Percentage of patients with infection by sex, type of admission, and length of stay.
Percentage of deaths attributable to female sex by length of stay and by adjustment for infection (weighted sum adjustment of attributable risk with case load weights).
Percentage of patients with infection by sex and by infection site.
Rogers MAM, Langa KM, Kim C, Nallamothu BK, McMahon LF, Malani PN, Fries BE, Kaufman SR, Saint S. Contribution of Infection to Increased Mortality in Women After Cardiac Surgery. Arch Intern Med. 2006;166(4):437-443. doi:10.1001/archinte.166.4.437
Women have higher mortality rates after coronary artery bypass graft (CABG) surgery compared with men. Explanations for this sex difference are controversial. The objective of this study was to assess whether infection contributes to the increased risk of mortality in women.
We conducted a cohort study of 9218 Michigan Medicare beneficiaries hospitalized for CABG surgery. The prevalence of infection at any site during hospitalization was determined. Patients were followed up for 100 days after surgery to assess vital status. Analyses were conducted using proportional hazards regression and population attributable risk.
Women hospitalized for CABG surgery were more likely to have an infection than men (16.1% vs 9.8%, P<.001), regardless of age, race, type of admission, hospital volume, or presence of comorbidities. Infections of the respiratory tract, urinary tract, digestive tract, and skin and subcutaneous tissue were more common in women than in men. The risk of death in men increased 3-fold with infection, whereas the risk in women increased 1.8-fold. The interaction between infection and sex on mortality was significant after adjusting for age, type of admission, and presence of comorbidities (P = .008). The unadjusted percentage of deaths attributable to female sex was 13.9%, which decreased to 0.3% when adjusted for infection. Of the excess deaths in women, 96% could be accounted for by the differential distribution of infection between the sexes.
The increased risk of mortality after CABG surgery in women may be explained by underlying differences in the prevalence of infection among men and women.
Coronary artery bypass graft (CABG) surgery is one of the most common surgical procedures in the United States, with more than half a million procedures performed annually.1 During hospitalization, women have a greater risk of death than men.2 Several explanations have been given for the higher mortality in women, including structural differences in artery size and greater prevalence of preoperative comorbidities among women.3- 5 Given the discrepant findings across populations, however, this issue remains controversial.
Infection has not been previously considered as an explanation for this sex effect in population-based studies, to our knowledge. Yet, chronic infections are associated with atherogenesis, the severity of coronary artery disease correlates with pathogen burden, and infection increases the risk of mortality in cardiac patients.6- 11 We hypothesized that a portion of the sex effect in mortality following CABG surgery may be explained by underlying differences in the prevalence and consequences of infection among men and women.
Subjects were Michigan residents, 65 years or older, who were Part A Medicare beneficiaries and who underwent CABG surgery (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes 36.10-36.19). A 100% sample of the Inpatient Standard Analytical File from the Centers for Medicare and Medicaid Services for 1997-1998 was obtained for Michigan residents, regardless of the location of the hospital where the surgery was performed. A retrospective cohort design was used, with a follow-up period of 100 days after the CABG procedure date. To accommodate the follow-up period and the 6-month period before hospitalization to assess comorbidities, patients with CABG procedures between July 1, 1997, and September 22, 1998, formed the cohort. In 16 patients who underwent 2 CABG procedures during this period, the first surgery was used for analysis. The Centers for Medicare and Medicaid Services provided written permission for use of the Medicare data files, and the study received institutional review board approval.
The prevalence of infection during the hospital stay was obtained using the ICD-9-CM codes within the Inpatient Standard Analytical File, including the primary diagnosis, up to 8 secondary diagnoses, and 1 injury code. All codes indicating infection were included, regardless of infection site. The codes listed in Table 1 are indicative only of those infection-related codes present in the subjects and are not an exhaustive list of all possible infection codes within the ICD-9-CM system. Information regarding time of onset of infection is unavailable in the Centers for Medicare and Medicaid Services files; therefore, infections that were present at the time of admission and those that occurred after admission were included. This measure is an approximation of period prevalence; it captures newly occurring and existing infections during the hospitalization period.
The outcome of interest was cumulative mortality during the 100-day period after the date of the CABG procedure, with the date of death determined from the Centers for Medicare and Medicaid Services denominator file. In-hospital mortality was also calculated, as well as 30-day cumulative mortality.
