KEY POINTS

Question: Is female sex associated with increased in-hospital mortality after open cardiac valve surgery? Findings: Relative to male patients, female patients showed an association with overall increased rates and confounder-adjusted odds of in-hospital death after multiple types of open cardiac valve surgeries. Meaning: Female patients may be more likely to die in-hospital after cardiac valve surgery in the United States.

See Article, page 941

Cardiac valvular disease affects millions of people in the Western world and is a major cause of morbidity and mortality.1,2 The economic burden on the United States for heart valve disease is estimated at $23.4 billion each year, with nearly 156,000 valvular surgeries being performed annually.1,3 When patients become increasingly symptomatic despite optimal medical therapy and/or valve disease becomes severe, heart surgery (repair or replacement) is recommended by the American College of Cardiology/American Heart Association (ACC/AHA) guidelines.2,4 Recent literature has shown that female sex is associated with adverse outcomes after cardiac operations such as coronary artery bypass grafting (CABG) or isolated single valve surgeries.5 While some studies suggest hypotheses for this sex-based difference in outcomes, previous literature that has evaluated this association has only studied a particular valve type rather than all valve surgeries (eg, isolated mitral valve repair) or has looked at outcomes in relation to sex surrounding isolated transcatheter valve replacements.6–8 However, there is a lack of large multicenter, multiregional studies evaluating how sex affects outcomes after open valve surgeries overall.9

We assessed whether female sex is associated with increased rates of in-hospital mortality (primary outcome) after open-heart valve surgery utilizing inpatient hospital administrative data available through the Healthcare Cost and Utilization Project (HCUP) State Inpatient Database (SID) from the Agency for Healthcare Research and Quality (AHRQ). We hypothesized that female sex is associated with increased in-hospital mortality after open cardiac valve surgery.

METHODS Data Extraction

After institutional review board (IRB) approval by Weill Cornell Medicine (IRB 1308014181), utilizing the HCUP SID, we conducted a retrospective cohort study of patients undergoing open cardiac valve surgery from 2007 to 2018 in Washington, Maryland, Kentucky, and Florida; 2007 to 2011 in California; and 2007 to 2016 in New York. The HCUP SID is a longitudinal database that captures nearly 95% of hospital inpatient stays and discharge data by state.10 This database includes >100 variables such as primary and secondary diagnoses, patient demographics, payment source, length of stay (LOS), procedures, admission and discharge status, and hospital characteristics.10 Because HCUP SID data are deidentified, the requirement for written informed consent was waived by the Weill Cornell Medicine IRB.

