Exploring variables related to medical surrogate decision-making accuracy during the COVID-19 pandemic.

OBJECTIVE: Surrogate medical decision making is common in the United States healthcare system. Variables that may influence surrogate decision making have been proposed. Little work has examined relations between these variables and outcomes of surrogate decision making. This study investigated whether surrogates' characteristics, including their knowledge of treatment options, and interpersonal factors predicted surrogates' accuracy and intervention selections. Specifically, predictor variables included: trust in the medical profession, trait-level anxiety, depression, anxiety about COVID-19, religiosity, perceived emotional support, understanding of treatment options, empathy, willingness to accept responsibility for medical decisions, reluctance to burden others, and perceived similarity between oneself and the patient.
METHODS: 154 pairs of patients and their surrogates completed an online survey. Patients indicated preferred treatments in hypothetical decision scenarios. Surrogates indicated the treatment that they thought the patient would prefer.
RESULTS: When taken all together in a predictive model, the variables significantly predicted surrogates' accuracy, F (6) = 3.03, R2 = .12, p = .008. Variables also predicted selection of intensive interventions, F (4) = 5.95, R2 = .14, p = .00. Surrogates reporting greater anxiety about COVID-19 selected more intensive interventions.
CONCLUSIONS: Consistent with prior research, this study found that surrogates' characteristics influence the interventions they choose, with anxiety about COVID-19 having considerable bearing on their chosen interventions. PRACTICE IMPLICATIONS: These findings can inform development of decision-making interventions to improve surrogates' accuracy. Providers may attend to variables highlighted by this study to support surrogates, particularly within the stressful context of COVID-19 and possible future pandemics.

Introduction

Decisions about medical treatment at end-of-life are some of the most challenging that a patient can face. End-of-life care is a timely topic in the United States, with surveys revealing increased concerns for maintaining personal control over the circumstances of one’s death and greater fears of pain and dependency at end-of-life [26]. The ongoing coronavirus (COVID-19) pandemic has further brought critical care services and end-of-life care to the forefront of public consciousness [32]. Given the considerable stress that the pandemic has placed on the United States healthcare system, clinicians are being urged to facilitate patient completion of advance care planning, in order to avoid life-sustaining treatment when it is unwanted and decrease demand on the intensive resources [12]. Advance care planning often involves patients designating a surrogate decision maker who can be consulted to make treatment decisions on their behalf if they are unable to express a choice. This situation occurs among nearly half of hospitalized older adults [46] and close to 50% of all patients considered to be at end-of-life [45].

Surrogates are most often encouraged to use the substituted judgment standard, specifying that they should make decisions in accordance with what the patient would have selected [45]. This approach raises surrogate decision-making accuracy as an issue, regarding whether surrogates adequately predict patient preferences through their decisions. Copious research suggests that surrogates are often inaccurate when predicting patient preferences. In one systematic review, Shalowitz et al. [36] concluded that individuals serving as surrogates predict patient preferences only slightly better than chance, with 68% accuracy on average.

Despite consensus that surrogate decision-making accuracy is a pertinent topic, little research has investigated what variables predict accuracy. While several studies of surrogate decision-making accuracy and theoretical models (e.g., [30], [46]) have proposed factors that may impede surrogate accuracy when identifying patient preferences, no research has yet formally investigated whether these factors predict surrogates’ accuracy.

The current study sought to replicate work by Hare et al. [22] in which they investigated surrogate medical decision-making accuracy using five hypothetical treatment decision vignettes to patient-surrogate pairs. Findings indicated that agreement between patients and their surrogates occurred, on average, 70% of the time. It is important to note that the study by Hare et al. [22] and the current study recruited patient-surrogate pairs from a healthy community sample. Therefore, individuals were assigned to the role of “patient” for the purposes of the study. The term “patients” with which they are labeled is not meant to denote the presence of any current illness or clinical features.

