Data Collection

This study was a retrospective, electronic chart review of patients who received a PICC while hospitalized between August 1, 2005 and November 1, 2005 at the Methodist University Hospital (MUH), a 652‐bed, urban, university‐affiliated, community hospital in Memphis, TN. Patients were identified through the use of a PICC database that is maintained by the nurses who routinely place the PICCs. The data collected included the date of insertion, the diameter of the catheter, the vein accessed, the position of the catheter tip, and the reason for PICC insertion. These factors as well as demographics were examined to determine whether they were associated with thrombosis. These data were linked with data from the ultrasound laboratory and nuclear medicine/radiology laboratory and with hospital discharge data. Data were recorded by trained research assistants and verified for accuracy by the study investigators. The institutional review board approved the study protocol prior to data collection.

Patients and Outcomes

All adult consecutive patients who had a PICC inserted during the study period and who did not have a UEDVT or pulmonary embolism (PE) at the time of PICC insertion were included in the study.

PICCs were placed using a modified Seldinger technique at the bedside with portable ultrasound guidance. The vessel of choice for insertion was the basilic vein. Confirmation of catheter tip placement in the lower third of the superior vena cava (SVC) was done with chest x‐ray prior to use of the PICC. The PICC manufacturer was Boston Scientific (Vaxcel with PASV; Natick, MA) and normal saline was used for routine flushing of the PICC.

Study Outcomes

Statistical Analyses

The incidence of VTE was reported as the proportion of patients who had a documented event during hospitalization, and also as the number of events per 1000 PICC‐days. Baseline characteristics were assessed as potential risk factors. Differences in proportions were tested with the chi square or Fisher exact test, and differences in means of the continuous variables were tested using the t‐test. Univariate analysis of symptomatic VTE by each potential risk factor was performed using logistic regression. Logistic regression models were used to simultaneously assess the relationship between the baseline factors and the probability of developing a thrombotic event. Using a backward elimination modeling strategy, only factors that maintained a P value of <0.05 were retained in the final model. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were calculated. SAS version 8.2 (SAS Institute, Inc., Cary, NC) was used for data analysis.

Patient Demographics and Baseline Characteristics

From August 1, 2005 to November 1, 2005, 954 PICCs were inserted in 777 patients. The demographics and baseline characteristics of the 777 patients are outlined in Table 1. History of cancer was present in 20.21% of the patients. History of VTE was present in 7.02% of the patients. The most common reason for PICC insertion was poor access in 90.6% of the PICCs. The most common medication infused through the PICC was antibiotics in 69.2% and intravenous hydration in 50.1%. The basilic vein was accessed in 90%. The tip location, as determined by chest X‐ray at the time of insertion, was the SVC in 85.3%. PICCs were in situ for an average of 9 days and most were 5‐French (Fr) catheters. The average duration of PICC placement was over 9 days, and the average length of stay slightly exceeded 16 days.

Outcomes: VTE

During their hospital stay, 38 patients experienced 1 or more VTEs, yielding an incidence of 4.89% (Table 2). A total of 7444 PICC‐days were recorded for 777 patients. This yields a rate of 5.10 VTEs/1000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 3.65 UEDVT/1000 PICC days. The mean length of stay was 26 days in those with VTE versus 15.8 days in those who did not develop VTE (P < 0.001). Average PICC‐days were also longer in those who developed VTE compared to those who did not (13 days vs. 9; P < 0.001).

VTE prophylaxis (using enoxaparin or heparin) was administered to 26% of patients from the time of PICC insertion until UEDVT occurred. Only 12.5% of those who developed PE were given VTE prophylaxis. All patients who developed a UEDVT or PE were treated with full anticoagulation except those with contraindications and patients with superficial upper extremity thrombosis. Five of the patients with UEDVT and 2 of the patients with PE died during their hospitalization.

Four of the 8 patients with PICC‐associated PE had bilateral lower extremity ultrasounds performed, and 3 of these were negative. The other 4 patients did not have ultrasounds performed.

Risk Factors

History of VTE was the strongest risk factor for VTE in univariate analysis, patients with a history of VTE being 10 times more likely to develop a PICC‐related VTE event (Table 3). PICC tip location was strongly associated with VTE. A noncentral location of the tip at the time of insertion was associated with a 2.34 (95% CI, 1.15‐4.75) higher risk of developing VTE during the hospitalization stay, compared to a central location (SVC or RA). The duration of PICC use was also associated with an increased risk of developing VTE (OR for a 10‐day increase in duration, 1.40 (95% CI, 1.07‐1.84), as it was the length of hospital stay (OR for a 10‐day increase in duration, 1.18 (95% CI, 1.05‐1.33). The duration of hospital stay was correlated with the duration of PICC use; however, its temporal relationship with a VTE was uncertain, as a prolonged hospital stay could be the consequence of a VTE event in some cases. In the multivariate analysis, history of VTE, PICC tip location, and length of stay retained statistical significance (Table 4); while our data are suggestive for an association with VTE events, the duration of PICC use did not maintain statistical significance, in the presence of PICC tip location. History of VTE remained the strongest risk factor for subsequent VTE (OR, 10.83; 95% CI, 4.89‐23.95); the location of the PICC tip was location highly associated with subsequent VTE (OR, 2.61; 95% CI, 1.28‐5.35).

Overall Findings

Our study is the first to examine both UEDVT and PE rates in hospitalized adult patients with PICC lines. We found that 4.89% of our patients experienced a thromboembolic complication during their hospitalization stay, yielding a rate of 5.1 VTE/1,000 PICC‐days and 3.63 UEDVT/1,000 PICC‐days. These figures are similar to or higher than those reported by other retrospective studies using symptom‐driven ultrasound diagnosis of UEDVT, reporting incidence figures ranging from 2 per 100 patients,3 to 2.47 per 100 patients,4 to 3.7 per 100 patients.5

There are several factors to consider when comparing our findings to previously published studies. First, our study population differs by being older and by including hospitalized patients with both acute and chronic medical illnesses, and with a wider spectrum of PICC indications. Second, the duration of PICC use is known to be associated with the development of symptomatic UEDVT.6 Several published studies do not report the duration of PICC use, thus making the comparison of incidence difficult to interpret. Third, the duration of patient follow‐up influences the likelihood that a symptomatic UEDVT is diagnosed; patient follow‐up time is oftentimes unspecified in some studies, thus making comparisons difficult. In contrast, our study reports the incidence of VTE during a clearly specified time window; ie, hospital stay.

Several studies that reported the incidence per 100 PICCs found incidences ranging from 3.9 per 100 PICCs7 and 4.66 per 100 PICCs.8 Similarly, studies of different patient populations and the use of systematic diagnostic and follow‐up methods have found very high rates of thrombosis (15.4% in a randomized clinical trial [RCT] of total parenteral nutrition [TPN],9 38% with systematic venography to diagnose UEDVT,10 and 64.52% in an RCT of intensive care unit [ICU] patients11).

PE

In the largest study to date of 2063 patients who received a PICC for intravenous antibiotic therapy, it was found that 2.5% developed upper extremity thrombosis; of these, 3.8% also had a positive VQ scan for PE.4 However, this study did not systematically review records for PE in those without upper extremity thrombosis. Our study showed that approximately 1% of those who received a PICC developed a PE during their hospitalization. In 3 of our 8 patients who developed a PE, bilateral lower extremity ultrasounds were negative for PE. None of the subjects who developed PE received ultrasounds of the upper extremities in order to look for deep vein thrombosis (DVT), possibly due to a lack of awareness of PICC‐associated VTE.

Risk Factors

Among the risk factors explored, we found that patients with previous history of a venous thromboembolic event are much more likely to develop a PICC‐related VTE. This is not surprising, as previous VTE has been identified a risk factor.1214 We also found that patients whose PICC was not confirmed to be in the SVC or at the junction of the SVC and the RA, were twice as likely to experience VTE. An increased risk associated with noncentral tip location was reported in 2 previous studies using venography to assess vein condition among infectious disease patients,15 and oncology patients.16 This is biologically plausible, since the smaller diameter vessels may be more conducive to thrombus formation than the SVC or RA. Other factors such as blood flow rate, turbulent flow, and endothelial injury may also play a role. As recognized by the National Association of Vascular Access Networks,17 tip positioning is influenced by catheter length, anthropometric measurements, and anatomical pathways. Central tip location, whenever possible, may be 1 of the few controllable risk factors for UEDVT. Decreasing the hospital‐acquired DVT events has been recognized as an important quality improvement, requiring an increased identification and treatment of high‐risk groups. Such hospital‐based approaches have been suggested as effective.18, 19 Both tip location and history of VTE are identifiable risk factors that may be readily incorporated into specific strategies to decrease hospital‐acquired DVT events.

While previously reported as a risk factor, left‐sided catheter location was not significantly associated with VTE in our study. Our data suggest that the duration of PICC use is associated with the risk of developing a VTE event during hospital stay, yet without reaching statistical significance. In the absence of information regarding the underlying comorbidities and/or reasons for hospitalization, interpretation of PICC line duration remains elusive.

Anticoagulant Prophylaxis

Most of the subjects in our cohort did not receive VTE prophylaxis. The efficacy of anticoagulant prophylaxis for preventing catheter‐associated VTE is controversial. In fact, current guidelines do not support VTE prophylaxis as a means to reduce rates of VTE associated with central venous catheters. Furthermore, recent controlled clinical trials of low‐molecular weight heparins at standard prophylactic doses and low doses of warfarin for VTE prevention in patients with central venous catheters undergoing cancer chemotherapy have not demonstrated reductions in VTE rates.2022 In contrast, a recent meta‐analysis suggests that anticoagulant prophylaxis is effective for preventing all catheter‐associated DVT in patients with central venous catheters, but the effectiveness for preventing symptomatic VTE, including PE, remains uncertain.23

Limitations

The small number of VTE and the lack of control for the underlying disease, concomitant therapy, and anticoagulant prophylaxis limited our ability to identify independent risk factors. Further studies are needed to confirm our finding that 1% of patients who receive a PICC will develop PE.

