Attention to concurrent care concerns (CCCs) by diagnosis. The association of CCCs with diagnosis was statistically significant (P= .001, Kruskal-Wallis test). DNR indicates do-not-resuscitate; HIV, human immunodeficiency virus.
Sulmasy DP, Sood JR, Ury WA. The Quality of Care Plans for Patients With Do-Not-Resuscitate Orders. Arch Intern Med. 2004;164(14):1573-1578. doi:10.1001/archinte.164.14.1573
Care plans for patients with do-not-resuscitate (DNR) orders often fail to define limits other than cardiopulmonary resuscitation and fail to address other patient care needs. We studied the explicitness and comprehensiveness of care plans for patients with DNR orders and what factors were associated with this aspect of the quality of their care.
A cross-sectional study was conducted at Georgetown University Medical Center (GUMC), Washington, DC, and St Vincent Catholic Medical Centers (SVCMC), St Vincent's Hospital–Manhattan, New York, NY. Participants included 189 consecutive medical inpatients with DNR orders.
A previously validated medical chart review technique termed concurrent care concerns (CCCs) measured whether 11 possible patient care needs had been addressed within 2 days of the DNR order. Reasons for the DNR order were documented in only 55% of cases, and a consent conversation was documented in only 69%. The mean number of total CCCs per DNR order was 1.55 (1.84 at GUMC and 1.29 at SVCMC; (P = .007). In a multivariate logistic regression analysis of low (≤1) vs high (≥2) CCCs, patients with malignancy (P = .002), higher APACHE III (Acute Physiology and Chronic Health Evaluation III) scores (P = .007), and a documented consent conversation (P = .009) and those at Georgetown (P = .005) were more likely to have high attention to CCCs. Patients with dementia were the least likely to have high attention to CCCs.
Documented consent conversations and care plans for patients with DNR orders are less than ideal. Care plans differ in quality by diagnosis, institution, and whether or not a consent conversation is documented. These observations might help to guide interventions that aim to improve the care of patients with DNR orders.
Patients with do-not-resuscitate (DNR) orders are very likely to die in the hospital, even when one adjusts for severity of illness, diagnosis, and other factors that might affect a patient's propensity to have a DNR order.1- 3 A DNR order is therefore an excellent marker for identifying inpatients at the very end of life,4 and studying such patients provides an excellent opportunity to learn about the quality of end-of-life care in the hospital.
A DNR order literally indicates only that if a patient sustains a full cardiopulmonary arrest, no resuscitative measures are to be undertaken.5- 7 However, it is often unclear whether the plan of care for a patient with a DNR order also includes limits on other life-sustaining treatments such as intubation for respiratory distress short of a full cardiopulmonary arrest or the use of pressor drugs. Care plans for patients with DNR orders should also attend to such patient needs as symptom relief, spiritual care, and the patient's preferred site of terminal care (eg, hospital or hospice). Whether these issues are addressed can be considered an indicator of the quality of end-of-life care. This has led our group and others8- 13 to recommend special order sheets, computer-order entry systems, and staff education designed to facilitate more explicit and comprehensive care plans.
The purposes of the present study were (1) to examine how explicit care plans are in defining the limits of life-sustaining treatments; (2) to examine how comprehensive the care plans are in addressing needs that are likely to be indicated for patients with DNR orders; and (3) to assess what clinical, sociodemographic, and professional factors were associated with more explicit and comprehensive care plans for patients with DNR orders.
The sample for this study was drawn from a larger study of hospitalized patients with serious illnesses and poor prognoses conducted at 2 urban teaching hospitals, Georgetown University Medical Center (GUMC) in Washington, DC, and Saint Vincent Catholic Medical Centers (SVCMC), St Vincent's Hospital–Manhattan, New York, NY. The larger study sample, described in greater detail elsewhere,14 consisted of consecutive English- or Spanish-speaking patients with DNR orders who could be spoken with or who had a surrogate who could be contacted between 2 and 5 days after the DNR order plus a convenience sample of similarly seriously ill English- orSpanish-speaking patients with poor prognoses who did not have DNR orders and could be spoken with or had a surrogate who could be contacted between 3 and 8 days after hospital admission. Of these patients or surrogates, 110 (30%) declined to participate. Participants did not differ significantly from nonparticipants in age, sex, race, or diagnosis in either group. However, there was a trend for GUMC nonparticipants with DNR orders to be more likely to be female (72% vs 58%; P = .06). The cross-sectional design of the study we report herein includes only those patients from this larger sample (n = 189) who had DNR orders and complete medical chart review data.
