ICU Occupancy and mechanical ventilator use in the United States

Abstract

Introduction

The occupancy of US intensive care units (ICUs) dictates care decisions for patients under routine circumstances and also has implications for disaster planning (1, 2). The United States has more intensive care unit (ICU) beds – approximately 20 per 100,000 population – than most other nations (3), and the number of ICU beds in the country has steadily increased over time (4, 5). Despite an overall occupancy estimated at 68% in 2005, many clinicians and policymakers, and perhaps even patients, perceive that the supply of US ICU beds is still insufficient to meet normal temporal variation in critical care demand (6). Such views are bolstered by reports of how limited bed availability may influence care decisions and concern regarding the associated outcomes (7–9).

The apparent need to triage patients under “routine” conditions suggests that a large increase in ICU beds and mechanical ventilators may be needed to accommodate surges in critical care demand brought on by mass casualties or epidemics. However, specific data to guide disaster planning have been limited. A recent taskforce on disaster planning estimated that fewer than 70% of occupied ICU beds were filled by patients requiring mechanical ventilation at any one time, but more precise estimates were not available (10, 11). Furthermore, it is unknown what proportion of admitted patients in US ICUs need ICU beds immediately (e.g., patients with respiratory failure) or might have their admissions deferred (e.g., planned admissions for patients following elective surgeries) under situations of unexpected need.

Current estimates of occupancy are also limited in that they do not consider potential variation in utilization patterns across ICUs of different type or size. Perhaps most importantly, robust data are lacking regarding the utilization of mechanical ventilators in the US. All of these details are essential to understanding the flexibility of a system composed of many ICUs. Therefore, to inform decisions on how best to plan for normal variation in critical care demand and to guide regional and national disaster planning (12), we sought to provide estimates of ICU bed occupancy across ICU types, use of mechanical ventilators, and to determine the proportion of ICU beds used for elective surgical admissions across a representative sample of US ICUs.

Methods

Analysis of bed use

We calculated occupancy/available beds for each hour of each day in the ICU by counting patients who were in the ICU on the first minute of each hour. For the 9.8% of patients who were missing a date or time of ICU discharge, we imputed these values based on the mean length of stay of patients in that individual ICU. We defined bed use (occupancy) as the percentage of beds in use out of the total number of “operational beds” as reported by each ICU. Operational beds are defined in Project IMPACT as “the total number of licensed ICU beds in your unit for which you maintain nursing staff and which are considered operational.”

We assessed the use of ICU beds across ICUs in three different ways. First, we examined occupancy in terms of the total “system” of ICUs by calculating percent occupancy as the total number of occupied ICU patients in the cohort in a given hour divided by the total number of operational ICU beds that hour. We used this approach to assess temporal trends in overall ICU occupancy during the three years of the study. We show this graphically with plots of all hourly data (Figure 1), accompanied by the moving weekly averages over the study period. To test for significant differences in occupancy over time, we created linear regression models clustered on ICU using the sandwich variance estimator and regressed hourly occupancy data on calendar year, which was modeled as a continuous variable. We also assessed differences in occupancy by day of the week (weekday versus weekend), and examined occupancy during days within the probable influenza seasons (December 1–March 31 of each year).

Open in a separate window

Figure 1

Percentage of all ICU beds (pooled data from all ICUs) that were occupied each hour 2005–2007. Smoothed lines are one week moving averages.

Second we calculated the mean hourly occupancy for each ICU, and summarized these data by presenting the mean (standard deviation (± SD)) and median (with interquartile range (IQR)) of these ICU-specific means across all ICUs. We explored variation in mean and median occupancy among hospital types (grouped as university affiliated, community not-for profit, community for-profit, or government), ICU types (grouped as medical, surgical, mixed, or cardio-thoracic), and ICU size (grouped as 5–14, 15–19, and 20+ beds, with groupings based on inspection of the distribution of ICU sizes and occupancies). To assess statistical differences between ICU and hospital types, we used linear regression clustered on ICU using the sandwich variance estimator, with ICU and hospital types and size as categorical variables. Third, we examined individual ICUs to determine the percentage of hours each ICU had 0, 1, 2, 3 or 4+ open beds, and the percentage of hours with different rates of occupancy (e.g., < 50% or > 90%). For units that reported periods of > 100% occupancy, we assigned zero available beds for those hours.

