Effectiveness of an Algorithmic Approach to Ventilator Withdrawal at the End of Life: A Stepped Wedge Cluster Randomized Trial

Abstract

Background:

The transition to spontaneous breathing puts patients who are undergoing ventilator withdrawal at high risk for developing respiratory distress. A patient-centered algorithmic approach could standardize this process and meet unique patient needs because a single approach (weaning vs. one-step extubation) does not capture the needs of a heterogenous population undergoing this palliative procedure.

Objectives:

(1) Demonstrate that the algorithmic approach can be effective to ensure greater patient respiratory comfort compared to usual care; (2) determine differences in opioid or benzodiazepine use; (3) predict factors associated with duration of survival.

Design/Settings/Measures:

A stepped-wedge cluster randomized design at five sites was used. Sites crossed over to the algorithm in random order after usual care data were obtained. Patient comfort was measured with the Respiratory Distress Observation Scale© (RDOS) at baseline, at ventilator off, and every 15-minutes for an hour. Parenteral morphine and lorazepam equivalents from the onset of the process until patient death were calculated.

Results:

Usual care data n = 120, algorithm data n = 48. Gender and race were evenly distributed. All patients in the usual care arm underwent a one-step ventilator cessation; 58% of patients in the algorithm arm were weaned over an average of 18 ± 27 minutes as prescribed in the algorithm. Patients had significantly less respiratory distress in the intervention arm (F = 10.41, p = 0.0013, effective size [es] = 0.49). More opioids (t = −2.30, p = 0.023) and benzodiazepines (t = −2.08, p = 0.040) were given in the control arm.

Conclusions:

The algorithm was effective in ensuring patient respiratory comfort. Surprisingly, more medication was given in the usual care arm; however, less may be needed when distress is objectively measured (RDOS), and treatment is initiated as soon as distress develops as in the algorithm.

Clinical Trial Registration number: NCT 03121391.

Background

It is estimated that 1 in 5 (20%) of the 2.4 million people who die each year in the United States will do so in an intensive care unit (ICU) or shortly after an ICU stay.¹ Many ICU deaths are preceded by the withdrawal of invasive mechanical ventilation as a palliative procedure, although the frequency of this decision and the subsequent patient care varies globally.² Withdrawal of mechanical ventilation, if not performed correctly, can lead to patient respiratory distress and suffering.

Withdrawal of invasive mechanical ventilation, the focus of this study, entails a transition to unsupported spontaneous breathing to afford a natural death. The most common approaches are terminal weaning, a staggered reduction of oxygen and ventilation, and terminal extubation in which a single step of ventilator off and endotracheal tube removal is done. Currently, patient care decisions are guided by clinician preferences rather than an evidence-based approach.³ Since the empirical evidence to guide the conduct of this common procedure is largely observational, clinicians rely on intuition, varying levels of experience, or local practice customs.

Small samples, observational studies, retrospective chart reviews, and qualitative descriptions characterize the body of evidence about processes for ventilator withdrawal.^4,5 Available evidence suggests that there is (1) a lack of a standard, objective measure for detecting respiratory distress to guide the process; (2) high variability in initiation and escalation of opioids and sedatives⁶; and (3) an inability to predict the method (weaning vs. extubation) that best ensures patient comfort without hastening death.^4,5 Only one small sample pilot investigation comparing one method to another has been conducted establishing preliminary efficacy of this algorithmic approach.⁷ The ideal best practice process for conducting ventilator withdrawal across a heterogeneous population must account for the variance in patient experience. A patient-centered algorithm guided by an objective measure of respiratory distress would provide the best standard. The purpose of this study was threefold: (1) to determine the effectiveness of a principal investigator (PI)-developed algorithm compared to usual care; (2) ascertain differences in medications used; and (3) investigate the impact of select variables on duration of survival.

