Treatments and outcomes of patients admitted to the intensive care unit (ICU) with preexisting treatment limitations.
aData were missing for 47 patients.
All plots represent point estimates and 95% CIs for each included ICU. A, Among 109 ICUs. B, Among 137 ICUs. C, Among 117 ICUs including only survivors of their hospitalization. D, Among 128 ICUs including only survivors of their hospitalization.
eFigure. Number of Patients, Intensive Care Units, and Hospitals in the Study
eTable 1. Strength of Patient Characteristics Alone in Predicting Events of Interest
eTable 2. Selected Center-Level Characteristics Associated With the Management of Patients Entering the Intensive Care Unit With Limits on Care
eTable 3. The Association of Year With Admission and Management of Critically Ill Patients With Preexisting Treatment Limitations
Customize your JAMA Network experience by selecting one or more topics from the list below.
Hart JL, Harhay MO, Gabler NB, Ratcliffe SJ, Quill CM, Halpern SD. Variability Among US Intensive Care Units in Managing the Care of Patients Admitted With Preexisting Limits on Life-Sustaining Therapies. JAMA Intern Med. 2015;175(6):1019–1026. doi:10.1001/jamainternmed.2015.0372
Although the end-of-life care patients receive is known to vary across nations, regions, and centers, these differences are best explored within a group of patients with presumably similar care preferences.
To examine the proportions of patients admitted to the intensive care unit (ICU) with limitations on life-sustaining treatments and the proportions of such patients who receive aggressive care across individual ICUs.
Design, Settings, and Participants
Retrospective cohort study using the Project IMPACT database (from April 1, 2001, to December 31, 2008) including 141 ICUs in 105 hospitals in the United States and 277 693 ICU patient visits. We used logistic regression analysis models adjusted for available patient characteristics and clustered visits by individual ICU. The full analysis was last performed in October 2014.
Main Outcomes and Measures
Outcomes included the provision of (1) cardiopulmonary resuscitation, (2) new forms of life support, and the (3) addition or (4) reversal of treatment limitations.
Of the ICU admissions evaluated, 4.8% (95% CI, 4.7%-4.9%) had previously established treatment limitations. Patients admitted with treatment limitations were more likely to be older with more functional limitations and comorbidities. Among patients who survived to hospital discharge, more experienced reversals of existing treatment limitations during the ICU stay (17.8% [95% CI, 17.0%-18.7%]) than additions of new limits (11.7% [95% CI, 11.1%-12.4%]) (P < .01). Among patients who died, 15.7% (95% CI, 14.7-16.8%) had received cardiopulmonary resuscitation. After risk adjustment, ICUs varied widely in the proportions of patients admitted with treatment limitations (median, 4.0%; range, <1.0%-20.9%), the proportions of those who received cardiopulmonary resuscitation (37.7% [95% CI, 3.8%-92.4%]), the proportions of new forms of life support (30.0% [95% CI, 6.0%-84.2%]), and, among survivors, the proportion who had new treatment limitations established (11.2% [95% CI, 1.9%-57.3%]) and reversal of treatment limitations during or following ICU admission (20.2% [95% CI, 1.8%-76.2%]). The observed variability could not be consistently explained using measurable center-level characteristics.
Conclusions and Relevance
Intensive care units vary dramatically in how they manage care for patients admitted with treatment limitations. Among patients who survive, escalations in the aggressiveness of care are more common during the ICU stay than are de-escalations in aggressiveness. This study cannot directly measure whether care received was consistent with patients’ preferences but suggests that ICU culture and physicians’ practice styles contribute to the aggressiveness of care.
Several aspects of end-of-life and critical care have been shown to vary across nations,1-3 regions,4-6 and centers.5,7-12 Such variability has been interpreted as suggesting that similar patients are treated differently owing to factors such as local policy, practice culture,12 or resource constraints. However, these studies have been limited by the inability to focus on patients with presumably similar preferences for end-of-life care. As a result, the observed variability may represent appropriate responses to patients’ preferred approaches to care. Examining populations of patients with specifically documented preferences, such as those with previously established do not resuscitate (DNR) orders or other preferences for limited life-sustaining treatments, may help differentiate due from undue variability.
