Page 193 - Computational Retinal Image Analysis

188 CHAPTER 10 Statistics in ophthalmology

• An alternative to complete case analysis is to impute the missing data.
Imputation replaces the missing data with some plausible value predicted
from that subject’s (or another subject’s) data. One method of imputation,
which is commonplace in ophthalmic literature, is that of last observation
carried forward (LOCF), where any missing data is replaced with the last
observed value for that patient. Such method assumes that there was no
change in the variable over time since the last observation was obtained and
recorded.
• There are multiple imputation (MI) methods that draw plausible values multiple
times from the observed distributions of relevant variables and aggregates
the results incorporating the differences between them in the estimates of
uncertainty. A multiple imputation method is a superior method to simple
imputation methods or to LOCF, but is only appropriate when the assumption of
MAR can be made [28].
• There are model based approaches, that that do not impute data but rather fit the
model to all the data available hence they do not omit any cases or patients with
missing data, i.e. all cases and patients data are utilized. Those methods are, for
example, generalized estimating equations (GEE), multivariate normal linear
models (MNLM) also referred to as mixed-model repeated measures (MMRM).
Such methods are only valid for MAR.
• Finally, there are model based approaches that do not impute data, they model
the data available as well as they model the drop-out (i.e. the missingness
mechanism). Some of the methods are selection models (SM), pattern-mixture
models (PMM) and joint modeling.
Appropriateness of the statistical analysis methods depends on the reason of data
missingness. The main points to remember are the following:
• In MCAR scenario, complete case analysis is valid, though there is a
loss of information. In a longitudinal study LOCF is valid if there are
not trends with time. While this method may be appropriate when there
is little missing data, it can lead to incorrect conclusions. If an available
case analysis is conducted, it is essential to examine reasons for data being
missing. If there are not many missing data, an available case analysis
with a valid assumption of data being MCAR may be unbiased (i.e. it
does not overestimate or underestimate a treatment difference or evidence
of association), but it will have lower power to detect a difference or
association than if all data were present [29].
• In MAR scenarios, CC, LOCF and GEE are invalid, as they can yield biased
estimates of associations. The methods MI, MNLM and weighted GEE may be
appropriate to use [30].
• In MNAR scenarios, CC, LOCF, GEE, MI and MNLM are not recommended to
use because they can lead to biased inference. SM, PMM or joint modeling may
be appropriate in this scenario.