Paracetamol and selective and non-selective non-steroidal anti-inflammatory drugs (NSAIDs) for the reduction of morphine-related side effects after major surgery: a systematic review

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 08/114/01. The contractual start date was in February 2009. The draft report began editorial review in August 2009 and was accepted for publication in October 2009. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

None

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

Morphine

Morphine is the most valuable opioid for severe postoperative pain relief. It is the gold standard against which the effectiveness of all other analgesics is compared. 7 Although there are several modes of administration, patient-controlled analgesia (PCA) has become the standard method of administering morphine after major surgery. 5 PCA involves the patient self-administering small doses of morphine by pressing a button connected to a programmable pump. The PCA device is programmed by the health-care provider to deliver a specific amount of medication (a ‘bolus’) upon each request by the patient. A continuous ‘background’ infusion may be administered in addition to patient-controlled bolus doses. In order to prevent an overdose of morphine, bolus doses are limited by a programmed ‘lockout interval’ during which subsequent requests are ignored. 7 PCA has been shown to provide marginally superior analgesia in comparison to other modes of administration, and patients report greater satisfaction with, and in general prefer, PCA. 8

Morphine exerts its analgesic effect by binding to specific opioid receptors in the brain and spinal cord that are involved in the perception of pain. This mode of action can also result in significant adverse effects. These include: respiratory depression, postoperative nausea and vomiting (PONV), sedation, bowel dysfunction (delayed gastric emptying, inhibition of bowel motility and constipation), urinary retention and pruritus. 1,9

Respiratory depression, though uncommon, is a potentially life-threatening adverse effect and of most concern to health-care professionals. 10 Meanwhile PONV, although self-limiting, is common, having an incidence of 30–67%, and is of most concern to patients. 1,11,12 Furthermore, PONV can delay postoperative recovery, which has consequences for the patient and also has an economic impact on health-care resources. 13

Paracetamol

Paracetamol (acetaminophen) is an analgesic and antipyretic with little anti-inflammatory effect, whose exact mode of action is currently unknown. It is the most widely used drug for pain relief. In order of increasing effectiveness, paracetamol can be administered rectally, orally and intravenously. 14 While all three modes of administration can achieve adequate plasma concentrations, there are differences in absorption and time to reach peak plasma levels. With rectal administration, absorption can be unpredictable with bioavailability ranging from 24% to 98%, varying with factors such as formulation of the suppositories, number used and the particle size of the paracetamol. 15 Paracetamol, at therapeutic doses, rarely results in adverse effects and, unlike NSAIDs, does not cause gastrointestinal ulceration or bleeding. 1 Propacetamol hydrochloride, an injectable prodrug of paracetamol, was the first form of paracetamol developed to be administered intravenously. 14,16 It is hydrolysed to paracetamol in the blood, with 2 g of propacetamol releasing 1 g of paracetamol. Propacetamol, though effective and generally well tolerated, is notable for adverse effects of localised pain at the injection site and contact dermatitis. Although licensed and available in other countries, including France and Belgium, it is not licensed in the UK (Table 1). However, an intravenous form of active paracetamol, Perfalgan^®, has been available in the UK since 2004. Studies have shown that compared to intravenous (i.v.) propacetamol, i.v. paracetamol is associated with a reduction in incidence of localised pain at the injection site and contact dermatitis. However, there is no significant difference in the incidence of other adverse effects. 14

Non-steroidal anti-inflammatory drugs (NSAIDs) and cyclo-oxygenase 2 (COX-2) inhibitors

Non-steroidal anti-inflammatory drugs are analgesic, anti-inflammatory, antiplatelet and antipyretic. In comparison to paracetamol, NSAIDs have been shown to offer superior postoperative pain relief. 17 They exert their analgesic effect by reducing the production of prostaglandins responsible for pain and inflammation. NSAIDs achieve this by inhibiting the enzyme COX-2, which is essential in the synthesis of these prostaglandins. NSAIDs vary in whether they selectively inhibit COX-2. Non-selective NSAIDs, such as ibuprofen and diclofenac, inhibit not only COX-2 but also cyclo-oxygenase 1 (COX-1). COX-1 is involved in the synthesis of prostaglandins that have a role in the maintenance and protection of the gastrointestinal (GI) tract, platelet adhesion and renal function. Non-selective NSAIDs are therefore associated with adverse GI effects, renal toxicity, prolonged bleeding time, bronchospasm and oedema. 1 Several NSAIDs are available for use in the postoperative setting (see Table 1).

Non-steroidal anti-inflammatory drugs, even when used in the short term, can cause GI adverse effects ranging from abdominal pain, dyspepsia and superficial erosions to serious GI complications such as perforated gastric ulcers and life-threatening GI haemorrhage. 18 Furthermore, the risk of a GI adverse event varies between NSAIDs, with the lowest risk associated with ibuprofen and the highest with ketorolac. 19 Renal toxicity is a noted adverse effect of NSAIDs. However, a systematic review found that the use of NSAIDs for postoperative pain relief in adults with normal renal function causes only a small, temporary effect on renal function. 20

A systematic review examining the use of NSAIDs after tonsillectomy, where perioperative bleeding is a serious complication, found that NSAIDs were statistically significantly associated with the need for reoperation due to bleeding [odds ratio (OR) 2.3; 95% confidence interval (CI) 1.12 to 4.83]. However, NSAIDs were not statistically significantly associated with intraoperative blood loss, postoperative bleeding and hospital admission. 21

Cyclo-oxygenase 2 inhibitors, also referred to as ‘COXIBs’ or ‘Cox-2 selective NSAIDs’ (see Table 1), were designed to selectively inhibit COX-2 only, thereby reducing GI bleeding and renal adverse effects. 4 However, the long-term use of COX-2 inhibitors is associated with increased incidence of thromboembolic events such as myocardial infarction and stroke, and they are as likely as non-selective NSAIDs to cause impaired renal function and oedema. 1,9 Over the past 5 years, two COX-2 inhibitors have been withdrawn from use worldwide: rofecoxib due to an increased risk of cardiovascular adverse effects, and valdecoxib due to an increased risk of severe skin reactions. 22,23

Previous systematic reviews

There are a number of previous relevant reviews assessing the effectiveness of adding a non-opioid to PCA morphine for pain relief and reduction of morphine-related side effects following surgery. Some reviews have focused on specific types of surgery, for example cardiothoracic surgery24 and lumbar spine surgery. 25 We have identified three previous systematic reviews that are not procedure specific and were all published in 2005: Remy et al. 26 investigated the effects of paracetamol on morphine consumption and associated adverse effects after surgery; Marret et al. ,27 from the same research group, investigated the effects of NSAIDs (including COX-2 inhibitors); and Elia et al. 28 investigated paracetamol, NSAIDs and COX-2 inhibitors.

The reviews by Remy et al. 26 and Elia et al. 28 both showed that paracetamol (including propacetamol) combined with PCA morphine results in a statistically significant reduction in morphine consumption in the first 24 hours following surgery: there was a pooled mean reduction of 9 mg and 8.3 mg respectively compared to PCA morphine alone (Tables 2 and 3). However, there was not a statistically significant reduction in the incidence of any morphine-related adverse effects including PONV, urinary retention, sedation, pruritus, apnoea or respiratory depression in either study. 26,28

Marret et al. 27 reported that, compared to PCA morphine alone, there was a statistically significant reduction in PONV, nausea alone, vomiting alone and sedation with NSAIDs in combination with PCA morphine (see Table 3). Non-selective NSAIDs and COX-2 inhibitors were combined for some analyses. Furthermore, regression analysis indicated a positive correlation between morphine consumption and the incidence of postoperative nausea or vomiting, though the size of the correlation was small (r² = 0.37 for nausea and r² = 0.27 for vomiting). There was no statistically significant decrease in the incidence of pruritus, urinary retention or respiratory depression when NSAIDs were added to PCA morphine. Data were not pooled for morphine consumption.

