Constructing Epidemiologic Cohorts from Electronic Health Record Data

Tailoring Risk Prediction Models to Local Populations

Importance

Risk estimation is an integral part of cardiovascular care. Local recalibration of guideline-recommended models could address the limitations of existing tools.

Objective

To provide a machine learning (ML) approach to augment the performance of the American Heart Association's Predicting Risk of Cardiovascular Disease Events (AHA-PREVENT) equations when applied to a local population while preserving clinical interpretability.

Design, Setting, and Participants

This cohort study used a New England-based electronic health record cohort of patients without prior atherosclerotic cardiovascular disease (ASCVD) who had the data necessary to calculate the AHA-PREVENT 10-year risk of developing ASCVD in the event period (2007-2016). Patients with prior ASCVD events, death prior to 2007, or age 79 years or older in 2007 were subsequently excluded. The final study population of 95 326 patients was split into 3 nonoverlapping subsets for training, testing, and validation. The AHA-PREVENT model was adapted to this local population using the open-source ML model (MLM) Extreme Gradient Boosting model (XGBoost) with minimal predictor variables, including age, sex, and AHA-PREVENT. The MLM was monotonically constrained to preserve known associations between risk factors and ASCVD risk. Along with sex, race and ethnicity data from the electronic health record were collected to validate the performance of ASCVD risk prediction in subgroups. Data were analyzed from August 2021 to February 2024.

Main Outcomes and Measures

Consistent with the AHA-PREVENT model, ASCVD events were defined as the first occurrence of either nonfatal myocardial infarction, coronary artery disease, ischemic stroke, or cardiovascular death. Cardiovascular death was coded via government registries. Discrimination, calibration, and risk reclassification were assessed using the Harrell C index, a modified Hosmer-Lemeshow goodness-of-fit test and calibration curves, and reclassification tables, respectively.

Results

In the test set of 38 137 patients (mean [SD] age, 64.8 [6.9] years, 22 708 [59.5]% women and 15 429 [40.5%] men; 935 [2.5%] Asian, 2153 [5.6%] Black, 1414 [3.7%] Hispanic, 31 400 [82.3%] White, and 2235 [5.9%] other, including American Indian, multiple races, unspecified, and unrecorded, consolidated owing to small numbers), MLM-PREVENT had improved calibration (modified Hosmer-Lemeshow P > .05) compared to the AHA-PREVENT model across risk categories in the overall cohort (χ23 = 2.2; P = .53 vs χ23 > 16.3; P < .001) and sex subgroups (men: χ23 = 2.1; P = .55 vs χ23 > 16.3; P < .001; women: χ23 = 6.5; P = .09 vs. χ23 > 16.3; P < .001), while also surpassing a traditional recalibration approach. MLM-PREVENT maintained or improved AHA-PREVENT's calibration in Asian, Black, and White individuals. Both MLM-PREVENT and AHA-PREVENT performed equally well in discriminating risk (approximate ΔC index, ±0.01). Using a clinically significant 7.5% risk threshold, MLM-PREVENT reclassified a total of 11.5% of patients. We visualize the recalibration through MLM-PREVENT ASCVD risk charts that highlight preserved risk associations of the original AHA-PREVENT model.

Conclusions and Relevance

The interpretable ML approach presented in this article enhanced the accuracy of the AHA-PREVENT model when applied to a local population while still preserving the risk associations found by the original model. This method has the potential to recalibrate other established risk tools and is implementable in electronic health record systems for improved cardiovascular risk assessment.

Absence of nonfatal suicidal behavior preceding suicide death reveals differences in clinical risks

Nonfatal suicidality is the most robust predictor of suicide death. However, only ~10% of those who survive an attempt go on to die by suicide. Moreover, ~50% of suicide deaths occur in the absence of prior known attempts, suggesting risks other than nonfatal suicide attempt need to be identified. We studied data from 4,000 population-ascertained suicide deaths and 26,191 population controls to improve understanding of risks leading to suicide death. This study included 2,253 suicide deaths and 3,375 controls with evidence of nonfatal suicidality (SUI_SI/SB and CTL_SI/SB) from diagnostic codes and natural language processing of electronic health records notes. Characteristics of these groups were compared to 1,669 suicides with no prior nonfatal SI/SB (SUI_None) and 22,816 controls with no lifetime suicidality (CTL_None). The SUI_None and CTL_None groups had fewer diagnoses and were older than SUI_SI/SB and CTL_SI/SB. Mental health diagnoses were far less common in both the SUI_None and CTL_None groups; mental health problems were less associated with suicide death than with presence of SI/SB. Physical health diagnoses were conversely more often associated with risk of suicide death than with presence of SI/SB. Pending replication, results indicate highly significant clinical differences among suicide deaths with versus without prior nonfatal SI/SB.

