Multimodal Analgesic Regimen for Spine Surgery: A Randomized Placebo-controlled Trial

Standardized care pathways like Enhanced Recovery After Surgery programs promise to improve perioperative outcomes.^1,2  Multimodal analgesic regimens are an essential component of Enhanced Recovery After Surgery programs, potentially reducing postoperative complications³  and improving recovery.^4–6  While multimodal analgesia is recommended by the American Society of Regional Anesthesia and Pain Medicine (Pittsburgh, Pennsylvania),⁷  perioperative opioid administration remains the dominant component of most pain management plans.⁸ 

Postoperative pain control is especially challenging in spine surgery patients. Acetaminophen,⁹  gabapentin,^10–12  lidocaine,^13,14  and ketamine^15–17  can potentially improve acute pain but are not consistently used. To improve pain control of spine surgery patients, our anesthesia and surgery teams devised a multimodal analgesic pathway that included acetaminophen, gabapentin, lidocaine, and ketamine.

The drugs, doses, and routes we considered were selected based on best available evidence and team consensus. Acetaminophen and gabapentin decrease perioperative opioid use, and perhaps reduce persistent incisional pain.^18,19  Even a single preoperative dose of acetaminophen reportedly improves postoperative pain control.²⁰  Similarly, single-dose gabapentin 600 mg is reported to reduce postoperative pain.²¹  Ketamine blocks N-methyl-d-aspartate excitatory glutamate receptors, and intraoperative infusions are widely reported to reduce postoperative opioid use.^22–24  Similarly, lidocaine infusion has analgesic, antihyperalgesic, and antiinflammatory properties, which help control postoperative pain.^14,25 

Our goal was to improve Quality of Recovery and pain control while minimizing opioid use.²⁶  We therefore tested the primary hypothesis that patients given multimodal analgesia consisting of oral gabapentin and acetaminophen combined with infusions of lidocaine and ketamine have superior Quality of Recovery scores 3 days after multilevel spine surgery. Secondarily, we tested the hypotheses that multimodal analgesia reduces opioid consumption and pain scores during the initial 48 postoperative hours; we also compared the opioid related side effects at first and second postoperative days.

Our exploratory outcomes were (1) postanesthesia care unit (PACU) length of stay; (2) postoperative nausea and vomiting 24 h after surgery; (3) need for acute pain consultation, as determined by clinicians; (4) patient satisfaction with pain management at discharge (numeric rating scale 1-100); (5) hospital length of stay; (6) Quality of Recovery at 1 month; (7) EuroQol (EQ-5D) at 3 months; (8) chronic postsurgical pain at 3 months; and (9) pain disability questionnaire at 3 months.

After Institutional Review Board (IRB No. 16-012) approval in July 2016, we conducted a double-blinded, placebo-controlled, parallel-group, randomized controlled trial at the Cleveland Clinic Main Campus (Cleveland, Ohio) from August 2016 to December 2018 (fig. 1). The trial was registered at ClinicalTrials.gov before the first patient was enrolled (NCT00517127, Principal Investigator: Kamal Maheshwari). We enrolled patients scheduled for multilevel posterior spine elective surgery who were at high risk for postoperative pain. We considered two or more of the following factors to constitute high risk^27,28 : age less than 60 yr; female sex; history of chronic pain exceeding 3 months; preoperative visual analog pain scale 7/10 or greater; high anxiety level based on the Amsterdam Preoperative Anxiety and Information Scale²⁹  when any of the questions were marked as 4 and 5 corresponding to very large and extreme level of worry; and complex spine surgery with instrumentation. We excluded patients who had allergies or hypersensitivities to lidocaine, ketamine, acetaminophen, or gabapentin; who had abused drugs within 6 months; who had heart failure with ejection fraction less than 30%; or who had liver dysfunction manifested by twice normal liver enzymes and International Normalized Ratio equal to or greater than 2.

The study statistician generated an allocation sequence table corresponding to 1:1 randomization with random-sized blocking and without stratification. An investigator screened patients in the preanesthesia evaluation clinic for potential eligibility and obtained informed consent during the visit. A total of 299 adults scheduled for multilevel posterior spine surgery were assigned to study groups with a Web-based system that concealed allocation until shortly before surgery.

