Sex-Based Differences in the Epidemiology, Clinical Characteristics, and Outcomes Associated with Worsening Heart Failure Events in a Learning Health System.
Written by: Chloe J LeungAnkeet S BhattAlan S GoRishi V ParikhElisha A GarciaDeborah LowAmanda R AllenJesse K FitzpatrickSirtaz AdatyaDana R SaxParag GoyalAnubodh S VarshneyAlexander T SandhuShanshan E GustafsonAndrew P Ambrosy
Find these authors on Roon
AB
Dr. Ankeet Bhatt
General Cardiology · Kaiser Permanente
SV
Dr. Sunny Varshney
Advanced Heart Failure · Stanford
Introduction
Heart failure (HF), a primary contributor to cardiovascular morbidity and mortality, is a leading cause of hospitalization and other healthcare utilization in the United States and worldwide (1). Patients may present with worsening HF (WHF) events (i.e., defined as a change in signs and symptoms necessitating urgent and/or emergent HF-related therapy) across a variety of diverse clinical settings, including hospitalizations, emergency department (ED) visits or observation stays, and in outpatient settings (2–4). Signs and symptoms of HF may manifest differently in women and men due to several factors associated with sex, including the prevalence of comorbid conditions (i.e., diabetes mellitus, obesity, hypertension), genetics, and reproductive factors (5). In addition, certain forms of HF, such as HF with preserved ejection fraction (HFpEF), may be more common among women than men (5). These differences and others may lead to variations in the population prevalence, presenting clinical setting, and outcomes among women and men experiencing WHF events; however, contemporary, real-world data examining sex-based differences in WHF have been generally lacking (6, 7).It was reported previously that validated rule-based natural language processing (NLP) algorithms can accurately identify WHF events and may better reflect the population burden of WHF hospitalizations as compared with diagnostic coding (8). Encounter types outside of traditional hospitalizations (i.e., ED visits, observation stays, and outpatient encounters) account for nearly half of all WHF events (9). However, little is known about how these patterns differ based on sex. We describe sex-based differences in the epidemiology, clinical characteristics, and outcomes associated with WHF across clinical settings (i.e., hospitalizations, ED visits/observation stays, and outpatient encounters) within a large learning health system.
Methods
Kaiser Permanente Northern California (KPNC) is a large integrated health care delivery system that currently offers comprehensive care (i.e., inpatient, ED, and outpatient encounters) to >4.6 million members through 21 hospitals and >260 freestanding clinics. KPNC membership is highly representative of local and statewide populations in terms of age, sex, race and ethnicity, and socioeconomic status (10–12). This study was approved by the KPNC institutional review board, and a waiver of consent was obtained due to the retrospective nature of this study.The study cohort included all adult (age ≥18 years) KPNC members with HF between January 1st, 2010 and December 31st, 2019. We created 10 calendar year cohorts of adults with prevalent HF on January 1st of each year (i.e., the index date), defined as having a prior hospitalization with a principal discharge diagnosis of HF and/or ≥3 outpatient visits with the following and codes (ICD-9: 398.91, 402.x1, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93 and 428.x and ICD-10: I09.81, I11.0, I11.9, I13.0, I13.1, I13.10, I13.11, I13.2, I50, I50.1, I50.2, I50.20, I50.21, I50.22, I50.23, I50.3, I50.30, I50.31, I50.32, I50.33, I50.4, I50.40, I50.41, I50.42, I50.43, I50.9, I97.13) at any time within 5 years prior to each index date. These codes have been validated by multiple health care delivery systems and demonstrated a positive predictive value of ≥95% (13, 14).
International Classification of Diseases, 9th (ICD-9)
10th (ICD-10) edition
Patients with end-stage kidney disease (i.e., defined as receipt of chronic dialysis or kidney transplant), stage D heart failure (i.e., defined as receipt of left-ventricular assistive device or heart transplant), less than 12 months of health plan membership prior to the index date, and unknown sex were excluded from the study.
Patients in each calendar year cohort were followed until December 31st of that year. Eligible patients remained represented in consecutive calendar year cohorts unless censored due to health plan disenrollment or death. Death was ascertained using health plan administrative and clinical databases, member proxy reporting, Social Security Administration vital status files, and state death certificate information.
