mim18 / loyalty_cohort

Uses heuristics to find patients who get the majority of their care from your hospital system

Loyalty Cohort Algorithm

Implements the loyalty cohort algorithm defined in External Validation of an Algorithm to Identify Patients with High Data-Completeness in Electronic Health Records for Comparative Effectiveness Research by Lin et al.

See appendix table S1 for the regression equation used to calculate the score.

See algorithm outline for more details on the algorithm.

Notes:

• Relies on i2b2 data using the ACT ontology.
• This version is MSSQL. Oracle refactor underway in a separate branch.
• (Maybe - The Loyalty Paths might be missing some sibling nodes).

From your ACT database...

To install:

1. Run ddl_xref_LoyaltyCode_paths.sql to create the xref_LoyaltyCode_paths table.
2. Run ddl_LU_CHARLSON.sql to create the LU_CHARLSON table.
3. Import the contents of xref_LoyaltyCode_paths.csv into the table, or run insert_xref_LoyaltyCode_paths.sql
4. Run insert_LU_CHARLSON.sql to populate the LU_CHARLSON table. Replace the prefix in the code pattern column if your data do not use ICD10CM: and ICD9CM:
5. Run ddl_dml_xref_LoyaltyCohort_PSCoeff.sql to create and load the Predicted Score coefficients table.
6. Run dbtype/LoyaltyCohortdbtype to create the usp_LoyaltyCohort_opt stored procedure.

To run:

1. Create a cohort filter table, defining the patients on which to compute loyalty scores. The three columns are:
   • patient_num: patient_num from the i2b2 tables
   • cohort_name: a name for the cohort. You can optionally compute several cohorts separately, but specifying different values for this.
   • index_dt: a date which is a reference point in time at which to compute the loyalty score. It is suggested to select a common recent point in time or to choose each patient's most recent visit date, for example.

DECLARE @cfilter udt_CohortFilter

INSERT INTO @cfilter (PATIENT_NUM, COHORT_NAME, INDEX_DT)
   select distinct patient_num, substring(cohort,1,charindex('202',cohort)-1) cohort, admission_date 
 FROM [FourCE_LocalPatientSummary] /*  4CE X.2 COHORT FOR EXAMPLE */

2. Customize the following statement and execute on your database to compute the loyalty cohort. EXEC [dbo].[usp_LoyaltyCohort_opt] @site='UKY', @lookbackYears=1, @demographic_facts=1, @gendered=0, @cohort_filter=@cfilter, @output=0 Note that steps 1-2 must be run in the same transaction.

3. Execute the following statement on your database to print the output that can be shared: select * from loyalty_dev_summary where Summary_Description='PercentOfSubjects'

4. We are collecting outputs of this script to compare heuristics. If participating, contact us for access and then paste the output of step 3 into the Google sheet here: https://docs.google.com/spreadsheets/d/1ubuRt_ffVcZiQgUdOmeMXxgjOdfkpQe2FNFyt0u2Un4/edit?usp=sharing

5. The script also outputs patient level data in loyalty_dev and loyalty_charlson_dev. These cohorts are dependent on lookbackYears but not demographic_facts.

Parameter description:

• site: A short (3-character) identifier for your site.
• lookbackYears: A number of years for lookback. The original algorithm used 1 year, but we have found 3- or 5-years are more accurate, because some preventitive care like PSA and Pap Smears do not occurr every year.
• demographic_facts: Set to 1 if demographic facts (e.g., age, sex) are stored in the fact table (rather than patient_dimension).
• gendered: Set to 1 to create a summary table (and cutoffs) that do not include male-only facts for female patients in the denominator and vice-versa.
• filter_by_existing_cohort and cohort_filter: If the first is 1, specify a table variable of (PATIENT_NUM, COHORT_NAME) in the second parameter.
• output: If 1, pretty-print the loyalty_dev_summary percentages after the run.