Organizational Characteristics of Intensive Care Units Related to Outcomes of Abdominal Aortic Surgery FREE

Morbidity and mortality rates in intensive care units (ICUs) vary widely among institutions.^{1 - 3} This variation may be related to differences in ICU structure and care processes.^{4 - 9} To assess and improve the quality of care in ICUs, it is necessary to understand how ICU structure and care processes are related to clinical and economic outcomes.^{10 - 13} We hypothesize that differences in organizational characteristics of ICUs are associated with significant differences in clinical and economic outcomes for patients undergoing high-risk surgical procedures that typically require postoperative care in an ICU.

Abdominal aortic surgery is a relatively common procedure that is performed in a variety of acute care hospitals with different ICU organizational characteristics. Patients admitted for abdominal aortic surgery routinely require ICU admission because of high postoperative morbidity and mortality.^{14 - 16} Patients undergoing abdominal aortic surgery thus provide an appropriate population in which to assess our hypothesis. The specific aim of this study was to determine whether differences in organizational characteristics of ICUs were associated with differences in in-hospital morbidity and mortality, hospital length of stay, and ICU days for abdominal aortic surgery patients.

Following approval from our institutional review board, we used nonconfidential patient data from the Uniform Health Discharge Data Set maintained by the Maryland Health Services Cost Review Commission (HSCRC), which contains information on all patients discharged from the 52 nonfederal short-stay hospitals in Maryland. We selected variables on the patient's age, sex, race, nature of admission, operating physician, vital status at discharge, hospital length of stay, ICU days, and codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) for the primary discharge diagnosis, the principal procedure, as many as 14 secondary discharge diagnoses, and as many as 14 secondary procedures.

We obtained information on all patients aged 30 years or older who were discharged from a Maryland hospital between January 1994 and December 1996 with a principal procedure code for abdominal aortic surgery (ICD-9-CM code 3844 for resection of abdominal aorta with replacement and ICD-9-CM code 3925 for aorto-iliac-femoral bypass). We excluded 9 patients who were younger than 30 years, all of whom had injury to a blood vessel (ICD-9-CM code 902). Our study had 3 primary outcome variables: in-hospital mortality, hospital length of stay, and ICU days. We obtained information about complications and comorbid diseases using the secondary diagnosis and procedure codes included in the HSCRC database. Complications were chosen by 2 intensivists (P.P. and T.D.) by reviewing ICD-9-CM prior to data abstraction for complications likely to be associated with in-hospital mortality; these were coded as dichotomous variables.

We adjusted for comorbid diseases, severity of illness, and hospital and surgeon volumes. We selected the Romano-Charlson comorbidity index to identify potentially important comorbid diseases.^{17 - 20} Instead of using a single comorbidity index in our analysis, each disease was included as a separate variable. To adjust for severity of illness, we classified patients as having a ruptured or unruptured aorta (ICD-9-CM code 441.3) and used the nature of admission field, coded at admission, to identify each case as elective, urgent, or emergent (Alan Dardik, MD, PhD, oral communication, 1998).²¹ We calculated the volume of aortic surgery performed by each hospital and each surgeon in the database.

We developed a questionnaire to obtain information about organizational characteristics of ICUs that provide care to abdominal aortic surgery patients. To identify specific ICU issues that might be relevant to the care of such patients, we used a previously developed questionnaire about ICU organization and staffing.²² The questionnaire used in our study had 32 items that evaluated ICU physician staffing, nurse staffing, and care processes. We established content validity by having the instrument reviewed independently by 5 intensive care physicians to determine if each question captured the intended domain.

We identified the ICU directors at the 46 Maryland hospitals that were recorded in the HSCRC database as having performed abdominal aortic surgery during our target period. For hospitals with more than 1 ICU, we identified the director of the ICU that cared for abdominal aortic surgery patients. We mailed each ICU director a letter explaining our study. One week later we mailed the, and 2 weeks later, a reminder letter. After another 2 weeks, we mailed a second questionnaire. Six weeks after the initial questionnaire mailing, the principal investigator (P.P.) called the ICU directors at nonresponding hospitals and encouraged them to respond. We linked our survey data to HSCRC data, but we excluded the names of hospitals and their ICU directors from the combined database used to conduct our analysis. The data abstractor was blinded to the hospital name, ICU characteristics, and patient outcome.

We performed a descriptive analysis of ICU characteristics (Table 1), patient characteristics (Table 2), patient outcomes (in-hospital mortality, hospital length of stay, and ICU days) and medical and surgical complications and interventions (Table 3).

