Page 311 - CyberSecurityEssentialsEnglish

Screening usually happens in two stages: title and abstract, then full text. In the first stage
we lean toward sensitivity so that a suitable study is not lost because of an incomplete
abstract. In the second we apply the criteria more strictly, recording the reason for each

exclusion. AI can rank records or highlight phrases, but the final decision must rest on the
text and the criteria, especially when the terminology or the designs are complex.

A PRISMA flow diagram shows how records move from identification through screening to

inclusion, with counts and reasons for exclusion [2][3]. The diagram does not turn a review
into a systematic one; it is a transparency device. Quality also comes from a clear protocol,
an appropriate search, reliable screening, careful extraction, an appraisal of the risk of bias,

and a synthesis proportionate to the data.

Relevance asks: does the source answer the question? Quality asks: was the study designed,
carried out, and reported in a way that reduces error and bias? A high-quality study may be
irrelevant to your context, and a highly relevant report may be methodologically weak. A

single label such as “peer-reviewed” or “not peer-reviewed” is therefore not enough.
The strength of evidence is not determined by design alone. A randomized trial may be weak
if it is small, loses participants, or measures an unsuitable outcome, and an observational

study may be very useful for understanding a rare harm or a real-world context. Evidence
hierarchies are a first guide; we then review the execution, the risks of bias, the precision,

the consistency, and the directness.

Bias can enter in how participants were selected, in how exposure and outcomes were
measured, in missing data, in confounding nobody controlled for, in selective reporting, and
in conflicts of interest. In qualitative studies we attend to the suitability of the sample, the
clarity of data collection, the researcher’s reflexivity, the depth of the analysis, and the

support of interpretations by quotations or other evidence. There is no single checklist for
every design, which is why reporting-guideline libraries help in finding the appropriate

instrument [7][8].

Publication status must be checked. A source may be a preprint that has not been peer-
reviewed, a conference version, a corrected article, a retracted paper, or one carrying an
expression of concern. A retraction does not mean every idea in the paper is wrong, but it is

a strong signal that requires understanding the reason and not citing the finding as valid
evidence without explanation.

Institutional reports and government documents need a slightly different appraisal: which
body produced it? What was the purpose? How were the data collected? Are the definitions

stable? Are there methodological annexes? Are the figures provisional or final? And is there
an interest in highlighting a particular result? An official source matters, but it is not immune
to limits of measurement, delay, or changing definitions.

310