Single injection peripheral nerve blocks are effective for postoperative pain after both upper and lower extremity surgery but are limited by the duration of action of local anesthetics. Continuous peripheral nerve blockade (CPNB) provides the potential benefits of single injection techniques (e.g., decreased pain, respiratory depression, and nausea/vomiting) well into the postoperative period (1,2). Furthermore, the introduction of mechanical and electronic pumps for the continuous infusion of local anesthetic has allowed patients to receive the benefits of continuous catheters after leaving the hospital (4–7).
Numerous clinical trials have been published examining the efficacy of CPNB for the treatment of postoperative pain after both upper and lower extremity surgery compared with systemic opioids. Although most randomized clinical trials conclude that CPNB decreases postoperative pain and opioid-related side effects when compared with opioids, studies have included a relatively small number of patients and the majority failed to show statistical significance during all time periods for reduced pain or side effects. The objectives of this meta-analysis were to evaluate data from randomized clinical trials (RCTs) to determine 1) efficacy of perineural catheters for reducing postoperative pain, 2) side effects (nausea/vomiting, sedation, pruritus, motor/sensory block), 3) opioid use, 4) and patient satisfaction compared with opioid analgesia.
A search of Ovid Medline to identify RCTs comparing CPNB with opioids for the management of postoperative pain from 1966 to the third week of May 2004 with the terms “Pain, postoperative” (13,752 articles) combined with “nerve block” (7399 articles) yielded 788 articles. Limiting this to RCTs, human and all adults (≥19 yr) yielded 236 articles. We reviewed each of the abstracts of these articles to determine if there was a description of the use of continuous peripheral nerve catheters for postoperative pain in one of the randomized groups and opioids (either oral or parenteral) in the other randomized group. This search identified 37 articles for further review of the full article to determine if inclusion criteria were met. A review of the author's files and references from the original search yielded an additional 7 articles for review. Inclusion criteria were a clearly defined anesthetic technique (combined general/regional anesthesia, general anesthesia [GA] alone, peripheral nerve block), randomized trial, adult patient population (≥18 yr old), CPNB (or analgesia) used postoperatively (intrapleural catheters were deemed not to be classified as a peripheral nerve catheter), and opioids administered for postoperative analgesia in groups not receiving peripheral nerve block. Exclusion criteria were no measurement of pain score that could be converted to visual analog scale (VAS) or no comparison of opioid to CPNB. Each article was reviewed by two separate authors with a third author used to resolve any disputes on the inclusion of any articles.
The analyzed items are reported in Table 1. Methodology and results were recorded for each study. Incidence rates for complications were recorded for each study as reported, using the larger number for nausea and vomiting if both were recorded. Although opioids were used in both treatment groups, a classification of opioid or CPNB was made to more easily indicate the two separate groups for the purpose of data analysis.
All recorded data were separated into individual subgroups as described in Table 1. Each VAS pain score was converted based on a 0 to 10 scale (all data measured using a 0–100 VAS scale were converted to a 0 to 10 scale). The standard error of each mean, if not given, was computed using the given standard deviation and sample size. If a median and interquartile range were reported, these were converted to a mean and standard deviation based on the assumption of a lognormal distribution of the original measure.
Overall, pooled means for each of the measures were computed by taking a sample size-weighted average of the reported (or estimated) means from each study as follows:
where i indexes a specific study, i is the study-specific mean, and ni the study specific sample size. The standard error (se) of each overall mean was computed by the following formula:
where se(i) is the study specific standard error of the mean. Articles that included data for multiple time points were included in the analysis. For VAS pain scores measured on postoperative days 1 through 3, time values of 24 h, 48 h, and 72 h were respectively assigned to account for differences in data observation among included studies. Opioid consumption other than IV morphine (e.g., oral, IM, or non-morphine IV opioid) was converted to IV morphine equivalents.
Data were extrapolated from graphs as needed using averaged manual measurements by two of the authors. VAS data were weighted by sample size, with the number reported for each article being the number of subjects in each treatment group. The VAS scores for each time period and complication rates were compared between treatment groups. Peripheral nerve catheter analgesia may not provide equivalent analgesia for all catheter locations; therefore, we subdivided the data by catheter location and compared perineural catheters to opioid analgesia. All peripheral catheter local anesthetic infusions were considered equivalent, including those with and without additives (e.g., clonidine). Similarly, all opioids were considered equivalent. Both rest and maximal pain were included in the analysis and were further subdivided based on catheter location.
