Table 4.

Patient medical history features and weights used in the final Mass General Brigham Biobank CRT^PFT+ algorithm for classification of COPD.

Model feature	Model weight	Variable type	Description
Intercept	− 1.871
everPFTlt70	1.750	NLP	Ever had a pulmonary function test with spirometry indicating pre-bronchodilator FEV1/FVC ratio < 0.7 OR post-bronchodilator FEV1/FVC ratio < 0.7
nCOPDGTE3_365	0.465	Coded	Ever diagnosed with 3 or more COPD-related ICD codes within any rolling time window of 365 days
everTiotropium	0.334	Coded	Ever been prescribed tiotropium
iNotWhite	− 0.239	Coded	Race category denoting whether subject is White or Not White
smkEver	0.175	NLP	Any current/former history of smoking
everdxAtPulmClinic	0.056	Coded	Ever diagnosed with a COPD-related ICD code at a pulmonary clinic
everCOPDmed	0.048	Coded	Ever been prescribed a medication used to treat COPD?
nmedLAMA	0.017	Coded	Total count of distinct prescription codes for long acting muscarinic antagonists in participant medical record for treatment of lung diseases
pftCount	− 0.016	Coded	Total count of any kind of pulmonary function test
ageCOPDt1Specific	0.013	Coded	Age (in years) at first ICD code specific to COPD
nCOPD_ICD	0.013	Coded	The COPD feature count of distinct dates on which a subject has a code from this feature ICD10: J40–J44 ICD9: 491, 492, 493.2, 496
nBronchitis	− 0.009	Coded	The count of distinct dates on which a subject has a Bronchitis ICD code ICD10: J40, J41, J42 ICD9: 490, 491
nBronchiectasis	− 0.008	Coded	The count of distinct dates on which a subject has a Bronchiectasis ICD code ICD10: J47 ICD9: 494
patient_dxenct	− 0.001	Coded	Total number of encounters (visits) per subject with a coded diagnosis (any diagnosis not limited to COPD)

The case assignment threshold for this model, holding specificity at 95%, was 0.754. For subjects who were missing PFT results, the everPFTlt70 variable was classified as ‘No’ in this model.