VTE were identified only when symptoms led to diagnostic testing, so our event rates likely underestimate the true rate of PICC‐associated VTE. It is estimated that subclinical thromboses can be found in 30% to 60% of all central catheters.10, 24 The clinical significance of silent thromboses remains to be established.

Strengths

The Society of Interventional Radiology recommends uniform reporting requirements to assist in study design and outcomes reporting on central venous access devices.25 We adhered to many of these guidelines which may facilitate comparison among studies from different institutions.

Our study is one of the largest studies to asses the rate of thrombosis among inpatients. We used broad eligibility criteria, included consecutive, unselected patients, in a large cohort of patients who were all hospitalized for acute medical illnesses. Furthermore, we used data collection that corroborated data across the entire range of hospital records, and as such we were able to determine accurately the incidence of thrombosis. UEDVT and PE were diagnosed objectively and, unlike previous studies, we also collected information on prophylaxis and treatment of VTE. To our knowledge, our study is the first to document an increased risk of symptomatic UEDVT and PE among acutely ill hospitalized patients who receive PICC.

Prior studies may be less applicable to PICC‐associated VTE in hospitalized patients because the populations included a mix of inpatients and outpatients. Our results may be more applicable to hospitalized patients who receive a PICC because more severely ill patients who are hospitalized may be at higher risk of thrombosis than outpatients. Also, previous studies included some patients who received relatively small, 3‐Fr to 4‐Fr single‐lumen PICCs, whereas the current practice is to insert larger, double‐lumen, 5‐Fr catheters. Today's PICCs may be more likely to result in vessel occlusion, venous stasis, and thrombosis.

In conclusion, we have shown that the incidence of VTE in our study population was 4.89% with a rate of 5.10 VTE/1,000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 365 UEVDVT/1000 PICC days. The most significant predictors of VTE were history of previous VTE and the location of the PICC tip at the time of insertion.

Data Collection

This study was a retrospective, electronic chart review of patients who received a PICC while hospitalized between August 1, 2005 and November 1, 2005 at the Methodist University Hospital (MUH), a 652‐bed, urban, university‐affiliated, community hospital in Memphis, TN. Patients were identified through the use of a PICC database that is maintained by the nurses who routinely place the PICCs. The data collected included the date of insertion, the diameter of the catheter, the vein accessed, the position of the catheter tip, and the reason for PICC insertion. These factors as well as demographics were examined to determine whether they were associated with thrombosis. These data were linked with data from the ultrasound laboratory and nuclear medicine/radiology laboratory and with hospital discharge data. Data were recorded by trained research assistants and verified for accuracy by the study investigators. The institutional review board approved the study protocol prior to data collection.

Patients and Outcomes

All adult consecutive patients who had a PICC inserted during the study period and who did not have a UEDVT or pulmonary embolism (PE) at the time of PICC insertion were included in the study.

PICCs were placed using a modified Seldinger technique at the bedside with portable ultrasound guidance. The vessel of choice for insertion was the basilic vein. Confirmation of catheter tip placement in the lower third of the superior vena cava (SVC) was done with chest x‐ray prior to use of the PICC. The PICC manufacturer was Boston Scientific (Vaxcel with PASV; Natick, MA) and normal saline was used for routine flushing of the PICC.

Study Outcomes

Statistical Analyses

The incidence of VTE was reported as the proportion of patients who had a documented event during hospitalization, and also as the number of events per 1000 PICC‐days. Baseline characteristics were assessed as potential risk factors. Differences in proportions were tested with the chi square or Fisher exact test, and differences in means of the continuous variables were tested using the t‐test. Univariate analysis of symptomatic VTE by each potential risk factor was performed using logistic regression. Logistic regression models were used to simultaneously assess the relationship between the baseline factors and the probability of developing a thrombotic event. Using a backward elimination modeling strategy, only factors that maintained a P value of <0.05 were retained in the final model. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were calculated. SAS version 8.2 (SAS Institute, Inc., Cary, NC) was used for data analysis.

Patient Demographics and Baseline Characteristics

From August 1, 2005 to November 1, 2005, 954 PICCs were inserted in 777 patients. The demographics and baseline characteristics of the 777 patients are outlined in Table 1. History of cancer was present in 20.21% of the patients. History of VTE was present in 7.02% of the patients. The most common reason for PICC insertion was poor access in 90.6% of the PICCs. The most common medication infused through the PICC was antibiotics in 69.2% and intravenous hydration in 50.1%. The basilic vein was accessed in 90%. The tip location, as determined by chest X‐ray at the time of insertion, was the SVC in 85.3%. PICCs were in situ for an average of 9 days and most were 5‐French (Fr) catheters. The average duration of PICC placement was over 9 days, and the average length of stay slightly exceeded 16 days.

Outcomes: VTE

During their hospital stay, 38 patients experienced 1 or more VTEs, yielding an incidence of 4.89% (Table 2). A total of 7444 PICC‐days were recorded for 777 patients. This yields a rate of 5.10 VTEs/1000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 3.65 UEDVT/1000 PICC days. The mean length of stay was 26 days in those with VTE versus 15.8 days in those who did not develop VTE (P < 0.001). Average PICC‐days were also longer in those who developed VTE compared to those who did not (13 days vs. 9; P < 0.001).

VTE prophylaxis (using enoxaparin or heparin) was administered to 26% of patients from the time of PICC insertion until UEDVT occurred. Only 12.5% of those who developed PE were given VTE prophylaxis. All patients who developed a UEDVT or PE were treated with full anticoagulation except those with contraindications and patients with superficial upper extremity thrombosis. Five of the patients with UEDVT and 2 of the patients with PE died during their hospitalization.

Four of the 8 patients with PICC‐associated PE had bilateral lower extremity ultrasounds performed, and 3 of these were negative. The other 4 patients did not have ultrasounds performed.

Risk Factors

History of VTE was the strongest risk factor for VTE in univariate analysis, patients with a history of VTE being 10 times more likely to develop a PICC‐related VTE event (Table 3). PICC tip location was strongly associated with VTE. A noncentral location of the tip at the time of insertion was associated with a 2.34 (95% CI, 1.15‐4.75) higher risk of developing VTE during the hospitalization stay, compared to a central location (SVC or RA). The duration of PICC use was also associated with an increased risk of developing VTE (OR for a 10‐day increase in duration, 1.40 (95% CI, 1.07‐1.84), as it was the length of hospital stay (OR for a 10‐day increase in duration, 1.18 (95% CI, 1.05‐1.33). The duration of hospital stay was correlated with the duration of PICC use; however, its temporal relationship with a VTE was uncertain, as a prolonged hospital stay could be the consequence of a VTE event in some cases. In the multivariate analysis, history of VTE, PICC tip location, and length of stay retained statistical significance (Table 4); while our data are suggestive for an association with VTE events, the duration of PICC use did not maintain statistical significance, in the presence of PICC tip location. History of VTE remained the strongest risk factor for subsequent VTE (OR, 10.83; 95% CI, 4.89‐23.95); the location of the PICC tip was location highly associated with subsequent VTE (OR, 2.61; 95% CI, 1.28‐5.35).

Overall Findings

Our study is the first to examine both UEDVT and PE rates in hospitalized adult patients with PICC lines. We found that 4.89% of our patients experienced a thromboembolic complication during their hospitalization stay, yielding a rate of 5.1 VTE/1,000 PICC‐days and 3.63 UEDVT/1,000 PICC‐days. These figures are similar to or higher than those reported by other retrospective studies using symptom‐driven ultrasound diagnosis of UEDVT, reporting incidence figures ranging from 2 per 100 patients,3 to 2.47 per 100 patients,4 to 3.7 per 100 patients.5

There are several factors to consider when comparing our findings to previously published studies. First, our study population differs by being older and by including hospitalized patients with both acute and chronic medical illnesses, and with a wider spectrum of PICC indications. Second, the duration of PICC use is known to be associated with the development of symptomatic UEDVT.6 Several published studies do not report the duration of PICC use, thus making the comparison of incidence difficult to interpret. Third, the duration of patient follow‐up influences the likelihood that a symptomatic UEDVT is diagnosed; patient follow‐up time is oftentimes unspecified in some studies, thus making comparisons difficult. In contrast, our study reports the incidence of VTE during a clearly specified time window; ie, hospital stay.

Several studies that reported the incidence per 100 PICCs found incidences ranging from 3.9 per 100 PICCs7 and 4.66 per 100 PICCs.8 Similarly, studies of different patient populations and the use of systematic diagnostic and follow‐up methods have found very high rates of thrombosis (15.4% in a randomized clinical trial [RCT] of total parenteral nutrition [TPN],9 38% with systematic venography to diagnose UEDVT,10 and 64.52% in an RCT of intensive care unit [ICU] patients11).

PE

In the largest study to date of 2063 patients who received a PICC for intravenous antibiotic therapy, it was found that 2.5% developed upper extremity thrombosis; of these, 3.8% also had a positive VQ scan for PE.4 However, this study did not systematically review records for PE in those without upper extremity thrombosis. Our study showed that approximately 1% of those who received a PICC developed a PE during their hospitalization. In 3 of our 8 patients who developed a PE, bilateral lower extremity ultrasounds were negative for PE. None of the subjects who developed PE received ultrasounds of the upper extremities in order to look for deep vein thrombosis (DVT), possibly due to a lack of awareness of PICC‐associated VTE.

