Relationship between Volume and Survival in Closed Intensive Care Units Is Weak and Apparent Only in Mechanically Ventilated Patients

CRITICAL illness represents an enormous clinical and economic burden on the healthcare system. There is an accepted relationship between hospital volume and survival in some conditions, related to the “practice makes perfect” concept. This concept logically applies to technically demanding cases and is especially relevant in the context of a shortage of specialists.

Increased case load is associated with improved outcomes in some areas of health care, including trauma, acute myocardial infarction, and many types of high-risk surgeries.^1–5  Recent studies have documented a relationship between volume and outcome in critical care,⁶  mostly in patients with acute respiratory failure, sepsis, or high risk for death.^7–10 

Although most intensive care unit (ICU) volume–outcome studies have shown a significant relationship, at least one population-based study in patients undergoing mechanical ventilation showed no significant volume–outcome relationship.¹¹  Some studies found weaker volume–outcome relationships, present only in high-risk patients (Simplified Acute Physiology Score [SAPS] >41) treated at ICUs treating high volumes of high-risk patients.⁸  The existence of a volume–outcome effect in selected hospitals may not reflect the true relationship in broader settings. Some studies demonstrate worse outcomes at very low-volume hospitals, but no relationship between volume and outcome in medium-, high-, or very high-volume hospitals.¹² 

These issues are important not only in the United States but also in other industrialized countries, all of which struggle with increasing demand for critical care in the setting of constrained resources. Although integral to healthcare systems in developed countries, intensive care varies considerably among countries.¹³  Factors related to variable outcomes include ICU bed rate, ICU models (open vs. closed), nurse–patient ratios, availability of coverage by trained ICU physicians 24 h a day, 7 days a week, and ICU overcrowding. The Spanish healthcare system is a mainly publicly funded hospital network characterized by closed-ICU models and 24/7 coverage by trained ICU physicians.¹⁴  On the basis of reports demonstrating better outcomes in high-volume hospitals, most from studies at American hospitals, opinion leaders propose to regionalize Spanish ICUs. Thus, it is of the utmost interest to know the relationship between ICU volume and outcome in our publicly funded healthcare system. The objective of the current study was to examine the relationship between ICU volume and outcome in the entire group of ICU patients and in the group of ICU patients receiving mechanical ventilation. We used data from the Sabadell Score clinical database to compare mortality rates and length of stay at hospitals that care for a high volume of patients with those at hospitals that care for a low volume of patients.

We analyzed data from a prospective, multicenter cohort of 4,001 adult patients admitted to 29 ICUs in Spain (see appendix for list of centers) during a 3-month period beginning March 1, 2008.¹⁵  The institutional review boards at each participating center approved the study protocol and waived the need for consent.

In a specific Web-based database, we recorded the following variables for each patient admitted to the ICU:¹  on ICU admission: age, sex, diagnosis, predicted risk of death (Acute Physiology and Chronic Health Evaluation II, SAPS II, or SAPS3 depending on the clinical routine of each ICU), source of admission, and do-not-resuscitate orders;²  during the ICU stay: tracheal intubation, mechanical ventilation, noninvasive mechanical ventilation, vasoactive drugs, parenteral nutrition, blood transfusion, dialysis, tracheostomy, acute renal failure or infection acquired in the ICU, and prognosis assessed by Sabadell Score;³  after discharge from the ICU: ICU readmission and outcome. In patients readmitted to the ICU, only the first admission was included in the analysis.

The ward team was unaware of the study, so the care of patients in the ward should represent the standard of treatment in most of our healthcare systems.

Follow-up included ICU readmission and up-to-90-day ward outcome, even in patients who were transferred to another acute care hospital.

The exposure variable was annualized ICU volume, defined as the extrapolated number of patients per year. Moreover, we analyzed this variable separately for the group of patients who received mechanical ventilation. The primary outcome was hospital mortality. The lengths of stay in the ICU and in the hospital among survivors were secondary outcomes. For the analysis of the association between ICU volume and outcome, volume was treated as a continuous variable in which the reference category was the ICU with the lowest volume; after categorizing volume into quartiles, the reference category was the quartile with the lowest volume. Physician staffing was not included as an independent variable because all the ICUs were organized as “closed units”; in other words, all had dedicated intensivists present 24 h per day, 7 days per week, as is the norm in our healthcare system.