Design

A stepped-wedge cluster design with repeated measures option was used (Fig. 1). The stepped-wedge cluster randomized trial is a form of cross-over design with unidirectional cross-over from control (usual care), to intervention (algorithm), but with randomization of when each cluster undertakes this transition.^8,9 All clusters started the trial in a control phase, then sequentially crossed over from the control group to the intervention group, until all sites received the intervention. Site simple randomization occurred assigning them to order of crossover to intervention; allocation was not concealed at the cluster level. This design was chosen to avoid contamination of the intervention, that is, patients and the medical, nursing, and respiratory therapist (RT) staff caring for them under different experimental conditions were not comingled. Importantly, the intervention was not removed once it was implemented and demonstrated to be effective, which alleviates ethical concerns.⁹ Outcomes were measured on the study participants in all clusters over time; hence, measurement of outcomes takes place at each step in the wedge; each cluster provides data points in the control and intervention conditions allowing each site to act as its own control.

FIG. 1.

Study diagram.

Algorithm interventions and procedures

The algorithm was PI-developed based on a large palliative care clinical experience. The algorithm is designed to be nurse-led with RT support. There are five processes: (1) prewithdrawal assessment (Richmond Agitation and Sedation Scale [RASS] and Respiratory Distress Observation Scale [RDOS]); (2) premedication decisions; (3) withdrawal method (weaning vs. one-step); (4) extubation considerations; and (5) post-withdrawal oxygen and opioid continuous infusion decisions. All 5 sites began the trial in the control arm with crossover to intervention after 13 patients (cluster 1), 26 patients (cluster 2), 39 patients (cluster 3), and so on (Fig. 1) (Details are in the Supplementary Material).

Eligibility and exclusion criteria

Adult patients aged ≥18 years, undergoing the withdrawal of invasive mechanical ventilation at the end of life were eligible for inclusion. Patients who were conscious and cognitively intact at the time of ventilator withdrawal were excluded because they constitute a different population from the usual patient undergoing this procedure and are likely to require a different approach. Patients who were planned to undergo organ donation after cardiac death following ventilator withdrawal were also excluded because we were not able to collect data from this population. Patients who were brain dead were excluded because they have no ability to experience respiratory distress. Patients with bulbar amyotrophic lateral sclerosis, C1–C4 quadriplegia, or locked-in syndrome secondary to brainstem disease were excluded because the respiratory distress measure is not valid under conditions that produce paralysis. Consent for observation of the ventilator withdrawal processes was obtained from the patient's legally authorized individual, usually family.

Settings

We enrolled study participants from five regional medical intensive care units (MICUs). All MICUs were closed units with attending intensivists, fellows, residents, and students rounding each day for medical decisions. Nurse-to-patient ratios were similar across sites. No site had a standardized ventilator withdrawal process.

Training

Critical care medical and nursing leadership supported this standard of care change. All nurses and RTs were trained to the algorithm when their site crossed over to implement the intervention.

Usual care group

Ventilator withdrawal was conducted by the usual personnel for that unit. A research assistant (RA) observed the patient from an unobtrusive yet unobstructed place in the patient's room for all observation measures at baseline, after every ventilator change, after the ventilator was turned off, and every 15 minutes after the ventilator was turned off for up to 2 hours.

Algorithm intervention completed by an ICU staff nurse with an ICU RT

The RN assigned to the patient began the processes with support from the assigned RT. As with the usual care group, the RA observed the process and collected data.

Ensuring fidelity to the intervention

The RN/RTs had laminated bedside cards with the processes and decision trees illustrated for reference. RN/RT fidelity to the algorithm was measured by RA observation and check off on a manualized version of the algorithm for all cases.

Measurements

Patient and illness characteristics

Sociodemographic information was obtained from the medical record. Consciousness was assessed with the Reaction Level Scale 85 (RLS85) that was developed for use in the ICU.^10,11 Scores range from 1 (alert with no delay in response) to 8 (unconscious with no response to painful stimuli).

