Table 1.

Different approaches for phenotypic discovery with the associated advantages and disadvantages.

References	Age (years)	Sample size	Methodology	Strengths	Limitations
(13)	1–6	54	Subjective sub-typing	- Phenotypes are observable expressions - Choice of cutoff guided by investigator expertise - Simple	- Predefined or hypothesized criteria needed - Rare patterns may be missed - Risk of over- or under- fitting as there are no objective statistical criteria for judging fit - Subjective cut-offs need to be recalibrated when new data becomes available - Un-validated cut-offs pose challenge for comparing findings across studies
(14)	1–6	826
(15)	1–6	6265	Latent class analysis	- Probabilistic class allocation. - No prior knowledge is needed. - Hidden patterns may be uncovered that could not be a priori. - Hypothesis generating - Objective statistical criteria for judging whether phenotypes represent true variation	- Discovered sub-types are latent and retrospective by nature - Within-class heterogeneity arising from individuals whose patterns do not exemplify any phenotype - Meaningful clinical interpretation required to explain the patterns - Number of derived phenotypes may be related to the frequency and timing of data collection - Unclear to what extent established phenotype labels convey temporal patterns
(10)	1–7	689
(16)	1–9	953
(17)	1–8	5760
	1–8	2810
(18)	1–8	1184
(19)	8–12	3890
(20)	3–5	946	Principal component analysis	- Accounts for - coexisting symptoms - Reduces the variable dimensions in complex diseases	- Difficult clinical interpretation - Not useful for categorical and longitudinal data unless properly specified
(21)	7–35	925	Exploratory factor analysis
(22)	6–18	613	Hierarchical clustering	- No a priori info about the number of classes required	- Risk of misclassifying distinct phenotypes that are present at low frequency