Risk Factors

Among the risk factors explored, we found that patients with previous history of a venous thromboembolic event are much more likely to develop a PICC‐related VTE. This is not surprising, as previous VTE has been identified a risk factor.1214 We also found that patients whose PICC was not confirmed to be in the SVC or at the junction of the SVC and the RA, were twice as likely to experience VTE. An increased risk associated with noncentral tip location was reported in 2 previous studies using venography to assess vein condition among infectious disease patients,15 and oncology patients.16 This is biologically plausible, since the smaller diameter vessels may be more conducive to thrombus formation than the SVC or RA. Other factors such as blood flow rate, turbulent flow, and endothelial injury may also play a role. As recognized by the National Association of Vascular Access Networks,17 tip positioning is influenced by catheter length, anthropometric measurements, and anatomical pathways. Central tip location, whenever possible, may be 1 of the few controllable risk factors for UEDVT. Decreasing the hospital‐acquired DVT events has been recognized as an important quality improvement, requiring an increased identification and treatment of high‐risk groups. Such hospital‐based approaches have been suggested as effective.18, 19 Both tip location and history of VTE are identifiable risk factors that may be readily incorporated into specific strategies to decrease hospital‐acquired DVT events.

While previously reported as a risk factor, left‐sided catheter location was not significantly associated with VTE in our study. Our data suggest that the duration of PICC use is associated with the risk of developing a VTE event during hospital stay, yet without reaching statistical significance. In the absence of information regarding the underlying comorbidities and/or reasons for hospitalization, interpretation of PICC line duration remains elusive.

Anticoagulant Prophylaxis

Most of the subjects in our cohort did not receive VTE prophylaxis. The efficacy of anticoagulant prophylaxis for preventing catheter‐associated VTE is controversial. In fact, current guidelines do not support VTE prophylaxis as a means to reduce rates of VTE associated with central venous catheters. Furthermore, recent controlled clinical trials of low‐molecular weight heparins at standard prophylactic doses and low doses of warfarin for VTE prevention in patients with central venous catheters undergoing cancer chemotherapy have not demonstrated reductions in VTE rates.2022 In contrast, a recent meta‐analysis suggests that anticoagulant prophylaxis is effective for preventing all catheter‐associated DVT in patients with central venous catheters, but the effectiveness for preventing symptomatic VTE, including PE, remains uncertain.23

Limitations

The small number of VTE and the lack of control for the underlying disease, concomitant therapy, and anticoagulant prophylaxis limited our ability to identify independent risk factors. Further studies are needed to confirm our finding that 1% of patients who receive a PICC will develop PE.

VTE were identified only when symptoms led to diagnostic testing, so our event rates likely underestimate the true rate of PICC‐associated VTE. It is estimated that subclinical thromboses can be found in 30% to 60% of all central catheters.10, 24 The clinical significance of silent thromboses remains to be established.

Strengths

The Society of Interventional Radiology recommends uniform reporting requirements to assist in study design and outcomes reporting on central venous access devices.25 We adhered to many of these guidelines which may facilitate comparison among studies from different institutions.

Our study is one of the largest studies to asses the rate of thrombosis among inpatients. We used broad eligibility criteria, included consecutive, unselected patients, in a large cohort of patients who were all hospitalized for acute medical illnesses. Furthermore, we used data collection that corroborated data across the entire range of hospital records, and as such we were able to determine accurately the incidence of thrombosis. UEDVT and PE were diagnosed objectively and, unlike previous studies, we also collected information on prophylaxis and treatment of VTE. To our knowledge, our study is the first to document an increased risk of symptomatic UEDVT and PE among acutely ill hospitalized patients who receive PICC.

Prior studies may be less applicable to PICC‐associated VTE in hospitalized patients because the populations included a mix of inpatients and outpatients. Our results may be more applicable to hospitalized patients who receive a PICC because more severely ill patients who are hospitalized may be at higher risk of thrombosis than outpatients. Also, previous studies included some patients who received relatively small, 3‐Fr to 4‐Fr single‐lumen PICCs, whereas the current practice is to insert larger, double‐lumen, 5‐Fr catheters. Today's PICCs may be more likely to result in vessel occlusion, venous stasis, and thrombosis.

In conclusion, we have shown that the incidence of VTE in our study population was 4.89% with a rate of 5.10 VTE/1,000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 365 UEVDVT/1000 PICC days. The most significant predictors of VTE were history of previous VTE and the location of the PICC tip at the time of insertion.

Data Collection

This study was a retrospective, electronic chart review of patients who received a PICC while hospitalized between August 1, 2005 and November 1, 2005 at the Methodist University Hospital (MUH), a 652‐bed, urban, university‐affiliated, community hospital in Memphis, TN. Patients were identified through the use of a PICC database that is maintained by the nurses who routinely place the PICCs. The data collected included the date of insertion, the diameter of the catheter, the vein accessed, the position of the catheter tip, and the reason for PICC insertion. These factors as well as demographics were examined to determine whether they were associated with thrombosis. These data were linked with data from the ultrasound laboratory and nuclear medicine/radiology laboratory and with hospital discharge data. Data were recorded by trained research assistants and verified for accuracy by the study investigators. The institutional review board approved the study protocol prior to data collection.

Patients and Outcomes

All adult consecutive patients who had a PICC inserted during the study period and who did not have a UEDVT or pulmonary embolism (PE) at the time of PICC insertion were included in the study.

PICCs were placed using a modified Seldinger technique at the bedside with portable ultrasound guidance. The vessel of choice for insertion was the basilic vein. Confirmation of catheter tip placement in the lower third of the superior vena cava (SVC) was done with chest x‐ray prior to use of the PICC. The PICC manufacturer was Boston Scientific (Vaxcel with PASV; Natick, MA) and normal saline was used for routine flushing of the PICC.

Study Outcomes

Statistical Analyses

The incidence of VTE was reported as the proportion of patients who had a documented event during hospitalization, and also as the number of events per 1000 PICC‐days. Baseline characteristics were assessed as potential risk factors. Differences in proportions were tested with the chi square or Fisher exact test, and differences in means of the continuous variables were tested using the t‐test. Univariate analysis of symptomatic VTE by each potential risk factor was performed using logistic regression. Logistic regression models were used to simultaneously assess the relationship between the baseline factors and the probability of developing a thrombotic event. Using a backward elimination modeling strategy, only factors that maintained a P value of <0.05 were retained in the final model. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were calculated. SAS version 8.2 (SAS Institute, Inc., Cary, NC) was used for data analysis.

Patient Demographics and Baseline Characteristics

From August 1, 2005 to November 1, 2005, 954 PICCs were inserted in 777 patients. The demographics and baseline characteristics of the 777 patients are outlined in Table 1. History of cancer was present in 20.21% of the patients. History of VTE was present in 7.02% of the patients. The most common reason for PICC insertion was poor access in 90.6% of the PICCs. The most common medication infused through the PICC was antibiotics in 69.2% and intravenous hydration in 50.1%. The basilic vein was accessed in 90%. The tip location, as determined by chest X‐ray at the time of insertion, was the SVC in 85.3%. PICCs were in situ for an average of 9 days and most were 5‐French (Fr) catheters. The average duration of PICC placement was over 9 days, and the average length of stay slightly exceeded 16 days.

Outcomes: VTE

During their hospital stay, 38 patients experienced 1 or more VTEs, yielding an incidence of 4.89% (Table 2). A total of 7444 PICC‐days were recorded for 777 patients. This yields a rate of 5.10 VTEs/1000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 3.65 UEDVT/1000 PICC days. The mean length of stay was 26 days in those with VTE versus 15.8 days in those who did not develop VTE (P < 0.001). Average PICC‐days were also longer in those who developed VTE compared to those who did not (13 days vs. 9; P < 0.001).

VTE prophylaxis (using enoxaparin or heparin) was administered to 26% of patients from the time of PICC insertion until UEDVT occurred. Only 12.5% of those who developed PE were given VTE prophylaxis. All patients who developed a UEDVT or PE were treated with full anticoagulation except those with contraindications and patients with superficial upper extremity thrombosis. Five of the patients with UEDVT and 2 of the patients with PE died during their hospitalization.

Four of the 8 patients with PICC‐associated PE had bilateral lower extremity ultrasounds performed, and 3 of these were negative. The other 4 patients did not have ultrasounds performed.

Risk Factors

History of VTE was the strongest risk factor for VTE in univariate analysis, patients with a history of VTE being 10 times more likely to develop a PICC‐related VTE event (Table 3). PICC tip location was strongly associated with VTE. A noncentral location of the tip at the time of insertion was associated with a 2.34 (95% CI, 1.15‐4.75) higher risk of developing VTE during the hospitalization stay, compared to a central location (SVC or RA). The duration of PICC use was also associated with an increased risk of developing VTE (OR for a 10‐day increase in duration, 1.40 (95% CI, 1.07‐1.84), as it was the length of hospital stay (OR for a 10‐day increase in duration, 1.18 (95% CI, 1.05‐1.33). The duration of hospital stay was correlated with the duration of PICC use; however, its temporal relationship with a VTE was uncertain, as a prolonged hospital stay could be the consequence of a VTE event in some cases. In the multivariate analysis, history of VTE, PICC tip location, and length of stay retained statistical significance (Table 4); while our data are suggestive for an association with VTE events, the duration of PICC use did not maintain statistical significance, in the presence of PICC tip location. History of VTE remained the strongest risk factor for subsequent VTE (OR, 10.83; 95% CI, 4.89‐23.95); the location of the PICC tip was location highly associated with subsequent VTE (OR, 2.61; 95% CI, 1.28‐5.35).

