Pain Assessment Is Associated with Decreased Duration of Mechanical Ventilation in the Intensive Care Unit: A Post HocAnalysis of the DOLOREA Study

The prospective, multicenter, observational DOLOREA study was conducted from January 2004 until January 2005, in 43 ICUs in France and 1 ICU in Luxembourg.14Patients 15 yr or older were enrolled in the study if admitted to the ICU for a foreseeable duration of MV of more than 24 h. Patients were excluded if they had severe brain injury on admission (defined by a Glasgow Coma Scale score of less than 9), or if MV was delayed for more than 24 h after admission to ICU. For the purpose of this post hoc  analysis, we excluded from the DOLOREA database those patients who had not received analgesia on D2 of their ICU stay.

A detailed description of data collection and quality control procedures is available elsewhere.14Briefly, each site had a dedicated individual who entered raw data into an electronic case report form (ClinInfo S.A., Lyon, France). For each patient, a set of variables was collected that included demographic characteristics, illness severity on admission as defined by the Simplified Acute Physiology Score II, and an individual Sequential Organ Failure Assessment score of 3 or 4 (i.e. , moderate-to-severe organ failure).19For each patient, we also recorded the instrument used to assess sedation and pain levels on D2 of ICU stay, the type of sedative and opioid drug used on D2, the cumulative amounts of these drugs during the previous 24 h, the use of nonopioids and neuromuscular blocking agents on D2, and the management of procedural pain on D2. In patients who were assessed using sedation scales, we defined a deep state of sedation by a Ramsay score of 5 or 6, a Richmond Agitation Sedation Scale score of −5 or −4, or a Sedation Agitation Scale score of 1 or 2. Information about the recruiting sites was collected regarding their resources, the existence of pain control protocols/guidelines, and whether they provided dedicated education for pain and sedation management.

Patients were followed up until death, until ICU discharge, or for 30 days in the ICU. The primary study outcomes included MV duration and duration of ICU stay for survivors. Patients were considered as candidates for weaning from the ventilator if they no longer had a high-grade fever, hemodynamic instability, or severely altered consciousness or if they exhibited adequate oxygenation with an inspired oxygen fraction less than 0.5 and positive end-expiratory pressure less than 5 cm H²O. Candidates for weaning were switched to pressure support ventilation followed by daily spontaneous breathing trials on a T-piece. The decision to extubate was based on simple bedside tolerance variables, including respiratory rate, arterial oxygen saturation, and the use of accessory respiratory muscles during T-piece trials. This protocol was applied uniformly across all sites.

The secondary study outcomes included mortality rates and, among survivors, the incidence of acquired complications: ventilator-associated pneumonia, gastroduodenal hemorrhage, venous thromboembolism, and colonization of central venous catheters. Ventilator-associated pneumonia was defined by a new parenchymal opacity in the lung on chest radiograph plus at least two of the following three criteria: (1) temperature less than 36°C or greater than 38°C, (2) leukocyte count less than 4,000/ml or greater than 10,000/ml, (3) purulent secretions from the endotracheal tube.20Gastroduodenal hemorrhage was defined by esophagogastroduodenoscopy, or by the combination of grossly visible blood from an enterally placed tube and subsequent transfusions of 2 or more units of packed erythrocytes. Thromboembolic events were defined by the presence of a venous thrombosis proven by Doppler ultrasonography or venography, or by the presence of a pulmonary embolism proven by pulmonary angiography, or contrast spiral computed tomography of the thorax. Central venous catheter colonization was defined by the isolation of at least one organism at a concentration of 10³or more colony-forming units/ml from a catheter tip culture.21 

Clinical practices regarding sedation and analgesia management were also compared between the two groups of patients, i.e. , patients assessed for pain versus  patients not assessed for pain on day 6 (D6) of their ICU stay. Patients were included in the D6 analysis if they had been mechanically ventilated on both D2 and D6, they received analgesia on both D2 and D6, and their allocated group on D2 was the same on D6.