Risk of Gastric Adenocarcinoma in a Multiethnic Population Undergoing Routine Care: An Electronic Health Records Cohort Study

BACKGROUND

Gastric adenocarcinoma (GAC) is often diagnosed at advanced stages and portends a poor prognosis. We hypothesized that electronic health records (EHR) could be leveraged to identify individuals at highest risk for GAC from the population seeking routine care.

METHODS

This was a retrospective cohort study, with endpoint of GAC incidence as ascertained through linkage to an institutional tumor registry. We utilized 2010 to 2020 data from the Palo Alto Medical Foundation, a large multispecialty practice serving Northern California. The analytic cohort comprised individuals ages 40-75 receiving regular ambulatory care. Variables collected included demographic, medical, pharmaceutical, social, and familial data. Electronic phenotyping was based on rule-based methods.

RESULTS

The cohort comprised 316,044 individuals and approximately 2 million person-years (p-y) of observation. 157 incident GACs occurred (incidence 7.9 per 100,000 p-y), of which 102 were non-cardia GACs (incidence 5.1 per 100,000 p-y). In multivariable analysis, male sex [HR: 2.2, 95% confidence interval (CI): 1.6-3.1], older age, Asian race (HR: 2.5, 95% CI: 1.7-3.7), Hispanic ethnicity (HR: 1.9, 95% CI: 1.1-3.3), atrophic gastritis (HR: 4.6, 95% CI: 2.2-9.3), and anemia (HR: 1.9, 95% CI: 1.3-2.6) were associated with GAC risk; use of NSAID was inversely associated (HR: 0.3, 95% CI: 0.2-0.5). Older age, Asian race, Hispanic ethnicity, atrophic gastritis, and anemia were associated with non-cardia GAC.

CONCLUSIONS

Routine EHR data can stratify the general population for GAC risk.

IMPACT

Such methods may help triage populations for targeted screening efforts, such as upper endoscopy.

Biases in Electronic Health Records Data for Generating Real-World Evidence: An Overview

Electronic Health Records (EHR) are increasingly being perceived as a unique source of data for clinical research as they provide unprecedentedly large volumes of real-time data from real-world settings. In this review of the secondary uses of EHR, we identify the anticipated breadth of opportunities, pointing out the data deficiencies and potential biases that are likely to limit the search for true causal relationships. This paper provides a comprehensive overview of the types of biases that arise along the pathways that generate real-world evidence and the sources of these biases. We distinguish between two levels in the production of EHR data where biases are likely to arise: (i) at the healthcare system level, where the principal source of bias resides in access to, and provision of, medical care, and in the acquisition and documentation of medical and administrative data; and (ii) at the research level, where biases arise from the processes of extracting, analyzing, and interpreting these data. Due to the plethora of biases, mainly in the form of selection and information bias, we conclude with advising extreme caution about making causal inferences based on secondary uses of EHRs.

Who benefits from green spaces? Surrounding greenness and incidence of cardiovascular disease in a population-based electronic medical records cohort in Madrid

The objective was to study the association between surrounding greenness and the incidence of cardiovascular diseases (CVD) with a four years follow-up in almost half a million high CVD-risk women and men, as well as its differential effect by area-level deprivation in Madrid. We analyzed 2015-2018 primary healthcare electronic medical records for 437,513 high CVD risk individuals representing more than 95% of the population of that age range residing in Madrid. The outcome variable was any cardiovascular event. We measured surrounding residence greenness at 200 m, 300 m, 500 m, and 1000 m through the Normalized Difference Vegetation Index (NDVI). We assessed socioeconomic deprivation through a census-based deprivation index. We estimated the 4-year relative risk of CVD by an increase in 0.1 units of NDVI and then stratified the models by quintiles of deprivation (Q5 the most deprived). We found that for every increase in 0.1 units of NDVI at 1000 m there was a 16% decrease in CVD risk (RR = 0.84 95% CI 0.75-0.94). CVD risk for the remaining distance exposures (at 200 m, 300 m, and 500 m) were none statistically significant. In general, the protective effect of green spaces was present in medium-deprivation areas and males, but the associations were inconsistent across deprivation levels. This study highlights the relevance of evaluating the interaction between physical and social urban components to further understand possible population prevention approaches for cardiovascular diseases. Future studies should focus on the mechanisms of context-specific interactions between social inequalities and green spaces' effects on health.