All patients received general inhalational anesthesia with endotracheal intubation and standard American Society of Anesthesiologists (Schaumburg, Illinois) monitors. Clinicians provided anesthesia care per clinical routine including hemodynamic monitoring, intraoperative administration of opioid medications, and muscle relaxants. Enrolled patients were instructed not to take any acetaminophen or gabapentin at home on the day of surgery. Patients assigned to the analgesic pathway were given single preoperative oral doses of 1 g acetaminophen²⁰  and 600 mg gabapentin²¹ ; they were also given an intraoperative infusion of 1.5 mg/kg/h lidocaine¹⁴  (maximum individual dose, 200 mg/h) and 5 µg/kg/min ketamine.³⁰  Ketamine was stopped at wound closure. Lidocaine was reduced to 1 mg/kg/h at wound closure and was continued for the initial hour in the PACU so long as the total duration did not exceed 8 h. Epidural analgesia or local wound infiltration were performed by the surgeons on a preference basis. Oral acetaminophen was continued in most patients after surgery. Gabapentin was continued at the discretion of the surgery team.

Patients assigned to the placebo group received placebo acetaminophen and gabapentin, along with placebo ketamine and lidocaine. No other aspect of intraoperative and postoperative care was controlled by the study protocol. All study medications were prepared by the Cleveland Clinic research pharmacy; clinicians, patients, and investigators were thus fully blinded.

Baseline characteristics as well as primary and exploratory outcomes were recorded by investigators. Pain scores and opioid consumption were obtained from patients’ electronic medical records.

Our primary outcomes were Quality of Recovery score (range, 0 to 150) on the third postoperative day. This survey is a valid, reliable, responsive, and multidimensional measure of Quality of Recovery after anesthesia and surgery. It includes physiologic values, functional recovery, and patient-reported outcomes.³¹ 

Postoperative pain was evaluated with numeric rating scores (0 to 10) at 15-min intervals for the initial two postoperative hours, and thereafter every 4 h by ward nurses. Time-weighted average pain scores were calculated over the initial 48 postoperative hours or until discharge, whichever came first. A time-weighted average pain score is equal to the sum of the portion of each time interval (as a decimal, such as 0.25 h) multiplied by the levels of the pain (0 to 10 numeric rating scores) during the time period divided by 48 h. Opioid consumption within the initial 48 h was converted to IV morphine equivalents using the conversions specified in Supplemental Digital Content 1 (https://links.lww.com/ALN/C178). Opioid-related symptom distress was evaluated on the first two postoperative mornings.³²  Chronic postsurgical pain was assessed by a phone call using numeric rating scale (0 to 10) where a score of 0 is “no pain” and a score of 10 is “pain as bad as it could be.” Postoperative nausea and vomiting was obtained from electronic health record nursing progress notes; severity was 0 = none, 1 = mild, 2 = moderate, and 3 = severe.

The analysis was intent-to-treat, and thus included all randomized patients. We assessed the balance of the two randomized groups on baseline and procedural characteristics using absolute standardized difference, defined as the absolute difference in means, mean ranks, or proportions divided by the pooled SD. Baseline variables with absolute standardized difference greater than 0.2 were considered to be imbalanced and would be adjusted for in primary, secondary, and exploratory analyses as well as sensitivity and post hoc analyses. SAS statistical software version 9.4 (SAS Institute, USA) for 64-bit Microsoft Windows (USA) was used for statistical analysis.

We estimated the effect of analgesic pathway care approach on Quality of Recovery scores with an independent samples t test, which was justified because the scores were normally distributed. We report differences in means Quality of Recovery scores, along with 95% CIs.

Among 299 randomized patients, 41 (14%) were discharged before the third postoperative day. Missing Quality of Recovery scores were imputed by multivariable single imputation with five imputations: the imputation model included all the baseline variables listed in table 1 together with exposure and all the primary, secondary, and exploratory outcomes. We ran two sensitivity analyses: (1) in the “complete cases” analysis, patients without 3-day postoperative Quality of Recovery score were excluded from the analysis; and (2) we assumed that patients who were discharged before completing the Quality of Recovery survey had the average score in their randomized group. We also conducted two separate post hoc subgroup analyses to assess whether the relationship between analgesic approaches and Quality of Recovery score differed in patients with histories of chronic pain or opioid use. The interaction terms between analgesic approaches and history of chronic pain and opioid use were added to a multivariable linear regression model, and significant interaction was claimed if the corresponding P value was more than 0.10.

