Results of the literature search.
Forest plot of pneumothorax rates following thoracentesis. The x-axis is drawn on a log scale. Studies are organized by whether real-time ultrasonography guidance was used during the procedures and then by quality score. Squares indicate the proportion of thoracenteses complicated by pneumothorax in each study; horizontal lines, the 95% confidence interval. Diamonds at the bottom of the each subgroup and the overall total at the bottom of the figure show the pooled estimates (with 95% confidence intervals) from the random-effects models. The study by Bass and White17 did not present data on the pneumothorax rate with or without ultrasonography guidance and is presented separately. All studies were included in the overall total analysis, with the study arms of the randomized controlled trials analyzed separately. *Randomized controlled trial comparing real-time ultrasonography-guided thoracentesis vs unguided thoracentesis. Study arms were analyzed separately in the overall total meta-analysis. †Study presented data on procedures performed with real-time ultrasonography guidance and on procedures performed without ultrasonography guidance, which are presented in the appropriate section. For the overall total meta-analysis, the studies are analyzed as single cohorts.
Gordon CE, Feller-Kopman D, Balk EM, Smetana GW. Pneumothorax Following ThoracentesisA Systematic Review and Meta-analysis. Arch Intern Med. 2010;170(4):332–339. doi:10.1001/archinternmed.2009.548
Little is known about the factors related to the development of pneumothorax following thoracentesis. We aimed to determine the mean pneumothorax rate following thoracentesis and to identify risk factors for pneumothorax through a systematic review and meta-analysis.
We reviewed MEDLINE-indexed studies from January 1, 1966, through April 1, 2009, and included studies of any design with at least 10 patients that reported the pneumothorax rate following thoracentesis. Two investigators independently extracted data on the pneumothorax rate, risk factors for pneumothorax, and study methodological quality.
Twenty-four studies reported pneumothorax rates following 6605 thoracenteses. The overall pneumothorax rate was 6.0% (95% confidence interval [CI], 4.6%-7.8%), and 34.1% of pneumothoraces required chest tube insertion. Ultrasonography use was associated with significantly lower risk of pneumothorax (odds ratio [OR], 0.3; 95% CI, 0.2-0.7). Lower pneumothorax rates were observed with experienced operators (3.9% vs 8.5%, P = .04), but this was nonsignificant within studies directly comparing this factor (OR, 0.7; 95% CI, 0.2-2.3). Pneumothorax was more likely following therapeutic thoracentesis (OR, 2.6; 95% CI, 1.8-3.8), in conjunction with periprocedural symptoms (OR, 26.6; 95% CI, 2.7-262.5), and in association with, although nonsignificantly, mechanical ventilation (OR, 4.0; 95% CI, 0.95-16.8). Two or more needle passes conferred a nonsignificant increased risk of pneumothorax (OR, 2.5; 95% CI, 0.3-20.1).
Iatrogenic pneumothorax is a common complication of thoracentesis and frequently requires chest tube insertion. Real-time ultrasonography use is a modifiable factor that reduces the pneumothorax rate. Performance of thoracentesis for therapeutic purposes and in patients undergoing mechanical ventilation confers a higher likelihood of pneumothorax. Experienced operators may have lower pneumothorax rates. Patient safety may be improved by changes in clinical practice in accord with these findings.
Medical errors have received increasing attention since the publication of the 1999 Institute of Medicine report To Err Is Human: Building a Safer Health System.1 Among medical errors, procedural complications are an important source of morbidity. Procedural complications were second in frequency only to medication errors among nonoperative adverse events in the Harvard Medical Practice Study.2Moreover, procedural complications confer a 17% excess mortality rate compared with control subjects matched by the Acute Physiology and Chronic Health Evaluation score.3 Patients who develop procedural complications have a 7-day increase in the length of inpatient stay and incur $12 913 in excess costs.3 Iatrogenic pneumothoraces resulting from thoracentesis increase morbidity, mortality, and length of hospitalization. Chest tube insertion may be required in up to 50% of cases, with a mean duration of placement of approximately 4 days.4,5
According to a 1998 National Center for Health Statistics study,6 physicians perform an estimated 173 000 thoracenteses annually in the United States. Although thoracentesis generally is considered technically straightforward, safe, and well tolerated,7 there is wide variation in published pneumothorax rates, ranging from 0%8 to 19%.9 Researchers have variably investigated the role of real-time ultrasonography guidance and operator experience as modifiable factors that may reduce pneumothorax rates following thoracentesis. Some uncertainty exists about the magnitude of the benefit of ultrasonography guidance in lowering pneumothorax rates following thoracentesis. Investigators have attempted to identify patient and procedural risk factors for the development of pneumothorax following thoracentesis, but results have been inconsistent.