This study also extended prior work by investigating the extent to which specific variables can predict accuracy. Frequently proposed variables were selected for use in this study, as determined by the relative frequency with which they had appeared in empirical and theoretical discussions of surrogate decision-making accuracy to date. Variables featured intrapersonal characteristics of surrogates and interpersonal variables operating between surrogates and other individuals who may be involved in the decision-making process. Specifically, predictor variables included: trust in the medical profession, trait-level anxiety, depression, anxiety about COVID-19, religiosity, perceived emotional support, understanding of treatment options, empathy, willingness to accept responsibility for medical decisions, reluctance to burden others, and perceived similarity between oneself and patient. Since data collection occurred during the COVID-19 pandemic, COVID-19-related anxiety was examined in light of findings that anxiety about COVID-19 is strongly related to broader mental health concerns, such as anxiety and depression [1], and general engagement in health care behaviors [38] . Given the relative dearth of research in this area, the study was broad and exploratory in scope.

Method

Recruitment and participation procedures

All procedures occurred through an online survey hosted through Qualtrics. Recruitment and participation procedures were approved by the West Virginia University Institutional Review Board. Healthy individuals residing in the community responded to the study advertisement and were assigned the role of “patient” for the purposes of this study. Patients with a surrogate, or who could identify an individual most likely to serve in this role, were initially recruited through Amazon Mechanical Turk and a university alumni and faculty listserv. Since issues related to medical treatment decisions and end-of-life are often most salient for middle-aged and older adults, a minimum age criterion of 45 years old was used for patients. Patients were provided with a link to an online survey to share with their surrogate partner for them to complete. Participants indicated their consent to participate by reading a consent form and proceeding to the next page of the survey. Surrogates were required to be over age 18, capable of reading English, and to have access to an email account. Data collection occurred between May 1, 2020 and August 1, 2020. All participants received an Amazon gift card.

First, all participants completed a demographic questionnaire (see Appendix A). Following the demographic questionnaire, participants completed the online survey. The first portion of the survey consisted of hypothetical medical decision vignettes (see Appendix B). Two versions of the survey were created, a surrogate version and a patient version, that differed based on the language used in the decision vignettes. Next, all participants completed measures corresponding to each of the predictor variables (see Appendix C, Appendix D, Appendix E, Appendix F, Appendix G, Appendix H, Appendix I, Appendix J). The order in which participants completed these measures was randomized.

Missing values analyses verified that patterns of missingness were consistent with missingness occurring completely at random or at random. An expectation maximization procedure was used to account for problematic missingness (greater than 5% of items in a measure; [18]).

Participants

One-hundred-and-sixty-one pairs, or 322 total participants, were recruited. All participants identified as residing in the United States. Seven pairs were excluded from subsequent analyses: three pairs due to the surrogate’s omission of a substantial portion of the survey, three due to the patient’s omission of a substantial portion of the survey, and one due to the surrogate responding incorrectly to more than one attention check item. Therefore, a final sample of 154 patient-surrogate pairs remained.

Materials

Criterion variable measure

Choice of end-of-life medical treatment

This study employed the same five medical treatment decision vignettes used by Hare et al. [22], with two additional vignettes pertaining to coronary bypass surgery and dialysis treatment for chronic kidney disease, which are commonly considered in advance directive treatment preference documentation ([52]; see Appendix B). In response to each vignette, respondents selected either to accept (scored as “1”) or decline (scored as “0”) the treatment offered in the vignette. Scores were summed across the vignettes to create a total intervention score, with higher scores indicating a greater tendency to accept presented interventions. To operationalize surrogates’ accuracy, patient-surrogate agreement scores were calculated for each pair. Vignettes on which the pair agreed were scored as 1 and summed. Agreement scores for each pair ranged from 0 to 7, with higher scores denoting greater accuracy by the surrogate. The kappa statistic also was calculated to measure agreement between patient-surrogate pairs. Kappa is a measure of agreement widely used in the surrogate decision-making accuracy literature to date [36], [37], since it provides a measure of agreement that is corrected for chance agreement.

