Use of Natural Language Processing to Improve Surveillance of Aortic Aneurysms in Women Deemed Ineligible by Current Guidelines

Natural language processing (NLP) programs utilize raw text language to detect disease processes. We sought to use NLP to identify patients with abdominal aortic aneurysms (AAAs) and determine the extent of gender disparities in AAA screening created by United States Preventative Services Task Force (USPSTF) and SVS (Society for Vascular Surgery) guidelines. Illuminate ActKnowledge NLP software was employed at a tertiary care facility to review all imaging studies and identify any patients with AAA. After identification and confirmation of an aneurysm, baseline characteristics were collected and analyzed by gender. After review of over 2.5 million radiology reports, the NLP software identified 11,353 (0.44%) patients (3866 [34%] females and 7486 [66%] males) with a possible AAA. Of these, 3859 were confirmed to have an aneurysm or ectasia, 907 female (24%) and 2952 male (76%) patients, with a corresponding positive screening rate of 0.1% and 0.25%, respectively (Table I). After excluding thoracic aneurysms and patients with aneurysms <3 cm, 1934 patients (445 [23%] female and 1489 [77%] male) comprised the final analysis cohort. Compared with men, women were less commonly smokers (75% vs 80%; P = .042), but did not differ in age, family history of AAA, or aneurysm diameter (P = .193, P = .948, and P = .145, respectively). Female patients were more likely to be racial minorities (Black, Asian, Native American, Other; P = .030) (Table I). The percent of patients that underwent active surveillance (P = .548), reason for non-surveillance (P = .486), and level of patient-provider engagement (P = .313) did not differ by gender (Table II). Women were excluded by USPSTF screening guidelines (100% vs 31% in males; P ≤ .001) and were more likely to miss screening within the window recommended by the SVS compared with men (28.8% vs 23.7%; P ≤ .030). Subsequent death was the most frequent reason for closing identified AAA cases (44% vs 41% of males; P = .5). A total of 825 patients (43%) had an aneurysm identified incidentally, and 451 patients (23%) were identified after previous loss to follow-up, representing 66% of the total cohort. Among these patients, women more frequently had aneurysms found incidentally on imaging (45.7% vs 41.9% of males; P < .001). Although SVS guidelines recommend AAA screening in women smokers, many non-vascular providers defer to USPSTF guidelines, resulting a screening gap that predominantly impacts women. NLP software offers an elegant solution, using existing EMR data to identify and address these gaps in AAA screening and surveillance.TableBaseline patient characteristicsFemale (%)Male (%)TotalRadiology Reports Reviewed Institutional Cohort:4,989,701Total eligible reports1,342,156 (52)1,224,477 (48)2,567,264Total AAA identified3866 (34)7486 (66)11,353Patients Closed907 (24)2952 (76)3,859Active Surveillance265 (25)812 (75)1077 Analysis Cohort:AAA Cohort Identified517 (14)1605 (76)2122Patients with dilation <3 cm (excluded from analysis)107 (57)81 (43)188Final Analysis Cohort445 (23)1489 (77)1934Female (%)N= 445Male (%)N=1489P value (95% CI)Age75 ± 10.175 ± 9.9P = .193P = .531 Age Ranges<65 years56 (13)306 (21)65-75 years158 (36)554 (37)>75 years231 (52)729 (49)History of smoking335 (75)1188 (80)P = .042Family history of AAA11 (3)36 (2)P = .948RaceP = .030 White312 (70.1)1124 (75.8) Black32 (7.2)90 (6.1) Asian35 (7.9)87 (5.9) Native American7 (1.6)6 (.4) Pacific Islander2 (.5)12 (.8) Other57 (12.8)160 (10.1) Decline to answer03 (.2)EthnicityP = .499 Non-Hispanic420 (94.6)1388 (93.7) Hispanic24 (5.4)93 (6.3)Aneurysm DiameterP = .145 3.0-3.9 cm251 (56.4)774 (52.0) 4.0-4.9 cm101 (22.7)354 (23.8) 5.0-5.5 cm29 (6.5)145 (9.7) 5.5 cm +64 (14.4)216 (14.5) Open table in a new tab Table IIComparison of screening results and guidelines by genderFemale (%)N= 445Male (%)N=1489P value (95% CI)Illuminate StatusP = .548 Open (Under surveillance)177 (39.8)616 (41.4) Closed268 (60.2)873 (58.6)Indication for ClosureP = .486 Deceased119 (44.4)359 (41.1) Unable to contact7 (2.6)20 (2.3) Patient declined23 (8.6)56 (6.4) Follows external Vascular Surgeon88 (32.8)328 (37.6) Follows external PCP31 (11.6)110 (12.6)Patient EngagementP = .313 Engaged (follows as recommended)146 (32.9)508 (34.1) Disengaged (lost to follow up)95 (21.4)356 (24.0) Incidental (AAA not previously identified)203 (45.7)622 (41.9)USPSTF Screening GuidelinesP = <.001 Captured (met criteria)01026 (68.9) Did not meet criteria445 (100)463 (31.1) Missed Screening (met criteria)0353 (23.7)P = <.001SVS Screening GuidelinesP = <.001 Captured (met criteria)346 (77.8)1026 (68.9) Did not meet criteria99 (22.3)463 (31.1) Missed Screening (met criteria)128 (28.8)353 (23.7)P = .030 Open table in a new tab