Thirty-four individuals were invited to participate in the Delphi. Thirty-one (91%) individuals participated in one or both rounds. Participants included eight (26%) personal experience experts and 26 (84%) professional experience experts (Table 2). Three panelists (10%) had overlapping personal and professional experience. Fifteen participants (48%) primarily practiced and/or worked in the United States, with the remainder in Europe (10, 29%), Canada (4, 13%), and Australia (2, 6%). Thirteen participants (42%) were female. Professional experience experts’ clinical specialties included adult and pediatric critical care as well as medical and anesthesia critical care. Clinical roles included physician, physical therapist, and nursing.

Round 1 was completed by 28 individuals (82% response rate). In Round 1 of the Delphi, participants scored seven attributes of approaches to handle truncation due to death in critical care clinical trials: accuracy, sensitivity, interpretability, clinical relevance, patient-centeredness, comparability, and familiarity (Table 1).

Accuracy and clinical relevance received the greatest proportion of responses as “Critical” (n = 26, 93%), followed by interpretability (n = 23, 82%), sensitivity (n = 20, 71%), and patient-centeredness (n = 20, 71%). Comparability and familiarity each garnered lower support as “Critical” (n = 14, 50% and n = 7, 25%, respectively) and the highest proportions of participants rating as “Not Important” (n = 3, 11%, and n = 9, 32%, respectively) (Figure 1). Seventeen additional characteristics were proposed by panelists, reviewed by the study team, and consolidated into three new attributes: practicality, mechanistic plausibility, and statistical simplicity.

Importance of attributes for approaches used to analyze patient-centered outcome data missing due to death. A Importance of Accuracy. B Importance of Interpretability. C Importance of Clinical Relevance. D Importance of Patient-Centeredness. E Importance of Sensitivity. F Importance of Practicality. G Importance of Mechanistic Plausibility. H Importance of Statistical Simplicity. aOne participant responded as “Unable to score” for this attribute in Round 1. bOne participant responded as “Unable to score” for this attribute in Round 2. cAttribute added for Round 2

The number of points allocated to a single attribute in the point allocation task in Round 1 ranged from 0 to 50. Accuracy received the greatest median total of points (20 (inter-quartile range (IQR) 14.5, 25), followed by clinical relevance (17, IQR 13.5, 20), patient-centeredness (15, IQR 12, 22.5), interpretability (13, IQR 10, 15), sensitivity (10, IQR 9, 15), comparability (10, IQR 5, 10), and familiarity (5, IQR 0, 10).

Based on scoring and comments from Round 1, several attribute definitions were revised for clarity, and two attributes (comparability and familiarity) were dropped from inclusion in Round 2 due to low scores (Table 1).

Round 2 was completed by 29 individuals (85% response rate), 26 (90%) of whom had also completed Round 1. In Round 2 of the Delphi, participants scored the retained and new attributes from Round 1. Four attributes met the threshold for consensus (accuracy, interpretability, clinical relevance, and patient-centeredness), and were deemed critical qualities of an approach used to account for truncation due to death in critical care clinical trials (Figure 1). Four attributes did not meet the threshold for consensus: sensitivity, practicality, mechanistic plausibility, and statistical simplicity. One panelist did not provide a score in Round 2 for sensitivity, citing concerns with the attribute definition.

When asked to rank attributes from most to least important, accuracy (n = 18, 62%) was most commonly ranked as the most important attribute, followed by clinical relevance (n = 6, 21%), patient-centeredness (n = 2, 7%), and mechanistic plausibility (n = 2, 7%) (Table 1). No participants ranked statistical simplicity, sensitivity, or practicality as the most important attribute. Statistical simplicity (n = 11, 38%), mechanistic plausibility (n = 8, 28%), and sensitivity (n = 5, 17%) were most commonly ranked as the least important attribute.

The number of points allocated to a single attribute in the point allocation task in Round 2 ranged from 0 to 50. Accuracy received the greatest median total of points (20, IQR 15, 25), followed by clinical relevance (15, IQR 10, 17), patient-centeredness (15, IQR 9, 19), interpretability (12, IQR 10, 15), sensitivity (10, IQR 5, 15), statistical simplicity (10, IQR 5, 12), practicality (10, IQR 5, 13), and mechanistic plausibility (10, IQR 5, 12) (Table 1).