Chronic obstructive pulmonary disease (COPD) is now the third leading cause of death in the United States,[1] and its rising mortality trend is unique among the top 5 causes of death.[2] Acute exacerbations of COPD (AECOPD) are important events in the natural history of COPD, accounting for 1.5 million emergency department (ED) visits and 726,000 hospitalizations each year in the United States.[3, 4] Given the significant morbidity and mortality from AECOPD, Healthy People 2020 lists reducing deaths, hospitalizations, and ED visits as the key objectives for COPD.[5]

Over the past 2 decades, noninvasive ventilation (NIV) has emerged as a potentially useful treatment modality in AECOPD patients with acute respiratory failure. Noninvasive ventilation commonly refers to positive‐pressure ventilatory support delivered through a nasal or full‐face mask, such as bilevel positive airway pressure.[6] A number of randomized controlled trials[7, 8, 9] and meta‐analyses[10] have suggested a mortality‐reduction benefit with NIV use compared with standard medical care in AECOPD. To our knowledge, however, very few small randomized controlled trials compared NIV vs invasive mechanical ventilation (IMV) head‐to‐head,[11, 12, 13] and a recent evidence review found only 5 studies (405 subjects) on this topic.[14] Collectively, the limited evidence from randomized trials showed that NIV use resulted in similar intensive care unit (ICU) and in‐hospital mortality, fewer complications (eg, ventilator‐associated pneumonia and sepsis), and shorter hospital length of stays (LOS). Given that these trials have a smaller sample size and tend to exclude older patient (age >75 years) or patients with multiple comorbidities, there is a need to better understand the adoption and effectiveness of NIV treatment for AECOPD in a much larger patient population in the real‐world setting using observational data.

To address these knowledge gaps in the literature, we analyzed data from a large, nationally representative ED and inpatient sample. The objective of the present analysis was 2‐fold: (1) to characterize the use of NIV and IMV in AECOPD patients with acute respiratory failure at a national level; and (2) to compare the effectiveness of NIV vs IMV in the real‐world setting.

METHODS

RESULTS

Patient and ED Characteristics

The 20062008 NEDS sample contained 67,651 ED visits for AECOPD with acute respiratory failure from 1594 US EDs. After the weighting procedure, there were an estimated 101,000 visits annually for AECOPD with acute respiratory failure from approximately 4700 US EDs. In the weighted analysis, the mean patient age of these visits was 68 years, and 56% were made by women. Ninety‐six percent were admitted to the hospital. Of these, the mortality rate was 9% and the mean hospital LOS was 7 days. Figure 1 shows the secular trends in NIV, IMV, and the combined use over the 3‐year study period. Use of IMV decreased from 28% in 2006 to 19% in 2008 (P<0.001), whereas NIV use increased slightly from 14% in 2006 to 16% in 2008 (P=0.049); the combined use of both ventilation modalities remained stable (4%). Inpatient mortality decreased from 10% in 2006 to 7% in 2008 (P<0.001).

Figure 1

Figure 2 shows that the frequency of NIV use (including combined use of NIV and IMV) varied widely between hospitals, ranging from 0% to 100% with a median of 11%. In the unweighted cohort of AECOPD with acute respiratory failure, 43% received some forms of ventilatory support. Table 1 shows the patient and hospital characteristics of the patients receiving ventilatory support: 36% received NIV, 56% received IMV, and 8% received combined use. In general, patients receiving combined use of NIV and IMV tended to have more comorbidities (eg, congestive heart failure and pneumonia) compared with the NIV‐alone or IMV‐alone groups. With respect to hospital characteristics, NIV was used more often in hospitals with higher volumes of COPD exacerbation and respiratory failure, in nonmetropolitan hospitals, and in hospitals in the Northeast.

Figure 2

Table 1.Patient and Hospital Characteristics According to Ventilatory Mode

	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value, A vs B	P Value, B vs C
NOTE: Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; ED, emergency department; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation.
Patient characteristics
Age, y,				<0.001	0.64
4049	5	5	5
5059	17	18	19
6069	31	33	33
7079	30	29	29
80	17	15	13
Female sex, %	57	53	54	<0.001	0.87
Quartile for median household income of patient ZIP code, $, %				<0.001	<0.001
138,999	30	34	29
39,00047,999	28	28	28
48,00062,999	24	22	24
63,000	18	15	19
Weekend admission, %	27	28	28	0.07	0.80
Insurance status, %				<0.001	0.91
Medicare	74	70	70
Medicaid	9	12	12
Private	12	13	13
Self‐pay	2	3	2
Other	2	2	2
Season, %				<0.001	0.16
Winter (January 1March 31)	29	32	31
Spring (April 1June 30)	24	25	26
Summer (July 1September 30)	22	20	19
Fall (October 1December 31)	25	22	24
No. of comorbidities, median (IQR)	4 (35)	4 (35)	4 (36)	<0.001	<0.001
Selected comorbidities, %
Hypertension	56	55	55	0.01	0.65
CHF	38	40	44	0.001	<0.001
Fluid and electrolyte disorders	37	44	49	<0.001	<0.001
Diabetes, uncomplicated	27	26	29	0.04	0.002
Pneumonia	19	34	39	<0.001	<0.001
Deficiency anemia	16	19	19	<0.001	0.39
Obesity	18	12	17	<0.001	<0.001
Depression	15	11	11	<0.001	0.54
Pulmonary circulatory diseases	15	11	14	<0.001	<0.001
Hospital characteristics
Annual ED visit volume, median (IQR)	42,704 (29,50562,470)	44,119 (29,89564,097)	46,695 (31,29866,235)	0.02	0.0003
Annual ED volume of COPD exacerbation with respiratory failure, median (IQR)	45 (2672)	42 (2368)	38 (2364)	<0.001	<0.001
Urban/rural and teaching status, %				<0.001	<0.001
Metropolitan nonteaching	53	52	47
Metropolitan teaching	31	35	39
Nonmetropolitan	16	13	13
US region, %				<0.001	<0.001
Northeast	28	16	36
Midwest	17	22	15
South	41	45	32
West	14	17	17

The unadjusted differences in outcomes are shown in Table 2. The combined‐use group had the highest inpatient mortality, longest LOS, and highest charges, followed by the IMV and NIV groups. In general, complications were few across all 3 groups, but the rate of iatrogenic pneumothorax was notably lower in the NIV group. Table 3 details the statistically significant predictors of NIV use in the propensity score model. Similar to the unadjusted analysis, older age, high‐income neighborhoods, Medicare insurance, and some comorbidities were positively associated with NIV use (eg, pulmonary circulatory disorders and liver disease), whereas a few comorbidities were negatively associated with NIV use (eg, pneumonia, and alcohol and drug abuse). With respect to hospital characteristics, higher case volumes of COPD exacerbation/respiratory failure, Northeastern and nonmetropolitan hospitals, and more recent years were associated with NIV use.