Predictor variable measures

In the regression analyses evaluating the prediction of surrogates’ accuracy and intervention selections, surrogates’ reports on the following variables were used as predictor variables.

Treatment understanding

In relation to the treatment decisions, participants also rated the extent to which they felt they understood the costs and benefits associated with the treatment offered in each vignette. Scores were summed across the seven vignettes to obtain a total “treatment understanding” score. Higher scores reflected greater self-assessed understanding.

Understanding of patient treatment preferences

Surrogates responded to a single item indicating the extent to which they had talked with their partner about end-of-life care preferences, using a 4-point scale.

COVID-19

To assess respondents’ current experience of the pandemic, participants indicated on a 10-point scale, with ratings ranging from “1 = not at all” to “10 = very much so,” their response to the following: “How worried are you about the COVID-19 pandemic?”.

Trust in the medical profession

The extent to which participants trusted the medical profession was assessed by the 11-item scale developed by Hall and colleagues (TIMPS; [21]). The measure demonstrates good psychometric properties, with high internal consistency reliability (α = 0.89; [21]).

Perceived emotional support

The Communication Based Emotion Support Scale (CBESS; Weber et al., [51]) assessed the degree to which participants perceive that they receive emotionally supportive messages from their partner in this study. The scale demonstrate strong psychometric properties, with good internal consistency reliability reported (α = 0.93; Rittenour and Martin, 2008).

Perceived similarity between self and patient

To assess surrogates’ perceived similarity to patients regarding values related to end-of-life treatment, the Quality of Life Values Inventory (QLVI; [8]) was used. As an additional measure of values related to quality of life, the 13-item Life Values measure designed by Schonwetter and colleagues (LVI; 1996) was used. The QLVI demonstrates adequate internal consistency reliability among community-dwelling samples (α = 0.61–0.67; [8], [9]).

To assess perceived similarity, participants rated each of the 18 items from both of these measures using a five-point Likert-type scale ranging from 1 (exactly the opposite of me) to 5 (exactly the same as me) to reflect the extent to which they thought their partner’s values regarding the item were similar to their own. Ratings on the items were summed to produce a total perceived similarity score.

Empathy

Participants’ levels of empathy, or concern for others, was measured by Davis’ [13] Interpersonal Reactivity Index (IRI). Higher scores indicate greater empathy. Satisfactory internal consistency reliability has been demonstrated for this measure (α = 0.71–0.77; [13]).

Willingness to assume full responsibility for medical decisions

The extent to which participants were willing to take full personal responsibility for medical decisions was assessed using the behavioral involvement subscale of the Health Opinion Survey (HOS; [25]). Higher scores indicate favorable attitudes towards self-directed decisions about medical treatment. The behavioral involvement scale demonstrates good internal consistency reliability (KR-20 = 0.74–0.75; Chewning et al., 2012).

Anxiety

The State-Trait Anxiety Scale, Trait Subscale (STAI-T; [41]) assessed trait-level anxiety. The STAI-T measures the respondent’s tendency to experience anxiety and perceive situations as threatening. The measure has high test-retest reliability, ranging from .73–86 [41].

Depression

The Center for Epidemiological Studies Depression Scale-Revised (CESD-R; [16]) assessed depressive symptoms consistent with DSM-V criteria for depression. Higher scores indicate greater depressive symptoms. The measure demonstrates high internal consistency reliability (α = −0.93, [48]).

Reluctance to burden others

This exploratory variable was assessed using the three-item measure developed by Winter and Parks (RBO; [53]). Surrogates responded to these items using the substituted judgment standard and provided the response they thought the patient would choose on a Likert-type scale ranging from 1 (not at all important) to 5 (very important). Ratings were summed to produce a total RBO score. Patients completed this measure as a self-report.

Religiosity

Religiosity was assessed by the Duke University Religion Index (DUREL; [24]). Higher scores indicate greater religiosity. The measure demonstrates adequate internal consistency reliability (α = 0.78–0.91; [24]).