Illness severity was quantified using the Simplified Acute Physiology Score II (SAPS II).^12–16 The SAPS II includes 17 variables: 12 physiology variables, age, and type of admission (scheduled surgical, unscheduled surgical, or medical), and 3 underlying disease variables (acquired immunodeficiency syndrome, metastatic cancer, and hematologic malignancy). SAPS II was calculated with variables from the first 24-hour interval after admission to the MICU.

Oxygenation performance was quantified using a peripheral oxygen saturation/fraction of inspired oxygen ratio (SpO₂/FiO₂) that was calculated using the patient's baseline (prewithdrawal) SpO₂ and FiO₂. SpO₂ was measured by oximetry, and FiO₂ was recorded from the patient's prewithdrawal ventilator settings.

Inference of the patient's ability to experience distress in the algorithm arm was assessed with the RASS. Scores ranging +3 to −3 signified ability to experience distress.

Medications

Opioids, benzodiazepines, propofol, and dexmedetomidine are routinely given to maintain patient comfort and achieve patient–ventilator synchrony in the ICU. Doses were calculated at two time points: (1) total dosage for the 24-hour period before ventilator withdrawal commenced; (2) the total dose administered in the interval from the first ventilator change initiating withdrawal until the patient death. Opioid doses were converted to parenteral morphine equivalents using the formula fentanyl dose ÷10 and hydromorphone × 7; benzodiazepines were converted to lorazepam equivalents using the formula midazolam dose ÷2 and alprazolam × 2 for comparison purposes.

Respiratory distress

The RDOS is an ordinal scale with eight observer-rated parameters: heart rate, respiratory rate, accessory muscle use, paradoxical breathing pattern, restlessness, grunting at end-expiration, nasal flaring, and a fearful facial display. Each parameter is scored from 0 to 2 points, and the points are summed. Scale scores range from 0 signifying no distress to 16 signifying the most severe distress. The scale was PI-developed and has undergone rigorous psychometric testing to establish reliability and validity and cut-point determinations.^17–20 RDOS was measured at baseline before any ventilator changes, immediately after the first ventilator change, immediately after every subsequent ventilator change, after the ventilator was turned off, and every 15-minutes for 2 hours after the ventilator was turned off. RDOS ≤3 signified respiratory comfort.

Duration of survival was the interval from the time of the first ventilator change until the patient died measured in minutes.

Sample size estimation

Based on findings from our pilot study, the effect size for RDOS was 0.38.⁷ This effect size was considered medium to large based on the F-statistic in analysis of variance (ANOVA).²¹ A repeated measures design with 1 between factor and 1 within factor has two groups with 35 participants each for a total of 70 participants were required for the study. This sample size was based on the correlation coefficient of 0.5 between the repeated measure, level of significance of 0.05%, and 80% power. Since the stepped-wedge cluster randomization method was used in this study, given the intracluster correlation coefficient of 0.05, the total number of subjects increased from 70 to 198, which was equivalent to recruiting about 40 subjects for each of the five designated clusters. The study was longitudinal; thus, some missing data were to be expected as some participants could die more rapidly yielding fewer measures. To allow for this and maintain the expected precision of estimators and the minimum detectable effect size, the required sample size was increased by 30% to a minimum total sample size of 260 subjects, which will result in 52 subjects per cluster.^21–23

Analysis

Demographic characteristics are summarized as mean and standard deviation or percentage, as applicable. A multivariate repeated-measures analysis of variance was used to determine the change in patient comfort, operationalized by the total score on the RDOS instrument, within patients, between patients in groups, and within-patients-by-between-patients interactions using the GLIMMIX procedure in SAS. A two-sample t-test was used to compare the total dosages of parenteral morphine equivalents, lorazepam equivalents, and duration of survival between the two groups. Multiple regression equations were calculated to evaluate the potential impact of select variables on survival duration.