Table 2.Clinical Outcomes Among COPD Patients With Acute Respiratory Failure According to Ventilatory Mode

Outcome	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value. A vs B	P Value, B vs C
NOTE: Abbreviations: COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation. *For privacy protection, we are not able to report cells in the tables 10 individual records.
Inpatient mortality, n (%)	825 (8)	2,454 (16)	407 (18)	<0.001	0.04
Hospital length of stay, median (IQR), d	5 (48)	8 (513)	10 (716)	<0.001	<0.001
Hospital charge per visit, median (IQR), $	26,002 (15,74744,638)	53,432 (31,99892,664)	64,585 (39,024110,336)	<0.001	<0.001
Complications*
Ventilator‐associated pneumonia, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.09	1.00
Facial injury, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.26	1.00
Iatrogenic pneumothorax, n (%)	10 (0.1)	90 (0.6)	14 (0.6)	<0.001	0.90

Table 3.Statistically Significant Predictors of NIV Use Alone

Patient Characteristics	Adjusted OR (95% CI)*	P Value
NOTE: Abbreviations: AIDS, acquired immune deficiency syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; ED, emergency department; NIV, noninvasive ventilation; OR, odds ratio; RA, rheumatoid arthritis. *Propensity score model included the following patient and hospital characteristics: age, sex, median household income, insurance status, weekend admission, season, comorbid conditions, US region, urban/rural and teaching status, annual ED volume, annual ED volume of COPD with respiratory failure, and calendar year. Nonsignificant predictors that are not reported in the table include sex, weekend admission, and the following comorbid conditions: peripheral vascular disorders, paralysis, uncomplicated diabetes, hypothyroidism, renal failure, peptic ulcer disease excluding bleeding, hypertension, lymphoma, AIDS, solid tumor without metastasis, and metastatic cancer.
Age, y
4049	1.00 (Reference)
5059	0.96 (0.84‐1.11)	0.61
6069	0.96 (0.84‐1.10)	0.56
7079	1.09 (0.94‐1.25)	0.25
80	1.30 (1.12‐1.52)	0.001
Quartile for median household income of patient ZIP code, $
138,999	1.00 (Reference)
39,00047,999	1.13 (1.05‐1.21)	0.001
48,00062,999	1.21 (1.12‐1.30)	<0.001
63,000	1.21 (1.11‐1.32)	<0.001
Insurance status
Medicare	1.00 (Reference)
Medicaid	0.79 (0.72‐0.88)	<0.001
Private	0.88 (0.81‐0.96)	0.004
Self‐pay	0.68 (0.56‐0.82)	<0.001
Other	0.88 (0.73‐1.07)	0.22
Season
Winter (January 1March 31)	1.00 (Reference)
Spring (April 1June 30)	1.06 (0.99‐1.14)	0.11
Summer (July 1September 30)	1.17 (1.08‐1.26)	<0.001
Fall (October 1December 31)	1.24 (1.15‐1.33)	<0.001
Comorbidity
CHF	0.90 (0.85‐0.95)	<0.001
Pulmonary circulatory disorders	1.40 (1.29‐1.52)	<0.001
Diabetes, complicated	1.25 (1.08‐1.44)	0.002
Liver disease	1.79 (1.40‐2.28)	<0.001
Coagulopathy	0.54 (0.46‐0.63)	<0.001
Obesity	1.52 (1.41‐1.65)	<0.001
Weight loss	0.50 (0.44‐0.57)	<0.001
Fluid and electrolyte disorders	0.84 (0.80‐0.89)	<0.001
Deficiency anemia	0.83 (0.78‐0.90)	<0.001
Alcohol abuse	0.66 (0.58‐0.76)	<0.001
Drug abuse	0.74 (0.62‐0.88)	0.001
Psychoses	1.22 (1.10‐1.37)	<0.001
Depression	1.45 (1.34‐1.57)	<0.001
Pneumonia	0.48 (0.45‐0.51)	<0.001
Valvular heart disease	0.87 (0.77‐0.97)	0.01
Neurological disorders	0.89 (0.80‐0.98)	0.02
RA/collagen vascular diseases	1.25 (1.02‐1.53)	0.04
Blood‐loss anemia	0.72 (0.53‐0.97)	0.03
Hospital characteristics
Annual ED visit volume, per 1000‐visit increase	0.997 (0.996‐0.998)	<0.001
Annual ED volume of COPD exacerbation with respiratory failure, per 10‐visit increase	1.03 (1.02‐1.03)	<0.001
Urban/rural and teaching status
Metropolitan nonteaching	1.00 (Reference)
Metropolitan teaching	0.91 (0.85‐0.97)	0.006
Nonmetropolitan	1.30 (1.20‐1.42)	<0.001
US region
Northeast	1.00 (Reference)
Midwest	0.44 (0.40‐0.48)	<0.001
South	0.54 (0.50‐0.58)	<0.001
West	0.51 (0.46‐0.56)	<0.001
Calendar year
2006	1.00 (Reference)
2007	1.30 (1.22‐1.39)	<0.001
2008	1.65 (1.54‐1.76)	<0.001

In terms of propensity score distributions (see Supporting Information, Figure E1, in the online version of this article), there was sufficient overlap of the NIV and IMV groups. After matching on propensity score for the NIV and IMV groups, the differences in baseline characteristics were all balanced (see Supporting Information, Table E1, in the online version of this article), as indicated by <10% standardized differences in all covariates between the 2 groups. Finally, in the propensity scorematched cohort (see Supporting Information, Table E2, in the online version of this article), NIV use remained associated with significantly lower inpatient mortality (risk ratio: 0.54; 95% CI: 0.50‐0.59, P<0.001), a shorter hospital LOS (mean difference, 3.2 days; 95% CI: 3.4 to 2.9 days, P<0.001), and lower hospital charges (mean difference, P<$35,012; 95% CI: $36,848 to $33,176, P<0.001), compared with IMV use. Use of NIV also was associated with a lower rate of iatrogenic pneumothorax than IMV use (0.05% vs 0.5%, P<0.001).

Using hospital preference for NIV vs IMV as an instrument, the instrumental analysis confirmed the benefits of NIV use, with a 5% reduction in inpatient mortality in the NIV‐preferring hospitals (risk difference, P<5%; 95% CI: P<1.8% to P<8.3%).

In the sensitivity analysis to assess the impact of an unmeasured confounder, the confounder would have had to have a very strong impact on outcome (risk ratio: 5) and a severe exposure‐confounder imbalance (odds ratio of exposure on confounder: 5) to reduce the observed association to 1.0. In other words, an individual unmeasured confounder is unlikely to explain the observed association.

DISCUSSION

In this nationally representative sample of 67,651 ED visits for AECOPD with acute respiratory failure, we found that NIV use was increasing from 2006 to 2008. However, the utilization of NIV remained low (16% in 2008) and varied widely by patient and hospital characteristic. As with all observational studies, causality cannot be inferred definitely; however, our study suggests that, NIV usecompared with IMV usewas associated with potentially important benefits: a reduction of inpatient mortality by 46%, shortened hospital LOS by 3 days, reduced hospital charges by approximately $35,000 per visit, and modestly reduced risk of iatrogenic pneumothorax.

A recent analysis using the US Nationwide Inpatient Sample has shown increasing use of NIV and concomitant decreasing mortality in AECOPD over time.[28] Our analysis confirmed these favorable trends in the United States using a much larger NEDS sample (28 million visits in the NEDS vs 8 million visits in the Nationwide Inpatient Sample per year). Despite these favorable trends, NIV was still underutilized for AECOPD with respiratory failure in the United States (16% in 2008) compared with major European countries (40%).[29] Although our study lacked clinical details to arrive at the optimal rate of NIV use, the low rate of NIV use is concerning and suggests room for improvement in NIV use in appropriate patients as outlined by the current COPD guidelines.[18, 30] Why is NIV not widely adopted, given its demonstrated efficacy? Previous surveys have identified several perceived reasons for low NIV use, including lack of physician knowledge, insufficient respiratory therapist training, inadequate equipment, and time required for setting up NIV.[29, 31, 32] Our study adds to the literature by showing the actual predictors of NIV use in the real world. Our data showed that the early adopters were hospitals with higher case volumes, and hospitals in the Northeast and in nonmetropolitan areas. A higher case volume has been linked with lower mortality in AECOPD (ie, practice makes perfect),[33] and frequent NIV use could explain the lower AECOPD mortality in highcase volume centers. Alternatively, smaller hospitals tend to have moonlighters working in EDs who may not be board certified in emergency medicine. Perhaps the logical next step is to conduct a qualitative study to understand the specifics of best practices and provider characteristics in these Northeastern, highercase volume centers. Another incentive to promote NIV use in clinical practice is the cost‐effectiveness associated with this intervention, as previous studies have shown that, compared with usual care, receiving NIV was associated with a reduction in costs, mainly through reduced use of the ICU.[34, 35]

Some patient factors associated with NIV use may be well justified. For example, older AECOPD patients may have an advance directive describing their treatment wishes (eg, do‐not‐intubate order),[36] and therefore NIV was preferred to IMV. Also, our data suggested AECOPD patients with a suspected pneumonia component were less likely to be placed on NIV, which is consistent with COPD guideline recommendations.[18, 30] As outlined in the current guidelines, the major contraindications to NIV include impending respiratory arrest, excessive respiratory secretions, massive gastrointestinal bleeding, recent facial trauma, or altered mental status.[18, 30] By contrast, some factors associated with NIV use may be targeted for intervention, such as lower rates of NIV use in the uninsured, patients who live in low‐income neighborhoods, and hospitals in US regions other than the Northeast.

