Abstract

MATERIALS AND METHODS

RESULTS

Enrollment

A total of 4195 patients had surgery and an ICU admission between 2012 and 2017. After exclusions outlined in Figure ​Figure11, the final cohort consisted of 1865 patients.

Open in a separate window

Figure 1.

Flow diagram of study enrollment. MGI = Michigan Genomics Initiative.

Demographic and Clinical Characteristics

Mean (sd) age was 56 years (sd, 15 yr), 910 (48.8%) were female, 92.4% were White, 52.3% were ASA class III, and 17 (0.01%) had emergency surgery. Preoperatively, 487 (28.6%) patients were taking prescribed opioids. Participants reported an average of 3.1 (sd, 3.0) of ten on the Brief Pain Inventory Scale, 48.7% had at least moderate anxiety, and 25.9% had at least moderate depression (Table ​Table11).

TABLE 1.

Descriptive and Univariable Analysis of Sociodemographic and Baseline Characteristics and Surgical Factors

Variable (Categorical or Continuous)	Frequency (%) or Mean (sd), n = 1865
Age (yr)a	56 (15)
Sex
Female	910 (48.8)
Male	955 (51.2)
Race
White	1724 (92.4)
African American	84 (4.5)
Other	57 (3.1)
Body mass indexa (kg/m2)	30.1 (7.3)
ASA physical status classification
Class 1	21 (1.3)
Class 2	470 (28.3)
Class 3	870 (52.3)
Class 4	302 (18.2)
Tobacco use
Current	416 (22.6)
Former	678 (36.8)
Never	751 (40.7)
Alcohol use
Never	381 (25.7)
Current or former	1102 (74.3)
Illicit drug use
Never	1604 (92.2)
Current or former	135 (7.8)
Insurance
Private	843 (57.8)
Medicaid/Medicare	616 (42.2)
At least moderate anxiety
No	752 (51.3)
Yes	713 (48.7)
At least moderate depression
No	1081 (74.1)
Yes	378 (25.9)
Life Satisfaction Scalea	6.7 (2.7)
Fibromyalgia Survey Scorea	6.1 (4.8)
Number of chronic pain areasa	3.5 (4.4)
Neuropathic Pain Scalea	6.6 (7.5)
Brief pain inventorya	3.1 (3.0)
Preoperative opioid use
No	1217 (71.4)
Yes	487 (28.6)
Major surgery
No	227 (13.6)
Yes	1441 (86.4)
Anesthesia duration (hr)a	7.4 (3.0)
> 24hr ventilator use in ICU
No	1724 (92.4)
Yes	141 (7.6)

Open in a separate window

^aContinuous variable: mean (sd); remaining variables categorical: frequency (%).

The most common types of surgery were neurosurgery (n = 579, 31.0%), cardiac (n = 343, 18.4%), otolaryngology/oral maxillofacial (n = 194, 10.4%), intra-abdominal (n = 174, 9.3%), spine (n = 163, 8.7%), interventional radiology (n = 70, 3.8%), plastic surgery (n = 62, 3.3%), and thoracic (n = 35, 1.9%). Most surgeries were classified as major (86.4%). Although surgical case listing and surgical service did not capture whether patients presented due to a trauma, very few cases would be expected given that there were only 17 emergency cases in the cohort. The anesthetic lasted on average 7.4 hours (sd, 3.0hr), and 7.6% of patients received greater than 24 hours of mechanical ventilation during the ICU course. Mean ICU length of stay was 2.7 days (sd, 3.8 d). Throughout the ICU stay, average OME use per day was 64.3mg (sd, 106.4mg).

Opioid Utilization During the Postoperative ICU Encounter

Figure ​Figure22 depicts the number of patients who received OME greater than 0 on each ICU-day. Opioid use decreases over the first 4 days postoperatively. The majority of patients receiving opioids in the ICU had undergone major surgery (Fig. ​(Fig.22A). Patients who received preoperative opioids were more likely to continue to receive opioids throughout their ICU stay than opioid-naive patients: on day 1, 29% of patients receiving opioids in the ICU had been taking preoperative opioids. By day 10, 57% of patients still receiving opioids in the ICU had been taking preoperative opioids (Fig. ​(Fig.22B).