Results

The cohort consisted of 226,942 patients admitted to 97 ICUs from January 1, 2005 through December 31, 2007 (Table 1). Majorities of patients were male, Caucasian, and were admitted for medical rather than post-surgical reasons. The mean age was 59.8 (± 18.3). Patients’ average MPM₀-III predicted probability of death was 13.9%, their observed hospital mortality was 14.1%, and they had a median ICU length of stay of 2.0 days (IQR 1.0–3.9).

Individual ICU Occupancy

The mean hourly occupancy for ICUs was 68.2% ± 21.3, with a median of 74.6% (IQR 53.4–86.1%, and full range of 17.4% to 109.0%) (Figure 2; eTable 2). There was variation in mean hourly occupancy across ICU types, from 62.9% (± 21.9) among mixed medical-surgical ICUs to 74.0% (± 17.6) among surgical ICUs (P = 0.036). Smaller ICUs had higher mean occupancy (75.8 ± 16.5%) for ICUs with 5–14 beds, versus 60.9 ± 22.1% for ICUs with 20+ beds, P < 0.001). ICUs in academic hospitals had the highest mean occupancy (78.7% (± 15.9)). All supplemental analyses yielded similar estimates of overall occupancy (eTable 3).

Open in a separate window

Figure 2

Mean hourly occupancy of ICUs (bars represent the standard deviations for the means across ICUs).

There was considerable variation in hourly bed availability across ICUs (Figure 3). Most ICUs 52 (53.6%) had ≥ 4 beds available at least 50% of the hours (light yellow), whereas only 5 (5.2%) ICUs had no beds available more than half of the hours (dark grey). Comparable assessments of hourly occupancy rate (instead of open beds), yielded similar variations across ICUs (eFigure 1).

Open in a separate window

Figure 3

Percentage of hours with available beds (0, 1, 2, 3, 4+) in each ICU. Each column represents a single ICU.

Use of mechanical ventilation

A mean of 39.5% (± 15.2) of ICU patients were mechanically ventilated in any given hour (eTable 4). Patients in mixed medical-surgical ICUs were the least likely to be receiving mechanical ventilation (36.4% (± 14.1)) and patients in surgical ICUs the most likely (46.7% (± 14.4), P = 0.012). ICUs in academic hospitals had the highest percentage of mechanically ventilated patients each hour (50.8% (± 13.5)).

At any time, 29.0% (± 15.9) of ICU beds were occupied by patients receiving mechanical ventilation (Figure 4; eTable 4). This ranged from a mean of 24.5% (± 15.3) in units with 20+ beds to 33.8% (± 15.7) of patients in units with 5–14 ICU beds, P = 0.004. ICUs in academic hospitals again had the highest percentage of beds filled with a patient requiring mechanical ventilation (mean 41.0% (± 15.3)).

Open in a separate window

Figure 4

Mean hourly occupancy of ICUs with mechanically ventilated patients (bars represent the standard deviations for the means across ICUs)

Discussion

In providing national estimates of real-time ICU occupancy and mechanical ventilator use in the US, this study yields four primary findings. First, at any given time, approximately two thirds of ICU beds are occupied, and approximately one third of these beds are occupied by patients requiring mechanical ventilation. These results suggest greater excess capacity in the US critical care supply than many have imagined. Second, this excess in critical care supply is increased if one considers the large proportion of potentially “discretionary admissions”. Third, there appears to be little seasonal variation in critical care occupancy or mechanical ventilator use. This should facilitate local planning for the provision of routine critical care, and regional or national planning for mass casualties, by suggesting that seasonal contingencies need not be considered.

Finally, we found considerable variation in occupancy across ICUs. In particular, our data support many clinicians’ perceptions that ICU bed allocation decisions are common in academic medical centers, where ICUs tend to operate at significantly higher occupancy rates. Although the high volume and expertise clustered in academic centers may be beneficial for patients on average (17), these units may be less able to safely accommodate sudden increases in critical care demand. This raises the possibility that when these units are near capacity, individual patients may be better served in one of the many available beds elsewhere.