Current guidelines recommend using NIV in AECOPD patients with early signs of respiratory failure, such as arterial pH of 7.257.35 or pCO₂ 45 mm Hg.[18, 30] When NIV is considered as the modality of ventilatory support, it should probably be used as early as possible,[37] because evidence suggests that delayed use of NIV may lead to severe respiratory acidosis and increased mortality.[38] Other than in ICUs, NIV can be used on general wards and in EDs that have adequate staff training and experience, because the success rates of NIV in these settings are similar to those reported in ICU studies.[8, 36, 39] In addition, NIV is more cost‐effective when performed outside the ICU.[35] In fact, studies have found a substantial portion of patients had NIV started in the ED (one‐fourth) and on the general ward (one‐fourth).[31, 40] Given the shortage of intensivists in the United States, hospitalists begin to play an important role in provision of critical care outside the ICU.[41] Once NIV is used, it is important to ensure that it is delivered effectively and monitored closely because NIV failure has been shown to be associated with high mortality.[28, 42]

This study has some potential limitations. First, we used administrative claims that lack clinical details such as data on arterial blood gases and severity scores, and thus potential residual confounding may exist. In our study, the IMV group may be sicker than the NIV group, which could partially explain the increased mortality with IMV. However, the propensity scores overlap to a great extent between the 2 study groups, suggesting that a strong confounding bias is less likely, given the observed covariates. Furthermore, the instrumental variable and sensitivity analyses taking into account unmeasured confounders still suggested the benefits of NIV. Second, the NEDS does not contain data on the location where NIV was initiated (eg, ED, ward, or ICU) or the timing of initiating NIV or IMV. As a result, for the combined‐use group, we could not further distinguish the group switching from NIV to IMV (ie, NIV failure)[42] or from IMV to NIV (ie, NIV as a weaning strategy).[43] Accordingly, we chose to focus on the comparativeness effectiveness of NIV vs IMV. Third, although the NEDS data have undergone quality‐control procedures,[44] some misclassification may exist in identifying patient population and interventions. Finally, the analysis may not reflect the most recent trend in NIV use, as the 2010 NEDS data have just been released. In addition, although the study is the largest to date on this topic, our findings may not be generalizable to EDs that were not part of the NEDS.

In summary, in this nationally representative ED and inpatient database, NIV use is increasing for AECOPD with acute respiratory failure; however, its adoption remains low and varies widely between US hospitals. Our observational study suggests that NIV appears to be more effective and safer than IMV in the real‐world setting. There is an opportunity to increase the use of NIV as recommended in guidelines and to promote the use NIV in replacement of IMV in patients with severe AECOPD. Given the increasing mortality burden of COPD, such a strategy may help reduce COPD mortality at the population level, thereby fulfilling the objectives of Healthy People 2020.

Disclosure

Partial results from this study were presented at the 2012 Society for Academic Emergency Medicine Annual Meeting, Chicago, Illinois, May 912, 2012. This project was supported by grant number R03HS020722 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. The authors have no conflicts of interest to disclose.

Chronic obstructive pulmonary disease (COPD) is now the third leading cause of death in the United States,[1] and its rising mortality trend is unique among the top 5 causes of death.[2] Acute exacerbations of COPD (AECOPD) are important events in the natural history of COPD, accounting for 1.5 million emergency department (ED) visits and 726,000 hospitalizations each year in the United States.[3, 4] Given the significant morbidity and mortality from AECOPD, Healthy People 2020 lists reducing deaths, hospitalizations, and ED visits as the key objectives for COPD.[5]

Over the past 2 decades, noninvasive ventilation (NIV) has emerged as a potentially useful treatment modality in AECOPD patients with acute respiratory failure. Noninvasive ventilation commonly refers to positive‐pressure ventilatory support delivered through a nasal or full‐face mask, such as bilevel positive airway pressure.[6] A number of randomized controlled trials[7, 8, 9] and meta‐analyses[10] have suggested a mortality‐reduction benefit with NIV use compared with standard medical care in AECOPD. To our knowledge, however, very few small randomized controlled trials compared NIV vs invasive mechanical ventilation (IMV) head‐to‐head,[11, 12, 13] and a recent evidence review found only 5 studies (405 subjects) on this topic.[14] Collectively, the limited evidence from randomized trials showed that NIV use resulted in similar intensive care unit (ICU) and in‐hospital mortality, fewer complications (eg, ventilator‐associated pneumonia and sepsis), and shorter hospital length of stays (LOS). Given that these trials have a smaller sample size and tend to exclude older patient (age >75 years) or patients with multiple comorbidities, there is a need to better understand the adoption and effectiveness of NIV treatment for AECOPD in a much larger patient population in the real‐world setting using observational data.

To address these knowledge gaps in the literature, we analyzed data from a large, nationally representative ED and inpatient sample. The objective of the present analysis was 2‐fold: (1) to characterize the use of NIV and IMV in AECOPD patients with acute respiratory failure at a national level; and (2) to compare the effectiveness of NIV vs IMV in the real‐world setting.

METHODS

RESULTS

Patient and ED Characteristics

The 20062008 NEDS sample contained 67,651 ED visits for AECOPD with acute respiratory failure from 1594 US EDs. After the weighting procedure, there were an estimated 101,000 visits annually for AECOPD with acute respiratory failure from approximately 4700 US EDs. In the weighted analysis, the mean patient age of these visits was 68 years, and 56% were made by women. Ninety‐six percent were admitted to the hospital. Of these, the mortality rate was 9% and the mean hospital LOS was 7 days. Figure 1 shows the secular trends in NIV, IMV, and the combined use over the 3‐year study period. Use of IMV decreased from 28% in 2006 to 19% in 2008 (P<0.001), whereas NIV use increased slightly from 14% in 2006 to 16% in 2008 (P=0.049); the combined use of both ventilation modalities remained stable (4%). Inpatient mortality decreased from 10% in 2006 to 7% in 2008 (P<0.001).

Figure 1

Figure 2 shows that the frequency of NIV use (including combined use of NIV and IMV) varied widely between hospitals, ranging from 0% to 100% with a median of 11%. In the unweighted cohort of AECOPD with acute respiratory failure, 43% received some forms of ventilatory support. Table 1 shows the patient and hospital characteristics of the patients receiving ventilatory support: 36% received NIV, 56% received IMV, and 8% received combined use. In general, patients receiving combined use of NIV and IMV tended to have more comorbidities (eg, congestive heart failure and pneumonia) compared with the NIV‐alone or IMV‐alone groups. With respect to hospital characteristics, NIV was used more often in hospitals with higher volumes of COPD exacerbation and respiratory failure, in nonmetropolitan hospitals, and in hospitals in the Northeast.