Indeed, such patients may be particularly vulnerable to variations in care delivery, and small studies13-16 of critically ill patients have demonstrated that DNR orders influence the site and type of care patients receive. A study17 from Australia and New Zealand found that nearly half of patients admitted to the intensive care unit (ICU) with preexisting limitations on treatments survive to hospital discharge and that nearly one-third are discharged directly home. This finding suggests that, although such patients’ preferences may preclude some treatments used commonly in the ICU, these patients may nonetheless benefit from critical care.18
No large-scale studies have described how US hospitals and ICUs manage care for patients with expressed preferences for treatment limitations. We therefore sought to determine the proportion of ICU patients who are admitted with existing treatment limitations and the variability in how such care for these patients is managed in the ICU. Specifically, we explored rates of cardiopulmonary resuscitation (CPR), initiation of new forms of life support, and, among surviving patients, the implementation of additional limits on care and reversal of previously expressed limits. We sought to identify characteristics of hospitals and ICUs associated with these practices.
Using the Project IMPACT database (Cerner Corporation), we conducted a retrospective cohort study of US ICU patient visits between April 1, 2001, and December 31, 2008. Project IMPACT was a voluntary, fee-based, ICU clinical information system used for benchmarking and research that used a trained data collector to input data regarding individual patients, processes of care, and center characteristics into a standardized, web-based instrument. IMPACT ICUs were nationally representative, and prior studies have validated key fields.7,19-21 The institutional review board of the University of Pennsylvania approved this study with a waiver of informed consent. Deidentified patient data were used for analysis, which was last performed in October 2014.
We excluded patients from analysis if they were ineligible for severity of illness assessment as calculated by the Mortality Prediction Model III or admitted to the ICU for the purposes of organ donation. To further augment the stability of statistical models, we excluded ICUs that collected data for less than 1 year, contributed fewer than 20 patient visits per quarter-year, or had fewer than 5 patient admissions with limitations on desired treatments during their participation in the database. These exclusions resulted in a refined population of 277 693 patient admissions within 141 ICUs in 105 hospitals (eFigure in the Supplement). Twenty-three of these hospitals had more than 1 ICU, with a maximum number of 4. We explored variation at the ICU level, including appropriate hospital-level characteristics (eg, academic affiliation or region) as qualities of each ICU. Patient admissions were classified as having a treatment limitation if, at the time of ICU admission, their code status was recorded as either DNR (defined as no chest compressions, no intubation, and no electrocardioversion), withholding therapy, or withdrawing therapy. Visits classified as withholding therapy may or may not have specified preferences to withhold chest compressions.
We explored several aspects of the care of critically ill patients with treatment limitations at the time of ICU entry. First, we explored the proportion of such individuals among all those admitted to ICUs. Second, we examined the proportions of patients admitted with treatment limitations (largely DNR orders) who received CPR and the proportions who received other new forms of life support during their ICU stays: (1) vasoactive medications, (2) mechanical ventilation, or (3) new hemodialysis, which excludes routine dialysis for patients receiving it before the ICU admission. Finally, we examined the proportions of patients who survived to hospital discharge who had further treatment restrictions placed during the hospitalization and who, during the ICU stay, had reversals of treatment limitations that were present at ICU admission. These outcomes could be measured only among patients surviving their ICU stay because Project IMPACT did not consistently collect code status designations at the time of death, with these data missing in 59.3% of patients admitted to the ICU with treatment limitations who died in the hospital. Patients’ treatment limitations were therefore recorded on ICU admission and, among surviving patients, at the time of ICU discharge. We designated patients who survived to hospital discharge as having further limitations on care placed if (1) the limitations on life-sustaining therapy at the time of ICU discharge were greater than those at the time of ICU admission or (2) if the patient received hospice services at the time of hospital discharge. Finally, we considered surviving patients to have undergone a reversal of treatment limitations if they were discharged from the ICU with less restrictive orders than at the time of ICU admission. Treatment limitation orders were not recorded at the time of hospital discharge.
We included the following variables in the model to account for differences in care owing to patient-level variation: sex, age, race, insurance type, functional status (degree of dependency) and location prior to ICU admission, type of ICU admission (medical, planned postoperative, or unplanned postoperative), diagnosis (1 of 9 categories, eg, respiratory arrest/failure, metabolic/renal, hemorrhage, or postoperative observation), and severity of illness (Mortality Prediction Model III score). In addition, preexisting and comorbid conditions, such as chronic cardiovascular, pulmonary, renal, or oncologic conditions, were included. The year of ICU admission was also included to account for any differences over time.
