Contextual factors related to ventilator mode during an embedded randomized trial: a longitudinal survey of clinicians

Abstract

Background

Pragmatic trials can efficiently generate comparative effectiveness evidence by embedding delivery of interventions assigned by the trial with the people and systems of clinical care. Delivering assigned treatments with high fidelity is essential to generating informative trial results, but which factors affect clinicians’ adherence to assigned treatments in pragmatic trials is not known.

Methods

Observational study assessing clinician perceptions of trial rationale, conduct, and interventions during a pragmatic, randomized trial of three ventilator modes (volume control vs. pressure control vs. adaptive pressure control) in a single intensive care unit. Clinicians were surveyed before and monthly during the 9-month trial. Survey instruments assessed readiness for the trial using the Hospital Change Readiness Questionnaire (scored from 7 to 49), Proctor’s Implementation Outcomes, and comfort with each mode on a Likert scale of 1–7. We hypothesized that perceptions would differ by ventilator mode, clinician type (respiratory therapists, advanced practice providers, fellow physicians, attending physicians) and over time. Comparisons between groups were performed by chi-squared and ANOVA tests, and tests for change over time were performed by generalized estimating equations.

Results

Overall, 82 of 120 clinicians (68%) responded to at least one survey, with a median response rate of 51% (IQR 48%-56%) per survey. Pre-trial readiness scores differed by clinician type and were lowest among fellows (median 43 [IQR 39–45]) and highest among respiratory therapists (49 [48–49]) (P < 0.01). During the trial, comfort differed by ventilator mode with a median of 7.0 (6.6-7.0) for volume control, 6.3 (6.0–7.0) for pressure control, and 6.6 (6.0–7.0) for adaptive pressure control (P < 0.01). Comfort with using each mode also differed by clinician type and over time. Perceptions of appropriateness (6.8 [6.3-7.0]), acceptability (6.3 [5.8-7.0]), and feasibility (6.5 [6.0–7.0]) for conducting the trial were high, similar between clinician types, and consistent over time.

Conclusions

For a pragmatic trial of ventilator modes, clinician perceptions of readiness, appropriateness, acceptability, feasibility, and comfort with trial interventions were high, and differences between modes and clinician types were small. To optimize trials embedded in healthcare systems, future work should survey across multiple sites and use interviews to attain a broader perspective on contextual factors affecting fidelity.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-026-14835-1.

Background

Pragmatic randomized clinical trials are capable of efficiently generating evidence to inform care by embedding study procedures with the people and the systems of clinical care [1, 2]. In an embedded pragmatic trial, the same clinicians who would administer the treatment in care outside of the trial are responsible for delivering the treatment assigned by the randomized trial. Barriers to clinicians delivering the assigned treatment threaten the informativeness of the trial by reducing separation between study groups [3–5] and increasing the probability the trial will fail to detect a true difference in outcomes between treatments, if one exists [6, 7].

Barriers to delivery of the assigned treatment include factors related to the intervention itself and to the clinical context in which the pragmatic trial is embedded [8, 9]. Clinician-related factors, in particular, are an important dimension of clinical context for these trials, and the barriers perceived by clinicians may vary by clinician type and over time with experience gained during the trial [10, 11]. Methods from implementation science could be applied to embedded randomized trials to understand which factors affect adherence to treatments assigned by the trial and delivered in clinical care, but few trials have done so in a rigorous manner [12–17].

In the pragmatic, embedded Mode of Ventilation During Critical Illness (MODE) trial, comparing ventilator mode choice in a single intensive care unit [18], the assigned intervention of ventilator mode setting (volume control vs. pressure control vs. adaptive pressure control) was at risk of low fidelity due to the continuous, dynamic, and multi-disciplinary nature of managing mechanical ventilator settings [12]. Clinicians may also differ from one another in their adherence to ventilator modes assigned by the trial due to their training, preferences, and familiarity with each mode, both at baseline and with experience during the trial. Whether clinician perceptions are contextual barriers to a trial comparing ventilator modes and whether those perceptions vary by clinician type and time is not known. During the MODE trial, we conducted an ancillary longitudinal study of clinician perspectives regarding factors that may affect adherence to the assigned study interventions prior to and during enrollment. We hypothesized that clinician perceptions would vary by clinician type and over time in ways that could inform conduct of future trials.