We assessed the effectiveness of analgesic pathway compared to placebo on cumulative opioid consumption and pain intensity scores within the first 48 h after surgery, using a joint hypothesis testing framework.³³  We considered the analgesic pathway to be superior to placebo on postoperative pain management if both opioid consumption (Supplemental Digital Content 1, https://links.lww.com/ALN/C178) and pain scores were noninferior (i.e., not worse) and at least one of the outcomes was superior for analgesic pathway patients. We defined the a priori noninferiority pain score delta as 1 point (on a scale of 0 to 10) and the opioid delta as 20% change from IV morphine equivalent doses.

We first estimated CIs for the treatment effect for both pain score and opioid consumption. Opioid consumption had highly skewed distribution, with many patients having consumption of zero. We therefore evaluated the difference in median IV morphine equivalent dose between the two groups using Hodges–Lehmann estimation of location shift.³⁴  To evaluate the difference in mean pain scores, we used a linear regression model to assess the exposure effect on the time-weighted average pain score outcome.

Joint hypothesis testing of pain score and opioid consumption was conducted at the overall 0.025 significance level (Bonferroni-corrected for two secondary hypotheses), and all tests were one-tailed in the direction favoring the novel pain management approach. Noninferiority of analgesic pathway versus placebo was assessed for each outcome at the 0.025 level (no Bonferroni correction) since noninferiority is required on both outcomes. Therefore, noninferiority (being “not worse”) was concluded for both outcomes at the significance level of 0.025 if the upper limit of 95% CI was below the corresponding noninferiority delta. If noninferiority on both outcomes was found, superiority was assessed on each. We adjusted for two outcomes for the superiority testing only, using a significance criterion of 0.0125 for each outcome (i.e., 0.025/2, Bonferroni correction), since superiority on either outcome would suffice. Superiority thus was claimed for a particular outcome if the 97.5% interval limits were below zero for both pain score and opioid consumption. No multiple testing adjustment was required for assessing both noninferiority and superiority since significance for both was required to reject the null hypothesis, and also because superiority falls in the noninferiority rejection region.

For post hoc analysis, we assessed whether the effect of the analgesic on 48-h pain management is different for opioid-naïve patients than patients who had a history of opioid use. Specifically, we added an interactions term between the exposure and history of opioid use in the pain and opioid models (with a criterion for significant interaction of P < 0.10). Additionally, we evaluated the effect of two study groups on pain management at 24 h postoperatively.

Opioid-related side effects scores were log-transformed to normalize the distributions and compared with a log-normal linear regression model; the percent difference in geometric means scores are reported, along with 95% CI and corresponding P values. Post hoc, we considered interactions between randomized assignment and several baseline risk factors for poor postoperative pain control, with the significance criterion for interactions being P < 0.10.

This study was designed to have 90% power at the 0.05 significance level to detect a ratio in geometric means of 1.15 or higher of Quality of Recovery score comparing the analgesic pathway to placebo, assuming that the Quality of Recovery coefficient of variation (SD/mean before log-transformation) was 0.43 for both groups. After accounting for an anticipated dropout rate of 10% (due to surgery cancelation, withdrawals and other unexpected events) and for two interim and one final analyses, we concluded that we would need to enroll a maximum of N = 440 patients for this study (N = 220 patients per group).

We used a group sequential design³⁵  to test for efficacy and futility to maintain the significance level for the primary outcome at 5% and the power at 90% across the interim analyses. Efficacy and futility are specific interim analysis terms used for overwhelmingly positive and convincing negative results.^35,36  Specifically, we employed a gamma spending function (gamma = –4 for efficacy and –1 for futility)³⁷ ; the originally planned efficacy and futility boundaries are displayed in figure 2. The final boundary for the primary outcome was P < 0.047 for efficacy.

After enrolling 299 patients, we recalculated the power of the independent samples t test using the actual study estimates of coefficient of variation (0.22). We did not use the observed treatment effect, but rather assessed power to detect the planned 15% relative increase in geometric mean Quality of Recovery score at the overall 0.05 significance level. With 299 patients, we had 99% power to detect a 15% relative increase in Quality of Recovery score.

The trial was stopped per protocol after the second interim analysis when results crossed our predefined futility boundary. At that point, there were 150 (50%) patients randomized to the analgesic pathway group and 149 (50%) patients randomized to the standard of care group (fig. 1). Demographic, baseline, and surgical characteristics are presented in table 1; all potential confounders were balanced at the baseline. About half of the patients were women, half were smokers, and 80% were American Society of Anesthesiologists Physical Status III.

