Table 2.

Ranked topics for different document types using the four models (MPT, DMR, TNG, and LDA) based on the topic hyperparameters. TNG and LDA do not use document metadata, so the prior for the topic distribution is the same for every document type. Incorporating phrases into the model results in more targeted topics and higher weighting for these topics based on document metadata. Incorporating document metadata allows for patient- and document-specific topics, while models without metadata use only one document distribution for all document types.

MPT		DMR
	Operative report		Operative report
3.93281	catheter, prepped, using, draped, inserted	1.814893	catheter, needle, using, vein, placement
1.692049	placed, operation, cranial_nerve, taken, used	1.66272	operation, placed, surgeon, incision, anesthesia
1.05243	placed, removed, using, performed, operation	0.843987	normal, csf, stomach, endoscope, consent
0.777849	tumor, using, dura, resection, resection_cavity	0.753292	tumor, using, resection, flap, intraoperative
0.666869	approximately, appeared, obtained, performed, felt	0.338279	eye, lens, anterior, tube, chamber
	Discharge summary		Discharge summary
3.182792	course, continued, home, hospitalization, follow-up	3.451551	course, started, continued, therapy, home
2.972885	course, follow, discharged, instructed, pain	1.947799	surgery, pain, stable, home, postoperative
1.716823	stroke, given, prior, acute, evidence	0.805114	stroke, cerebral, artery, ct, middle
1.55142	transfer, course, started, tube, continued	0.653268	tumor, glioblastoma, resection, radiation, multiforme
1.273632	able, daily, therapy, activities, improved	0.572004	stroke, mri, mca, tpa, arm
	Progress note		Progress note
1.327833	continue, plan, bid, inpatient, hr	0.671341	continue, plan, bid, bp, hr
1.09157	plan, year-old, clear, edema, currently	0.357264	plan, inpatient, stroke, seen, discussed
0.766955	bid, given, continue, qd, old	0.347626	outpatient, able, therapy, good, having
0.685851	intact, year-old, normal, seen, mri_scan	0.340108	medicine, stable, continue, htn, pain
0.674692	likely, given, iv, blood_pressure, inpatient	0.286204	clear, abdomen, extremities, rate, regular
TNG		LDA
	All report types		All report types
0.52712	given, lower_extremity, stroke, upper_extremity, approximately	0.10219	given, evidence, therapy, significant, possible
0.42026	able, normal_limits, daily, acute_distress, therapy	0.09007	able, having, problems, normal, days
0.40148	glioblastoma_multiforme, brain, decadron, mri, dilantin	0.08062	clear, rate, abdomen, extremities, regular
0.37318	denies, pain, clear, shortness, systems	0.05552	continue, plan, stable, inpatient, bp
0.32127	currently, negative, plan, continue, year-old	0.05437	allergies, family, social, systems, denies