Wilfred L. Rosenbaum

Publication Decisions Revisited: The Effect of the Outcome of Statistical Tests on the Decision to Publish and Vice Versa

Abstract This article presents evidence that published results of scientific investigations are not a representative sample of results of all scientific studies. Research studies from 11 major journals demonstrate the existence of biases that favor studies that observe effects that, on statistical evaluation, have a low probability of erroneously rejecting the so-called null hypothesis (H 0). This practice makes the probability of erroneously rejecting H 0 different for the reader than for the investigator. It introduces two biases in the interpretation of the scientific literature: one due to multiple repetition of studies with false hypothesis, and one due to failure to publish smaller and less significant outcomes of tests of a true hypotheses. These practices distort the results of literature surveys and of meta-analyses. These results also indicate that practice leading to publication bias have not changed over a period of 30 years.

Analysis of the relationship between smokeless tobacco and cancer based on data from the National Mortality Followback Survey

This study investigates the potential link between the use of smokeless tobacco and oral cancer and cancer of digestive organs. The combined data of the National Mortality Followback Survey (NMFS), a probability sample of the U.S. deaths, and the coincident National Health Interview Survey (NHIS), a probability sample of the living, non-institutionalized U.S. population, are used to compute risk estimates for cancer, oral cancer, and cancer of the digestive organs associated with use of smokeless tobacco based on a cross sectional study design, simultaneously controlled for potential confounding from active smoking, alcohol consumption, and occupational exposure. Use of smokeless tobacco (either as snuff or chewing tobacco) does not increase the risk of oral cancer or cancer of the digestive organs. Alcohol emerges as a major risk factor for oral cancer with a strong dose-response relationship between the amount of drinking and risk. The same is true to a lesser extent for cancer of the digestive organs. Smoking is associated with increased risk of oral cancer but not of cancer of the digestive organs. Blue collar, technical, and service workers have significantly increased risk of cancer of the digestive organs relative to professional, managerial, and clerical workers, but not of oral cancer. Differences between findings based on the NMFS/NHIS and those obtained from other data very likely are due to inadequate control for confounding. Other reasons for differences between the NMFS/NHIS data and other studies are discussed.

Bias in the attribution of lung cancer as cause of death and its possible consequences for calculating smoking-related risks.

Most published calculations of mortality risk, especially those for lung cancer associated with smoking, are based almost exclusively on the underlying cause as recorded on death certificates. Such risk calculations implicitly assume that the conditional probability of recording lung cancer as the underlying cause of death, given that it really is the underlying cause, is the same for all exposure groups. If these probabilities are not equal for all exposure groups, we call the resulting bias a cause of death attribution bias. We analyzed the 1986 National Mortality Followback Survey, a sample of 18,733 U.S. death certificates, and the 1954–1962 Dorn study, a follow-up study of approximately 250,000 holders of U.S. Veterans Life Insurance. Both data sets include information on the smoking habits of decedents and on the underlying and contributing causes of their deaths. We found that lung cancer as an underlying cause is recorded with a much smaller relative frequency if the decedent is known to be a never-smoker and with a much larger relative frequency when the decedent is known to be a smoker. On the other hand, lung cancer as a contributing cause is recorded with a much larger frequency if the decedent is known to be a never-smoker and with a much smaller frequency when the decedent is known to be a smoker. The reverse is true for cancers other than of the lung. There is no similar pattern related to smoking for other causes of death (specifically for myocardial infarction, other chronic ischemic heart disease, diabetes, or cerebrovascular disease). This pattern provides evidence of a possible bias because knowledge of a decedents smoking status appears to influence the designation of lung cancer or some other cancer as the underlying cause or a contributing cause of death. This bias is especially strong when the choice of possible underlying causes of death is limited to one of a number of cancers. Insofar as calculations of lung cancer risk utilize exclusively recorded underlying causes, the observed attribution bias must result in an overestimate of the lung cancer mortality rate for smokers. (Epidemiology 1992;3:11–16)

An alternative explanation for the apparent elevated relative mortality and morbidity risks associated with exposure to environmental tobacco smoke.