Quality of recovery at 3 days, our primary outcome, is reported by study groups in table 2, figure 3, and Supplemental Digital Content 2 (https://links.lww.com/ALN/C179; individual questions of Quality of Recovery survey). The treatment effect on Quality of Recovery outcome was not significant (P = 0.920), with the estimated difference in means of 0 (95% CI, –6 to 6). Results of both sensitivity analyses were consistent with primary analysis. Per post hoc analysis, there were no significant interactions between analgesic approaches and history of chronic pain or history of opioid use (P = 0.733 and P = 0.696), meaning the relationship between postoperative analgesic approaches and Quality of Recovery outcome was not modified by these two factors.

Pain management within the initial 48 postoperative hours was not superior with the analgesic pathway per our predefined joint hypothesis rules (table 3): both 48-h pain and opioid were noninferior in the analgesic pathway group (both noninferiority P < 0.001); however, neither of two outcomes was significantly lower in the analgesic pathway group (superiority P = 0.175 in opioids and P = 0.094 in pain). Per post hoc analysis, there were no significant interactions between history of opioid use and the exposure on 48-h pain scores outcome (P = 0.775) and on opioids consumption outcome (P = 0.202).

Both 24-h pain scores and opioid use were noninferior in patients randomized to the analgesic pathway (both noninferiority P < 0.001). Pain scores were significantly lower in the analgesic pathway group (superiority P = 0.025), but not by a clinically important amount. Opioid-related side effects (table 4) were similar in each group on both the first (P = 0.819) and second (P = 0.530) postoperative days.

None of the exploratory outcomes including PACU length of stay, postoperative nausea and vomiting, patient satisfaction with pain management at discharge from the hospital, Quality of Recovery score at 1 month, and health-related quality of life EQ-5D at 3 months differed by clinically important amounts (table 5).

In spine surgery patients at high risk of postoperative pain, a multimodal analgesic pathway that included acetaminophen, gabapentin, lidocaine, and ketamine did not improve Quality of Recovery. Presumably, Quality of Recovery was similar in the treatment and placebo groups because there were no clinically important differences in the most obvious mediators, postoperative opioid consumption and pain scores. It was surprising that the combination of four established analgesics proved ineffective in our patients. There are at least four potential explanations, including the surgical setting; the analgesic combination, doses, and routes; trial design; and the chosen primary outcome.

We enrolled patients at high risk for postoperative pain, and studied an operation that is especially painful. Spine surgery differs from most other surgical procedures in provoking both neuropathic and muscle pain. Neuropathic pain is notoriously difficult to treat, and spasmodic muscle pain also responds poorly to many conventional analgesics. It is thus possible that the combination we tested would have been more effective for other procedures. In contrast, local anesthetic infiltration or epidural analgesia might have been more effective in our patients.

Each of the four drugs we tested is reported to improve postoperative analgesia.^9,14,17,21  Two of the drugs, acetaminophen and gabapentin, were given orally as a single preoperative dose. While there are reports supporting this approach,^9,10,21  both are relatively short-acting, and it seems somewhat unlikely that either would provide clinically important analgesia for several postoperative days. Each presumably provided a degree of preemptive analgesia, but most trials do not identify long-term benefit from preemptive analgesia.³⁸  Acetaminophen could have been given intravenously, but there is no evidence that efficacy is improved.³⁹ 

From a practical perspective, ketamine and lidocaine were our most important interventions. Both were given by infusion throughout surgery, providing a reasonable exposure. The dose of lidocaine, 1.5 mg/kg/h, is similar to that used in many previous trials that reported benefit, and is probably near the safety threshold. The dose of ketamine, 5 µg/kg/min (about 0.3 mg/kg/h in a 70-kg patient), was on the low end of the typical range for use as a perioperative analgesic adjuvant. However, it is well within recent consensus guidelines suggesting that 0.1 to 0.5 mg/kg/h provides a good balance between analgesia and adverse effects.⁴⁰  Our trial was predicated on the assumption that each of the four tested drugs is individually effective. In fact, the evidence for effectiveness is relatively weak. It is also possible that benefit would have been observed at a higher dose, with some risk of toxicity.

