Type/Design/Method entity/field?

[roll]

Background
Related to #21 issue.

Topic

Method - some trial implementation rules

Our ideal data model doesn't have a Method entity or field. It looks it's a vital part of a trial description. Example of method: Randomised - random number generation + admission order

Proposed solutions

• we can have entity Method with m2m relation with Trial (as in Cochrane database)
• we can have field Trial.method implemented as an array of strings

See also - https://github.com/okfn/opentrials-cochrane-schizophrenia-data/wiki

[pwalsh]

Similar comments here to those I made on #7. m2m seems wasteful if we are not expecting commonality - ie - that each trial just has its own text description(s) of method. Can you run some queries to see actually if there is any consistency here across trials?

BTW @roll this is great work, well done.

[danfowler]

@pwalsh @roll There are only 43 different method descriptions spread across 8851 studies, so I think it might be OK to do the m2m option.

[pwalsh]

Is there a further way to break down these strings into more structured data?

Example: @roll has an example which essentially is 'random'. If we look at these 43 manually, I'm curious if these are just textual variations on themes that could be enum-ed.

[roll]

@pwalsh But it's just one dataset. I think to make it more structured we have to have a very deep understanding of the domain and other dataset examples. To do not run into troubles further.

All of this looks like point 15 of http://www.who.int/ictrp/network/trds/en/ (Study type):

• (from Design)

"RCT"
"CCT"

• (from Method)

"Cluster randomisation (events)"
"Cluster randomisation (wards)"
"Cluster randomisation (families)"
"Cluster randomisation (homes)"
"Crossover"
"Blinding - open-label"
"Blinding - unclear"
"Blinding - double"
"Blinding - single"
"Cluster randomisation (mental health workers)"
"Blinding - triple"
"Blinding - multi-blind"
"Cluster randomisation (general practices)"
"Cluster randomisation (mental health centres)"
"Factorial"
"Unclear"
"Unclear - requires translation"
"Non blinded"
"Randomised- no further info"
"Block randomisation"
"Randomised - admission order"
"Randomised- tossing a coin"
"Randomised- drawing straws"
"Randomised - date of admission, odd or even"
"Randomised - consultation order"
"Randomised - random number generation"
"Randomised - number of admission, odd or even"
"Randomised - computer random number generation"
"Randomised - admission  and consultation order"
"Randomised - discharge order"
"Randomised - admission order, even or odd"
"Randomised - admission order and tossing a coin"
"Randomised - admission number, odd or even"
"Randomised - selection order"
"Randomised - random number generation + admission order"
"Randomised - random number generation table"
"Randomised - random number, odd or even"
"Randomised - admission order and random number generation"
"Randomised - case note numbers"
"Randomised - admission order + stratified"
"Randomised - block randomisation"
"Randomised - card"
"Quasi randomization"

[roll]

But here we see more structure - https://clinicaltrials.gov/ct2/show/results/NCT00922272?term=schizophrenia&rank=121&sect=X30156

Study Type - Interventional
Study Design - Allocation: Randomized; Endpoint Classification: Safety/Efficacy Study; Intervention Model: Parallel Assignment; Masking: Double Blind (Subject, Caregiver, Investigator, Outcomes Assessor); Primary Purpose: Treatment

[pwalsh]

Those methods, I'm inclined to say we make a list of types, and have a method description field.

EG:

"Randomised - admission  and consultation order"
"Randomised - discharge order"

becomes:

• method_type: "randomised" (enum or some other structure)
• method_description: "admission and consultation order" or "discharge order"

WDYT?

[pwalsh]

method_type might become method_design or study_design if we want to align with other data sets... not sure.

[roll]

I see it in the same way on the dataset we have (about randomized/blinded/etc + type). Just a little bit afraid we don't see the whole picture.

It looks like - #6 (comment) - it's kinda a reference for our problem. Here is some study types (1 type per study) and study designs (tags).

[roll]

Cochrane's RCT, CCT (from Dan):

http://handbook.cochrane.org/chapter_6/box_6_3_a_cochrane_definitions_and_criteria_for_randomized.htm