Methods

We conducted a longitudinal study of clinicians who cared for patients enrolled in the MODE trial. The primary exposure variables were clinician type, study month (1 to 9), and ventilator mode assigned during the study month (volume control, pressure control, adaptive pressure control). Outcomes of interest were perceptions of the trial and trial procedures as well as perceptions of the study interventions (ventilator modes), assessed by surveys.

Population

The MODE trial was conducted in the medical intensive care unit of an academic medical center from November 1, 2022 to July 31, 2023. During the trial, the entire study unit was assigned to use one of three study modes (volume control, pressure control, and adaptive pressure control, NCT05563779) for all eligible patients receiving invasive mechanical ventilation (cluster-level allocation), and the assigned mode changed each month in random sequence (cluster-level crossover) in permuted blocks of 3 for the 9 months of the trial. If clinicians felt a different mode was necessary for optimal care of a given patient (e.g., to maintain lung-protective tidal volume goals, to improve patient comfort) they were permitted to make changes as necessary.

We distributed surveys by email to all clinicians who worked in the study unit during the MODE trial and held a clinical role that involved adjusting mechanical ventilators, which included respiratory therapists, advanced practice providers, in-training fellow physicians, and supervising attending physicians. Respiratory therapists have completed a respiratory therapy degree (2 years) and are responsible for managing therapies like mechanical ventilation through protocolized patient assessments at least every 4 h. Advanced practice providers have completed training as a nurse practitioner or physician’s assistant (2 to 3 years) and care for patients as primary clinicians. Fellow physicians have completed medical school (4 years) and a residency (3 years) and are training in pulmonary and critical care medicine (3 years). Attending physicians have completed medical school, residency, and fellowship and are responsible for supervising the clinical team and guide overall patient care.

At each study timepoint, surveys were distributed to all clinicians with experience relevant to that timepoint. The pre-trial survey was distributed immediately before the enrollment began in the MODE trial to all clinicians who were expected to work during the trial, and the during-trial survey was distributed in the final 5 days of each of the study months to all clinicians who were known to have worked during that month.

Survey instrument

Clinicians were surveyed using brief and repeated instruments throughout the MODE trial, distributed before the trial began and monthly during the trial [19]. The surveys asked respondents for basic demographic information (age, clinical role, years in practice) at the first survey response, and then subsequently asked about perceptions of the MODE trial and trial procedures using both validated and trial-specific questions. The pre-trial instrument contained some questions unique to the pre-trial instrument, and all surveys contained 5 consistent questions repeated throughout the study. (All survey questions are provided in the Supplementary Appendix)

The pre-trial survey included the Hospital Change Readiness Questionnaire, a 7-question instrument validated for inpatient staff to assess variability in perceptions of readiness for the trial with responses ranging from 7 (indicating the lowest readiness for the trial) to 49 (indicating the highest readiness for the trial) [20]. The pre-trial surveys also asked clinicians about their attitudes toward the study interventions. Respondents were asked to select which mode they believed was better for most patients (if any) and which mode they preferred to use for most patients (if any).

All during-trial survey instruments included questions assessing perceptions of the trial as a whole with select measures from Proctor’s Implementation Outcomes (appropriateness, acceptability, and feasibility) [21, 22], and comfort using the study mode assigned during that month, which solicited responses on a Likert-scale (1, strongly disagree to 7, strongly agree). Finally, the respondents were asked whether they had experienced any of 6 common barriers to delivering the assigned mode to patients.

Procedures

Surveys were distributed by email and responses were collected through the secure REDCap online database [23]. At each timepoint, reminders were distributed to clinicians who had not responded within 4 days of the initial survey invitation. Prior to receiving the survey, respondents were provided a brief summary of the study and consent was obtained through the REDCap instrument. This study design and survey instruments were approved prior to the start of data collection by the Vanderbilt University Medical Center Institutional Review Board (#221622) to be conducted with a modification of informed consent.