We addressed potential confounding due to variation in the case mix by controlling for the severity of illness and other variables related to the outcome of critical care. The severity of illness was determined by Acute Physiology and Chronic Health Evaluation II, SAPS II, or SAPS3 scores on the day of admission according to each ICU’s routine approach. Other risk-adjustment variables included age, acute renal failure, ICU-acquired infection, mechanical ventilation, blood transfusion, parenteral nutrition, vasoactive drugs, and noninvasive ventilation.

Categorical variables were expressed as percentages and compared using chi-square tests. Normally distributed continuous variables were expressed as means and SDs and compared by one-way ANOVA; nonnormally distributed continuous variables were expressed as medians and interquartile range and compared by Kruskal–Wallis analysis. Statistical significance was set at a P value of less than 0.05.

The relationship between hospital mortality and ICU volume, both in the entire population and in the group of patients who received mechanical ventilation, was assessed by simple linear regression by least squares fitting.

ICUs were grouped in quartiles according to the number of patients admitted, considering the total number of patients and mechanically ventilated patients separately. The association of the variables with hospital mortality was assessed with backward multiple logistic regression. The discrimination of the multivariate model was assessed using the area under the receiver operating characteristic curve. Accuracy was considered to be good if the area under the curve was more than 0.75 and excellent if more than 0.85. All statistical analyses were performed with STATA (version 10.0) software (StataCorp, College Station, TX). We assessed the sensitivity of our findings by repeating the primary analysis under varying assumptions about the study population in a sensitivity analysis for mortality in the hospital.

Table 1 shows the characteristics of the participating hospitals. Most hospitals were funded by the public healthcare system, but the geographic location and academic status varied considerably.

Compared with ICUs with low volume, ICUs with high volume had more hospital beds and were more likely to have residency programs. The median annualized ICU volume was 464 (interquartile range: 280–728) patients per year.

Table 2 shows the patients’ clinical characteristics by ICU volume quartiles. We studied 4,001 patients (mean age, 61 ± 17 yr; mean risk of death, 23% [range at hospital level, 14–46%]). Observed hospital mortality was 19% (range at hospital level, 11–35%), resulting in an observed versus predicted mortality ratio, or standardized mortality ratio (SMR) of 0.81 (range at hospital level, 0.5–1.3). Although no other correlations with volume were observed, age and severity of illness were higher at hospitals with very low ICU volume than at hospitals in the other quartiles. Diagnoses at admission varied widely among quartiles, but did not correlate with ICU volume. Coronary patients were less common in quartile 3, and neurological patients were less common in quartile 2. Most trauma and postsurgical patients were quartile 3, which correlated with higher use of mechanical ventilation and blood transfusion despite lower severity of illness.

The ICU length of stay, ICU readmission, and ICU mortality were very similar in all volume quartiles. Although we found a slight trend toward lower hospital mortality in higher volume ICUs, when we adjusted for predicted hospital mortality, the SMR in higher volume ICUs was very similar to that of the other groups. Figure 1 is a scatter plot showing the distribution of SMR among ICUs and differences in volume and severity of illness.

We constructed a logistic multiple regression model for variables associated with hospital mortality, including all the variables that were significant in the univariate analysis and quartiles of ICU volume. Table 3 shows the variables selected by the model (age, predicted mortality, mechanical ventilation, vasoactive drugs, acute renal failure, and ICU-acquired infection); the model discarded ICU volume. The discrimination of the model was excellent (area under the curve = 0.86).