Open in a separate window

Figure 2.

Opioid administration over the first 10 d of the ICU course. A, Opioid administration was stratified by major vs. minor surgery. B, Opioid administration was stratified by preoperative opioid use.

Nonopioid Drugs Utilization During the Postoperative ICU Encounter

Figure ​Figure33 depicts nonopioid medications administered over the first 10 days in the ICU. Acetaminophen was given on average to 72% of participants every day, followed by nonsteroidal anti-inflammatory drugs (NSAIDs) (38%), and anticonvulsants including gabapentin/pregabalin (24%). The proportion of patients receiving each drug remained relatively constant over time. Propofol was administered to 27% of patients on the first ICU-day and decreased below 10% on subsequent days.

Open in a separate window

Figure 3.

Number of participants receiving nonopioid medications per ICU-day stratified by medication class. Background bar chart depicts total number of participants present in the ICU who received postoperative opioids. Anticonvulsants include gabapentin, pregabalin, carbamazepine, and lamotrigine. Anxiolytic/antidepressants include selective serotonin reuptake inhibitors, serotonin and norepinephrine reuptake inhibitors, and tricyclic antidepressants. Sedatives include benzodiazepines and diphenhydramine. Muscle relaxants include cyclobenzaprine, baclofen, and tizanidine. Ketamine administration was not depicted because it was administered in less than 1% of participants. NSAID = nonsteroidal anti-inflammatory drug.

Factors Associated With Opioid Use in the ICU

The initial multivariable regression model for backward variable selection consisted of all variables in Table ​Table11 except sex, age, and body mass index, based on screening criteria of p value of less than 0.15 by univariable linear regression. The final model meeting minimum BIC included eight predictors, all of which, except for anxiety, were significantly associated with mean daily OME (Table ​Table22). For their effect on mean daily OME use, the parameter estimates were converted back by power of 10: the effect size for age on OME was 0.98, implying that OME decreased by 2% for every 1-year increase in age. Preoperative opioid use increased OME by a factor of 1.67, implying a 67% increase in OME compared with those who did not use preoperative opioids. Current/former drug use increased OME by 30% compared with those who do not use drugs. Preoperative centralized pain increased OME by 3% for every additional point on the Fibromyalgia Survey Score Questionnaire. Anesthesia duration increased OME by 3% for every additional hour of anesthesia. Receiving mechanical ventilation greater than 24 hours increased OME by 85% compared with those who did not receive mechanical ventilation. Major surgery increased OME by 62% compared with those who underwent minor surgery.

TABLE 2.

Final Multivariable Linear Regression for Mean Daily Oral Morphine Equivalent Use During the ICU Course With Predictors of Sociodemographic, Baseline Characteristics, and Surgical Factors

Variable	Beta Estimatea (95% CI)	p	Effect Sizeb (95% CI)
Age (yr)	–0.007 (–0.009 to –0.005)	< 0.0001c	0.98 (0.98–0.99)
Current or former illicit drug use	0.12 (0.01–0.23)	0.04c	1.30 (1.01–1.68)
Preoperative opioid use	0.22 (0.16–0.29)	< 0.0001c	1.67 (1.43–1.96)
At least moderate anxiety	0.02 (–0.04 to 0.08)	0.57	1.04 (0.91–1.19)
Fibromyalgia Survey Score	0.01 (0.007–0.02)	< 0.0001c	1.03 (1.02–1.05)
Anesthesia duration (hr)	0.01 (0.002–0.02)	0.02c	1.03 (1.00–1.05)
> 24hr ventilator use in ICU	0.27 (0.15–0.38)	< 0.0001c	1.85 (1.41–2.42)
Major surgery	0.21 (0.12–0.30)	< 0.0001c	1.62 (1.33–1.97)

Open in a separate window

^aParameter estimates are for log₁₀ transformed oral morphine equivalent (OME).