Results

The proposal was registered with clinicaltrials.gov. The study was approved as a single IRB by the Institutional Review Board at the PI's University with reliance agreements from four sites and IRB approval from the remaining site; 120 patients were enrolled in the control arm with 48 in the intervention. Participants were enrolled from April 21, 2017 to October 25, 2020. The study was suspended in 2020 due to the coronavirus disease 2019 (COVID-19) pandemic and risk for the RAs resulting in an uneven sample leaving fewer patients in the intervention arm.

Patient characteristics

Patients in the control arm ranged 28–97 years, 29–83 years in the intervention arm (Table 1). There were more men in the control arm. Patients ranged from drowsy (RLS level 2) to unconscious with no response to deep pain (RLS level 8) and were on average unconscious in both arms. The MICU admitting diagnoses were typical with most being respiratory failure, septic shock, and cardiac arrest in both arms. All patients in both arms died as expected.

Table 1.

Participant Characteristics

Variable	Control n = 120	Intervention n = 48
Age, mean (SD)	68 (14.5)	61 (11.5)
Gender, % male	56	38
Race, %	White, 58 Black, 42	White, 60 Black, 40
Illness severity, mean (SD) SAPS II	64 (16.9)	77 (18.6)
Consciousness, mean (SD) Reaction Level Scale	5 (2.2)	6 (1.9)
Baseline SpO2/FiO2, mean (SD)	214 (74.9)	175 (82.2)
Admission diagnosis, %
Respiratory failure	43	29
Septic shock	24	31
Cardiac arrest	16	15
Liver failure	6	15
Stroke	5	3
Heart failure	2	2
Days ventilated, mean (SD)	6.8 (5.5)	7.5 (6.8)

Fi0₂, fraction of inspired oxygen; SAPS, Simplified Acute Physiology Score; SD, standard deviation; Sp0₂, peripheral oxygen saturation.

Patient responses

As indicated in Figure 2, the pattern of change in RDOS scores from the time the ventilator was turned off to 45 minutes was not significant within both groups, that is, no significant interaction between time and group. However, there was a significant difference in mean RDOS scores between the two groups (F = 10.41, p = 0.0013, es = 0.49), that is, the RDOS scores were higher for the control group overtime. Only the first 45 minutes are illustrated as the numbers of deaths increased over time leaving more unevenness in the groups. All patients in the control arm underwent terminal extubation, and RDOS scores ranged from 0 to 16 from ventilator off until last RDOS measure at 120 minutes averaging 3.75. In the intervention arm, more than half (58%) underwent terminal weaning, as prescribed in the algorithm for patients estimated to experience distress, with an average time of 18 ± 27 minutes. RDOS scores during weaning (from baseline to last ventilator change before time off) in the intervention arm ranged 0–11 with an average of 2.34 before ventilator off and an average of 2.95 after the ventilator was turned off (Fig. 3).

FIG. 2.

Change in mean RDOS over time by group. RDOS, Respiratory Distress Observation Scale.

FIG. 3.

Change in mean RDOS over time among intervention arm patients who were weaned.

Medication use

There were no differences in opioid (p = 0.872), or lorazepam (p = 0.781) use in the 24 hours before ventilator withdrawal (Table 2). More opioids and benzodiazepines were given in the control arm during and after withdrawal; only morphine or fentanyl were given in the intervention arm as prescribed in the algorithm. Few patients in either arm received propofol or dexmedetomidine.

Table 2.

Medication Use

Variable	Control (n = 120)	Intervention (n = 48)	p
Dosages in the 24-hour interval before withdrawal
Parenteral morphine equivalents, mg (SD)	88.95 (182.28)	93.96 (179.49)	0.872
Parenteral lorazepam equivalents, mg (SD)	3.06 (13.93)	3.89 (24.2)	0.781
Dosages from beginning of withdrawal until patient death
Parenteral morphine equivalents, mg (SD)	191 (640)	50 (119)	0.02
Parenteral lorazepam equivalents, mg (SD)	4.9 (17.6)	1.3 (4.4)	0.04
Propofol, mg (SD)	6.6 (40.3)	10.3 (31.6)	0.28
Dexmedetomidine, mg (SD)	0.015 (.168)	0.001 (.006)	0.28

Bold values represent statistical significance.

