Patient control preferences for medical decision making before and after evaluation for left ventricular assist device

Abstract

Understanding patients' preferred roles in medical decision making (i.e., passive, collaborative, active) is important to personalized care and patient engagement. Patient control preferences have been described for many treatment decisions, but their stability over time has not been characterized, particularly for major medical events with long-term implications. We prospectively surveyed 233 patients at the initiation of evaluation for a left ventricular assist device, and 1 and 6 months later, including collection of the Control Preferences Scale. Collaborative and active preferences were most common initially, followed by a shift towards more active. Approximately half of patients reported a different control preference in follow up. Patients with higher income and education levels were more likely to prefer an active role. These findings suggest that most patients want to be engaged in shared decision making, but to what degree is varied, can change over time, and is influenced by social determinants of health.

1. Introduction

A patient's preference regarding their involvement in clinical decision making can be broadly categorized as passive, collaborative, or active [1]. Multiple studies demonstrate that patients who are actively involved in making medical decisions achieve enhanced clinical outcomes and are more satisfied compared to patients who are less active [2]. It is important for clinicians to understand the patient's control preferences to conduct optimal shared decision making. Data on control preferences in patients faced with a life-threatening illness remain limited, and how preferences evolve over time is relatively unknown. Patients faced with decisions regarding left ventricular assist devices (LVAD) provide a unique population to study control preferences in a high-acuity, high- risk decision-making context. While we know using decision aids can improve shared decision making [3], understanding the patient's level of involvement in the LVAD decision had not been examined. Thus, we aimed to understand patient's control preferences during LVAD decision making and changes over time.

2. Methods

The DECIDE-LVAD Trial is a stepped-wedge, cluster randomized trial aimed at evaluating the effectiveness of a shared decision support tool for patients being considered for LVAD at 6 diverse centers across the United States. Patients were enrolled at the time of LVAD evaluation initiation and completed a baseline survey within 3 days of enrollment (and prior to treatment decision). Follow-up surveys were collected at 1 month and 6 months post-enrollment; medical record data were collected at baseline, 1 month, and 6 months. Full methods published previously [3]. As part of the patient surveys, the Control Preferences Scale (CPS), a validated measure of patients' desired level of involvement in decision making, was collected. The CPS is a single-item 5-point scale to indicate a patient's self-reported preferred role in making decisions about their health care, ranging from passive (“I prefer to leave all decisions regarding treatment to my doctor”) to active (“I prefer to make the final selection about which treatment I will receive”), representing 5 levels of involvement in medical decision making [4]. Demographics including patient age, gender, race, relationship status, education level, and total household income were also collected.

For the purposes of modeling and interpretability, the 5-level control preference responses were grouped into three categories of active, collaborative and passive roles for descriptive analysis, and further collapsed into binary outcome categories of active/collaborative and passive for modeling. We modeled the associations between demographic factors and control preferences at baseline. A separate (univariable) logistic regression model (of the probability that preferred role = ‘Passive’) was fit to estimate unadjusted odds ratios (OR) for each demographic characteristic, with corrections for clustering using a random intercept for site and robust standard errors adjusted for small sample size according to the method of Morel, Bokossa, and Neerchal [5]. For education, in order to obtain valid estimates for the subgroup category with only two participants having a college degree or higher and preferring a passive control role, the model was specified as an exact logistic regression adjusted by a fixed instead of random effect of site to correct for clustering. Estimates of this fixed effects model are therefore specific to the 6 clinical sites included in the sample, rather than generalizable as in the random intercept models.

Patients missing key variables of interest were excluded from the primary analysis. The model for income excludes 20 participants with missing income data (8.6 % of the total sample; 18 in the ‘Active/Collaborative role’ group and 2 in the ‘Passive role’ group). A sensitivity analysis was also performed treating ‘missing’ as a separate category in the model for income.

Analyses were performed using SAS version 9.4, and the Sankey diagram was produced in R version 4.1.1 using the ‘plotly’ package [8].

3. Results

Of 248 participants enrolled in the main study, the current sample included 233 who completed the baseline CPS. By 6 months, 32 % (n = 75) of those 233 patients were missing CPS data, most commonly due to patient death (n = 33, 44 %). Other reasons for missing data included study withdrawal (n = 9, 12 %), survey noncompletion (n = 17, 22.7 %), or loss to follow-up (n = 16, 21.3 %).