Complications and side effects reported in the studies were recorded and analyzed, although not every article reported all possible complications. A complication rate of 0% was not assumed if studies failed to report a particular complication.
The level of significance for all tests was set at an α level of 0.05 and variances were not assumed to be equal. A Kolmogorov test showed that the data were normally distributed. One-way analysis of variance tests, weighted by sample size, were used to compare VAS pain scores. A χ2 analysis was used to analyze complication data between treatment groups. The resulting odds ratio was used to compute number needed to treat (NNT) for complication data. All statistical analyses performed used SPSS 11.5.1. (SPSS Inc., Chicago, IL).
Cost data for local anesthetic infusion pumps were pooled from that reported by Ilfeld et al. (8) for C-Bloc (I-Flow Corporation) $250, AutoMed 3200 (AceMedical/Algos) $200, and Accufuser Plus (McKinley Medical) $260 disposable pumps. The cost for peripheral nerve catheter sets (Arrow International) quoted from a company representative was $60. Daily costs of morphine and bupivacaine are negligible and accurate data could not be located in the literature to compare with our institutional costs. Cost data for prevention of nausea and vomiting were pooled from recent publications examining a variety of antiemetic strategies (dexamethasone, dolasetron, droperidol, and ondansetron) in Europe and North America (9–12). The average incremental cost for one additional nausea-free and vomiting-free patient was $40.49 (United States dollars). Cost data for the treatment of pruritus were not found in the literature, although the costs of naloxone and diphenhydramine (two of the commonly used drugs for the treatment of pruritus) are minimal. Cost data for sedation are not available. The actual cost associated with catheter placement was not found in the literature.
At the completion of the data collection and analyses, additional analyses were done to assess the validity of the conclusions that were drawn. An analysis of the file drawer problem was performed to calculate how many subjects that displayed no difference between treatment groups would be needed to invalidate our results.
A funnel plot was also created to indicate any publication bias or other biases in our meta-analysis (English language, citation, and multiple publications) (13). The natural logarithm of relative VAS scores versus precision of the studies was used to create the funnel plot. Relative VAS was calculated by dividing the mean VAS score at 24 h for patients who received a peripheral nerve catheter by the mean VAS score at 24 h for patients who received an opioid for each trial. A relative VAS of <1 signified better analgesia with peripheral nerve catheter techniques. The precision of studies was calculated by using the inverse of the standard deviation of the mean difference between the same VAS scores.
A total of 19 articles, enrolling 603 patients, were ultimately included in the meta-analysis (Appendix). The characteristics of included studies, which also contain additional data (demographics and study location) recorded but not necessarily quantified for analysis, are shown in Table 2. Included articles came primarily from hospitals in Europe (58%) and North America (38%). More studies involved lower extremity surgery (60%) than upper extremity (40%) with femoral nerve/lumbar plexus the most common catheter location for analgesia (51%) followed by interscalene (35%). Randomized clinical trials comparing perineural catheters with opioids were very limited for other locations (13%). Individual study results were summarized according to upper extremity (Table 3) and lower extremity (Table 4).
Studies in the analysis included 11 with data obtained by intention-to-treat (all enrolled patients were included in the data analysis with no treatment failures), whereas only adequately functioning catheters were included in the remaining 8 studies. A total of 13 patients were withdrawn from the catheter group after randomization and 7 excluded from the opioid group in these 8 studies. Ten additional patients were withdrawn before randomization. Overall, there were 10 catheter placement failures, 11 catheter dislodgements, and 2 patients in the catheter groups excluded for other reasons, whereas 5 patients in the opioid group were withdrawn because of nausea and 2 for failure to complete surveys.
When all studies and observations were combined, perineural analgesia provided better postoperative analgesia compared with opioids (P < 0.001). This effect was seen for all time periods measured for both mean VAS (Fig. 1) and maximum VAS (Fig. 2) at 24 h (P < 0.001), 48 h (P < 0.001), and 72 h (mean VAS only) (P < 0.001) postoperatively. When analyzed by catheter location, perineural analgesia provided superior analgesia to opioids (P < 0.05) for all locations and time periods (Table 5). The funnel plot demonstrated a relative VAS of less than one for all trials, indicating better analgesia with the CPNB technique. Analysis of the file drawer problem indicates 148 subjects who displayed no difference between treatment groups would be needed to invalidate our results. This indicates that this meta-analysis is not very resistant to the file drawer problem, which is likely attributed to the small overall number of studies with small numbers of study subjects in each.
