Over the past 50 years, critical care medicine has rapidly developed into a complex, resource-intensive, and multidisciplinary field. The care of patients has evolved with implementation of new monitoring devices and therapies based on the best available evidence. In addition, care has been affected by the introduction of new team members dedicated to the care of critically ill patients and specific protocols for care. In the setting of ever-changing practice, it is important to ask whether outcomes for our patients have improved.
Outcomes Measured in Critically Ill Populations
The most consistently described outcome in both observational and interventional studies is mortality, which is variably reported as intensive care unit (ICU) mortality, in-hospital mortality, or mortality within a fixed time limit (most often between 28 and 90 days, but sometimes longer ). This chapter focuses primarily on short-term mortality, still the most commonly used measure of success.
Mortality as an outcome has the advantages of objectivity and ease of measurement, but it may not be appropriate for every study, such as in studies of palliative care when unchanged or higher mortality may be acceptable. A focus on reporting mortality can misrepresent the effect of a given intervention if the period of measurement is too short (failing to identify all related mortalities) or too long (introducing confounding from other sources of mortality). Moreover, mortality may not be the focus of an intervention or improvement initiative.
Many other endpoints have been used to assess outcome in critically ill patients ( Table 3-1 ). Data on these endpoints may be more difficult to obtain but may hold greater significance for patients and their caregivers. The strength of these different approaches to outcome lies in the delineation of clear administrative, policy, and economic implications and the ability to determine if these variables overlap with patient-centered outcomes (such as length of stay in the hospital).
| Mortality | Processes of Care and Resource Use | Measures Related to Short- and Long-Term Quality of Life |
|---|---|---|
| ICU Hospital 28 or 30 days 60 or more days | ICU length of stay Hospital length of stay Time on a ventilator Ventilator-(or other) free days Iatrogenic complications Location after acute hospital discharge Long-term health care utilization Hospital costs Hospital readmission | ICU length of stay Hospital length of stay Time on a ventilator Ventilator-(or other) free days Iatrogenic complications Location after acute hospital discharge Physical or functional disability Hospital costs Hospital readmission Quality of dying and death Family satisfaction with ICU care |
Sources of Data
A wealth of data from various sources can be used to study critical care outcomes, including administrative data, prospectively collected clinical data, and control arms of randomized trials. Each data source has inherent strengths and weaknesses that may bias the conclusions regarding trends in mortality over time.
Administrative data are readily available from various government, public, and private sources but have important limitations. The quality of the data relies on documentation and coding by clinicians. Data acquired in this way may have low sensitivity for specific diagnoses and may be variable across individual physicians and hospitals. A related concern is the potential for “upcoding,” the practice of billing for more expensive diagnoses and services than provided. This (illegal) practice can create biases toward higher severity of illness. Changes in coding standards or payment incentives also may alter the use of a given diagnostic code without a change in true incidence of the condition. Finally, “extraction,” the identification of certain combinations of signs, symptoms, and diagnostic terms, may be used to identify complex clinical conditions from within administrative datasets. The algorithms used in this process vary in sensitivity and specificity, with consequences for measured incidence and outcomes. Outcomes derived from administrative databases are most meaningful when their data extraction methods have been validated with multiple clinical datasets and with consistent coding practices over time.
Clinical observational data can be used to study various risk factors and outcomes, but data collection is expensive and time consuming. Often, such data reflect the experience of either a single center or a few centers, and result may be poorly generalizable to other patients or institutions. Outcomes among patients randomized to receive placebo or “usual care” in controlled trials may be extrapolated to describe the natural history of a given condition. Data collected in this setting are prospective, clinically relevant, and frequently validated. However, because these patients must meet specific study inclusion criteria, they may differ significantly from the larger pool of critically ill patients with respect to severity of illness, age, comorbid disease, and sites of care delivery. Moreover, such studies frequently exclude patients with poor prognoses. A consistent outcome trend in all types of available data increases confidence in the conclusions drawn. When such consistency does not occur (i.e., a trend is apparent in one data type but not discernable in another), these concerns must be weighed for each study to adjudicate its quality.