The likelihood that a given patient will have a requested ICU admission delayed or refused is a function of both occupancy and the total number of beds in the system (18). One unexpected finding was that smaller ICUs in our cohort tended to have higher occupancy rates than did larger ICUs. Queuing theory suggests that ICUs with more beds can operate at higher occupancy than can ICUs with fewer beds while maintaining the same probability of a delayed admission (i.e., a bottleneck) (18, 19). However, we found that the opposite pattern occurs in US ICUs. These results suggest that regional reallocation of critical care beds, without increasing the total supply of beds, may reduce the odds of delayed or deferred ICU admissions in both surge and non-surge situations. Implementation of improved and formalized triage and transfer systems, such as are used in some other countries (20), might also promote efficiency.

Prior studies have assessed daily occupancy using billing data, but with little distinction among types of ICUs, and without detailed information on the use of mechanical ventilation. For example, administrative data does not capture patients’ exact duration of mechanical ventilation (21). This may account for the differences between our estimates of the use of mechanical ventilators in US ICUs in comparison to estimates that have guided disaster planning to date (10). Our data suggest that the number of appropriately trained staff to care for these patients and manage their ventilators is far more likely to be the limiting factor than is the availability of fixed resources such as ICU beds or ventilators in the provision of critical care in the US during mass casualties (22, 23).

Recent evidence from Canada showed that patients experiencing sudden clinical deterioration on a hospital floor when fewer (or no) ICU beds were available were more likely to receive palliation instead of ICU admission, but were not more likely to die (1). This suggests that some patients who are admitted to the ICU when beds are plentiful may be too sick to benefit from intensive care (24). There are also many patients admitted to ICUs in the US who may be too well to benefit (25), as evidenced by hospitals admitting many patients with a low predicted risk of death (≤ 2%) (26). Our study complements these findings in showing that up to a third of patients admitted to ICU beds might be considered discretionary admissions. Among these, elective surgical patients represent the clearest group for potential deferral or diversion during mass casualties. We also considered medical patients admitted for observation as a potentially discretionary group because only 10–15% of such patients eventually require active treatment (15, 16).

Our study has limitations. ICUs in this study did not cluster geographically in a way that permits determination of the slack present within a given local area or region. Thus, although Project IMPACT provides the best available data for estimating national ICU occupancy and ventilator use, we cannot assess how ICU occupancy is balanced across ICUs within a small geographic area. Moreover, data suggest that ICU beds are not distributed equally in the US based on the underlying population in a given region (27). Prudent use of inter-hospital transfers (28), and of ICU beds within a hospital or local area may promote more efficient use of the system’s existing capacity (29).

Although Project IMPACT comprises a large and diverse sample of US ICUs – including ICUs in 35 states and the District of Columbia – it is not a random sample (30). In particular, the ICUs are skewed towards larger, academic ICUs compared with the country as a whole. (23) Given that academic ICUs were more likely to operate at higher occupancy, and with a greater proportion of patients mechanically ventilated, this means we have likely erred on the side of overestimating both occupancy and use of mechanical ventilators. The higher occupancy in academic ICUs compared with other types of ICUs also suggests that studies that solely involve academic ICUs may provide skewed estimates of use of mechanical ventilation and bed pressure.

A few ICUs reported very high occupancy rates, including rates greater than 100%. Although we cannot establish why these numbers arose, we speculate these are high-occupancy units that also commonly “board” patients in other parts of the hospital (31). We also cannot be certain that every ICU bed is equipped to support a patient requiring mechanical ventilation. However, a recent survey of ventilators available in US acute care hospitals found that approximately one (0.96) ventilator existed for every adult ICU bed in the country (32). Finally, the calculated occupancy rates depend on the number of “operational” beds available at any given time. In particular, we found some discrepancies; for the units where we were able to confirm the number of operational beds, in a small proportion the correct bed numbers were different from the original data in Project IMPACT. Our sensitivity analysis using only ICUs with validated bed numbers yielded very similar calculations of overall occupancy, as well as reported ranges within ICUs, supporting the robustness of our estimates. We cannot rule out the possibility, however, that the units we were unable to contact had systematic errors in reported bed numbers.

Using our occupancy estimates in combination with data regarding the total number of US ICU beds (5), we estimate that there are approximately 20,000 adult ICU beds available at any given time. As this number represents 8.5 available ICU beds per 100,000 adult population, the excess capacity in the US system is greater than the total capacity in many developed countries (3). This study shows that this substantial excess in capacity is consistent over time, and provides explanations for why triage is nevertheless needed in certain institutions; redistributing existing beds from low to high occupancy units may alleviate critical care bottlenecks.

Supplementary Material

Acknowledgments

Footnotes

References