Page 194 - Computational Retinal Image Analysis

5 Missingness of data 189

5.3 Words of caution for dealing with missing data
Clearly there are situations where there is no information about those who are miss-
ing. In such cases, we would recommend drawing attention to the presence of miss-
ing data and the fact that it was not possible to investigate further. By doing this,
readers are aware of the potential for bias. Best and worst-case scenarios could be
considered to show how conclusions might have differed under such circumstances.
E.g, in a study comparing drugs A and B, and where the primary outcome is treat-
ment success, a best case scenario might be that all those lost to follow-up on treat-
ment A were successes while all those on treatment B were failures. A worst-case
scenario would reverse these assumptions.
The consequence of missingness clearly increase with larger number of missing
data. If the numbers lost are smaller this clearly limits their likely impact on study
conclusions, yet even small numbers can alter study conclusions [18].
It is important to note that different missing data analytical approaches can yield
different conclusions. For example, in a detailed study Fielding et al. [31] describes
the REFLUX trial which randomized 357 participants with gastro-oesophageal re-
flux disease to surgery or medicine and had an overall high response rate of 89%, i.e.
11% of patients did not have a record of the primary outcome. The authors exam-
ined the impact of missing data on a quality of life outcome measure, the EuroQuol
EQ-5D which is the primary outcome of a large clinical trial currently being con-
ducted on patients with glaucoma. Fielding et al. explored eight different approaches
to missing data and show that while two approaches gave statistically significant
results, six did not; and that for the statistically significant models, one estimated an
effect that was of clinical significance, the other did not. Choice of analysis method
for missing data can thus impact on conclusions. A word of caution on complexity of
advanced statistical methods is provided by Streiner, however: “the easy methods are
not good and the good ones … are not easy” [32].
In summary, it is the best to take preventive measures to avoid missing data.
This can be done by closely following the study protocol and taking preventative
measures to avoid mistakes, this also includes the recording of the reasons for the
missingness. If missing data occur then it is essential to visualize summaries or plots
for the pattern of the missingness. It is important to remember that if the assumptions
made in relation to missingness are incorrect, then the analyses may lead to mislead-
ing results and conclusions. Therefore, it is crucial to report the amount and the pat-
tern of the missingness and to report the methods used to handle the missing data in
the analyses. For cohort studies it is advisable to follow the reporting guidelines of
Strengthening the Reporting of Observational studies in Epidemiology (STROBE)
STROBE [33–35]. For randomized clinical trials it is advisable to follow the guide-
lines of CONsolidated Standard of Reporting Trials (CONSORT) [36] and for diag-
nostic studies to follow the STARD publication standard [37]. This will ensure that
missing data are reported with enough detail to allow readers to assess the validity
of the results. It is important to remember, that incomplete data and the statistical
methods used to deal with the missing data can lead to bias, or can be inefficient.