Low Tidal Volume Ventilation Is Poorly Implemented for Patients in North American and United Kingdom ICUs Using Electronic Health Records

Background Low tidal volume ventilation (LTVV; < 8 mL/kg predicted body weight [PBW]) is a well-established standard of care associated with improved outcomes. This study used data collated in multicenter electronic health record ICU databases from the United Kingdom and the United States to analyze the use of LTVV in routine clinical practice. Research Question What factors are associated with adherence to LTVV in the United Kingdom and North America? Study Design This was a retrospective, multicenter study across the United Kingdom and United States of patients who were mechanically ventilated. Methods Factors associated with adherence to LTVV were assessed in all patients in both databases who were mechanically ventilated for > 48 h. We observed trends over time and investigated whether LTVV was associated with patient outcomes (30-day mortality and duration of ventilation) and identified strategies to improve adherence to LTVV. Results A total of 5,466 (Critical Care Health Informatics Collaborative [CCHIC]) and 7,384 electronic ICU collaborative research database [eICU-CRD] patients were ventilated for > 48 h and had data of suitable quality for analysis. The median tidal volume (VT) values were 7.48 mL/kg PBW (CCHIC) and 7.91 mL/kg PBW (eICU-CRD). The patients at highest risk of not receiving LTVV were shorter than 160 cm (CCHIC) and 165 cm (eICU-CRD). Those with BMI > 30 kg/m2 (CCHIC OR, 1.9 [95% CI, 1.7-2.13]; eICU-CRD OR, 1.61 [95% CI, 1.49-1.75]) and female patients (CCHIC OR, 2.39 [95% CI, 2.16-2.65]; eICU-CRD OR, 2.29 [95% CI, 2.26-2.31]) were at increased risk of having median VT > 8 mL/kg PBW. Patients with median VT < 8 mL/kg PBW had decreased 30-day mortality in the CCHIC database (CCHIC cause-specific hazard ratio, 0.86 [95% CI, 0.76-0.97]; eICU-CRD cause-specific hazard ratio, 0.9 [95% CI, 0.86-1.00]). There was a significant reduction in VT over time in the CCHIC database. Interpretation There has been limited implementation of LTVV in routine clinical practice in the United Kingdom and the United States. VT > 8 mL/kg PBW was associated with worse patient outcomes.

10] The availability of pooled, anonymized electronic health records offers the opportunity to examine routine clinical practice across multiple centers and longitudinally.We investigated the adherence to LTVV in multiple centers in a retrospective observational study using pooled data from intensive care electronic health records in the United Kingdom and United States. 11,12The aim of the current study was to determine the factors associated with implementation of LTVV and whether these factors were consistent across the United Kingdom and the United States.We hypothesized that baseline clinical features could identify patients who were at risk of not receiving LTVV, and failure to implement LTVV was associated with adverse patient outcomes in routine clinical practice.

Study Design and Data Sources
This retrospective, multicenter analysis included ventilated ICU patients whose data were recorded in one of two databases: one from the United Kingdom and the other from the United States.The UK data were from 11 ICUs from five academic centers (Critical Care Health Informatics Collaborative [CCHIC]) that included 47,391 patient episodes.The US data were obtained from the electronic-ICU collaborative research database (eICU-CRD) and were collected from academic and nonacademic medical centers that used the tele-ICU program called eICU (electronic ICU, Philips Healthcare), which included 200,859 patient episodes.Ethical approvals and data governance details are included in e-Appendix 1.

Inclusion and Exclusion Criteria
The study included all adult (aged > 16 years) intubated patients who had undergone a period of invasive mechanical ventilation and had corresponding arterial blood gas, ventilation (tidal volume [V T ] and positive end-expiratory pressure), height, and sex data.Patient height and sex were used to calculate predicted body weight (PBW). 1 Patients who were ventilated for < 48 h and those who died within 48 h of ICU admission were excluded.Patients with a height < 1.2 m were excluded due to the potential inaccuracy of the PBW formulae and the nonlinear relationship between anatomical dead space and height. 13

Outcomes
The primary outcome was adherence to LTVV and identification of the factors associated with adherence in a multicenter setting.Secondary outcomes included the association between LTVV and all-cause ICU mortality and duration of mechanical ventilation, changes in V T over time (2014 to 2019), and adherence to LTVV during periods of acute hypoxemic respiratory failure (AHRF).