Oxygen use

More oxygen was placed after extubation in the control arm (65%) than in the intervention arm (27%; χ² = 19.23, p < 0.001). The algorithm prescribes room air following extubation.

Duration of survival

The time from when the ventilator was turned off until patient death was measured in minutes. Patients in the control arm lived longer than in the intervention arm (Table 3). Covariates in the regression model included consciousness (RLS85), SpO₂:FiO₂ ratio, vasopressor use, opioid use, and illness severity (SAPS II). Using multiple regression, duration was shorter for patients who had been on vasopressors (t = 2.23, p = 0.028) and for those with worse oxygenation (t = 3.33, p = 0.001). Duration was longer for patients receiving opioids, every 1 mg resulted in 2.8 minutes survival (t = 8.12, p < 0.001) (Table 4).

Table 3.

Duration of Survival from Ventilator Off Until Death

	Control (n = 120)	Intervention (n = 48)	p
Time in minutes, (SD)	1711.55 (3385.40)	721.19 (2153.38)	0.013
Duration log (SD)	5.44 (2.27)	4.02 (1.97)	<0.001

Bold values represent statistical significance.

Table 4.

Regression Analysis Predicting Survival Duration

Variable	B	SEB	Beta	t	p
Oxygenation	8.66	2.50	0.22	3.33	0.001
Vasopressors	−914.59	410.91	0.15	−2.23	0.028
Opioids	2.81	0.35	0.53	8.12	<0.001
R2 = 0.39, F = 23.48, p < 0.001)

B: unstandardized regression coefficient; Beta: standardized regression coefficient.

SEB, standard error bands.

Discussion

Algorithmic approach effectiveness was established in this first, rigorous, clinical trial of a ventilator withdrawal intervention. Our findings comport well with our efficacy pilot study.⁷ In this algorithm, patients who may experience distress are weaned and those unlikely to experience distress may have a one-step terminal extubation, thus both commonly employed approaches may be appropriate according to the patient's context, which is ability to experience distress.

Patients undergoing ventilator withdrawal are heterogeneous. For example, patients choosing ventilator withdrawal for themselves are awake and aware, able to report dyspnea, and often completely dependent on the ventilator, many of whom are residing at home or in long-term care facilities.²⁴ A majority undergoing withdrawal, however, are critically ill, cognitively impaired or unconscious, unable to self-report dyspnea, and may or may not be able to experience respiratory distress depending on the severity of unconsciousness.²⁵ Nevertheless, patients undergoing ventilator withdrawal are at risk for severe respiratory distress in response to respiratory failure if their level of consciousness affords the ability to experience distress. Patients who cannot self-report symptom distress can be at risk for undertreatment because clinicians may underestimate the patient experience.^26,27 Conversely, mechanically ventilated patients may be vulnerable to overtreatment of anticipated distress when medication regimens are not evidence-based or guided by an objective measure of patient distress. Overtreatment may lead to a perception of hastening death which is not an aim of this palliative process.²⁸ Thus, the anticipated experience of the patient will vary greatly.

In the control arm, we found that all patients in each of the study sites were undergoing the same single process, terminal extubation, regardless of their underlying ability to experience distress. Weaning affords an opportunity to capture the earliest signs of respiratory distress as oxygen and ventilation are withdrawn in a staggered fashion; the process can be paused while the patient is medicated before proceeding. Terminal extubation is too rapid if the patient can experience distress since the patient is quickly transitioned from supportive ventilation/oxygenation to spontaneous breathing. As seen with our findings, the rapid escalation to moderate or severe distress may lead to difficulties in achieving patient comfort and perhaps requiring more medication.