Descriptive statistics included frequencies and percentages for categorical variables, and medians and interquartile range, i.e. , 25th and 75th percentiles, for continuous variables. Baseline characteristics were compared using chi-square tests for categorical variables and the nonparametric Kruskal–Wallis test for continuous variables. ICU mortality rates, duration of MV, and duration of ICU stay were compared using discrete time survival logistic hazard models. All observations were censored at 30 days. The duration of MV and the duration of ICU stay were compared among survivors.

Because some patients might have experienced a primary study outcome before pain assessment (i.e. , death, weaning from MV, or discharge from the ICU), we used a landmark analysis with prespecified landmarks on D2 of the ICU stay, as before.22This landmark analysis minimized the impact of survivor bias by comparing study outcomes of patients with no primary study outcomes on D2.

As individual physician discretion directed pain assessment, unadjusted comparisons of outcomes between patients with and without pain assessment might be confounded by imbalances in baseline characteristics. To address this issue, we performed a propensity score analysis.23Conceptually, the propensity score corresponds to the conditional probability of exposure to a treatment given the observed characteristics of a patient. Stratifying on the propensity score tends to balance all observed characteristics that are used to construct the score and, in this way, approximates the conditions of random treatment.23In practice, we derived a propensity score for pain assessment on D2 using a full, nonparsimonious logistic regression model that included patient baseline characteristics (age, sex, weight, Simplified Acute Physiology Score II, admission source, Sequential Organ Failure Assessment score 3 or 4 on admission, chronic heart failure, chronic respiratory failure, active cancer, diabetes, regular psychoactive drug use, cirrhosis, chronic renal failure, stroke), ICU characteristics (university affiliated, number of beds ≥ 12, ICU nurse-to-bed ratio > 4, dedicated education, protocol use), treatment with analgesics (morphine, sufentanil, fentanyl, remifentanil, other opioids, paracetamol, nefopam, ketamine, other nonopioids, procedural pain treatment), treatment with sedatives (midazolam, propofol, other sedatives), and treatment with neuromuscular blocking agents. We used a linear spline model to adjust for the confounding effect of age with knots at 40 and 70 yr, and Simplified Acute Physiology Score II with knots at 50 and 75 yr, respectively. The model yielded a c  statistic of 0.87, indicating a good ability to differentiate between patients with and without pain assessment. Each patient was assigned a propensity score, which ranged from 0.01 to 0.99 and which reflected the conditional probability of pain assessment on D2, given his baseline characteristics. We then stratified patients by quintiles of increasing propensity score. To validate our propensity score adjustment, we checked for the absence of significant residual imbalances in baseline characteristics after adjusting for quintile of propensity score and for adequate overlap of propensity score between the two groups within each quintile. We then estimated the odds ratios of study outcomes associated with pain assessment on D2 after adjusting for the quintile of propensity score. All P  values were two-tailed, and P ≤ 0.05 was considered statistically significant. Analyses were performed using Stata version 9.0 (StataCorp, College Station, TX).

Of the 1,381 patients in the DOLOREA database, 1,144 mechanically ventilated patients satisfied the inclusion criteria for the 48-h landmark analysis. This included patients who were assessed for pain on D2 (n = 513 patients, 45% of the population) and patients who were not assessed for pain on D2 (n = 631 patients, 55% of the population) (fig. 1). Patients had been assessed for pain using the following instruments: the behavioral pain scale24(451/513, 49% of patients), the Harris scale25(98/513, 19% of patients), the visual analog scale (71/513, 14% of patients), the verbal descriptor scale (64/513, 12% of patients), and the numeric rating scale (24/513, 5% of patients).

Table 1shows baseline characteristics of the patients. No significant differences were found between the two groups regarding their characteristics, with the exception of age higher than 75 yr, admission source, chronic heart failure, and use of psychoactive drugs. The two groups had comparable illness severity on admission. Patients with pain assessments were more likely to be admitted to university-affiliated sites, to sites with more resources, and to sites with more protocols and dedicated pain education. There was 24-h in-house intensivist coverage for all participating sites.

