Introduction

Critical care medicine has developed as a specialty designed to save immediately threatened identifiable lives. The prevailing ethos, therefore, has been akin to what Jonsen called the “rule of rescue” [1], or “the powerful human proclivity to rescue a single identified endangered life, regardless of cost, at the expense of any nameless faces who will therefore be denied health care” [2]. Increasingly, however, critical care clinicians are pressured to contain costs and manage scarce resources in light of rising demands for and costs of care they provide [3–5]. Thus, physicians are being asked to consider limiting services for their patients to benefit society more broadly [4].

Although it is clear that critical care clinicians must routinely ration intensive care unit (ICU) resources [6–12], few studies have examined how clinicians balance their duties to try to save identifiable lives with their additional responsibilities to promote, or at least not contravene, societal interests [13]. Surveys of intensivists [14, 15] and other physicians [16, 17] have evaluated clinicians’ perceptions of what resources they limit and how commonly they do so. However, such studies are limited by the difficulty of understanding the questions being asked, and by the social desirability of providing certain answers [18].

An alternative approach to exploring how clinicians balance conflicting duties between the rule of rescue and societal interests is to explore the decision processes of clinicians facing a choice that vividly pits these duties against each other. Thus, to advance understanding of the tension between serving an identifiable patient or anonymous patients more broadly, we presented ICU clinicians with vignettes that forced them to reveal [19], rather than merely state, how they would weigh these conflicting mandates. Because we were specifically interested in how clinicians balance such tensions when purely utilitarian arguments unambiguously support prioritizing society’s interests, we used a scenario in which one available ICU bed could be allocated to a severely ill patient with a small chance for meaningful recovery, or to a patient nearing death or having recently died who could donate his organs post-mortem to benefit others. Although perhaps uncommon, the Society of Critical Care Medicine Ethics Committee’s consensus statement reflects the plausibility of this scenario, noting that although ICU beds generally should be reserved for patients with reasonable prospects of substantial recovery, “an exception may be [made for] patients being supported for organ donation” [12].

Methods and materials

The University of Pennsylvania Institutional Review Board approved this study.

Questionnaire

We developed a questionnaire containing scenarios that forced clinicians to prioritize either (1) an identifiable cancer patient for whom ICU admission was unlikely to be beneficial, or (2) a deceased or unequivocally dying patient for whom ICU admission would serve to benefit multiple anonymous organ recipients. After pilot testing among local ICU physicians and nurses, the final questionnaire presented all clinicians with two scenarios that differed only in the ages (50 or 62) of the two patients described. These were chosen to provide a plausible basis for varying one of our independent variables of interest—the number of life-years to be gained by patients awaiting organ transplantation (30 or 5, respectively) if the bed was allocated to the potential organ donor. To mitigate the potential for ordering effects, we used random number generation to determine the order in which scenarios were presented to each clinician.

To explore the extent to which patients’ vital statuses influenced clinicians’ prioritization of potential organ donors [20], we randomly assigned each clinician to receive scenarios in which the potential donor had been declared dead on the basis of neurological criteria (donor after neurological determination of death, or DNDD) or in which the potential donor was still alive but had previously stated wishes to have life-sustaining therapy withdrawn, and to become a donor after circulatory determination of death (DCDD). Thus, each clinician received one set of two scenarios out of four possible pairs that differed only in the social benefit that would accrue from prioritizing the potential organ donor (30 or 5 extra years); and in the type of donor (DCDD vs. DNDD, which was held constant in each clinician’s two scenarios (Fig. 1).

Fig. 1
figure 1

Scenarios

Full size image

The questionnaire also assessed clinicians’ perceptions of organ donation and conflicts of interest in end of life care; their perceptions of Organ Procurement Organizations (OPOs); and demographic characteristics, for a total of 11 questions.

Survey administration

We administered the questionnaire to 2,206 academic ICU physicians with presumed valid email addresses in a national database of academic intensivists [21], and 988 ICU nurses randomly selected from the American Association of Critical Care Nurses mailing list with valid postal addresses. The physician survey was administered via internet and the nurse survey was delivered by post; they were otherwise identical. We provided financial incentives to improve response rates.