Figure 2

Table 1.Patient and Hospital Characteristics According to Ventilatory Mode

	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value, A vs B	P Value, B vs C
NOTE: Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; ED, emergency department; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation.
Patient characteristics
Age, y,				<0.001	0.64
4049	5	5	5
5059	17	18	19
6069	31	33	33
7079	30	29	29
80	17	15	13
Female sex, %	57	53	54	<0.001	0.87
Quartile for median household income of patient ZIP code, $, %				<0.001	<0.001
138,999	30	34	29
39,00047,999	28	28	28
48,00062,999	24	22	24
63,000	18	15	19
Weekend admission, %	27	28	28	0.07	0.80
Insurance status, %				<0.001	0.91
Medicare	74	70	70
Medicaid	9	12	12
Private	12	13	13
Self‐pay	2	3	2
Other	2	2	2
Season, %				<0.001	0.16
Winter (January 1March 31)	29	32	31
Spring (April 1June 30)	24	25	26
Summer (July 1September 30)	22	20	19
Fall (October 1December 31)	25	22	24
No. of comorbidities, median (IQR)	4 (35)	4 (35)	4 (36)	<0.001	<0.001
Selected comorbidities, %
Hypertension	56	55	55	0.01	0.65
CHF	38	40	44	0.001	<0.001
Fluid and electrolyte disorders	37	44	49	<0.001	<0.001
Diabetes, uncomplicated	27	26	29	0.04	0.002
Pneumonia	19	34	39	<0.001	<0.001
Deficiency anemia	16	19	19	<0.001	0.39
Obesity	18	12	17	<0.001	<0.001
Depression	15	11	11	<0.001	0.54
Pulmonary circulatory diseases	15	11	14	<0.001	<0.001
Hospital characteristics
Annual ED visit volume, median (IQR)	42,704 (29,50562,470)	44,119 (29,89564,097)	46,695 (31,29866,235)	0.02	0.0003
Annual ED volume of COPD exacerbation with respiratory failure, median (IQR)	45 (2672)	42 (2368)	38 (2364)	<0.001	<0.001
Urban/rural and teaching status, %				<0.001	<0.001
Metropolitan nonteaching	53	52	47
Metropolitan teaching	31	35	39
Nonmetropolitan	16	13	13
US region, %				<0.001	<0.001
Northeast	28	16	36
Midwest	17	22	15
South	41	45	32
West	14	17	17

The unadjusted differences in outcomes are shown in Table 2. The combined‐use group had the highest inpatient mortality, longest LOS, and highest charges, followed by the IMV and NIV groups. In general, complications were few across all 3 groups, but the rate of iatrogenic pneumothorax was notably lower in the NIV group. Table 3 details the statistically significant predictors of NIV use in the propensity score model. Similar to the unadjusted analysis, older age, high‐income neighborhoods, Medicare insurance, and some comorbidities were positively associated with NIV use (eg, pulmonary circulatory disorders and liver disease), whereas a few comorbidities were negatively associated with NIV use (eg, pneumonia, and alcohol and drug abuse). With respect to hospital characteristics, higher case volumes of COPD exacerbation/respiratory failure, Northeastern and nonmetropolitan hospitals, and more recent years were associated with NIV use.

Table 2.Clinical Outcomes Among COPD Patients With Acute Respiratory Failure According to Ventilatory Mode

Outcome	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value. A vs B	P Value, B vs C
NOTE: Abbreviations: COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation. *For privacy protection, we are not able to report cells in the tables 10 individual records.
Inpatient mortality, n (%)	825 (8)	2,454 (16)	407 (18)	<0.001	0.04
Hospital length of stay, median (IQR), d	5 (48)	8 (513)	10 (716)	<0.001	<0.001
Hospital charge per visit, median (IQR), $	26,002 (15,74744,638)	53,432 (31,99892,664)	64,585 (39,024110,336)	<0.001	<0.001
Complications*
Ventilator‐associated pneumonia, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.09	1.00
Facial injury, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.26	1.00
Iatrogenic pneumothorax, n (%)	10 (0.1)	90 (0.6)	14 (0.6)	<0.001	0.90

Table 3.Statistically Significant Predictors of NIV Use Alone

Patient Characteristics	Adjusted OR (95% CI)*	P Value
NOTE: Abbreviations: AIDS, acquired immune deficiency syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; ED, emergency department; NIV, noninvasive ventilation; OR, odds ratio; RA, rheumatoid arthritis. *Propensity score model included the following patient and hospital characteristics: age, sex, median household income, insurance status, weekend admission, season, comorbid conditions, US region, urban/rural and teaching status, annual ED volume, annual ED volume of COPD with respiratory failure, and calendar year. Nonsignificant predictors that are not reported in the table include sex, weekend admission, and the following comorbid conditions: peripheral vascular disorders, paralysis, uncomplicated diabetes, hypothyroidism, renal failure, peptic ulcer disease excluding bleeding, hypertension, lymphoma, AIDS, solid tumor without metastasis, and metastatic cancer.
Age, y
4049	1.00 (Reference)
5059	0.96 (0.84‐1.11)	0.61
6069	0.96 (0.84‐1.10)	0.56
7079	1.09 (0.94‐1.25)	0.25
80	1.30 (1.12‐1.52)	0.001
Quartile for median household income of patient ZIP code, $
138,999	1.00 (Reference)
39,00047,999	1.13 (1.05‐1.21)	0.001
48,00062,999	1.21 (1.12‐1.30)	<0.001
63,000	1.21 (1.11‐1.32)	<0.001
Insurance status
Medicare	1.00 (Reference)
Medicaid	0.79 (0.72‐0.88)	<0.001
Private	0.88 (0.81‐0.96)	0.004
Self‐pay	0.68 (0.56‐0.82)	<0.001
Other	0.88 (0.73‐1.07)	0.22
Season
Winter (January 1March 31)	1.00 (Reference)
Spring (April 1June 30)	1.06 (0.99‐1.14)	0.11
Summer (July 1September 30)	1.17 (1.08‐1.26)	<0.001
Fall (October 1December 31)	1.24 (1.15‐1.33)	<0.001
Comorbidity
CHF	0.90 (0.85‐0.95)	<0.001
Pulmonary circulatory disorders	1.40 (1.29‐1.52)	<0.001
Diabetes, complicated	1.25 (1.08‐1.44)	0.002
Liver disease	1.79 (1.40‐2.28)	<0.001
Coagulopathy	0.54 (0.46‐0.63)	<0.001
Obesity	1.52 (1.41‐1.65)	<0.001
Weight loss	0.50 (0.44‐0.57)	<0.001
Fluid and electrolyte disorders	0.84 (0.80‐0.89)	<0.001
Deficiency anemia	0.83 (0.78‐0.90)	<0.001
Alcohol abuse	0.66 (0.58‐0.76)	<0.001
Drug abuse	0.74 (0.62‐0.88)	0.001
Psychoses	1.22 (1.10‐1.37)	<0.001
Depression	1.45 (1.34‐1.57)	<0.001
Pneumonia	0.48 (0.45‐0.51)	<0.001
Valvular heart disease	0.87 (0.77‐0.97)	0.01
Neurological disorders	0.89 (0.80‐0.98)	0.02
RA/collagen vascular diseases	1.25 (1.02‐1.53)	0.04
Blood‐loss anemia	0.72 (0.53‐0.97)	0.03
Hospital characteristics
Annual ED visit volume, per 1000‐visit increase	0.997 (0.996‐0.998)	<0.001
Annual ED volume of COPD exacerbation with respiratory failure, per 10‐visit increase	1.03 (1.02‐1.03)	<0.001
Urban/rural and teaching status
Metropolitan nonteaching	1.00 (Reference)
Metropolitan teaching	0.91 (0.85‐0.97)	0.006
Nonmetropolitan	1.30 (1.20‐1.42)	<0.001
US region
Northeast	1.00 (Reference)
Midwest	0.44 (0.40‐0.48)	<0.001
South	0.54 (0.50‐0.58)	<0.001
West	0.51 (0.46‐0.56)	<0.001
Calendar year
2006	1.00 (Reference)
2007	1.30 (1.22‐1.39)	<0.001
2008	1.65 (1.54‐1.76)	<0.001

In terms of propensity score distributions (see Supporting Information, Figure E1, in the online version of this article), there was sufficient overlap of the NIV and IMV groups. After matching on propensity score for the NIV and IMV groups, the differences in baseline characteristics were all balanced (see Supporting Information, Table E1, in the online version of this article), as indicated by <10% standardized differences in all covariates between the 2 groups. Finally, in the propensity scorematched cohort (see Supporting Information, Table E2, in the online version of this article), NIV use remained associated with significantly lower inpatient mortality (risk ratio: 0.54; 95% CI: 0.50‐0.59, P<0.001), a shorter hospital LOS (mean difference, 3.2 days; 95% CI: 3.4 to 2.9 days, P<0.001), and lower hospital charges (mean difference, P<$35,012; 95% CI: $36,848 to $33,176, P<0.001), compared with IMV use. Use of NIV also was associated with a lower rate of iatrogenic pneumothorax than IMV use (0.05% vs 0.5%, P<0.001).