Results

Surrogate characteristics

Surrogates ranged in age from 22 to 88 years old (M = 47.38 years; SD = 9.98) and reported an average of 15 years of education (SD = 4.32). Over three-quarters identified as Christian (57.8%) or Roman Catholic (18.8%). In terms of reported relation to their patient partner, most (83.8%) surrogates were spouses or romantic partners, 7.1% were parents, 3.9% were siblings, 3.9% were close friends, and.6% were adult children or another family member. Surrogates reported relative familiarity with the role of serving as a surrogate, with 77.9% indicating that they had served as a surrogate in a formal capacity. See Table 1 for further demographic information.

Table 1

Surrogate demographic characteristics.

Characteristic	N	%
Sex
Female	116	75.3%
Male	38	24.7%
Race/ethnicity
White/Caucasian (not Hispanic)	131	85.1%
Black/African-American	9	5.8%
Asian-American	9	5.8%
Hispanic	3	1.9%
Native American/Pacific Islander	1	.6%
Mixed	1	.6%
Religion
Agnostic	4	2.6%
Atheist	9	5.8%
Christian	89	57.8%
Jewish	3	1.9%
Muslim	3	1.9%
Other	4	2.6%
Protestant	5	3.2%
Roman Catholic	29	18.8%
Seventh-Day Adventist	2	1.3%
Spiritual but not religious	5	3.2%
Marital Status
Single	9	5.8%
Married/committed relationship	139	90.3%
Divorced	2	1.3%
Separated	1	.6%
Widowed	3	1.9%

Patient characteristics

Patients ranged in age from 45 to 71 years old (M = 51.24, SD = 5.52) and reported 15.5 years of education (SD = 4.66), on average. Similar to the surrogate sample, over three-quarters of patients identified as Christian (58.4%) or Roman Catholic (14%). See Table 2 for further demographic information.

Table 2

Patient demographic characteristics.

Characteristic	N	%
Sex
Female	44	28.8%
Male	109	71.2%
Race/ethnicity
White/Caucasian (not Hispanic)	132	86.3%
Black/African-American	7	4.6%
Asian-American	10	6.5%
Hispanic	2	1.3%
Native American/Pacific Islander	1	.7%
Mixed	1	.7%
Religion
Agnostic	6	3.9%
Atheist	14	9.2%
Christian	89	58.2%
Jewish	1	.7%
Muslim	4	2.6%
Other	2	1.3%
Protestant	9	5.9%
Roman Catholic	22	14.4%
Seventh-Day Adventist	3	2.0%
Spiritual but not religious	3	2.0%
Marital Status
Single	6	3.9%
Married/committed relationship	134	87.6%
Divorced	7	4.6%
Separated	3	2.0%
Widowed	3	2.0%

Surrogates tended to approve administration of treatment, with an average treatment intervention score of 5.63 on the 7-point intervention selection scale. Notably, 42.9% of surrogates indicated that they would provide treatment in all 7 vignettes. For patients, there was also a tendency to approve interventions, with an average score of 5.46%, and 39.2% of patients desiring treatment in all vignettes. Surrogates’ accuracy when predicting treatment selections was worse than that which would be expected by chance, kappa = 0.268, p < .000. Guidelines from Cicchetti and Sparrow [11] classify this level of agreement as “poor.” Of the 154 pairs, 46.1% agreed on all seven decision scenarios; therefore, 53.9% of the sample featured a surrogate inaccurately predicting the patient’s preferences on at least one treatment decision. When predicting preferences, surrogates could err by “overtreating,” or providing interventions that the patient would not have selected, or by “undertreating” and withholding a desired intervention. Of the 80 pairs in which the surrogate inaccurately predicted preferences for at least one scenario, 49 (61.2%) featured the surrogate undertreating, while 31 (38.8%) indicated overtreating.