For the physician survey, we sent an email to potential respondents, providing an embedded link to a Web-based questionnaire. Up to two reminder emails were sent at 5-day intervals to non-responders.

For the nurse survey, we sent a copy of the questionnaire by post, and a pre-stamped return envelope. A follow-up mailing was sent 3 weeks later to non-responders. In both surveys, responses received within 2 weeks of the final mailing were included for analysis.

Analyses

We calculated the response rate as the number of respondents completing at least 8 of the 11 items, divided by the sum of known eligible respondents plus respondents of unknown eligibility [22].

We used multivariable logistic regression to identify characteristics of clinicians and elements of the scenarios that were independently associated with choices to prioritize management of the cancer patient rather than the potential organ donor (“rule of rescue decisions”). The scenario elements of hypothesized importance (life-years gained by organ donor management and whether the donor was a DNDD or DCDD) were forced into the model. Clinician characteristics and the interaction term between clinician type and life-years gained were included as covariates if they were related to the outcome at p < 0.20 in a univariable model, or if, when included in the multivariable model, they modified the coefficient for the life-years gained by organ donor management by at least 15%. All models were clustered on respondent and used robust variance estimates to account for the non-independence of the two responses per clinician.

We used Stata 10.1 software (Stata Corp., College Station, TX) to construct all models.

Statistical power

The study was designed to detect at least a 10% absolute difference in the proportions of clinicians making rule of rescue decisions when doing so would require foregoing 30 versus 5 extra life-years for patients awaiting transplantation. Conservatively assuming that 50% of clinicians would make such choices in the 5 life-year condition, and using a two-sided α = 0.05, we estimated that 816 completed responses were needed to provide at least 80% power to detect this difference.

Qualitative data analysis

To better understand clinicians’ reasons for making rule of rescue decisions, we asked clinicians to describe their rationale via open-ended responses. One investigator (RK) reviewed clinicians’ responses line by line, and used thematic data analysis and constant comparison techniques to code and sort data into discrete themes that emerged from these responses [23].

Results

Among 2,206 physicians and 988 nurses asked to participate, 684 physicians and 438 nurses consented and completed at least 8 of the 11 items (response rates = 31.0 and 44.3%, respectively). As we report elsewhere [24], there was no evidence of non-response bias. This conclusion stems from our observation that no differences were found in our outcome variables among groups of clinicians displaying different response rates as a result of being randomly assigned to receive different response incentives [24].

The demographic and clinical characteristics of the 1,122 total respondents are shown in Table 1.

Table 1 Characteristics of respondents
Full size table

Proportions of clinicians making rule of rescue decisions

Physicians were more likely than nurses to adhere to the rule of rescue (45.9 vs. 32.6%, difference = 13.2%; 95% CI 9.1–17.3%) (Fig. 2). Clinicians’ decisions were partially influenced by the life-years to be gained through organ donor management, with clinicians being less likely to make rule of rescue decisions in the scenarios in which organ donation would yield 30, as compared with 5, additional life-years for transplant recipients (43.1% versus 38.5%, difference 4.6%; 95% CI 0.5–8.6%). In these analyses of clinicians’ responses across both scenarios, 7.7% (95% CI 6.2–9.1%) of physicians and 3.8% (95% CI 2.5–5.0%) of nurses prioritized the cancer patient when 5 extra life-years could be gained through organ donor management and prioritized the potential organ donor when 30 extra life-years could be gained (p = 0.008 for the difference in proportions). By contrast, in among-subjects analyses restricted to the scenarios clinicians were randomly assigned to receive first, this utilitarian benefit did not influence the probability that physicians (46.5 vs. 43.6%, p = 0.44) or nurses (30.8 vs. 32.7%, p = 0.66) would make rule of rescue decisions.