Trends in Mortality
Critical care outcomes are generally studied with one of three approaches: examining outcomes among patients receiving ICU care for any reason, limiting evaluation to a specific subgroup of patients admitted to ICUs (e.g., septic shock requiring mechanical ventilation), or focusing on a specific critical illness that might necessitate admission to an ICU for a proportion of the patients (e.g., severe sepsis).
Trends for Patients Admitted to Intensive Care Units
Data showing trends over time for all ICU patients are sparse. Recent studies in which outcomes were examined over the past two decades have identified consistent changes in patient demographics and severity of illness. These differences must be accounted for when an attempt is made to determine whether outcomes have improved. A study by Zimmerman et al. examined trends in in-hospital mortality among 482,601 patients admitted to U.S. ICUs between 1988 and 2012. Despite increases in severity of illness and patient age over the study period, the investigators found significant decreases in all-cause acute hospital mortality as well as in ICU and hospital lengths of stay. However, these observed improvements were partially attributable to higher rates of discharge to skilled nursing facilities. Mortality in such facilities is known to be high; therefore, although these data are clear in showing a decrease in acute hospital mortality for ICU patients over this period, we cannot conclusively determine whether overall short-term mortality decreased.
Likewise, in a retrospective analysis of a large ICU patient database in Australia and New Zealand between 2000 and 2012, Kaukonen and colleagues observed decreased crude and adjusted in-hospital mortality and, with the exception of patients with severe sepsis or septic shock (who were more likely, over time, to be discharged home), increasing rates of discharge to rehabilitation facilities. In the United Kingdom, work by Hutchings et al. demonstrated lower risk–adjusted ICU and hospital mortality for critically ill patients between 2000 and 2006 despite a constant severity of illness. This decrease in mortality was specifically attributed to changes in the system of care, including an increase in the number of ICU beds in the country and other systems interventions, such as critical care networks and rapid response teams.
Perhaps the most compelling evidence of improving short-term mortality for critically ill patients is the “drift” or “fade” of severity of illness scores over time. Many of these scores (e.g., the Simplified Acute Physiology Score [SAPS] and the Acute Physiology and Chronic Health Evaluation [APACHE] ) have been recalibrated multiple times over the past 20 to 30 years to maintain predictive accuracy. The model drift (in general) has been toward overprediction of mortality, leading to a progressive overestimation of predicted mortality that affects the accuracy of severity of illness adjustments between historical cohorts. Although subtle shifts in case mix may account for some of these changes, this trend adds weight to the suggestion in the studies previously described that overall short-term mortality has decreased over time.
Trends for Specific Critical Illnesses
Changes in outcomes have been assessed for many ICU-specific illnesses. This chapter focuses on two common diagnoses: septic shock and acute respiratory distress syndrome (ARDS). A systematic review by Friedman and Vincent published in 1998 examined trends in septic shock mortality with 131 studies published between 1958 and 1997. The authors found an overall mortality of 49.7%, decreasing mortality over time, and changes in infection site and causative organisms; however, they noted significant heterogeneity in definitions of disease and severity of illness between studies. Because the American College of Chest Physicians’ and Society of Critical Care Medicine’s 1991 European Consensus Conference definitions of sepsis, severe sepsis, and septic shock have been widely adopted, comparison of outcomes over time has become a little easier, although patient populations in individual studies remain heterogeneous because of variable interpretation of aspects of the definition, such as “hypotension” and “unresponsive to adequate resuscitation.” An additional marker of possible decreasing mortality for patients with septic shock is that the mortality for the usual care arms of studies designed to capture this population has steadily decreased over time.
As with septic shock, the assessment of ARDS mortality is confounded by changes in clinical definitions with time, and trends for mortality associated with ARDS are even less consistent. A study by Milberg et al. analyzed the etiology of ARDS and outcomes in the Harborview Medical Center ARDS registry and found decreases in crude and adjusted mortality between 1983 and 1993 despite increasing severity of illness. Since the publication of that study, our understanding of the pathophysiologic features of ARDS and the role of ventilator-induced lung injury in patients susceptible to ARDS has significantly grown. The resultant implications for ventilator management and adjunct interventions for ARDS may affect outcomes and outcomes assessment. Despite advances in understanding and options for care, recent evidence in temporal ARDS outcomes does not consistently demonstrate large improvements in mortality.