There were no major complications reported in any of the 19 studies. Twelve of the 19 studies (63%) reported at least one minor complication, with sedation occurring most frequently overall. Motor block was the adverse effect most attributable to peripheral nerve block (P < 0.001) whereas nausea/vomiting, sedation, and pruritus all occurred more commonly with opioid analgesia (P < 0.001) (Table 6). The NNT was calculated for nausea/vomiting, sedation, and pruritus with 4, 4, and 6 patients receiving perineural analgesia expected to result in one less patient with nausea/vomiting, sedation, and pruritus, respectively, compared with opioid analgesia (Table 6).
Four trials measured patient satisfaction on a VAS and demonstrated a higher composite mean VAS satisfaction for catheters 9.6 (n = 93) (95% confidence interval [CI], 9.5–9.7) compared with opioids 7.1 (n = 90) (95% CI, 6.9–7.2). Total opioid consumption for both groups for the duration of catheter use was calculated for 12 of 19 studies. Seven studies either failed to document total opioid consumption for both groups or did not provide the data in a manner that could be converted for direct comparison. Total opioid consumption over 48 h was significantly less (P < 0.001) with the use of perineural analgesia (20.8 mg morphine, n = 165 patients [95% CI 18.5–23.1]) compared with opioid analgesia (54.1 mg morphine, n = 174 patients [95% CI, 50.8–57.4]). A limited cost analysis demonstrated an average cost of the catheter kit and disposable infusion pumps to total $296.67 (US dollars). Based on a NNT of 4, the use of CPNB analgesia would result in an average savings of $10.12 (US dollars) per patient for the treatment of nausea/vomiting.
CPNB has been proposed to offer similar benefits to single injection techniques extending well into the postoperative period in both the ambulatory and inpatient settings. We performed a meta-analysis of RCTs and found that, when compared with opioid (parenteral and oral), perineural analgesia with local anesthetic provided significantly better analgesia for postoperative pain. Improvements in analgesia were noted through postoperative day 3. When analyzed according to catheter location, (e.g., interscalene, femoral, popliteal) and type of pain assessment (rest versus maximal pain), CPNB provided superior postoperative analgesia compared with opioids. Perineural analgesia also resulted in fewer minor complications, including nausea/vomiting, pruritus, and sedation, and improved patient satisfaction.
Single injection peripheral nerve blocks have been demonstrated to provide superior pain control and decreased side effects compared with the use of opioids (14,15). These techniques are limited by the relatively short (12–24 hours) duration of analgesia provided by a single injection nerve block. Although epidural patient-controlled analgesia (PCA) and IV PCA provide adequate analgesia for inpatients with postoperative pain, these techniques are unsuitable for postoperative pain management after outpatient surgical procedures. More than 40% of patients undergoing ambulatory orthopedic procedures experience moderate to severe pain postoperatively (16). There are several reasons that perineural catheters may provide better postoperative analgesia compared with parenteral opioids, including the fact that local anesthetics may attenuate or block painful input into the central nervous system.
The overall incidence of side effects was similar to that noted in reviews of opioid analgesia and single injection peripheral nerve blocks. An expected incidence of 15% for pruritus (17) and 50% for nausea/vomiting (18) are similar to results obtained for our opioid group. The use of CPNB resulted in less side effects (nausea/vomiting, sedation, and pruritus) than opioid analgesia. CPNB do not generally obviate the need for opioids entirely. The description of opioid use varied significantly in the RCTs, making it difficult to analyze data on reduced opioid consumption; however, a statistically significant decrease in opioid requirements was noted overall with the use of CPNB. The use of opioids in this group may contribute to the sedation, pruritus, and possibly nausea/vomiting that were noted in patients with CPNB analgesia. The group receiving CPNB reported more frequent motor block, although the incidence (31.4%) was less than would be expected for single injection blocks, likely as a result of decreased concentration and smaller volumes of local anesthetic associated with continuous infusions.