To our knowledge, no systematic review of the pneumothorax rate of thoracentesis exists. Our objectives were to conduct a systematic review and meta-analysis of the mean pneumothorax rate following thoracentesis and the procedure- and patient-related factors associated with the development of pneumothorax, and to identify modifiable risk factors that could lead to improved patient safety.
We performed a MEDLINE search of English-language articles from January 1, 1966, through April 1, 2009. Search terms included the Medical Subject Headings terms pneumothorax, ultrasound, ultrasonography, complications, medical errors, risk, and injuries and the free-text terms thoracentesis, thoracocentesis, and error. The full search strategy is available on request from the authors. We identified additional references through a manual search of the bibliographies of retrieved articles. Figure 1 shows the flow of articles. We identified 448 potentially eligible studies. Of these, we excluded 41 duplicate references. After screening titles and abstracts, we deemed 342 publications ineligible as not reporting complications of thoracentesis. We retrieved and reviewed the remaining 65 studies for possible data extraction.
Two of us (C.E.G. and D.F.-K.) independently reviewed the 65 retrieved studies to determine their eligibility for our review. Because our focus was on the development of pneumothorax following thoracentesis, we included only those studies in which routine chest radiography was performed in more than 95% of subjects. We included only articles that (1) provided explicit criteria for the diagnosis of postprocedural pneumothorax, (2) clearly stated patient selection criteria, (3) defined the primary operator of the procedure, and (4) enrolled at least 10 patients. We included prospective and retrospective studies but excluded letters to the editor, editorials, review articles, position statements, abstracts, and studies that did not report complication rates. Using these criteria, we excluded an additional 41 studies (Figure 1). The remaining 24 studies formed the basis of our review.
The same 2 of us independently extracted available data about complication rates and patient and procedural risk factors for pneumothorax. We resolved any discrepancies by consensus among all authors. We established an a priori 5-point scale for study quality (Table 1). We focused our analysis on the rate of pneumothorax following thoracentesis and did not systematically review other reported complications, as these were not reported consistently in the source studies.
We performed meta-analysis of the pneumothorax event rates using the random-effects model by DerSimonian and Laird10 and a software program (Metaanalyst, beta 3.0; Schmid CH, Wallace B, Lau J, Trikalinos TA, Tufts Medical Center [http://tuftscaes.org/meta_analyst/]). We tested for heterogeneity using the Cochran χ2 statistic.
To explore relationships between pneumothorax rates and a priori selected procedural and patient characteristics known to be associated with increased complication rates following other procedures,11- 13 we performed subgroup meta-analyses. Patient factors included sex, pleural effusion size, loculation of effusion, and site of procedure (inpatient, outpatient, or intensive care unit [ICU]). Procedural factors included the use of real-time ultrasonography guidance, level of operator experience, number of needle passes, and whether the procedure was performed for diagnostic or therapeutic purposes. When studies reported pneumothorax rates for thoracenteses with and without putative risk factors, we determined the odds ratio (OR) of pneumothorax for those risk factors. However, when the study design precluded comparative analysis, we instead calculated summary pneumothorax rates across all studies reporting data on specific risk factors using random-effects model meta-analysis (ie, we combined all studies that reported pneumothorax rates with ultrasonography guidance and separately combined all studies that reported on unguided thoracentesis). We used a z score to calculate the statistical significance of differences in summary pneumothorax rates between studies with and without specific risk factors (indirect comparisons).
We selected cutoffs to categorize subgroups after considering the distribution of our data and after reviewing relevant previous literature. For operator experience, we defined less experienced operators as physicians in residency training compared with pulmonary medicine or radiology faculty. We considered thoracentesis to be therapeutic when source studies reported that the primary purpose of the procedure was therapeutic. Typically, this involved drainage of greater volumes of fluid and larger pleural effusions than diagnostic thoracentesis, but this was not universal. Conversely, we defined diagnostic thoracenteses as those performed primarily for diagnostic purposes. We defined small needles as those smaller than 20 gauge and included larger needles and needle catheter sets in our definition of large needles. We defined follow-up procedures as a second or greater procedure in the same patient, without regard to whether the first procedure was described in the study. Reporting of effusion size was heterogeneous, so we defined a large effusion as involving greater than one-third of a hemithorax. This threshold was based on the distribution of effusion size across source studies.