Using a valid, reliable objective measure of patient respiratory distress (RDOS) is an important aspect of this algorithmic approach. Some patients do not require medication, and for those who do, the RDOS guides initiation, titration, and achievement of clinical endpoints. Robert et al. defined patient discomfort as “gasping” or “bronchial obstruction” or elevated Behavior Pain Scale scores, which are less sensitive indicators of respiratory distress than the RDOS.³ Mazer et al. also used the Behavior Pain Scale.²⁹

Robert et al. reported more apparent distress among patients in their observation study who underwent terminal extubation compared to patients who were weaned.³ In their study, the units included used either weaning or one-step extubation, and patient respiratory distress was not the major study aim with no reliable measure of respiratory distress. We also found more respiratory distress among the control arm patients in both the pilot and this study all of whom underwent terminal extubation.⁷

The algorithm prescribes only morphine or fentanyl by intravenous bolus in response to an RDOS >3 as these opioids have the strongest evidence base for treating dyspnea in patients who are at the end of life.³⁰ Billings recommended preemptive sedation and analgesia when withdrawing mechanical ventilation considering the patient risk for unrecognized, untreated respiratory distress.³¹ Because some patients have no distress during this palliative procedure, preemptive sedation may lead to overmedication of some patients. This algorithm prescribes pausing the procedure when RDOS escalates >3 and immediately administering an opioid bolus; this approach may obviate the concern about undertreatment. Significantly, more opioids were given in the control arm, although there was more unrelieved distress. Overuse of opioids among patients with little or no distress may account for this difference. In addition, when distress is severe more analgesia is needed to restore comfort, and patients in the control arm had a more profound and rapid escalation from baseline to distress when the ventilator was turned off.

More benzodiazepines were given in the control arms of both the pilot study⁷ and in this investigation than in the intervention arms. Benzodiazepines do not have an evidence base as a first-line treatment of dyspnea.³² In a secondary analysis of the control arm of this study, Obarzanek et al. found medication use across study sites was highly variable, and patients were nearly three times less likely to achieve respiratory comfort with lorazepam administration (p = 0.039).²⁸ Overuse of benzodiazepines for dyspnea has been reported among hospice clinicians.³³

The algorithm prescribes room air after ventilator withdrawal because we established in a previous study that patients who are near death with no respiratory distress do not need oxygen.³⁴ Most patients in the control arm received supplemental oxygen after extubation. This may result from responding to declining oxygenation as measured by pulse oximetry, rather than recognizing that imminent death is preceded by desaturation, and in the face of no distress, oxygen may serve only to prolong dying without promoting comfort. Ceasing continuous oxygen saturation monitoring is recommended.

In our study, patients with more distress in the control arm lived significantly longer than the patients in the intervention arm. This finding corresponds with Robert et al.³ We suspect that a sympathetic response to respiratory distress may account for the longer survival. Promoting longer survival is not a goal nor is hastening death when withdrawing life-sustaining treatment. Hypoxemia and hypotension contributed to shorter survival as was found in other studies.^35–37 Patients lived longer if they received more opioids as was also seen in Mazer et al.²⁹ This finding should alleviate clinician fears that giving opioids during ventilator withdrawal risks hastening patient death.

Our study was limited by an uneven sample between groups that was necessitated by the COVID pandemic. Data were collected from five hospitals within the same region limiting generalizability. Furthermore, we did not enroll any patients with COVID, thus, do not know if this approach is effective in that patient context.

Summary

This nurse-led, RT-supported algorithmic approach to ventilator withdrawal guided by a sensitive measure of respiratory distress (RDOS) is more effective than unstandardized usual care because patients had less respiratory distress in the intervention arm with less medication needed. Promises are made to patient's families when the ICU team shares a decision to withdraw mechanical ventilation that patient respiratory comfort will be the goal.

Authors' Contributions

M.L.C.: conception of the study design, analyses, and interpretation. H.N.Y.: conception of the study design, analyses, and interpretation

Funding Information

National Institute of Nursing Research (NR 015768).

Author Disclosure Statement

No competing financial interests exist.

Supplementary Material

Supplementary Material

CONSORT Extension for Cluster Trials 2012 Checklist