Bivariate correlation analyses revealed significant relations between accuracy and several of the variables individually, see Table 4. Accuracy did not significantly differ across demographic variables including the surrogate’s gender, marital status, ethnicity, religious preference, or relationship to the patient.

Table 4

Correlations between selected surrogate intrapersonal and interpersonal variables.

	Age	Education	COVID Worry	Similar Values to Patient	Willingness to take Decisional Responsibility	Trust	Anxiety	Empathy	Depression	Religiosity	Reluctance to Burden Others	Emotional Support	Treatment Knowledge
Age	1	.12	-0.10	-0.10	.06	-0.02	-0.20*	.15	-0.10	.02	.07	-0.04	.00
Education	-0.08	1	-0.12	-0.08	-0.03	-0.06	-0.05	-0.16	-0.10	-0.03	.01	-0.08	-0.11
COVID Worry	-0.10	-0.12	1	.02	.05	.16*	.30**	.11	.31**	-0.16*	-0.03	-0.06	-0.15
Similar Values to Patient	-0.10	-0.08	.02	1	-0.25**	.42**	-0.26**	.57**	-0.24**	23**	.64**	.57**	.36**
Willingness to take Decisional Responsibility	.06	-0.03	.05	-0.26**	1	-0.11	.35**	-0.42**	.34**	.07	-14	-0.50**	-0.23**
Trust in Medical Profession	-0.02	-0.06	.16*	.42**	-0.11	1	.04	.24**	.20*	.20*	.36**	.41**	.14
Anxiety	-0.20*	-0.04	.30**	-0.26**	.36**	.04	1	-0.20*	.71**	.01	-0.26**	-0.50**	-0.31**
Empathy	-0.15	-0.16	.11	.57**	-0.42**	.24**	-0.20*	1	-0.14	.13	.33**	.55**	.16*
Depression	-0.10	-0.10	.31**	-0.24**	.34**	.19	.71**	-0.14	1	.10	-0.20*	-0.37**	-0.32**
Religiosity	.02	-0.03	.16*	.23**	.07	.20*	.01	.13	.11	1	.20*	-0.04	.19*
Reluctance to Burden Others	.07	.01	-0.03	.64**	-0.14	.36**	.26**	.33**	-0.20*	.20*	1	.41**	.24**
Emotional Support	-0.04	-0.08	-0.06	.57**	-0.50**	.41**	-0.50**	.55**	-0.37**	-0.04	.41**	1	.40**
Treatment Knowledge	.00	-0.11	-0.15	. 37**	-0.23**	.14	-0.31**	.16*	-0.32**	.19*	.24**	. 40**	1

Note. ** p < .01 (2-tailed).

* p < .05 (2-tailed).

None of the variables appeared as unique predictors in a multiple linear regression analysis evaluating the prediction of surrogates’ accuracy; see Table 3. However, the overall predictive model was significant, F(6) = 3.03, R = .12, p = .008. The following predictor variables were used: surrogates’ ratings of trust, religiosity, empathy, perceived similarity to patients on end-of-life values, perceived degree of patients’ reluctance to burden others, and differences between patients and surrogates on end-of-life values. Collinearity diagnostics, variance inflation factors, and tolerance values were within normal limits. Conditioning plots and Levene’s test supported assumptions of equality of variances and normal distributions ( Table 5).

Table 3

Percent of patient-surrogate pairs who agreed on each decision vignette.

Treatment Vignette	% Agreement
Ventilation	75.6%
CPR	83.6%
Feeding Tube	76.1%
Chemotherapy	81.3%
Amputation	82.8%
Cardiac Bypass	88.8%
Dialysis	85.1%

Table 5

Means and standard deviations for key study variables, as reported by surrogates.