Twenty-four studies8,9,14- 35 provided data on the pneumothorax rate following 6605 unique thoracentesis procedures. Table 2 summarizes the study design, patient population, pneumothorax rate, and quality score for eligible studies. The studies included 11 prospective cohort studies, 11 retrospective analyses, and 2 randomized controlled trials9,28 that compared ultrasonography-guided thoracentesis by radiologists with unguided bedside thoracentesis by internists. Sample sizes ranged from 23 to 605 patients undergoing 23 to 941 (median, 227) thoracenteses. Eighteen studies included patients on a general medicine service, 5 studies8,22,23,29,33 enrolled patients in an ICU undergoing mechanical ventilation, and 1 study20 evaluated outpatients in a special procedures clinic. The primary operators of procedures were internal medicine residents (30.0% of studies), pulmonary faculty (20.0%), radiology faculty (30.0%), or a combination of these groups (20.0%). Ultrasonography guidance was used in 16 studies,8,9,14,15,20,22,25,27- 35 including the treatment arm of the 2 randomized trials. Twelve studies achieved a high quality score of 4 or 5 based on our predefined criteria; 5 of these studies met all criteria.
A total of 349 pneumothoraces occurred among 6605 thoracenteses, yielding a summary pneumothorax rate (by meta-analysis) of 6.0% (95% CI, 4.6%-7.8%), as shown in Figure 2. Pneumothorax rates among individual studies varied from 0% to 19.2% and were statistically heterogeneous across studies (P < .001). Chest tube insertion was required in 1.7% of all thoracenteses to evacuate symptomatic pneumothoraces (Table 2). Therefore, 34.1% of pneumothoraces following thoracenteses required chest tube insertion.
Pneumothorax rates were similar in prospective and retrospective studies (5.1% and 5.5%, respectively) but were lower in studies with a quality score of 4 or 5 (4.8%) than in studies with lower quality scores (7.0%); however, this difference was nonsignificant. Studies published after 2000 had significantly lower pneumothorax rates (2.9%) than earlier studies (6.8%) (P = .003). Pneumothorax rates were also significantly lower in studies of more than 200 thoracenteses (4.7%) than in smaller studies (9.2%) (P = .002).
We calculated ORs for procedure- and patient-related risk factors for pneumothorax from studies that reported pneumothorax rates for procedures with and without the particular risk factor (Table 3). We determined summary estimates of pneumothorax rates and 95% confidence intervals (CIs) for the same risk factors (Table 4) when the study design precluded our measuring ORs.
Among 16 study cohorts, investigators estimated the effect of ultrasonography-guided thoracentesis on pneumothorax rates and reported significantly lower rates than among 13 cohorts with unguided thoracentesis (Table 4 and Figure 2) (4.0% vs 9.3%, P = .001). In 6 comparative studies9,14,15,28,33,34 that reported pneumothorax rates with and without ultrasonography guidance (Table 3), ultrasonography-guided thoracentesis was associated with a significantly lower risk of pneumothorax than unguided thoracentesis (OR, 0.3; 95% CI, 0.2-0.7). Among these studies, the 2 randomized controlled trials9,28 found a similar effect size, but the difference was not significant (OR, 0.3; 95% CI, 0.0-2.8).
Pneumothorax rates were lower for procedures performed by experienced operators than for procedures performed by less experienced clinicians (3.9% vs 8.5%, P = .04). When we restricted our analysis to 4 studies that provided direct comparisons, this relationship was nonsignificant (OR, 0.7; 95% CI, 0.2-2.3). No differences were noted in pneumothorax rates between experienced radiologists and experienced pulmonologists (4.3% vs 4.4%, P = .81). Pneumothorax rates were higher for therapeutic thoracenteses than for diagnostic procedures (8.4% vs 5.2%, P = .12); this difference was statistically significant among studies that provided direct comparisons (OR, 2.6; 95% CI, 1.8-3.8). The risk of pneumothorax was increased when larger needles or catheters were used compared with smaller needles (OR, 2.5; 95% CI, 1.1-6.0). Pneumothorax rates were similar for catheters and for larger needles (8.3% vs 5.9%, P = .19), but the available studies did not allow a direct comparison. Catheter use was associated with a nonsignificant increased risk of pneumothorax compared with the use of smaller needles (OR, 1.9; 95% CI, 0.5-7.2). The studies reporting data for different needle sizes did not supply information about the indication for thoracentesis or effusion size, so we were unable to further examine these factors.
