Artificial intelligence approaches for phenotyping heart failure in U.S. Veterans Health Administration electronic health record.

AIMS: Heart failure (HF) is a clinical syndrome with no definitive diagnostic tests. HF registries are often based on manual reviews of medical records of hospitalized HF patients identified using International Classification of Diseases (ICD) codes. However, most HF patients are not hospitalized, and manual review of big electronic health record (EHR) data is not practical. The US Department of Veterans Affairs (VA) has the largest integrated healthcare system in the nation, and an estimated 1.5 million patients have ICD codes for HF (HF ICD-code universe) in their VA EHR. The objective of our study was to develop artificial intelligence (AI) models to phenotype HF in these patients. METHODS AND RESULTS: The model development cohort (n = 20 000: training, 16 000; validation 2000; testing, 2000) included 10 000 patients with HF and 10 000 without HF who were matched by age, sex, race, inpatient/outpatient status, hospital, and encounter date (within 60 days). HF status was ascertained by manual chart reviews in VA's External Peer Review Program for HF (EPRP-HF) and non-HF status was ascertained by the absence of ICD codes for HF in VA EHR. Two clinicians annotated 1000 random snippets with HF-related keywords and labelled 436 as HF, which was then used to train and test a natural language processing (NLP) model to classify HF (positive predictive value or PPV, 0.81; sensitivity, 0.77). A machine learning (ML) model using linear support vector machine architecture was trained and tested to classify HF using EPRP-HF as cases (PPV, 0.86; sensitivity, 0.86). From the 'HF ICD-code universe', we randomly selected 200 patients (gold standard cohort) and two clinicians manually adjudicated HF (gold standard HF) in 145 of those patients by chart reviews. We calculated NLP, ML, and NLP + ML scores and used weighted F scores to derive their optimal threshold values for HF classification, which resulted in PPVs of 0.83, 0.77, and 0.85 and sensitivities of 0.86, 0.88, and 0.83, respectively. HF patients classified by the NLP + ML model were characteristically and prognostically similar to those with gold standard HF. All three models performed better than ICD code approaches: one principal hospital discharge diagnosis code for HF (PPV, 0.97; sensitivity, 0.21) or two primary outpatient encounter diagnosis codes for HF (PPV, 0.88; sensitivity, 0.54). CONCLUSIONS: These findings suggest that NLP and ML models are efficient AI tools to phenotype HF in big EHR data to create contemporary HF registries for clinical studies of effectiveness, quality improvement, and hypothesis generation.

Assessment of Lung Cancer Risk Among Smokers for Whom Annual Screening Is Not Recommended.

Importance: The US Preventive Services Task Force does not recommend annual lung cancer screening with low-dose computed tomography (LDCT) for adults aged 50 to 80 years who are former smokers with 20 or more pack-years of smoking who quit 15 or more years ago or current smokers with less than 20 pack-years of smoking. Objective: To determine the risk of lung cancer in older smokers for whom LDCT screening is not recommended. Design, Settings, and Participants: This cohort study used the Cardiovascular Health Study (CHS) data sets obtained from the National Heart, Lung and Blood Institute, which also sponsored the study. The CHS enrolled 5888 community-dwelling individuals aged 65 years and older in the US from June 1989 to June 1993 and collected extensive baseline data on smoking history. The current analysis was restricted to 4279 individuals free of cancer who had baseline data on pack-year smoking history and duration of smoking cessation. The current analysis was conducted from January 7, 2022, to May 25, 2022. Exposures: Current and prior tobacco use. Main Outcomes and Measures: Incident lung cancer during a median (IQR) of 13.3 (7.9-18.8) years of follow-up (range, 0 to 22.6) through December 31, 2011. A Fine-Gray subdistribution hazard model was used to estimate incidence of lung cancer in the presence of competing risk of death. Cox cause-specific hazard regression models were used to estimate hazard ratios (HRs) and 95% CIs for incident lung cancer. Results: There were 4279 CHS participants (mean [SD] age, 72.8 [5.6] years; 2450 [57.3%] women; 663 [15.5%] African American, 3585 [83.8%] White, and 31 [0.7%] of other race or ethnicity) included in the current analysis. Among the 861 nonheavy smokers (<20 pack-years), the median (IQR) pack-year smoking history was 7.6 (3.3-13.5) pack-years for the 615 former smokers with 15 or more years of smoking cessation, 10.0 (5.3-14.9) pack-years for the 146 former smokers with less than 15 years of smoking cessation, and 11.4 (7.3-14.4) pack-years for the 100 current smokers. Among the 1445 heavy smokers (20 or more pack-years), the median (IQR) pack-year smoking history was 34.8 (26.3-48.0) pack-years for the 516 former smokers with 15 or more years of smoking cessation, 48.0 (35.0-70.0) pack-years for the 497 former smokers with less than 15 years of smoking cessation, and 48.8 (31.6-57.0) pack-years for the 432 current smokers. Incident lung cancer occurred in 10 of 1973 never smokers (0.5%), 5 of 100 current smokers with less than 20 pack-years of smoking (5.0%), and 26 of 516 former smokers with 20 or more pack-years of smoking with 15 or more years of smoking cessation (5.0%). Compared with never smokers, cause-specific HRs for incident lung cancer in the 2 groups for whom LDCT is not recommended were 10.54 (95% CI, 3.60-30.83) for the current nonheavy smokers and 11.19 (95% CI, 5.40-23.21) for the former smokers with 15 or more years of smoking cessation; age, sex, and race-adjusted HRs were 10.06 (95% CI, 3.41-29.70) for the current nonheavy smokers and 10.22 (4.86-21.50) for the former smokers with 15 or more years of smoking cessation compared with never smokers. Conclusions and Relevance: The findings of this cohort study suggest that there is a high risk of lung cancer among smokers for whom LDCT screening is not recommended, suggesting that prediction models are needed to identify high-risk subsets of these smokers for screening.