Medical residents at GUMC had received education about patient care plans for patients with DNR orders,12 but no specific education about palliative care or pain management. By contrast, residents at SVCMC had received a pain management curriculum,15 but no education about care plans for patients with DNR orders. Both institutions are tertiary-care major teaching hospitals for medical schools with a mix of private patients and those treated by the house staff service. This project was approved by the institutional review boards at both institutions.
Medical charts were reviewed for documentation of the DNR order, documented consent for the DNR order, attention to what we have termed the concurrent care concerns (CCCs), and clinical and sociodemographic data, including severity of illness as measured by APACHE III (Acute Physiology and Chronic Health Evaluation III) score.16 Diagnosis was coded as the patient's primary underlying condition. Attending signature was defined as a signed attending order or a cosignature within 2 days of the order having been written by a house officer. Attending physician type (generalist vs specialist) was coded according to departmental privilege lists and confirmed by a chief resident who was not a subject in the study.
The CCCs refer to an extensively used and previously validated medical chart review method with high interrater reliability (κ scores, 0.77-0.85).8,17,18 The method was validated by interviewing the patient, attending physician, intern, and nurse and comparing their understandings of the care plan with that written in the medical chart as assessed by our method.18
The CCCs assess whether there is medical chart documentation of attention to 11 indicators of the explicitness and comprehensiveness of the care plans within 2 days of the writing of the DNR order. The 7 discrete items measuring the explicitness of limits on life-sustaining therapy included use of intubation, dialysis, blood products, antibiotics, pressors, artificial hydration, and artificial nutrition. The 4 measures of comprehensiveness of the care plans include attention to analgesic and sedative needs, consideration of hospice, consideration of spiritual needs, and consideration of a decrease in the frequency of vital signs.
Attention to a CCC is evidenced in 1 of the following 3 ways: (1) an explicit order limiting an intervention (such as "do not intubate") or an explicit order for a positive patient intervention (such as "chaplain visit please"); (2) a documented conversation with the patient or surrogate indicating some discussion of whether that particular intervention was appropriate; or (3) a note documenting a team discussion or the thoughts of the attending physician considering whether the intervention was genuinely appropriate given the patient's situation, even if the ultimate decision was to provide it. For example, a simple order for a blood transfusion or the placement of a feeding tube is not scored as attention to the concern unless there is an indication that someone assessed the treatment in light of an overall care plan and the goals of treatment for a patient and at least hypothetically considered not giving the treatment.
We used descriptive statistics and nonparametric hypothesis testing. We used χ2 and Spearman ρ for bivariate analyses. We analyzed differences in total CCCs between hospitals using the Wilcoxon rank sum test and differences according to diagnosis using Kruskal-Wallis test. Because CCCs are highly skewed, we dichotomized total CCCs per DNR order as low (≤1) vs high (≥2) and estimated a logistic regression model of factors associated with low vs high attention to CCCs. The SPSS software package (SPSS Inc, Chicago, Ill) was used for all analyses.
As shown in Table 1, the mean age of our patients was 71.6 years; 62% were women; and 39% were nonwhite minorities. The most common diagnosis among these patients was malignancy, and 6% had a primary diagnosis of dementia. These patients were, in general, quite sick, with a mean APACHE III acute physiology score (APS) of 33.84. Forty percent had completed advance directives on the chart (35% had health care proxies, and 18% had living wills).
Of these 189 patients, 2% had unwritten DNR orders (ie, they were understood by the team to be ineligible for cardiopulmonary resuscitation even though a formal DNR order had never been written in the medical chart). Most orders (66%) were written by house officers (and, in 1 case, a medical student). Only 70% of orders were signed or countersigned by the attending physician. The reasons for the DNR order (eg, patient preferences or futility) were documented in only 55% of cases, and only 69% had documentation of a consent conversation with a patient or a surrogate.
In most cases (73%), the staff attended to at least 1 CCC, with a mean of 1.55 CCCs per DNR order. Staff at GUMC attended to significantly more CCCs than did the staff at SVCMC (1.84 vs 1.29; P = .005).
Table 2 shows the percentage of patients for whom each individual CCC was addressed. The most common explicitly defined limit other than cardiopulmonary resuscitation was intubation (42% of patients). Documentation of attention to patients' analgesic needs was present in only 37% of charts (the data were collected in 1998 and 1999, before the present pain standard of the Joint Commission on the Accreditation of Healthcare Organizations). Only 7% of the charts revealed explicit attention to patient spiritual needs, and only 1% attended to whether the frequency of vital signs should be reduced.