The review by Elia et al. 28 assessed the effect of the non-selective NSAIDs and COX-2 inhibitors separately. There was a statistically significant reduction in morphine consumption with NSAIDs in combination with PCA morphine compared to PCA morphine alone (10.3 mg with single doses, 18.3 mg with continuous infusion, and 19.7 mg with multiple dose regimens). There was also a statistically significant reduction in sedation and PONV but not for nausea or vomiting alone, though the trend was towards reduction (see Table 2). 28 In contrast, whilst COX-2 inhibitors in combination with PCA morphine resulted in a statistically significant reduction in morphine consumption compared to PCA morphine alone, there was no statistically significant reduction in any morphine-related adverse effects (Table 2). 28

Any reduction in morphine-related adverse effects needs to be balanced against the possible adverse effects of the non-opioid analgesic. The reviews by Marret et al. 27 and Remy et al. 26 did not consider this issue. In the review by Elia et al. 28 the use of NSAIDs was associated with a statistically significant increase in the incidence of surgical bleeding complications (Table 4). COX-2 inhibitors were associated with a statistically significant increase in renal failure, but not surgical bleeding complications (Table 4).

In summary, the existing systematic reviews suggest that paracetamol, NSAIDs and COX-2 inhibitors all reduce morphine consumption in the first 24 hours following surgery, but only NSAIDs appear to reduce morphine-related adverse effects. However, the relative effects of the non-opioids are unclear.

Definition of decision problem

The problem faced by decision-makers in health care is which class of non-opioid analgesic (paracetamol, NSAID or COX-2 inhibitor) is the most effective at reducing morphine consumption and associated adverse effects when used as part of multimodal analgesia following major surgery. Any benefits in terms of reduction in morphine-related adverse effects need to be balanced against the potential risk of adverse effects of the non-opioid analgesic.

The scope of the review

We were commissioned to undertake a short report, building on earlier reviews of paracetamol and NSAIDs, to conduct an analysis comparing the morphine-sparing effects of these drugs following major surgery.

Of the available reviews we elected to update the Elia et al. 28 review. This was a good-quality review with appropriate searches and clearly defined inclusion criteria that used appropriate methods to reduce error and bias in study selection and data extraction. Study quality was assessed and taken into consideration in the synthesis. The search date for the Elia review is more recent by 7 months than the other two reviews and as a result captured more trials from 2003 and 2004. The Remy and Marret reviews used a quality score as an inclusion criterion for their review; however, we preferred to include all the randomised evidence, as Elia had done, to maximise the evidence available. In addition, we also had access to the individual trial data from the Elia review, which included the adverse effects of the non-opioid analgesics as well as morphine-related adverse effects.

The earlier three reviews, including the Elia review, did not compare the three classes of non-opioid analgesics to each other, possibly a reflection of the limited number of trials making direct comparisons. The main aim of the current review was to assess the relative effectiveness of paracetamol, NSAIDs and COX-2 inhibitors. The focus was the relative effectiveness of the drug classes and not individual drugs within the classes. The ideal evidence to address the decision problem posed would be a synthesis of three-arm trials comparing paracetamol versus NSAID versus COX-2 inhibitor. In terms of the current review, we were aware that although there was a reasonable body of evidence comparing each of the three analgesic classes to placebo, it was likely that the quantity of evidence directly comparing the three drug classes would be limited. We therefore undertook a mixed treatment comparison (MTC) to derive results for the relative effectiveness of the three non-opioid analgesics in the first 24 hours following surgery.

An MTC is an extension or generalisation of traditional meta-analysis in which trials comparing the same intervention and same comparator are pooled to estimate an overall treatment effect. An MTC overcomes the limitations of standard meta-analysis in cases where there are no or limited trials making the relevant head-to-head comparison or where the decision problem requires the comparison of several interventions. 29,30 In addition, a ranking of interventions based on the probability that each treatment is best can be produced,31 which can be of particular value where several treatment options are under consideration.

Search strategy

MEDLINE, EMBASE and the Cochrane Central Register of Controlled trials (CENTRAL) were searched for the period January 2003 to February 2009. The search strategy for each database is reported in Appendix 1. The start search date was January 2003 to overlap with Elia et al. 28 (search end July 2004) to allow for late indexing of studies. Published and unpublished studies were eligible and no language restrictions were applied. In addition, the reference lists of relevant systematic reviews were checked to identify relevant studies.

Titles and abstracts were examined for relevance by two researchers, and all papers identified by either researcher as potentially relevant were ordered. Full papers were examined for relevance by two researchers independently, based on the inclusion criteria below. Disagreements were resolved by consensus and if necessary through discussion with a third researcher.

Inclusion and exclusion criteria

The inclusion criteria followed those of Elia et al. 28 except where indicated below. Studies were included if they met the following criteria:

Population Adults who had undergone major surgery and were receiving PCA morphine for postoperative pain were included. Studies using PCA opioids other than morphine, intrathecal opioids or peripheral nerve blocks were excluded.

Interventions Studies of paracetamol (including propacetamol), non-selective NSAID or COX-2 inhibitor given in addition to PCA morphine were included. The COX-2 inhibitors rofecoxib and valdecoxib were not included as these are no longer licensed in the UK. Although propacetamol is not licensed in the UK it was included as it is a prodrug of paracetamol and we anticipated that there would be few trials available of paracetamol used as licensed in the UK.

Comparator treatment PCA morphine plus placebo or PCA morphine plus a different non-opioid class (paracetamol, NSAID or COX-2 inhibitor) were included. Studies using a no treatment comparator were excluded.

Outcomes Only studies that reported cumulative morphine consumption for the first 24 hours following surgery were included. The other outcomes of interest were: morphine-related adverse effects (respiratory depression, nausea, vomiting, PONV, urinary retention, pruritus, dizziness, sedation, including drowsiness or somnolence, and bowel dysfunction) and non-opioid-related adverse effects. The presumption was made that pain was adequately controlled with PCA morphine in both arms of the trial; therefore pain was not included as an outcome.

Study design Randomised controlled trials (RCTs) with at least 10 participants per treatment group were included.

Data extraction

The data previously extracted by Elia et al. 28 formed the basis for the update (http://anesthesiologie.hug-ge.ch/data.htm). The data from the earlier review were not available as data files, therefore the data were extracted directly from the papers. These were then checked by a second researcher against the original paper and the data extracted by Elia et al. 28 Where Elia et al. had obtained data directly from authors, these data were used for the current review. For some of the studies from the earlier review, missing data could not be obtained directly from the authors and data were then estimated from a graph. New studies were also extracted by one researcher and checked by a second. Authors of trials published since the review by Elia et al. were contacted for additional information where necessary. The data extracted from the individual studies are provided in Appendix 9.

For 24-hour morphine consumption (i.e. morphine consumption in the first 24 hours following surgery), the mean and standard deviation (SD) were extracted for the intervention and comparator. The number of events was extracted for morphine-related and non-opioid analgesic-related adverse effects. Where the denominator for adverse effects reported by the primary study authors was the number of patients in the analysis, this was extracted. This replicated the approach by Elia et al. 28 Some of the studies reported adverse effects beyond the immediate 24-hour period or were not explicit about the cut-off used. In these instances adverse events for the whole period were recorded to avoid loss of data from these studies.

Study quality

Study quality was assessed using the same modified seven-point four-item Oxford scale32 used by Elia et al. 28 This scale assesses whether randomisation, concealment of allocation, double blinding and the flow of patients within a study are adequately described or not (see Appendix 5). The minimum score attainable on the scale is zero and the maximum score is seven.

Methods for synthesis

Quantity and quality of research available

The searches identified 4357 potentially relevant references (Figure 1). On the basis of screening titles and abstracts, 147 full papers were ordered for further assessment. In addition 52 papers from the Elia et al. 28 review were ordered for screening making a total of 199 full papers. Of the 199 full papers, 139 were excluded because they did not meet the inclusion criteria; reasons for exclusion are reported in Appendix 3. One hundred and twenty-seven of these papers were new studies, of which two40,41 were excluded due to retraction by the respective journals early in 2009 because of falsification of data. 42,43 We were not able properly to assess for inclusion one Turkish language study due to problems in getting a translator,44 and one Bulgarian language study45 as the journal was not held by the British Library. Twenty new studies met the inclusion criteria.

FIGURE 1.

Selection of studies.