Statistical analysis

To test for differences by clinician type, chi-squared and analysis of variance tests were performed for categorical and continuous variables, respectively. The mean value from each clinician was used when the same clinician provided responses over multiple surveys. To test for differences over time, a generalized estimating equation to account for within-subject correlation was used. The association between longitudinal measures and adherence to treatment assignment was explored visually. Due to the exploratory nature of this analysis, no adjustments were made for multiple testing, and all analyses were considered hypothesis-generating. All statistical analyses were performed on SAS Studio 3.81 (SAS Institute Inc., Cary, NC).

Results

From October 2022 to July 2023, surveys were distributed at 10 time points to a total of 120 eligible clinicians. A total of 82 clinicians (68%) responded to at least 1 survey. Across the 10 time-points, the median response rate was 51% (interquartile range [IQR], 48–56%) per survey (Table S1 in the Supplementary Appendix). Of all respondents, 23% were attending physicians, 17% were fellow physicians, 16% were advanced practice providers, and 44% were respiratory therapists (Table 1).

Table 1.

Demographics of the survey respondents

	Overall	Attending Physicians	Fellow Physicians	Advanced Practice Providers	Respiratory Therapists
Response to any survey	n = 82 /120 (68%)	n = 19/21 (90%)	n = 14/17 (82%)	n = 13/19 (68%)	n = 36/63 (57%)
Age, years, median (IQR)	35.5 (33–49)	40 (36–47)	32 (30–33)	37.5 (33-40.5)	34 (32–42)
Time since terminal degree, years, median (IQR)	9 (5–13)	13 (10–20)	5 (4–5)	7 (6–11)	9.5 (3.5–14.5)

Definitions of Abbreviations: IQR: Interquartile range

Age is missing for 8 respondents

Time since terminal degree is missing for 6 respondents

Before the trial began, Hospital Change Readiness scores among clinicians had a median of 44 (IQR, 42 to 49) of 49 points with a higher score indicating greater perception of readiness, and the scores differed by role (P < 0.01 for difference across 4 roles). Fellows reported the lowest readiness with a median score of 43, and respiratory therapists reported the highest readiness with a median of 49 (Table 2). Most clinicians expressed the belief that no single mode was better for most patients (75%), but approximately half expressed a preference for an individual mode. Volume control was the most common preference (42%). No clinicians expressed a belief that pressure control was better or that pressure control was their preference.

Table 2.

Clinician attitudes and perceptions of MODE trial and trial interventions prior to the trial start

	Overall	Attendings	Fellows	APPs	RTs	P
	n = 34	n = 8	n = 12	n = 6	n = 8
Hospital Change Readiness Score for MODE trial (out of 49), Median (IQR)	44 (42–49)	46 (43–49)	43 (39–45)	43 (40–49)	49 (48–49)	<0.01
Belief regarding study interventions* (n = 32), No. (%)
Volume control is better	4 (13%)	0 (0%)	1 (9%)	1 (17%)	2 (25%)
Pressure control is better	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Adaptive pressure control is better	4 (13%)	0 (0%)	1 (9%)	0 (0%)	3 (38%)
No mode is better	24 (75%)	7 (100%)	9 (82%)	5 (83%)	3 (38%)
Preference regarding study modes, No. (%)
Preference for volume control	15 (44%)	2 (25%)	9 (75%)	2 (33%)	2 (25%)
Preference for pressure control	0 (0%)	0 (0%)	0 (0%)	0 (0%)	0 (0%)
Preference for adaptive pressure control	4 (12%)	0 (0%)	1 (8%)	0 (0%)	3 (38%)
No preference for one mode	15 (44%)	6 (75%)	2 (17%)	4 (67%)	3 (38%)

Abbreviations: APP: Advanced Practice Provider (e.g. Nurse Practitioner, Physician’s Assistant); RT: Respiratory Therapist; IQR: Interquartile range; No.: Number

*1 Attending and 1 Fellow declined to answer this question

Perceptions of appropriateness (median 6.8 [IQR, 6.3-7.0]), acceptability (6.3 [5.8-7.0]), and feasibility (6.5 [6.0–7.0]) for conducting the randomized trial were high, similar between clinician type, and similar over time (Fig. 1, Table S2 in the Supplementary Appendix).

Fig. 1.