Among patients needing mechanical ventilation (n = 1,923; 48% of the total), the predicted risk of death was 32% (range at hospital level, 14–60%) and observed hospital mortality was 30% (range at hospital level, 12–61%), resulting in an SMR of 0.96 (range at hospital level, 0.5–1.7; table 4). Figure 2 is a scatter plot showing the distribution of SMR among ICUs and differences in volume and severity of illness for mechanically ventilated patients. We found a trend toward older and sicker mechanically ventilated patients in low- and medium–low-volume ICUs.

In all ICU volume quartiles, we found very similar ICU readmission rates. However, ICU length of stay varied widely, although it did not correlate with volume. Raw ICU and hospital mortality were inversely correlated with volume; but, after we adjusted for expected mortality, this correlation disappeared, although ICU and hospital mortality in very low-volume ICUs remained higher.

We constructed a logistic multiple regression model for variables associated with hospital mortality, including all the variables that were significant in the univariate analysis and quartiles of ICU mechanical ventilation volume. Table 5 shows the variables selected by the model (age, predicted mortality, parenteral nutrition, vasoactive drugs, acute renal failure, ICU-acquired infection, and the first quartile, but not the other quartiles). The discrimination of the model was very good (area under the curve = 0.84).

To explore the sensitivity of our findings, we repeated the analysis with varying assumptions about the patient population (tables 6 and 7). Our results were not affected by exclusion of the variables related to ICU-specific treatments or by the exclusion of the ICU-acquired complications (infections and acute renal failure), which may be associated with poor performance. Our results were not affected by using a multilevel model with patients as the basic observation and hospital as the second-level of aggregation, with quartile of hospital volume as a hospital-level characteristic in a random-intercept model (fig. 1, see Supplemental Digital Content 1, http://links.lww.com/ALN/A951, which is a figure demonstrating an intraclass correlation coefficient of 3%). ICU volume was also modeled as a continuous predictor, but failed to prove to be a significant factor (table 1, see Supplemental Digital Content 2, http://links.lww.com/ALN/A952, which is a table of variables associated with mortality).

The current data failed to demonstrate an association between higher ICU volume and lower risk-adjusted mortality in ICU patients. We found only slightly worse outcome for patients receiving mechanical ventilation in very low-volume (<100 patients/yr) ICUs.

There are many possible causes of a relationship between ICU volume and outcome among patients receiving critical care. It has been suggested that high-volume ICUs may improve outcomes by implementing a broad range of best practices, including higher nurse-to-patient ratios, multidisciplinary care teams, lung-protective ventilation strategies, and protocols for sedation and weaning. Nevertheless, recent studies found no relationship between hospital volume and adherence to clinical guidelines.¹⁶  Clinicians may gain experience in the care of the critically ill by treating more severe patients, but the ratio of clinicians to patients is by no means correlated with ICU volume. More experienced clinicians may be better at recognizing and treating the complications of critical illness or they may be better at translating evidence into practice.¹⁷  Thus, our universal intensivist coverage may improve overall outcome; however, patients at high-volume hospitals may be exposed to physicians in-training more frequently than those in low-volume hospitals. Recent reports have demonstrated that periods of major change in resident surgical staff are associated with increased risk-adjusted in-hospital mortality after complex cardiac operations.¹⁸  It is conceivable that a similar relationship might exist in the ICU.

A closed-ICU model has been reported to improve outcome in critically ill patients.^19,20  In our healthcare system, nearly all ICUs are staffed by intensivists 24 h a day, 7 days a week,¹⁴  whereas in the United States most small-community hospitals lack full coverage by trained intensivists. Some observational studies have shown that the presence of a trained intensivist is associated with lower ICU mortality.²¹  Consequently, some groups advocate using the permanent presence of intensivists as a measure of hospital quality.²²  Moreover, although patients with acute lung injury cared for in-closed model ICUs are more likely to receive lower tidal volume mechanical ventilation, the difference in delivered tidal volume did not completely account for the improved mortality observed in closed-model ICUs.¹⁹ 

If outcome is partly related to exportable factors such as protocols, guidelines, and multidisciplinary care models, then centers with low volume of mechanically ventilated patients might achieve the same outcome as the others by adopting these practices.²² 

Like all analyses of the association between volume and outcome, our study cannot determine the direction of the association.²³  Some investigators suggested that higher-quality hospitals might attract more patients on the basis of superior care when performance data are publicly reported.⁹  Nevertheless, our mainly public healthcare system is not influenced by this issue, as each hospital provides care for its catchment area, and very few patients seek care outside their own area.