Although the proportion of patients receiving continuous opioids was comparable between the two groups, patients with pain assessments were more likely to receive fentanyl, higher dosages of sufentanil, and lower dosages of remifentanil during the previous 24 h (table 2). More patients with pain assessments were treated with nonopioids, such as paracetamol and nefopam, than those without, and they received more often multimodal analgesia in the ICU, i.e. , an association of opioids and nonopioids. They were also more likely to have dedicated pain treatment during procedural pain events, such as endotracheal suctioning and mobilization during standard care.

We also noticed markedly different sedation management between the two groups. Patients with pain assessments on D2 were less likely to receive hypnotic drugs, midazolam in particular, and they received lower daily doses of midazolam (table 3). Of the 928 patients who received hypnotic drugs on D2, those with pain assessments were more likely to be assessed for sedation as well (91% vs.  30%; P < 0.01). Sites used the Ramsay scale, the Richmond Agitation–Sedation Scale, the Sedation–Agitation Scale, and other instruments to assess sedation. In patients who were assessed for sedation, the proportion of patients in a deep sedative state was similar between the two groups. Finally, the use of neuromuscular blocking agents was significantly reduced in patients with pain assessment (table 3).

No significant difference in mortality was found between the two groups of patients, allowing comparisons of MV duration and duration of stay among survivors (table 4). In univariate analysis, patients assessed for pain on D2 had a shorter duration of MV (8 vs.  11 days; P < 0.01) and duration of ICU stay (13 vs.  18 days; P < 0.01). While adjusting for quintile of propensity score to rule out all differences in baseline characteristics between the two study groups (data not shown), pain assessment was associated with increased odds of weaning from the ventilator (odds ratio, 1.40; 95% confidence interval, 1.00–1.98) and discharge from the ICU (odds ratio, 1.43; 95% confidence interval, 1.02–2.00). Because MV duration and duration of ICU stay were determined for survivors, the comparisons of ICU- and MV-free days led to the same conclusions between the two study groups (data not shown). A correct overlap of the propensity scores was observed between the two study groups within each quintile (table 5).

Of the 1,144 patients included in this study, 653 patients were still mechanically ventilated and receiving analgesia on D6. Of these, we excluded 114 because of the absence of MV on D2 (n = 23 patients), the absence of analgesia use on D2 (n = 7 patients), and a crossover between the 2 groups of patients between D2 and D6 (n = 84 patients). No significant differences were found in baseline characteristics for those 84 crossover patients and the 539 analyzed patients (data not shown). The group of patients with pain assessments on D6 included 229 patients (42% of the population), and the group with no pain assessments on D6 included 310 patients (58% of the population). Most of the differences between the two groups of patients regarding pain and sedation management on D2 agreed with those found on D6 (table 6). Patients with pain assessment on D6 were more likely to be assessed for sedation and for procedural pain. Also, they received fewer sedatives and lower daily doses of midazolam compared with patients not assessed for pain. In patients assessed for sedation, the proportion of patients in a deep sedative state was similar between the two groups.

In this cohort study of mechanically ventilated ICU patients receiving analgesia, pain assessment on D2 of the ICU stay was associated with marked differences in sedation and analgesia management. Such patients had more frequent sedation level evaluations, fewer hypnotics and neuromuscular blocking agents, lower daily midazolam doses, more nonopioids, and more care for procedural pain than those whose pain was not assessed. Most of these differences still persisted on D6 of the ICU stay. After multiple adjustments for severity factors, pain and sedation medications, and ICU characteristics, these data reveal the use of pain assessments as an independent factor in reducing the MV duration and the duration of ICU stay. Although association does not prove a causal relation, the consistency of the results regarding the types of adjustments and the known effects of sedative use on MV duration suggest a link between pain assessment and patient outcome. Specifically, pain assessment might result in more attention toward pain and sedation management, in a multimodal approach, that in turn might reduce the duration of MV and duration of ICU stay. These findings, drawn from a large database of patient-based current practices, strongly argue in favor of the routine use of dedicated instruments to assess both pain and sedation in mechanically ventilated patients.