We performed bivariate analysis to evaluate the association between the independent variables and each of the dependent variables. We used the t test or Wilcoxon rank sum test to evaluate the association between continuous dependent variables and categorical independent variables. We used simple linear regression to evaluate the association between continuous dependent and independent variables; the χ² test to evaluate the association between categorical dependent and independent variables; and simple logistic regression to evaluate the association between categorical dependent variables and continuous independent variables. We modeled hospital and surgeon volume as dichotomous variables, using a Lowess smoothing curve,²³ and defined low volume as fewer than 36 cases per year for hospitals and fewer than 8 cases per year for surgeons. Thresholds for aggregating survey data were determined by exploration of the distribution of responses for each item and exploration of the relationship between independent and dependent variables using smoothing splines.²⁴

We performed a 2-stage multivariate analysis for each of our main outcome variables. In the first stage, we sought to identify patient characteristics that were independently related to each outcome variable. In the second stage, we sought to identify ICU characteristics that were independently related to each outcome variable after adjusting for differences in patient characteristics. Because the risk for an outcome, including the primary outcome variables and complications, was clustered within a hospital, we obtained robust variance estimates and used multilevel modeling to evaluate the data hierarchically (characteristics of patients and hospitals on outcomes).²⁵ In the first stage, we adjusted for patient age, sex, race (white vs nonwhite), nature of admission (emergent vs elective and urgent vs elective), type of aneurysm (ruptured vs unruptured), comorbidity (yes vs no for each of the 10 diseases in the Romano-Charlson index), surgeon volume (<8 vs ≥8 cases per year), and hospital volume (<36 vs ≥36 cases per year). Thresholds for surgeon and hospital volume are consistent with previously published literature.¹⁴

We used multiple logistic regression to evaluate the relationship between ICU and patient characteristics and the categorical outcome variables for in-hospital mortality and complications. We used multiple linear regression to evaluate the relationship between ICU and patient characteristics and the continuous dependent variables for hospital length of stay and ICU days. To meet the assumption of normality of the linear regression model, we performed a log transformation on hospital length of stay and ICU days. We evaluated the effect of the log transformation by using the residual plot, the quantile-quantile plot, and the Shapiro-Wilks test.²⁶ We obtained the maximum likelihood estimated mean percentage increase in each continuous outcome variable by taking the antilog of the linear regression coefficients for each independent variable. We included each independent variable with P<.05 in the bivariate analysis in the multivariate model. If 2 independent variables were highly correlated with each other, we excluded the variable with the largest variance from the multivariate analysis.

All reported P values are 2-tailed.P values were considered significant if they were less than .05. We used STATA Version 5.0 software (Stata Corp, College Station, Tex) to perform all calculations.

We received completed questionnaires from the ICU directors at 39 (85%) of the 46 hospitals that performed abdominal aortic surgery during the study period. Table 1 describes the characteristics of the responding ICUs.

Responding hospitals accounted for 2606 (87%) of the 2987 patients receiving abdominal aortic surgery during the study period and also included the 2 largest medical centers in Maryland. These 2 centers performed 313 (12%) of the cases included in our database but neither performed the largest number of cases. Table 2 describes the unadjusted outcomes and characteristics of the 2987 patients aged 30 years or older who had abdominal aortic surgery in responding and nonresponding hospitals between 1994 and 1996. The 2987 patients in our database had the following primary ICD-9-CM diagnoses: 65% had a code for aortic aneurysm; 16% for atherosclerosis; 16% for arterial embolism and thrombosis; and 3% for complications of a surgical procedure.

The in-hospital mortality rate varied among hospitals from 0% to 66%. In the bivariate analysis, a number of ICU characteristics were associated with increased risk of in-hospital mortality after abdominal aortic surgery. Odds ratios (ORs) and 95% confidence intervals (CIs) for these characteristics included not having a full-time ICU medical director (OR, 2.1; 95% CI, 1.2-3.5), having less than 50% of ICU attendings certified in critical care (OR, 2.0; 95% CI, 1.4-3.0), not having daily rounds by an ICU physician (OR, 3.0; 95% CI, 2.1-4.3), and having a decreased ICU nurse-patient ratio in the evening (OR, 1.9; 95% CI, 1.2-3.0). The other ICU characteristics were not associated with in-hospital mortality.

In the multivariate analysis, (Table 4), not having daily rounds by an ICU physician vs having daily rounds by an ICU physician was independently associated with an increased risk of in-hospital mortality (OR, 3.0; 95% CI, 1.9-4.9).