Twelve of the 52 studies included in the earlier review were excluded from the current review. Four were of valdecoxib or rofecoxib, which are no longer licensed in the UK;46–49 three had a no treatment comparison group (i.e. no placebo or active intervention);50–52 in one the NSAID was given in conjunction with another analgesic;53 in one a variety of opioids were administered via PCA;54 one was based upon an abstract for which a full paper was published since the searches undertaken by Elia;28 and one by Reuben55 was excluded as it was retracted by the journal early in 2009 due to falsification of data. 42 We also decided to exclude a further paper by this author. 56 This paper has not been retracted but, because we were aware of at least 12 papers by Reuben that had definitely been withdrawn, and at the time of the analysis were unable to establish with certainty the veracity of this second paper, we excluded it from the review. 57

When the relevant studies from the earlier review (n = 40) and those identified from our own searches (n = 20) were combined there were a total of 60 included studies. Two of the included studies were non-English language, one being Greek and the other German. 58,59

Study characteristics

There were no studies located that directly compared all three classes of drug (NSAID, COX-2 inhibitor and paracetamol) and none that compared COX-2 to paracetamol (Table 5). One study directly compared COX-2 inhibitor to NSAID (and placebo);60 and there were five studies that directly compared NSAID and paracetamol (three also had a placebo arm61–63 and two did not64,65). Placebo was the only comparator in 15 studies of COX-2 inhibitors, in 32 studies of NSAIDs and in seven studies of paracetamol (Table 5).

The characteristics of the included studies are summarised in Table 6. All of the participants were receiving PCA morphine for at least 24 hours following major surgery. A range of different surgeries were undertaken across the studies, and sometimes within studies, including thoracic, orthopaedic, gynaecological, obstetric and general surgery. General anaesthesia was most commonly used (see Appendix 9 for further details of anaesthesia). The number of participants in the included studies ranged from 20 to 514, and over 40% of studies had 20 or fewer participants in each comparison group.

The type of drug, dosing regimen and mode of administration of COX-2 inhibitors and NSAIDs varied between studies. The dosing regimen for each study is provided in Table 6, and details of the dosing regimen, by drug type, are provided in Appendix 4.

The COX-2 inhibitors investigated were parecoxib (11 studies),58,60,69,71–76,78,79 celecoxib (three studies),67,68,70 and etoricoxib (two studies). 66,77 In four COX-2 inhibitor studies, participants were randomised to different doses of COX-2 (dose ranging studies),66,71,73,78 and in four they were randomised to receive the COX-2 at different times such as before or after surgery (timing studies). 70,74,76,79 Celecoxib and etoricoxib were both administered orally as single doses; celecoxib at a dose of 200 mg or 400 mg and etoricoxib at a dose of 120 mg or 180 mg. In all the studies of parecoxib, the drug was administered intravenously; lower dose studies used a single dose of 40 mg or 20 mg at 12-hourly intervals, higer dose studies used 40 mg at 6-hourly intervals or 40 mg at 12-hourly intervals (see Table 6 and Appendix 4).

There were 11 different NSAIDs: ketorolac (13 studies),60,80–84,88–90,96,103,105,108 diclofenac (nine studies),61,63,65,80,86,92,98,100,110 tenoxicam (four studies),87,93,97,107 ketoprofen (four studies),62,91,95,102 lornoxicam (four studies),59,94,95,109 ibuprofen (three studies),64,99,101 indometacin (one study),104 meloxicam (one study),106 naproxen (one study),85 dexketoprofen (one study),91 and piroxicam (one study). 87 There were five NSAID dose-ranging studies;83,90,96,103,105 one timing study;81 and four studies that compared different NSAIDs. 81,87,91,95

Ketorolac was administered using intravenous, intranasal and intramuscular methods and was predominantly given in multiple doses or by continuous infusion. A single dose (30 mg and 60 mg) of ketorolac was used in two studies. The multidose regimen for ketorolac varied widely (see Appendix 4); intravenous doses ranged from 15 mg at 6-hourly intervals to 60 mg starting dose plus 30 mg every 6 hours; intranasal doses ranged from 10 mg to 30 mg every 8 hours; and intramuscular doses ranged from 1.5 mg every 6 hours (plus a starting dose of 6 mg) to 30 mg every 6 hours (plus a starting dose of 60 mg). The continuous infusion dose also varied (see Appendix 4).

There was less variability within the remaining NSAIDs. Diclofenac was most commonly administered rectally, using a multiple dose regimen, but some studies also used oral, intravenous and intramuscular methods. The rectal doses ranged from 75 mg at 12-hour intervals to 100 mg at 8-hour intervals but were mainly at the lower dose (see Appendix 4) and did not vary widely. Tenoxicam was administered as a single dose in three studies, ranging from 20 to 40 mg and in the fourth study 40 mg every 24 hours. Administration was predominantly intravenous. Ketoprofen was administered using a multiple dose regimen of 50 mg every 6 hours or 100 mg every 12 hours or in one study a single 100-mg dose. Administration was intravenous and intramuscular. Lornoxicam was administered as a single dose of 8 mg, 8 mg every 8 hours, and 8 mg every 12 hours following an initial 16-mg dose. Administration was intravenous and intramuscular. Ibuprofen was administered as a 1600-mg dose before surgery and at 24 hours, 400 mg every 6 hours, and 500 mg every 8 hours. The remaining NSAIDs were investigated in single trials only. With the exception of dexketoprofen (50 mg every 12 hours), they were given as single doses: indometacin 75 mg; meloxicam 15 mg (rectal); naproxen 550 mg; and piroxicam (40 mg).

There were 12 studies of paracetamol and the prodrug propacetamol: seven of paracetamol61,64,65,111,113,115,116 and six of propacetamol16,62,63,112,114,116 (one of which compared propacetamol and paracetamol116). In all the studies, propacetamol was administered intravenously in doses of 2 g (which releases 1 g of paracetamol) every 6 hours. The paracetamol doses were 0.5 g every 4 hours (oral administration), 1.0 g every 6 hours (oral and rectal administration) and 1.3 g every 8 hours (rectal administration).

Study quality

All the included studies were RCTs with a placebo or active comparator. Full details of the validity assessment are presented in Appendix 5. The quality of reporting was variable between studies and across the criteria. Seven studies received the maximum possible score for each of the criteria: randomisation, allocation concealment, double blinding and description of flow of participants through the study. 63–65,72,79,94,109 The method of randomisation was described and adequate in 57% of studies and mentioned in the remaining studies (this was a minimum criterion for inclusion). Allocation concealment was the most poorly reported criterion: 60% of studies did not describe allocation concealment and 40% did so. No mention was made of blinding in 10% of studies; 48% mentioned double blinding and 42% described an adequate method of blinding. There was no description of flow of participants in 20% of studies, it was described but incomplete in 32% and described and adequate in 48%.

Assessment of effectiveness

Morphine consumption

There was considerable variability in the baseline morphine consumption: the simple mean in the placebo group was 45.26 mg (SD 22.23), and ranged from a minimum of 8.6 mg (SD 5.2) to a maximum of 141.5 mg (SD 74.9). There were five studies where the placebo group had a 24-hour morphine consumption of less than 20 mg67,89,94,109,111 and five with morphine consumption greater than 70 mg. 75,79,85,90,100 There was no apparent pattern amongst these studies in terms of age of participants, type of surgery, size of morphine bolus or length of lockout.

Mixed treatment comparison

A connected network for the four treatment classes was formed for cumulative 24-hour morphine consumption, allowing a comparison between all four classes to be made for this outcome (Figure 2). There were 56 studies in the network, which included comparisons with both placebo and other active treatments. Table 22 in Appendix 6 contains details of the specific studies included in the network. Two studies were excluded because they reported median morphine consumption,83,98 one because a variance was not available from the paper,91 and one because the number analysed was unclear. 58

FIGURE 2.

Network for 24-hour morphine consumption

In Figure 2the numbers represent the number of studies in which the two treatments were compared. If a study compared three treatments, it will be counted three times.