Duration of MV and duration of stay in the ICU have become reliable markers for determining the effects of sedatives and analgesics in mechanically ventilated patients.16–18Several randomized controlled trials have used these criteria to determine the efficacy of various strategies in optimizing sedation and analgesia: protocol-directed sedation according to consciousness levels,26,27sedation-based analgesia,28,29daily interruption of sedation,30–32combined sedation and ventilator weaning protocol,33,34and spontaneous breathing trial during ventilatory support.35,36Researchers have also conducted before-and-after studies to determine the impact of implementing protocols on the duration of MV and duration of ICU stay in single centers.15,37–40Common to all of these studies is the repeated measurement of the level of conscious levels—and possibly pain—to achieve the desired target. In routine clinical practice, pain and sedation measurement rates remain unsatisfactory.10–14For this reason, we took a different approach to investigating a possible link between clinical practices—assessing pain in a cohort of mechanically ventilated patients—and changes in patient outcome. Despite the well-known limitations of cohort studies compared with randomized controlled trials, cohort studies do provide useful information. In particular, they help to determine the benefits and dangers of medical interventions, provided they respect the guidelines designed to strengthen the quality of reporting in observational studies.41 

In the ICU, pain can result from several sources, including surgical incisions, traumatic injuries, occult infections, immobility, and ICU procedures.1,42Forty-five percent of patients reported moderate to severe pain (visual analog scale > 50 mm) during removal of the endotracheal tube.43Pain assessment in verbal patients is straightforward, whereas assessing pain in patients who are unable to communicate presents a great challenge. In the DOLOREA study, self-rating scales increased in use on subsequent days, as the level of hypnotic use decreased and more patients became able to communicate.14This resulted in no pain assessment for most nonverbal ICU patients. However, clinical instruments have been developed to address this issue, including the behavioral pain scale24and the Critical Care Pain Observation Tool,44both of which show good validity and reliability.9By using appropriate instruments to systematically assess pain in nonverbal patients, 50% of patients reported moderate to severe pain.15Our cohort of mechanically ventilated ICU patients, without severe head injuries, was mostly nonverbal, as 80% received sedatives on D2. We therefore hypothesized that measuring pain in those patients would imply the concomitant use of an instrument for measuring sedation, resulting in possible changes in the drug administration.

In this study, patients assessed for pain were less likely to be treated with hypnotic drugs, and they received 30% smaller daily doses of midazolam in comparison with the patients not assessed for pain. There were marginal differences between the two groups regarding their medical conditions. In addition, daily sedative interruptions were not performed. Although being one means to achieve reduced hypnotic doses, daily sedative interruption is not suitable for all mechanically ventilated patients.32Therefore, these differences between the two groups of patients could result from higher sedation assessment rates leading to more restricted sedative use. Previous studies found a 30–50% reduction in the mean daily and cumulative dosages of midazolam after the implementation of an algorithm. Staff repeated pain and sedation measurements every 3 h in the intervention phase to target the sedation level and adjust drug doses accordingly.38,39When staff systematically assessed pain and agitation at rest and 30 min after any procedure, the duration of continuous drug infusions was reduced by 30%.15 

In our cohort of patients, pain assessment was also associated with better procedural pain management and with significant changes in the use of other drugs on D2 (nonopioids, neuromuscular blocking agents), in line with current recommendations.45–48Because more than 90% of our patients received opioids in the two groups, it could be postulated that nonopioids were used more often because pain assessment was provided in the group “Yes.” Some evidence indicates that nefopam, a centrally acting nonopioid analgesic that inhibits serotonin and norepinephrine reuptake, may reduce postoperative morphine consumption and morphine-related side effects in surgical patients.49Although no impact can be inferred from 12% of patients treated with nefopam, this suggests that benefits could be expected from combining the use of nonopioids with opioids in the ICU setting. Our results also indicate that implementing ICU protocols and increasing education about pain and sedation could significantly help caregivers to follow national guidelines.