A number of postoperative complications from our preselected list were independently associated with increased in-hospital mortality after abdominal aortic surgery. These included acute myocardial infarction (OR, 10.6; 95% CI, 5.1-21.9), cardiac arrest (OR, 91.0; 95% CI, 35.0-397.0), acute renal failure (OR, 8.3; 95% CI, 3.9-17.8), septicemia (OR, 7.9; 95% CI, 4.3-14.6), platelet transfusion (OR, 4.5; 95% CI, 1.6-12.7), reintubation (OR, 3.1; 95% CI, 2.2-4.3), reoperation for bleeding (OR, 2.5; 95% CI, 1.3-4.9), and surgical complications after a procedure, defined as hemorrhage, laceration, or disruption of wound (OR, 3.7; 95% CI, 2.2-6.0). Table 3 summarizes the rates of occurrence of these postoperative complications in the abdominal aortic surgery cases.

To explore how an ICU characteristic might be related to in-hospital mortality, we examined the association between complications and the ICU characteristic significantly associated with in-hospital mortality in the multivariate analysis. After adjusting for patient characteristics, hospital, and surgeon volume, we found that lack of daily rounds by an ICU physician was independently associated with an increased risk of cardiac arrest (OR, 2.9; 95% CI, 1.2-7.0), acute renal failure (OR, 2.2; 95% CI, 1.3-3.9), septicemia (OR, 1.8; 95% CI, 1.2-2.6), platelet transfusion (OR, 6.4; 95% CI, 3.2-12.4), and reintubation (OR, 2.0; 95% CI, 1.0-4.1), but was not associated with an increased risk of surgical complications (Table 3).

In the bivariate analysis, ICU characteristics associated with increased hospital length of stay included not having a full-time ICU medical director (mean increase, 10%; 95% CI, 4%-16%), having an ICU nurse-patient ratio of less than 1:2 during the evening (mean increase, 17%; 95% CI, 1%-35%), not having monthly review of ICU morbidity and mortality (mean increase, 18%; 95% CI, 8%-27%), and routinely extubating aortic surgery patients in the operating room (mean increase, 23%; 95% CI, 8%-40%).

In the multivariate analysis, shown in Table 4, ICU characteristics independently associated with increased hospital length of stay for abdominal aortic surgery cases were having an ICU nurse-patient ratio of less than 1:2 in the evening (mean increase, 20%; 95% CI, 7%-33%), not having monthly review of ICU morbidity and mortality (mean increase, 15%; 95% CI, 4%-25%), and having aortic surgery patients routinely extubated in the operating room (mean increase, 11%; 95% CI, 2%-21%). After adjusting for patient characteristics, hospital volume, and surgeon volume, we found that routinely extubating patients in the operating room was independently associated with increased risk of reintubation (OR, 1.7; 95% CI, 1.3-2.1) and postoperative pulmonary complications (OR, 1.9; 95% CI, 1.7-2.3).

In the bivariate analysis, the ICU characteristics associated with increased ICU days included not having daily rounds by an ICU physician (mean increase, 65%; 95% CI, 35%-96%), having the surgeon or both the surgeon and ICU physician manage the patient vs having the ICU physician manage the patient in the ICU (mean increase, 39%; 95% CI, 18%-55%), and having an ICU nurse-patient ratio of less than 1:2 during the day (mean increase, 29%; 95% CI, 1%-68%).

In the multivariate analysis, as shown in Table 4, the ICU characteristics independently associated with increased ICU days for abdominal aortic surgery cases included not having daily rounds by an ICU physician (mean increase, 83%; 95% CI, 48%-126%), and having an ICU nurse-patient ratio of less than 1:2 during the day (mean increase, 49%; 95% CI, 17%-91%).

Surgeons who performed fewer than 8 cases per year had a higher mean in-hospital mortality rate than surgeons who performed 8 or more (10% vs 5%; P = .003). However, when we adjusted for differences in patient characteristics using multivariate logistic regression, there was no significant difference in in-hospital mortality between surgeons who performed fewer than 8 and those who performed 8 or more cases per year.

Hospitals that had fewer than 36 cases of abdominal aortic surgery per year had a higher mean in-hospital mortality rate than hospitals that had 36 or more cases per year (8% vs 5%; P = .005). When we adjusted for differences in patient characteristics using multivariate logistic regression, we found that hospitals that had fewer than 36 cases per year had a significantly higher in-hospital mortality rate than hospitals that had 36 or more. As shown in Table 4, hospital volume also was inversely associated with in-hospital mortality after adjusting for differences in both ICU and patient characteristics.