Table 1. - Patient and Hospital Characteristics for Patients Undergoing Open Valve Surgery (In-Hospital Mortality) Variable Male (n = 164,511) Female (n = 108,443) P value Age in years (mean ± SD) 66.3 ± 13.8 67.6 ± 14.3 <.001 Payer  Medicare 95,259 (57.9%) 70,543 (65.1%) <.001  Medicaid 10,835 (6.59%) 9311 (8.59%)  Private 50,359 (30.6%) 24,847 (22.9%)  Other 3965 (2.41%) 1530 (1.41%)  Self-pay/no charge 4093 (2.49%) 2212 (2.04%) Race/ethnicity  White 125,092 (76.0%) 77,976 (71.9%) <.001  Black 8528 (5.18%) 8389 (7.74%)  Hispanic 13,229 (8.04%) 9860 (9.09%)  Other 11,820 (7.18%) 8675 (8.00%)  Missing 5842 (3.55%) 3543 (3.27%) State of hospital  CA 29,376 (17.9%) 18,891 (17.4%) <.001  FL 52,427 (31.9%) 33,654 (31.0%)  KY 10,336 (6.28%) 7491 (6.91%)  MD 10,440 (6.35%) 6920 (6.38%)  NY 45,358 (27.6%) 31,995 (29.5%)  WA 16,574 (10.1%) 9492 (8.75%) CBSAa  Non-CBSA (rural) 6094 (3.73%) 3810 (3.53%) .028  Micropolitan statistical area 9991 (6.11%) 6600 (6.12%)  Metropolitan statistical area 147,325 (90.2%) 97,446 (90.3%) Elixhauser comorbidities  Congestive heart failure 64,737 (39.4%) 46,839 (43.2%) <.001  Cardiac arrhythmia 54,830 (33.3%) 37,330 (34.4%) <.001  Valve disease 147,091 (89.4%) 98,817 (91.1%) <.001  Pulmonary circulatory disease 26,562 (16.1%) 24,510 (22.6%) <.001  Peripheral vascular disease 34,755 (21.1%) 16,872 (15.6%) <.001  Hypertension, uncomplicated 80,889 (49.2%) 55,383 (51.1%) <.001  Hypertension, complicated 31,894 (19.4%) 17,378 (16.0%) <.001  Paralysis 679 (0.41%) 374 (0.34%) .006  Other neurological disorder 4676 (2.84%) 3187 (2.94%) .143  Chronic pulmonary disease 46,816 (28.5%) 38,363 (35.4%) <.001  Diabetes, uncomplicated 32,034 (19.5%) 21,803 (20.1%) <.001  Diabetes, complicated 8943 (5.44%) 5485 (5.06%) <.001  Hypothyroidism 11,410 (6.94%) 20,953 (19.3%) <.001  Renal failure 29,892 (18.2%) 15,316 (14.1%) <.001  Liver disease 4795 (2.91%) 2647 (2.44%) <.001  Peptic ulcer disease 1040 (0.63%) 789 (0.73%) .003  HIV/AIDS 417 (0.25%) 132 (0.12%) <.001  Lymphoma 1140 (0.69%) 707 (0.65%) .21  Metastatic cancer 317 (0.19%) 204 (0.19%) .823  Solid tumor without metastasis 2177 (1.32%) 834 (0.77%) <.001  Rheumatoid arthritis/collagen vascular disease 2741 (1.67%) 5069 (4.67%) 0  Coagulopathy 16,040 (9.75%) 9681 (8.93%) <.001  Obesity 23,511 (14.3%) 19,881 (18.3%) <.001  Weight loss 4366 (2.65%) 3207 (2.96%) <.001  Fluid and electrolyte disorders 14,518 (8.82%) 10,673 (9.84%) <.001  Blood loss anemia 1021 (0.62%) 849 (0.78%) <.001  Deficiency anemia 2875 (1.75%) 2838 (2.62%) <.001  Alcohol abuse 6543 (3.98%) 1093 (1.01%) <.001  Drug abuse 4765 (2.90%) 2593 (2.39%) <.001  Pyschoses 655 (0.40%) 439 (0.40%) .811  Depression 9097 (5.53%) 10,873 (10.0%) <.001 Case type  Elective 84,400 (51.3%) 55,986 (51.6%) .01  Emergency/urgent 50,443 (30.7%) 33,393 (30.8%)  Other/unknown 29,668 (18.0%) 19,064 (17.6%) Year  2007 16,597 (10.1%) 11,220 (10.4%) <.001  2008 17,305 (10.5%) 11,681 (10.8%)  2009 18,006 (11.0%) 12,411 (11.4%)  2010 18,237 (11.1%) 12,285 (11.3%)  2011 18,609 (11.3%) 12,382 (11.4%)  2012 12,225 (7.43%) 8355 (7.70%)  2013 12,717 (7.73%) 8494 (7.83%)  2014 13,211 (8.03%) 8476 (7.82%)  2015 12,837 (7.80%) 8062 (7.43%)  2016 11,465 (6.97%) 7032 (6.48%)  2017 6612 (4.02%) 3999 (3.69%)  2018 6690 (4.07%) 4046 (3.73%) Hospital volume  First quartile 1345 (0.82%) 979 (0.90%) .062  Second quartile 12,393 (7.53%) 8309 (7.66%)  Third quartile 35,447 (21.5%) 23,314 (21.5%)  Fourth quartile 115,326 (70.1%) 75,841 (69.9%) Median household income state quartile for patient zip code  First quartile 34,283 (20.8%) 25,393 (23.4%) <.001  Second quartile 39,474 (24.0%) 26,894 (24.8%)  Third quartile 42,070 (25.6%) 27,571 (25.4%)  Fourth quartile 44,574 (27.1%) 26,253 (24.2%)  Missing 4110 (2.50%) 2332 (2.15%)

Pearson’s χ2 and ANOVA were conducted for comparison between groups. For nonparametrically distributed data, Kruskal-Wallis tests were conducted. All statistical tests were 2-sided.