Defining AHRF
We assessed the management of ventilated patients with AHRF, defined as a PaO 2 :FIO 2 ratio < 300 mm Hg while receiving positive end-expiratory pressure $ 5 cm H 2 O and FIO 2 $ 0.4.The PaO 2 :FIO 2 ratios were calculated only for those patients receiving FIO 2 $ 0.4 due to concerns regarding the utility of this measure at low FIO 2 . 14Mild, moderate, and severe AHRF were defined according to Berlin ARDS definition thresholds for PaO 2 :FIO 2 if there were two corroborating blood gas analyses within a 12 h window, an approach used in contemporaneous ARDS studies. 15 6 h window following a qualifying blood gas finding was captured, and the median V T from this period was reported.Because of the anonymization process, we were unable to determine whether patients had been diagnosed with ARDS.

Ventilation Periods and Parameters
Values from periods of noninvasive ventilation were excluded.Ventilatory data were available across the databases at differing measurement frequencies, with some recorded at minute-by-minute intervals and others hourly or less frequently.To unify our analyses across both databases, ventilatory data were abstracted at 1 h

Take-home Points
Study Question: What were the factors associated with adherence to low tidal volume ventilation in the United Kingdom and North America?Results: Shorter patients (most often female) and those with BMI > 30 kg/m 2 were most at risk of not receiving low tidal volume ventilation, and initial ventilator settings were often not altered even in the context of worsened respiratory failure.Interpretation: There has been limited implementation of low tidal volume ventilation in routine clinical practice in the United Kingdom and the United States, and tidal volume > 8 mL/kg predicted body weight was associated with worse patient outcomes.
in Medicine (A.E.), University of Cambridge, Cambridge, England; and the Bloomsbury Institute of Intensive Care Medicine (S.H.), University College London, London, England.intervals using the median value for multiple observations in 1 h.New episodes of ventilation were demarcated by pauses in V T recordings that were > 24 h.Missing values were not imputed.Mechanical ventilation modes were not available in the CCHIC database and were sparsely recorded in the eICU-CRD.We restricted our analyses to periods of controlled ventilation determined by the set and observed ventilator rates, allowing for a 15% difference between these values to determine periods that were likely to be controlled.For example, if the ventilator rate was set to 12 breaths per minute, we would allow up to 14 breaths per minute to define the given period as controlled mode ventilation.Because plateau pressures were not recorded in the CCHIC database, inferences relating to driving pressure or static compliance could not be made.
To identify opportunities to improve practice, we examined the V T during the first 6 h, 24 h (day 1), between 24 and 48 h (day 2), and during periods of the worst AHRF to determine if initial settings were changed over time, or whether there was a persistence of initial practice.Changes in V T were mapped by using Sankey diagrams.
The optimum starting V T was calculated (in milliliters) for patients using cumulative distribution curves based on the heights of patients who were mechanically ventilated in each database.The aim was to reduce the rate of patients receiving > 8 mL/kg PBW to < 10% of patients.