Clinician perceptions of appropriateness, acceptability, and feasibility of the MODE trial during the trial. Shown are the responses to survey questions regarding the appropriateness of the MODE trial, acceptability of letting the trial randomize mode assignment, and feasibility of the MODE trial. Responses were on a Likert Scale from 1 (strongly disagree) to 7 (strongly agree) and values presented here are the mean and 95% confidence intervals of responses at each timepoint (Table S2 in Supplementary Appendix). Data was collected prior to the start of the trial and subsequently at the end of each of the 9 months of the trial. In total, 67 respondents contributed at least once, and details on the number of respondents at each timepoint are available in Table S1 of the Supplementary Appendix. To test for change over time while accounting for within-subject correlation, a generalized estimating equation was used and P-values are provided for each measure

During the trial, perceptions of comfort with ventilator modes differed by the mode with a median of 7.0 (6.6-7.0) (Likert scale of 7) for volume control, 6.3 (6.0–7.0) for pressure control, and 6.6 (6.0–7.0) for adaptive pressure control (P < 0.01) (Table 3). Mean comfort also differed by role in pressure control, for which comfort was lowest among fellows (median 5.3 (5.0-5.7)) and highest among respiratory therapists (median 7.0 (6.3-7.0)) (P < 0.01). Small but statistically significant increases in comfort were observed over time in the volume control and pressure control groups (Fig. 2 and Table S3 in the Supplementary Appendix). Overall adherence to the assigned mode in the trial was 88.6% of all eligible ventilator-delivered breaths, was high for all three modes but was lowest in the volume control group (less than 80% in each block).

Table 3.

Clinician comfort with use of the study modes during the trial

Mean value for each respondent (out of 7)	Overall	Attendings	Fellows	APPs	RTs	P
	n = 80	n = 19	n = 14	n = 13	n = 34
Comfort with using*…
Volume control, median (IQR)	7.0 (6.6-7.0)	7.0 (7.0–7.0)	6.7 (6.3-7.0)	7.0 (6.6–6.6)	7.0 (7.0–7.0)	0.77
Pressure control, median (IQR)	6.3 (6.0–7.0)	6.0 (6.0–7.0)	5.3 (5.0-5.7)	6.3 (6.0–7.0)	7.0 (6.3-7.0)	< 0.01
Adaptive pressure control, median (IQR)	6.6 (6.0–7.0)	6.0 (6.0–7.0)	6.0 (5.7–6.5)	6.8 (6.0–7.0)	7.0 (6.5-7.0)	0.03

Abbreviations: APP: Advanced Practice Provider (e.g. Nurse Practitioner, Physician’s Assistant); RT: Respiratory Therapist; IQR: Interquartile range

N refers to number of respondents that provided a response to any question related to comfort at any survey timepoint. Not all respondents provided an answer for all modes because questionnaires issued during the trial only asked about comfort with one of the three modes

*Comfort was graded by respondents on a Likert scale with 1 = strongly disagree, 4 = neither agree nor disagree, and 7 = strongly agree. For respondents that provided multiple responses for comfort with the same mode, the mean of their responses was used in this analysis

Fig. 2.

Clinician comfort with use of trial interventions and adherence to the assigned intervention during the trial. A represents the comfort with each of the trial ventilator modes reported by clinicians over time during the MODE trial, organized by trial block. Clinicians were surveyed prior to the beginning of the trial and at multiple time points during the trial. Each trial block was 3 months in duration, during which a different mode was assigned for each month and clinicians were surveyed at the end of each month. Responses were on a Likert Scale from 1 (strongly disagree) to 7 (strongly agree) and values presented are the mean and 95% confidence intervals of responses for each mode over the trial block (Table S3 in the Supplementary Appendix). To test for change over time while accounting for within-subject correlation, a generalized estimating equation was used, and p-values are provided for each ventilator mode group. B presents the recorded adherence to the assigned intervention (ventilator mode) during the trial. Adherence is reported as the percentage of breaths that patients in each study group received (recorded every minute with 1,119,699 values) using the assigned mode, rather than either of the other two study modes. Breaths were recorded from enrollment until the first of extubation, hospital discharge, death, or the end of the study month. Within each block of 3 months, the order of mode assignment was randomized but presented here in consistent order for presentation purposes

Clinicians reported barriers to adherence to the assigned mode during the trial. Overall, most clinicians reported some barrier to pressure control (68%), and few reported barriers to adaptive pressure control (30%) (Table 4). One of the most common reasons was the behavior of other clinicians (49% for volume control), but very few clinicians reported their own lack of knowledge or experience as a barrier. Other barriers were related to differences in properties inherent to each ventilator mode and differed between modes.