Baseline patient control preference for their role in LVAD decision making spanned the available options in a bell-shaped distribution with skew towards an active role; patients most commonly desiring an active-collaborative approach (Table 1). Over time, roughly half of the patients changed their stated preference from one category (active, collaborative, passive) to another (Fig. 1). Descriptive analysis showed a small trend over time towards more active roles (Table 1).

Table 1.

Control Preferences Scale item response (preferred role) by visit, among patients with non- missing item response at baseline and each visit (1- and 6-month time points).

Preferred Role	Item Response, n (%)	Baseline (N = 233)	Month 1 (N = 176)	Month 6 (N = 158)
Active	I prefer to make the final selection about which treatment I will receive	21 (9.0 %)	29 (16.5 %)	17 (10.8 %)
	I prefer to make the final selection of my treatment after seriously considering my doctor's opinion	92 (39.5 %)	77 (43.8 %)	74 (46.8 %)
Collaborative	I prefer that my doctor and I share responsibility for deciding which treatment is best for me	85 (36.5 %)	52 (29.5 %)	54 (34.2 %)
Passive	I prefer that my doctor makes the final decision about which treatment will be used, but seriously considers my opinion	20 (8.6 %)	12 (6.8 %)	9 (5.7 %)
	I prefer to leave all decisions regarding treatment to my doctor	15 (6.4 %)	6 (3.4 %)	4 (2.5 %)

Fig. 1.

Sankey diagram depicting active, collaborative, and passive roles that patients elected to take at baseline and at time points of 1 and 6 months.

BL = baseline, M1 = month 1, M6 = month 6; green = active role, purple = collaborative role, orange = passive role, light gray = missing data, dark gray = patient death. Numbers in parentheses represent the total count of participants in the given response category at the given time point. For participants with missing responses at month 1, flows were drawn directly from their baseline to month 6 categories. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

At baseline, higher total household income and higher attained level of education were associated with lower odds of passive role preference. Age, gender, race/ethnicity, relationship status, and LVAD treatment were not statistically significantly related to preferred control roles (Table 2).

Table 2.

Demographic characteristics overall and by preferred control role (active/collaborative vs. passive) at baseline.

Characteristic, n (%)	Total (N = 233)	Preferred role		Logistic regressiona
		Active/Collaborative (N = 198)	Passive (N = 35)	Odds Ratio (95 % CI)	P value
Age quartile (years)					0.3427
28 to 55	54 (23.2 %)	44 (22.2 %)	10 (28.6 %)	0.93 (0.28, 3.04)	0.9037
56 to 65	57 (24.5 %)	49 (24.7 %)	8 (22.9 %)	0.67 (0.21, 2.13)	0.4931
66 to 70	66 (28.3 %)	60 (30.3 %)	6 (17.1 %)	0.41 (0.13, 1.31)	0.1324
>70 (ref.)	56 (24.0 %)	45 (22.7 %)	11 (31.4 %)	reference
Gender					0.5986
Female	35 (15.0 %)	28 (14.1 %)	7 (20.0 %)	1.52 (0.32, 7.23)	0.5986
Male (ref.)	198 (85.0 %)	170 (85.9 %)	28 (80.0 %)	reference
Race/Ethnicity					0.9105
White non-Hispanic (ref.)	186 (80.9 %)	158 (81.0 %)	28 (80.0 %)	reference
Other	44 (19.1 %)	37 (19.0 %)	7 (20.0 %)	1.07 (0.34, 3.36)	0.9105
Relationship status					0.8247
Married/partnered (ref.)	174 (75.0 %)	147 (74.6 %)	27 (77.1 %)	reference
Divorced/separated/widowed/single	58 (25.0 %)	50 (25.4 %)	8 (22.9 %)	0.87 (0.26, 2.97)	0.8247
Education levelb					0.0073
High school grad/GED or less (ref.)	87 (37.5 %)	71 (36.0 %)	16 (45.7 %)	reference
Some college	83 (35.8 %)	66 (33.5 %)	17 (48.6 %)	1.12 (0.49, 2.58)	0.9196
College graduate or more	62 (26.7 %)	60 (30.5 %)	2 (5.7 %)	0.14 (0.02, 0.65)	0.0059
Total household incomec					0.0464
≤$60,000 per year (ref.)	143 (67.1 %)	116 (64.4 %)	27 (81.8 %)	reference
>$60,000	70 (32.9 %)	64 (35.6 %)	6 (18.2 %)	0.40 (0.16, 0.99)	0.0464

CI = confidence interval; GED = General Education Development Test; ref. = reference level.