Statistical Methods
Descriptive statistics are reported as median with interquartile range, mean with SD, and counts with percentages.When comparing subgroups of patients, we used ORs, t tests for normally distributed data, and Wilcoxon rank sum tests for non-normally distributed data.ICU mortality, LTVV use, and duration of ventilation were evaluated by using generalized mixed linear models, attributing random effects to the hospital identifier.The interclass correlation coefficient was used to report the explained variance from random effects.Ventilation duration was analyzed by using a log-linked Poisson distribution, and results are reported as the relative risk ratios (RRRs).Because data from some hospitals were sparse, Markov chain Monte Carlo methods with no U-turn sampling (RStan, 1,000 iterations, 500 warmups, two chains) were used to ensure model convergence and check the results from the generalized linear mixed effects models.
Age, sex, height, ethnicity, Acute Physiology and Chronic Health Evaluation (APACHE) scores (APACHE II in CCHIC, APACHE IV in eICU-CRD), and type of admission (medical/surgical) were used as covariates as these were consistently measured and relatively complete in both databases.To assess the interactions between covariates, log odds (logit) plots were used.The effect sizes and statistical significance of interaction terms were checked and the threshold values for effects calculated.To aid interpretability, generalized linear models with random effects were not fitted with interactions.To ensure that we were not inadvertently unmasking the effects of closely related baseline variables (sex and height) when estimating the primary outcome, we performed a separate subgroup analysis of patients stratified according to sex.ORs were expressed as point estimates with 95% confidence limits, and the threshold P value for significance was < .05.The change in practice over time was examined in the CCHIC database by calculating the median V T for a given patient, weighted by duration of ventilation in each month.A linear model for weighted median V T against time was used to estimate this change, which was expressed as milliliters per kilogram PBW per quarter (3-month period).Results were visualized by using a locally estimated scatterplot smoothing curve.
To determine the association between LTVV and patient outcomes, a competing risk model was used to estimate the subdistribution hazards for patients who remained ventilated, had died, or had been extubated 30 days following initiation of invasive ventilation.Adjustments were made for age and APACHE score.These were described using cumulative incidence function curves with censoring at day 30.
Data queries were performed with PgAdmin 4 (version 3.0, PostgreSQL Global Development Group) and R version 3.6.2(R Core Team, R Foundation for Statistical Computing).Data curation codes for the eICU-CRD are included in e-Appendix 2. Given that the two databases differed in terms of their respective time periods covered, populations, data quality, and size, direct comparisons between them were not made.

Factors Associated With Adherence to LTVV
We found that taller patients, following adjustment for sex, age, patient type (medical or surgical), APACHE score, and treating hospital, had a higher likelihood of receiving LTVV in both databases (Fig 2A , Table 2).A 10 cm height increment corresponded to an increased probability of LTVV in both the CCHIC (OR, 1.08; 95% CI, 1.07-1.09)and eICU-CRD (OR, 1.08; 95% CI, 1.08-1.09)databases (Table 2) and a reduction in median V T per ventilation episode by 0.27 mL/kg PBW.
A primary admission diagnosis that was medical in nature was associated with an increased likelihood of receiving LTVV, even when adjustments for an interaction between height and sex were considered: OR for LTVV 1.18 (95% CI, 1.02-1.36,P ¼ .02) in CCHIC and 1.3 (95% CI, 1.13-1.49,P < .0001)for eICU-CRD (Table 2).
There was a general failure to adjust V T in response to deterioration in oxygenation status (Figs 3C, 3D).Within a 6 h window following each patient's worst recorded PaO 2 :FIO 2 value, relative proportions of V T > 8 mL/kg PBW decreased by 14.9% in the CCHIC database and slightly increased by 0.4% in the eICU-CRD database.

LTVV Adherence at Different Times During Patients' Ventilation Episodes
Little variation in V T was observed over the course of individual ventilation episodes.In the CCHIC database at 24 h, there was a 12.9% decrease in the proportion of V T that was consistent with LTVV compared with the initial 6 h period.However, in the eICU-CRD, there was a slight increase in the adoption of LTVV at 24 h (0.8%), compared with the first 6 h (Fig 3D (Table 4).This association did not cross the threshold for significance in the eICU-CRD (SHR, 0.9; 95% CI, 0.81-  5).

Temporal Trends Show a Reduction in V T
There was an easily discernible trend in the CCHIC database, with a significant reduction in median V T for both male (-0.