Overall Findings

Our study is the first to examine both UEDVT and PE rates in hospitalized adult patients with PICC lines. We found that 4.89% of our patients experienced a thromboembolic complication during their hospitalization stay, yielding a rate of 5.1 VTE/1,000 PICC‐days and 3.63 UEDVT/1,000 PICC‐days. These figures are similar to or higher than those reported by other retrospective studies using symptom‐driven ultrasound diagnosis of UEDVT, reporting incidence figures ranging from 2 per 100 patients,3 to 2.47 per 100 patients,4 to 3.7 per 100 patients.5

There are several factors to consider when comparing our findings to previously published studies. First, our study population differs by being older and by including hospitalized patients with both acute and chronic medical illnesses, and with a wider spectrum of PICC indications. Second, the duration of PICC use is known to be associated with the development of symptomatic UEDVT.6 Several published studies do not report the duration of PICC use, thus making the comparison of incidence difficult to interpret. Third, the duration of patient follow‐up influences the likelihood that a symptomatic UEDVT is diagnosed; patient follow‐up time is oftentimes unspecified in some studies, thus making comparisons difficult. In contrast, our study reports the incidence of VTE during a clearly specified time window; ie, hospital stay.

Several studies that reported the incidence per 100 PICCs found incidences ranging from 3.9 per 100 PICCs7 and 4.66 per 100 PICCs.8 Similarly, studies of different patient populations and the use of systematic diagnostic and follow‐up methods have found very high rates of thrombosis (15.4% in a randomized clinical trial [RCT] of total parenteral nutrition [TPN],9 38% with systematic venography to diagnose UEDVT,10 and 64.52% in an RCT of intensive care unit [ICU] patients11).

PE

In the largest study to date of 2063 patients who received a PICC for intravenous antibiotic therapy, it was found that 2.5% developed upper extremity thrombosis; of these, 3.8% also had a positive VQ scan for PE.4 However, this study did not systematically review records for PE in those without upper extremity thrombosis. Our study showed that approximately 1% of those who received a PICC developed a PE during their hospitalization. In 3 of our 8 patients who developed a PE, bilateral lower extremity ultrasounds were negative for PE. None of the subjects who developed PE received ultrasounds of the upper extremities in order to look for deep vein thrombosis (DVT), possibly due to a lack of awareness of PICC‐associated VTE.

Risk Factors

Among the risk factors explored, we found that patients with previous history of a venous thromboembolic event are much more likely to develop a PICC‐related VTE. This is not surprising, as previous VTE has been identified a risk factor.1214 We also found that patients whose PICC was not confirmed to be in the SVC or at the junction of the SVC and the RA, were twice as likely to experience VTE. An increased risk associated with noncentral tip location was reported in 2 previous studies using venography to assess vein condition among infectious disease patients,15 and oncology patients.16 This is biologically plausible, since the smaller diameter vessels may be more conducive to thrombus formation than the SVC or RA. Other factors such as blood flow rate, turbulent flow, and endothelial injury may also play a role. As recognized by the National Association of Vascular Access Networks,17 tip positioning is influenced by catheter length, anthropometric measurements, and anatomical pathways. Central tip location, whenever possible, may be 1 of the few controllable risk factors for UEDVT. Decreasing the hospital‐acquired DVT events has been recognized as an important quality improvement, requiring an increased identification and treatment of high‐risk groups. Such hospital‐based approaches have been suggested as effective.18, 19 Both tip location and history of VTE are identifiable risk factors that may be readily incorporated into specific strategies to decrease hospital‐acquired DVT events.

While previously reported as a risk factor, left‐sided catheter location was not significantly associated with VTE in our study. Our data suggest that the duration of PICC use is associated with the risk of developing a VTE event during hospital stay, yet without reaching statistical significance. In the absence of information regarding the underlying comorbidities and/or reasons for hospitalization, interpretation of PICC line duration remains elusive.

Anticoagulant Prophylaxis

Most of the subjects in our cohort did not receive VTE prophylaxis. The efficacy of anticoagulant prophylaxis for preventing catheter‐associated VTE is controversial. In fact, current guidelines do not support VTE prophylaxis as a means to reduce rates of VTE associated with central venous catheters. Furthermore, recent controlled clinical trials of low‐molecular weight heparins at standard prophylactic doses and low doses of warfarin for VTE prevention in patients with central venous catheters undergoing cancer chemotherapy have not demonstrated reductions in VTE rates.2022 In contrast, a recent meta‐analysis suggests that anticoagulant prophylaxis is effective for preventing all catheter‐associated DVT in patients with central venous catheters, but the effectiveness for preventing symptomatic VTE, including PE, remains uncertain.23

Limitations

The small number of VTE and the lack of control for the underlying disease, concomitant therapy, and anticoagulant prophylaxis limited our ability to identify independent risk factors. Further studies are needed to confirm our finding that 1% of patients who receive a PICC will develop PE.

VTE were identified only when symptoms led to diagnostic testing, so our event rates likely underestimate the true rate of PICC‐associated VTE. It is estimated that subclinical thromboses can be found in 30% to 60% of all central catheters.10, 24 The clinical significance of silent thromboses remains to be established.

Strengths

The Society of Interventional Radiology recommends uniform reporting requirements to assist in study design and outcomes reporting on central venous access devices.25 We adhered to many of these guidelines which may facilitate comparison among studies from different institutions.

Our study is one of the largest studies to asses the rate of thrombosis among inpatients. We used broad eligibility criteria, included consecutive, unselected patients, in a large cohort of patients who were all hospitalized for acute medical illnesses. Furthermore, we used data collection that corroborated data across the entire range of hospital records, and as such we were able to determine accurately the incidence of thrombosis. UEDVT and PE were diagnosed objectively and, unlike previous studies, we also collected information on prophylaxis and treatment of VTE. To our knowledge, our study is the first to document an increased risk of symptomatic UEDVT and PE among acutely ill hospitalized patients who receive PICC.

Prior studies may be less applicable to PICC‐associated VTE in hospitalized patients because the populations included a mix of inpatients and outpatients. Our results may be more applicable to hospitalized patients who receive a PICC because more severely ill patients who are hospitalized may be at higher risk of thrombosis than outpatients. Also, previous studies included some patients who received relatively small, 3‐Fr to 4‐Fr single‐lumen PICCs, whereas the current practice is to insert larger, double‐lumen, 5‐Fr catheters. Today's PICCs may be more likely to result in vessel occlusion, venous stasis, and thrombosis.

In conclusion, we have shown that the incidence of VTE in our study population was 4.89% with a rate of 5.10 VTE/1,000 PICC‐days. There were 27 patients who had a UEDVT over the 7444 PICC days yielding a rate of 365 UEVDVT/1000 PICC days. The most significant predictors of VTE were history of previous VTE and the location of the PICC tip at the time of insertion.

Patients with Known VTE

Suggested indications for the use of vena cava filters in patients with proven VTE are listed in Table 3. For patients at risk for either recurrent or severe bleeding (eg, multiple falls, recurrent gastrointestinal or intracranial hemorrhage) or most patients who have failed treatment with therapeutic anticoagulation, a permanent filter is usually the preferred mechanical option. However, for certain conditions (such as Trousseau's syndrome, heparin‐induced thrombocytopenia, antiphospholipid syndrome, or anatomic abnormalities such as thoracic outlet syndrome‐Paget‐von Schroetter syndrome, or May‐Thurner syndrome‐iliac vein compression syndrome), vena cava filters have been shown either to be ineffective or to worsen thrombosis. In these cases, alternative therapies must be used, based on the underlying disorder and the clinical situation.

A retrievable filter should only be considered in patients who have a transient contraindication to anticoagulation (Table 5). Such contraindications include isolated but treatable episodes of hemorrhage, urgent surgeries, or procedures associated with a high risk of bleeding, and trauma. The risk of recurrent VTE in the absence of anticoagulation has been estimated at 40% in the first month after VTE and then 10% during the second and third months.22 Therefore, it is reasonable to place a retrievable filter in perioperative patients who cannot be treated with therapeutic anticoagulation during the first 30 days after an acute VTE. If more than 30 days have passed since the thrombotic event, a filter is probably not necessary for patients who will have temporary interruptions in anticoagulation therapy. Instead, bridging anticoagulation (eg, unfractionated heparin [UFH] or low molecular weight heparin [LMWH]) can be given while warfarin is being held prior to surgery. Then, the patient can be transitioned back to warfarin therapy with prophylactic and then therapeutic LMWH or UFH in the postoperative period.

Controversy remains regarding the use of retrievable filters in patients with calf vein DVT. It also exists for patients with massive or submassive PE who are receiving anticoagulation therapy but are at high risk for poor outcomes should another PEeven if smalloccur while they are on anticoagulation therapy. Vena cava filters are generally not recommended for patients with distal VTE unless they have a persistent contraindication to anticoagulation therapy and have shown clot propagation on serial duplex studies. At least 1 institution, however, has noted an increased use of filter placement in this population since the advent of retrievable filters.23 Randomized controlled trials and practice guidelines are still lacking in this area. Therefore, there is currently insufficient evidence to recommend retrievable filters for distal VTE.

There is also insufficient evidence to recommend filters for patients with massive or submassive PE who can tolerate anticoagulation therapy. Only 1 registry study has compared patients with massive PE (defined by a systolic blood pressure <90 mmHg at presentation) who were treated with vena cava filters to those who were not.24 Though there was a reduction in recurrent PE and mortality at 90 days in patients who received filters, this result requires further confirmation due to the small number of patients who received filters (11 patients) and a possible selection bias (patients who received filters were, on average, 16 years younger than those who did not). More evidence will be needed to weigh not only the cost but the risks of filter insertion (such as insertion site hematoma, increased incidence of DVT, or contrast nephropathy) against any benefit. Until then, routine filter use in patients with massive or submassive PE cannot be routinely recommended, but may be considered in those with massive PE and impending hemodynamic collapse.