The specific combination of drugs we evaluated has not previously been rigorously tested, although many trials evaluated each individually. Why the combination would prove ineffective, given the apparent benefit from the individual components, remains unclear. A remote possibility is that the drugs interacted antagonistically. However, it seems unlikely that antagonistic interactions explain our negative results. We evaluated the combination of all four drugs simultaneously, but an alternative would have been to use a factorial design that would evaluate individual effects. With enough patients, about four times the number we studied, it would have been possible to formally evaluate interactions among the drugs. In contrast to our rigorous blinded randomized design, many previous studies reporting benefit used scientifically weak before–after designs⁴¹  that suffer from the Hawthorne effect, regression to the mean, and time-dependent confounding (e.g., practice improvements over time).

We enrolled 299 patients, making our trial considerably larger than nearly all previous perioperative trials of the drugs we tested. The trial was stopped because results were well into the futility range. Our results showing lack of benefit from multimodal analgesic pathway are robust (a difference in quality of recovery of 0 units), rather than representing an underpowered study or uncertain results.

Spine surgery is often performed to limit disability and improve quality of life, and we believe it should be the goal. Therefore, we chose Quality of Recovery as the primary outcome, which is a valid, reliable, and responsive measure of functional recovery after surgery.^26,42  The Quality of Recovery instrument evaluates the following domains: pain, physical comfort, physical independence, psychologic support, and emotional state. Our trial evaluated four analgesics, so the most obvious benefit would be in the pain domain. Consequent opioid sparing might improve ambulation and thus the physical independence domain. Opioid sparing might also improve emotional state. The effects of ketamine are less obvious and might either improve or worsen emotional state. In fact, pain scores were not improved and opioid use was comparable in both trial groups. It is therefore unsurprising that Quality of Recovery was also similar in each group.

Multimodal analgesia is a component of many care pathways or Enhanced Recovery After Surgery protocols. The assumption is that combining various nonopioid analgesics will reduce the need for opioids and consequently opioid-related complications. For example, the American Society of Anesthesiologists Taskforce on Acute Pain Management recommends multimodal analgesia, a recommendation largely based on class C evidence defined as “informal opinion.”⁷  In most cases, the assumption that multimodal analgesia is beneficial has not been rigorously tested—which was the basis for our trial. Unfortunately, widespread implementation of Enhanced Recovery After Surgery programs is often based on weak evidence.⁴³  Our negative findings suggest that care pathways should be formally tested in each clinical context. For example, Bragato and Jacobs noted success with care pathways in orthopedic units, but not in trauma units, and cautioned against universal implementation of care pathways in all clinical areas.⁴⁴ 

We did not control postoperative use of acetaminophen and gabapentin. However, use was similar in each group. For example, about 90% of the patients in each group were given acetaminophen, and about 40% of each group was given gabapentin. Pain scores at rest were evaluated by nurses in their clinical routine. Possibly pain scores would have differed with activity if analgesia was disproportionately effective at higher pain intensities. We included patients at moderate-high risk of pain, who are most likely to benefit from multimodal analgesia. Although it seems unlikely, patients at lower or much higher risk might possibly benefit.

To summarize, use of multimodal analgesic pathway based on preoperative single-dose acetaminophen and gabapentin, and intraoperative infusions of lidocaine and ketamine, did not improve day 3 Quality of Recovery or reduce pain scores or 48-h opioid consumption. This combination of four analgesics was not beneficial for patients having multilevel spine surgery. The lack of benefit in spine surgery patients suggests that multimodal analgesia should be formally tested in each clinical context.

Support was provided solely from institutional and/or departmental resources.

Dr. Krishnaney reports a relationship as a consultant with Stryker Corp. (Kalamazoo, Michigan). Dr. Machado reports a relationship as a consultant with Abbott (Abbott Park, Illinois), Medtronic (Minneapolis, Minnesota), Enspire (Austin, Texas), and Cardionomic Inc. (New Brighton, Minnesota) intellectual property distribution rights. Dr. Rosenquist is a member of the Clinical Events Review Committee, Mainstay Medical (Dublin, Ireland); member of the Board of Directors for KURE Pain Holdings, LLC (Wilmington, Delaware); receives royalties from and is an educational content developer for UpToDate (Waltham, Massachusetts); receives royalties from Oxford University Press (Oxford, United Kingdom) and Springer (Berlin, Germany); and received honoraria for content development and presentation of risk evaluation and mitigation strategy material from the American Society of Anesthesiologists (Schaumburg, Illinois). The other authors declare no competing interests.

Full protocol available at: maheshk@ccf.org. Raw data available at: maheshk@ccf.org.