We explored center-level characteristics, which included those specific to the ICU or hospital, as independent predictors of the outcomes, including those directly available in the database, such as the ICU’s critical care staffing model (closed or open), size of the ICU and hospital, availability of a step-down unit or hospitalist services, academic affiliation, region, and environment (urban, suburban, or rural setting). We constructed 2 additional variables. The first was a technology index describing the degree of advanced resources available at the centers evaluated including surgical subspecialty services, cardiology services, and imaging modalities. We weighted each of 9 resources equally, resulting in each center receiving a technology index score of 0 to 9. We also explored the role of ICU census in relationship to our selected outcomes. To do this, the ICU census in the 2 hours before each patient’s admission was measured and coded as a proportion of that ICU’s median patient census during the duration of that ICU’s participation in the database. This patient-specific relative census was used to determine whether a limitation in available resources at the time of admission may influence admission decisions and the services provided to this patient population.
Descriptive statistics and comparisons were summarized using a 2-tailed t test or χ2 methods as appropriate. To evaluate the performance of a model comprising patient characteristics in predicting the likelihood of the selected care patterns, we measured the area under the receiver operating characteristic curve for predicting each outcome (eTable 1 in the Supplement). In addition, we confirmed measurement in the main model using intraclass correlation coefficients (ICCs). The ICC does not decrease after adjusting for patient characteristics (null model ICC = 0.19, after addition of patient characteristics ICC = 0.18). Furthermore, we performed a sensitivity analysis eliminating all hospitals containing more than 1 ICU to remove any residual clustering by hospital.
For each care pattern, we created logistic models that contained the set of patient-level characteristics. The ICU was first entered as a fixed effect to quantify center-level differences in the proportion of each outcome. The predicted probability for each ICU from this model is referred to as the adjusted ICU-level proportion of each outcome of interest. These proportions are estimated using the margins after estimation package in Stata, version 12.1 (StataCorp), and 95% CIs for these estimated proportions were calculated using a modified form of the Δ method.22 In the event the ICU had a single event with the selected outcome, the ICU was dropped from the fixed-effect model given modeling limitations. In the case of admission practices, we performed both ICU- and hospital-level analysis given that triage and admission practices likely reflect hospital and community practice rather than ICU-specific practice. There were no marked differences between the analyses. In our estimate, the range of predicted proportions of patients admitted with limitations on preferred treatments among ICUs is less than 1% to 20.9% and among hospitals is less than 1% to 17.2%. For consistency with the remainder of the results, we report only the ICU-level analyses below.
Once variation among centers was revealed through the fixed-effects modeling, we turned our attention to the quantification of possible associations of center-level characteristics with each outcome. Thus, the ICU was modeled as a random effect, which accounts for clustering of patient outcomes from each ICU by estimating a latent distribution of all outcomes, thereby permitting estimation of associations with center-level characteristics of interest.
Overall, 4.8% (95% CI, 4.7%-4.9%) of ICU admissions were patients with preexisting limits on care. Such patients had a median age of 78 years, 59.4% had dependent care needs at the time of admission, and nearly all had preexisting conditions, the most common of which were chronic respiratory disease (13.8%), chronic renal impairment (13.3%), solid organ cancer (10.9%), and chronic cardiovascular symptoms (8.8%) (Table). Most of these patients entered the ICU via the emergency department (52.5%), with substantial proportions transferred to the ICU from the wards (18.3%) and following a procedure (14.4%).
The median Mortality Prediction Model III score for the estimated risk of death at the time of ICU admission was 30.2% (interquartile range, 20.3%-47.0%). The actual in-hospital mortality rate for patients admitted to the ICU with treatment limitations was 34.6% (Figure 1). Of patients who survived to hospital discharge, the median ICU length of stay was 45 hours (interquartile range, 25-76 hours) with a median hospital length of stay of 168 hours (96- 288 hours). Most of those surviving to hospital discharge required further inpatient care in a setting such as a nursing facility or rehabilitation center (52.7%), but nearly one-third were discharged directly home (33.1%). (Figure 1) Approximately 10% were discharged into hospice care, which may have been provided at home or in an inpatient hospice facility.