^bThe effect size for mean daily OME = 10^β where β is the beta estimate derived for log₁₀(OME). The effect size value was rounded to two decimal points for reporting purpose. For instance, the effect size of age on mean daily OME implies a 2% decrease in OME for every one-year increase in age. The effect size of preoperative opioid use implies a 67% increase in OME compared with those who did not use preoperative opioids.

^cSignificance determined by p < 0.05.

DISCUSSION

In this large prospective cohort study of patients who were admitted to the ICU after surgery and received postoperative opioids, we found that patients who received preoperative opioids were more likely to continue to receive opioids throughout their ICU stay compared with opioid naive patients. The strongest risk factors of opioid use in the ICU were receiving mechanical ventilation (which increased OME requirements by 85% compared with those not mechanically ventilated), receiving preoperative opioids, and undergoing major surgery. In genetic association analyses of three pain candidate genes, we did not find any statistically significant associations with OME use in the ICU. The strengths of our study included the large cohort size and the availability of detailed baseline preoperative pain and mental/physical characteristics. There has been no study to our knowledge looking at predictors of opioid use during the ICU course specifically. Such genetic analysis in the ICU population has also not been done before.

The experience of pain in the ICU is common (7, 9, 31) and often a traumatic experience (7, 32). Although the treatment of pain is important to improve certain outcomes such as reduce ICU length of stay, duration of mechanical ventilation, and sedation requirements (7, 9), exposure to opioids in the ICU is not without risk. Receiving opioids in the ICU is associated with an increased risk of delirium (33), which in turn can increase cognitive decline (34) and mortality (35). The simple exposure to opioids after a routine procedure can lead to increased odds of continued opioid use (36) and 4% of ICU survivors developed opioid dependence after discharge (6). Risk factors for pain and opioid use vary depending on whether they focus on postsurgical pain or pain after critical illness. Our study, which focused on risk factors during the ICU period specifically, found overlapping risk factors with postsurgical pain (e.g., younger age, preexisting pain, preexisting opioid use, illicit drug use, surgery type, and longer surgery duration) (37–42), and pain after critical illness (e.g., duration of mechanical ventilation) (3). Mechanical ventilation having the largest effect size in our study could perhaps be explained by clinicians having a different threshold for prescribing opioids in mechanically ventilated patients, either as a result of clinical practice guidelines recommending an analgesia-first sedation approach (8), a need for ventilatory synchrony, and/or less concern regarding respiratory depression. The significant influence of mechanical ventilation on ICU opioid use can potentially explain why one particular study, which had a larger proportion of mechanically ventilated patients than in our study, did not find preadmission opioid use as a risk factor for ICU opioid use despite preadmission opioids being associated with worse pain after ICU discharge (43). Furthermore, although psychologic risk factors are important in the development of acute and chronic post-surgical pain (37, 38, 41, 42) and long term after an ICU course (5, 6, 44), psychologic factors did not have a significant impact during the ICU course itself where patients may be sedated and less aware of their surroundings. At the time of patient recruitment for this study, there was no protocol in place in the ICU for the management of pain or administration of opioids. As per routine practice, parenteral opioids are often given postoperatively until patients can transition to oral opioids, and parenteral opioids may be continued for breakthrough pain. A large proportion of patients in our cohort received nonopioid analgesics postoperatively, particularly acetaminophen, NSAIDs, and gabapentinoids. Individual clinician practice variation and judgment, as well as variability in interpreting clinical orders (e.g., how often pain is assessed) is something that is difficult to control and could potentially have had an impact on opioids consumed. Nevertheless, this variability in prescribing and administering opioids may enhance the generalizability of our study.