Information was obtained regarding sex, age, race, type of admission (elective, urgent, or emergency), type of hospital room (private room only, 2 persons per room maximum during stay, or ≥3 persons per room at any time during stay), hospital location (Michigan or out of state), length of stay (LOS) during hospitalization (in days), hospital volume (number of CABG procedures performed annually during the study period), surgeon Medicare volume (number of CABG procedures recorded for a given surgeon among Part A Medicare Michigan beneficiaries during the study period), and recent skilled care (hospitalization or skilled nursing facility stay ≤180 days before hospitalization).
To assess the prevalence of comorbid conditions among the patients, the method by Elixhauser et al12 was used. This method, developed for large administrative data sets, allows construction of 30 comorbidity measures using ICD-9-CM codes and has been shown to predict mortality in hospitalized patients.13 Information was obtained for diagnoses listed during the hospitalization for the CABG procedure and for any other hospitalizations during the 6-month period before this stay.
There were 4 main hypotheses tested in this study. (1) There are sex differences in the prevalence of infection among patients hospitalized for cardiac surgery. (2) Hospitalized patients with an infection are more likely to die than patients without an infection. (3) The strength of the association between infection and mortality is different in men and women. (4) The increased mortality in women after CABG surgery may be partially explained by differences in infection across sex.
The analyses proceeded in 3 phases. The first phase was an investigation of the association between sex and infection. The relative risk of infection for women compared with men was calculated, with 95% confidence intervals (CIs). Extended Mantel-Haenszel tests were used to assess differences in infection across age, LOS, and ordinal categorical variables, while χ2 tests were used to assess differences in nominal categorical data, with α = .05 (2-sided). The second phase was an examination of the relation between infection and mortality. A Cox proportional hazards regression model was used for the determination of adjusted hazard ratios, with clustering by hospital and robust calculation of the variance-covariance matrix.14 The third phase was an assessment of the mortality risk (crude and adjusted for infection) attributable to female sex. The population attributable risk of mortality due to female sex was calculated, as was the number of excess deaths in women.15 Adjustment of attributable risk was conducted using the weighted sum approach, with the proportion of individuals who died in each stratum serving as weights (case load method).15 This yielded an asymptotically unbiased estimator of attributable risk and controlled for underlying differences due to confounding, effect modification, or both. Asymptotic variance of the adjusted attributable risk for cohort studies was used to calculate 95% CIs.16 Attributable risks were stratified by LOS, by presence or absence of 1 or more of the Elixhauser comorbidities, as well as for specific comorbidities (diabetes mellitus, renal failure, chronic pulmonary disease, protein-calorie malnutrition, fluid and electrolyte disorder, deficiency anemia, paralysis, or other neurological disease) that were a priori suspected of increasing the risk of infection.
There were 9218 Michigan Medicare patients who underwent a CABG procedure during the study period, 5749 men (62.4%) and 3469 women (37.6%). Of these patients, 1122 (12.2%) had an infection during their hospitalization for cardiac surgery, including 34 (0.4%) with an admission diagnosis of infection. Most patients were not black and received care in state, with almost half admitted following recent skilled care (Table 2).
Women were significantly more likely to be infected than men (16.1% vs 9.8%, P<.001). The sex-specific prevalence of infection is shown in Figure 1, stratified by LOS. Although the prevalence of infection increased with age, urgency of admission, and LOS (P<.001 for all), the excess likelihood of infection among women persisted regardless of age, race, type of admission, LOS, hospital and surgical volume, recent skilled care, and presence of various comorbidities (Table 3). The proportion of women infected was greater than the proportion of men infected for each of the Elixhauser comorbidities, with one exception: the prevalence of infection among patients with rheumatoid arthritis or collagen vascular diseases was 10.5% for women and 18.7% for men.
In-hospital mortality was 11.9% in patients with an infection compared with 4.0% in patients without an infection (relative risk, 3.0; 95% CI, 2.5-3.6), while 30-day mortality was 9.2% vs 4.5% (relative risk, 2.0; 95% CI, 1.6-2.5). Mortality within the 100-day period after CABG surgery was 16.5% in those with an infection compared with 6.2% in those without an infection (relative risk, 2.7; 95% CI, 2.3-3.1). This relation, however, was stronger in men than in women (Table 4). Men with an infection had a 3-fold (95% CI, 2.2-3.9) higher mortality risk compared with men without an infection, whereas women with an infection had a 1.8-fold (95% CI, 1.5-2.2) higher mortality risk compared with women without an infection. The interaction between infection and sex remained significant after adjusting for age, type of admission, and presence of comorbidities (P = .008). This interaction was evident during stratification; mortality was 17.2% in men who were infected, 15.8% in women who were infected, 5.2% in men who were not infected, and 7.9% in women who were not infected.