Also shown in Table 2 is the breakdown of attention to individual CCCs by hospital. Although the trend is generally that GUMC paid more attention to most CCCs, only the differences in intubation (P<.001) and pressors (P<.001) were significant. The one notable exception was that at SVCMC, significantly more attention was given to patients' analgesic needs (46% vs 26%; P<.001).
Figure 1 shows the breakdown of the significant differences by diagnosis, with the highest mean number of CCCs per DNR order associated with a diagnosis of cancer (2.08) and the fewest associated with a diagnosis of dementia (0.67). The association of CCCs with diagnosis was statistically significant (P = .001, Kruskal-Wallis test).
In bivariate analysis, hospital (P = .003), age (P = .008), diagnosis (P = .006), APS (P = .002), the presence of a consent conversation (P<.001), and attending physician (private vs house staff service; P = .08) showed an association or a trend for association with low vs high CCCs. Not significant were sex, race, religion, religiosity, insurance, and advance directive status.
The distribution of CCCs was highly skewed. Therefore, we elected to perform multivariate analyses by dichotomizing into low (≤1) vs high (≥2) number of CCCs per DNR order. We estimated a multivariate logistic regression model that included all variables that had a P value of .10 or less in bivariate analyses against the dependent variable of low vs high CCCs. For these analyses, diagnosis was coded as malignancy, human immunodeficiency virus or AIDS, dementia, or other. As shown in Table 3, attending physician type and age did not prove to be independently significant in multivariate modeling. However, patients with a higher APS, patients with malignancy, patients at GUMC, and patients whose medical charts documented a consent conversation with the patient or a surrogate were all independently more likely to be associated with greater attention to CCCs.
Our results indicate that the quality of care plans for patients with DNR orders was overall rather low, as evidenced by a lack of explicit delineation of limits on other life-sustaining treatments and a lack of comprehensive attention to patients' CCCs. Our data also seem to indicate that documentation of the reasons for the DNR order and consent conversations was also less than ideal. Finally, the quality of these care plans was associated with severity of illness, diagnosis, hospital, and documentation of a consent conversation.
Despite years of experience and education about DNR orders throughout the United States, 45% of the medical charts did not document the reasons for the DNR order, 31% failed to document an informed consent discussion, and 30% were neither signed by an attending physician nor countersigned within 2 days of having been written by a house officer. This is particularly noteworthy in that one of the study sites (SVCMC) is in New York State, which has an explicit DNR order statute that requires compliance with these widely accepted moral guidelines.7 The rates of noncompliance with documentation of consent that we have noted are consistent with rates documented at other institutions and are essentially the same as they were 25 to 30 years ago.19,20 Rates of unwritten DNR orders were negligible, a finding consistent with rates reported in these previous studies.
Rates of attention to the CCCs of patients with DNR orders were also rather low. We have previously shown that even if the attending physician reports having considered these CCCs, the fact that documentation about how to proceed is not recorded in the patients' medical charts is strongly associated with a lack of clear guidance about these issues in the minds of patients, nurses, and house officers.18 Do-not-resuscitate orders do not entail other limits on care and should not affect other aspects of care. For example, a DNR order by itself does not imply that these patients ought not be intubated for respiratory distress short of a full cardiopulmonary arrest or receive pressor drugs in the intensive care unit for hypotension. Despite many articles during the past decade describing the importance of care plans that clarify these issues,8- 13 our results suggest that this has not become part of routine practice.
The fact that relatively more attention was paid to CCCs at GUMC 3 years after the termination of an educational effort among house officers suggests that education can have sustained, if modest, effects. One year after the completion of this program of education and computerized order entry, the mean rate at GUMC had been 2.9 CCCs per DNR order.12 In the present study, the mean rate was 1.8. It should be pointed out that by 1998, GUMC had abandoned the computer order entry system that had automatic reminders for house officers regarding CCCs. Although some attenuation of the educational effect is reasonable to expect, demonstrated sustained changes in practice 3 years after an educational effort are rare. Perhaps successive generations of house officers were teaching this idea to their younger confreres. Nevertheless, 2.9 CCCs per DNR order might be considered suboptimal. All of these findings suggest a need for ongoing education.
We found independent associations between CCCs and several clinical and sociodemographic factors. Severity of illness was correlated with total number of CCCs. It should not be surprising that patients with greater severity of illness had documented attention to more CCCs. The APS is higher if the patient has problems with various organ systems, eg, renal insufficiency, hypotension, anemia, or hypoxemia. One would therefore expect it to be more likely that the staff would consider the wisdom of specific therapies that are indicated for each of the respective organ systems that contribute points to the APS. For example, the staff might be more likely to address the CCC of dialysis if the patient were to have renal insufficiency, or to consider the wisdom of continued use of blood products if the patient were anemic.