Most patients (77.4%) entered the ICU with DNR orders, which indicated preferences prohibiting chest compressions, intubation, and electrical cardioversion. The remainder of the patients had limits on acceptable therapies ranging from prohibitions on specific therapies, such as dialysis or nutritional support (21.3%), to an expressed preference for comfort measures only (3.9%). Among all patients admitted with treatment limitations, 23.3% (95% CI, 22.6%-24.0%) received CPR in the ICU. Among the patients with documented DNR orders (ie, excluding the 2850 patients with treatment limitations that did not specify preferences regarding chest compressions), 24.6% (95% CI, 23.8%-25.5%) received CPR and 15.7% (95% CI, 14.7%-16.8%) of those who died had received CPR. During the ICU admission, 40.9% (95% CI, 40.1%-41.7%) of patients who entered with treatment limitations received 1 or more forms of life support including vasoactive medications (27.9%-29.4%), mechanical ventilation (19.2%-20.5%), or initiation of renal replacement therapies (1.9%-2.4%) (Figure 1). A total of 11.7% (95% CI, 11.1%-12.4%) of surviving critically ill patients who entered the ICU with treatment limitations had further limitations placed before ICU discharge. A greater proportion (17.8% [95% CI, 17.0%-18.7%]; P < .001) of such patients had a reversal of previous treatment limitations during the ICU stay.
After adjusting for patient characteristics, wide variation remained in the predicted proportions of patient admissions with treatment limitations among all admissions to a particular ICU, ranging from less than 1% to 20.9% (median, 4.0%). After adjustment, ICUs also varied in the predicted proportion of such patients who received new forms of life-sustaining therapy, ranging from 6.0% to 84.2% (median, 30.0%) (Figure 2). The adjusted ICU-specific proportion of patients entering the ICU with treatment limitations who received CPR ranged from 3.8% to 92.4% (median, 37.7%) (Figure 2). Among patients entering the ICU with treatment limitations who survived to hospital discharge, ICUs also varied widely in the proportion of such patients who had additional limitations on life-sustaining therapies placed during or following care in the ICU, ranging from 1.9% to 57.3% (median, 11.2%), and who had reversal of limitations present on admission during or following ICU care, with a range of 1.8% to 76.2% (median, 20.2%) (Figure 2).
Assessment of the measured ICU-level characteristics failed to identify any characteristics that were consistently associated with the admission and management practices for critically ill patients with preexisting limitations on desired therapies (eTable 2 in the Supplement). Intensive care unit management by a critical care physician was associated with greater odds that patients admitted with treatment limitations would ultimately receive new forms of life support (odds ratio, 1.35; 95% CI, 1.19-1.53). A suburban setting, when compared to an urban setting, was associated with greater odds that patients surviving the ICU stay would have new treatment limitations established (odds ratio, 1.50; 95% CI, 1.02-2.22). However, there was no consistent effect on the likelihood of admission, provision of new forms of life support, or further treatment limitations based on academic affiliation, technologies available, or size of the hospital and ICU. The relative ICU census at the time of patient admission also demonstrated no effect.
In the sensitivity analysis eliminating the 23 of 105 hospitals (21.9%) containing more than 1 ICU, the magnitude of variability among ICUs remained largely unchanged for each outcome. For example, the range of predicted proportions of patients admitted with limitations on preferred treatments among ICUs remained the same, indicating that the eliminated ICUs were not at the extremes. We also explored whether there were any significant changes over time in our analyses by examining the year of study as a fixed effect. The results revealed little change in our selected outcomes over the included years (eTable 3 in the Supplement).
This study describes the critical care management for patients entering US ICUs with expressed preferences to avoid certain aggressive therapies and demonstrates considerable variability among ICUs in the types of care delivered to this population. Trends in the management of this specific patient population demonstrate that ICU care is more likely to result in escalation rather than de-escalation of treatment intensity among patients with previously expressed wishes limiting therapy. In addition, the wide center-level variation we found across these clinical conditions reveals that a particular patient may be admitted and treatment managed differently depending on the center.
We found that nearly 25% of ICU patients receive care that differs from what was apparently desired at the time of ICU admission, and that more than 15% of DNR patients who died in the ICU did so after receiving CPR. We cannot determine what proportion of such changes in care are attributable to high-quality shared decision making leading to updated patient-centered care plans. However, the observed high rates of care discordant with earlier preferences, reversals of treatment limitation decisions, and unexplained variability among ICUs in each of these domains suggests opportunities for improvement in ICU-based decision making about goals of care between ICU clinicians and family members.
The population we explored may be particularly susceptible to variations in clinical culture since ICUs and providers have well-established differences in their individual approaches to end-of-life care.4,9-12,16 Although we found that ICUs vary dramatically in the management of this patient population, we were unable to identify factors that systematically explain most of the center-level variation among ICUs, which is consistent with prior studies seeking to explore variation in care.7,9 As with any study of variation, there is no ideal proportion of critically ill patients with treatment limitations whose treatment should be managed with ICU care or new forms of life support nor an established ideal proportion who should receive CPR or additional limitations on care. However, this study adds to an evolving evidence base that suggests the need to better understand how ICU culture develops and to identify whether potential differences among providers within an institution may drive some of this variability across centers.