Variable	Range of possible scores	M	SD
Empathy (IRI)	28–140	95.17	11.13
Trust in the medical profession (TIMPS)	11–55	38.78	6.82
Religiosity (DUREL)	6–30	17.29	5.60
Anxiety (STAI-T)	20–80	50.80	9.83
COVID-19-related anxiety	1–10	7.76	1.00
Desire for autonomy on decisions (HOS)	9–81	23.95	5.24
Depression (CESD-R)	20–60	23.76	5.88
Emotional support from partner (CBESS)	13–65	39.00	4.65
Communication with partner about preferences	1–4	3.04	.75
Self-rated physical health	1–5	3.63	.840
Understanding of treatment options	7–35	25.49	5.88
Partner’s reluctance to burden others (RBO)	5–15	11.34	2.40
Values similarity to partner (QLVI/LVI)	18–90	70.02	11.13
Treatment intervention score	0 – 7	5.63	1.40

The variables significantly predicted treatment selection, or surrogates’ tendencies to approve treatment in the decision vignettes; F (4) = 5.95, R 2 = .14, p = .00. Of the predictor variables, surrogates’ reports of COVID-19-related anxiety (B = 0.2.19, p < .005) appeared as a unique predictor.

Discussion and conclusion

Discussion

In concordance with Hare et al. [22], and evidence from the surrogate decision-making accuracy literature [40], surrogates predicted patients’ preferences with poor accuracy. This finding adds to the body of evidence suggesting that individuals who make decisions for patients in end-of-life situations may not be trusted to consistently select the patient’s desired treatment (Torke, 2008). The level of agreement found in this study is notably lower than that reported in a systematic review of the literature by Shalowitz et al. [36], [37]; 68% agreement, on average), yet it is consistent with levels reported by more recent investigations of surrogate decision-making accuracy (e.g., [5]).

Among pairs in which surrogates inaccurately predicted preferences on one or more decisions, there was a tendency for surrogates to undertreat and fail to provide treatment that the patient would have desired. The current findings are consistent with literature indicating that surrogates are only moderately successful when gauging levels of pain experienced by the patients they serve [14] and that surrogates tend to underestimate patients’ perceptions of an acceptable quality of life [5].

There are undoubtedly mediating and moderating factors that may influence surrogates’ propensities to undertreat. The “disability paradox” [47], speaks to the mechanism by which undertreatment errors may occur, since individuals may fail to appreciate the ease with which others can adapt to poor health conditions and situations of impairment. Certainly, the issue of undertreatment raises ethical issues that have prompted medical ethicists (e.g., [20]) to advocate for revising standards of surrogate decision making beyond the substituted judgment standard, perhaps more strictly enforcing a standard of “best interest” standard that restricts surrogates from violating explicit treatment norms of the medical profession.

The discrepancy in treatment decisions between patients and surrogates is particularly alarming given that surrogates reported that they felt they understood the costs and benefits related to the treatments relatively well, and that they had discussed end-of-life preferences with patients to some extent. Prior research documents a propensity for surrogates to feel more equipped to make decisions when they have discussed patients’ preferences in advance (e.g., [3]).

However, the evidence regarding whether discussion-based interventions targeting end-of-life preferences improves surrogates’ accuracy is mixed [29]. The current study underscores how surrogates’ confidence in the value of prior discussions about end-of-life care is not strongly tied to their accuracy. Surrogates may misinterpret higher-frequency discussions as being of higher-quality, producing an inflated assessment of one’s knowledge of treatment options and patient preferences. Values related to end-of-life decisions are complex, and surrogates may unintentionally adopt a myopic view of treatment decisions [35]. For instance, surrogates may fail to appreciate the number and variety of values implicated in a decision and focus too narrowly on a single identified value, such as the desire to sustain life at any cost [44].

Several interesting relations emerged to enhance understandings of potential mechanisms underpinning surrogates’ accuracy. First, the relation between accuracy and empathy is consistent with understandings of empathy as a construct that promotes perspective-taking [28]. Moreover, the observed role of empathy may have been particularly pronounced given the age of the recruited sample (M age = 47 years for patients, 51 years for surrogates), as empathy tends to peak in middle-adulthood (O’Brien et al., [31]).