Genetic factors can also influence one’s response to pain. Genetic variations have been shown to explain up to three quarters of the variance in pain responses in animal studies (45). Having a better understanding of genetic risk factors can lead to a more individualized approach to understanding and treating pain. The candidate genes FKBP5, OPRM1, and COMT were chosen based on prior associations with pain: FKBP5 is a key stress response regulator that has led to the development of chronic pain and somatic symptoms after traumatic events and persistent pain of neuropathic and inflammatory origin (12, 46); OPRM1 polymorphism is associated with higher opioid requirements in acute postoperative pain (14); and COMT haplotypes are associated with chronic pain after trauma (13). Preexisting factors such as childhood trauma and lower socioeconomic status have been shown to moderate the relationship between FKBP5 genetic risk factors and chronic pain (47, 48). The interaction of genes with environmental factors is an important consideration for future studies (49, 50) and could explain in part why our findings were not significant. The nature of the pain experienced is also an important consideration. Diatchenko et al (51) found that variations in the COMT gene were only associated with sensitivity to certain types of pain (e.g., painful heat stimuli but not mechanical or ischemic pain). This complex nature of pain is also reflected in the fact that preoperative centralized pain was a significant predictor of ICU OME consumption in our study whereas preoperative pain severity from the Brief Pain Inventory was not. With such complex traits such as postsurgical ICU pain, multiple genes are likely simultaneously contributing to pain phenotypes and opioid consumption. Although additional candidate genes could be tested in the ICU population (e.g., MC1R can modulate pain sensitivity and µ-opioid analgesia; ADRB2 can affect β2-adrenergic receptor signaling; HTR2A can affect serotonin receptor signaling; CYP2D6 can affect the metabolism of opioids; and SCN9A encodes a voltage-gated sodium channel subunit, which can affect intracellular neuronal signaling) (15, 45), future studies investigating genome-wide associations with postoperative ICU pain are also needed.

Our findings need to be interpreted in the context of several limitations. This is a single-center study of post-surgical ICU patients, which may not be generalizable to all critically ill patients. Although patient enrollment and the baseline preoperative survey were done prospectively, critically ill patients were identified retrospectively by professional billing codes, and certain datapoints extracted retrospectively could not be obtained with additional granularity (e.g., total preoperative opioid doses). For opioids given in the ICU, we combined all opioids into OME to facilitate analysis instead of analyzing individual opioids given, as is commonly done in the literature (28, 29). Additionally, we did not extract postoperative pain scores since we believed that opioid administration would be titrated to maintain a pain score within the desired range, and we did not extract intraoperative regional/epidural techniques used, as these occurrences were very low. Finally, a vast majority of our population was White, and the genetic analysis was done in patients of European ancestry, which may not be generalizable to other populations.

CONCLUSIONS

In summary, opioid use during a postoperative ICU course is prevalent, and those who were taking preoperative opioids were more likely to continue to receive opioids throughout their ICU stay than opioid-naive patients. Younger age, preoperative opioid use, preoperative centralized pain, illicit drug use, major surgery, longer anesthetic duration, and ICU mechanical ventilation were associated with opioid use during a postoperative critical care course. The selected genetic variants of FKBP5, COMT, and OPRM1, previously shown to be associated with pain outcomes in other settings, were not associated with OME use in this study. Evaluation of the impact of gene-gene interactions, gene-environment moderations on opioid use in the ICU, and genome-wide associations would be important future considerations.

ACKNOWLEDGMENTS

We acknowledge the Michigan Genomics Initiative participants, Precision Health at the University of Michigan, the University of Michigan Medical School Central Biorepository, and the University of Michigan Advanced Genomics Core for providing data and specimen storage, management, processing, and distribution services, and the Center for Statistical Genetics in the Department of Biostatistics at the School of Public Health for genotype data curation, imputation, and management in support of the research reported in this publication. We also acknowledge Stephanie E. Moser, PhD, Courtney Cole, Kathy Scott, BSN, RN, Ryan Scott, MPH, and Wei-Quan Fang, PhD, for their work and expertise in database management and statistics.

Footnotes

REFERENCES