The pooled mean baseline morphine consumption was 37.43 mg (SE 2.0). There was a statistically significant reduction (5% level) in mean cumulative 24-hour morphine consumption with paracetamol, NSAIDs and COX-2 inhibitors compared to placebo; that is, the credibility intervals did not cross the line of no effect (zero) (see column 3 in Table 7). The difference ranged from a mean reduction of 6.34 mg for paracetamol to 10.92 mg for COX-2 inhibitors compared to placebo. The mean reduction compared to placebo for NSAIDs was similar to that of COX-2 inhibitors. Comparison of the active treatments shows that although NSAIDs and COX-2 inhibitors were both significantly better than paracetamol, there was no statistically significant difference between NSAIDs and COX-2 inhibitors (MD –0.74; 95% CrI –3.03 to 1.56).

Table 7 - Mixed treatment comparison meta-analysis 24-hour morphine consumption (pairwise comparisons)

The first treatment is the intervention and the second is the control. The negative mean difference indicates theintervention was more effective than the control treatment.

This is the pooled mean from a random effects model. The 45.26 mg quoted in the text above is a simple mean.

Comparison	Baseline morphine consumption: mean mg (SE)	Mean difference: mg (95% CrI)
Placebo	37.43 (2.00)a	0
Paracetamol vs placebo		–6.34 (–9.02 to –3.65)
NSAID vs placebo		–10.18 (–11.65 to –8.72)
COX-2 vs placebo		–10.92 (–12.77 to –9.08)
NSAID vs paracetamol		–3.85 (–6.80 to –0.89)
COX-2 vs paracetamol		–4.58 (–7.83 to –1.35)
COX-2 vs NSAID		–0.74 (–3.03 to 1.56)

The MTC analysis also produced data on the probability of each intervention being the most effective. Based on these data, COX-2 inhibitors had the highest probability of being the best (Table 8): there was a 74% chance that this drug class is the most effective treatment for reducing 24-hour morphine consumption. A probability of less than 95% indicated some uncertainty and reflected the finding of no statistically significant difference between COX-2 inhibitors and NSAIDs. The residual deviance (186) was larger than the number of study arms indicating that the model is not a perfect fit to the data.

Table 8 - Mixed treatment comparison meta-analysis of 24-hour morphine consumption (probability of being best treatment)

116 arms; residual deviance 186.

The second column shows the probability (p) that each treatment is the most effective one.

Treatment (n of studies)	p best (%)
Placebo (54)	0
Paracetamol (12)	0
NSAID (35)	26
COX-2 (15)	74

Sensitivity analyses

Baseline morphine consumption

Sensitivity analyses were run that included a covariate to adjust for baseline morphine consumption using the network of 56 studies. The analyses evaluated the impact of baseline morphine consumption on the treatment effect for each treatment compared to placebo, and calculated the treatment effect at a placebo morphine consumption level of 37.43 mg.

Three models were run that involved independent, exchangeable and common interaction assumptions. The number of trial arms, the DIC and the residual deviance (RD) are reported in Appendix 7, Table 31. The residual deviance shows that the models with a covariate are close to the number of arms in the study and are a good fit. The DIC is considerably lower for each of the models adjusting for baseline morphine consumption than the DIC for the model with no adjustment (Appendix 7, Table 31). There is little difference in the DIC between the three models adjusting for baseline morphine consumption. As the model with an exchangeable interaction assumption had the lowest DIC, the mean pairwise differences for this model are reported in Table 9 along with those for the model with no baseline adjustment. The covariate coefficients were all statistically significantly different from zero at a 5% level (Appendix 7, Table 31).

Table 9 - 24-hour morphine consumption adjusted and unadjusted for baseline morphine consumption

The first treatment is the intervention and the second is the control. The negative mean difference indicates that the intervention was more effective than the control treatment.

Comparison	Unadjusted mean difference, mg (95% CrI)	Adjusted (exchangeable interaction) mean difference, mg (95% CrI)
Paracetamol vs placebo	–6.34 (–9.02 to –3.65)	–8.68 (–11.43 to –5.94)
NSAID vs placebo	–10.18 (–11.65 to –8.72)	–9.45 (–10.90 to –8.01)
COX-2 vs placebo	–10.92 (–12.77 to –9.08)	–10.67 (–12.42 to –8.94)
NSAID vs paracetamol	–3.85 (–6.80 to –0.89)	–0.77 (–3.75 to 2.21)
COX-2 vs paracetamol	–4.58 (–7.83 to –1.35)	–1.99 (–5.24 to 1.24)
COX-2 vs NSAID	–0.74 (–3.03 to 1.56)	–1.22 (–3.43 to 1.00)

When the model was adjusted for baseline morphine consumption, the results were broadly similar to those of the unadjusted model indicating that the results were robust. COX-2 inhibitors still had the highest probability of being the most effective treatment for reducing 24-hour morphine consumption (Table 10). The main change was that whilst there was still a statistically significant reduction in morphine consumption with all three drugs compared to placebo, the mean difference for paracetamol compared to placebo was larger than in the unadjusted analysis. Any benefits of NSAIDs and COX-2 inhibitors over paracetamol were marginal and no longer statistically significant (see Table 9) and the probabilities for NSAIDs and paracetamol being best were now similar (Table 10).

Table 10 - Results from mixed treatment comparison analysis of 24-hour morphine consumption (probability of being best treatment)

Treatment (n of studies)	Unadjusted, p best (%)	Adjusted, p best (%)
Placebo (54)	0	0
Paracetamol (12)	0	10
NSAID (35)	26	11
COX-2 (15)	74	79

Individual drugs

The main purpose of the review was to compare the three classes of analgesic: paracetamol, NSAIDs and COX-2 inhibitors. An MTC was also conducted by individual drug to explore the appropriateness of the assumption made when grouping all types of NSAIDs together, all types of COX-2 inhibitors, and grouping paracetamol with propacetamol. This sensitivity analysis used the single outcome of 24-hour morphine consumption. A connected network was formed consisting of the same 56 studies that were in the main analysis for 24-hour morphine consumption. The model was also adjusted for baseline morphine consumption and hence the treatment effect results are calculated for a placebo morphine consumption of 37.43 mg. There were 15 individual drugs in the analysis plus placebo: two paracetamol (paracetamol and propacetamol), 10 NSAIDs and three COX-2 inhibitors. The residual deviance (130.2) was greater than the number of trial arms (120 arms) in the analysis indicating that the model is not a perfect fit to the data: this may be due to the large number of treatments in the analysis and the fact that four of the drugs were only included in one trial each.

The drug with the best effectiveness estimate was naproxen, although the probability of it being the most effective, 41%, is very low (Table 11). This reflects the degree to which the 95% credibility intervals of the drugs overlap, particularly for naproxen, diclofenac, indometacin, piroxicam, meloxicam and celecoxib.

Table 11 - Mixed treatment comparison analysis of 24-hour morphine consumption by individual drug

CrI, credibility interval; SD, standard deviation.

The second column shows the probability (p) that each treatment is the most effective one.

Treatment (n of studies)	Mean difference, mg (95% CrI)	p best (%)
Placebo (54)
Paracetamol
Paracetamol (7)	–7.96 (–11.59 to –4.35)	0
Propacetamol (6)	–8.73 (–12.24 to –5.20)	0
NSAIDs
Diclofenac (8)	–16.05 (–20.41 to –11.75)	27
Ibuprofen (3)	–7.30 (–13.36 to –1.27)	0
Indometacin (1)	–11.32 (–30.64 to 7.41)	24
Ketoprofen (3)	–8.11 (–11.52 to –4.78)	0
Ketorolac (12)	–10.58 (–13.55 to –7.60)	0
Lornoxicam (4)	–7.86 (–10.39 to –5.40)	0
Meloxicam (1)	–4.81 (–17.13 to 7.77)	2
Naproxen (1)	–16.73 (–23.48 to –9.78)	41
Piroxicam (1)	–8.05 (–17.99 to 1.80)	3
Tenoxicam (4)	–8.38 (–12.45 to –4.35)	0
COX-2 inhibitor
Celecoxib (3)	–12.55 (–15.74 to –9.33)	2
Etoricoxib (2)	–8.13 (–11.50 to –4.79)	0
Parecoxib (10)	–10.94 (–13.64 to – 8.22)	0

The results indicate that the decision to group together propacetamol and paracetamol in one class seems to have been reasonable: the mean difference in morphine consumption was similar for the two drugs and the credibility intervals overlapped (Table 11). This would be expected given that propacetamol is a prodrug of paracetamol. Similarly, the decision to group together COX-2 inhibitors is also shown to be reasonable: the mean reduction in morphine consumption ranged from 8.13 to 12.55 mg and the credibility intervals for celecoxib, etoricoxib and parecoxib overlapped (Table 11). The performance of individual NSAIDs was more variable than within the other two classes. For four of the drugs the analysis is based on single trials and for three of these there was no statistically significant difference between the drug and placebo. The reduction in morphine consumption compared to placebo ranged from 4.81 to 16.73 mg for individual NSAIDs and the credibility interval (CrI) for some NSAIDs barely overlapped. These findings suggest that there may be variability in the effectiveness of individual NSAIDs.