We found that patients assessed for pain required fewer hypnotics, lower midazolam doses, shorter MV durations by 3 days, and shorter ICU stays by 5 days. The relation between continuous hypnotic infusions and MV duration has been explored by daily sedative interruptions, which resulted in lower daily midazolam doses and, concomitantly, reduced MV durations.30Using the concept of sedative interruption, the use of propofol instead of lorazepam (a long-acting benzodiazepine) further reduced MV duration.50Both the dose and the type of daily hypnotics can affect MV duration, a statement supported by our findings. The comparable proportion of patients in a deep sedative state (47% vs.  49% on D2) indicates that, for those patients assessed for sedation, both groups had similar sedation requirements, thus excluding a potential confounding factor affecting MV duration. This assertion is somewhat hampered by the low proportion of patients assessed for sedation in the group with no pain assessment (30%). Interestingly, few studies have investigated whether the liberal use of opioids has a similar effect on patient outcome. In the study by Kress et al. ,30the shorter MV duration in the intervention group resulted from lower doses of either morphine or midazolam. De Jonghe et al.  38reported no changes in either the daily fentanyl doses or in the duration of fentanyl administration during the intervention phase, which resulted in 5 fewer days of MV. In our study, we found marginal differences in opioid use between the two groups, suggesting that those agents may have a less important role per se  in the duration of MV than hypnotic drugs. The high concomitant rates of sedation assessments as well as the optimal use of analgesics at rest and during procedures should surely prevent the inappropriate use of hypnotics to treat agitation and other sources of patient discomfort. With a restricted use of hypnotics in sedation-based analgesia, results showed differences in MV duration from opioids with different pharmacologic profiles.28,29 

The current analysis has several limitations, in addition to those of the DOLOREA study that are discussed elsewhere.14First, although we made every effort to adjust for illness severity, treatments, and ICU characteristics, we could not randomly assign pain assessment. We used propensity-adjusted analysis to eliminate imbalances between the groups of patients for measured characteristics of both patients and sites. Apart from the highest quintile, the overlap of propensity scores between the two groups was satisfactory, indicating no imbalances in baseline characteristics. Interestingly, no significant first-order interaction was found (table 5), suggesting that the relation between pain assessment and the shorter duration of MV and duration of ICU stay was homogeneous across the quintiles of propensity scores. However, we could not exclude unrecorded confounding factors that may have influenced patient outcome, as recently pointed out.51Our finding that differences in pain and sedation management between the two groups still persisted on D6 gives some reliability to the results found on D2. Second, the odds ratios and their 95% confidence limits for MV duration and duration of ICU stay were close to one in the propensity score-adjusted analysis. Therefore, the data may not be compelling enough to strongly recommend pain assessment as a way of improving ICU patient outcomes. However, an adequately powered randomized controlled trial exploring the impact of pain assessment in the ICU on patient outcomes is ethically impossible to conduct, because pain and sedation assessments are strongly recommended in national guidelines.45Third, although we did study individual complications as a matter of routine, we cannot exclude possible interrelations between multiple complications within individuals. Fourth, we did not study the pharmacokinetic interactions between hypnotics and opioids or the pharmacogenetics of opioids that may have influenced the sedative/analgesic drug effects.52In particular, there were more elderly patients with no pain assessment, which could explain, in part, findings on MV duration due to their sensitivity to drugs.53However, the impact of age was balanced between the two groups with the use of a linear spline model in our adjustments.

In conclusion, our analysis shows an association between assessing pain in mechanically ventilated ICU patients and changes in clinical practice. In particular, staff assessed sedation more frequently and reduced hypnotic use. After adjustments, these changed clinical practices were associated with a shorter duration of MV and a reduced duration of stay in the ICU. Therefore, pain assessment in nonverbal ICU patients, although difficult, must be promoted together with sedation measurements in a multimodal approach, to avoid inappropriate use of hypnotic drugs to treat or mask pain.

The authors thank Kathleen Puntillo, R.N., C.N.S., D.N.Sc. (Professor, Department of Physiological Nursing, University of California, San Francisco, California), for her helpful comments on the manuscript; and Céline Genty, B.Sc. (Statistician), Carole Rolland, B.Sc. (Technician), and Marilyne Blanc, B.Sc. (Technician), Clinical Research Center INSERM 003, Albert Michallon Hospital, Grenoble, France, for their help in managing the data.