When randomized controlled trials are considered in isolation, short-term mortality among patients with ARDS does appear to be improving. Examining 2451 patients enrolled in ARDS Network randomized controlled trials, Erickson and colleagues found decreased raw (from 35% to 26%) and adjusted 60-day mortality despite increased severity of illness; this trend was evident even with inclusion of patients who received high tidal volume ventilation (12 mL/kg), a finding that led the authors to conclude that observed decreases in mortality were due to generalized improvements in critical care delivery at participating hospitals rather than specific interventions for ARDS.
Two systematic reviews (incorporating both trial and observational evidence) on ARDS mortality provide conflicting results. One reported a 1.1% annual decrease in mortality between 1994 (the year of publication of the European-American Consensus definitions ) and 2006, whereas the other found no significant change in mortality among 18,900 patients. Moreover, an observational study of 514 patients with ARDS in Olmsted County, Minnesota, between 2001 and 2008 also failed to identify a significant change in hospital mortality over time.
ARDS remains a heterogeneous syndrome involving subjective assessment and many causes. These inconsistencies may explain the conflicting conclusions in different studies. The development of electronic “sniffers”—programs that automatically process real-time clinical data from electronic medical records to alert clinicians to the potential presence of ARDS —may provide more consistent identification of patients and thus a more accurate assessment of trends in mortality.
Trends for Diagnoses with Variable Admission to Intensive Care Units
The decision to admit a given patient to an ICU is multifactorial. For example, many patients with severe sepsis are admitted to ICUs, but many patients with the same diagnosis are cared for in emergency departments, hospital wards, or step-down units. Mortality in these alternative treatment sites may be substantial.
Several large observational studies have described the epidemiologic features of severe sepsis in the United States over the past 30 years. Serial analyses of the Agency for Healthcare Research and Quality’s Nationwide Inpatient Sample (NIS) database, which includes data from 1993 to 2010, demonstrate increases in measured incidence of severe sepsis and severity of illness, as well as decreased hospital mortality. The largest of these studies, by Stevenson and colleagues, included both NIS data collected between 1993 and 2009 and a meta-analysis of more than 14,000 patients enrolled in usual care or placebo arms of 36 multicenter randomized controlled trials worldwide. The authors observed differences in effect size between observational and trial data but consistent, significant decreases in overall mortality, regardless of data or the administrative coding method used. Likewise, in a study with clinical and administrative data sampled from a cohort of more than 1 million patients admitted to two U.S. medical centers between 2003 and 2012, Rhee et al. found decreased hospital mortality among patients with severe sepsis.
A study of 92,000 adults with severe sepsis admitted to 240 ICUs in England, Wales, and Northern Ireland between 1996 and 2004 identified an increasing proportion of ICU admissions with sepsis. Mean patient age increased over time, but there was no change in severity of illness (as described by the APACHE II score) or the extent of organ dysfunction on admission. Importantly, unadjusted ICU and hospital mortality also were unchanged. Data from Australia and New Zealand in which 100,000 ICU patients with severe sepsis were examined between 2000 and 2012 similarly showed an increasing rate of ICU admissions with severe sepsis. However, this study found decreasing rates of crude and adjusted mortality that paralleled overall ICU mortality trends and increased rates of discharge to home.
The “Will Rogers phenomenon,” in which earlier diagnosis of a given condition leads to an observed increase in measured incidence and decreased mortality, may play a role in observed increases in severe sepsis incidence. Growing clinician and hospital awareness of severe sepsis with an emphasis on early diagnosis and intervention may decrease observed overall severe sepsis mortality because of the addition of a group of patients with less severe disease and lower expected mortality to a pool of previously identified, sicker patients. Appropriate risk adjustment may help to minimize this issue, but such a phenomenon remains a concern.
Related posts:
What Lessons Have Intensivists Learned During the Evidence-Based Medicine Era?
Is Extracorporeal Life Support an Evidence-Based Intervention for Critically Ill Adults with ARDS?
What Fluids Should I Give to the Critically Ill Patient? What Fluids Should I Avoid?
Should Exacerbations of COPD Be Managed in the Intensive Care Unit?
Are Anti-inflammatory Therapies in ARDS Effective?
Has Outcome in Sepsis Improved? What Has Worked? What Has Not Worked?
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