Individual RCTs suggest additional benefits including improved patient satisfaction (6,19), better sleep patterns (6,7), improved rehabilitation (1,20), and shorter hospital stays (1,19). Improved patient satisfaction in our study was statistically significant; however, data came from only 4 studies. Further studies to confirm these benefits, in addition to the advantage of decreased pain and decreased opioid-related side effects, may be incorporated into a cost-benefit analysis.
There are several limitations to this study, including those that relate to the general use of a meta-analysis and others which pertain specifically to the issue examined (opioid versus perineural catheter analgesia). The clinical significance of our findings with regard to decreases in mean and maximal VAS may not correlate with the finding of a statistically significant decrease in pain with the use of perineural catheters. Although an approximate 50% reduction in VAS scores was noted at all time periods, a decrease of 2 cm (on a 0–10 cm scale) may not be clinically meaningful (e.g., a 50% reduction from a VAS of 0.5 to 0.25 is likely to be clinically insignificant whereas a decrease in VAS from 10 to 5 is likely to have a large clinical impact). Reductions in VAS scores of the magnitude observed in our study are consistent with values that have been suggested to be a clinically relevant reduction in pain scores (21,22).
Although the data were weighted by trial size, they were not weighted by the quality of the RCTs used nor were they assessed in a blinded fashion. The effects of assessing quality of the RCTs or assessment in a blinded fashion of the estimate of intervention efficacy reported in a meta-analysis are unclear (23). This may play a significant role in the analysis, as several of the trials were of poor quality. Only 11 of the 19 studies were double-blind with catheters placed in both groups. One study did not involve continuous infusion or patient-controlled intermittent bolus of local anesthetic but rather a physician-controlled bolus every 6 hours (24). This study was among those that failed to show a benefit for the use of perineural catheters, and exclusion of the results would only strengthen the overall conclusion that analgesia is improved with perineural catheters.
We limited our meta-analysis to English language articles. This may introduce publication bias if only positive findings are published primarily in English language journals (25). The effect of excluding non-English language trials on the results of a meta-analysis is equivocal; however, some data suggest that omission of trials published in non-English journals may have little effect on the summary treatment effects and may result in a more conservative estimate of treatment effect (26). Our funnel plot showed little publication bias or other biases because almost all data points were plotted around a relative VAS of <1.0.
In examining the included studies for methodology, there was no consistency in analgesic regimen for either the opioid or peripheral nerve catheter group. The opioid group included a variety of opioids, routes of administration (oral, parenteral), and frequency of administration, whereas the catheter group included different local anesthetics (bupivacaine and ropivacaine), concentrations (ranging from 0.125% to 0.5%), infusion rates, boluses, and catheter locations. Both groups also commonly had supplemental analgesics administered, including various nonsteroidal antiinflammatory drugs. This variability in protocol could affect the results of the study, although a trend towards a benefit with perineural analgesia appears to occur regardless of the analgesic regimen. Further studies to determine the ideal local anesthetic, concentration, infusion rate, bolus dose, and additives for each catheter site and surgical location are still needed to determine the optimal use of CPNB.
In summary, we performed a meta-analysis of RCTs to determine the analgesic efficacy of postoperative perineural catheter analgesia compared with opioid. Peripheral nerve catheter analgesia provided a statistically and clinically significant improvement in postoperative pain control compared with opioids and a decrease in opioid-related side effects. This effect was seen at all time periods studied after surgery, for mean and maximal pain, and with all catheter locations.
The authors wish to thank Brian Ilfeld, MD, Assistant Professor, Department of Anesthesiology, University of Florida, Gainesville, Florida, for taking the time from his busy schedule to offer his thoughts for this meta-analysis.
Appendix: Articles Meeting Inclusion Criteria
- Borgeat A, Perschak H, Bird P, et al. Patient-controlled interscalene analgesia with ropivacaine 0.2% versus patient-controlled IV analgesia after major shoulder surgery: effects on diaphragmatic and respiratory function. Anesthesiology 2000;92:102–8.
- White PF, Issioui T, Skrivanek GD, et al. The use of a continuous popliteal sciatic nerve block after surgery involving the foot and ankle: does it improve the quality of recovery? Anesth Analg 2003;97:1303–9.
- Griffith JP, Whiteley S, Gough MJ. Prospective randomized study of a new method of providing postoperative pain relief following femoropopliteal bypass. Br J Surg 1996;83:1735–8.