Initial V T Values and Strategies to Improve Practice
The median initial V T (during the first 6 h of ventilation) for male patients was 492 mL and 545 mL in the CCHIC and eICU-CRD databases, respectively.For female patients, these values were 426 mL and 464 mL (CCHIC/ eICU-CRD).This translated to > 42%/75% of female patients in the CCHIC/eICU-CRD databases receiving V T > 8 mL/kg PBW from initiation of invasive mechanical ventilation.We calculated that to ensure > 90% of people received LTVV, initial V T values of 370 mL for female and 494 mL for male patients should be used (   3]).Male patients were more likely to receive LTVV (unadjusted model) but not if their height was considered.Taller patients were consistently associated with receiving LTVV across all models regardless of their sex with consistent effect sizes.B, Interaction between height and sex on the log odds of receiving LTVV for each database.If these lines were parallel, this would have suggested that there was no interaction between height and sex.The lines happen to intersect each other, and the log odds equals zero line (dashed line, corresponding to an OR of 1) at 160 cm and 165 cm in the CCHIC/eICU-CRD databases, respectively.These values can be interpreted as the threshold height at which both female and male patients were likely to receive LTVV.Ventilated female patients shorter than these heights were less likely to receive LTVV than male patients of the same height.CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; LTVV ¼ low tidal volume ventilation.
chestjournal.orgReference groups were "White/White British" (CCHIC) or "Caucasian" (eICU-CRD).We have presumed an age of 90 years in these patients and used the median value to describe the population average.c In the United Kingdom, "Asian" ethnicity refers to those who identify as originating from South Asia/the Indian subcontinent.
d Missing data are due to differences in how patient features were coded in each of the two databases.
e Percentages for comorbidities apply to the number of patients within each organ system category, not the number of patients in total.Many patients had multiple comorbid conditions.Respiratory Distress Syndrome Clinical Trial Network.A, Female patients who were mechanically ventilated consistently received higher tidal volume, on average, than male patients who were mechanically ventilated.B, This was more apparent in female patients with higher BMIs and was observed in both databases.C, There appeared to be no change in tidal volume, in either database, for patients with different grades of acute hypoxemic respiratory failure.These grades were determined based on consistent PaO 2 :FIO 2 ratios in a 12 h window, while receiving FIO 2 $ 40% and $ 5 cm H 2 O positive endexpiratory pressure.D, There was little variation in tidal volume for patients, at 24 h following admission or when patients experienced their worst period of acute hypoxemic respiratory failure following this 24 h period.CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; PBW ¼ predicted body weight.

and non-ARDS (Practice of Ventilation in Critically Ill
Patients Without ARDS at Onset of Ventilation [ProVENT], 7.9 mL/kg PBW) 17 populations.The proportion of patients who received LTVV was higher in eICU-CRD (43.5%) than in CCHIC (33.4%).Both values were consistent with the results from pooled analyses of non-ARDS patients in high-income countries (44.5%) 18 and patients in the United Kingdom with ARDS (35.9%-38.5%). 19A similar discrepancy in the implementation of LTVV between European and non-European high-income countries has previously been described by Laffey et al. 20 Another strength of the current study was the opportunity to observe the changes in longitudinal trends in V T in a multicenter setting.The fall in V T over time ( Fig 5) shows that improvements in LTVV implementation are being adopted across institutions.[23] We noted a paucity of variation in V T when comparing the initial values (first 6 h) vs the 24 h and 48 h V T , as well as during periods of the most severe hypoxemia within each ventilation episode.This result underscores the importance of a precise initial approach to mechanical ventilation, as we found poor practices tended to persist.Higher initial V T has previously been associated with significantly increased mortality. 24e initiation of mechanical ventilation and admission to an ICU are often emergency situations, frequently occurring outside of normal working hours.Accurate height measurements, necessary for V T titration, may not be readily available.To help address this issue, we calculated starting values in milliliters for patients in each study population.This may help mitigate the risk of exposing patients to harmful V T in the absence of precise height data and help overcome other obstacles to implementation of LTVV. 25,26Interventions to enhance adherence to LTVV have been shown to be cost-effective. 8ur proposition offers a simple step that may help to highlight the importance of regular evaluation of V T .
Previous studies have shown disparities in LTVV implementation based on sex, with female patients less likely to receive LTVV.This has been associated with increased mortality in patients with ARDS. 27e did not observe a difference in mortality based on sex in either database (e-Table 1).This was consistent with the findings in other non-ARDS populations in whom similar differences in V T between sexes have been described.We found that patients receiving LTVV were more likely to have a greater duration of ventilation.We hypothesized that these patients were those undergoing prolonged weaning (eg, with delayed respiratory or neurologic recovery) who may have self-regulated their V T to their normal physiologic range of 6 to 8 mL/kg PBW or who may have had poor pulmonary compliance that required a prolonged period of lung-protective ventilation.The latter case could be an example of reverse   chestjournal.orgcausality that we could not address due to anonymization.
There are several important limitations to our findings that should be considered.Our study was a post hoc analysis, and due to data limitations, we were unable to accurately abstract ventilation modes to determine whether patients received controlled or spontaneous ventilatory modes.We were also unable to evaluate the effects of driving pressure or mechanical power on patient outcome because the required variables were not available in the databases.We chose not to impute missing data, as attempts to do so induced overrepresentation of high V T in the imputed values.
The anonymization process meant that the absence of clinical notes prevented us from accounting for other reasons that might cause a high V T to be recorded (eg, a drained pneumothorax).
Illness severity was reported using APACHE scores and were recorded reliably in both databases.However, APACHE scores do not evaluate organ dysfunction beyond the initial 24 h period.Because of missing data across each organ system domain, we could not reliably extract other illness severity scores (eg, Sequential Organ Failure Assessment) that may have captured peak illness severity more accurately.