Qualifying clinical encounters included all hospitalizations (i.e., defined as admissions lasting >24 hours), ED visits including observation stays, and outpatient encounters (i.e., defined as an urgent care visit or a clinical appointment with a primary care provider or a cardiologist) with a diagnosis code for HF. To specifically identify clinical encounters for WHF, we used an operational definition including ≥1 qualifying clinical encounter, ≥1 symptom (i.e., dyspnea, orthopnea, paroxysmal nocturnal dyspnea, fatigue, weight gain, and/or tachypnea), ≥2 objective findings (i.e., tachycardia [heart rate >100 bpm], elevated BNP [≥100 ng/L], and/or chest x-ray findings [pulmonary edema, pleural effusion, and/or cardiomegaly]) including ≥1 sign (i.e., lower extremity edema, pulmonary rales/wheezing, jugular venous distension, third heart sound [S3 gallop], hepatomegaly, and/or abdominal swelling), and new administration of intravenous loop diuretics (i.e., ≥2 doses if hospitalized or ≥1 dose if non-hospitalized) and/or new hemodialysis/continuous renal replacement therapy. For outpatient encounters, we also defined a change in HF-related therapy as either new initiation and/or augmentation of oral diuretics leveraging both structured (i.e., pharmacy dispensing data) and unstructured (i.e., written provider documentation) data elements. These diagnostic criteria are based on a standardized definition of inpatient and outpatient WHF previously developed and validated by a consensus panel of trialists with expertise in clinical endpoint classification in collaboration with the U.S. Food and Drug Administration (15). This multidimensional definition is specific to WHF and has been shown to be accurate and reproducible in multiple pivotal trials of investigational drugs and devices seeking regulatory approval (16–20).
Rules-based NLP queries were developed using Linguamatics i2E (IQVIA, Cambridge, U.K.) for each diagnostic criterion using custom ontologies, and queries were derived and validated in KPNC against a “gold standard” consisting of manual records review and validation of >500 cases of suspected WHF across clinical settings by two physicians, with final adjudication by a cardiologist if discrepancies existed. The NLP algorithm achieved high accuracy (>95%) for WHF (8).
The KPNC Epic-based EHR system was the primary source of hospitalization data, provider progress notes, patient demographic information (i.e., age and self-reported ethnicity), and cardiac imaging reports (i.e., echocardiograms and chest X-rays). Data on inpatient medications, laboratory values, vital signs, and procedures were also obtained from EHR databases. To supplement these records, the KPNC Virtual Data Warehouse was also used to ascertain concurrent comorbidities, outpatient medications, and outpatient laboratory values. Comorbid conditions were noted within 5 years of the index date. Baseline laboratory results were defined as the most recent outpatient, non-emergency values within 7-365 days before the index date.
Data on left ventricular ejection fraction (LVEF), was assessed using the most recent value within 2 years of each index date from structured results of echocardiograms, radionuclide scintigraphy, other nuclear imaging modalities, and left ventriculography, or extracted from echocardiogram reports with rules-based NLP algorithms. Patients were grouped by LVEF, classifying them as having HF with reduced EF (HFrEF, defined as <40%), HF with mildly reduced EF (HFmrEF, defined as 40-49%), or HF with preserved EF (HFpEF, defined as ≥50%) congruent with updated U.S. and European guidelines for HF management (21, 22).
Analyses were performed using SAS software version 9.4 (Cary, N.C.) at KPNC’s Division of Research (Oakland, CA). This report follows the guidelines for reporting observational studies as outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement (https://www.equator-network.org/reporting-guidelines/strobe/).
We first described characteristics of all unique patients at their first calendar year cohort. We compared differences across sex strata using standardized differences between groups using Cohen’s D for continuous and binary variables and Cramer’s V for categorical variables. Differences were considered significant if the effect size estimate was ≥0.10, indicating a 10% difference between the groups. To distinguish discrete WHF episodes during follow-up within each year, we applied standardized rules to collapse immediately consecutive encounters (i.e., ED visit followed by hospitalization for WHF). ED visits/observation stays for WHF occurring within 3 days before hospitalization or after discharge were removed and indicated as a hospitalization for WHF. Outpatient WHF encounters within 3 days before or after the remaining ED visits/observation stays were indicated as ED/observation stays; consecutive outpatient WHF encounters occurring within 3 days of each other were collapsed.
With the resulting discrete episodes of inpatient, ED/observation, and outpatient WHF episodes of care, we calculated annual incidence rates of WHF episodes per 100 person-years by encounter type and sex. Temporal trends were evaluated using a calendar year term in Poisson models with a robust variance estimator accounting for repeated individuals across calendar years. We also estimated the crude risks of subsequent events within 30 days following each WHF episode, including all-cause hospitalizations, all-cause ED visits/observation stays, hospitalizations for WHF, ED visits/observation stays for WHF, and death, stratified by sex and initial WHF encounter type. We examined the association between sex at the index encounter and 30-day outcomes using nested logistic regression models sequentially adjusted for sociodemographics (i.e., age and race and/or ethnicity), LVEF, comorbidity index (23), vital signs (i.e., heart rate and systolic blood pressure), renal function (i.e., serum creatinine), and medications (i.e., beta-blocker, renin-angiotensin-system inhibitors [RASi], and mineralocorticoid receptor antagonists [MRAs]). A generalized estimating equations approach was used to account for repeated observations (i.e., index encounters) among individuals.