Using hospital preference for NIV vs IMV as an instrument, the instrumental analysis confirmed the benefits of NIV use, with a 5% reduction in inpatient mortality in the NIV‐preferring hospitals (risk difference, P<5%; 95% CI: P<1.8% to P<8.3%).

In the sensitivity analysis to assess the impact of an unmeasured confounder, the confounder would have had to have a very strong impact on outcome (risk ratio: 5) and a severe exposure‐confounder imbalance (odds ratio of exposure on confounder: 5) to reduce the observed association to 1.0. In other words, an individual unmeasured confounder is unlikely to explain the observed association.

DISCUSSION

In this nationally representative sample of 67,651 ED visits for AECOPD with acute respiratory failure, we found that NIV use was increasing from 2006 to 2008. However, the utilization of NIV remained low (16% in 2008) and varied widely by patient and hospital characteristic. As with all observational studies, causality cannot be inferred definitely; however, our study suggests that, NIV usecompared with IMV usewas associated with potentially important benefits: a reduction of inpatient mortality by 46%, shortened hospital LOS by 3 days, reduced hospital charges by approximately $35,000 per visit, and modestly reduced risk of iatrogenic pneumothorax.

A recent analysis using the US Nationwide Inpatient Sample has shown increasing use of NIV and concomitant decreasing mortality in AECOPD over time.[28] Our analysis confirmed these favorable trends in the United States using a much larger NEDS sample (28 million visits in the NEDS vs 8 million visits in the Nationwide Inpatient Sample per year). Despite these favorable trends, NIV was still underutilized for AECOPD with respiratory failure in the United States (16% in 2008) compared with major European countries (40%).[29] Although our study lacked clinical details to arrive at the optimal rate of NIV use, the low rate of NIV use is concerning and suggests room for improvement in NIV use in appropriate patients as outlined by the current COPD guidelines.[18, 30] Why is NIV not widely adopted, given its demonstrated efficacy? Previous surveys have identified several perceived reasons for low NIV use, including lack of physician knowledge, insufficient respiratory therapist training, inadequate equipment, and time required for setting up NIV.[29, 31, 32] Our study adds to the literature by showing the actual predictors of NIV use in the real world. Our data showed that the early adopters were hospitals with higher case volumes, and hospitals in the Northeast and in nonmetropolitan areas. A higher case volume has been linked with lower mortality in AECOPD (ie, practice makes perfect),[33] and frequent NIV use could explain the lower AECOPD mortality in highcase volume centers. Alternatively, smaller hospitals tend to have moonlighters working in EDs who may not be board certified in emergency medicine. Perhaps the logical next step is to conduct a qualitative study to understand the specifics of best practices and provider characteristics in these Northeastern, highercase volume centers. Another incentive to promote NIV use in clinical practice is the cost‐effectiveness associated with this intervention, as previous studies have shown that, compared with usual care, receiving NIV was associated with a reduction in costs, mainly through reduced use of the ICU.[34, 35]

Some patient factors associated with NIV use may be well justified. For example, older AECOPD patients may have an advance directive describing their treatment wishes (eg, do‐not‐intubate order),[36] and therefore NIV was preferred to IMV. Also, our data suggested AECOPD patients with a suspected pneumonia component were less likely to be placed on NIV, which is consistent with COPD guideline recommendations.[18, 30] As outlined in the current guidelines, the major contraindications to NIV include impending respiratory arrest, excessive respiratory secretions, massive gastrointestinal bleeding, recent facial trauma, or altered mental status.[18, 30] By contrast, some factors associated with NIV use may be targeted for intervention, such as lower rates of NIV use in the uninsured, patients who live in low‐income neighborhoods, and hospitals in US regions other than the Northeast.

Current guidelines recommend using NIV in AECOPD patients with early signs of respiratory failure, such as arterial pH of 7.257.35 or pCO₂ 45 mm Hg.[18, 30] When NIV is considered as the modality of ventilatory support, it should probably be used as early as possible,[37] because evidence suggests that delayed use of NIV may lead to severe respiratory acidosis and increased mortality.[38] Other than in ICUs, NIV can be used on general wards and in EDs that have adequate staff training and experience, because the success rates of NIV in these settings are similar to those reported in ICU studies.[8, 36, 39] In addition, NIV is more cost‐effective when performed outside the ICU.[35] In fact, studies have found a substantial portion of patients had NIV started in the ED (one‐fourth) and on the general ward (one‐fourth).[31, 40] Given the shortage of intensivists in the United States, hospitalists begin to play an important role in provision of critical care outside the ICU.[41] Once NIV is used, it is important to ensure that it is delivered effectively and monitored closely because NIV failure has been shown to be associated with high mortality.[28, 42]

This study has some potential limitations. First, we used administrative claims that lack clinical details such as data on arterial blood gases and severity scores, and thus potential residual confounding may exist. In our study, the IMV group may be sicker than the NIV group, which could partially explain the increased mortality with IMV. However, the propensity scores overlap to a great extent between the 2 study groups, suggesting that a strong confounding bias is less likely, given the observed covariates. Furthermore, the instrumental variable and sensitivity analyses taking into account unmeasured confounders still suggested the benefits of NIV. Second, the NEDS does not contain data on the location where NIV was initiated (eg, ED, ward, or ICU) or the timing of initiating NIV or IMV. As a result, for the combined‐use group, we could not further distinguish the group switching from NIV to IMV (ie, NIV failure)[42] or from IMV to NIV (ie, NIV as a weaning strategy).[43] Accordingly, we chose to focus on the comparativeness effectiveness of NIV vs IMV. Third, although the NEDS data have undergone quality‐control procedures,[44] some misclassification may exist in identifying patient population and interventions. Finally, the analysis may not reflect the most recent trend in NIV use, as the 2010 NEDS data have just been released. In addition, although the study is the largest to date on this topic, our findings may not be generalizable to EDs that were not part of the NEDS.

In summary, in this nationally representative ED and inpatient database, NIV use is increasing for AECOPD with acute respiratory failure; however, its adoption remains low and varies widely between US hospitals. Our observational study suggests that NIV appears to be more effective and safer than IMV in the real‐world setting. There is an opportunity to increase the use of NIV as recommended in guidelines and to promote the use NIV in replacement of IMV in patients with severe AECOPD. Given the increasing mortality burden of COPD, such a strategy may help reduce COPD mortality at the population level, thereby fulfilling the objectives of Healthy People 2020.

Disclosure

Partial results from this study were presented at the 2012 Society for Academic Emergency Medicine Annual Meeting, Chicago, Illinois, May 912, 2012. This project was supported by grant number R03HS020722 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. The authors have no conflicts of interest to disclose.