Mortality risk attributable to female sex within the 100-day period after CABG surgery was 13.9% (95% CI, 8.1%-19.6%) and decreased to 0.3% (95% CI, −5.8% to 6.3%) after adjustment for infection (Table 5). The unadjusted excess number of deaths due to female sex in this population was 95, which decreased to 4 after adjustment for infection. Therefore, 96% of the sex effect on 100-day mortality could be explained by the underlying differences in infection in this population. Additional adjustment for age altered the attributable risks, but to a lesser degree than infection alone. Stratification by comorbidities yielded similar results.
Because the likelihood of infection increased with LOS, attributable risk was calculated for patients based on LOS (Figure 2). In each stratum, the attributable risk due to sex decreased after adjustment for infection. For patients who were discharged less than 5 days after admission, the attributable risk declined from 20.1% to 10.4% after adjustment for infection. For patients with longer LOS (5-8, 9-12, or >12 days), female sex was not a risk factor for mortality after adjustment for infection.
For patients who underwent elective surgery only, the unadjusted mortality risk attributable to female sex was 16.3% and declined to 0.3% after adjustment for infection. For patients with urgent admissions, the attributable risk decreased from 14.6% to 4.3% after adjustment for infection. For patients with emergency admissions, it decreased from 6.0% to −4.1% after adjustment for infection.
Respiratory tract infections were most common, occurring in 7.4% of patients, followed by urinary tract infections (4.7%) and digestive tract infections (2.2%). There were differences between men and women by type of infection (Figure 3). Respiratory tract infections, urinary tract infections, and digestive tract infections were more common in women than in men (P<.001, P<.001, and P = .006, respectively). Skin and subcutaneous tissue infections were also more common in women than in men (1.3% vs 0.6%, P<.001). One hundred twenty-six patients had postoperative infection (ICD-9-CM code 998.5), which affected 1.8% of women and 1.1% of men (P = .01). However, there were other types of infection that were similar across sex. The frequency of septicemia (ICD-9-CM code 038) was similar across sex, as was infection due to an internal prosthetic device, implant, or graft (ICD-9-CM code 996.6), which was reported in 0.5% of women and in 0.5% of men. Infections of the heart (pericarditis, endocarditis, and myocarditis) were recorded in 0.6% of men and in 0.7% of women. Mediastinitis was reported in 12 patients, 0.2% of men and 0.1% of women.
Our data suggest that the increased mortality in women after CABG surgery may be largely explained by differences in the underlying distribution of infection between men and women. Women who had CABG surgery were more likely to have an infection than men; it was this differential prevalence in infection that may have contributed to the excess deaths in women. The predominance of infection in women was evident across age, race, and other subgroups of patients, as well as across various infection sites.
Our results indicate that there were 2 competing forces that influenced the sex effect on mortality. The first was the increased prevalence of infection among women, and the second was the increased risk of mortality in men who were infected. The first put women at a disadvantage, while the second was detrimental to men. When these 2 competing forces were considered, the sex effect on mortality was substantially reduced toward zero. This highlights the importance of considering underlying factors, including effect modifiers, that may contribute to observed differences in mortality. In this population, the 95 excess deaths in women could be largely explained by infection; with adjustment, the number of excess deaths in women decreased to 4. The pathophysiology of infectious processes clearly places infection in the causal pathway and is more relevant to postsurgical death than known physiological differences between the sexes. However, small coronary artery size, more common in women than in men, also has been shown to be associated with greater postoperative mortality.3,4 Therefore, it would be prudent to investigate the relation, if any, between vessel size and infection, as well as the effect on postsurgical mortality.
Although infection in hospitalized patients has long been recognized as a risk factor for postoperative death, risk-based models for predicting mortality after CABG surgery have focused on preoperative risk factors, including sex but not infection.17 Although infection provides a possible explanation for the disparity between men and women in mortality after CABG surgery, such an explanation has not yet been recognized, despite evidence of greater infection in women after CABG surgery.18- 22 Studies18- 22 have shown that female sex is an independent risk factor for hospital-acquired infection after cardiac surgery. Other studies23- 25 of single-site infections in patients after cardiac surgery, however, have not shown an increased incidence in women. If a systemic response to sepsis affects mortality risk, investigations that solely target single-site infections may potentially misclassify exposure in individuals with infection at other sites and artificially reduce the estimate of effect toward the null.