Table 4.

Perceptions of barriers to delivery of the study modes during the trial

Barriers:	Overall*	Volume Control	Pressure Control	Adaptive Pressure Control
	n = 283	n = 95	n = 84	n = 104
Clinician factors -
My own lack of knowledge/experience, No. (%)	18 (6%)	1 (1%)	11 (13%)	6 (6%)
Other team members, No. (%)	78 (28%)	47 (49%)	21 (25%)	10 (10%)
Intervention factors -
Patient discomfort, No. (%)	40 (14%)	24 (25%)	7 (8%)	9 (9%)
Maintaining lung protective ventilation, No. (%)	43 (15%)	18 (19%)	22 (26%)	3 (3%)
Additional ventilator adjustments, No. (%)	31 (11%)	4 (4%)	24 (29%)	3 (3%)
Additional ventilator monitoring, No. (%)	41 (14%)	0 (0%)	17 (20%)	24 (23%)
Other, No. (%)	9 (3%)	3 (3%)	4 (5%)	2 (2%)
None, No. (%)	150 (53%)	50 (53%)	27 (32%)	73 (70%)

Abbreviations: No.: Number

*Data represents all survey responses across the duration of the study. Clinicians may have contributed multiple responses if they were eligible at multiple survey time-points

Discussion

This longitudinal study examining clinician perceptions during an embedded randomized trial of ventilator modes in an intensive care unit has two major findings. First, clinician perceptions of readiness, appropriateness, acceptability, feasibility, and comfort with the study interventions were high, often at the ceiling of the validated measurement scales, and differences between clinician types and over time during the trial were small in magnitude. Second, the ventilator mode for which clinicians expressed the highest preference (volume control) also had the lowest adherence when assigned by the trial, suggesting that factors beyond clinician preference may drive adherence to the assigned ventilator mode. These findings have important implications for future embedded trials of ventilator mode and for the use of implementation science methods more broadly to optimize randomized trials embedded in critical care.

The Hospital Change Readiness Score was used in this study to determine if modifiable barriers were perceived prior to the trial, because the score is tailored to assess readiness for change among busy, inpatient clinicians and has been validated in this setting [20]. The high degree of readiness expressed by respiratory therapists informed trial planning, but the small difference between groups is difficult to interpret. Furthermore, determinants of readiness for a clinical trial may differ from those of readiness for other changes to clinical care [24].

Perceptions of appropriateness, acceptability, and feasibility were high among clinicians, with nearly all respondents expressing agreement at all timepoints. The ceiling effect with these instruments suggests that their utility may be limited in the setting of pragmatic trials, which are already designed to minimize known barriers. Different instruments or methods such as semi-structured interviews may be needed to understand implementation determinants for pragmatic trial interventions.

Regarding perceptions of the interventions, clinicians expressed high levels of comfort with each mode, differing by mode and by clinician type, but these differences were small between the levels of “agree” and “strongly agree” on the Likert scale. Similarly, the apparent increase in comfort during the trial with volume control and pressure control modes may represent a learning curve, but the change is small in magnitude and may be obscured by a ceiling effect. Prior studies have characterized clinician preferences for choosing ventilator modes in clinical care outside of research and found wide variation in mode use overall and for specific patient populations [25, 26]. Clinician comfort and mode selection in clinical care may differ from their perceptions of using a ventilator mode assigned by a randomized trial. Clinicians in this study expressed a high degree of comfort with using all three ventilator modes, consistent with the perceptions of trial feasibility, but these measures may vary across other sites and clinicians.

