Clinical Feature Vectors (aka, Clinical Co-occurrence Tables)

[karafecho]

Green Team is exploring the use of patient feature vectors (aka, clinical co-occurrence tables) to answer CQs via Translator services. The hope is that the data are rich enough to answer certain types of CQs in a manner that is regulatory compliant and free of IRB concerns. If successful, this approach will allow full execution of CQ1, CQ4, and CQ5.

[karafecho]

CMAQ output for 2010 and 2011 can be found here: https://github.com/ResearchSoftwareInstitute/greendatatranslator/issues/20

[karafecho]

Approach 1:

‘Study’ periods: 1/1/2010 - 12/31/2010 (‘study 1’) and 1/1/2011 - 12/31/2011 (‘study 2’) (two years for which exposures data are available)

Age at ‘study’ start date: calculated from birth date and binned as 0...2, 3...17, 18...34, 35...50, 51...69, 70+ years

Asthma-like phenotype: Default per patient pull criteria (ICD codes for asthma-like conditions)

Exposure to particulate matter: average exposure over ‘study’ period, binned as quintiles

Exposure to ozone: same as for particulate matter

Responsiveness to treatment: #ED/inpatient visits for respiratory issue(s) (defined by same ICD codes used to pull patients with an asthma-like phenotype) over ‘study’ period

Select asthma-related meds administered/prescribed over 'study' period (0 = not administered/prescribed; 1 = administered/prescribed at least one time) (see list of relevant meds below)

Additional variables of interest for state-trait stratification: Obesity: Obesity ICD code OR BMI >= 30 over ‘study’ period; Sex; Race; Ethnicity

Patient feature vector, one for year 2010 and one for year 2011: V(patient, time) = f(Sex, Race, Obesity, Age at study start date, #ED/inpatient visits for respiratory issue, year avg PM2.5 exposure, year avg ozone exposure, meds administered/prescribed over study period)

7 column headers!

Approach 2:

‘Study’ period: 1/1/2010 - 12/31/2010 (‘study 1’) and 1/1/2011 - 12/31/2011 (‘study 2’) (two years for which exposures data are available)

Age at ‘study’ start date: calculated from birth date and binned as 0...2, 3...17, 18...34, 35...50, 51...69, 70+ years

Asthma-like phenotype: Default per patient pull criteria (ICD codes for asthma-like conditions)

Exposure to particulate matter: maximum one-hour exposure over 24-hour period before ED/inpatient visit(s) for respiratory issue(s), binned as quintiles

Exposure to ozone: same as for particulate matter

Responsiveness to treatment: ED/inpatient visits for respiratory issue(s) (defined by same ICD codes used to pull patients with an asthma-like phenotype) and outpatient visits over ‘study’ period

Additional variables of interest for state-trait stratification: Obesity: Obesity ICD code OR BMI >= 30 over ‘study’ period; Sex; Race, Ethnicity

Visit feature vector, one for year 2010 and one for year 2011: V(visit, time) = f(Visit type, Sex, Race, Ethnicity, Obesity, Age at visit, 24 h PM2.5 exposure, 24 h ozone exposure)

6 column headers!

Medications prescribed to patients with asthma-like phenotype:

prednisone fluticasone mometasone budesonide beclomethasone ciclesonide flunisolide albuterol metaproterenol diphenhydramine fexofenadine cetirizine ipratropium salmeterol arformoterol formoterol indacaterol theophylline omalizumab mepolizumab

[stevencox]

Described this approach to NCATS during Reasoner demo.

[karafecho]

Created two spreadsheets (patient level, visit level) to describe our approach and define relevant variables. Reviewed plan with team (Emily, James, Hao, Kimberly, Sarav) and confirmed that everyone is clear with the plan and has all data, resources, etc. required to generate first two clinical feature vector tables.

[empfff]

Clinical data piece done... handed off to @jameschump for linkage to environmental data. James and I will go over the final vectors before we provision to Gamma team.

[karafecho]

Architecture/approach to be presented to the CDWH Oversight Committee for approval to move forward with action plan.

[karafecho]

See #122.

[karafecho]

@stevencox : can we close this issue and leave #122? They seem a bit redundant.

[stevencox]

Sounds good.