Results
From January 1st, 2010 to December 31st, 2019, we identified 1,122,368 discrete clinical encounters with a diagnosis code for HF including 167,219 (15%) hospitalizations, 221,601 (20%) ED visits/observation stays, and 733,548 (65%) outpatient encounters (Figure 1). A total of 124,479 (11.1%) clinical encounters met the standardized consensus definition for WHF, occurring in 102,116 unique patients of whom 48,543 (47.5%) were women and 53,573 (52.5%) were men (Table 1). Women were older compared to men (76±13 years vs. 72±14; standardized difference: 0.28). The cohort consisted of 63.7% White, 9.7% Black, and 12.4% Hispanic patients; distribution of race and ethnicity was similar between men and women. Cardiovascular comorbidities, including the proportion of patients with hypertension and diabetes mellitus, and mean body mass index were also similar by sex. Frequent presenting signs and symptoms included shortness of breath, rales, and lower extremity edema. Additional clinical characteristics stratified by sex are listed in Table 1.
Women had a higher mean LVEF compared to men (55±13% vs. 49±14%, standardized difference: 0.44) at the time of presenting WHF encounter. Women were less likely to have HFrEF<40% (11% vs. 20%) and HFmrEF 40-49% (9% vs. 14%) and more likely to have HFpEF ≥50% (56% vs. 43%) compared with men (Table 1). Presenting clinical setting for WHF was similar among women and men: hospitalizations (36.8% vs. 37.7%), ED visits or observation stays (11.8% vs. 13.4%), and outpatient encounters (4.4% vs. 4.9%) (Figure 1, Supplemental Table 1).
The incidence of WHF events increased over time, from 25 events per 100 person-years in 2010 to 33 events per 100 person-years in 2019. Women and men experienced similar increases in the annual incidence of any WHF episode over the study period (women: 26 to 34 events per 100 person-years; men: 25 to 33 events per 100 person-years; Figure 2). Increases were observed across all clinical settings. Temporal trends in the annual incidence of WHF in men and women were also similar across LVEF strata (Supplemental Figure 1, Supplemental Figure 2).
Following an index WHF event, 30-day all-cause hospitalizations across all presenting clinical settings ranged from 13.7%-20.8% and 30-day all-cause ED visits ranged from 10.7-19.2%. All-cause and HF-specific healthcare utilization rates and mortality by index WHF event are shown in Figure 3. Compared with men, women had similar or modestly lower crude risks of all-cause hospitalization following an index ED visit/observation stay (15.5% vs. 17.0%) or outpatient encounter (11.0% vs. 13.6%). Women also had similar or modestly lower crude risks of an all-cause ED visit (12.6% vs. 13.4%), ED visit/observation stay for WHF (17.8% vs. 19.4%), or outpatient encounters (9.0% vs. 10.3%) for WHF following an index hospitalization. Crude mortality rates were also modestly lower in women compared to men after an index WHF hospitalization (11.9% vs. 13.9%) or an ED visit/observation stay (3.9% vs. 4.7%) for WHF. However, after adjustment for sociodemographics, vital signs, laboratory values, systolic function, and guideline-directed medical therapy for HF, odds of all-cause or HF hospitalizations and ED visits within 30 days of a WHF encounter were similar between men and women (p-value >0.05 for all comparisons) (Figure 4, Supplemental Table 2,). Adjusted odds of 30-day death remained lower in women vs. men following an index hospitalization (adjusted odds ratio [aOR] 0.88, 95% Confidence Interval 0.93) or ED visit/observation stay (aOR 0.86, 95% CI: 0.75, 0.98) for WHF but not for outpatient encounters (aOR 1.02, 95% CI: 0.86, 1.20).
Discussion
We systematically analyzed sex-related differences in the epidemiology, clinical characteristics, and outcomes associated with WHF events including >1 million encounters over a 10-year period in a real-world, diverse, and community-based cohort. Although women and men were different in terms of sociodemographics and clinical characteristics, we found that the proportions of hospitalizations, ED visits, and outpatient encounters for WHF were comparable by sex using validated NLP-based algorithms. Furthermore, the incidence of WHF events increased substantially across clinical settings and to a similar extent over time in both women and men. However, following an index WHF event, women had lower rates of healthcare utilization and mortality within 30 days, and the protective association with survival persisted even after comprehensively adjusting for observed differences in baseline clinical characteristics.