Fig. 2
figure 2

Proportions of clinicians choosing to allocate the last ICU bed to the identifiable patient with a grave prognosis. The “life-years lost to society” represents what was presented to clinicians regarding the social consequences of choosing not to allocate the last ICU bed to manage a potential organ donor. Error bars represent the 95% CIs around the unadjusted proportions, MD doctor of medicine, RN registered nurse

Full size image

Factors associated with rule of rescue decisions

In univariable analyses, physicians, men, and clinicians who believed that OPO staff did not respect patients’ interests were each significantly more likely to make rule of rescue decisions (p < 0.001 for each) (Table 2). To investigate the possibility that results were influenced by particular concerns with organ donor management, rather than reflective of the underlying tension between attempting to rescue a cancer patient versus a potential organ donor, we restricted analyses to the 551 physicians (80.6%) and 396 nurses (90.4%) who felt that OPOs respected patients’ wishes. In these restricted analyses, 40% (95% CI = 36–44%) of physicians and 29% (95% CI = 25–34%) of nurses still made rule of rescue decisions.

Table 2 Factors associated with participants’ willingness to allocate the last ICU bed to an identifiable patient
Full size table

In the multivariable model (Table 2), being a physician (OR = 1.47; 95% CI = 1.04–2.09), believing that OPO staff do not respect patients’ interests (OR = 1.84; 95% CI = 1.31–2.58), and the life-years to be gained through organ donation (OR = 0.89; 95% CI = 0.80–0.98), retained significant associations with rule of rescue decisions. After adjustment for men’s increased probabilities of being physicians, a trend remained towards men being more likely to make rule of rescue decisions (OR = 1.39; 95% CI = 1.00–1.94).

Influence of donor status

Clinicians’ decisions to allocate a scarce bed to a patient unlikely to benefit from it were not influenced by whether the potential organ donor was already dead (DNDD) or dying (DCDD) (OR = 1.11; 95% CI = 0.88–1.40). When managing the organ donor would yield 5 life-years gained for those awaiting transplantation, 46% (95% CI = 41–51%) of physicians and 35% (95% CI = 29–41%) of nurses made rule of rescue decisions when the donor was a potential DNDD, and 51% (95% CI = 46–57%) of physicians and 32% (95% CI = 26–39%) of nurses made rule of rescue decisions when the donor was a potential DCDD. Similar proportions were observed when managing the organ donor would yield 30 extra life-years for those awaiting transplantation.

Reasons for prioritizing identifiable patients

Table 3 describes themes that emerged from clinicians’ open-ended responses regarding rule of rescue decisions. The frequencies of these themes reflect proportions of the 353 physicians and 149 nurses who prioritized the cancer patient under at least one of the two scenarios. The majority of these clinicians (68% overall) explained their choice by reflecting on their professional duties of fidelity to identifiable living patients. The other explanations were cited substantially less commonly (Table 3). Clinicians who prioritized the organ donor in both scenarios typically based such judgments on the expected utilitarian benefit and/or perceptions regarding the futility of ongoing critical care for the cancer patient.

Table 3 Clinicians’ stated reasons for allocating the last ICU bed to an identifiable patient
Full size table

Discussion

This study suggests that many ICU clinicians feel compelled to provide salvage critical care for identifiable patients with grave prognoses even when doing so measurably contravenes society’s interests. Adherence to this rule of rescue accords with traditional norms of prioritizing one’s own patients, but contrasts with recent conceptualizations of medical professionals’ social responsibilities [13, 25].

These findings suggest that many ICU clinicians are not prepared to abrogate their perceived duties to patients in front of them. We might expect greater reluctance to abandon duties to identifiable patients when clinicians must allocate beds already occupied by their own patients in efforts to make room for others. That clinicians are unwilling to abrogate the rule of rescue to promote sizeable increases in the life spans of patients awaiting transplantation suggests potential barriers to implementing ICU triage protocols based on potential benefit rather than pre-established relationships with patients, or to proposals to reduce or eliminate care processes for cost containment.

We found that clinicians were not strictly utilitarian in their judgments. In among-subjects analyses of clinicians’ initial instincts, clinicians’ judgments were not influenced by whether donor management would yield 5 or 30 additional life-years to patients awaiting transplantation. By contrast, in analyses of how clinicians responded across the two scenarios, we found that focusing clinicians’ attention on this utilitarian contrast swayed minorities of decisions. Thus, using two sets of results that complement each other, we conclude either that ICU clinicians’ judgments are decidedly non-utilitarian, or that clinicians failed to attend to the magnitude of the benefit until they were directly alerted to it by our manipulating the value.