Abbreviations: AIDS, acquired immunodeficiency syndrome; ANOVA, analysis of variance; CBSA, core base statistical area; HIV, human immunodeficiency virus; SD, standard deviation.

aCBSA divides counties into: metropolitan, micropolitan, and outside CBSAs (non-CBSAs). Metropolitan refers to counties with >50,000 residents. Micropolitan refers to counties of 10,000 to 49,999 residents. Metropolitan and micropolitan areas are considered areas that have a population nucleus and adjacent communities that have a high degree of integration with the CBSA. Non-CBSAs are often considered rural. Source: https://www.hcup-us.ahrq.gov/db/vars/siddistnote.jsp?var=pl_cbsa.

Open valve surgery was defined by specific International Classification of Diseases (ICD-9 and ICD-10) procedure modification codes (Supplemental Digital Content 1, Table 1a, https://links.lww.com/AA/D957; Supplemental Digital Content 2, Table 1b, https://links.lww.com/AA/D958). Variables collected included patient-specific demographic data, patient comorbidities present on admission (Table 1), diagnoses, procedure codes, hospital-specific data (surgical volume and state), information concerning postsurgical hospital course (including complications and total LOS), and disposition at discharge. ICD codes for complications are listed in Supplemental Digital Content 3, Table 2a, https://links.lww.com/AA/D959, and Supplemental Digital Content 4, Table 2b, https://links.lww.com/AA/D960.

Exclusion Criteria

Patients <18 years of age, patients with missing demographic data, and those undergoing transcatheter valve (nonopen valve) interventions were excluded from the study (Supplemental Digital Content 2, Table 1b, https://links.lww.com/AA/D958).

Statistical Analysis

The primary objective of this study was to estimate the confounder-adjusted association between sex and in-hospital mortality (as recorded and coded by HCUP SID) after open cardiac valve surgery. The secondary outcome was hospital LOS. To account for competing risk of mortality in the measure of LOS, we top-coded the values of LOS of any patient who died to be the equivalent of the longest LOS for patients in the population alive at discharge.

Bivariate analyses were conducted for our primary outcome measure. Pearson’s χ2 test was used for categorical variables, and analysis of variance (ANOVA) was used for parametric continuous variables. The Kruskal-Wallis test was used for continuous variables that violated assumptions of normality. We assessed the association between sex and our primary outcome of in-hospital mortality using a multivariable generalized estimating equation model to account for intrahospital practice patterns. We specified a logit link, binomial distribution of the outcome, and exchangeable correlation structure. The primary outcome reflected mortality across the entire cohort, regardless of procedure type. In addition, a series of exploratory multivariable models stratified by individual valve type subgroup were fitted for this outcome. This included aortic valve repair, aortic valve replacement, mitral valve repair, mitral valve replacement, surgery on multiple valves, other single valve surgery (pulmonic valve repair, pulmonic valve replacement, tricuspid repair, or tricuspid replacement), and valve surgery plus CABG surgery. Results from these stratified models were limited because of limitations in statistical power secondary to post hoc stratification; however, they represent trends in findings as corroborative evidence. We further assessed the interaction between sex and 3 baseline factors in separate models: sex with valve procedure type, sex with hospital state, and sex with valve procedure hospital volume in quartiles. The latter 2 models were intended to assess the interaction of hospital characteristics with sex. If the models with interaction terms had superior fits to the main model (based on a likelihood ratio test), a linear combination of coefficients was calculated in the categories in which the interaction term was significant to report the association between sex and mortality within levels of the interacting variable. Multivariable models included the following covariates: sex (our primary variable of interest), primary insurance payer (unordered categories: Medicare, Medicaid, private insurance, uninsured, and other types of insurance); race/ethnicity (unordered categories: White, Black, Hispanic, and other); state quartile of median income for the patient’s zip code of residence; age; 31 individual Elixhauser comorbidity measures (coded for the presence of disease comorbidity, yes/no)11; year of surgery; admission type (elective, emergency/urgent, and other/unknown); any complication during the inpatient stay; state of the hospital where the surgery was performed (Washington, Maryland, Kentucky, Florida, New York, and California); and hospital volume of open valve surgeries (in quartiles based on the distribution of volume in our sample).