In the present analysis, we found that the proportion of patients with a WHF event with HFpEF was approximately 10-15% higher in women compared to men. HFpEF is a heterogeneous syndrome with a distinctly different underlying pathophysiology, clinical presentation, and response to therapy compared to HFrEF (24). For this reason, it was our a priori hypothesis that women and men might have differing prevalence of signs and symptoms of HF. However, using rule-based NLP, we found that among patients with a WHF event, the distribution of individual signs and symptoms of volume overload was highly similar. This contrasts with other clinical conditions (i.e., acute myocardial infarction) where women present distinctly differently than men and healthcare providers may be less likely to attribute their prodromal symptoms to heart disease and make a timely and accurate diagnosis (25). It has previously been reported that a more holistic NLP-based approach disproportionately increased the perceived burden of hospitalizations for WHF among those with HFpEF compared to the primary discharge diagnosis alone (5). Thus, it is worth highlighting that we did not redemonstrate this finding when we compared NLP to diagnostic coding for WHF hospitalizations in women versus men.
Estimates of the lifetime risk of HF have ranged from 20-50% and are similar among women and men with the same racial and ethnic background (26). The present analysis extends the existing evidence-base by examining sex-related differences in surrogates of morbidity in a contemporary, diverse, and community-based setting. Although it is well-established that hospitalization for HF is a sentinel event in the natural history of the condition, it was previously reported that approximately 50% of WHF events occur in a non-hospitalized setting (i.e., ED visits and outpatient encounters) and carry a similarly dismal prognosis (9). Given the delayed age of onset and the higher prevalence of HFpEF among women, which is predominantly managed in the outpatient setting (24), we theorized that the rate of hospitalizations for WHF might be lower for women and the rate of ED visits and outpatient encounters for WHF might be higher. However, we did not document any sex-related disparities in the incidence of WHF overall nor by clinical encounter type (i.e., hospitalizations vs. ED visits vs. outpatient encounters). Hence, although men may be at heightened risk for incident HF starting at a younger age, women with prevalent HF have a comparable risk of morbidity and should not be falsely labelled as “low risk” with respect to WHF events.
Although women on average were approximately four years older than men, we found that women had marginally lower 30-day mortality rates following an index hospitalization or ED visit/observation stay for WHF compared to men. This is a somewhat paradoxical observation as age is known to be one of the strongest predictors of future morbidity and mortality. These findings are likely explained, at least in part, by men being more likely to have an underlying ischemic etiology, as evidenced by higher rates of coronary revascularization, and a lower LVEF, both of which are well-known to be associated with an overall worse prognosis (27–31). However, women experienced better short-term survival even after comprehensively adjusting for sociodemographics, vital signs, laboratory values, systolic function, and receipt of guideline-directed medical therapy for HF. Future research is necessary to fully elucidate the protective association of female sex with mortality following an index WHF event, which may be attributable to other clinical, social, behavioral, and/or biological factors not captured by EHR data.
Our study has several limitations. First, although we applied a rigorous and standardized definition for WHF based on expert consensus, FDA guidance, and previous study, it was not feasible to incorporate several potential therapeutic interventions for WHF including IV vasoactive medications, temporary mechanical circulatory support, and/or renal replacement therapy. These treatments were inconsistently documented and/or lacked diagnostic specificity. Second, we collapsed and prioritized clinical encounters occurring within the same 72-hour timeframe to identify unique WHF events to best resolve lack of agreement surrounding what constitutes discrete episodes of care. Third, intensification of oral diuretics in the outpatient setting may be communicated directly to patients and not necessarily lead to a prescription change and/or be clearly documented in the EHR. We used NLP to review all provider documentation for newly prescribed and/or augmented oral diuretics, but this challenge likely limited the sensitivity of algorithms for identifying outpatient WHF. Fourth, a subset of clinical encounters may have occurred at non-KPNC facilities with limited documentation access in the KPNC EHR. However, prior studies have shown that >95% of events are captured by the KPNC EHR due to the exclusive relationship between the health plan, members, and providers at KPNC (13, 32, 33). Finally, because this is an observational EHR-based study, our results may be susceptible to selection bias arising from socially-produced sex or gender disparities in access to care, care-seeking behavior, provider diagnosis, and practice patterns that we were unable to account for or measure in this analysis.
Conclusions
In conclusion, using validated NLP-based algorithms to analyze state-of-the-art EHR data, we found few sex-based differences in the epidemiology of WHF within a real-world, diverse, and community-based cohort. However, following an index WHF event, women had lower unadjusted rates of healthcare utilization and mortality and following comprehensive adjustment, a persistent protective association of female sex was seen on survival. Future research is necessary to elucidate the underlying social, behavioral, and/or biological factors contributing to the observed sex-related differences to provide more personalized care for women and men living with HF.
SV
Dr. Sunny Varshney
Cardiology · Advanced Heart Failure & TransplantationStanford MedicinesunnyStanford, CA