^aOdds ratios are unadjusted, estimated by a separate logistic regression model (of probability that preferred role = ‘Passive’) for each characteristic. All models except for education include corrections for clustering using a random intercept for site and robust standard errors adjusted for small sample size according to the method of Morel, Bokossa, and Neerchal [^5].

^bDue to having a cell size <5, the model for education was specified using exact logistic regression, adjusted by a fixed effect of site to correct for clustering. Without a random site effect, these estimates are specific to the 6 particular clinical sites included in the sample (results of this model are not generalizable to other clinical sites).

^cModel for income excludes 20 participants with missing income data (8.6 % of the total sample; 18 in the ‘Active/Collaborative role’ group and 2 in the ‘Passive role’ group). In a sensitivity analysis treating missing as a separate income category in the model, results were the same (the p-value for the comparison between non-missing income categories was nearly identical, and the difference between the ‘missing’ category and reference level was not statistically significant, though estimates were not valid due to small cell size).

4. Discussion

This study describes patient preferences for control in medical decision making over the 6 months following the initiation of an evaluation for LVAD, a particularly high intensity medical decision in the setting of life-threatening illness. It demonstrated that roughly half of the participants changed their preferred roles during the study. Descriptive analysis showed trends of increased percentages of patients preferring active roles from baseline to follow-up time points. This qualitatively suggests that many patients preferred an active role as their plans of care progressed. Perhaps an increase in quality of life post-LVAD treatment (due to improved heart failure symptoms) affected patients' outlook on their health care and subsequently their preferred role in decisions about that health care. This indicates that clinicians should practice shared decision making over the entire course of patient care and be aware of and sensitive to changes in communication styles and patient needs. The findings also suggest that patients with higher income and education levels are more likely to prefer an active role in high-stakes decisions such as LVAD implantation. College educated patients prefer an active role significantly more than their counterparts, suggesting that engaging non-college educated patients in a meaningful shared decision-making discussion is crucial to ensure their values are considered. This highlights the need for personalized care and tailored levels of patient engagement for such a high-stakes decision as LVAD.

Our findings are comparable to those from other medical decision-making studies. One study of adult patients found that younger age, female sex, lower morbidity, and higher education status were associated with preference of an active role [6]. Another study investigated the control preferences of surrogate decision makers in the intensive care setting. Surrogates' control preferences were associated with age, personality traits and education level [7]. Our study goes beyond these studies to provide a longitudinal look at the stability of control preferences following progression into advanced heart failure and confronting a major medical decision with long-term implications. Not only are patients' CPS followed over time, but CPS is also collected prior to treatment decision, capturing real-time preferred role data in the midst of such a complex medical decision process of LVAD decision making.

4.1. Limitations

This study was a secondary analysis of data collected in the DECIDE-LVAD trial, but this was a prespecified secondary analysis. This study is applicable to patients in the unique situation of considering LVAD implantation; thus, applying the results to a broader population should be done with caution. Missing participant data led to a cohort of patients whose CPS is unknown and could have been more passive, and that missing data was concentrated among the participants who did not undergo implantation of LVAD, in part because they were more likely to die [3]. In collecting patients' reported role in decision making, we are unable to capture what level of control actually occurred in the clinical encounter; however, their preferred role did strongly align with their reported actual control level during the decision. Finally, the population consisted largely of white males, as is the case in essentially all LVAD studies, limiting analysis regarding the effect of race/ethnicity and gender on control preferences. Having a homogenous sample prevents us from understanding ethnic/cultural differences in patient's preferred role in decision making.

5. Conclusion

Most patients in high stress situations want to be engaged in shared decision making. The degree of desired involvement, however, is varied and may be affected by various social determinants of health.

Funding

Research reported in this publication was funded through Patient-Centered Outcomes Research Institute (PCORI) Awards (CDR-1310-06998). The views, statements, opinions in this work are solely the responsibility of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute (PCORI), its Board of Governors or Methodology Committee.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr. Allen reports grant funding from AHA, NIH, and PCORI, and reports consulting fees from ACI Clinical, Amgen, Boston Scientific, Cytokinetics, and Novartis. All other authors declare that they have no conflicts of interest.