The variation in attention to CCCs according to diagnosis raises potential concerns with the quality of care rendered to certain groups of patients with DNR orders. Several studies have previously shown that rates of DNR orders vary according to diagnosis.21- 23 A previous study has shown that among patients who have DNR orders, attention to CCCs also varies according to diagnosis.8 Certain diseases, such as malignancies, might be considered terminal, and the clinicians who care for these patients might have learned to anticipate the course of their end-of-life care, at least by the time that DNR orders have been written, and therefore plan accordingly. However, these same care issues also arise for patients with conditions such as Alzheimer disease. In patients with dementia, for instance, respiratory distress may develop owing to pneumonia and intubation might be indicated, or hypotension might develop owing to a urinary tract infection and pressors might be indicated. However, our study shows that the care plans of such patients tend not to specify how clinicians should proceed in such circumstances. A potential remedy for this might lie in the outpatient or nursing home setting, using planning mechanisms such as the Physician Orders for Life-Sustaining Treatment form to anticipate the needs of these patients before hospitalization.24
The fact that attention to CCCs was lower for patients for whom there was no documentation of a consent discussion with the patient or a surrogate deserves comment. Perhaps the failure to document a consent conversation indicates that the patient or surrogate was genuinely not involved and that the health professionals did not wish to address any further limits other than a DNR order. Alternatively, failure to document a consent conversation might indicate that patient or surrogate involvement in decision making was merely perfunctory and staff who treat consent in such a perfunctory manner might be less likely to engage in more expansive planning for the care of the patient. Evidence demonstrates that greater patient involvement in health care decision making is associated with better health care outcomes.25 Similarly, it may be that greater patient or surrogate involvement in end-of-life care decision making is associated with better-quality end-of-life care and that our data reflect this.
We realize that this study has limitations. Because it was conducted at only 2 hospitals and only among medical inpatients, its generalizability may be limited. However, most patients still die in hospitals.26(p39) Our variable of interest is not an outcome and has not been associated with an outcome and thereby formally validated as a process measure of the quality of care.27,28 However, CCCs constitute an explicit, a priori measure of the process of care, and particularly at the end of life, such measures may be important in their own right.26(pp135-139) The CCC method is one of only a few medical chart review measures of the process of care at the end of life that has been validated by staff and patient interviews. The explicit delineation of limits on life-prolonging treatments and comprehensive attention to patient needs included among the CCCs all seem to be important aspects of care, even if they have not been explicitly linked to patient outcomes. For example, the facts that the patient's level of pain has not been addressed and that plans for hospice or limits on care other than cardiopulmonary resuscitation have not been clarified seem to have face validity as deficiencies in quality. Not all CCCs will be of immediate relevance to each patient's care, but in almost all cases, several will be appropriate (eg, dialysis and administration of blood products will be more salient for one; hydration and antibiotic therapy, for another; and ventilation and pain control, for almost all). Because the measure of total CCCs does not capture the many possible patient-specific scenarios, the method trades patient specificity for ease of administration. Because of this, it is more useful in quality improvement than in the evaluation of individual cases. Finally, we recognize that our cross-sectional design limits causal inferences.
This study demonstrates the utility of a medical chart review method for measuring one aspect of the quality of end-of-life care. The fact that documented consent discussions are associated with more explicit and comprehensive care plans makes sense, but warrants further study. The fact that patients with malignancy have the greatest attention to CCCs whereas patients with dementia have far less also warrants attention if we are to improve the quality of care for patients at the end of life. A previous study has demonstrated in randomized controlled trials that house staff education improves attention to CCCs, and the present study suggests that such results are (at least somewhat) sustained. Overall, our results show that there is room for improvement in the care of patients with DNR orders, and suggest that ongoing staff education will be needed.
Correspondence: Daniel P. Sulmasy, OFM, MD, PhD, The John J. Conley Department of Ethics, St Vincent Catholic Medical Centers, St Vincent's Manhattan, 153 W 11th St, New York, NY 10011 (firstname.lastname@example.org).
Accepted for publication October 24, 2003.
This study was supported by the Open Society Institute, New York, NY, through a Project on Death in America Faculty Scholar's Award (Dr Sulmasy) and by a generous grant from the Altman Foundation, New York.
This study was presented in abstract form at the Society of General Internal Medicine Annual National Meeting; May 3, 2002; Atlanta, Ga.
We thank Maike Rahn, MS, Jessica McIlvane, PhD, Vic Tolentino, MPH, JD, Peter Pasley, MD, and Sr Grace Henke, SC, RN, EdD, for their assistance with data collection and management.