Accuracy was also positively correlated with religiosity, such that surrogates with greater self-reported religiosity were more accurate. This is the first study to specifically examine religiosity relative to surrogate decision-making accuracy. An explanation for this relation could lie in previously demonstrated positive associations between religiosity and personality variables not assessed by this study, such as conscientiousness [23]. Surrogates who report greater religiosity may be those who are more conscientious, and who therefore completed the decision activity with greater care and attention to patient preferences.

Surrogates who were more accurate predictors of patient preferences also tended to be those who reported greater trust in the medical profession. Researchers have proposed that low trust in the medical profession may hinder surrogates’ decision making (e.g., [46]). As the first investigation to specifically examine trust in the medical profession as related to surrogate decision-making accuracy, this study provides preliminary support for these conjectures.

Perceiving the patients’ values as more similar to one’s own (as measured by ratings on the Quality of Life and Life Values Index) was also associated with accuracy. This is consistent with prior research evaluating surrogates’ execution of the substituted judgment standard and their reflections of the process. Surrogates often state that they rely on known mutual interests between themselves and the patient to guide their decisions (Vig et al., [49]). However, researchers caution that high perceived similarity can undermine decision making if surrogates mistakenly interpret their own values as identical to those of the patient [17].

The selected variables significantly predicted surrogates’ treatment selections. A unique predictor variable was COVID-19-related anxiety, such that surrogates with greater anxiety about COVID-19 selected more intensive treatments for patients. Although current understandings of COVID-19-related anxiety are nascent, this specific type of anxiety demonstrates positive relations with anxiety and depressive disorder symptoms [1], which was also evident in this study. Moreover, anxiety about COVID-19 is associated with feelings of hopelessness and the use of maladaptive coping mechanisms (e.g., drugs and alcohol; [27]). Importantly, these negative emotions and coping behaviors pertain to oneself and one’s current situation, not involving another person. In this study, surrogates were acting on behalf of the patient, a person to whom they were presumably close. Evidence from the self-other decision-making literature reflects that individuals often choose different decisions for another person than they would for themselves; for instance, they are more likely to make a “safer” choice than they would for themselves in situations where risk aversion is valued [43]. Therefore, surrogates may have chosen an intensive treatment intervention to “save” the patient in these vignettes, despite feeling pessimistic about their personal circumstances in the distressing context of a global pandemic.

Participants’ approaches to the decision vignette regarding ventilation in particular must be considered in the unique context of COVID-19, a disease with symptoms primarily affecting the respiratory system, that can produce severe complications such as pneumonia that require the use of mechanical ventilation [15]. During the pandemic, estimates of the number of needed ventilators in the United States have ranged from several hundred thousand to over one million [19]. The country’s insufficient ability to meet this demand was well-broadcast and concerns for the adequate provision of critical care supplies have been raised [33]. Heightened public attention towards mechanical ventilation may have prompted participants to reexamine their views towards this treatment option in particular, such that patients’ personal desires to receive ventilation may have shifted, unbeknownst to their surrogates. Therefore, surrogates may have been less accurate on this decision because they chose decisions based on understandings of what patients may have selected before the pandemic.

As with any study, limitations exist. First, hypothetical scenarios can only model a real-world decision-making scenario, and these findings are therefore limited with respect to generalizability. Surrogates’ accuracy may be lower when working with a hypothetical situation that does not pertain to the patient’s actual condition. With regards to the hypothetical situations used in the current survey, the question concerning treatment knowledge and understanding may have been confusing to participants. Phrasing regarding “costs and benefits” of treatment options may have led participants to solely consider financial costs of treatments, rather than more comprehensive risks and burdens. However, logistical constraints of gathering data at the exact moment at which surrogate decision making takes place, particularly during a pandemic limiting access to healthcare facilities, impeded this possibility.