Chronic obstructive pulmonary disease (COPD) is now the third leading cause of death in the United States,[1] and its rising mortality trend is unique among the top 5 causes of death.[2] Acute exacerbations of COPD (AECOPD) are important events in the natural history of COPD, accounting for 1.5 million emergency department (ED) visits and 726,000 hospitalizations each year in the United States.[3, 4] Given the significant morbidity and mortality from AECOPD, Healthy People 2020 lists reducing deaths, hospitalizations, and ED visits as the key objectives for COPD.[5]

Over the past 2 decades, noninvasive ventilation (NIV) has emerged as a potentially useful treatment modality in AECOPD patients with acute respiratory failure. Noninvasive ventilation commonly refers to positive‐pressure ventilatory support delivered through a nasal or full‐face mask, such as bilevel positive airway pressure.[6] A number of randomized controlled trials[7, 8, 9] and meta‐analyses[10] have suggested a mortality‐reduction benefit with NIV use compared with standard medical care in AECOPD. To our knowledge, however, very few small randomized controlled trials compared NIV vs invasive mechanical ventilation (IMV) head‐to‐head,[11, 12, 13] and a recent evidence review found only 5 studies (405 subjects) on this topic.[14] Collectively, the limited evidence from randomized trials showed that NIV use resulted in similar intensive care unit (ICU) and in‐hospital mortality, fewer complications (eg, ventilator‐associated pneumonia and sepsis), and shorter hospital length of stays (LOS). Given that these trials have a smaller sample size and tend to exclude older patient (age >75 years) or patients with multiple comorbidities, there is a need to better understand the adoption and effectiveness of NIV treatment for AECOPD in a much larger patient population in the real‐world setting using observational data.

To address these knowledge gaps in the literature, we analyzed data from a large, nationally representative ED and inpatient sample. The objective of the present analysis was 2‐fold: (1) to characterize the use of NIV and IMV in AECOPD patients with acute respiratory failure at a national level; and (2) to compare the effectiveness of NIV vs IMV in the real‐world setting.

METHODS

RESULTS

Patient and ED Characteristics

The 20062008 NEDS sample contained 67,651 ED visits for AECOPD with acute respiratory failure from 1594 US EDs. After the weighting procedure, there were an estimated 101,000 visits annually for AECOPD with acute respiratory failure from approximately 4700 US EDs. In the weighted analysis, the mean patient age of these visits was 68 years, and 56% were made by women. Ninety‐six percent were admitted to the hospital. Of these, the mortality rate was 9% and the mean hospital LOS was 7 days. Figure 1 shows the secular trends in NIV, IMV, and the combined use over the 3‐year study period. Use of IMV decreased from 28% in 2006 to 19% in 2008 (P<0.001), whereas NIV use increased slightly from 14% in 2006 to 16% in 2008 (P=0.049); the combined use of both ventilation modalities remained stable (4%). Inpatient mortality decreased from 10% in 2006 to 7% in 2008 (P<0.001).

Figure 1

Figure 2 shows that the frequency of NIV use (including combined use of NIV and IMV) varied widely between hospitals, ranging from 0% to 100% with a median of 11%. In the unweighted cohort of AECOPD with acute respiratory failure, 43% received some forms of ventilatory support. Table 1 shows the patient and hospital characteristics of the patients receiving ventilatory support: 36% received NIV, 56% received IMV, and 8% received combined use. In general, patients receiving combined use of NIV and IMV tended to have more comorbidities (eg, congestive heart failure and pneumonia) compared with the NIV‐alone or IMV‐alone groups. With respect to hospital characteristics, NIV was used more often in hospitals with higher volumes of COPD exacerbation and respiratory failure, in nonmetropolitan hospitals, and in hospitals in the Northeast.

Figure 2

Table 1.Patient and Hospital Characteristics According to Ventilatory Mode

	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value, A vs B	P Value, B vs C
NOTE: Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; ED, emergency department; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation.
Patient characteristics
Age, y,				<0.001	0.64
4049	5	5	5
5059	17	18	19
6069	31	33	33
7079	30	29	29
80	17	15	13
Female sex, %	57	53	54	<0.001	0.87
Quartile for median household income of patient ZIP code, $, %				<0.001	<0.001
138,999	30	34	29
39,00047,999	28	28	28
48,00062,999	24	22	24
63,000	18	15	19
Weekend admission, %	27	28	28	0.07	0.80
Insurance status, %				<0.001	0.91
Medicare	74	70	70
Medicaid	9	12	12
Private	12	13	13
Self‐pay	2	3	2
Other	2	2	2
Season, %				<0.001	0.16
Winter (January 1March 31)	29	32	31
Spring (April 1June 30)	24	25	26
Summer (July 1September 30)	22	20	19
Fall (October 1December 31)	25	22	24
No. of comorbidities, median (IQR)	4 (35)	4 (35)	4 (36)	<0.001	<0.001
Selected comorbidities, %
Hypertension	56	55	55	0.01	0.65
CHF	38	40	44	0.001	<0.001
Fluid and electrolyte disorders	37	44	49	<0.001	<0.001
Diabetes, uncomplicated	27	26	29	0.04	0.002
Pneumonia	19	34	39	<0.001	<0.001
Deficiency anemia	16	19	19	<0.001	0.39
Obesity	18	12	17	<0.001	<0.001
Depression	15	11	11	<0.001	0.54
Pulmonary circulatory diseases	15	11	14	<0.001	<0.001
Hospital characteristics
Annual ED visit volume, median (IQR)	42,704 (29,50562,470)	44,119 (29,89564,097)	46,695 (31,29866,235)	0.02	0.0003
Annual ED volume of COPD exacerbation with respiratory failure, median (IQR)	45 (2672)	42 (2368)	38 (2364)	<0.001	<0.001
Urban/rural and teaching status, %				<0.001	<0.001
Metropolitan nonteaching	53	52	47
Metropolitan teaching	31	35	39
Nonmetropolitan	16	13	13
US region, %				<0.001	<0.001
Northeast	28	16	36
Midwest	17	22	15
South	41	45	32
West	14	17	17

The unadjusted differences in outcomes are shown in Table 2. The combined‐use group had the highest inpatient mortality, longest LOS, and highest charges, followed by the IMV and NIV groups. In general, complications were few across all 3 groups, but the rate of iatrogenic pneumothorax was notably lower in the NIV group. Table 3 details the statistically significant predictors of NIV use in the propensity score model. Similar to the unadjusted analysis, older age, high‐income neighborhoods, Medicare insurance, and some comorbidities were positively associated with NIV use (eg, pulmonary circulatory disorders and liver disease), whereas a few comorbidities were negatively associated with NIV use (eg, pneumonia, and alcohol and drug abuse). With respect to hospital characteristics, higher case volumes of COPD exacerbation/respiratory failure, Northeastern and nonmetropolitan hospitals, and more recent years were associated with NIV use.

Table 2.Clinical Outcomes Among COPD Patients With Acute Respiratory Failure According to Ventilatory Mode

Outcome	NIV Alone (A) (n=10,032)	IMV Alone (B) (n=15,427)	Combined Use (C) (n=2311)	P Value. A vs B	P Value, B vs C
NOTE: Abbreviations: COPD, chronic obstructive pulmonary disease; IMV, invasive mechanical ventilation; IQR, interquartile range; NIV, noninvasive ventilation. *For privacy protection, we are not able to report cells in the tables 10 individual records.
Inpatient mortality, n (%)	825 (8)	2,454 (16)	407 (18)	<0.001	0.04
Hospital length of stay, median (IQR), d	5 (48)	8 (513)	10 (716)	<0.001	<0.001
Hospital charge per visit, median (IQR), $	26,002 (15,74744,638)	53,432 (31,99892,664)	64,585 (39,024110,336)	<0.001	<0.001
Complications*
Ventilator‐associated pneumonia, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.09	1.00
Facial injury, n (%)	10 (0.1)	10 (0.1)	10 (0.5)	0.26	1.00
Iatrogenic pneumothorax, n (%)	10 (0.1)	90 (0.6)	14 (0.6)	<0.001	0.90

