Statistical Details for Developmental Predictors

Calculating Developmental Scores

Developmental scores were calculated in two steps. First, each child was given points for each milestone listed in the survey which was met. The number of points granted for each milestone is the age at which the milestone is met on average in typically developing children. In order to account for some omissions, a child was given points for all skills below the highest skill attained. (For example, if a parent reported an age at which the child walked, then points were also given for sitting and standing.) The sum of all points for all skills was calculated for each child.

The second step involved correcting for how old the child was when the skills were learned. The age at which a given milestone is attained by typical children was divided by the age at which it was attained by the respondent child in question. This ratio was averaged over all skills for which data was present, and gives an estimate of how delayed in acquiring skills the child is.

The total number of points was multiplied by this correction for the degree of delay to attain a developmental score.

Calculating Feature Scores

Feature scores were calculated by summing the following:

Feature	Both Eyes	One Eye	None	Unsure/No answer
Legally blind	3	2	0	1
Coloboma	3	2	0	1
Lacunae	3	2	0	1

Feature	Yes		No	Unsure/No Answer
Complete ACC	2		0	1
Polymicrogyria	2		0	1
Hemivertibrae	2		0	1

plus 5 and then subtracting the number of months at which seizures began, or zero when unknown. (The 5 is added to keep all numbers positive.)

The specific features were chosen both because they had some discrimination power, but also because these are features which can typically be assessed at the time of diagnosis.

Unanswered or unknown questions are assigned a score between that of "yes" and "no" answers to certain features, in an attempt to reduce the effect of questions which were left unanswered. Only children over the age of two were included in the analysis in order to eliminate some uncertainty in development. Many children continue to acquire the milestones listed after the age of two, so this cutoff is not ideal, but making it much higher reduces the statistics too significantly to be useful.

Pearson Product Moments

Pearson product moments were calculated to measure the degree of correlation between development and various features. Below is a table summarizing the Pearson r values.

Feature	Pearson's r	n	p
Age when problem first noted	.5	55	<0.01
Age at seizure onset	.45	47	<0.01
Legally blind	.4	59	<0.01
Coloboma	.29	55	<0.05
Chorioretinal lacunae	.33	72	<0.01
Microcephaly	.31	64	<0.02
Complete ACC	.21	69	<0.10
Polymicrogyria	.40	40	<0.01
Scoliosis	.24	68	<0.05
Hemivertebrae	.26	64	<0.05
Feature score	.5	64	<0.01

Note that for eye findings it was possible to have the abnormality in both eyes, one eye, or neither eye. This allowed for a meaningful correlation with chorioretinal lacunae, as all resondents had them in at least one eye, but several only had them in one eye.

Other Techniques

More sophisticated statistical techniques to determine development based on features were tried, but none provided as good of a correlation as simply adding up the number of features. A nearest-neighbor technique was attempted, finding the nearest neighbors to a given child in the multi-dimensional space defined by the above features. This resulted in an insignificant correlation. The second technique involved calculating the average developmental score for children with and without each feature separately. This would be the expected developmental score for a child with or without this feature. For each child, the expected developmental scores for each feature was determined based on wether or not the child had that feature. Then all expected scores for all features for that child were averaged to give an overall expected developmental score. While this did produce a significant correlation, it was so weak as to not be useful.