It is possible that infection at any site may raise the baseline risk of mortality in older hospitalized patients, as our data suggest. Although a systemic inflammatory response has been shown to increase the risk of coronary events and death,26 it is unclear the extent to which an inflammatory response due to infection and the duration of such a response (acute or chronic) may precipitate death after cardiac surgery. It has been suggested that older patients with sepsis and concomitant conditions may undergo a prolonged hypoimmune state that increases mortality risk.27 Milazo et al28 found that elevated C-reactive protein levels before CABG surgery increased the risk of recurrent myocardial infarction. However, Marica et al29 did not find such a relation between preoperative C-reactive protein and cardiac events (including death), although they excluded individuals with acute infection from their analyses. Leukocyte count, an indicator of systemic inflammation, was recently found to be significantly associated with mortality after CABG surgery, with a 30-day mortality of 7.6% in patients in the highest leukocyte quartile compared with 2.6% in patients in the lower quartiles combined (P<.001).30
Although it is possible that the sex difference in the prevalence of infection found in our study could be due to differential coding of infection by sex (systematic undercoding of infections in men or systematic overcoding of infections in women), others have found a greater prevalence of infection after surgery.18- 22 Specific types of infection, such as urinary tract infections, have been widely reported to be more prevalent among women than among men, occurring at an annual incidence of 12% in women and of 3% in men.31 Moreover, the survival advantage of women, once infected, has been previously recognized,32- 34 and the percentage of patients infected after CABG surgery in our study (12.2%) was similar to that found in other studies.18,19 On the other hand, underreporting of infection when using administrative databases is a possibility. The addition of information regarding antimicrobial use to the ICD-9-CM code of postoperative infection has been shown to increase the proportion of patients identified with surgical site infection.35
With the information available, we were unable to discern whether the onset of infection was before admission or developed during the hospital stay. Data suggest that the proportion of patients with overt infection at admission is not likely to be large. The prevalence of infection increased with LOS, from 2.6% among those who spent less than 5 days in the hospital to 33.4% among those with stays of more than 12 days. Only 0.4% of the patients had an admission diagnosis indicative of infection. Furthermore, the scheduling of elective surgery would be expected to be postponed if an infection was readily apparent at the time of admission. With the database used, the extent to which unrecognized infection or low-grade infections were present at the time of admission could not be determined. Therefore, confirmatory studies using clinically robust data would be useful.
The findings of this study are not likely due to increased LOS in women. The reduction in attributable risk with adjustment for infection was evident, even after exclusion of the patients with longer LOS. However, the difference in mortality between men and women was not entirely explainable by infection status for those patients with short stays (<5 days). Short-stay patients are more likely to represent individuals at the extremes, those with favorable prognoses or those who died during the immediate postoperative period. Additional investigation would be necessary to clarify the factors that affect mortality in such patients.
Replication of these findings would be advisable in other databases with clinically relevant predictors of mortality, such as ejection fraction and extent of disease, and with time-related factors, such as antibiotic prophylaxis. Furthermore, because the Medicare data used in this study represent patients hospitalized in 1997-1998, it would be desirable to replicate these analyses using more recent hospitalizations.
The major findings are that women hospitalized for CABG surgery are more likely to have an infection than men, that infection at any site increases the baseline risk of mortality (although the degree of increase is greater in men than in women) and that the increased mortality in women after CABG surgery diminishes after adjustment for underlying sex differences in infection. Clearly, infection is an important factor when evaluating the association between sex and mortality after CABG surgery. As a direct indicator of patient safety, infection at any site, regardless of time of onset, warrants additional investigation as a potential contributor to mortality.
Correspondence: Mary A. M. Rogers, PhD, Division of General Medicine, Department of Internal Medicine, University of Michigan, 300 N Ingalls, Suite 7E07, Ann Arbor, MI 48109-0429 (firstname.lastname@example.org).
Accepted for Publication: August 29, 2005.
Author Contributions: Dr Rogers had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosure: None.
Funding/Support: This project was supported by the Veterans Affairs Medical Center/University of Michigan Patient Safety Enhancement Program and by grant 2002-0013 from The John A. Hartford Foundation, New York, NY. Dr Langa was supported by Career Development Award K08 AG19180 from the National Institute on Aging, Bethesda, Md, and by the Paul B. Beeson Career Development Awards in Aging Research Program, New York. Dr Saint was supported by a Career Development Award (RCD-00-006) from the Health Services Research and Development Program, Department of Veterans Affairs, Washington, DC, and by Patient Safety Developmental Center Grant P20-HS11540 from the Agency for Healthcare Research and Quality, Rockville, Md.
Role of the Sponsor: The funding agencies had no role in the design of the study, conduct of the study, analyses or interpretation of the data, or preparation, review, or approval of the manuscript.
Acknowledgment: We are indebted to Geir Egil Eide, MSc, for his statistical advice and to Steven Bernstein, MD, MPH, for his helpful comments.