Joachim Calderon, M.D. (Hôpital Cardiologique, Bordeaux, France), Laurent Beydon, M.D. (Centre Hospitalier Universitaire, Angers, France), Charles Cerf, M.D. (Hôpital Henri Mondor, Créteil, France), Philippe Barbe, M.D. (Centre Hospitalier, Chambéry, France), Stéphane Winnock, M.D. (Centre Hospitalier Universitaire, Bordeaux, France), Jean Hébert, M.D. (Centre Hospitalier Universitaire, Caen, France), Moldi Hamrouni, M.D. (Hôpital Louis Pasteur, Chartres, France), Dominique Guelon, M.D. (Hôpital Gabriel Montpied, Clermont-Ferrand, France), Stéphanie Artigues, M.D. (Hôtel Dieu, Clermont-Ferrand, France), Irène Messant, M.D. (Hôpital Général, Dijon, France), Elsa Brocas, M.D. (Centre Hospitalier Louise Michel, Evry, France), Pierre Lavagne, M.D. (Hôpital Michallon, Grenoble, France), Christiane Hercule, M.D. (Hôpital Cardiovasculaire et Pneumologique Louis Pradel, Lyon, France), Serge Duperret, M.D. (Hôpital Croix Rousse, Lyon, France), Khalid Berrada, M.D. (Hôtel Dieu, Lyon, France), Jean-Michel Grozel, M.D. (Centre Hospitalier Lyon Sud, Lyon, France), Gérald Chanques, M.D. (Hôpital St Eloi, Montpellier, France), Sarah Valette, M.D. (Hôpital Lapeyronie, Montpellier, France), Francine Bonnet, M.D., Laurent Benayoun, M.D. (Hôpital Beaujon, Clichy, France), Adrien Decorps-Declère, M.D. (Hôpital Antoine Beclère, Clamart, France), Jacques Duranteau, M.D. (Hôpital Bicêtre, Le Kremlin-Bicêtre, France), Jean Mantz, M.D., Hervé Quintard, M.D. (Hôpital Bichat Claude Bernard, Paris, France), Jean-Francois Timsit, M.D., Bruno Mourvilliers, M.D. (Hôpital Bichat Claude Bernard, Paris, France), Christian Rathat, M.D. (Hôpital Avicenne, Bobigny, France), Rémy Gauzit, M.D. (Hôpital Jean Verdier, Bondy, France), Andrea Passard, M.D. (Hôpital Ambroise Paré, Boulogne-Billancourt, France), Jean-Paul Perez, M.D. (Hôpital d'Instruction des Armées, Clamart, France), Laurent Dubé, M.D. (Centre Hospitalier Universitaire, Angers, France), Michèle Binhas, M.D. (Hôpital Henri Mondor, Créteil, France), Anne-Claire Lukaszewicz, M.D. (Hôpital Lariboisière, Paris, France), Jean-Luc Leguillou, M.D. (Institut Mutualiste Montsouris, Paris, France), Laurent Jacob, M.D. (Hôpital Saint Louis, Paris, France), Sonia Elmetaoua, M.D. (Hôpital Tenon, Paris, France), Jean-Paul Bleichner, M.D. (Hôpital Ponchaillou, Rennes, France), Igor Auriant, M.D., Gaëlle Demeilliers-Pfister, M.D. (Hôpital Charles Nicolle, Rouen, France), Philippe Mahul, M.D. (Hôpital Nord, St Etienne, France), Guy Freys, M.D. (Hôpital de Hautepierre, Strasbourg, France), Pascale Sanchez, M.D. (Hôpital Purpan, Toulouse, France), Francois Lagarrigue, M.D. (Hôpital Trousseau, Tours, France), Bruno Raynard, M.D. (Institut Gustave Roussy, Villejuif, France), Emmanuelle Nalet, M.D. (Centre Hospitalier de la Région Annecienne, Annecy, France), Marc Feissel, M.D. (Centre Hospitalier Général, Belfort, France), Jérôme Baudot, M.D. (Centre Hospitalier Henri Duffaut, Avignon, France), Marc Klop, M.D. (Clinique Sainte Thérèse, Luxembourg).