Search PubMed⌕ Search

Biomedical subjects

J M Robins

Publications and source records attributed to J M Robins.

At least 19 recordsLinked to original sources
PubMed1992

Identifiability and exchangeability for direct and indirect effects.

We consider the problem of separating the direct effects of an exposure from effects relayed through an intermediate variable (indirect effects). We show that adjustment for the intermediate variable, which is the most common method of estimating direct effects, can be biased. We also show that even in a randomized crossover trial of exposure, direct and indirect effects cannot be separated without special assumptions; in other words, direct and indirect effects are not separately identifiable when only exposure is randomized. If the exposure and intermediate never interact to cause disease and if intermediate effects can be controlled, that is, blocked by a suitable intervention, then a trial randomizing both exposure and the intervention can separate direct from indirect effects. Nonetheless, the estimation must be carried out using the G-computation algorithm. Conventional adjustment methods remain biased. When exposure and the intermediate interact to cause disease, direct and indirect effects will not be separable even in a trial in which both the exposure and the intervention blocking intermediate effects are randomly assigned. Nonetheless, in such a trial, one can still estimate the fraction of exposure-induced disease that could be prevented by control of the intermediate. Even in the absence of an intervention blocking the intermediate effect, the fraction of exposure-induced disease that could be prevented by control of the intermediate can be estimated with the G-computation algorithm if data are obtained on additional confounding variables.

Algorithms↗
PubMed1992

G-estimation of the effect of prophylaxis therapy for Pneumocystis carinii pneumonia on the survival of AIDS patients.

AIDS Clinical Trial Group Randomized Trial 002 compared the effect of high-dose with low-dose 3-azido-3-deoxythymidine (AZT) on the survival of AIDS patients. Embedded within the trial was an essentially uncontrolled observational study of the effect of prophylaxis therapy for pneumocystis carinii pneumonia on survival. In this paper, we estimate the causal effect of prophylaxis therapy on survival by using the method of G-estimation to estimate the parameters of a structural nested failure time model (SNFTM). Our SNFTM relates a subject's observed time of death and observed prophylaxis history to the time the subject would have died if, possibly contrary to fact, prophylaxis therapy had been withheld. We find that, under our assumptions, the data are consistent with prophylaxis therapy increasing survival by 16% or decreasing survival by 18% at the alpha = 0.05 level. The analytic approach proposed in this paper will be necessary to control bias in any epidemiologic study in which there exists a time-dependent risk factor for death, such as pneumocystis carinii pneumonia history, that (A1) influences subsequent exposure to the agent under study, for example, prophylaxis therapy, and (A2) is itself influenced by past exposure to the study agent. Conditions A1 and A2 will be true whenever there exists a time-dependent risk factor that is simultaneously a confounder and an intermediate variable.

Acquired Immunodeficiency Syndrome↗
PubMed1992

Estimating exposure effects by modelling the expectation of exposure conditional on confounders.

In order to estimate the causal effects of one or more exposures or treatments on an outcome of interest, one has to account for the effect of "confounding factors" which both covary with the exposures or treatments and are independent predictors of the outcome. In this paper we present regression methods which, in contrast to standard methods, adjust for the confounding effect of multiple continuous or discrete covariates by modelling the conditional expectation of the exposures or treatments given the confounders. In the special case of a univariate dichotomous exposure or treatment, this conditional expectation is identical to what Rosenbaum and Rubin have called the propensity score. They have also proposed methods to estimate causal effects by modelling the propensity score. Our methods generalize those of Rosenbaum and Rubin in several ways. First, our approach straightforwardly allows for multivariate exposures or treatments, each of which may be continuous, ordinal, or discrete. Second, even in the case of a single dichotomous exposure, our approach does not require subclassification or matching on the propensity score so that the potential for "residual confounding," i.e., bias, due to incomplete matching is avoided. Third, our approach allows a rather general formalization of the idea that it is better to use the "estimated propensity score" than the true propensity score even when the true score is known. The additional power of our approach derives from the fact that we assume the causal effects of the exposures or treatments can be described by the parametric component of a semiparametric regression model. To illustrate our methods, we reanalyze the effect of current cigarette smoking on the level of forced expiratory volume in one second in a cohort of 2,713 adult white males. We compare the results with those obtained using standard methods.