The secondary outcome was hospital LOS. To account for competing risk of mortality in the measure of LOS, we top-coded the values of LOS of any patient who died to be the equivalent of the longest LOS for patients in the population alive at discharge. We fit a multivariable Cox proportional hazards model for LOS contained the aforementioned covariates. In this model, the outcome was time to discharge alive, and deaths were censored at the longest LOS of any living patient. This semiparametric regression model makes no assumptions for the distribution of failure times.

Finally, we conducted 2 sensitivity analyses. In the first, we estimated the average treatment effect of sex using inverse probability-weighted regression adjustment, in which the weights created were the estimated inverse probabilities of being female. Variables in the multivariable model to create the probability weights included all covariates in the main model, with sex specified as the outcome. We subsequently used these inverse probability weights in a weighted multivariable logistic regression model of in-hospital mortality.12 Next, we calculated the E value to show the minimum unmeasured confounding that would be required (with the treatment and outcome measures) to explain away the association of interest. In our study, the E value quantifies the magnitude of unobserved confounding that would be required to nullify the relationship between sex and in-hospital mortality.13

This article adheres to the applicable Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.14 All clinical outcomes and subgroup analyses were defined a priori at the initiation of the study design. All P values were 2-sided, and a P value of <.05 was used to indicate statistical significance in all tests. Results are reported as n (percentage), mean (standard deviation), median (interquartile range [IQR]), and adjusted odds ratio (aOR) with 95% confidence intervals (CIs). All statistical analyses were performed using SAS version 9.4 (SAS Institute) and Stata SE, version 16.

RESULTS Demographics

A total of 272,954 patients in 345 separate hospitals underwent open valve repair and/or replacement between 2007 and 2018 who met the inclusion criteria. Table 1 shows the demographics and clinical characteristics of the patient population by sex. The majority of the patients were men: 164,511 (60.27%). The mean age of male patients was 66.3 ± 13.8 years, and the mean age of female patients was 67.6 ± 14.3 years. Female patients in the sample population were primarily insured by Medicare (70,543; 65.1%) and predominantly White (77,976; 71.9%). Similarly, male patients in the sample population were primarily insured by Medicare (95,259; 57.9%) and predominantly White (125,092; 76%).

The data were also stratified by individual valve type, which included aortic valve replacement (N = 75,565), aortic valve repair (N = 4661), mitral valve replacement (N = 59,392), mitral valve repair (N = 20,461), surgery on multiple valves with repairs or replacements (N = 17,377), CABG plus valve surgery (N = 89,639), and other single valve surgeries (N = 5859). Other single valve surgery was defined as the tricuspid and pulmonic valves, as these valves on their own were not a large enough group to be analyzed (pulmonic valve repair [N = 95]; pulmonic valve replacement [N = 1798]; tricuspid valve repair [N = 1993]; tricuspid valve replacement [N = 2539]).

Outcomes

Unadjusted rates of in-hospital mortality and LOS are shown in Table 2. Of 272,954 patients in the cohort, there were a total of 11,793 deaths (4.32%), including 6282 male patients (3.82%) and 5511 female patients (5.08%). The median LOS for female patients was 9 days (IQR, 6–15), and the median LOS for male patients was 8 days (IQR, 6–13). Table 2 also shows the unadjusted outcomes of any complication and total charges.

Table 2. - Unadjusted Associations Between Sex and Outcomes Unadjusted outcome Male (n = 164,511) Female (n = 108,443) P value Death 6282 (3.82%) 5511 (5.08%) <.001 Any complication 87,386 (53.1%) 60,081 (55.4%) <.001 Total charge 2018 in US dollars 194,254 (123,146–307,575) 195,953 (124,190–312,614) <.001 LOS (IQR) 8.0 (6.0–13.0) 9.0 (6.0–15.0) <.001

Pearson’s χ2 and ANOVA were conducted for comparison between groups. For nonparametrically distributed data, Kruskal-Wallis tests were conducted. All statistical tests were 2-sided.

