PubMed Health⌕ Search

Biomedical subjects

M D Edwardes

Publications and source records attributed to M D Edwardes.

8 recordsLinked to original sources
PubMed2000

The generalization of the odds ratio, risk ratio and risk difference to r x k tables.

Familiar measures of association for 2 x 2 tables are the odds ratio, the risk ratio and the risk difference. Analagous measures of outcome-exposure association are desirable when there are several degrees of severity of both exposure and disease outcome. One such measure (alpha), which we label the general odds ratio (OR(G)), was proposed by Agresti. Convenient methods are given for calculation of both standard error and 95 per cent confidence intervals for OR(G). Other approaches to generalizing the odds ratio entail fitting statistical models which might not fit the data, and cannot handle some zero frequencies. We propose a generalization of the risk ratio (RR(G)) following the statistical approaches of Agresti, Goodman and Kruskal. A method of calculating the standard error and 95 per cent confidence interval for RR(G) is provided. A known statistic, Somers' d, fulfils the characteristics necessary for a generalized risk difference (RD(G)). These measures have straightforward interpretations, are easily computed, are at least as precise as other methods and do not require fitting statistical models to the data. We also examine the pooling of such measures as in, for example, meta-analysis.

Anti-Ulcer Agents↗
PubMed2000

GEE analysis of negatively correlated binary responses: a caution.

The method of generalized estimating equations has become almost standard for analysing longitudinal and other correlated response data. However, we have found that if binary responses have less than binomial variation over clusters, and are modelled using exchangeable correlations, prevailing software implementations may give unreliable results. Bounding the negative correlation away from its theoretical minimum may not always be a satisfactory solution. In such instances, using the independence working correlation structure and robust SEs is a more trustworthy alternative.

Binomial Distribution↗
PubMed1999

A physician-centred intervention to shorten hospital stay: a pilot study.

BACKGROUND: Studies of length of stay (LOS) in hospital usually focus on physician-independent factors. In this study, the authors identified physician-dependent factors and tested an intervention aimed at them to determine its effect on LOS. METHODS: A prospective comparison of LOS on 2 general medical wards in a tertiary care teaching hospital before and after the intervention. The pre-intervention (control) period and the intervention period were each 4 weeks. The intervention consisted of a checklist for planning management and discharge. RESULTS: Overall, the mean LOS was shorter during the intervention period than during the control period, but the difference was not statistically significant (12.0 and 14.4 days respectively, p = 0.13). The difference was significant on ward A (11.0 v. 14.7 days respectively, p = 0.02) but not on ward B (13.0 and 14.0 days respectively, p = 0.90). INTERPRETATION: An intervention at the level of the admitting physician may help to shorten LOS on a general medical ward.

Case Management↗
PubMed1998

External comparisons from nested case-control designs.

The nested case-control design, used to sample within cohorts, is usually employed for internal comparisons. We propose to use this design for external comparisons. We present two probability-weighted estimators of the expected number of cases under a given exposure, based on external rates, for two versions of the nested case-control design. These estimators are used, along with their variance estimators, to form confidence intervals for standardized mortality ratios. The estimators are practically unbiased, whereas the naive estimator that treats the nested case-control sample as a random sample of the cohort is clearly biased. An estimator from the alternative Cox model-based approach is found to be substantially biased when applied in this context. Comparing the proposed estimators for nested case-control designs to a corresponding estimator for the case-cohort design, we found that the correlation between follow-up time and exposure time (that is, the amount of time under the exposure effect) has an impact on which type of design is more efficient for external comparisons. A small correlation favors the case-cohort design and a large correlation the nested case-control design. We examine empirical properties of these estimators through computer simulations, using a cohort study of the incidence of second cancer in 2,189 patients with Hodgkin's disease.

Bias↗
PubMed1997

Adjusted odds ratios for case-control studies with missing confounder data in controls.

Nonexperimental studies using computerized databases often give rise to missing or partially available information on confounders. A frequent situation occurs when data on exposure are available for all subjects of a case-control study, but data on confounders are available only for the cases but not for the controls. In that situation, the fact of confounding can be verified by assessing the association between exposure and a confounder in the cases, but the data are insufficient to produce an adjusted estimate of the relative risk if confounding is found to be present. We propose simple conditions under which an adjusted estimate of the relative risk can be obtained when data on a confounder are available only for the cases, and we derive formulae for the estimator and its confidence limits. The method requires an external estimate of the confounder prevalence or, additionally, of the confounder-exposure odds ratio. We illustrate the technique with data from a nested case-control study of the risk of acute cardiac death associated with the use of bronchodilator drugs within a cohort of 12,301 asthmatics, with smoking as the confounder of interest.

Asthma↗
PubMed1995

A confidence interval for Pr(X < Y) - Pr(X > Y) estimated from simple cluster samples.

Distribution-free confidence intervals based on Somers' d (a simple function of the Mann-Whitney U statistic) are developed for Pr(X < Y) - Pr(X > Y), where X and Y are the ranks of any two observations from two independent populations. The approach accommodates a complex sampling design, and explicit formulas are given for simple cluster sampling. The method also accommodates simple progressive left- and right-censoring. The accuracy of the interval is shown to improve when the tanh-1 transform is used. A bootstrap solution was tried and did not perform as well as the proposed solution.

Analysis of Variance↗