Analysis of Variance↗
PubMed1991

Empirical-Bayes adjustments for multiple comparisons are sometimes useful.

Rothman (Epidemiology 1990;1:43-46) recommends against adjustments for multiple comparisons. Implicit in his recommendation, however, is an assumption that the sole objective of the data analysis is to report and scientifically interpret the data. We concur with his recommendation when this assumption is correct and one is willing to abandon frequentist interpretations of the summary statistics. Nevertheless, there are situations in which an additional or even primary goal of analysis is to reach a set of decisions based on the data. In such situations, Bayes and empirical-Bayes adjustments can provide a better basis for the decisions than conventional procedures.

Bayes Theorem↗
PubMed1990

Adjusting for early treatment termination in comparative clinical trials.

In clinical trials of long-term therapies, patients often terminate their treatments earlier than planned. When analysing time-to-failure data, one approach to account for early treatment termination censors failure at the time of termination of therapy. In general, however, this does not produce valid inferences about the distribution of time to failure that would have occurred had treatment not been terminated. In contrast, intent-to-treat analyses, which are based on time to failure regardless of whether and when treatment is terminated, always produce valid inferences about the unconditional distribution of time to failure. Early treatment termination does not distort the size (type I error rate) of intent-to-treat tests but can cause a loss in power. Modifications to ordinary logrank tests can be used to recover some of the lost power without affecting test size, and can be most useful when the proportion of at-risk patients still taking their treatment changes substantially during periods when failures are observed. Extensions of the modified test to include strata are straightforward, although important design questions require further research.

Data Interpretation, Statistical↗
PubMed1990

The behaviour of common measures of association used to assess a vaccination programme under complex disease transmission patterns--a computer simulation study of malaria vaccines.

Case-control studies have been evoked as important alternatives to randomized clinical trials in the evaluation of infectious disease interventions. Using computer simulations, we compare the behaviour of common measures of association derived from case-control studies in the context of a malaria vaccine programme administered under complex transmission conditions. Several simplifying assumptions of previous workers have been relaxed and the simulated conditions are endemic rather than epidemic. The common estimators of association used in case-control studies remain unbiased only in limited circumstances. The term dependent happenings, first defined by Ross in 1916, is resurrected. Since the number of people becoming infected is dependent on the number of people already infected, control programmes in infectious diseases produce direct as well as indirect effects. Three different study designs with different pairs of comparison populations are defined. The choice of comparison population can be used to differentiate direct from indirect effects. In order to clarify the direct effects of a vaccination programme the comparison groups must be subjected to identical transmission intensities. In contrast, the referent group must remain unaffected by consequences of the intervention to determine indirect effects.

Brazil↗
PubMed1989

Estimability and estimation of excess and etiologic fractions.

This paper describes conditions under which epidemiologic data can provide estimates of the excess fraction (proportionate increase in caseload due to an exposure) and the etiologic fraction (fraction of cases caused by exposure). The excess fraction can be estimated under essentially the same conditions often cited for general study validity. In contrast, estimation of the etiologic fraction will usually require very specific non-identifiable assumptions about exposure action and interactions, although one can derive simple lower and upper bounds for the fraction from survival comparisons. Since the etiologic fraction is equivalent to the probability of causation, our results have implications for injury compensation in lawsuits involving the probability of causation.

Epidemiologic Methods↗
PubMed1989

Work related decrement in pulmonary function in silicon carbide production workers.

The relation between pulmonary function, cigarette smoking, and exposure to mixed respirable dust containing silicon carbide (SiC), hydrocarbons, and small quantities of quartz, cristobalite, and graphite was evaluated in 156SiC production workers using linear regression models on the difference between measured and predicted FEV1 and FVC. Workers had an average of 16 (range 2-41) years of employment and 9.5 (range 0.6-39.7) mg-year/m3 cumulative respirable dust exposure; average dust exposure while employed was 0.63 (range 0.18-1.42) mg/m3. Occasional, low level (less than or equal to 1.5 ppm) sulphur dioxide (SO2) exposure also occurred. Significant decrements in FEV1 (8.2 ml; p less than 0.03) and FVC (9.4 ml; p less than 0.01) were related to each year of employment for the entire group. Never smokers lost 17.8 ml (p less than 0.02) of FEV1 and 17.0 (p less than 0.05) of FVC a year, whereas corresponding decrements of 9.1 ml (p = 0.12) in FEV1 and 14.4 ml (p less than 0.02) in FVC were found in current smokers. Similar losses in FEV1 and FVC were related to each mg-year/m3 of cumulative dust exposure for 138 workers with complete exposure information; these findings, however, were generally not significant owing to the smaller cohort and greater variability in this exposure measure. Never smokers had large decrements in FEV1 (40.7 ml; p less than 0.02) and FVC (32.9 ml; p = 0.08) per mg-year/m3 of cumulative dust exposure and non-significant decrements were found in current smokers (FEV1: -7.1 ml; FVC: -11.7 ml). A non-significant decrement in lung function was also related to average dust exposure while employed. No changes were associated with SO(2) exposure or and SO(2) dust interaction. These findings suggest that employment in SiC production is associated with an excessive decrement in pulmonary function and that current permissible exposure limits for dusts occurring in this industry may not adequately protect workers from developing chronic pulmonary disease.

