A total of 718 citations were retrieved from the systematic search: 683 from the electronic search, 33 from reference screening, and two from annual meeting websites (Fig. 1). From these citations, 12 potentially eligible randomized controlled studies were identified, but four were excluded because of non-randomization for the therapeutic regimen,32 data invalidated by the author (did not perform the study),33 or absence of a control group.34,35 Therefore, eight studies, including a total of 232 patients, met inclusion criteria for this systematic review (Table 1).36–43
Seven studies were published in English36–40,42,43 and one in Turkish.41 Six trials were conducted in North America36–39,42,43 and two were conducted in Asia.40,41 Seven were presented as full publications37–43 and one was presented as an abstract.36
The eight studies only included patients who were older than 18 years of age with multiple rib fractures. One study only enrolled patients who were suffering from flail chest,41 while three other studies only included patients with unilateral rib fractures.39,40,43 Two studies only included patients under mechanical ventilation.41,43 In all studies, patients had a normal level of consciousness.36–43 Exclusion criteria were mentioned in six studies: spinal cord injury or spine fracture,36,37,42 need for mechanical ventilation,40 or contraindications to epidural catheter insertion, such as hemodynamic instability and coagulopathy.36–38
Three studies compared the effect of thoracic epidural analgesia (TEA) with local anesthetics to parenteral opioids,37,38,41 two studies compared TEA with local anesthetics to interpleural analgesia,39,40 and three studies compared epidural analgesia with opioids to parenteral opioids.36,42,43
Five studies reported data on at least one of the primary outcomes: mortality (n = 4),37–40 ICU LOS (n = 5),36–38,41,43 hospital LOS (n = 5),36–38,42,43 and duration of mechanical ventilation (n = 3)37,41,43 (Table 2). The number of studies reporting secondary outcomes were: quality of pain control (n = 5)36,38–40,42 (Table 3), nosocomial pneumonia (n = 1),37 and respiratory function (n = 5).38–40,42,43
Clarification on the outcome data and the quality of methodology of six studies was obtained from the corresponding authors.33,36–40 Additional data on 17 new patients who were enrolled following publication were obtained for one abstract.36 However, since we could not obtain any standard deviation of the outcomes of interest, this study could not be included in the pooled analyses.36 For another study, standard deviations for ICU LOS and hospital LOS were calculated from raw data provided by the corresponding author.38
Of the four studies considered to be of high methodologolgical quality,37–40 only two reported data on one of our primary outcomes37,38 (Table 4). Attrition information was available for five studies: no drop-outs were observed in three studies,37,39,40 and drop-outs were excluded from the analyses in two studies.36,38 Appropriate method of allocation concealment was confirmed in four studies.37–40 Only one study mentioned cross-over between groups, and it was also the only study using intention-to-treat analyses.37 All studies were single centre studies.
Among the four studies that reported data on mortality, only three were parallel group studies where mortality could be attributed to the intervention.37–39 No significant effect on mortality was observed with the use of epidural analgesia compared to other analgesic modalities (89 patients, OR = 1.6, 95% CI [0.3, 9.3], I
2 = 0.0% [P = 0.63]) (Fig. 2).
ICU length of stay
No statistically significant reduction in ICU LOS was observed with the use of epidural analgesia compared to other analgesic modalities (125 patients, WMD −3.72, 95% CI [−11.43, 3.99], I
2 = 94.8% [P = 0.34]). Again, pooled data of all studies using TEA with local anesthetics compared to parenteral opioids (n = 3) did not show significant reduction in ICU LOS (97 patients, WMD 0.31, 95% CI [−1.44, 2.05], I
2 = 0% [P = 0.73]),37,38,41 similarly with pooled data from studies of high quality methodology (n = 2) (70 patients, WMD 0.01, 95% CI [−3.54, 3.57], I
2 = 0% [P = 0.99])37,38 (Fig. 2).
Hospital length of stay
No statistically significant reduction in hospital LOS was observed with the use of epidural analgesia compared to other analgesic modalities (130 patients, WMD −6.68, 95% CI [−19.76, 6.40], I
2 = 95.2% [P = 0.32]). Even when only pooled data of studies using TEA with local anesthetics compared to parenteral opioids were analyzed (n = 2), no significant reduction was observed (70 patients, WMD 1.55, 95% CI [−2.72, 5.82], I
2 = 0% [P = 0.48])37,38 (Fig. 2). These last two studies were also of high quality methodology.
Duration of mechanical ventilation
When the data of three studies were pooled, no statistically significant reduction of duration of mechanical ventilation was observed (101 patients, WMD −7.51, 95% CI [−16.25, 1.23], I
2 = 90.8% [P = 0.09]).37,41,43 However, when only the results from studies using TEA with local anesthetics compared to parenteral opioids were pooled (n = 2), the duration of mechanical ventilation was significantly reduced by 4.17 days (73 patients, WMD −4.17, 95% CI [−5.45, −2.88], I
2 = 0% [P < 0.01])37,41 (Fig. 2).
For all secondary outcomes, data synthesis and meta-analysis were not possible due to missing data on either standard deviation of the mean or on heterogeneous outcome measures.
In one study (46 patients), the incidence of nosocomial pneumonia was reduced by epidural analgesia compared to parenteral opioids.37 When adjusted for pulmonary contusions, flail chest, presence of chest tubes, and APACHE II scores, the OR for pneumonia in the parenteral opioids group was 6.0 (95% CI [1.0–35], [P = 0.05]), as presented by the authors in their manuscript. Information on how the analysis was adjusted could not be obtained.
Four studies reported pain scores either at rest or with coughing or deep breathing36,38,39,42 (Table 3). The pain scores in one additional study were reported on a different categorical pain scale, which precluded comparison with other studies.40 Pain at rest, at 24, 48, and 72 h, was not significantly reduced by epidural analgesia compared to other analgesic modalities in any study.38,39,42 However, statistically significant improvement in pain control was observed with cough or deep breathing in three studies.36,38,39
Vital capacity was not altered by epidural analgesia.39,42,43 In three studies, there was no difference between the maximal inspiratory pressure with epidural analgesia and with other analgesic modalities at 24, 48, and 72 h38,39,42; however, statistical significance in one study was only observed at 72 h.38 In three studies, the tidal volume reported during unassisted breathing38,39,43 was larger with epidural analgesia. No statistically significant difference in respiratory rate, PaO2, and PaCO2 were observed in three studies.40,42,43
Five studies reported adverse events,37–39,42,43 two studies reported none,36,40 and one study did not report adverse events41 (Table 5). Compared to other analgesic modalities, hypotension was significantly associated with the use of thoracic epidural analgesia with local anesthetics (three studies, 99 patients, OR 13.76, 95% CI [2.89, 65.51], I
2 = 0% [P < 0.01]).37–39 Nausea and vomiting were observed more often with parenteral opioids.37,42 Pruritus, decreased level of consciousness, and urinary retention were reported with both epidural and intravenous opioids.37,42,43 A mild allergic reaction and a superficial infection at the catheter site were reported in one patient in the epidural group.37 No study reported any serious adverse effects with the use of epidural analgesia, such as epidural hematomas, epidural abscess, spinal cord injury, or any permanent neurological symptoms. One case of respiratory depression was reported with the use of intravenous PCA.38
Available data from the primary studies did not allow performing additional sensitivity analyses as per the a priori hypotheses. However, the observed statistical heterogeneity was explained with the sensitivity analyses performed (I
2 = 0% in sensitivity analyses).31
Due to the small number of primary studies and the small number of participants in each study, the evaluation of potential publication bias could not be optimally performed with either visual or statistical evaluation.