This study also shows that many patients entering the ICU with treatment limitations in place have outcomes following critical care that would be desirable to most individuals. As expected, those entering the ICU with expressed preferences for limited treatment were older, had more comorbid conditions and lower functional ability, and had a higher predicted risk of death from the acute critical illness. Nonetheless, nearly two-thirds of such patients survived their hospitalization, and one-third of the survivors were discharged directly to home. These findings lend support to the concept that the ICU may be an appropriate level of care for patients with treatment limitations, that such patients should not be uniformly triaged away from critical care in an effort to avoid an ICU-based death or a trial of advanced therapies,18 and that specific treatment preferences based on patients’ individual goals and values should be elicited before providing or denying critical care.
Finally, we found that nearly 60% of patients admitted to the ICU with treatment limitations did not receive new forms of life-sustaining therapies during the admission. Although some of these patients may have received other forms of care that could be provided only in the ICU, many of these patients may have received equally effective treatment without ICU admission. A more complete understanding of the triage process for patients with expressed preferences for limitations on aggressive therapies may lead to improvements in ICU resource utilization and improved provision of patient-centered care.
This study is limited in that we were unable to assess moment-to-moment changes in patient care preferences or the mechanisms of care escalation, de-escalation, and discordance with previously expressed wishes. Second, although decisions made during an ICU stay are likely to affect patients’ subsequent care, we could not evaluate changes in levels of care before or following the ICU admission. Third, unidentified patient factors may inflate the variation attributed to centers. However, the patient-characteristic models we generated demonstrate adequate fit to conclude that considerable amounts of the variation in care are not attributable to differences in case mix (eTable 1 in the Supplement). Fourth, because we cannot compare outcomes among patients receiving treatment in an ICU with those admitted elsewhere, we cannot quantify the benefit attributable to ICU admission for these patients.
In addition, despite our data suggesting that several center-level characteristics that may be hypothesized to influence the management of ICU patients generally did not do so, there may be unmeasured ICU-, hospital-, or community-level characteristics that lead to systematic differences in how the care for patients with limitations on life support is managed. For example, the database did not collect information on patients’ use of palliative care, chaplaincy, or other supportive services or the availability of such resources within centers. The availability and use of such resources may be a marker of or contribute to practice norms supportive of palliative care. The database also did not include patients’ socioeconomic demographics, such as educational level or income, which may be associated with treatment limitations and aggressiveness of care. Data on patients’ long-term outcomes following hospital discharge were also unavailable.
Finally, the data were collected over a several-year period ending in 2008. Advances in palliative care research and practice since that time may have decreased variation in the provision of such care as more providers and centers may have adopted more palliation-supportive practice norms. Alternatively, variation may have increased if the divide between the 2 extremes of approach has widened. Research investigating the effect of this increased attention to family- and patient-centered ICU care on practice variation is needed.10
Patients with preexisting treatment limitations represent a small but important minority of critically ill patients admitted to US ICUs. Many of these patients have favorable outcomes following ICU admission, but their care trajectories may be heavily influenced by the ICU in which they receive care. By examining how treatment escalations and de-escalations are decided upon for this patient population at different ICUs, future research may identify targets for interventions that promote more patient-centered decision making.
Accepted for Publication: November 17, 2014.
Corresponding Author: Joanna L. Hart, MD, MSHP, Perelman School of Medicine, University of Pennsylvania, 3400 Spruce St, Gates 5044, Philadelphia, PA 19104 (email@example.com).
Published Online: March 30, 2015. doi:10.1001/jamainternmed.2015.0372.
Author Contributions: Drs Hart and Halpern had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Hart, Quill, Halpern.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Hart, Harhay.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Hart, Harhay, Gabler, Ratcliffe.
Study supervision: Halpern.
Conflict of Interest Disclosures: None reported.
Funding/Support: Dr Hart was supported by Institutional National Research Service Award grant T32 HL098054 from the National Heart, Lung, and Blood Institute administered by the National Institutes of Health, Mr Harhay was supported by National Cancer Institute grant R01CA159932 (Dr Halpern), and Dr Halpern was supported by a grant from the Otto Haas Charitable Trust.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Previous Presentations: Preliminary results of the study were presented in poster format at the American Thoracic Society International Conference; May 21, 2013; Philadelphia, Pennsylvania; and the American Thoracic Society International Conference; May 18, 2014; San Diego, California.
Create a personal account or sign in to: