: Angioplasty is the preferred initial treatment in patients with disabling claudication and a femoropopliteal stenosis or occlusion and in those with chronic critical ischemia and a stenosis. Bypass surgery is the preferred initial treatment in patients with chronic critical ischemia and a femoropopliteal occlusion.

Fulminant human herpesvirus six encephalitis in a human immunodeficiency virus-infected infant.

Self-limited involvement of the central nervous system (CNS) is a relatively common complication of primary infection with human herpesvirus six (HHV-6) in normal children. We describe an HIV-infected infant who developed fulminant encephalitis as a complication of HHV-6 infection. Immunohistochemical staining of CNS tissue demonstrated productive infection of all CNS cell-types. Analysis of the infected brain tissue by the polymerase chain reaction (PCR) confirmed the presence of a dense HHV-6 infection in the tissue, and demonstrated that the virus present in the CNS tissue was predominantly the A variant of HHV-6. This is the first demonstration of invasive tissue disease caused by HHV-6 in an HIV-infected infant.

Defining, identifying, and studying high-risk families: developing cohorts for epidemiologic study.

Many issues arise in planning epidemiologic studies of individuals at high risk for developing hereditary cancers. The most important are (a) determination of the information that can best be studied in epidemiologic settings; (b) selection of proper study designs; (c) acknowledgment of the ethical, psychosocial, and legal issues that will arise in these studies; and (d) anticipation of the logistical issues involved in large, multicenter studies. The breakout session "Developing Cohorts for Epidemiologic Study: Defining and Identifying High-Risk Families" examined these issues, and the results of that session are summarized here. There was general consensus that little information exists regarding the prevalence of genetic mutations that predispose individuals to increased cancer risk, the risks conferred by specific mutations and by gene-environment interactions, and the efficacy of potential interventions. Adequately controlled observational and randomized studies provide the best mechanism for obtaining this information, despite the considerable ethical and strategic difficulties that will arise in the planning and conduct of such studies.

An analytical approach to quantitative reconstruction of non-uniform attenuated brain SPECT.

An analytical approach to quantitative brain SPECT (single-photon-emission computed tomography) with non-uniform attenuation is developed. The approach formulates accurately the projection-transform equation as a summation of primary- and scatter-photon contributions. The scatter contribution can be estimated using the multiple-energy-window samples and removed from the primary-energy-window data by subtraction. The approach models the primary contribution as a convolution of the attenuated source and the detector-response kernel at a constant depth from the detector with the central-ray approximation. The attenuated Radon transform of the source can be efficiently deconvolved using the depth-frequency relation. The approach inverts exactly the attenuated Radon transform by Fourier transforms and series expansions. The performance of the analytical approach was studied for both uniform- and non-uniform-attenuation cases, and compared to the conventional FBP (filtered-backprojection) method by computer simulations. A patient brain X-ray image was acquired by a CT (computed-tomography) scanner and converted to the object-specific attenuation map for 140 keV energy. The mathematical Hoffman brain phantom was used to simulate the emission source and was resized such that it was completely surrounded by the skull of the CT attenuation map. The detector-response kernel was obtained from measurements of a point source at several depths in air from a parallel-hole collimator of a SPECT camera. The projection data were simulated from the object-specific attenuating source including the depth-dependent detector response. Quantitative improvement (>5%) in reconstructing the data was demonstrated with the nonuniform attenuation compensation, as compared to the uniform attenuation correction and the conventional FBP reconstruction. The commuting time was less than 5 min on an HP/730 desktop computer for an image array of 1282*32 from 128 projections of 128*32 size.

Quantitative reconstruction for myocardial perfusion SPECT: an efficient approach by depth-dependent deconvolution and matrix rotation.

An efficient reconstruction method for myocardial perfusion single-photon emission computed tomography (SPECT) has been developed which compensates simultaneously for attenuation, scatter, and resolution variation. The scattered photons in the primary-energy-window measurements are approximately removed by subtracting the weighted scatter-energy-window samples. The resolution variation is corrected by deconvolving the subtracted data with the detector-response kernel in frequency space using the depth-dependent frequency relation. The attenuated photons are compensated by recursively tracing the attenuation factors through the object-specific attenuation map. An experimental chest phantom with defects inside myocardium was used to test the method. The attenuation map of the phantom was reconstructed from transmission scans using a flat external source and a high-resolution parallel-hole collimator of a single-detector system. The detector-response kernel was approximated from measurements of a point source in air at several depths from the collimator surface. The emission data were acquired by the same detector setting. A computer simulation using similar protocols as in the experiment was performed. Both the simulation and experiment showed significant improvement in quantification with the proposed method, as compared to the conventional filtered-backprojection technique. The quantitative gain by the additional deconvolution was demonstrated. The computation time was less than 20 min on a HP/730 desktop computer for reconstruction of a 1282 x 64 array from 128 projections of 128 x 64 samples.

Weighted least squares analysis of repeated categorical measurements with outcomes subject to nonresponse.

In this paper, we describe a two-step weighted least squares method for analyzing repeated categorical outcomes when some individuals are not observed at all times of follow-up. Other weighted least squares methods for analyzing repeated measures data with missing responses have previously been proposed by Koch, Imrey, and Reinfurt (1972, Biometrics 28, 663-692) and Woolson and Clarke (1984, Journal of the Royal Statistical Society, Series A 147, 87-99). These methods give consistent estimators if the responses are missing completely at random, as discussed in Rubin (1976, Biometrika 63, 581-592). We propose a two-step method that will give consistent results under the weaker condition of missing at random, and compare it with the other two methods.

Parameter estimation and tissue segmentation from multispectral MR images.

A statistical method is developed to classify tissue types and to segment the corresponding tissue regions from relaxation time T(1 ), T(2), and proton density P(D) weighted magnetic resonance images. The method assumes that the distribution of image intensities associated with each tissue type can be expressed as a multivariate likelihood function of three weighted signal intensity values (T(1), T(2), P(D)) at each location within that tissue regions. The method further assumes that the underlying tissue regions are piecewise contiguous and can be characterized by a Markov random field prior. In classifying the tissue types, the method models the likelihood of realizing the images as a finite multivariate-mixture function. The class parameters associated with the tissue types (i.e. the weighted intensity means, variances and correlation coefficients of the multivariate function, as well as the number of voxels within regions of the tissue types of are estimated by maximum likelihood. The estimation fits the class parameters to the image data via the expectation-maximization algorithm. The number of classes associated with the tissue types is determined by the information criterion of minimum description length. The method segments the tissue regions, given the estimated class parameters, by maximum a posteriori probability. The prior is constructed by the tissue-region membership of the first- and second-order neighborhood. The method is tested by a few sets of T(1), T(2), and P(D) weighted images of the brain acquired with a 1.5 Tesla whole body scanner. The number of classes and the associated class parameters are automatically estimated. The regions of different brain tissues are satisfactorily segmented.