Interpretation
LTVV was poorly implemented across multiple ICUs in the United States and the United Kingdom.This was consistent with a failure to account for patient height when setting V T values.This oversight was the principal contributory factor to female patients and those with higher BMI being exposed to higher V T .In the UK cohort, exposure to V T > 8 mL/kg PBW was associated with increased risk of 30-day mortality.

Figure 2 -
Figure 2 -A, B, Analysis of factors predicting whether patients received LTVV.A, The forest plot shows the odds of patient receiving LTVV based on sex or height as univariate analysis or if adjusted for other covariates: age, ethnicity, Acute Physiology and Chronic Health Evaluation score, admission type, and hospital location as a random effect (other effect sizes not shown for clarity [Table3]).Male patients were more likely to receive LTVV (unadjusted model) but not if their height was considered.Taller patients were consistently associated with receiving LTVV across all models regardless of their sex with consistent effect sizes.B, Interaction between height and sex on the log odds of receiving LTVV for each database.If these lines were parallel, this would have suggested that there was no interaction between height and sex.The lines happen to intersect each other, and the log odds equals zero line (dashed line, corresponding to an OR of 1) at 160 cm and 165 cm in the CCHIC/eICU-CRD databases, respectively.These values can be interpreted as the threshold height at which both female and male patients were likely to receive LTVV.Ventilated female patients shorter than these heights were less likely to receive LTVV than male patients of the same height.CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; LTVV ¼ low tidal volume ventilation.

Figure 3 -
Figure 3 -A-D, Analysis of tidal volume management in different patient groups.The median tidal volume for each patient's ventilation period was calculated and converted to milliliters per kilogram by using the PBW formula of the National Heart, Lung, and Blood Institute-funded Acute

Figure 6 -
Figure6-Cumulative density plot showing the percentage of patients who would be in receipt of tidal volume > 8 mL/kg PBW based on the height and sex of patients in each database.To reduce the risk of patients receiving > 8 mL/kg PBW to < 10% in the absence of the patient's height, the initial tidal volume should be set to < 370 mL for female patients and < 494 mL for male patients.CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; PBW ¼ predicted body weight. ).
Figure1-Study flow diagram outlining data cleaning steps for each database and the numbers included in the analysis.a Values in these boxes refer to patient episodes, not unique patients.b Some patients were identified as having periods of both invasive ventilation and NIV.These patients were excluded to avoid misattributing measured volumes during NIV periods.NIV ¼ noninvasive ventilation; PEEP ¼ positive end-expiratory pressure; QC ¼ quality control.