Our results agree with those of two previous studies that found no association between volume and outcome in critically ill patients.^7,24  Many patients are admitted to the ICU for observation only, especially those not requiring mechanical ventilation or other interventions. As Kahn et al.⁹  pointed out, an association between ICU volume and outcome would be unlikely in a population of patients who are at low risk for death and who do not receive active intervention in the ICU. Nevertheless, the ICU case mix is highly influenced by the rate of ICU beds per population, as was recently demonstrated by Wunsch et al.,²⁵  who reported a clear-cut difference in the severity of illness, age, length of ICU stay, and outcome between ICUs in the United States and United Kingdom. Conceivably, our Spanish healthcare system is more similar to the United Kingdom system than to the U.S. system.

The predicted risk of death was inversely associated with volume, both in the general ICU population and in mechanically ventilated patients, being higher in very low-volume ICUs. It is commonly accepted that mortality-prediction models are reliable at any level of severity. Moreover, the concentration of very sick patients in a given unit has been suggested to increase ICU-acquired infections by cross-contamination and that this increase is even greater in cases of nursing shortage or excess ICU workload,²⁶  which can worsen the outcome of severely ill patients.

Nguyen et al.²⁷  recently reviewed three alternative organizational models that may expand access to high-quality critical care: tiered regionalization, ICU telemedicine, and quality improvement through regional outreach. They conclude that existing evidence does not strongly support the exclusive use of a particular model. We reemphasize the need for strong data about the relationship between ICU volume and outcome in a given healthcare system before deciding whether ICU regionalization is the best solution.

The hospitals analyzed in this study were not a random sample of all hospitals in Spain. After a formal country-based invitation, each hospital decided to participate in the Sabadell score study. Nevertheless, the wide variation in ICU size, academic affiliation, and population coverage strongly suggests that this sample could be representative of a large proportion of Spanish ICUs.

The short time span of our study (3 months in the spring) may introduce a seasonal bias, mainly in the proportion of mechanically ventilated patients or those undergoing scheduled surgery. The use of different scoring systems may be a source of bias, but SMR was stable irrespective of the scoring system used in our original description.¹⁵  Moreover, no association was found between hospital volume and the scoring system used.

Patient referral practices could influence the results of this study. Commonly, high-volume hospitals are both more likely to receive patients transferred from another hospital and less likely to transfer patients from their ICUs than are other hospitals. Because transferred patients have a higher rate of death than predicted, the referral bias would tend to make high-volume hospitals seem to have worse risk-adjusted mortality.²⁸ 

The worsened risk-adjusted outcomes at ICUs with a very low volume of mechanically ventilated patients may simply reflect less-accurate coding of the severity of illness (i.e., designating an illness as less severe than it is) at these centers.²⁹  Nevertheless, investigators in each ICU were unaware of plans for this secondary analysis, making any pretended “up-coding” in high-performing hospitals unlikely. Additionally, our public system offered no budgetary incentives to hospitals based on coding strategies, which could bias our results; nevertheless, we have no data about quality control of the accuracy of coding. Very recently, Breslow and Badawi³⁰  stated that “simple SMRs are disproportionately affected by outcomes in high-risk patients, and differences in population composition, even when performance is otherwise identical, can result in different SMRs.” Consequently, we cannot rule out whether our observed worse SMR in low-volume hospitals was an expression of poor performance or simply a mathematical coupling due to a higher proportion of high-risk patients.

In conclusion, we confirmed the wide variability in outcome among ICUs, even after adjusting for severity and confounding factors, but we found no relationship between ICU volume and outcome in our healthcare system. Only mechanically ventilated patients in very low-volume centers experienced slightly worse outcome.