The number of needle passes required to complete thoracentesis was reported in 3 studies.8,14,21 Pneumothorax rates were higher for procedures requiring 2 or more passes than for those completed in 1 pass (6.6% vs 3.5%, P = .42). The risk of pneumothorax was higher with 2 or more needle passes (OR, 2.5; 95% CI, 0.3-20.1), but this relationship was nonsignificant. There was no difference in pneumothorax rates for first-time procedures vs follow-up procedures.
Although only 3 studies14,21,27 reported the association between periprocedural symptoms and pneumothorax, the development of any symptom (cough, dyspnea, or chest pain) conferred a markedly higher risk of pneumothorax (OR, 26.6; 95% CI, 2.7-262.5). The risk of pneumothorax also was substantially increased when operators witnessed the aspiration of air into the pleural space (Tables 3 and 4).
In contrast to procedure-related factors, few patient factors were associated with increased likelihood of pneumothorax (Tables 3 and 4). The pneumothorax rate was similar for large and small effusion size and for men and women. Pneumothorax rates were similar for procedures performed among non-ICU inpatients, ICU inpatients, and outpatients. In 2 studies22,32 describing thoracentesis in patients with and without mechanical ventilation, the risk of pneumothorax was increased with mechanical ventilation (OR, 4.0; 95% CI, 0.95-16.8). However, this relationship may have been confounded because the comparison groups of nonventilated patients were not ICU patients. No study directly compared pneumothorax rates in ICU patients with and without mechanical ventilation. Investigating whether other factors may have confounded this relationship, we found that the pneumothorax rate in mechanically ventilated patients with and without ultrasonography guidance were not significantly different (2.9% vs 6.3%, P = .20).
The studies variably reported complications of thoracentesis other than pneumothorax. Hemothoraces were reported in 6 patients from 4 studies.17,21,26,34 Vasovagal response to thoracentesis was described in 2 studies.19,27 Subcutaneous or subpleural hematomas were described in 5 studies.9,20,21,27,28 Reexpansion pulmonary edema developed in 2 patients from 1 study.27 Splenic34 and hepatic26 lacerations were described in 1 study each, and splenic aspiration without injury was described in 1 study.26 A sheared-off catheter was described in 1 study.34 Reported patient symptoms included anxiety, local pain at the procedural site, dyspnea, cough, and pleuritic chest pain.
We present the first systematic review and meta-analysis to date of the pneumothorax rate following thoracentesis. Overall, 6.0% of thoracenteses were complicated by the development of pneumothorax, and 34.1% of pneumothoraces (1.7% of all thoracenteses) required chest tube insertion. Pneumothorax following thoracentesis is an important cause of morbidity and likely results in increased length of stay for hospitalized patients. Because substantial heterogeneity existed in the pneumothorax rates across studies, we explored patient and procedural risk factors for the development of pneumothorax.
Real-time ultrasonography-guided thoracentesis was associated with a significantly lower risk of pneumothorax compared with unguided thoracentesis and was the strongest predictor of low pneumothorax rates. The pneumothorax rate seemed to be lower when more experienced clinicians were the primary operators of the procedure, but there were no differences between experienced radiology and pulmonary medicine faculty. Important factors that increase the risk of pneumothorax include therapeutic indication for the thoracentesis, witnessed aspiration of air, and any periprocedural symptoms. Although not statistically significant, other possible predictors of pneumothorax include the need for 2 or more needle insertions and concurrent mechanical ventilation. These findings are consistent with those observed for central venous catheter insertion, in which ultrasonography guidance,36,37 more experienced operators,13 and fewer needle passes12 confer lower complication rates.
Our results highlight the importance of trainee supervision during thoracentesis because of lower pneumothorax rates with direct ultrasonography guidance and with more experienced operators. In many academic medical centers, trainees perform thoracenteses without supervision from expert faculty members. Some internal medicine residents report a lack of comfort in performing bedside procedures, despite having completed a moderate number by the end of their residency.38- 40 One study39 found that residents required completion of a mean of 8.1 (95% CI, 4.0-16.5) thoracenteses before reporting being comfortable with the procedure. Supporting this finding, only 79% of internal medicine program directors surveyed believed that their graduating residents had mastered thoracentesis.41 In another study,42 surveyed internists believed that clinicians must perform 5 to 10 thoracenteses to attain competence and must perform 1 to 5 annually to maintain competence. Notably, 66% of internists in the study still reported performing thoracentesis in their clinical practice. Although our analysis was not designed to determine operator comfort level with thoracentesis, previous research has demonstrated that resident comfort with thoracentesis is increased by faculty supervision.40 These results may provide additional support of the recent concept of medical procedure service staffed by expert faculty43 and of the role of medical simulation for thoracentesis in the education of trainees.44
Some of the relationships we observed may have been subject to confounding and resultant false-positive or false-negative associations. For example, the relationship between the risk of pneumothorax and operator experience may have been confounded by the use of ultrasonography. Bias also may have existed such that more dangerous procedures in mechanically ventilated ICU patients or in those with small loculated effusions were performed by more experienced operators or with real-time ultrasonography guidance, resulting in false-negative associations between these factors and the risk of pneumothorax. Pneumothorax rates also were lower in more recent studies, larger studies, and studies of higher quality. The secular trends may represent more careful patient selection, increased use of real-time ultrasonography guidance, or greater use of expert faculty as the primary operator of thoracenteses.
