01.06.2013 Views

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

680 <strong>Statistical</strong> methods <strong>in</strong> epidemiology<br />

1 543 1 96 0 0901 ˆ 1 366 and 1 720:<br />

The correspond<strong>in</strong>g limits for c are the exponentials, 3 92 and 5 58. The po<strong>in</strong>t<br />

estimate and the confidence limits are very similar <strong>in</strong>deed to those derived us<strong>in</strong>g Woolf 's<br />

method. Details of the calculation are not given here but (19.24) gives SE (ln RMH)<br />

ˆ 0 0977 to give 95% confidence limits of 3 86 and 5 67, similar to both sets of limits<br />

derived earlier.<br />

Another frequently occurr<strong>in</strong>g situation is when the factor, <strong>in</strong>stead of hav<strong>in</strong>g<br />

just two levels, presence or absence, has three or more levels that have a def<strong>in</strong>ite<br />

order. For example, <strong>in</strong>stead of classify<strong>in</strong>g people as non-smokers or smokers, the<br />

smokers could be divided further accord<strong>in</strong>g to the amount smoked. Such a<br />

situation was discussed <strong>in</strong> §15.7. Mantel (1963) extended the Mantel±Haenszel<br />

procedure to provide a test for trend and to comb<strong>in</strong>e a number of subsets of the<br />

data (15.21). An example of the method is given <strong>in</strong> Example 15.8.<br />

The methods discussed above are powerful <strong>in</strong> the commonly occurr<strong>in</strong>g<br />

situation where it is required to analyse the association between a disease and<br />

a s<strong>in</strong>gle exposure factor, mak<strong>in</strong>g allowance for one other factor, or at most a few,<br />

and where the effect of the other factors may be adequately represented by<br />

subdivision of the data <strong>in</strong>to strata. The methods are <strong>in</strong>convenient where several<br />

exposure factors are of <strong>in</strong>terest and for each factor allowance must be made for<br />

the others, as was the case <strong>in</strong> Example 14.1. Also, where several factors are to be<br />

adjusted for, stratification would <strong>in</strong>volve a large number of strata; <strong>in</strong> the limit<br />

each stratum would conta<strong>in</strong> only one subject, so that comparisons with<strong>in</strong> strata<br />

would be impossible. In these cases more general methods must be used and<br />

logistic regression (§14.2) is appropriate. This is because of the comb<strong>in</strong>ation of<br />

two features. First, as observed above, the odds ratio can be estimated from a<br />

case±control study and, secondly, the logarithm of the odds ratio is a difference<br />

of logits. For a case-control study all the coefficients of (14.6) may be estimated<br />

except for the constant term b 0, which is distorted by the <strong>in</strong>vestigator's choice of<br />

the size of the case and control groups. The estimate of a regression coefficient<br />

measures the logarithm of the odds ratio for a change of unity <strong>in</strong> the correspond<strong>in</strong>g<br />

variable, and so the exponentials of the regression coefficients give estimated<br />

odds ratios or approximate relative risks.<br />

It was observed above that, for a case±control study where a matched control<br />

is chosen for each case, the Mantel±Haenszel method may be applied, and leads<br />

to McNemar's test. When the number of cases available is limited, the precision<br />

of a study may be <strong>in</strong>creased, to some extent, by choos<strong>in</strong>g more than one matched<br />

control per case. The Mantel±Haenszel method can be applied, treat<strong>in</strong>g each set<br />

of a case and its controls as a subset. The analysis may be simplified by group<strong>in</strong>g<br />

together sets provid<strong>in</strong>g identical <strong>in</strong>formation; for example, with three controls<br />

per case those sets where the case and just two of the three controls are positive<br />

for the exposure factor all provide identical <strong>in</strong>formation. This is an extension of

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!