- Capdevila X, Barthelet Y, Biboulet P, et al. Effects of perioperative analgesic technique on the surgical outcome and duration of rehabilitation after major knee surgery. Anesthesiology 1999;91:8–15.
- Cuignet O, Pirson J, Boughrouph J, Duville D. The efficacy of continuous fascia iliaca compartment block for pain management in burn patients undergoing skin grafting procedures. Anesth Analg 2004;98:1077–81.
- Chudinov A, Berkenstadt H, Salai M, et al. Continuous psoas compartment block for anesthesia and perioperative analgesia in patients with hip fractures. Reg Anesth Pain Med 1999;24:563–8.
- Borgeat A, Schappi B, Biasca N, Gerber C. Patient-controlled analgesia after major shoulder surgery: patient-controlled interscalene analgesia versus patient-controlled analgesia. Anesthesiology 1997;87:1343–7.
- Ilfeld BM, Morey TE, Enneking FK. Continuous infraclavicular brachial plexus block for postoperative pain control at home: a randomized, double-blind, placebo-controlled study. Anesthesiology 2002;96:1297–304.
- Ilfeld BM, Morey TE, Wang RD, Enneking FK. Continuous popliteal sciatic nerve block for postoperative pain control at home: a randomized, double-blind, placebo-controlled study. Anesthesiology 2002;97:959–65.
- Ilfeld BM, Morey TE, Wright TW, et al. Continuous interscalene brachial plexus block for postoperative pain control at home: a randomized, double-blind, placebo-controlled study. Anesth Analg 2003;96:1089–95.
- Ganapathy S, Wasserman RA, Watson JT, et al. Modified continuous femoral three-in-one block for postoperative pain after total knee arthroplasty. Anesth Analg 1999;89:1197–202.
- Spansberg NL, Anker-Moller E, Dahl JB, et al. The value of continuous blockade of the lumbar plexus as an adjunct to acetylsalicyclic acid for pain relief after surgery for femoral neck fractures. Eur J Anaesthesiol 1996;13:410–2.
- Lehtipalo S, Koskinen LO, Johansson G, et al. Continuous interscalene brachial plexus block for postoperative analgesia following shoulder surgery. Acta Anaesthesiol Scand 1999;43:258–64.
- Borgeat A, Tewes E, Biasca N, Gerber C. Patient-controlled interscalene analgesia with ropivacaine after major shoulder surgery: PCIA vs. PCA. Br J Anaesth 1998;81:603–5.
- Edwards ND, Wright EM. Continuous low-dose 3-in-1 nerve blockade for postoperative pain relief after total knee replacement. Anesth Analg 1992;75:265–7.
- Hirst GC, Lang SA, Dust WN, et al. Femoral nerve block. Single injection versus continuous infusion for total knee arthroplasty. Reg Anesth 1996;21:292–7.
- Serpell MG, Millar FA, Thomson MF. Comparison of lumbar plexus block versus conventional opioid analgesia after total knee replacement. Anesthesia 1991;46:275–7.
- Singelyn FJ, Deyaert M, Joris D, et al. Effects of IV patient-controlled analgesia with morphine, continuous epidural analgesia, and continuous three-in-one block on postoperative pain and knee rehabilitation after unilateral total knee arthroplasty. Anesth Analg 1998;87:88–92.
- Klein SM, Grant SA, Greengrass RA, et al. Interscalene brachial plexus block with a continuous catheter insertion system and a disposable infusion pump. Anesth Analg 2000;91:1473–8.
1. Singelyn FJ, Deyaert M, Joris D, et al. Effects of intravenous patient-controlled analgesia with morphine, continuous epidural analgesia, and continuous three-in-one block on postoperative pain and knee rehabilitation after unilateral total knee arthroplasty. Anesth Analg 1998;87:88–92.
2. Capdevila X, Barthelet Y, Biboulet P, et al. Effects of perioperative analgesic technique on the surgical outcome and duration of rehabilitation after major knee surgery. Anesthesiology 1999;91:8–15.
4. Chelly JE, Delaunay L, Williams B, Borghi B. Outpatient lower extremity infusions. Best Pract Res Clin Anaesthesiol 2002;16:311–20.