Adult↗
PubMed1989

Respiratory symptoms associated with low level sulphur dioxide exposure in silicon carbide production workers.

Relations between pulmonary symptoms and exposure to respirable dust and sulphur dioxide (SO2) were evaluated for 145 silicon carbide (SiC) production workers with an average of 13.9 (range 3-41) years of experience in this industry. Eight hour time weighted average exposures to SO2 were 1.5 ppm or less with momentary peaks up to 4 ppm. Cumulative SO2 exposure averaged 1.94 (range 0.02-19.5) ppm-years. Low level respirable dust exposures also occurred (0.63 +/- 0.26 mg/m3). After adjusting for age and current smoking status in multiple logistic regression models, highly significant, positive, dose dependent relations were found between cumulative and average exposure to SO2, and symptoms of usual and chronic phlegm, usual and chronic wheeze, and mild exertional dyspnoea. Mild and moderate dyspnoea were also associated with most recent exposure to SO2. Cough was not associated with SO2. No pulmonary symptoms were associated with exposure to respirable dust nor were any symptoms attributable to an interaction between dust and SO2. Cigarette smoking was strongly associated with cough, phlegm, and wheezing, but not dyspnoea. A greater than additive (synergistic) effect between smoking and exposure to SO2 was present for most symptoms. These findings suggest that long term, variable exposure to SO2 at 1.5 ppm or less was associated with significantly raised rates of phlegm, wheezing, and mild dyspnoea in SiC production workers, and that current threshold limits for SO2 may not adequately protect workers in this industry.

Adult↗
PubMed1989

Designs for synthetic case-control studies in open cohorts.

Several designs are proposed for case-control studies within cohorts when the cohort is open to late entry. These and previously proposed designs are examined with respect to consistency and efficiency of relative risk parameter estimation, and a small simulation study is reported. If study costs increase in proportion to the total number of "at-risk" controls, the most efficient design, Design C, is as follows. For a case failing at time t, controls are selected at random (and without regard to "at-risk" status) from among cohort members who are (i) known not to have failed prior to t and (ii) have not been previously selected as controls. At each t, control sampling proceeds until a prespecified number of controls who are "at risk" at t have been obtained. The efficiency advantage of Design C over that of the standard case-control design proposed by Thomas (in Appendix to Liddell, McDonald, and Thomas, 1977, Journal of the Royal Statistical Society, Series B 140, 469-490) will often be small. If, on the other hand, the costs increase in proportion to the number of distinct "at-risk" controls, Design C is no longer the most efficient design. In this case, several alternative designs are proposed.

Biometry↗
PubMed1988

Confidence intervals for causal parameters.

Consider an unbiased follow-up study designed to investigate the causal effect of a dichotomous exposure on a dichotomous disease outcome. Under a deterministic outcome model, a standard '95 per cent binomial confidence interval' may fail to cover the causal parameter of interest at the nominal rate when we take the causal parameter to be a parameter associated with the observed study population (regardless of whether the observed study population was sampled from a larger superpopulation). I propose new interval estimators that, in this setting, improve upon the performance of the standard 'binomial confidence interval.'

Biometry↗
PubMed1988

Conceptual problems in the definition and interpretation of attributable fractions.