This study demonstrates that there is significant variation in the outcomes of abdominal aortic surgery patients and significant variation in ICU organizational characteristics in Maryland hospitals that perform this relatively common surgery. More importantly, our results indicate that ICU organizational characteristics are related to differences in in-hospital mortality, ICU days, and hospital length of stay. Such information may provide direction regarding ways to further improve the outcomes for patients who have high-risk operations such as abdominal aortic surgery. Because the 5-year relative survival of aortic aneurysm patients, especially octogenarians, is good and supports surgery,²⁷ strategies to reduce in-hospital mortality become increasingly important.

Daily rounds by an ICU physician were associated with a 3-fold reduction in in-hospital mortality for abdominal aortic surgery patients. This finding is consistent with an emerging body of evidence that suggests using full-time intensive care physicians can reduce in-hospital mortality.^{28 - 29} We found that daily rounds by an ICU physician were associated with reduced risk of several specific medical complications and interventions that an intensivist would likely affect but were not associated with reduced risk of surgical complications. Daily rounds by an ICU physician may be a marker for team care, and this model can be widely applied because our results were not predicated on the presence of residents. Our study is unique because we evaluated mortality in a high-risk population, adjusted for differences in comorbidity and severity of illness, used multilevel modeling, and included data from 2606 patients from 39 hospitals, which provided us with the statistical power to detect clinically significant associations between organizational characteristics of ICUs and outcomes. Previous studies have had less power for detecting differences because they included many patients with a relatively low risk of in-hospital mortality and adjusted for differences in risk across patient populations, which may distort the relationship between ICU organizational characteristics and outcomes.^{1 ,3 ,30}

We also found that variation in organizational characteristics of ICUs was associated with differences in resource use for patients undergoing abdominal aortic surgery in Maryland. A decreased ICU nurse-patient ratio during the day or evening was associated with increased ICU days and hospital length of stay, respectively. In addition, monthly review of ICU morbidity and mortality was associated with decreased hospital length of stay, while routinely extubating these patients in the operating room was associated with increased hospital length of stay. To further explore this relationship, we found that routine extubation was independently associated with increased risk of reintubation and postoperative pulmonary complications. These results are consistent with several small studies that showed that the addition of ICU specialists decreased ICU resource use.^{10 - 13 ,31}

Daily rounds by an ICU physician may be as important as the experience of the surgeon performing a high-risk operation .^{14 - 16 ,32 - 33} Although surgeon volume was not significantly related to outcomes after adjusting for patient characteristics, increased hospital volume was associated with better outcomes for abdominal aortic surgery patients. If the vast majority of surgeons performing this procedure have sufficient experience, the best way to improve outcomes may be to improve postoperative ICU care. It is concerning that hospital volume was independently associated with in-hospital mortality and 60% (n = 1793) of patients were operated on at low-volume hospitals. This is an area worthy of further study.

A closed ICU requires full-time, qualified critical care physicians and the commitment from both physicians and administrators to reorganize the ICU. We did not attempt to classify ICUs as open vs closed because these terms are poorly defined and may not reflect the actual process of care delivery.

We recognize several limitations to our study. The first is that coding of comorbid diseases and complications in the HSCRC database may not be as accurate as the coding of the principal procedure. An analysis the quality of the Maryland HSCRC database indicates a low error rate for coding the types of outcome variables used in this analysis.³⁴ In this study, the ICU data were prospective in that we collected the data, while the patient data were retrospective in that we retrieved the data from an existing secondary data set. It is possible that different methods for obtaining these 2 types of data may have introduced some bias, but this potential bias should have been minimized by our keeping the data abstraction blinded to hospital name, ICU characteristics, and patient outcome. However, we found no evidence of systematic differences in coding of comorbid diseases between hospitals with and without daily rounds by an ICU physician. The percentage of patients with chronic renal disease and mild liver disease (complications associated with in-hospital mortality) were exactly the same at hospitals with and without daily rounds by an ICU physician (4% and 1%, respectively), while prior myocardial infarction was coded more frequently at hospitals without daily rounds by an ICU physician and severe diabetes mellitus was coded more frequently at hospitals with daily rounds. Therefore, misclassification of comorbid diseases could not significantly bias the results because coding of comorbid diseases at hospitals with and without daily rounds by an ICU physician is not a confounding variable. Additionally, random misclassification of comorbid diseases would bias our study toward the null hypothesis by making it harder to identify differences in outcomes between study groups.