Abbreviations: ANOVA, analysis of variance; IQR, interquartile range; LOS, length of stay.

The confounder-aORs from the multivariable logistic regression models estimating the association between patient sex and in-hospital mortality are shown in Table 3. Our analysis found that female patients have an estimated 41% greater odds of in-hospital mortality after open cardiac valve surgery than male patients after adjusting for multiple patient- (demographics and comorbidities), hospital-, procedure-, and initial admission-related characteristics listed in the Methods section (aOR, 1.41; 95% CI, 1.35–1.47; P < .001).

Table 3. - Confounder-Adjusted Associations Between Sex and In-Hospital Mortality: Overall and by Procedure Population Sample size n (%) of in-hospital mortality within the analysis group aOR of female versus male (95% CI) P value for aOR of female versus male Overall population N = 272,954 11,793 (4.32) 1.41 (1.35–1.47) <.001 Overall population (IPTW: ATE weights) N = 272,954 11,793 (4.32) 1.42 (1.37–1.49) <.001 Aortic valve repair n = 4661 336 (7.21) 0.87 (0.67–1.14) .320 Aortic valve replacement n = 75,565 1954 (2.59) 1.38 (1.25–1.52) <.0001 Valve + CABG n = 89,639 5312 (5.93) 1.64 (1.54–1.74) <.0001 Mitral valve replacement n = 59, 392 2396 (4.03) 1.22 (1.12–1.34) <.0001 Mitral valve repair n = 20, 461 281 (1.37) 1.26 (0.98–1.64) .075 Multiple-valve surgeries n = 17,377 1257 (7.23) 1.38 (1.22–1.57) <.0001 Other single-valve surgery n = 5859 257 (4.39) 1.10 (0.82–1.46) .529 Models were adjusted for: primary insurance payer (unordered categories: Medicare, Medicaid, private insurance, uninsured, and other types of insurance); race/ethnicity (unordered categories: White, Black, Hispanic, and other); state quartile of median income for the patient’s zip code of residence; age; 31 individual Elixhauser comorbidity measures (coded for the presence of disease comorbidity yes/no)11; year of surgery; admission type (elective, emergency/urgent, and other/unknown); any complication; state of hospital where surgery was performed; and hospital volume of open-valve surgeries (in quartiles based on the distribution of volume in our sample). P value for interaction term of sex and procedure type (a joint Wald test)

Abbreviations: aOR, adjusted odds ratio; ATE, average treatment effect; CABG, coronary artery bypass grafting; CI, confidence interval; IPTW, inverse probability of treatment weighting; OR, odds ratio.

In addition to estimating the association between sex and mortality for the entire cohort of valve surgery patients, we performed stratified analyses of this association based on the type of valve operation (eg, aortic valve and mitral valve replacements and repairs). The Figure and Table 3 present the confounder aORs for the overall population and each type of valve surgery; the latter also reports the percentage of valve surgeries performed based on the total analysis group. The effect of sex on in-hospital mortality was significantly different between procedure types (P < .001). After stratification by type of valve surgery, increased estimated odds of inpatient mortality were found for (P < .001) female patients undergoing isolated aortic valve replacement, multiple valve surgery, isolated mitral valve replacement, and valve surgery with CABG (all P < .001; Table 3; Figure). Female patients did not have increased odds of in-hospital mortality when stratified by valve type for isolated mitral valve repair, isolated aortic valve repair, or any other isolated valve repair (Table 3; Figure).

Figure.:

Confounder-adjusted associations between sex and in-hospital mortality: overall and by procedure forest plot (adjusted odds ratios; 95% confidence intervals). CABG indicates coronary artery bypass grafting.

A likelihood ratio test showed that a model containing an interaction term between sex and surgery type was a significantly better fit than the main model at predicting in-hospital mortality. Female patients undergoing aortic valve replacement were less likely than male patients undergoing valve and CABG surgeries during the same inpatient stay to die in the hospital (aOR, 0.78; 95% CI, 0.72–0.84; P < .001). Conversely, female patients undergoing multivalve surgery were more likely than male patients undergoing valve and CABG surgeries to die in the hospital (aOR, 1.67; 95% CI, 1.53–1.84; P < .001). No statistically significant difference for an interaction term was found between female and male patients for aortic valve repair, mitral valve replacement, or other isolated valve surgeries using this same model (Supplemental Digital Content 5, Table 3, https://links.lww.com/AA/D961).