Regarding statistical limitations, the reported regression coefficient values for the multiple linear regression models were relatively modest in magnitude; therefore, important predictor variables may have been missed. The surrogate decision-making process is complex and involves the interplay between multiple variables. Future research will be important to identify other related variables in an effort to better account for variance in surrogates’ accuracy when predicting patient treatment preferences. The kappa statistic also has limitations with regards to its interpretation as a measure of chance-corrected agreement (e.g., the “kappa paradox,” [10]). However, this statistic was reported in order to maintain consistency with the surrogate decision-making accuracy literature to date (e.g., Shalowitz, 2006; [6], [30], [40]) and enable comparisons with prior studies.

With respect to demographic characteristics the recruited sample underrepresents the diverse population of patients, and their associated surrogates, who are encountered in United States intensive care units where end-of-life surrogate decision-making occurs [2]. Patient-surrogate pairs of minority backgrounds are important to prioritize in future work, as medical decision-making tendencies can vary considerably across groups (e.g., [50]). Moreover, the recruited sample was not an ideal representation of real-world surrogate decision makers, as surrogates tend to be older (e.g., mean age of 59.6 years reported by Snyder et al. [42]) and are most often adult children of the patient [42]. The current surrogate sample was younger than average and consisted mostly of spouses; therefore, they may have a different understanding of patients’ wishes and responded to scenarios differently than surrogates encountered in real medical decision-making situations. Additionally, the individuals assigned the role of “patient” for the purposes of this study were healthy, community-dwelling adults. Patients for whom surrogates most often make decisions are those with chronic conditions, severe illness, or frailty (Vig et al., [49]). Therefore, “patients” in the current study were not wholly representative of those involved in real-world surrogate medical decision-making situations.

Work conducted during the COVID-19 pandemic may prioritize the psychometrically sound assessment of COVID-19-related anxiety, a variable with demonstrated importance in this study. At the time of study design, no validated assessment measures of this construct existed, and a single-item measure was used. Since completion of data collection, researchers have begun to design and validate assessment measures to more comprehensively capture COVID-19-related anxiety (e.g., Coronavirus Anxiety Screener; [27]). Ultimately, more research is needed regarding the relation between perceptions of COVID-19 and treatment preferences, specifically with regard to self-other decision-making discrepancies.

Conclusion

This study examined decision making between patient-surrogate pairs on hypothetical treatment decision vignettes. In addition to examining the relation between accuracy and selected variables, the study evaluated the extent to which surrogates accurately predicted patients’ preferences. Findings further amplify concerns for surrogates’ accuracy, and their ability to execute the substituted judgment standard, by revealing inconsistencies in patient-surrogate agreement when selecting treatment interventions. Certain interpersonal and intrapersonal variables, as reported by surrogates, predicted accuracy across all decision vignettes. These variables include surrogates’ religiosity, trust in the medical profession, empathy, perceived similarity to patients on values related to end-of-life, and perceptions of patients as being more reluctant to burden others. Surrogates’ interpersonal and intrapersonal characteristics also influence the interventions they select for patients, with anxiety about COVID-19 having considerable bearing on surrogates’ chosen interventions.

Practice implications

The selected variables in this study significantly predicted surrogates’ intervention selections on behalf of patients. Of the predictor variables, surrogates with greater anxiety about COVID selected more intensive treatments for patients. Findings from this broad, exploratory study can be used to further the development of targeted decision-making interventions to improve surrogates’ accuracy. Providers may attend to these variables when working with surrogates in their decision-making role. For instance, during the current pandemic, providers may specifically attend to facilitating surrogates’ trust in the medical profession and ability to cope with anxiety surrounding COVID-19. These findings speak to the importance of considering decision-specific support for surrogates when making decisions on behalf of patients. For instance, ample patient decision guides and tools exist to aid with medical decisions, such as those relating to mechanical ventilation or dialysis [7], and could be similarly adapted for use with surrogates.

CRediT authorship contribution statement

Spalding Rachael: Conceptualization, Methodology, Data Analysis, Writing. Edelstein Barry: Supervision, Securing funding, Reviewing, Editing.

Declaration of Competing Interest

I have no conflicts of interest to disclose.