Quality

A sensitivity analysis was conducted to evaluate the impact of study quality on the results, as defined in Chapter 2 (Methods). This was done in two ways, both of which also adjusted for baseline morphine consumption. Firstly, the MTC analysis was run on the subset of studies that were recorded as good quality, i.e. studies reporting an adequate method of blinding (see Appendix 7, Table 33, for results). Secondly, a model was run using all of the studies and adding a dummy variable to account for study quality. When the dummy variable was 0 this represented a quality study. Three assumptions were again tested regarding the interaction of the dummy variable with the treatments. None of the models adjusting for study quality are an improvement over the model adjusted for baseline morphine consumption alone based on the DIC (Appendix 7, Table 32). The exchangeable interaction model had the lowest DIC (Appendix 7, Table 32) and the results from this model are reported in Tables 12 and 13. The covariate coefficients were not statistically significantly different from zero at a 5% level (Appendix 7, Table 32).

Table 12 - 24-hour morphine consumption adjusted for quality and baseline morphine consumption (pairwise comparisons)

The first treatment is the intervention and the second is the control. The negative mean difference indicates the intervention was more effective than the control treatment.

Comparison	Unadjusted results: mean difference, mg (95% CrI)	Adjusted for quality and baseline morphine consumption: mean difference, mg (95% CrI)
Placebo
Paracetamol vs placebo	–6.34 (–9.02 to –3.65)	–9.01 (–12.01 to –6.01)
NSAID vs placebo	–10.18 (–11.65 to –8.72)	–10.17 (–12.37 to –7.99)
COX-2 vs placebo	–10.92 (–12.77 to –9.08)	–12.03 (–15.73 to –8.46)
NSAID vs paracetamol	–3.85 (–6.80 to –0.89)	–1.17 (–4.31 to 1.98)
COX-2 vs paracetamol	–4.58 (–7.83 to –1.35)	–3.02 (–7.24 to 1.02)
COX-2 vs NSAID	–0.74 (–3.03 to 1.56)	–1.86 (–5.34 to 1.39)

Table 13 - 24-hour morphine consumption adjusted for quality and baseline morphine consumption (probability of being most effective treatment)

Treatment (n of studies)	Unadjusted, p best (%)	Adjusted for quality and baseline morphine consumption, p best (%)
Placebo (54)	0	0
Paracetamol (12)	0	5
NSAID (35)	29	11
COX-2 (15)	71	84

The results were broadly similar to those of the unadjusted model indicating that the results from the main analysis are reasonably robust (Tables 12 and 13). Based on the pairwise comparisons (Table 12) there was still a statistically significant reduction in morphine consumption with all three drugs compared to placebo, though the mean difference for paracetamol compared to placebo was larger than in the unadjusted analysis. The difference between NSAIDs and COX-2 inhibitors remained small and not statistically significant, and the benefits of NSAIDs and COX-2 inhibitors over paracetamol were marginal and no longer statistically significant. These differences were apparent in the first sensitivity analysis using baseline morphine consumption only, therefore the impact of quality was minimal.

Direct comparisons

Data on cumulative mean morphine consumption were available from five studies that directly compared paracetamol and NSAIDs,61–65 and for one study that directly compared COX-2 inhibitors and NSAIDs. 60 Cumulative 24-hour morphine consumption was statistically significantly lower with NSAIDs compared to paracetamol, with a mean reduction of 9.76 mg (95% CI –18.69 to –0.82) (Figure 3). However, there was evidence of moderate statistical heterogeneity (I² = 49%).

FIGURE 3.

Cumulative 24-hour morphine consumption (non-steroidal anti-inflammatory drug vs paracetamol).

Based on a single study,60 there was no statistically significant difference in cumulative 24-hour morphine consumption between COX-2 inhibitors and NSAIDs (MD –1.40; 95% CI –7.60 to 4.80) (Figure 4).

FIGURE 4.

Cumulative 24-hour morphine consumption (cyclo-oxgenase 2 inhibitor vs non-steroidal anti-inflammatory drugs).

Morphine-related adverse effects

Nausea and postoperative nausea and vomiting (PONV)

Mixed treatment comparison

Studies reporting postoperative nausea alone were pooled with studies that reported nausea and/or vomiting (PONV) as a combined outcome. A connected network for the four classes of drugs was formed, which consisted of 43 trials (Figure 5). Details of the studies included in the network are provided in Appendix 6, Table 23.

FIGURE 5.

Network for nausea and postoperative nausea and vomiting.

The pairwise ORs and the 95% CrI are reported in Table 14, where the first treatment in the first column is the intervention and the second is the control. An OR of less than 1.0 indicates that the intervention performed better than the control.

Table 14 - Nausea and postoperative nausea and vomiting (pairwise comparisons)

The first treatment in the first column is the intervention and the second is the control. An OR less than 1 indicates that the intervention performed better than the control.

Comparison	Pairwise odds ratio (OR) and 95% CrI
Paracetamol vs placebo	1.00 (0.60 to 1.53)
NSAID vs placebo	0.70 (0.53 to 0.88)
COX-2 vs placebo	0.88 (0.61 to 1.25)
NSAID vs paracetamol	0.74 (0.44 to 1.17)
COX-2 vs paracetamol	0.93 (0.51 to 1.63)
COX-2 vs NSAID	1.28 (0.81 to 1.97)

Non-steroidal anti-inflammatory drugs performed best for this outcome compared to placebo, with an odds ratio of 0.70, and this was the only comparison that was statistically significant. COX-2 inhibitors were slightly less effective than NSAIDs, and there was almost no difference between paracetamol and placebo (Table 14). These results are reflected in the probability of NSAIDs being the most effective treatment for reducing nausea or PONV: there was a 78% chance that this was the most effective treatment for this outcome (Table 15). In total, 88 trial arms were included in the analysis, of which 86 had at least one outcome event. The residual deviance (96.64) was similar to the number of arms that had at least one event, which indicates a good model fit.

Table 15 - Nausea and postoperative nausea and vomiting (probability of being best treatment)

86 armsa; residual deviance 96.64.

Refers to the number of arms with at least one event.

The second column shows the probability (p) that each treatment is the most effective one.

Treatment (n of studies)	p best (%)
Placebo (41)	0
Paracetamol (9)	7
NSAID (27)	78
COX-2 (11)	15

Direct comparisons

Data on nausea or PONV were available from four studies that directly compared paracetamol and NSAID. 62–65 Data from the sole study reporting postoperative nausea alone,64 was pooled with those from the three studies that reported PONV. 62,63,65 NSAIDs were slightly more effective than paracetamol in reducing nausea and PONV [risk ratio (RR) 0.78]; however, this was not statistically significant (95% CI 0.51 to 1.20). There was no statistical heterogeneity (I² = 0%) (Figure 6).

FIGURE 6.

Nausea and postoperative nausea and vomiting (non-steroidal anti-inflammatory drugs vs paracetamol).

Sensitivity analysis

As a sensitivity analysis, an MTC was undertaken for nausea alone, vomiting alone and PONV alone, and the results were similar. In each of these separate analyses NSAIDs had the highest probability of being the most effective treatment (ranging from 50% to 84%) (Appendix 8, Table 34). There were differences in the size of the OR for some of the comparisons, and the benefit with NSAIDs compared to placebo was statistically significant for PONV but not nausea alone or vomiting alone (Appendix 8, Table 35).