Results of a sensitivity analysis using inverse probability of treatment weights for regression adjustment produced similar results compared to the main model: the aOR for in-hospital mortality for female patients compared to male patients was 1.42 (95% CI, 1.37–1.49; P < .001), compared to 1.41 (95% CI, 1.35–1.47; P < .001) in the main model (Table 3). The E value estimate for the primary outcome was 2.17 (lower limit of the 95% CI, 2.04). Therefore, an unmeasured confounder would have to be associated with both sex and in-hospital mortality, with a risk ratio ≥2.17 for the observed association between female sex and in-hospital mortality to be fully explained away.

DISCUSSION

Our results show that compared to male patients, female open cardiac valve surgical patients had increased unadjusted rate and confounder-adjusted odds of in-hospital mortality after surgery. Our findings were statistically significant after controlling for multiple patient- (demographics and comorbidities), hospital-, procedure-, and admission-related factors. Stratified analyses confirmed increased mortality odds in female patients across multiple valve surgery subtypes, including aortic and mitral valve replacements and combined CABG-valve procedures, suggesting that excess mortality risk extends across numerous female valve surgical populations. In light of the improvements in cardiac surgery outcomes in recent decades,15 it is particularly notable that the adjusted in-hospital mortality risk for open valve surgery is calculated to be >30% greater for female patients than male patients, driven largely by open valve replacement surgeries. However, at subgroup analyses, the association of sex was not consistent in all surgical subcategories. Female patients undergoing valve repair rather than valve replacement were not found to have significantly higher in-hospital mortality. Given that millions of female patients with severe valvular disease will need open cardiac valve surgery in the coming years,1 further understanding and optimization of outcomes in female patients after open cardiac surgery are important.

The findings of this study align with several previous studies showing that female patients have worse outcomes after cardiac surgery and specific types of open valve surgery.16–19 In 2019, Giustino et al18 found that female patients with severe ischemic mitral regurgitation experienced a higher all-cause mortality (27.1% female all-cause mortality versus 17.4% male all-cause mortality; P = .03) and worse quality of life (QOL) at 2 years (lower mean European QOL-5 dimensions score: 69.9 ± 18.9 female versus 74.6 ± 17.2 male; P = .03) after mitral valve surgery using the Cardiothoracic Surgical Trials Network Severe Ischemic Mitral Regurgitation (CTSN SIMR) trial database. Similarly, in 2017, a study on isolated surgical aortic valve replacement in severe aortic stenosis utilizing an administrative claims database over 12 years, sampling 166,809 patients, showed a statistically significant 20% increase in confounder-adjusted odds of in-hospital mortality in female compared to male patients.19 In contrast, a smaller single-institution study in Germany evaluating isolated tricuspid valve surgery of 92 patients did not find a difference in surgical outcomes based on sex.20 Similarly, a study with 5582 patients from the Netherlands again found no statistical difference in in-hospital mortality based on sex for isolated or concomitant tricuspid valve surgery.21 While several studies, such as those mentioned above, have focused on various subtypes of cardiac surgery such as CABGs, isolated valve (eg, aortic valve alone) repair or replacement, or percutaneous interventions,19,22,23 our study is among the largest in scope to examine the association between patient sex and in-hospital mortality risk across multiple categories of open cardiac valve surgery.

Our results were largely driven by differences in in-hospital mortality between sex in aortic and mitral valve replacements, valve with CABG surgeries, and multiple valve surgeries; in the smaller populations of in mitral repairs, aortic valve repairs, or other single valve surgery (11.36% of total), there was no difference in mortality. The CIs are wide, suggesting a lack of precision; future large-scale studies should be performed in this population to understand the association between sex and outcomes after isolated valve repair.

Strengths of this study included a large sample size and high statistical power, allowing for multiple adjustments for known confounders and subgroup analyses stratified by valve surgery type. Additionally, the timeline of data collect