Prophylaxis in High‐Risk Patients

Controversy also exists in the use of retrievable filters in patients without VTE who are at high risk for thromboembolic events. Currently, there are no randomized controlled trials that have established the efficacy of retrievable filters as prophylaxis in these patients. However, there are a number of prospective and retrospective studies that examine this topic, particularly in trauma patients.

During Filter Placement

Complications related to both retrievable and nonretrievable filter placement are rare but have been documented in several studies. Failure of the filter to deploy properly has been reported.21 The same study also noted pneumothorax as a complication in some patients whose filters were inserted via the jugular vein.21 Therefore, location of access and retrieval should be an important consideration for patients with significant underlying pulmonary disease. Insertion site thrombosis and arteriovenous fistula formation have been reported primarily with permanent filters31, 32; that risk could be extrapolated to retrievable filters given that the method of placement is the same. Iodine contrast‐induced nephropathy is of concern for high‐risk patients, although the procedure can be performed using gadolinium‐based contrast, carbon dioxide contrast, or without contrast (under ultrasound guidance).

During Filter Retrieval

Filter tilting and clot trapping under the filter that occurs during the filter removal process are infrequent causes of non‐retrieval. Tilting of the filter sometimes can pose problems, but if this occurs, the filter can be repositioned so that the degree of tilt no longer precludes removal. Severe cases of tilting that lead to nonretrieval are very rare. When thrombus is trapped in the filter (Figure 3), retrieval often depends on the amount of thrombus. A visual scale to assist in judgment of thrombus volume has been developed to assist in retrieval decision‐making.33 In some cases, catheter‐directed thrombolysis has been used to facilitate thrombus dissolution.34

Figure 3

VTE After Placement

Table 2 lists the incidence of VTE after retrievable filter placement. The overall incidence of PE is low, but that of DVT varies widely. These data raise the possibility that some filters may not be removed due to the occurrence of a new DVT, thereby becoming permanent filters with the associated risks of recurrent DVT, caval thrombosis, and PE. Only a few studies have investigated the differences in the rate of PE between permanent and retrievable filters and have shown no differences.29 The long‐term complication rates of retrievable filters and how they may differ from permanent filters warrants further investigation.

Some studies have also noted the development of PE after filter retrieval.35, 36 It is possible that a subclinical DVT was present at the time of removal or that the filter was retrieved before the risk of thrombosis had resolved. Therefore, consideration should be given to the use of duplex ultrasound evaluation for DVT prior to filter removal to ensure that patients with active thrombosis receive therapeutic anticoagulation for an appropriate duration.

Because of the concern for DVT and PE associated with retrievable filters, anticoagulation should ideally occur before and after retrieval, once the bleeding risk has become acceptable. Consensus guidelines support this practice,5, 37 though one systematic review has found insufficient evidence regarding the use of anticoagulation in patients with vena cava filters.4 Retrospective reviews have shown that filters can be both placed and removed without bleeding complications, even in patients who are therapeutically anticoagulated with warfarin and/or LMWH.38, 39 Further investigation would be useful to confirm whether this is an effective approach to VTE prevention at the time of retrieval.

Other Adverse Events

Other complications that have been associated with retrievable filters include migration, fracture, infection, and perforation. It may be difficult to estimate the true incidence of these complications, as most of the literature on this topic comes from case reports. Vena cava perforation with hooks may be not uncommon but in most cases is not clinically significant.40 Filter fracture is more common but rarely reported. Filter migration toward the heart is a very rare but potentially life‐threatening complication. The Recovery filter was taken off the market due in part to concerns about migration.26 As the use of retrievable filters increases, complications related to filters will need to be monitored.

Ongoing and Future Research

Other types of removable filters are currently in development. Convertible filters that can be converted into a stent once they are no longer needed are under investigation. Other devices, such as absorbable or drug‐eluting filters, are also being studied.5 In addition, there is ongoing research to better characterize the safety and efficacy of available filters. The Prevention du Risque d'Embolie Pulmonaire par Interruption Cave (PREPIC) 2 will assess their use in the first prospective, randomized, controlled trial of retrievable filters in patients with acute VTE receiving anticoagulation (http://www.clinicaltrials.gov; Identifier: NCT00457158). Other studies include an evaluation of the long‐term outcomes of patients with retrievable filters who failed retrieval (http://www.clinicaltrials.gov; Identifier: NCT00163956) and a comparison of Gnther Tulip and OptEase filters (http://www. clinicaltrials.gov; Identifier: NCT00588757). Randomized controlled trials are still needed to evaluate the efficacy of prophylactic filter placement in high‐risk patients. Studies that examine intention to retrieve vs. actual and recommended retrieval rates would provide valuable information on practice patterns.

Patients with Known VTE

Suggested indications for the use of vena cava filters in patients with proven VTE are listed in Table 3. For patients at risk for either recurrent or severe bleeding (eg, multiple falls, recurrent gastrointestinal or intracranial hemorrhage) or most patients who have failed treatment with therapeutic anticoagulation, a permanent filter is usually the preferred mechanical option. However, for certain conditions (such as Trousseau's syndrome, heparin‐induced thrombocytopenia, antiphospholipid syndrome, or anatomic abnormalities such as thoracic outlet syndrome‐Paget‐von Schroetter syndrome, or May‐Thurner syndrome‐iliac vein compression syndrome), vena cava filters have been shown either to be ineffective or to worsen thrombosis. In these cases, alternative therapies must be used, based on the underlying disorder and the clinical situation.

A retrievable filter should only be considered in patients who have a transient contraindication to anticoagulation (Table 5). Such contraindications include isolated but treatable episodes of hemorrhage, urgent surgeries, or procedures associated with a high risk of bleeding, and trauma. The risk of recurrent VTE in the absence of anticoagulation has been estimated at 40% in the first month after VTE and then 10% during the second and third months.22 Therefore, it is reasonable to place a retrievable filter in perioperative patients who cannot be treated with therapeutic anticoagulation during the first 30 days after an acute VTE. If more than 30 days have passed since the thrombotic event, a filter is probably not necessary for patients who will have temporary interruptions in anticoagulation therapy. Instead, bridging anticoagulation (eg, unfractionated heparin [UFH] or low molecular weight heparin [LMWH]) can be given while warfarin is being held prior to surgery. Then, the patient can be transitioned back to warfarin therapy with prophylactic and then therapeutic LMWH or UFH in the postoperative period.

Controversy remains regarding the use of retrievable filters in patients with calf vein DVT. It also exists for patients with massive or submassive PE who are receiving anticoagulation therapy but are at high risk for poor outcomes should another PEeven if smalloccur while they are on anticoagulation therapy. Vena cava filters are generally not recommended for patients with distal VTE unless they have a persistent contraindication to anticoagulation therapy and have shown clot propagation on serial duplex studies. At least 1 institution, however, has noted an increased use of filter placement in this population since the advent of retrievable filters.23 Randomized controlled trials and practice guidelines are still lacking in this area. Therefore, there is currently insufficient evidence to recommend retrievable filters for distal VTE.

There is also insufficient evidence to recommend filters for patients with massive or submassive PE who can tolerate anticoagulation therapy. Only 1 registry study has compared patients with massive PE (defined by a systolic blood pressure <90 mmHg at presentation) who were treated with vena cava filters to those who were not.24 Though there was a reduction in recurrent PE and mortality at 90 days in patients who received filters, this result requires further confirmation due to the small number of patients who received filters (11 patients) and a possible selection bias (patients who received filters were, on average, 16 years younger than those who did not). More evidence will be needed to weigh not only the cost but the risks of filter insertion (such as insertion site hematoma, increased incidence of DVT, or contrast nephropathy) against any benefit. Until then, routine filter use in patients with massive or submassive PE cannot be routinely recommended, but may be considered in those with massive PE and impending hemodynamic collapse.

Prophylaxis in High‐Risk Patients

Controversy also exists in the use of retrievable filters in patients without VTE who are at high risk for thromboembolic events. Currently, there are no randomized controlled trials that have established the efficacy of retrievable filters as prophylaxis in these patients. However, there are a number of prospective and retrospective studies that examine this topic, particularly in trauma patients.

During Filter Placement

Complications related to both retrievable and nonretrievable filter placement are rare but have been documented in several studies. Failure of the filter to deploy properly has been reported.21 The same study also noted pneumothorax as a complication in some patients whose filters were inserted via the jugular vein.21 Therefore, location of access and retrieval should be an important consideration for patients with significant underlying pulmonary disease. Insertion site thrombosis and arteriovenous fistula formation have been reported primarily with permanent filters31, 32; that risk could be extrapolated to retrievable filters given that the method of placement is the same. Iodine contrast‐induced nephropathy is of concern for high‐risk patients, although the procedure can be performed using gadolinium‐based contrast, carbon dioxide contrast, or without contrast (under ultrasound guidance).

During Filter Retrieval

Filter tilting and clot trapping under the filter that occurs during the filter removal process are infrequent causes of non‐retrieval. Tilting of the filter sometimes can pose problems, but if this occurs, the filter can be repositioned so that the degree of tilt no longer precludes removal. Severe cases of tilting that lead to nonretrieval are very rare. When thrombus is trapped in the filter (Figure 3), retrieval often depends on the amount of thrombus. A visual scale to assist in judgment of thrombus volume has been developed to assist in retrieval decision‐making.33 In some cases, catheter‐directed thrombolysis has been used to facilitate thrombus dissolution.34

Figure 3

VTE After Placement

Table 2 lists the incidence of VTE after retrievable filter placement. The overall incidence of PE is low, but that of DVT varies widely. These data raise the possibility that some filters may not be removed due to the occurrence of a new DVT, thereby becoming permanent filters with the associated risks of recurrent DVT, caval thrombosis, and PE. Only a few studies have investigated the differences in the rate of PE between permanent and retrievable filters and have shown no differences.29 The long‐term complication rates of retrievable filters and how they may differ from permanent filters warrants further investigation.