Sedation

Mixed treatment comparison

A connected network for the four classes of drugs was formed for sedation, which consisted of 19 studies (Figure 7). Details of the studies included in the network are provided in Appendix 6, Table 25.

FIGURE 7.

Network for sedation.

The pairwise ORs (95% CrI) are reported in Table 16. There was no statistically significant difference between any intervention and control in reducing morphine-related sedation: there was a trend towards paracetamol performing more poorly than placebo, and COX-2 inhibitors more poorly than NSAIDs, with wide CrIs indicating considerable uncertainty, and NSAIDs and COX-2 inhibitors performing better than placebo and paracetamol.

Table 16 - Sedation (pairwise comparisons)

The first treatment in the first column is the intervention and the second is the control. An OR less than 1 indicates that the intervention has performed better than the control.

Comparison	Pairwise odds ratio (OR) and 95% CrI
Paracetamol vs placebo	1.62 (0.32 to 5.02)
NSAID vs placebo	0.53 (0.20 to 1.01)
COX-2 vs placebo	0.63 (0.18 to 1.49)
NSAID vs paracetamol	0.51 (0.08 to 1.63)
COX-2 vs paracetamol	0.63 (0.07 to 2.33)
COX-2 vs NSAID	1.40 (0.30 to 4.31)

Non-steroidal anti-inflammatory drugs performed best for this outcome: there was a 53% chance that NSAIDs are the most effective treatment for reducing sedation (Table 17). This is a low probability, which reflects the considerable overlap in the CrIs for the treatment effect estimates (Table 16).

Table 17 - Sedation (probability of being best treatment)

31 armsa; residual deviance 41.44.

Refers to the number of arms with at least one event.

The second column shows the probability (p) that each treatment is the most effective one.

Treatment (n of studies)	p best (%)
Placebo (19)	0
Paracetamol (4)	6
NSAID (12)	53
COX-2 (9)	41

In total, 40 arms were included in the analysis, of which 31 had at least one outcome event. The residual deviance was 41.44. This was similar to the number of data points with at least one event (31), therefore demonstrating a good fit of the model to the data.

Direct comparisons

Data were available on sedation from two studies that directly compared paracetamol and NSAIDs. 62,63 There was a trend towards NSAIDs being more effective than paracetamol in reducing sedation (RR 0.35); however, this was not statistically significant (95% CI 0.04 to 3.00) (Figure 8). Statistical heterogeneity was low (I² = 17%).

FIGURE 8.

Sedation (non-steroidal anti-inflammatory drugs vs paracetamol).

Other morphine-related side effects

In addition to the main morphine-related outcomes reported above, the effect of adding any of the three classes of non-opioid analgesics to PCA morphine, on reduction of respiratory depression, urinary retention, pruritus, bowel dysfunction and dizziness were also investigated. The full results of these analyses are reported in Appendix 8, and a summary is provided in Table 18. When taken together, these results present a complex picture of which drug was the most effective in reducing morphine-related side effects. Based on the pairwise comparisons, there were no statistically significant differences between intervention and control with the exception of pruritus, where there was a statistically significant improvement with paracetamol and NSAIDs compared to placebo (Appendix 8, Table 40). This is reflected in the low probabilities for the outcomes, which ranged from 43% to 73% (Table 18); a probability of being best of less than 95% indicates no statistically significant difference at a 95% level between the best treatment and at least one comparator.

Table 18 - Summary of probability of being the most effective treatment for reduction of secondary outcomes

✓ = intervention with the highest probability of being the most effective intervention (probability).

Outcome	Placebo	Paracetamol	NSAID	COX-2	Comments
Respiratory depression			✓ (43%)		One COX-2 study in network
Urinary retention				✓ (61%)
Pruritus		✓ (73%)
Bowel dysfunction		✓ (58%)			No COX-2 studies in network
Dizziness				✓ (56%)

Adverse effects of non-opioid analgesics

As would be expected it was not possible to form a network for the analgesic-related adverse effects. The most commonly reported adverse effects were those associated with NSAIDs. Studies reported adverse events for the first 24–48 hours after surgery.

Bleeding

The primary analgesic-related adverse effect of interest was surgical bleeding. This outcome was not reported in any of the paracetamol studies; and although it was reported in a single study comparing COX-2 inhibitor to placebo, there were zero events in each group. Five of the remaining six studies,69,72,78–80,89 all comparing an NSAID to placebo, reported zero events in each of the placebo arms therefore a pooled estimate could not be calculated. In addition, this outcome was defined differently across studies and the number of events overall was small. In the NSAID group 2.4% of participants experienced surgery-related bleeding, compared to 0.4% in the placebo group (Table 20).

Table 20 - Surgery-related bleeding problems

Study	Definition of bleeding event	Placebo: number of events/number analysed	NSAID: number of events/number analysed	COX-2: number of events/number analysed
Balestrieri 199781	Clinically significant bleeding	0/82	4/166
Cassinelli 200884	Epidural hematoma	1/12	0/13
Gillies 198790	Postoperative bleeding	0/18	1/39
Hanna 200391	Postoperative haemorrhage	0/54	1/114
Hodsman 198792	Reoperation due to bleeding	0/32	2/33
Plummer 1996101	Intraoperative bleeding	0/57	2/57
Tang 200278	Bleeding problems	0/18		0/37
	Total	1/273 (0.4%)	10/422 (2.4%)	0/37 (0%)

It was also not possible to construct a network for gastrointestinal bleed. This outcome was not reported in any paracetamol studies. For the three studies available with this outcome there were zero events in four of the six arms. 66,79,89 Among participants in the NSAID group, 2.3% experienced GI bleeding compared to 0% with placebo (Table 21).

Table 21 - Gastrointestinal bleeding

Study	Definition of bleeding event	Placebo: number of events/number analysed	NSAID: number of events/number analysed	COX-2: number of events/number analysed
Hanna 200391	GI bleeding	0/54	3/114
Plummer 1996101	GI haemorrhage	0/57	1/57
Siddiqui 200877	GI bleeding	0/100		0/100
	Total	0/211 (0%)	4/171 (2.3%)	0/100 (0%)

Oliguria and renal failure

Six studies (535 participants) reported on renal dysfunction; five compared NSAID to placebo48,72,79,85,90 and one compared COX-2 to placebo. 66 There was a single event, described as transient oliguric renal failure, in a patient receiving NSAID. 97

Four studies reported on oliguria;69,91,95,98 all comparing NSAID to placebo. There was no statistically significant difference between NSAID and placebo, though there was a trend towards an increase in oliguria with NSAID (Figure 9).

FIGURE 9.

Oliguria (non-steroidal anti-inflammatory drugs vs placebo).

Principal findings

All three classes of non-opioid analgesic reduced mean cumulative morphine consumption. From the main analysis, PCA morphine with COX-2 inhibitors reduced morphine consumption by 10.9 mg, followed by NSAIDs with a 10.2 mg reduction and paracetamol with a 6.3 mg reduction compared to PCA morphine alone; these all had narrow CrIs (unadjusted results). Based on the average baseline morphine consumption of 37.43 mg, this equates to a 29.2% (COX-2 inhibitors), 27.2% (NSAIDs) and 16.9% (paracetamol) reduction in morphine consumption in the 24 hours immediately following surgery. However, from a clinical perspective, the actual reduction in morphine consumption seems modest and arguably of questionable clinical significance.

Although NSAIDs and COX-2 inhibitors were both superior to paracetamol in the main analysis, the reduction in morphine consumption with COX-2 inhibitors compared to NSAIDs was marginal, with a mean difference of less than 1 mg of morphine (mean difference –0.74 mg; 95% CrI –3.03 to 1.56) which is not of clinical significance. This is reflected in the finding that, although COX-2 inhibitors had the highest probability of being most effective, this probability (74%) was lower than 95%, thereby indicating uncertainty. The sensitivity analyses for 24-hour morphine consumption, taking into account study quality and baseline morphine consumption, showed the results of the main analysis to be robust. The analysis of individual drugs (as opposed to drug class) suggested that it was reasonable to group the drugs into three classes, though there appeared to be possible inconsistency across different NSAIDs. The sensitivity analyses are discussed in further detail below (see Strengths and limitations of the assessment).