Table 3.Statistically Significant Predictors of NIV Use Alone

Patient Characteristics	Adjusted OR (95% CI)*	P Value
NOTE: Abbreviations: AIDS, acquired immune deficiency syndrome; CHF, congestive heart failure; CI, confidence interval; COPD, chronic obstructive pulmonary disease; ED, emergency department; NIV, noninvasive ventilation; OR, odds ratio; RA, rheumatoid arthritis. *Propensity score model included the following patient and hospital characteristics: age, sex, median household income, insurance status, weekend admission, season, comorbid conditions, US region, urban/rural and teaching status, annual ED volume, annual ED volume of COPD with respiratory failure, and calendar year. Nonsignificant predictors that are not reported in the table include sex, weekend admission, and the following comorbid conditions: peripheral vascular disorders, paralysis, uncomplicated diabetes, hypothyroidism, renal failure, peptic ulcer disease excluding bleeding, hypertension, lymphoma, AIDS, solid tumor without metastasis, and metastatic cancer.
Age, y
4049	1.00 (Reference)
5059	0.96 (0.84‐1.11)	0.61
6069	0.96 (0.84‐1.10)	0.56
7079	1.09 (0.94‐1.25)	0.25
80	1.30 (1.12‐1.52)	0.001
Quartile for median household income of patient ZIP code, $
138,999	1.00 (Reference)
39,00047,999	1.13 (1.05‐1.21)	0.001
48,00062,999	1.21 (1.12‐1.30)	<0.001
63,000	1.21 (1.11‐1.32)	<0.001
Insurance status
Medicare	1.00 (Reference)
Medicaid	0.79 (0.72‐0.88)	<0.001
Private	0.88 (0.81‐0.96)	0.004
Self‐pay	0.68 (0.56‐0.82)	<0.001
Other	0.88 (0.73‐1.07)	0.22
Season
Winter (January 1March 31)	1.00 (Reference)
Spring (April 1June 30)	1.06 (0.99‐1.14)	0.11
Summer (July 1September 30)	1.17 (1.08‐1.26)	<0.001
Fall (October 1December 31)	1.24 (1.15‐1.33)	<0.001
Comorbidity
CHF	0.90 (0.85‐0.95)	<0.001
Pulmonary circulatory disorders	1.40 (1.29‐1.52)	<0.001
Diabetes, complicated	1.25 (1.08‐1.44)	0.002
Liver disease	1.79 (1.40‐2.28)	<0.001
Coagulopathy	0.54 (0.46‐0.63)	<0.001
Obesity	1.52 (1.41‐1.65)	<0.001
Weight loss	0.50 (0.44‐0.57)	<0.001
Fluid and electrolyte disorders	0.84 (0.80‐0.89)	<0.001
Deficiency anemia	0.83 (0.78‐0.90)	<0.001
Alcohol abuse	0.66 (0.58‐0.76)	<0.001
Drug abuse	0.74 (0.62‐0.88)	0.001
Psychoses	1.22 (1.10‐1.37)	<0.001
Depression	1.45 (1.34‐1.57)	<0.001
Pneumonia	0.48 (0.45‐0.51)	<0.001
Valvular heart disease	0.87 (0.77‐0.97)	0.01
Neurological disorders	0.89 (0.80‐0.98)	0.02
RA/collagen vascular diseases	1.25 (1.02‐1.53)	0.04
Blood‐loss anemia	0.72 (0.53‐0.97)	0.03
Hospital characteristics
Annual ED visit volume, per 1000‐visit increase	0.997 (0.996‐0.998)	<0.001
Annual ED volume of COPD exacerbation with respiratory failure, per 10‐visit increase	1.03 (1.02‐1.03)	<0.001
Urban/rural and teaching status
Metropolitan nonteaching	1.00 (Reference)
Metropolitan teaching	0.91 (0.85‐0.97)	0.006
Nonmetropolitan	1.30 (1.20‐1.42)	<0.001
US region
Northeast	1.00 (Reference)
Midwest	0.44 (0.40‐0.48)	<0.001
South	0.54 (0.50‐0.58)	<0.001
West	0.51 (0.46‐0.56)	<0.001
Calendar year
2006	1.00 (Reference)
2007	1.30 (1.22‐1.39)	<0.001
2008	1.65 (1.54‐1.76)	<0.001

In terms of propensity score distributions (see Supporting Information, Figure E1, in the online version of this article), there was sufficient overlap of the NIV and IMV groups. After matching on propensity score for the NIV and IMV groups, the differences in baseline characteristics were all balanced (see Supporting Information, Table E1, in the online version of this article), as indicated by <10% standardized differences in all covariates between the 2 groups. Finally, in the propensity scorematched cohort (see Supporting Information, Table E2, in the online version of this article), NIV use remained associated with significantly lower inpatient mortality (risk ratio: 0.54; 95% CI: 0.50‐0.59, P<0.001), a shorter hospital LOS (mean difference, 3.2 days; 95% CI: 3.4 to 2.9 days, P<0.001), and lower hospital charges (mean difference, P<$35,012; 95% CI: $36,848 to $33,176, P<0.001), compared with IMV use. Use of NIV also was associated with a lower rate of iatrogenic pneumothorax than IMV use (0.05% vs 0.5%, P<0.001).

Using hospital preference for NIV vs IMV as an instrument, the instrumental analysis confirmed the benefits of NIV use, with a 5% reduction in inpatient mortality in the NIV‐preferring hospitals (risk difference, P<5%; 95% CI: P<1.8% to P<8.3%).

In the sensitivity analysis to assess the impact of an unmeasured confounder, the confounder would have had to have a very strong impact on outcome (risk ratio: 5) and a severe exposure‐confounder imbalance (odds ratio of exposure on confounder: 5) to reduce the observed association to 1.0. In other words, an individual unmeasured confounder is unlikely to explain the observed association.

DISCUSSION

In this nationally representative sample of 67,651 ED visits for AECOPD with acute respiratory failure, we found that NIV use was increasing from 2006 to 2008. However, the utilization of NIV remained low (16% in 2008) and varied widely by patient and hospital characteristic. As with all observational studies, causality cannot be inferred definitely; however, our study suggests that, NIV usecompared with IMV usewas associated with potentially important benefits: a reduction of inpatient mortality by 46%, shortened hospital LOS by 3 days, reduced hospital charges by approximately $35,000 per visit, and modestly reduced risk of iatrogenic pneumothorax.

A recent analysis using the US Nationwide Inpatient Sample has shown increasing use of NIV and concomitant decreasing mortality in AECOPD over time.[28] Our analysis confirmed these favorable trends in the United States using a much larger NEDS sample (28 million visits in the NEDS vs 8 million visits in the Nationwide Inpatient Sample per year). Despite these favorable trends, NIV was still underutilized for AECOPD with respiratory failure in the United States (16% in 2008) compared with major European countries (40%).[29] Although our study lacked clinical details to arrive at the optimal rate of NIV use, the low rate of NIV use is concerning and suggests room for improvement in NIV use in appropriate patients as outlined by the current COPD guidelines.[18, 30] Why is NIV not widely adopted, given its demonstrated efficacy? Previous surveys have identified several perceived reasons for low NIV use, including lack of physician knowledge, insufficient respiratory therapist training, inadequate equipment, and time required for setting up NIV.[29, 31, 32] Our study adds to the literature by showing the actual predictors of NIV use in the real world. Our data showed that the early adopters were hospitals with higher case volumes, and hospitals in the Northeast and in nonmetropolitan areas. A higher case volume has been linked with lower mortality in AECOPD (ie, practice makes perfect),[33] and frequent NIV use could explain the lower AECOPD mortality in highcase volume centers. Alternatively, smaller hospitals tend to have moonlighters working in EDs who may not be board certified in emergency medicine. Perhaps the logical next step is to conduct a qualitative study to understand the specifics of best practices and provider characteristics in these Northeastern, highercase volume centers. Another incentive to promote NIV use in clinical practice is the cost‐effectiveness associated with this intervention, as previous studies have shown that, compared with usual care, receiving NIV was associated with a reduction in costs, mainly through reduced use of the ICU.[34, 35]

Some patient factors associated with NIV use may be well justified. For example, older AECOPD patients may have an advance directive describing their treatment wishes (eg, do‐not‐intubate order),[36] and therefore NIV was preferred to IMV. Also, our data suggested AECOPD patients with a suspected pneumonia component were less likely to be placed on NIV, which is consistent with COPD guideline recommendations.[18, 30] As outlined in the current guidelines, the major contraindications to NIV include impending respiratory arrest, excessive respiratory secretions, massive gastrointestinal bleeding, recent facial trauma, or altered mental status.[18, 30] By contrast, some factors associated with NIV use may be targeted for intervention, such as lower rates of NIV use in the uninsured, patients who live in low‐income neighborhoods, and hospitals in US regions other than the Northeast.