5. Grant SA, Nielsen KC, Greengrass RA, et al. Continuous peripheral nerve block for ambulatory surgery. Reg Anesth Pain Med 2001;26:209–14.
6. Ilfeld BM, Morey TE, Wang RD, Enneking FK. Continuous popliteal sciatic nerve block for postoperative pain control at home: a randomized, double-blinded, placebo-controlled study. Anesthesiology 2002;97:959–65.
7. Ilfeld BM, Morey TE, Wright TW, et al. Continuous interscalene brachial plexus block for postoperative pain control at home: a randomized, double-blinded, placebo-controlled study. Anesth Analg 2003;96:1089–95.
8. Ilfeld BM, Morey TE, Enneking FK. New portable infusion pumps: real advantages or just more of the same in a different package? Reg Anesth Pain Med 2004;29:371–6.
9. Zarate E, Watcha MF, White PF, et al. A comparison of the costs and efficacy of ondansetron versus dolasetron for antiemetic prophylaxis. Anesth Analg 2000;90:1352–8.
10. Frighetto L, Loewen PS, Dolman J, Marra CA. Cost-effectiveness of prophylactic dolasetron or droperidol vs rescue therapy in the prevention of PONV in ambulatory gynecologic surgery. Can J Anaesth 1999;46:536–43.
11. Hill RP, Lubarsky DA, Phillips-Bute B, et al. Cost-effectiveness of prophylactic antiemetic therapy with ondansetron, droperidol, or placebo. Anesthesiology 2000;92:958–67.
12. Pueyo FJ, Lopez-Olaondo L, Sanchez-Ledesma MJ, et al. Cost-effectiveness of three combinations of antiemetics in the prevention of postoperative nausea and vomiting. Br J Anaesth 2003;91:589–92.
13. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629–34.
14. Biboulet P, Morau D, Aubas P, et al. Postoperative analgesia after total-hip arthroplasty: Comparison of intravenous patient-controlled analgesia with morphine and single injection of femoral nerve or psoas compartment block: a prospective, randomized, double-blind study. Reg Anesth Pain Med 2004;29:102–9.
15. Hadzic A, Arliss J, Kerimoglu B, et al. A comparison of infraclavicular nerve block versus general anesthesia for hand and wrist day-case surgeries. Anesthesiology 2004;101:127–32.
16. Rawal N, Hylander J, Nydahl PA, et al. Survey of postoperative analgesia following ambulatory surgery. Acta Anaesthesiol Scand 1997;41:1017–22.
17. Kjellberg F, Tramer MR. Pharmacological control of opioid-induced pruritus: a quantitative systematic review of randomized trials. Eur J Anaesthesiol 2001;18:346–57.
18. Tramer MR, Walder B. Efficacy and adverse effects of prophylactic antiemetics during patient-controlled analgesia therapy: a quantitative systematic review. Anesth Analg 1999;88:1354–61.
19. White PF, Issioui T, Skrivanek GD, et al. The use of a continuous popliteal sciatic nerve block after surgery involving the foot and ankle: does it improve the quality of recovery? Anesth Analg 2003;97:1303–9.
20. Chelly JE, Greger J, Gebhard R, et al. Continuous femoral blocks improve recovery and outcome of patients undergoing total knee arthroplasty. J Arthroplasty 2001;16:436–45.
21. Farrar JT, Portenoy RK, Berlin JA, et al. Defining the clinically important difference in pain outcome measures. Pain 2000;88:287–94.
22. Farrar JT, Berlin JA, Strom BL. Clinically important changes in acute pain outcome measures: a validation study. J Pain Symptom Manage 2003;25:406–11.
23. Jadad AR, Moore RA, Carroll D, et al. Assessing the quality of reports of randomized clinical trials: is blinding necessary? Control Clin Trials 1996;17:1–12.
24. Serpell MG, Millar FA, Thomson MF. Comparison of lumbar plexus block versus conventional opioid analgesia after total knee replacement. Anaesthesia 1991;46:275–7.
25. Moher D, Pham B, Lawson ML, Klassen TP. The inclusion of reports of randomised trials published in languages other than English in systematic reviews. Health Technol Assess 2003;7:1–90.
26. Juni P, Holenstein F, Sterne J, et al. Direction and impact of language bias in meta-analyses of controlled trials: empirical study. Int J Epidemiol 2002;31:115–23.