Non-steroidal anti-inflammatory drugs had the highest probability (78%) of reducing nausea or PONV. There was a statistically significant improvement for this outcome with NSAIDs added to PCA morphine compared to PCA morphine alone (OR 0.7; 95% CrI 0.53 to 0.88). However, the credibility intervals for the comparisons between NSAIDs and paracetamol and NSAIDs and COX-2 inhibitors covered the possibility of an increase in nausea and vomiting as well as a decrease with NSAIDs. For example, the OR for NSAIDs compared to paracetamol was 0.74 (indicating a reduction with NSAIDs) but there was a 95% probability that this would fall between a reduction (0.44) and a small increase in nausea and vomiting (1.17). This is reflected in the result that the probability of NSAIDs being best, 78%, was less than 95%. Similarly, for sedation NSAIDs had the highest probability of being the most effective at reducing sedation but the probability of it being best was low, 53%, reflecting the CrIs for the pairwise comparisons between NSAIDs and the other interventions, which allowed for the possibility of an increase in sedation as well as a decrease.

When secondary morphine-related outcomes were considered, the drug that had the highest probability of being the most effective varied by outcome. NSAIDs had the highest probability of being the best in reducing respiratory depression, paracetamol had the highest probability of reducing pruritus and bowel dysfunction, and COX-2 inhibitors had the highest probability of being best in reducing urinary retention and dizziness. The probabilities that these drugs were best were low. As with the primary morphine-related adverse effects, generally, the CrIs for many of the individual pairwise comparisons were broad and covered the possibility of an increase in the particular adverse event, as well as a reduction, for the drug with the highest probability of being best.

Any benefits in reduction of morphine-related adverse effects must be balanced against any potential adverse effects associated with the non-opioid analgesics. The review could only explore this in a limited way. Given the different adverse event profiles of the three drug classes, it was not possible to form a network to carry out a comparison similar to that undertaken for the other outcomes. Many studies did not report adverse effects associated with the analgesics. As would be expected, few studies of paracetamol reported adverse events because at therapeutic doses such effects are rare. Most of the adverse events reported were from NSAID studies. Approximately 2% of study participants treated with NSAIDs experienced some type of bleeding event, and a similar proportion experienced GI bleeding. Oliguria was reported for 4% of NSAID patients, and there was one case of transient renal failure. However, it needs to be kept in mind that these figures are based on trials with a selected population and therefore may underestimate the number of events that might occur in a general population. In addition, the included studies were powered (where reported) to detect a difference in morphine consumption and not differences in analgesic-related adverse effects.

Consistency with direct comparisons

The results from the MTC are consistent with the direct evidence synthesis and the direct evidence available from previous reviews. Two previous reviews comparing paracetamol to placebo found that while paracetamol combined with PCA morphine reduced 24-hour morphine consumption compared to PCA morphine alone, there was no benefit in terms of a reduction in morphine-related adverse effects. 21,23 The reduction in morphine consumption with paracetamol in the current review was slightly smaller than the two earlier reviews but the confidence intervals from the three reviews have a good overlap. A previous review found that there was a statistically significant reduction in morphine consumption when NSAIDs and COX-2 inhibitors were added to PCA morphine compared to PCA morphine alone. 23 There was a reduction in PONV and sedation with NSAIDs but there was no statistically significant difference in any morphine-related adverse effects with COX-2 inhibitors.

It is not surprising that we did not find any studies directly comparing all three non-opioid analgesics. There were also few studies available that directly compared any two of the three analgesics. There were five comparing NSAIDs and paracetamol, and a single study comparing a COX-2 inhibitor to an NSAID. We did not find any studies comparing a COX-2 inhibitor and paracetamol. The results from the synthesis of the direct comparison studies were consistent with the results of the MTC. There was a statistically significant reduction in morphine consumption and a trend towards improvement in nausea and vomiting and sedation with NSAID compared to paracetamol, which was not statistically significant. The single study comparing a COX-2 inhibitor and an NSAID reported no statistically significant difference in 24-hour morphine consumption; data on morphine-related side effects were not available.

Strengths and limitations of the assessment

Previous reviews have investigated the effectiveness of paracetamol, NSAIDs and COX-2 inhibitors compared to placebo19,21–23,106 but we are not aware of any previous systematic reviews that have investigated the relative effectiveness of these non-opioid analgesics using appropriate statistical methods. By using currently developing methods of synthesis of direct and indirect evidence to investigate the relative effectiveness of the drug classes, the current review extends the work undertaken in a previous systematic review.

As expected, we found limited direct evidence comparing the three non-opioid analgesics. Therefore, the MTC allowed us to maximise the usefulness of the available network of evidence. This review has also provided an opportunity to update the evidence on multimodal analgesia following major surgery. Twenty new trials were included and we were able to exclude trials by Scott S Reuben from the analysis, which were based on falsified data, as well as COX-2 inhibitors that are no longer licensed for use.

A key factor to consider in evaluating the strengths and limitations of the assessment undertaken is whether the assumption that there were no systematic differences between the trials that investigated each analgesic (exchangeability) was reasonable. Based on a qualitative examination of the trials we believe this was a reasonable assumption: the inclusion criteria for the review were narrow and all the participants were adults undergoing major surgery and receiving PCA morphine in the 24 hours following surgery. We also used a random effects model to allow for any possible heterogeneity. However, this approach does not explain heterogeneity and we found considerable variability across the trials in baseline morphine consumption (based on placebo control group), which had not been anticipated. This variation may be due to differences in surgery, the exact regimen under which morphine was administered or study population such as ethnicity or age. If an interaction did exist between drug class and morphine consumption then the main results could be misleading as the exchangeability assumption would not be met. An interaction could arise, for example, where a particular drug class was used in trials where it was anticipated that pain levels could be high (and therefore morphine consumption high) due to the severity of pain anticipated. We therefore conducted a post hoc sensitivity analysis to explore this further. This replaced the originally planned sensitivity analysis based on type of surgery.

The adjustment of the 24-hour morphine consumption model, for baseline morphine consumption, did not alter the results in terms of which drug class had the highest probability of being most effective. The treatment effect estimates of NSAIDs and paracetamol became closer but COX-2 inhibitors still had the highest treatment effect estimate with a similar probability of being the most effective, 79%. This adjusted analysis did show a greater reduction in morphine consumption with paracetamol compared to placebo, and the differences between the active interventions in the pairwise comparisons were no longer statistically significant. The reduction in morphine consumption with NSAIDs and COX-2 inhibitors compared to paracetamol were smaller and non-significant, though the direction of the effect continued to favour these two drugs over paracetamol. This sensitivity analysis showed the results of the main analysis to be robust to variation in baseline morphine consumption.

Although the sensitivity analysis we undertook does support the robustness of the results of the main analysis, it was only undertaken as an exploratory analysis and the results should not be considered definitive. The feasibility of incorporating covariates in a mixed treatment comparison has been demonstrated,38 though the approach is not in common use and the methods are continually being developed. First, the analysis is based on summary data and the comparisons are not based on randomised groups as in a trial. There may be unknown confounding factors that influence the relationship between the covariates used and 24-hour morphine consumption. This is a limitation of all meta-analyses based on aggregate data and can only be resolved through the analysis of independent patient data from the included studies. Second, because morphine consumption is both an outcome and a covariate in this analysis, there is a risk of regression to the mean:117,118 the regression model made the assumption that there was no uncertainty in the measurement of baseline morphine consumption and the baseline morphine consumption, derived from the placebo control group, also formed part of the outcome (morphine consumption). 117 Third, two of the studies included in the model did not have a placebo control group; therefore, it was necessary to make an estimate of the baseline morphine consumption for these two trials.

