3.

6 Sample size
For assessing prevalence: The sample size required for determination of the prevalence was
based on the formula for single sample cross-sectional studies in which the primary outcome
is expressed as a proportion [32]. The formula is as follows:

n = (Z α + Z β)2 * P(100-P)
δ2
Where
Zα = The standard normal deviate value corresponding 95% confidence (1.96)
Zβ ═ The standard normal deviate value corresponding 80% power (0.84)
P = The estimated prevalence of primary outcome under study. A P of 3% was
selected based on a study that showed prevalence of pre-diabetes in the
general population in Southern Uganda to be 3% [33]
δ = The level of sampling error acceptable for this study (2%).
D = Design effect (A design effect of 2.5 was factored into the equation because
of the fact that analysis was to be conducted at multiple levels of strata.
Analysis at multiple levels of strata would increase the sampling errors during
the stratified analysis. 2.5 is the estimated standard error of the sample
estimate [34]. The two levels of strata were: Estimation of general prevalence
of dysglycaemia the first level and estimation of the factors associated with
dysglycaemia the second level of strata.

Substituting into this formula gave a sample size of 634 participants. Adjusting for 10% non-
response, gave a total of 688 participants.

For assessing associated factors: The sample size for the determination of factors associated
with dysglycaemia was based on a formula for comparative studies in which the outcome is a
proportion [35]. The structure of this formula is given as follows:

𝑍1−𝛼/2 . √2𝑃̅ (1 − 𝑃̅) − 𝑍𝛽 . √𝑃1 (1 − 𝑃1 ) + 𝑃2 (1 − 𝑃2 )

𝑛=
(𝑃2 − 𝑃1 )2
Where:
P1 The expected proportion of the outcome in the non-exposed group
P2 The expected proportion of the outcome in the exposed group
𝑃̅ The simple average of the two proportions
Zα The standard normal deviate at a given level of confidence (for a 2 sided test)
Z1-β The standard normal deviate at a given level of power (one sided)

This is the formula that underlies the sample size computations used by Epi-Info in
computing sample sizes for comparative studies. Using Statcalc for Epi Info Version 3.2, the
following parameters were assumed:
• A level of significance (α) of 0.05 (i.e. 95% confidence, hence Zα=1.96),
• Power to detect significant differences between the two comparison groups (1-β) of 80%
(hence Z1-β= 0.86),
• Estimated prevalence of dysglycaemia in overweight adolescents (14.7%) [24]
• Estimated minimum effect size (difference in prevalence of dysglycaemia in exposed and
in unexposed that the study should be able to detect) (12%)
The computed sample size was therefore 101 adolescents. Adjusting for a 10% possible loss
of records, the sample size was therefore 110 respondents.