Some studies have also noted the development of PE after filter retrieval.35, 36 It is possible that a subclinical DVT was present at the time of removal or that the filter was retrieved before the risk of thrombosis had resolved. Therefore, consideration should be given to the use of duplex ultrasound evaluation for DVT prior to filter removal to ensure that patients with active thrombosis receive therapeutic anticoagulation for an appropriate duration.

Because of the concern for DVT and PE associated with retrievable filters, anticoagulation should ideally occur before and after retrieval, once the bleeding risk has become acceptable. Consensus guidelines support this practice,5, 37 though one systematic review has found insufficient evidence regarding the use of anticoagulation in patients with vena cava filters.4 Retrospective reviews have shown that filters can be both placed and removed without bleeding complications, even in patients who are therapeutically anticoagulated with warfarin and/or LMWH.38, 39 Further investigation would be useful to confirm whether this is an effective approach to VTE prevention at the time of retrieval.

Other Adverse Events

Other complications that have been associated with retrievable filters include migration, fracture, infection, and perforation. It may be difficult to estimate the true incidence of these complications, as most of the literature on this topic comes from case reports. Vena cava perforation with hooks may be not uncommon but in most cases is not clinically significant.40 Filter fracture is more common but rarely reported. Filter migration toward the heart is a very rare but potentially life‐threatening complication. The Recovery filter was taken off the market due in part to concerns about migration.26 As the use of retrievable filters increases, complications related to filters will need to be monitored.

Ongoing and Future Research

Other types of removable filters are currently in development. Convertible filters that can be converted into a stent once they are no longer needed are under investigation. Other devices, such as absorbable or drug‐eluting filters, are also being studied.5 In addition, there is ongoing research to better characterize the safety and efficacy of available filters. The Prevention du Risque d'Embolie Pulmonaire par Interruption Cave (PREPIC) 2 will assess their use in the first prospective, randomized, controlled trial of retrievable filters in patients with acute VTE receiving anticoagulation (http://www.clinicaltrials.gov; Identifier: NCT00457158). Other studies include an evaluation of the long‐term outcomes of patients with retrievable filters who failed retrieval (http://www.clinicaltrials.gov; Identifier: NCT00163956) and a comparison of Gnther Tulip and OptEase filters (http://www. clinicaltrials.gov; Identifier: NCT00588757). Randomized controlled trials are still needed to evaluate the efficacy of prophylactic filter placement in high‐risk patients. Studies that examine intention to retrieve vs. actual and recommended retrieval rates would provide valuable information on practice patterns.

Patients with Known VTE

Suggested indications for the use of vena cava filters in patients with proven VTE are listed in Table 3. For patients at risk for either recurrent or severe bleeding (eg, multiple falls, recurrent gastrointestinal or intracranial hemorrhage) or most patients who have failed treatment with therapeutic anticoagulation, a permanent filter is usually the preferred mechanical option. However, for certain conditions (such as Trousseau's syndrome, heparin‐induced thrombocytopenia, antiphospholipid syndrome, or anatomic abnormalities such as thoracic outlet syndrome‐Paget‐von Schroetter syndrome, or May‐Thurner syndrome‐iliac vein compression syndrome), vena cava filters have been shown either to be ineffective or to worsen thrombosis. In these cases, alternative therapies must be used, based on the underlying disorder and the clinical situation.

A retrievable filter should only be considered in patients who have a transient contraindication to anticoagulation (Table 5). Such contraindications include isolated but treatable episodes of hemorrhage, urgent surgeries, or procedures associated with a high risk of bleeding, and trauma. The risk of recurrent VTE in the absence of anticoagulation has been estimated at 40% in the first month after VTE and then 10% during the second and third months.22 Therefore, it is reasonable to place a retrievable filter in perioperative patients who cannot be treated with therapeutic anticoagulation during the first 30 days after an acute VTE. If more than 30 days have passed since the thrombotic event, a filter is probably not necessary for patients who will have temporary interruptions in anticoagulation therapy. Instead, bridging anticoagulation (eg, unfractionated heparin [UFH] or low molecular weight heparin [LMWH]) can be given while warfarin is being held prior to surgery. Then, the patient can be transitioned back to warfarin therapy with prophylactic and then therapeutic LMWH or UFH in the postoperative period.

Controversy remains regarding the use of retrievable filters in patients with calf vein DVT. It also exists for patients with massive or submassive PE who are receiving anticoagulation therapy but are at high risk for poor outcomes should another PEeven if smalloccur while they are on anticoagulation therapy. Vena cava filters are generally not recommended for patients with distal VTE unless they have a persistent contraindication to anticoagulation therapy and have shown clot propagation on serial duplex studies. At least 1 institution, however, has noted an increased use of filter placement in this population since the advent of retrievable filters.23 Randomized controlled trials and practice guidelines are still lacking in this area. Therefore, there is currently insufficient evidence to recommend retrievable filters for distal VTE.

There is also insufficient evidence to recommend filters for patients with massive or submassive PE who can tolerate anticoagulation therapy. Only 1 registry study has compared patients with massive PE (defined by a systolic blood pressure <90 mmHg at presentation) who were treated with vena cava filters to those who were not.24 Though there was a reduction in recurrent PE and mortality at 90 days in patients who received filters, this result requires further confirmation due to the small number of patients who received filters (11 patients) and a possible selection bias (patients who received filters were, on average, 16 years younger than those who did not). More evidence will be needed to weigh not only the cost but the risks of filter insertion (such as insertion site hematoma, increased incidence of DVT, or contrast nephropathy) against any benefit. Until then, routine filter use in patients with massive or submassive PE cannot be routinely recommended, but may be considered in those with massive PE and impending hemodynamic collapse.

Prophylaxis in High‐Risk Patients

Controversy also exists in the use of retrievable filters in patients without VTE who are at high risk for thromboembolic events. Currently, there are no randomized controlled trials that have established the efficacy of retrievable filters as prophylaxis in these patients. However, there are a number of prospective and retrospective studies that examine this topic, particularly in trauma patients.

During Filter Placement

Complications related to both retrievable and nonretrievable filter placement are rare but have been documented in several studies. Failure of the filter to deploy properly has been reported.21 The same study also noted pneumothorax as a complication in some patients whose filters were inserted via the jugular vein.21 Therefore, location of access and retrieval should be an important consideration for patients with significant underlying pulmonary disease. Insertion site thrombosis and arteriovenous fistula formation have been reported primarily with permanent filters31, 32; that risk could be extrapolated to retrievable filters given that the method of placement is the same. Iodine contrast‐induced nephropathy is of concern for high‐risk patients, although the procedure can be performed using gadolinium‐based contrast, carbon dioxide contrast, or without contrast (under ultrasound guidance).

During Filter Retrieval

Filter tilting and clot trapping under the filter that occurs during the filter removal process are infrequent causes of non‐retrieval. Tilting of the filter sometimes can pose problems, but if this occurs, the filter can be repositioned so that the degree of tilt no longer precludes removal. Severe cases of tilting that lead to nonretrieval are very rare. When thrombus is trapped in the filter (Figure 3), retrieval often depends on the amount of thrombus. A visual scale to assist in judgment of thrombus volume has been developed to assist in retrieval decision‐making.33 In some cases, catheter‐directed thrombolysis has been used to facilitate thrombus dissolution.34

Figure 3

VTE After Placement

Table 2 lists the incidence of VTE after retrievable filter placement. The overall incidence of PE is low, but that of DVT varies widely. These data raise the possibility that some filters may not be removed due to the occurrence of a new DVT, thereby becoming permanent filters with the associated risks of recurrent DVT, caval thrombosis, and PE. Only a few studies have investigated the differences in the rate of PE between permanent and retrievable filters and have shown no differences.29 The long‐term complication rates of retrievable filters and how they may differ from permanent filters warrants further investigation.

Some studies have also noted the development of PE after filter retrieval.35, 36 It is possible that a subclinical DVT was present at the time of removal or that the filter was retrieved before the risk of thrombosis had resolved. Therefore, consideration should be given to the use of duplex ultrasound evaluation for DVT prior to filter removal to ensure that patients with active thrombosis receive therapeutic anticoagulation for an appropriate duration.

Because of the concern for DVT and PE associated with retrievable filters, anticoagulation should ideally occur before and after retrieval, once the bleeding risk has become acceptable. Consensus guidelines support this practice,5, 37 though one systematic review has found insufficient evidence regarding the use of anticoagulation in patients with vena cava filters.4 Retrospective reviews have shown that filters can be both placed and removed without bleeding complications, even in patients who are therapeutically anticoagulated with warfarin and/or LMWH.38, 39 Further investigation would be useful to confirm whether this is an effective approach to VTE prevention at the time of retrieval.

Other Adverse Events

Other complications that have been associated with retrievable filters include migration, fracture, infection, and perforation. It may be difficult to estimate the true incidence of these complications, as most of the literature on this topic comes from case reports. Vena cava perforation with hooks may be not uncommon but in most cases is not clinically significant.40 Filter fracture is more common but rarely reported. Filter migration toward the heart is a very rare but potentially life‐threatening complication. The Recovery filter was taken off the market due in part to concerns about migration.26 As the use of retrievable filters increases, complications related to filters will need to be monitored.