We have argued that the concept of attributable fraction requires separation into the concepts of excess fraction, etiologic fraction, and incidence-density fraction. These quantities do not necessarily approximate one another, and the etiologic fraction is not generally estimable without strong biologic assumptions. For these reasons, care is needed in deciding which (if any) of the concepts is appropriate for a particular application. It appears that the excess fraction (like incidence proportion) will be most relevant in situations that require only consideration of whether disease occurs by a particular time. In situations that require consideration of when disease occurs, direct measures of effect on incidence time may be as relevant as or more relevant than any attributable fraction. To avoid technical complications, we have not discussed additional problems of causal attribution that can arise when exposure has multiple levels or is sustained over time, and the estimation problems that can arise when considering case-control studies, competing risks, or differential censoring. For more detailed discussions of such problems and proposed solutions, see references 11-20.

Epidemiology↗
PubMed1987

Analysis of proportionate mortality data using logistic regression models.

When only proportionate mortality data are available to an investigator studying the effect of an exposure on a particular cause of death, controls must be selected from among persons dying of other causes believed to be uninfluenced by the exposure under study. When qualitative or quantitative estimates of exposure history can be obtained for the deceased individuals, it is shown that one can use logistic regression models for the mortality odds to efficiently estimate the effect of exposure while controlling for relevant confounding factors by incorporating a priori information on baseline mortality rates available from US life tables. The proposed method is used to reanalyze data from a cohort of arsenic-exposed workers in a Montana copper smelter.

Adult↗
PubMed1987

Use of MR imaging in an outpatient MR center.

Indications for MR examinations and patient characteristics are evaluated for 4561 MR examinations performed at a freestanding outpatient MR imaging center between May 1984 and June 1986. Hospitalized patients accounted for less than 3% of the case load. Examinations of the head and spine accounted for 60% and 31% of the work load, respectively. Patients 65 years or older made up 15% of the case load during 1984 and 1985 and 21% in 1986. Referrals from neurologists, internists, and neurosurgeons accounted for 56%, 11% and 9% of patients, respectively. The percentage of patients who had CT, myelography, and other imaging procedures performed before referral for MR imaging declined significantly between 1984 and 1986. Indications for examination were mostly neoplastic diseases; degenerative diseases of the CNS, including multiple sclerosis; other disorders of the CNS; and disk diseases. Approximately 40% of all examinations were interpreted as normal. The number of patients referred for degenerative intervertebral disk disorders increased substantially between 1984 and 1985. This study documents the increasing acceptance of MR imaging as an important primary imaging technique for a variety of conditions, particularly those of the brain and spine.

Aged↗
PubMed1986

Estimation of ventilatory capacity in subjects with unacceptable lung function tests.

Based on pulmonary function data collected annually for six years on 540 Vermont granite workers, FEV1 in survey 1 was estimated by extrapolating back from subsequent measurements. The extrapolation method was found to fit the observed data of subjects with reproducible initial values very well (R2 = 0.87). Extrapolated FEV1s for workers unable to perform an adequate pulmonary function test according to the standards of the American Thoracic Society were compared to extrapolated values in the rest of the cohort. After adjusting for confounding, subjects with test failure in survey 1 had a lower extrapolated FEV1 than the rest of the cohort (p = 0.07). The mean extrapolated FEV1 of the 71 workers with an initial test failure was only 95% of a predicted value derived from the group with reproducible data, and the per cent predicted decreased from 98% to 71% as the number of test failures in the follow-up surveys increased (p = 0.0004). The American Thoracic Society and the Epidemiology Standardization Project currently recommend that test failures be excluded from the analysis of epidemiological data. Our findings suggest that alternative strategies for handling non-reproducible lung function may need to be explored in order to avoid selection bias.

Adult↗
PubMed1986

Identifiability, exchangeability, and epidemiological confounding.

Non-identifiability of parameters is a well-recognized problem in classical statistics, and Bayesian statisticians have long recognized the importance of exchangeability assumptions in making statistical inferences. A seemingly unrelated problem in epidemiology is that of confounding: bias in estimation of the effects of an exposure on disease risk, due to inherent differences in risk between exposed and unexposed individuals. Using a simple deterministic model for exposure effects, a logical connection is drawn between the concepts of identifiability, exchangeability, and confounding. This connection allows one to view the problem of confounding as arising from problems of identifiability, and reveals the exchangeability assumptions that are implicit in confounder control methods. It also provides further justification for confounder definitions based on comparability of exposure groups, as opposed to collapsibility-based definitions.

Bayes Theorem↗