Insofar as industrial and other blue collar workers are more likely to bring home toxic materials on their person, and also are more likely to smoke than those in other occupations, members of a household are more likely to be subject to paraoccupational exposure and belong to lower socioeconomic strata if the household contains a smoker than if the household does not contain a smoker. Thus observed differences in risk of mortality or morbidity ascribed to ETS on the basis of a comparison of households with and without smokers may be partly or entirely due to differences in paraoccupational exposure and socioeconomic strata. Similarly, differences in mortality and morbidity ascribed to paraoccupational exposure may be partly or entirely due to differences in ETS exposure that are also related to social class and to types of occupation. Unfortunately, there are no data now in existence that could help determine separately the effects of these major confounded variables. There exists, then, a situation in which two explanations are advanced for respiratory diseases among members of a household, each based on similar study populations but focused on different major risk variables: ETS on the one hand, socioeconomic status and paraoccupational exposure on the other. Properly focused investigations need to be initiated.

Difficulties of T1 brain MRI segmentation techniques

This paper looks at the difficulties that can confound published T1-weighted Magnetic Resonance Imaging (MRI) brain segmentation methods, and compares their strengths and weaknesses. Using data from the Internet Brain Segmentation Repository (IBSR) as a gold standard, we ran three different segmentation methods with and without correcting for intensity inhomogeneity. We then calculated the similarity index between the brain masks produced by the segmentation methods and the mask provided by the IBSR. The intensity histograms under the segmented masks were also analyzed to see if a Bi-Gaussian model could be fit onto T1 brain data. Contrary to our initial beliefs, our study found that intensity based T1-weighted segmentation methods were comparable or even superior to, methods utilizing spatial information. All methods appear to have parameters that need adjustment depending on the data set used. Furthermore, it seems that the methods we tested for intensity inhomogeneity did not improve the segmentations due to the nature of the IBSR data set.

Smoking and hospital utilization

Remaining lifetime hospital days (RLHD) are used as estimates of possible differences in medical care costs between ever smokers and never smokers. Hospital usage by age in days per person per year comes from the 1970 U.S. National Health Interview Survey (NHIS) of some 40,000 households. Life table analysis for relative longevity of ever smokers and never smokers is based on mortality ratios presented in the American Cancer Societys Million Person Study. Results are similar to those obtained by Leu and Schaub for Swiss medical costs. There is no consistent increase in RLHD for ever smokers. In fact, male ever smokers older than 44 years and female ever smokers older than 38 years can expect fewer RLHDs than never smokers.

Classification and performance of denoising algorithms for low signal to noise ratio magnetic resonance images

The generation of magnitude magnetic resonance images comprises a sequence of data encodings or transformations, from detection of an analog electrical signal to a digital phase/frequency k-space to a complex image space via an inverse Fourier transform and finally to a magnitude image space via a magnitude transformation and rescaling. Noise present in the original signal is transformed at each step of this sequence. Denoising MR images from low field strength scanners is important because such images exhibit low signal to noise ratio. Algorithms that perform denoising of magnetic resonance images may be usefully classified according to the data domain on which they operate (i.e. at which step of the sequence of transformations they are applied) and the underlying statistical distribution of the noise they assume. This latter dimension is important because the noise distribution for low SNR images may be decidedly non-Gaussian. Examples of denoising algorithms include 2D wavelet thresholding (operates on the wavelet transform of the magnitude image; assumes Gaussian noise), Nowaks 2D wavelet filter (operates on the squared wavelet transform of the magnitude image; assumes Rician noise), Alexander et. al.s complex 2D filters (operates on the wavelet transform of the complex image space; assumes Gaussian noise), wavelet packet denoising (wavelet packet transformation of magnitude image; assumes Rician noise) and anisotropic diffusion filtering (operates directly on magnitude image; no assumptions on noise distribution). Effective denoising of MR images must take into account both the availability of the underlying data, and the distribution of the noise to be removed. We classify a number of recently published denoising algorithms and compare their performance on images from a 0.35T permanent magnet MR scanner.© (2000) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.