Ongoing and Future Research

Other types of removable filters are currently in development. Convertible filters that can be converted into a stent once they are no longer needed are under investigation. Other devices, such as absorbable or drug‐eluting filters, are also being studied.5 In addition, there is ongoing research to better characterize the safety and efficacy of available filters. The Prevention du Risque d'Embolie Pulmonaire par Interruption Cave (PREPIC) 2 will assess their use in the first prospective, randomized, controlled trial of retrievable filters in patients with acute VTE receiving anticoagulation (http://www.clinicaltrials.gov; Identifier: NCT00457158). Other studies include an evaluation of the long‐term outcomes of patients with retrievable filters who failed retrieval (http://www.clinicaltrials.gov; Identifier: NCT00163956) and a comparison of Gnther Tulip and OptEase filters (http://www. clinicaltrials.gov; Identifier: NCT00588757). Randomized controlled trials are still needed to evaluate the efficacy of prophylactic filter placement in high‐risk patients. Studies that examine intention to retrieve vs. actual and recommended retrieval rates would provide valuable information on practice patterns.

Study Site

We studied CVC placement and thoracenteses performed by internal medicine residents at a 556‐bed Boston teaching hospital in 2003‐2004. During the 9‐month study period, 63 PGY1 residents (16 in a 1‐year preliminary program) and 95 PGY2 and PGY3 residents were enrolled in the program.

The MPS was staffed by hospitalists and pulmonologists skilled in teaching and performing 4 common inpatient procedures: CVC placement, thoracentesis, lumbar puncture, and paracentesis. We chose to study only the first 2 procedures because supervision of CVCs and thoracenteses by pulmonologists was available 24 hours a day in this initial year of the MPS. The other procedures were supervised by hospitalists during business hours only at the time. Ultrasound guidance was available for all procedures, supervised or not. At the time of the study, the residency program recommended consulting the MPS for procedures, but this was not mandatory. A resident electing to use the MPS to supervise a procedure on her own patient would page the MPS physician. If she were performing a procedure for the first time, she was required to review an online multimedia curriculum and complete a 5‐question cognitive test. She would then perform the procedure while supervised by the MPS physician, who would complete a checklist evaluation of the resident's performance online. All residents performing procedures, regardless of use of the MPS, would also complete procedure logs online to document procedural experience for the American Board of Internal Medicine requirements.

Study Design and Data Sources

We prospectively collected data from resident procedure logs from July 2003 through April 2004. We elicited the following information from the residents for each procedure: name of operator, year of training, date of procedure, patient's medical record number, name of attending or fellow supervisor, procedure, immediate complications (pneumothorax, bleeding, other, or none), self‐reported level of urgency (emergent, urgent, elective), time of day, procedure location, and the number of such procedures completed previously. We categorized level of supervision as: (1) MPS‐supervised if a pulmonary attending or interventional pulmonary fellow were listed as the supervisor (entailing formal faculty development as MPS faculty, resident use of the curriculum, and completion of faculty evaluations with structured feedback); (2) informally supervised if nonpulmonary attendings or fellows were involved (who may not supervise the entire procedure and would not complete a faculty evaluation); and (3) unsupervised if a resident physician or no supervisor was identified. Faculty development involved a single training session with the interventional pulmonary fellows and attendings and focused on optimal procedural teaching. During the session, we described the structure of the MPS, provided the curricular materials available to the residents, and reviewed the faculty evaluation forms in depth.

We abstracted patient characteristics (age, race/ethnicity, type of insurance, length of stay) from the electronic medical record. We performed retrospective chart reviews to record patient comorbidities (as defined by modified Deyo criteria3), to determine the number of medications associated with the risk of bleeding (such as anticoagulants and antiplatelet agents), and to discover complications that arose after the procedure was logged, including delayed bleeding, pneumothorax, or infection (localized site infection or line‐related bloodstream infection).

Data Analyses

We tabulated characteristics of residents (training level, gender, and self‐reported number of procedures), procedures (procedure type, procedure location, level of urgency, time of day), and patients (number of comorbidities and number of medications that promote bleeding) by use of the MPS. We combined resident‐reported (ie, immediate) complications and delayed complications (identified on retrospective chart review), stratified by use of the MPS. We also performed a subgroup analysis of non‐MPS procedures by comparing resident, procedure, and patient characteristics by presence or absence of informal supervision.

We created a univariable logistic regression model to examine factors associated with elective use of the MPS. We dichotomized the following independent variables: resident characteristics (PGY status, female gender, first time performing the procedure), patient characteristics (nonwhite race, female gender, Medicaid recipient, 3 or more comorbidities, any bleeding medication), and procedure characteristics (intensive care unit procedures, nonelective procedures, procedures performed between 11 _PM and 8 _AM). We also included 2 patient‐related interval variables (age and length of stay) in the univariable logistic regression model. We created a multivariable logistic regression model with backward elimination (P < 0.05) using the same independent variables as in the univariable analyses, to identify factors associated with use of the MPS, clustering by resident. We repeated this method to create a multivariable model to examine factors associated with the use of informal supervision among non‐MPS procedures. Analyses used Stata 7.0 (StataCorp, College Station, TX).

The study protocol was approved in advance by the hospital investigational review board.

Resident Characteristics

Sixty‐nine residents reported procedures during the 9‐month study period (Table 1). Thirty (43%) residents were PGY1 and 36 (52%) were female. Twelve (17%) residents performed the procedure for the first time.

Patient Characteristics

The 134 patients in the study had a mean age of 65.6 years. One‐half of patients were female, and 34% were nonwhite. The principal insurer was Medicare (57%); 24% were privately insured, and 17% received Medicaid. The mean length of stay was 18.4 days (range, 0‐98 days).

MPS and Non‐MPS Procedures

As detailed in the bivariate analyses in Table 2, residents performed 191 procedures (156 CVCs and 35 thoracenteses). PGY1 residents performed approximately one‐half of the 79 MPS procedures. Fifty‐one (65%) of the 79 MPS procedures were CVC placements and 28 (35%) were thoracenteses (P < 0.001). MPS procedures were less often performed in the emergency department than non‐MPS procedures (1% versus 21%, P < 0.001). There was no significant difference in the percentage of MPS and non‐MPS procedures by time of day. Patients whose procedures were supervised by the MPS had on average 3.0 comorbidities, while patients who underwent non‐MPS procedures had 2.6 comorbidities (P = 0.02). Complications occurred in 11 (14%) of MPS and 22 (20%) of non‐MPS procedures, a statistically nonsignificant difference.

In the univariable analysis, the only variable associated with elective use of the MPS was the presence of 3 or more comorbidities (oodds ratio [OR], 2.3; 95% confidence interval [CI], 1.2‐4.1). In the multivariable analysis, residents were more likely to use the MPS when patients had 3 or more comorbidities (OR, 2.1; 95% CI, 1.2‐3.5) and less likely to use the MPS when procedures were either urgent or emergent (OR, 0.4; 95% CI, 0.2‐0.8).

Unsupervised and Informally Supervised Procedures

Table 3 shows the results of the bivariate analyses of the characteristics of the 112 procedures that were unsupervised or supervised by non‐MPS physicians. Twenty‐seven (24%) were informally supervised by nonpulmonary attendings. Residents who had performed more than 6 procedures previously were more likely to be informally supervised than not supervised at all (P = 0.001). More informally supervised procedures were performed in the emergency department (41%) than in other settings (P = 0.01). There were no significant differences in year of training, gender, urgency, time of day, complications, comorbidities, or bleeding medications.

In the multivariable analysis, the only factor associated with the use of informal supervision (rather than absent supervision) was patient gender; informal supervision was less likely with female patients (OR, 0.3; 95% CI, 0.1‐0.8).

Study Site

We studied CVC placement and thoracenteses performed by internal medicine residents at a 556‐bed Boston teaching hospital in 2003‐2004. During the 9‐month study period, 63 PGY1 residents (16 in a 1‐year preliminary program) and 95 PGY2 and PGY3 residents were enrolled in the program.

The MPS was staffed by hospitalists and pulmonologists skilled in teaching and performing 4 common inpatient procedures: CVC placement, thoracentesis, lumbar puncture, and paracentesis. We chose to study only the first 2 procedures because supervision of CVCs and thoracenteses by pulmonologists was available 24 hours a day in this initial year of the MPS. The other procedures were supervised by hospitalists during business hours only at the time. Ultrasound guidance was available for all procedures, supervised or not. At the time of the study, the residency program recommended consulting the MPS for procedures, but this was not mandatory. A resident electing to use the MPS to supervise a procedure on her own patient would page the MPS physician. If she were performing a procedure for the first time, she was required to review an online multimedia curriculum and complete a 5‐question cognitive test. She would then perform the procedure while supervised by the MPS physician, who would complete a checklist evaluation of the resident's performance online. All residents performing procedures, regardless of use of the MPS, would also complete procedure logs online to document procedural experience for the American Board of Internal Medicine requirements.

Study Design and Data Sources

We prospectively collected data from resident procedure logs from July 2003 through April 2004. We elicited the following information from the residents for each procedure: name of operator, year of training, date of procedure, patient's medical record number, name of attending or fellow supervisor, procedure, immediate complications (pneumothorax, bleeding, other, or none), self‐reported level of urgency (emergent, urgent, elective), time of day, procedure location, and the number of such procedures completed previously. We categorized level of supervision as: (1) MPS‐supervised if a pulmonary attending or interventional pulmonary fellow were listed as the supervisor (entailing formal faculty development as MPS faculty, resident use of the curriculum, and completion of faculty evaluations with structured feedback); (2) informally supervised if nonpulmonary attendings or fellows were involved (who may not supervise the entire procedure and would not complete a faculty evaluation); and (3) unsupervised if a resident physician or no supervisor was identified. Faculty development involved a single training session with the interventional pulmonary fellows and attendings and focused on optimal procedural teaching. During the session, we described the structure of the MPS, provided the curricular materials available to the residents, and reviewed the faculty evaluation forms in depth.