Our study incorporated several design elements that fostered a productive exploration of clinicians’ tensions in prioritizing patients. First, unlike the aforementioned problems with studying rationing due to fiscal scarcity, our scenario required clinicians to consider how best to allocate truly scarce professional resources [26] and ICU beds [7, 10]. Because ICU clinicians make such allocation decisions daily [8], they may more readily recognize them, and thus report their behaviors more accurately. Second, albeit uncommon, our organ donor scenario was specifically chosen to vividly contrast the small benefits for clinicians’ identifiable patients and potentially substantive public health benefits. Third, because ICU clinicians are unlikely to meet organ recipients, this scenario forced clinicians to consider the interests of their own identifiable patients versus truly anonymous beneficiaries of their decisions, making a more stark exploration of the tensions clinicians experience than would occur if clinicians were asked to trade off benefits among several of their own patients. Fourth, by focusing on two patients in the ER competing for a single ICU bed, our study prevented clinicians from relying on simple allocation rules such as first-come, first-served, and instead forced them to consider the underlying tension between the good of a single identifiable patient or multiple anonymous ones [27].

This study also has limitations. First, we obtained only moderate response rates. However, non-response bias does not appear to have affected this study [24]. Second, some clinicians’ decisions to manage potential organ donors may have been partially influenced by the desire to promote these donors’ interests in bequeathing gifts post-mortem. However, if such motives were present, a purer scenario that avoided all contamination of social benefits with individual benefits might have found even greater reluctance to abrogate perceived duties to individual patients.

Third, although the use of thought experiments to understand clinical decisions between individuals and groups has a robust history [27], it is possible that some clinicians did not identify the scenario as a moral contrast between prioritizing the one or the many, but rather as a logistical matter of which patients are optimally managed in various places in a hospital. Indeed, the qualitative portion of our study revealed that 161 of 502 clinicians (32.1%) who made rule of rescue decisions did so in part because they believed the organ donor could be suitably managed elsewhere. We cannot determine how many of these clinicians created this explanation to justify uncomfortable decisions, but even after excluding these respondents, 31.9% of the remaining clinicians made rule of rescue decisions.

Fourth, although the use of vignettes has been shown to produce valid reflections of real-world decision-making in other settings [19], it remains possible that clinicians’ responses to the hypothetical scenarios used in this study may not capture the decisions they would make if actually faced with these situations. Additionally, these specific scenarios may make the results less generalizable even within the same study population. The purpose of the study was not to quantify the proportions of clinicians making rule of rescue decisions, but rather to demonstrate the tension clinicians experience between their roles of dual advocacy. If clinicians are not willing to abrogate their responsibilities to identifiable patients in favor of clear societal benefits, they may be less willing to forgo rule of rescue decisions when faced with actual patients and less clear social benefit.

Fifth, studying predominantly professional society members’ choices in an organ donor scenario could reduce the generalizability of the results. Because such societies educate members about and promote the societal benefits of organ donor management, our study may in fact underestimate the proportions of all ICU clinicians who would make rule of rescue decisions.

Finally, our specification of life-years gained by prioritizing organ donor management represents utilitarian benefit that would not be available when clinicians make actual allocation decisions. We made this choice to decrease the possibility that clinicians’ varying knowledge of organ donor management would influence the choices we were interested in, and to enable specific study of the extent to which pure utilitarian considerations enter into these choices. The observation that only 14% of clinicians who made rule of rescue decisions (Table 3) did so in part due to uncertainty regarding the prognostic estimates we provided suggests that this strategy was largely successful.

Conclusions

In summary, these results expand on prior studies of ICU rationing decisions [16–18, 28] by utilizing an experimental design to enable clinicians to reveal, rather than merely state, their allocation preferences [20], and by eliciting open-ended responses to identify why clinicians would prioritize individual patients. Taken together, these mixed qualitative and quantitative results provide empirical verification of predictions that the rule of rescue represents a substantial and persistent barrier to the efficient allocation of scarce resources and may impede efforts to improve patient triage or constrain costs [29, 30]. Future research—perhaps using ethnographic or epidemiologic methods to characterize real-world allocation decisions—is needed to identify the actual frequencies with which clinicians prioritize individual patients when doing so carries real social costs, and to further explore factors related to the observed variability in allocation patterns among clinicians.