Assessing Different Approaches to Leveraging Historical Smoking Exposure Data to Better Select Lung Cancer Screening Candidates: A Retrospective Validation Study.

INTRODUCTION: There is mounting interest in the use of risk prediction models to guide lung cancer screening. Electronic health records (EHRs) could facilitate such an approach, but smoking exposure documentation is notoriously inaccurate. While the negative impact of inaccurate EHR data on screening practices reliant on dichotomized age and smoking exposure-based criteria has been demonstrated, less is known regarding its impact on the performance of model-based screening. AIMS AND METHODS: Data were collected from a cohort of 37 422 ever-smokers between the ages of 55 and 74, seen at an academic safety-net healthcare system between 1999 and 2018. The National Lung Cancer Screening Trial (NLST) criteria, PLCOM2012 and LCRAT lung cancer risk prediction models were validated against time to lung cancer diagnosis. Discrimination (area under the receiver operator curve [AUC]) and calibration were assessed. The effect of substituting the last documented smoking variables with differentially retrieved "history conscious" measures was also determined. RESULTS: The PLCOM2012 and LCRAT models had AUCs of 0.71 (95% CI, 0.69 to 0.73) and 0.72 (95% CI, 0.70 to 0.74), respectively. Compared with the NLST criteria, PLCOM2012 had a significantly greater time-dependent sensitivity (69.9% vs. 64.5%, p < .01) and specificity (58.3% vs. 56.4%, p < .001). Unlike the NLST criteria, the performances of the PLCOM2012 and LCRAT models were not prone to historical variability in smoking exposure documentation. CONCLUSIONS: Despite the inaccuracies of EHR-documented smoking histories, leveraging model-based lung cancer risk estimation may be a reasonable strategy for screening, and is of greater value compared with using NLST criteria in the same setting. IMPLICATIONS: EHRs are potentially well suited to aid in the risk-based selection of lung cancer screening candidates, but healthcare providers and systems may elect not to leverage EHR data due to prior work that has shown limitations in structured smoking exposure data quality. Our findings suggest that despite potential inaccuracies in the underlying EHR data, screening approaches that use multivariable models may perform significantly better than approaches that rely on simpler age and exposure-based criteria. These results should encourage providers to consider using pre-existing smoking exposure data with a model-based approach to guide lung cancer screening practices.

Derivation and validation of a machine learning record linkage algorithm between emergency medical services and the emergency department.

OBJECTIVE: Linking emergency medical services (EMS) electronic patient care reports (ePCRs) to emergency department (ED) records can provide clinicians access to vital information that can alter management. It can also create rich databases for research and quality improvement. Unfortunately, previous attempts at ePCR and ED record linkage have had limited success. In this study, we use supervised machine learning to derive and validate an automated record linkage algorithm between EMS ePCRs and ED records. MATERIALS AND METHODS: All consecutive ePCRs from a single EMS provider between June 2013 and June 2015 were included. A primary reviewer matched ePCRs to a list of ED patients to create a gold standard. Age, gender, last name, first name, social security number, and date of birth were extracted. Data were randomly split into 80% training and 20% test datasets. We derived missing indicators, identical indicators, edit distances, and percent differences. A multivariate logistic regression model was trained using 5-fold cross-validation, using label k-fold, L2 regularization, and class reweighting. RESULTS: A total of 14 032 ePCRs were included in the study. Interrater reliability between the primary and secondary reviewer had a kappa of 0.9. The algorithm had a sensitivity of 99.4%, a positive predictive value of 99.9%, and an area under the receiver-operating characteristic curve of 0.99 in both the training and test datasets. Date-of-birth match had the highest odds ratio of 16.9, followed by last name match (10.6). Social security number match had an odds ratio of 3.8. CONCLUSIONS: We were able to successfully derive and validate a record linkage algorithm from a single EMS ePCR provider to our hospital EMR.

Neighborhood Disadvantage and Lung Cancer Incidence in Ever-Smokers at a Safety Net Health-Care System: A Retrospective Study.

BACKGROUND: Neighborhood circumstances have an influence on multiple health outcomes, but the association between neighborhood conditions and lung cancer incidence has not been studied in sufficient detail. The goal of this study was to understand whether neighborhood conditions are independently associated with lung cancer incidence in ever-smokers after adjusting for individual smoking exposure and other risk factors. METHODS: A cohort of ever-smokers aged ≥ 55 years was assembled from 19 years of electronic health record data from our academic community health-care system. Patient demographic characteristics and other measures known to be associated with lung cancer were ascertained. Patient addresses at their index visit were geocoded to the census block group level to determine the area deprivation index (ADI), drawn from 5-year estimates from the American Community Survey. A multivariate Cox proportional hazards model was fit to assess the association between ADI and time to lung cancer diagnosis. Tests of statistical significance were two-sided. RESULTS: The study included 19,867 male subjects and 21,748 female subjects. Fifty-three percent of the patients were white, 38% were black, and 5% were Hispanic. Of these, 1,149 developed lung cancer. After adjusting for known risk factors, patients residing in the most disadvantaged areas had a significantly increased incidence of lung cancer compared with those in the least disadvantaged areas (hazard ratio, 1.29; 95% CI 1.07-1.55). CONCLUSIONS: Census-derived estimates of neighborhood conditions have a powerful association with lung cancer incidence, even when adjusting for individual variables. Further research investigating the mechanisms that link neighborhood conditions to lung cancer is warranted.

Learning a Health Knowledge Graph from Electronic Medical Records.

Demand for clinical decision support systems in medicine and self-diagnostic symptom checkers has substantially increased in recent years. Existing platforms rely on knowledge bases manually compiled through a labor-intensive process or automatically derived using simple pairwise statistics. This study explored an automated process to learn high quality knowledge bases linking diseases and symptoms directly from electronic medical records. Medical concepts were extracted from 273,174 de-identified patient records and maximum likelihood estimation of three probabilistic models was used to automatically construct knowledge graphs: logistic regression, naive Bayes classifier and a Bayesian network using noisy OR gates. A graph of disease-symptom relationships was elicited from the learned parameters and the constructed knowledge graphs were evaluated and validated, with permission, against Google's manually-constructed knowledge graph and against expert physician opinions. Our study shows that direct and automated construction of high quality health knowledge graphs from medical records using rudimentary concept extraction is feasible. The noisy OR model produces a high quality knowledge graph reaching precision of 0.85 for a recall of 0.6 in the clinical evaluation. Noisy OR significantly outperforms all tested models across evaluation frameworks (p < 0.01).