Because of the potential for confounding relationships between the patient- and procedure-related risk factors for pneumothorax, we attempted to analyze combinations of variables. This analysis was limited because the study design, collection, and presentation of data in the source studies did not allow analysis of most combinations. To lessen bias, we limited analyses to direct comparisons only. Ultrasonography guidance reduced the risk of pneumothorax for large and small effusions. In the unadjusted analysis, the risk of pneumothorax was similar for both effusion sizes; this observation persisted after adjusting for ultrasonography guidance. The absence of patient-level data limited our ability to perform direct comparisons of therapeutic thoracentesis, operator experience, needle size, or other combined analyses of factors associated with pneumothorax in univariate analysis. A meta-analysis of patient-level data using the primary data from these studies or new randomized trials would be necessary to determine which factors are confounded and which are independent predictors of pneumothorax.
This study had several limitations. Publication bias is a frequent concern with meta-analysis, although there is no consensus that methods to measure the bias (such as funnel plots) are reliable.45 In addition, few studies reported data on important risk factors such as operator experience, number of needle passes, and effusion size, leading to uncertainty about the true effect of these variables. It is possible that some of these factors would be associated with higher risk of pneumothorax if more studies reported data on these variables. For example, a recent study46 (which did not meet our prespecified inclusion criteria because only 60% of patients underwent postthoracentesis chest radiography) demonstrated increased risk of pneumothorax for increasing volumes of drained effusions. Moreover, few studies directly compared pneumothorax outcomes in patients with or without putative risk factors. Although the results of our direct and indirect analyses were largely concordant, most of the analyses of risk factors relied heavily on indirect comparisons across studies. These analyses, in particular, can only be considered hypothesis-generating. Future research should report data for all putative risk factors that we identified and preferably should be of randomized controlled trial study design or should allow multivariable analyses of patient-level data to more fully account for unmeasured confounders and to measure independent effects of the various risk factors.
Despite these limitations, this is the first summary to date of a large number of studies that allows an estimate of the pneumothorax rate following thoracentesis and an analysis of risk factors for its development. The substantial between-study heterogeneity allowed us to investigate relationships between procedural and patient risk factors and the development of pneumothorax. The most important strategy to reduce pneumothorax rates is the use of ultrasonography guidance. Uniform use of ultrasonography guidance across institutions would reduce the burden of pneumothorax and allow a greater degree of safety for this commonly performed procedure. Other risk factors include mechanical ventilation, therapeutic procedures, and symptoms during thoracentesis. Pneumothorax rates are also lower for experienced clinicians, regardless of specialty, and may be lower when only 1 needle pass is required. Future research should clarify which factors are independent predictors through randomized controlled trials or analysis of patient-level data in observational studies. Institutional policies that require supervision or performance of thoracentesis by experienced operators may lower the pneumothorax rate. We encourage institutions to consider a policy of uniform use of ultrasonography guidance for thoracentesis.
Correspondence: Craig E. Gordon, MD, MS, Renal Section, Department of Medicine, Boston University Medical Center, Boston, MA 02118 (firstname.lastname@example.org).
Accepted for Publication: September 14, 2009.
Author Contributions: Dr Gordon had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Gordon, Feller-Kopman, and Smetana. Acquisition of data: Gordon and Feller-Kopman. Analysis and interpretation of data: Gordon, Feller-Kopman, Balk, and Smetana. Drafting of the manuscript: Gordon, Feller-Kopman, and Smetana. Critical revision of the manuscript for important intellectual content: Gordon, Feller-Kopman, Balk, and Smetana. Statistical analysis: Gordon and Balk. Study supervision: Feller-Kopman and Smetana.
Financial Disclosure: Dr Feller-Kopman has received lecture fees from SonoSite, Inc. Dr Smetana serves as a consultant to SafeMed.