The third point above contributes to the difficulty in accounting for regression to the mean in the model. Paracetamol was the comparator in the two trials without placebo. If placebo had been included in these trials, then the difference in morphine consumption between placebo and paracetamol could be calculated using estimates of the paracetamol treatment effect difference compared to placebo and the paracetamol covariate interaction. These were estimated by running the model without these two studies. These estimates were considered likely to be reasonably good because they were estimated using 54 trials that included placebo out of a total of 56 trials, which included 10 trials comparing paracetamol with placebo. That is, most of the data available were included and adequate paracetamol versus placebo data were available. However, the two trials were excluded in deriving these estimates and ideally the baselines for these two studies would be determined within the model including all trials. The best way to address the problem of trials not having a placebo control group is an area of ongoing work. 38

A final point to consider in the interpretation of the adjusted results is that the results presented are the mean values for the covariate, i.e. for the overall mean value of morphine consumption. This allows comparison of the results with those for the base-case model that did not adjust for baseline morphine consumption. However, the baseline morphine covariate was statistically significant, indicating that the higher the expected baseline morphine consumption, the greater the reduction in morphine will be. Effect differences at different levels of baseline morphine consumption have not been evaluated.

The main analysis was based on the assumption that it was reasonable to group individual drugs into classes. In many respects this was necessary to address the decision problem presented. There was variability between the three drug classes in the number of drugs investigated, and for some of the individual drugs there was variability in total dose, methods of administration timing and number of doses. In particular there were a large number of different NSAIDs. By pooling these as one class the assumption was made that the different NSAIDs used equivalent and optimal doses, which may not be the case. Even within some of the NSAIDs, particularly ketorolac, there was considerable variability. This was less of an issue with the COX-2 inhibitors and paracetamol. There were only three COX-2 inhibitors and the paracetamol class was made up of paracetamol and propacetamol. There was also less variability in dosage. The sensitivity analysis by individual drug (also adjusted for morphine consumption) suggested variability between NSAIDs in the size of the reduction in morphine consumption. The mean reduction in morphine consumption ranged from 4.1 mg for meloxicam to 16.7 mg for naproxen, and the CrIs for some NSAIDs barely overlapped. Due to time constraints we were not able to investigate whether this also applied to the morphine-related adverse effects and this would benefit from further investigation, though such an analysis may be constrained by the network available. The treatment effect across COX-2 inhibitors was consistent, indicating that the decision to treat them as a class was reasonable: the mean reduction in morphine consumption ranged from 8.1 mg to 12.6 mg and the CrIs for celecoxib, etoricoxib and celecoxib overlapped. Similarly the decision to group propacetamol and paracetamol was reasonable: the mean reduction in morphine consumption was 8.0 mg for paracetamol and 8.7 mg for propacetamol and there was good overlap in the CrIs.

Taking the evidence as a whole, a key finding was the disparity between the results for morphine consumption and morphine-related adverse effects. There was robust evidence of a reduction in morphine consumption with the addition of any of the non-opioid analgesics to PCA morphine but the evidence for reduction in morphine-related adverse effects was more equivocal. This dissonance between morphine consumption and related adverse effects has been noted in previous reviews. 20,23,107 A number of reasons have been suggested. One possibility is that the size of the reduction in morphine consumption was not sufficient to decrease morphine-related adverse effects. 23 The poor quality of adverse event data in many trials and the possibility that the trials are underpowered to detect a reduction in adverse events may be other factors. 107 There is a possibility that the analyses for morphine-related adverse effects were underpowered as the trials included in the review were generally powered to detect a difference in morphine consumption or, in a few instances, pain. However, against this, there was a reasonable body of evidence available for nausea and vomiting at least. Given that morphine consumption alone is not a clinically meaningful outcome, future trials should use one or more morphine-related adverse effects as the primary outcome and power calculations for the trial should be based on these outcomes and not morphine consumption alone. Also, due to time constraints we limited our sensitivity analyses to the outcome for which we had the most substantial set of data (24-hour morphine consumption) and therefore most complete network. There would be value in exploring whether taking baseline morphine consumption into account alters the results for morphine-related adverse effects. Furthermore, time constraints prevented us from evaluating the individual drug treatment effects for the morphine-related adverse effects. Given the variability in the treatment effects of individual NSAIDs in reducing morphine consumption, it is possible that the difference in the mix of individual drugs between the analyses (the relative number of studies per individual drug) may partly explain this dissonance in the results. This may warrant further investigation.

Finally, this review focused specifically on the morphine-sparing effects of the three analgesics. For the purposes of the review, the assumption was made that, because patients were receiving PCA morphine, optimum analgesia should be maintained and pain control should be the same in all arms of a trial. This does not take into account any differences there may be in the synergistic action between morphine and the three drug classes which may result in differences in pain control. Regardless of any reduction in morphine consumption, the improvement of analgesia post-surgery through the addition of a non-opioid to PCA morphine post-surgery is of clinical importance. This is likely to be of value to the patient beyond the immediate 24 hours following surgery and is itself an important research question.

Implications for service provision

Non-steroidal anti-inflammatory drugs, COX-2 inhibitors and paracetamol reduced PCA morphine consumption by 6.3 mg to 10.9 mg, compared to placebo, in the first 24 hours following major surgery. However, the reduction was modest for all three drug classes and probably of limited clinical significance. The difference between NSAIDs and COX-2 inhibitors was marginal and not statistically significant. Although NSAIDs and COX-2 inhibitors were both more effective than paracetamol the differences in morphine consumption compared to paracetamol were small, especially when baseline morphine consumption was taken into consideration: the adjusted results suggest a mean difference of less than 2 mg of morphine when each of the drug classes was compared to each other.

Non-steroidal anti-inflammatory drugs were ranked best for reducing nausea and vomiting and sedation, and for the former there was a statistically significant improvement over placebo. However, the confidence intervals for the difference between NSAIDs and paracetamol and COX-2 inhibitors for these outcomes indicate the possibility of an increase in incidence of these outcomes as well as a decrease. Although NSAIDs were marginally better at reducing the primary morphine-related adverse effects of interest, the results do not strongly favour one class of non-opioid analgesic. Paracetamol was ranked lower than NSAIDs and COX-2 inhibitors for each of the primary outcomes, therefore NSAIDs or COX-2 inhibitors might arguably be considered a preferential option. However, any benefit provided by these analgesics in terms of morphine sparing needs to be balanced against any adverse effects related to the analgesics themselves. There was a small number of surgical bleeding, GI bleeding events and oliguria for participants treated with an NSAID.

Taking the evidence as a whole, the uncertainty suggested by the size of the probabilities of being most effective, the small reductions in morphine consumption, and the wide CIs for the adverse effects outcomes, there does not appear to be a strong case for suggesting routine addition of any of the three non-opioids to PCA morphine in the 24 hours immediately after surgery. In addition, there does not appear to be a strong case for favouring one drug class above the others.

Suggested research priorities

There would be value in extending the analyses undertaken in this review to explore whether taking baseline morphine consumption into account alters the results for morphine-related adverse effects. Given the evidence that there may be variability in the effects of individual NSAIDs, further evidence synthesis on the NSAID data would be helpful, in particular exploration of any variation in the impact on morphine-related adverse effects.

There does not appear to be a compelling case for a further trial comparing these three analgesic classes, given the overlap between the non-opioid analgesics and their different benefits. It is likely that such a trial would have to be very large to detect statistically significant differences between the treatments and any differences might not be clinically meaningful. However, any future trials testing new analgesics in conjunction with morphine should focus on morphine-related adverse effects, ensuring that the power calculation is based on key morphine-related adverse effects rather than morphine consumption.

Contribution of authors

Catriona McDaid (Research Fellow) was involved in writing the protocol, study selection, data extraction, quality assessment, data analysis and report writing. Emma Maund (Research Fellow) was involved in writing the protocol, study selection, data extraction, quality assessment, data analysis and report writing. Stephen Rice (Research Fellow) contributed to the protocol, conducted the MTC analysis and was involved in writing sections of the report. Kath Wright (Information Specialist) devised the research strategy, carried out the literature searches and wrote the search methodology sections of the report. Brian Jenkins (Senior Lecturer in Anaesthetics) provided clinical input throughout the project, and commented on the protocol and drafts of the report. Nerys Woolacott (Senior Research Fellow) provided input at all stages of the review and commented on the protocol and drafts of the report.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.