It is unlikely that our results are significantly biased by misclassification of complications. We reviewed a random sample of 25 medical records from 1 hospital and found the coding of complications associated with in-hospital mortality to be 96% accurate. We helped establish construct validity for the coding of complications by evaluating only complications that were independently associated with in-hospital mortality.³⁵ Additionally, the bias from coding of complications is likely minimal because the association between daily rounds by an ICU physician and complications was consistent across several medical complications and interventions that an intensivist would likely affect but was not found with surgical complications. This type of statewide administrative database has unique value because it includes data on all patients receiving care in nonfederal hospitals in the state and permits comparisons among hospitals that would otherwise be difficult to obtain.^{36 - 39} The findings regarding demographic and clinical factors that are associated with reduced complications of abdominal aortic surgery are remarkably consistent for several of the medical complications we examined.

The second limitation relates to adjustment for severity of illness. We did not use a systematic scoring system, such as the Second Acute Physiology and Chronic Health Evaluation (APACHE II), that requires medical record review to adjust for severity of illness.⁴⁰ Rather, we classified patients as having ruptured vs unruptured aneurysms and elective, urgent, or emergent admissions; the coding for these variables was found to be more than 98% accurate (Alan Dardik, MD, PhD, oral communication, 1998).²¹ While the use of APACHE II data may have reduced bias in the severity of illness adjustment, it is not routinely collected at most hospitals and, thus, was not available for our analysis. However, patients having abdominal aortic surgery are unique because the variables for ruptured vs unruptured aneurysm and elective, urgent, or emergent admission seem to account for severity of illness relatively well, as evidenced by their ORs for in-hospital mortality. Most patients (91%) had unruptured aneurysms and would be expected to have normal physiology on admission. Therefore, we would expect the systematic error introduced by our inability to use APACHE II to be minimal. The study would be biased toward the null hypothesis if the misclassification of severity of illness was random.¹⁴ Had we studied only hospitals that had APACHE II data, we would have been limited to a small number with relatively sophisticated ICUs that would have likely biased the results and reduced generalizability. Had we prospectively obtained APACHE II data from all 46 hospitals in Maryland, the cost and duration of the study would have been significantly increased.

The third limitation is that we did not adjust for differences in pre-ICU care. We were not able to adjust for changes in surgical technique over time; however, there were no major technical advances in aortic surgery during our study. Endoluminal stents were not available during the study. Moreover, there is no known association between surgical approach and in-hospital mortality. We were also not able to adjust for the type of anesthesia patients received. However, most of the differences in pre-ICU care would affect surgical complications more than medical complications, and we were more interested in the association between ICU organization and medical complications.

A fourth limitation is our inability to adjust for differences in post-ICU care. However, we did adjust for hospital volume and used multilevel modeling that may account for some of the differences in post-ICU care. Although the data set used in this study does not provide information about ICU mortality, we believed the use of in-hospital mortality as an outcome would be less biased than ICU mortality because ICU mortality is affected by decisions to discharge patients from the ICU. We did not control for discharge decisions and, thus, cannot associate the location of death with the location in which care was received.

The fifth limitation is the reliability and validity of our ICU survey instrument. We relied on the ICU directors to describe the characteristics of their ICUs. Most questions, including daily rounds by an ICU physician, were derived from a previously validated questionnaire.³ Because most ICUs have 1 director, we were not able to obtain interrater reliability. We tested the interrater reliability on 2 sets of intensivists (the 2 codirectors of 1 ICU and the director and staff intensivist at another) and found 100% agreement and 97% agreement, respectively, on the responses to the questions. Our validity evaluation focused on content validity that we established by having 5 intensive care physicians independently review the instrument. We assumed the responding ICU medical directors would answer accurately, since we preserved hospital confidentiality. The sixth limitation is that the survey was administered at the end of the 3 years for which we had patient data. It is unlikely that hospitals dramatically changed their ICU organization during this period, but if this occurred, we would have been less likely to find an association between ICU organizational characteristics and outcome. The final limitation is that we focused on 1 surgical procedure in 1 state, thereby limiting our ability to apply these findings to other procedures and other states. However, our study is strengthened by inclusion of survey data from 39 of the 46 hospitals that performed abdominal aortic surgery in Maryland.

Despite these limitations, our study results have significant implications for clinicians, hospital administrators, and policymakers because they point to several aspects of ICU care that potentially could be modified to decrease in-hospital mortality, complications, and length of stay of postoperative care for high-risk patients. The results suggest that decreasing ICU nursing staff below a certain level may lead to increased costs of care and length of stay for patients. Surgeons and anesthesiologists should reassess reasons for extubating abdominal aortic surgery patients in the operating room because this may be associated with increased morbidity in some populations. However, additional research is needed to determine whether implementation of specific measures, such as daily rounds by an ICU physician, can decrease morbidity and mortality, as well as length of stay, for high-risk surgical patients in places where such measures are not in place. Meanwhile, patients should consider how ICUs are organized when choosing a hospital in which to have a major surgery.