TABLE 1 ]
Characteristics of Ventilated Patients Within Each Database

TABLE 2 ]
Factors Associated With Administration of Low Tidal Volume Ventilation Effect sizes were estimated by using a mixed effects logistic regression model.APACHE ¼ Acute Physiology and Chronic Health Evaluation; CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; ICC ¼ inter-class correlation (a measure of the proportion of the variance explained by the grouping structure, which in this case is the treating hospital).aValues< 0.0001 have been abbreviated to 0.00 for display in the table.bAPACHEII score was used for CCHIC, APACHE IV for eICU-CRD.c

TABLE 1 ]
(Continued) APACHE ¼ Acute Physiology and Chronic Health Evaluation; CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ eICU collaborative research database; IQR ¼ interquartile range; NA ¼ not available/missing data; OR ¼ operating room; PACU ¼ postanesthesia care unit.89 years are considered potentially identifying data by the Health Insurance Portability and Accountability Act of 1996 and thus are coded ">89." a In the United Kingdom, teaching hospital refers to an academic health center.b Ages >

TABLE 3 ]
Tidal Volumes During Invasive Mechanical Ventilation Data are presented as mean AE SD unless indicated otherwise.All tidal volumes are in milliliters per kilogram (PBW), which was calculated by using the National Heart, Lung, and Blood Institute-funded Acute Respiratory Distress Syndrome Clinical Trial Network formula.AHRF ¼ acute hypoxemic respiratory failure (PaO 2 :FIO 2 < 300 mmHg on FiO 2 $ 0.4 with $5 cmH 2 O positive end-expiratory pressure); CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; PBW ¼ predicted body weight.

TABLE 4 ]
Effect of LTVV on Outcomes of Patients Mechanically Ventilated for > 48 h From Both DatabasesOutcomes at day 30 were assessed by using competing outcomes and expressed as SHRs.For example, with respect to the outcome of being extubated by day 30, death was used as a competing hazard.Although LTVV was associated with a lower cause-specific hazard of death by day 30, patients managed in this way were not more likely to be extubated.APACHE ¼ Acute Physiology and Chronic Health Evaluation; CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; LTVV ¼ low tidal volume ventilation; PBW ¼ predicted body weight; SHR ¼ cause-specific hazard

mL/kg PBW Median VT < 8 mL/kg PBW Median VT > 8 mL/kg PBW
Cumulative incidence plots for patients with a median V T above or below 8 mL/kg PBW showing the relative incidence of each competing outcome (death, still ventilated, extubated) at day 30 after starting ventilation.The cause-specific hazard ratio for death at 30 days was lower in patients receiving LTVV in the CCHIC database (SHR, 0.86; 95% CI 0.76-0.97;P < .001)but not in the eICU-CRD database (SHR, 0.9 [95% CI, 0.81-1.01;P ¼ .06]).LTVV was, however, associated with a lower probability of extubation by day 30 in both databases (CCHIC SHR, 0.69 [95% CI, 0.64-0.75,P < .001];eICU-CRD SHR, 0.88 [95% CI, 0.83-0.92,P < .001]).CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; PBW ¼ predicted body weight; SHR ¼ cause-specific hazard ratio; V T ¼ tidal volume. 28 Trends (locally estimated scatterplot smoothing) in administered tidal volume over time, for both male and female patients in the Critical Care Health Informatics Collaborative database.The median tidal volume for each patient's ventilation period was calculated and aggregated with all the other mechanically ventilated patients that month using a weighted mean.A locally estimated scatterplot smoothing curve was fitted to the time series to show changes in practices over time with bandwidth that included 65% of local points.Following December 2016, there was a consistent and significant decrease in median tidal volume per ventilation episode for both male patients (-0.19 mL/kg PBW per quarter; 95% CI, 0.08-0.29)and female patients (-0.26 mL/kg PBW per quarter; 95% CI, 0.07-0.46).PBW ¼ predicted body weight.

TABLE 5 ]
Association Between LTVV and Duration of Ventilation in Patients Ventilated for > 48 h Length of stay data are modeled using the log-linked Poisson distribution.APACHE ¼ Acute Physiology and Chronic Health Evaluation; CCHIC ¼ Critical Care Health Informatics Collaborative; eICU-CRD ¼ electronic ICU collaborative research database; LTVV ¼ low tidal volume ventilation.
a Values < 0.0001 have been abbreviated to 0.00 for display in the table.b Reference groups were "White/White British" (CCHIC) or "Caucasian" (eICU-CRD) c APACHE-II score was used for CCHIC, APACHE-IV for eICU-CRD.