We abstracted patient characteristics (age, race/ethnicity, type of insurance, length of stay) from the electronic medical record. We performed retrospective chart reviews to record patient comorbidities (as defined by modified Deyo criteria3), to determine the number of medications associated with the risk of bleeding (such as anticoagulants and antiplatelet agents), and to discover complications that arose after the procedure was logged, including delayed bleeding, pneumothorax, or infection (localized site infection or line‐related bloodstream infection).

Data Analyses

We tabulated characteristics of residents (training level, gender, and self‐reported number of procedures), procedures (procedure type, procedure location, level of urgency, time of day), and patients (number of comorbidities and number of medications that promote bleeding) by use of the MPS. We combined resident‐reported (ie, immediate) complications and delayed complications (identified on retrospective chart review), stratified by use of the MPS. We also performed a subgroup analysis of non‐MPS procedures by comparing resident, procedure, and patient characteristics by presence or absence of informal supervision.

We created a univariable logistic regression model to examine factors associated with elective use of the MPS. We dichotomized the following independent variables: resident characteristics (PGY status, female gender, first time performing the procedure), patient characteristics (nonwhite race, female gender, Medicaid recipient, 3 or more comorbidities, any bleeding medication), and procedure characteristics (intensive care unit procedures, nonelective procedures, procedures performed between 11 _PM and 8 _AM). We also included 2 patient‐related interval variables (age and length of stay) in the univariable logistic regression model. We created a multivariable logistic regression model with backward elimination (P < 0.05) using the same independent variables as in the univariable analyses, to identify factors associated with use of the MPS, clustering by resident. We repeated this method to create a multivariable model to examine factors associated with the use of informal supervision among non‐MPS procedures. Analyses used Stata 7.0 (StataCorp, College Station, TX).

The study protocol was approved in advance by the hospital investigational review board.

Resident Characteristics

Sixty‐nine residents reported procedures during the 9‐month study period (Table 1). Thirty (43%) residents were PGY1 and 36 (52%) were female. Twelve (17%) residents performed the procedure for the first time.

Patient Characteristics

The 134 patients in the study had a mean age of 65.6 years. One‐half of patients were female, and 34% were nonwhite. The principal insurer was Medicare (57%); 24% were privately insured, and 17% received Medicaid. The mean length of stay was 18.4 days (range, 0‐98 days).

MPS and Non‐MPS Procedures

As detailed in the bivariate analyses in Table 2, residents performed 191 procedures (156 CVCs and 35 thoracenteses). PGY1 residents performed approximately one‐half of the 79 MPS procedures. Fifty‐one (65%) of the 79 MPS procedures were CVC placements and 28 (35%) were thoracenteses (P < 0.001). MPS procedures were less often performed in the emergency department than non‐MPS procedures (1% versus 21%, P < 0.001). There was no significant difference in the percentage of MPS and non‐MPS procedures by time of day. Patients whose procedures were supervised by the MPS had on average 3.0 comorbidities, while patients who underwent non‐MPS procedures had 2.6 comorbidities (P = 0.02). Complications occurred in 11 (14%) of MPS and 22 (20%) of non‐MPS procedures, a statistically nonsignificant difference.

In the univariable analysis, the only variable associated with elective use of the MPS was the presence of 3 or more comorbidities (oodds ratio [OR], 2.3; 95% confidence interval [CI], 1.2‐4.1). In the multivariable analysis, residents were more likely to use the MPS when patients had 3 or more comorbidities (OR, 2.1; 95% CI, 1.2‐3.5) and less likely to use the MPS when procedures were either urgent or emergent (OR, 0.4; 95% CI, 0.2‐0.8).

Unsupervised and Informally Supervised Procedures

Table 3 shows the results of the bivariate analyses of the characteristics of the 112 procedures that were unsupervised or supervised by non‐MPS physicians. Twenty‐seven (24%) were informally supervised by nonpulmonary attendings. Residents who had performed more than 6 procedures previously were more likely to be informally supervised than not supervised at all (P = 0.001). More informally supervised procedures were performed in the emergency department (41%) than in other settings (P = 0.01). There were no significant differences in year of training, gender, urgency, time of day, complications, comorbidities, or bleeding medications.

In the multivariable analysis, the only factor associated with the use of informal supervision (rather than absent supervision) was patient gender; informal supervision was less likely with female patients (OR, 0.3; 95% CI, 0.1‐0.8).

Study Site

We studied CVC placement and thoracenteses performed by internal medicine residents at a 556‐bed Boston teaching hospital in 2003‐2004. During the 9‐month study period, 63 PGY1 residents (16 in a 1‐year preliminary program) and 95 PGY2 and PGY3 residents were enrolled in the program.

The MPS was staffed by hospitalists and pulmonologists skilled in teaching and performing 4 common inpatient procedures: CVC placement, thoracentesis, lumbar puncture, and paracentesis. We chose to study only the first 2 procedures because supervision of CVCs and thoracenteses by pulmonologists was available 24 hours a day in this initial year of the MPS. The other procedures were supervised by hospitalists during business hours only at the time. Ultrasound guidance was available for all procedures, supervised or not. At the time of the study, the residency program recommended consulting the MPS for procedures, but this was not mandatory. A resident electing to use the MPS to supervise a procedure on her own patient would page the MPS physician. If she were performing a procedure for the first time, she was required to review an online multimedia curriculum and complete a 5‐question cognitive test. She would then perform the procedure while supervised by the MPS physician, who would complete a checklist evaluation of the resident's performance online. All residents performing procedures, regardless of use of the MPS, would also complete procedure logs online to document procedural experience for the American Board of Internal Medicine requirements.

Study Design and Data Sources

We prospectively collected data from resident procedure logs from July 2003 through April 2004. We elicited the following information from the residents for each procedure: name of operator, year of training, date of procedure, patient's medical record number, name of attending or fellow supervisor, procedure, immediate complications (pneumothorax, bleeding, other, or none), self‐reported level of urgency (emergent, urgent, elective), time of day, procedure location, and the number of such procedures completed previously. We categorized level of supervision as: (1) MPS‐supervised if a pulmonary attending or interventional pulmonary fellow were listed as the supervisor (entailing formal faculty development as MPS faculty, resident use of the curriculum, and completion of faculty evaluations with structured feedback); (2) informally supervised if nonpulmonary attendings or fellows were involved (who may not supervise the entire procedure and would not complete a faculty evaluation); and (3) unsupervised if a resident physician or no supervisor was identified. Faculty development involved a single training session with the interventional pulmonary fellows and attendings and focused on optimal procedural teaching. During the session, we described the structure of the MPS, provided the curricular materials available to the residents, and reviewed the faculty evaluation forms in depth.

We abstracted patient characteristics (age, race/ethnicity, type of insurance, length of stay) from the electronic medical record. We performed retrospective chart reviews to record patient comorbidities (as defined by modified Deyo criteria3), to determine the number of medications associated with the risk of bleeding (such as anticoagulants and antiplatelet agents), and to discover complications that arose after the procedure was logged, including delayed bleeding, pneumothorax, or infection (localized site infection or line‐related bloodstream infection).

Data Analyses

We tabulated characteristics of residents (training level, gender, and self‐reported number of procedures), procedures (procedure type, procedure location, level of urgency, time of day), and patients (number of comorbidities and number of medications that promote bleeding) by use of the MPS. We combined resident‐reported (ie, immediate) complications and delayed complications (identified on retrospective chart review), stratified by use of the MPS. We also performed a subgroup analysis of non‐MPS procedures by comparing resident, procedure, and patient characteristics by presence or absence of informal supervision.

We created a univariable logistic regression model to examine factors associated with elective use of the MPS. We dichotomized the following independent variables: resident characteristics (PGY status, female gender, first time performing the procedure), patient characteristics (nonwhite race, female gender, Medicaid recipient, 3 or more comorbidities, any bleeding medication), and procedure characteristics (intensive care unit procedures, nonelective procedures, procedures performed between 11 _PM and 8 _AM). We also included 2 patient‐related interval variables (age and length of stay) in the univariable logistic regression model. We created a multivariable logistic regression model with backward elimination (P < 0.05) using the same independent variables as in the univariable analyses, to identify factors associated with use of the MPS, clustering by resident. We repeated this method to create a multivariable model to examine factors associated with the use of informal supervision among non‐MPS procedures. Analyses used Stata 7.0 (StataCorp, College Station, TX).

The study protocol was approved in advance by the hospital investigational review board.

Resident Characteristics

Sixty‐nine residents reported procedures during the 9‐month study period (Table 1). Thirty (43%) residents were PGY1 and 36 (52%) were female. Twelve (17%) residents performed the procedure for the first time.

Patient Characteristics

The 134 patients in the study had a mean age of 65.6 years. One‐half of patients were female, and 34% were nonwhite. The principal insurer was Medicare (57%); 24% were privately insured, and 17% received Medicaid. The mean length of stay was 18.4 days (range, 0‐98 days).

MPS and Non‐MPS Procedures

As detailed in the bivariate analyses in Table 2, residents performed 191 procedures (156 CVCs and 35 thoracenteses). PGY1 residents performed approximately one‐half of the 79 MPS procedures. Fifty‐one (65%) of the 79 MPS procedures were CVC placements and 28 (35%) were thoracenteses (P < 0.001). MPS procedures were less often performed in the emergency department than non‐MPS procedures (1% versus 21%, P < 0.001). There was no significant difference in the percentage of MPS and non‐MPS procedures by time of day. Patients whose procedures were supervised by the MPS had on average 3.0 comorbidities, while patients who underwent non‐MPS procedures had 2.6 comorbidities (P = 0.02). Complications occurred in 11 (14%) of MPS and 22 (20%) of non‐MPS procedures, a statistically nonsignificant difference.