Page 21 - PATIENT REGISTRY DATA FOR RESEARCH: A Basic Practical Guide

of missing data in the registry, which the researcher will have to decide whether to wait for a

future data installation, or to proactively fill in these missing data, or to impute a median

value for all the missing data, or to simply declare them as missing data if the minimum

sample size deployed by the study is already adequate.

2.3 Estimate the Minimum Sample Size Required for a Registry Study

Sample size calculation is an important consideration because all researchers have to

enrol a minimum number of patients or to retain enough patients in the study in order to

attain the desired power of a registry study. Generally, it will not be necessary to perform a

sample size calculation if the registry is created using a census approach. However, many

problems can potentially arise when missing data are found in the registry. If this happens,

then it will be necessary to estimate the minimum sample size required for the registry-based

study even if the registry has been created by using the census or population-based approach.

In a similar vein, it is also always necessary for the researcher to give due consideration to the

sample size calculation of a registry-based study if the registry has been created by collecting

a sample of the total population data.

A notably significant observation that universally applies to all clinical research is that

sample size required for a study will always depend on the study objective. For example,

when calculating the sample size required for an objective of estimating the prevalence rate

or the population mean, the values of alpha, power of the study and margin of error are three

important determinants which must be set beforehand. Based on the findings provided by

Cochran's (1977), the sample size required to estimate the prevalence rate within a population

will be 384 if the researcher has set the values of alpha, power and margin error at 0.05, 80%

and 5.0% respectively (Krejcie & Morgan, 1970). Since the values of alpha and power are

always set at 0.05 and 80.0% respectively and the value of margin of error of 5% is generally