IMPORTANT NOTES
Always remember to access and analyze data on the CAMH server (eg in SCC)
Never save data on your personal computer
Data can be accessed/shared via the Y Drive (see instructions in the Powerpoint below) or via postgres (instructions below)
For more information and best practices, see:
Accessing data via a data mart in postgres
The following datasets are available via postgres:
DATA (POSTGRES) |
---|
1 <Id> schema = fml01_cmh, table = DCRestraint_20220826 FALSE 2 <Id> schema = fml01_cmh, table = DCDASAIV_20220519 FALSE 3 <Id> schema = fml01_cmh, table = medication_20220610 FALSE 4 <Id> schema = fml01_cmh, table = DCED_WaitTime_20220505 FALSE 5 <Id> schema = fml01_cmh, table = DCIncidentBehaviorsAndAssaults_20220531 FALSE 6 <Id> schema = fml01_cmh, table = encounter_20220516 FALSE 7 <Id> schema = fml01_cmh, table = HealthEquity_20220516 FALSE 8 <Id> schema = fml01_cmh, table = DCOrders_PICU_20220512 FALSE 9 <Id> schema = fml01_cmh, table = encounter_20220516_bkp FALSE |
To access this data, you must have been provided access by IMG staff (to request access, please let Marta or Laura know). You will also need to install the DBI package in R (https://dbi.r-dbi.org/). Once access is granted and the package is installed, you can use the following code to access these data:
R code |
---|
# import library # Connect to a specific postgres database DBI::dbListObjects(con, DBI::Id(schema = 'fml01_cmh')) #gives you a list of available databases # as an example, we will query all data from the restraints dataset: |
Data analysis plan
Predictors
We will train models on patient demographics (HealthEquity), diagnosis (encounter), admission information (encounter), structured risk assessments (DCDASAIV) and unstructured risk assessments (eg clinical notes/FMTs, when available). We will use these predictors recorded on each day a DASA was administered to predict our composite outcome (occurring between one DASA score and the next DASA score).
Outcomes
Our models will predict a composite outcome made up of the occurrence of a violent/aggressive incident (recorded in SCORES/STRIDES, DCIncidentBehaviorsAndAssaults), a restraint event (differentiating between mechanical or chemical only), and/or a seclusion event (restraints/seclusions are from the structured portions of the powerforms, DCRestraint).
Our data log which documents initial processing tasks, assigned team members, and completion dates can be found here: