Data were extracted from the electronic medical records of 427 patients who underwent open (n = 300) and laparoscopic (n = 127) upper gastrointestinal surgery at the Royal North Shore Hospital between November 2015 and October 2020. Most patients were ASA physical status 2 or 3 across all cohorts (Table 1).

Table 1 Patient characteristics

Neuraxial techniques for postoperative analgesia were more common in open surgical approaches, thoracic epidural (58%) and intrathecal morphine (21%), versus non-neuraxial (21%). By contrast, 85% of patients who underwent laparoscopic surgery had non-neuraxial analgesic management. The predominance of patients undergoing pancreatoduodenectomy (84%) or oesophagectomy (96%) received thoracic epidural analgesia. Most patients (87%) undergoing less extensive procedures, such as hiatus hernia repairs, were more likely to receive non-neuraxial analgesia as their primary analgesic modality, and rarely received intrathecal morphine (8%) or thoracic epidural analgesia (5%) (Table 1).

For thoracic epidural analgesia, 0.1% bupivacaine was co-administered with 2 mcg/mL of fentanyl and adrenaline administered as an adjunct. Epidural infusion used a combination of programmed intermittent epidural boluses and patient controlled epidural analgesia. For spinal analgesia, intrathecal morphine was co-administered with local anaesthetic without other additives.

Intraoperative oral morphine equivalent dose was highest in the intrathecal morphine group, median, interquartile range (IQR) 98 (60–120) mg, compared to the non-neuraxial 50 (IQR 30–100) mg and thoracic epidural 20 (IQR 20–60) mg analgesia groups (Table 2). More than 90% of patients who received a neuraxial technique had a planned high dependency unit admission postoperatively (92% intrathecal morphine, 93% thoracic epidural analgesia) compared to 57% of patients in the non-neuraxial cohort.

Table 2 Perioperative outcomes

Inability to site an epidural catheter occurred in 11 patients planned for thoracic epidural analgesia, 9 of whom subsequently successfully received an intrathecal morphine injection (Table 2). More than one attempt at siting the thoracic epidural catheter occurred in 28% of cases. All patients planned for an intrathecal morphine injection had such, however, 19% of patients required more than one attempt to access the intrathecal space. The median (range) dose of intrathecal morphine was 350 (150–500) mcg. Epidural analgesia was ceased earlier than planned in 32% of cases, mainly because of inadequate epidural function (25%). Analgesic adjuncts, including non-steroidal anti-inflammatory agents (COX-1 or -2 inhibitors), ketamine, regional blocks or elastomeric local anaesthetic wound infusion devices (Painbusters™), were utilised more frequently in the non-neuraxial and intrathecal morphine groups than in the thoracic epidural group (Table 2).

Primary outcome: oral morphine equivalent doseNon-neuraxial versus intrathecal groups

Simulation of patient allocation from the non-neuraxial group to the intrathecal morphine group predicts no overall clinically significant difference in oral morphine consumption for both laparoscopic (− 5.0 mg [− 18.3 to 9.3]) and open (− 6.0 mg [− 31.9 to 22.4]) surgeries (Table 3 and Additional file 1a Figure S18), reflecting a probability of reducing the oral morphine equivalent dose requirement by at least 10 mg (Pr(Diff \(\le -10 \mathrm)\)) of 22% and 38%, respectively.

Table 3 Predicted median oral morphine equivalent dose (mg)

However, specific to postoperative day 0, the probability of a clinically relevant reduction in opioid consumption for laparoscopic surgery was 92% (− 20.4 mg [− 35.3 to − 5.9], Pr(Diff \(\ge -\) 10 mg) = 0.92) and 87% for open surgery (− 20.5 mg [− 40.0 to − 3.6], Pr(Diff \(\ge -\) 10 mg) = 0.87). By postoperative day 2 and 3, the trend appeared to have reversed in favour of non-neuraxial analgesia, although the median doses and their 95% credible intervals remained clinically not significant (Table 3).

Similar patterns were observed in the subgroup analysis by surgical types (Table 3 and Additional file 1a Figure S19). There was no overall clinically relevant reduction in oral morphine equivalent consumption if one were to have received intrathecal morphine instead of non-neuraxial analgesia, being − 6.2 mg (− 31.1 to 20.0, Pr(Diff \(\le -10 \mathrm\)) = 0.38), − 2.1 mg (− 51.3 to 47.5, Pr(Diff \(\le -10\mathrm\)) = 0.36) and − 4.6 mg (− 18.6 to 9.5, Pr(Diff \(\le - 10 \mathrm\)) = 0.22) for type 1, 2 and 3 surgeries, respectively. Clinically relevant reduction in opioid requirement was evident on postoperative day 0 for type 1 [− 21.5 mg (− 38.5 to − 3.5, Pr(Diff \(\le -10 \mathrm\)) = 0.90] and type 3 [− 19.8 mg (− 36.1 to − 5.9, Pr(Diff \(\le -10 \mathrm\)) = 0.91] surgeries.

Thoracic epidural analgesia versus non-neuraxial groups

The probability of clinical significance in favour of thoracic epidural analgesia was greater than 97% for all analyses. Overall, movement from non-neuraxial to thoracic epidural analgesia resulted in a median oral morphine equivalent opioid reduction of 38.5 mg (− 47.2 to − 29.4, Pr(Diff \(\le -10\mathrm\)) = 1.00) in laparoscopic surgery and 67.2 mg (− 86.3 to − 47.2, Pr(Diff \(\le -10 \mathrm\)) = 1.00 in open surgery (Table 3 and Additional file 1a Figure S18). This effect was sustained for all postoperative time points, with postoperative day 1 predicting the greatest reduction in opioid consumption if one were to receive thoracic epidural analgesia instead of a non-neuraxial technique for both laparoscopic (− 59.5 mg (− 80.2 to − 43.6), Pr(Diff \(\le -10 \mathrm\)) = 1.0) and open (− 105.5 mg (− 145.0 to − 67.4), Pr(Diff \(\le - 10 \mathrm\)) = 1.00) surgeries.

Subgroup analysis based on the extent of surgery determined an overall clinically relevant reduction in median oral morphine equivalent dose across all groups if one switched the analgesic technique from non-neuraxial to thoracic epidural analgesia (Table 3 and Additional file 1a Figure S19), specifically − 63.1 mg (− 82.0 to − 46.4, Pr(Diff \(\le - 10\mathrm\)) = 1.00), − 48.5 mg (− 99.6 to − 11.8, Pr(Diff \(\ge \le - 10 \mathrm\)) = 0.98) and − 37.6 mg (− 46.2 to − 28.6, Pr(Diff \(\le - 10 \mathrm\)) = 1.00) for type 1, 2, and 3 surgeries, respectively. Again the effect was evident at all postoperative time points, with the greatest impact on postoperative day 1 (type 1: − 101.0 mg (− 138.7 to − 67.3), Pr(Diff \(\le -10 \mathrm\)) = 1.0; type 2: − 65.9 mg (− 152.5 to − 6.5), Pr(Diff \(\le - 10\mathrm\)) = 0.98; type 3: − 56.2 mg (− 74.3 to − 37.1), Pr(Diff \(\le - 10\mathrm\)) = 1.0) with the effect gradually attenuating by postoperative day 3 (type 1: − 42.7 mg (− 78.5 to − 9.3), Pr(Diff \(\le - 10\mathrm\)) = 0.97; type 2: − 37.5 mg (− 132.1 to 57.4), Pr(Diff \(\le - 10\mathrm\)) = 0.79; type 3: − 25.2 mg (− 39.8 to − 11.0), Pr(Diff \(\le - 10\mathrm\)) = 0.98).

Intrathecal versus epidural groups

Simulation of patient allocation from the intrathecal morphine to epidural analgesia group found an overall reduction in median opioid requirements in both laparoscopic [− 41.8 mg (− 66.2 to − 20.1, Pr(Diff \(\le - 10 \mathrm\)) = 1.00)] and open surgery [− 66.4 mg (− 87.9 to − 47.0, Pr(Diff \(\le -10 \mathrm\)) = 1.00)](Table 3 and Additional file 1a Figure S18). This effect was least prominent on postoperative day 0 for both surgical approaches, laparoscopic [− 9.3 mg (− 28.8 to 10.8, Pr(Diff \(\le - 10 \mathrm\)) = 0.47)] versus open [− 16.1 mg (− 31.3 to − 2.6, Pr(Diff \(\le - 10 \mathrm\)) = 0.83)], the former of which was not clinically significant. From postoperative day 1 onwards, the prominence in the opioid reductions were sustained, possibly reflecting the increased requirements when the intrathecal morphine wears off. This pattern is again observed with subgroup analysis by surgical extent (Table 3 and Supplementary Material 1a Figure S19).

Primary outcome: rest and dynamic pain scores

The rest pain score most frequently measured was 4 or less at all time points in the intrathecal morphine and thoracic epidural analgesia groups (Fig. 1). Rest pain scores of greater or equal to 5, were more frequently measured in the non-neuraxial group up until postoperative day 1, after which, the highest frequency pain score in this group was 0. Dynamic pain scores were more frequently greater than 5 at all time points in the non-neuraxial group. Early dynamic pain scores (postoperative day 0 and day 1) were more frequently 0 for intrathecal morphine and thoracic epidural analgesia, the latter remaining 0 until postoperative day 2. Thereafter, dynamic pain scores were more frequently greater than 5 in both the intrathecal morphine and thoracic epidural analgesia groups.

Fig. 1

Frequency of recorded a rest and b dynamic pain scores. Frequency of recorded pain scores over different postoperative days (0 to 3) across three analgesic techniques. The X axis denotes the numerical rating scale (0–10), Y axis denotes proportion of patients in that group. The red bars represent the pain score with the most counts

Non-neuraxial versus intrathecal morphine group

Overall, simulation of allocation from a non-neuraxial technique to intrathecal morphine, or vice versa, had minimal impact on the probability of poorly controlled rest (median change − 1%, 95% credible interval (CrI) − 0.09 to 0.08) and dynamic (− 2%, 95% Crl − 0.11 to 0.08) pain, according to extent of surgery or surgical approach (Tables 4 and 5, Fig. 2). A 12% (− 0.26 to 0.01) reduction in the probability of poor rest pain control was evident on postoperative day 0 and 4% reduction (− 0.16 to 0.09) on postoperative day 1. Reversal in favour of the non-neuraxial group was evident by postoperative day 3. Subgroup analysis by surgical approach and type of surgery had similar results. An early benefit of intrathecal morphine at reducing the probability of poorly controlled dynamic pain scores (0–20% reduction) is evident across all types of surgery and approaches, with reversal in favour of non-neuraxial analgesia by postoperative day 3.

Table 4 Overall change in probabilities of poor rest pain controlTable 5 Probability of poor dynamic pain controlFig. 2

Probability of poor rest (a) or dynamic (b) pain control if one were to have a different analgesic technique. The plot shows the posterior predictive distribution of the probability of poor pain control of the original analgesic technique, the posterior predictive distribution of the probability of poor pain control if one were to have a different analgesic technique, and the distribution of the differences in the probability of poor pain control (grey). The black error bars reflect the 95% credible intervals

Non-neuraxial versus thoracic epidural analgesia group

Where a non-neuraxial technique is chosen, simulation of movement to thoracic epidural analgesia is predicted to reduce the median probability of poorly controlled rest pain overall by 8% (-0.16 to -0.01), with the greatest impact on day 0 (− 21% [− 0.35 to − 0.10]) (Fig. 2 and Table 4). No difference in the probability of poorly controlled rest pain was calculated from postoperative day 2 onwards following pancreato-duodenectomy and oesophagectomy, and by postoperative day 3, an increase was calculated following laparoscopic (5% [− 0.06 to 0.17]) and open (6% [− 0.14 to 0.22]) surgery. This finding is similar when the movement between analgesia types is considered in reverse.

A 12% reduction (− 0.21 to − 0.02) in the overall probability of poorly controlled dynamic pain was found when moving from non-neuraxial to thoracic epidural analgesia (Fig. 2 and Table 5). The impact is most substantial on postoperative day 0 (20% reduction [− 0.35 to − 0.08]) and predicted to continue until postoperative day 2. Similar findings in favour of thoracic epidural analgesia are evident when moving from thoracic epidural analgesia to non-neuraxial management, and when separating groups by surgical approach. Analysis based on extent of surgery found that by postoperative day 3 there was no difference in the probability of poorly controlled dynamic pain evident between the analgesic techniques for type 1 surgeries. There is an early reduction in the probability of poorly controlled dynamic pain when moving from non-neuraxial to thoracic epidural analgesia (33% [− 0.67 to − 0.33)] following pancreato-duodenectomy and oesophagectomy, however, there is no difference from postoperative day 1 onwards. When calculating the reverse transition from thoracic epidural to non-neuraxial, the benefit remains until postoperative day 2. Similar findings are seen for type 3 surgery.

Intrathecal morphine versus thoracic epidural analgesia groups

Simulation of allocation of patients from the intrathecal morphine to thoracic epidural analgesia is predicted to reduce the probability of median poorly controlled rest pain by 7% overall (− 0.16 to 0.02), whilst increasing it by 7% when the reverse is considered (− 0.01 to 0.15) (Fig. 2 and Table 4). In the least extensive category of major surgery (classed as type 3), there was no predicted difference in the probability of poorly controlled rest pain when moving from thoracic epidural analgesia to intrathecal morphine. A consistent benefit (ranging from 0 to 17%) in favour of thoracic epidural analgesia was seen until postoperative day 2 when considering the reverse movement between analgesia groups.

The probability of overall median poorly controlled dynamic pain is − 10% (− 0.22 to 0.00) when simulating movement from intrathecal morphine to thoracic epidural analgesia, with consistent findings until postoperative day 3 and when simulating the reverse movement, from thoracic epidural analgesia to intrathecal morphine (Fig. 2 and Table 5). Subgroup analysis shows the benefit in favour of thoracic epidural analgesia is consistent in magnitude across all types of surgery.

Secondary outcomes

The incidence of respiratory depression, sedation, and ileus were highest in the intrathecal morphine group (9%, 8%, and 10% respectively) compared to the thoracic epidural analgesia (6%, 2%, and 7%) and non-neuraxial groups (5%, 1%, and 1%) (Table 2). Postoperative nausea and vomiting were prevalent across groups; non-neuraxial (55%), intrathecal morphine (62%), and thoracic epidural analgesia (68%) (Table 2). Hypotension occurred more frequently in the thoracic epidural analgesia group (63%), compared to intrathecal morphine (30%) and the non-neuraxial group (20%).

A greater proportion of patients in the neuraxial cohorts were admitted to the high dependency unit postoperatively, thoracic epidural analgesia (94%), intrathecal morphine (93%), non-neuraxial (58%), where they also remained for longer. The median (IQR) high dependency unit length of stay for thoracic epidural analgesia was 3 (2–4) days, intrathecal morphine was 1 (1–2) day, non-neuraxial analgesia was 1 (0–1) days (Table 2). Overall hospital length of stay was also longest in the thoracic epidural analgesia cohort (epidural analgesia 13 (9–22) days versus intrathecal morphine 8 (6–12) days versus non-neuraxial 5 (3–8) days). In-hospital mortality was less than 2% across all cohorts. Discharge home with opioids was required in 44% (thoracic epidural analgesia) and 54% (intrathecal morphine and non-neuraxial groups) of patients (Table 2). At the time of data extraction, mortality over the 5-year period was highest in the thoracic epidural analgesia group (19%), compared to the intrathecal morphine (16%) and non-neuraxial groups (13%).