The ventilator mode for which clinicians expressed the highest preference and comfort (volume control) also had the lowest adherence to mode assignment. These findings are consistent with Gibbs and colleagues, who compared use of volume control and adaptive pressure control at a single center with a concurrent survey of clinicians regarding mode preference, comfort, and thematic analyses of free-text responses [12]. This incongruity between clinician perceptions and fidelity in the trial has multiple potential causes. First, clinicians may be a poor judge of their own knowledge and ability to use the modes. Second, a small number of clinicians may determine care for a large number of patients. Both possibilities could also explain the high proportion of respondents citing other clinicians as a barrier. Third, intervention-factors may be driving the difference in adherence, rather than clinician-factors. For example, volume control is associated with patient-ventilator dyssynchrony events which may lead clinicians to more often change modes away from volume control in the trial, thereby reducing adherence [27]. Outside of research, clinicians may not find the mode changes harmful or relevant to their preferences for each mode. While perceptions of comfort with a study intervention may capture valuable differences in some trials, other factors are needed to explain the differences in adherence during this trial of ventilator modes – it is not known whether clinicians could have predicted or perceived this difference.

This study has multiple strengths. We used validated measures and received responses from more than 50% of those surveyed. Multiple types of clinicians were responsible for the management of mechanical ventilation in the study unit during the trial, and each was well represented among respondents. the use of repeated measures over time and across clinician types provides a depth of data to understand the responses across stakeholders. Finally, this use of contextual assessment in embedded pragmatic trials represents an important new application for these methods. We make a distinction between the use of implementation science methods to assess adherence to the assigned intervention in an embedded pragmatic randomized trial (which facilitates generation of data for effectiveness) and type 1 implementation-effectiveness hybrid trials, which assess the delivery of interventions within a trial in order to apply the intervention after the trial (which provides insights if the intervention is found to be effective) [28]. Instruments designed specifically for embedded pragmatic trials are needed to guide strategies and adaptations that can improve the adherence to the assigned intervention. Such tools could optimize the separation between groups, which is a prerequisite to generating informative results from randomized trials capable of improving clinical care.

This study has limitations. In voluntary surveys of busy clinicians, response biases are likely present. The survey respondents represent the clinicians who were involved in the trial at the study site, but different types of clinicians may be involved in a similar trial at other sites, limiting the generalizability of these findings. Results from the quantitative survey measures demonstrated a ceiling-effect which reduced variability in responses and did not provide dimensionality needed to guide significant implementation strategies or adaptations in this pilot trial. The statistical analyses performed were not adjusted for multiple-testing and should be interpreted as hypothesis-generating. Different instruments and qualitative methods may better capture important factors that affect delivery of study interventions.

Conclusions

Longitudinal surveys conducted using established implementation questionnaires of clinicians before and during an embedded pragmatic trial can provide granular data on the clinical context of the research. Within a randomized trial of ventilator modes, perceptions of readiness, appropriateness, acceptability, feasibility, and comfort with trial interventions were high, and differences between clinician types were small in magnitude. Future work should consider collecting data across multiple sites, incorporating additional domains, or employing interviews to attain a fuller picture of contextual factors important for fidelity in embedded clinical trials.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MODE

Mode of Ventilation During Critical Illness Trial

IQR

Interquartile range

Author contributions

KS, MS, JC, AM conceived the research question and study design; KS, AM designed the study and survey instruments; KS distributed the surveys and collected data; KS and AM analyzed the data; KS led the preparation of the manuscript; all authors interpreted data, have drafted the work or subsequently revised it, and read and approved the final manuscript.

Funding

Data collection in this study was performed with the use of the Research Electronic Data Capture (REDCap) tool, which was developed and maintained with grant support (UL1 TR000445) from the National Center for Advancing Translational Sciences to the Vanderbilt Institute for Clinical and Translational Research. During the conduct of the research, KPS was supported in part by the NIH/NHLBI (T32HL087738), LMB was supported in part by the NHLBI (K12HL137943) and the American Association of Critical Care Nurses, JAP was supported in part by the NIH (K23 AG073529), SCD was supported in part by the NIH (T32GM108554), MWS and TWR was supported in part by the NIH/NCATS (UL1 TR002243), and JDC was supported the NIH/NCATS (U24TR004737-02) and PCORI (BPS-2024C1-37478).

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to the potentially identifiable nature of the full dataset format, but subsets of the data are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved to be conducted with an alteration of informed consent (Vanderbilt University Medical Center Institutional Review Board, Integrated Sciences Committee; IRB#: 221622), and all work related to the study was performed in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.