Current guidelines recommend using NIV in AECOPD patients with early signs of respiratory failure, such as arterial pH of 7.257.35 or pCO₂ 45 mm Hg.[18, 30] When NIV is considered as the modality of ventilatory support, it should probably be used as early as possible,[37] because evidence suggests that delayed use of NIV may lead to severe respiratory acidosis and increased mortality.[38] Other than in ICUs, NIV can be used on general wards and in EDs that have adequate staff training and experience, because the success rates of NIV in these settings are similar to those reported in ICU studies.[8, 36, 39] In addition, NIV is more cost‐effective when performed outside the ICU.[35] In fact, studies have found a substantial portion of patients had NIV started in the ED (one‐fourth) and on the general ward (one‐fourth).[31, 40] Given the shortage of intensivists in the United States, hospitalists begin to play an important role in provision of critical care outside the ICU.[41] Once NIV is used, it is important to ensure that it is delivered effectively and monitored closely because NIV failure has been shown to be associated with high mortality.[28, 42]

This study has some potential limitations. First, we used administrative claims that lack clinical details such as data on arterial blood gases and severity scores, and thus potential residual confounding may exist. In our study, the IMV group may be sicker than the NIV group, which could partially explain the increased mortality with IMV. However, the propensity scores overlap to a great extent between the 2 study groups, suggesting that a strong confounding bias is less likely, given the observed covariates. Furthermore, the instrumental variable and sensitivity analyses taking into account unmeasured confounders still suggested the benefits of NIV. Second, the NEDS does not contain data on the location where NIV was initiated (eg, ED, ward, or ICU) or the timing of initiating NIV or IMV. As a result, for the combined‐use group, we could not further distinguish the group switching from NIV to IMV (ie, NIV failure)[42] or from IMV to NIV (ie, NIV as a weaning strategy).[43] Accordingly, we chose to focus on the comparativeness effectiveness of NIV vs IMV. Third, although the NEDS data have undergone quality‐control procedures,[44] some misclassification may exist in identifying patient population and interventions. Finally, the analysis may not reflect the most recent trend in NIV use, as the 2010 NEDS data have just been released. In addition, although the study is the largest to date on this topic, our findings may not be generalizable to EDs that were not part of the NEDS.

In summary, in this nationally representative ED and inpatient database, NIV use is increasing for AECOPD with acute respiratory failure; however, its adoption remains low and varies widely between US hospitals. Our observational study suggests that NIV appears to be more effective and safer than IMV in the real‐world setting. There is an opportunity to increase the use of NIV as recommended in guidelines and to promote the use NIV in replacement of IMV in patients with severe AECOPD. Given the increasing mortality burden of COPD, such a strategy may help reduce COPD mortality at the population level, thereby fulfilling the objectives of Healthy People 2020.

Disclosure

Partial results from this study were presented at the 2012 Society for Academic Emergency Medicine Annual Meeting, Chicago, Illinois, May 912, 2012. This project was supported by grant number R03HS020722 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. The authors have no conflicts of interest to disclose.

The practice of bloodletting, performed using sticks, thorns, bones, or anything sharp, probably began in Egypt about 3,000 years ago.[1] The practice continued in Greece, where Hippocrates recommended bloodletting to balance the body's four humorsblood, phlegm, yellow bile, and black bileand continued during Roman times under the influence of Galen. In the United States, perhaps the most infamous use of bloodletting was when doctors reportedly bled as much as 5 U of blood from George Washington before he died from what was probably either acute epiglottitis or streptococcal pharyngitis.[2, 3] Although many infectious organisms, especially malaria parasites, require iron to proliferateand therefore may be less virulent in iron‐deficient people[4]acute near‐exsanguination undoubtedly did more harm than good in the elderly ex‐president.

But the practice of bloodletting continued and even flourished. In 1833 alone, France reportedly imported more than 40 million leeches to assist in bloodletting,[5] which oftentimes was thought to be sufficiently aggressive only when the patient actually fainted. Enthusiasm for bloodletting declined in the second half of the 19th century, influenced in part by a nonrandomized study that compared mortality rates among patients who were bled early in their illness with those who were bled later.[6] Nevertheless, Sir William Osler still recommended small amounts of bloodletting for pneumonia in his last edition of his famous textbook, The Principles and Practice of Medicine, published in 1920.[7] By 1927, however, the first edition of the Cecil's A Textbook of Medicine thankfully no longer recommended venesection except to treat conditions such as pulmonary edema.[8]

Why would I start this essay with a history of bloodletting? Surely, one might argue, nothing could be less relevant to a modern discussion of the quality of in‐hospital medical care. The substantial literature on quality improvement emphasizes the practical implementation of strategies to increase the appropriate adherence to processes that are known to improve outcomes. A number of common quality measures quickly come to mind: the use of aspirin, ‐blockers, angiotensin‐converting enzyme inhibitors, and statins in post‐myocardial infarction patients without contraindications,[9, 10] the rapid initiation of appropriate antibiotics to patients with community‐acquired pneumonia,[11] and early endoscopy for patients with acute upper gastrointestinal hemorrhage.[12] I could go on and on, listing in‐hospital interventions supported by class 1 evidence from more than one definitive randomized trial. In essentially all of these situations, the creation of quality metrics, often accompanied by measurement and feedback, have improved adherence and undoubtedly saved lives. But although adherence has improved, the explosion in evidence‐based medicine means that even the best hospitals may be in perpetual catch‐up mode as they try to ensure adherence with the next wave of improvement interventions.

Unfortunately, every now and then a lot of attention is paid to meeting a quality metric that turns out to be misguided. Perhaps the best recent in‐hospital example was the metric of prophylactic ‐blocker use before major noncardiac surgery. Although this recommendation initially appeared to be based on reasonable data,[13] the large Perioperative Ischemic Evaluation Study (POISE) trial showed that reductions in rates of myocardial infarction were more than offset by an increased risk of stroke and other complications; therefore, average‐risk patients actually did worse, not better, with the ‐blocker regimen used in the trial. Although some have questioned whether these results were a function of the precise ‐blocker regimen that was used, the results of POISE are actually remarkably consistent with prior data on the risk of myocardial infarction and stroke.[14, 15] What was really different was the relative importance of these and other end points in patients whose risk of cardiac death was lower than those of higher risk patients in prior studies. But more recently, an even more disturbing reality has emerged: a number of key reports on which the guidelines were based came from an investigator whose publications included data that could not be confirmed when his studies were reviewed by his home institution.[16] Regardless of the precise reasons, we no longer routinely recommend an intervention that at one time was a key quality indicator.

The ‐blocker fiasco brings me back to bloodletting. In the early 19th century, a hypothetical visionary physician interested in quality improvement would likely have looked for ways to improve the efficiency and reduce the cost of bloodletting. Perhaps the leeches could be bigger, hungrier, or applied in a more effective fashion? Or perhaps vacuum tubes would have been invented sooner?

Of course, I am overemphasizing to prove a point. I truly believe that more and more of what we recommend is based on solid evidence to document, at least for now, that we are doing the right thing. If we do it more often and in more people, net benefit will be realized.

What does all this mean for the future of hospital medicine and its emerging research endeavors? For me at least, the message is pretty clear. First, we must be careful not to over extrapolate from limited studies in high risk patients, or we will jump to more conclusions like we did with ‐blockers. Second, most advances in medicine require new and better data.

How can new clinical data be generated most quickly and efficiently? One‐off studies at individual institutions are logistically and financially challenging, whereas an enduring research infrastructure is a treasure that can study a series of questions as they arise. The Thrombolysis in Myocardial Infarction Study Group has published scores of papers looking at a series of interventions in patients with acute myocardial infarction and the acute coronary syndrome.[17, 18] The Acute Respiratory Distress Syndrome Network has demonstrated the value of lower tidal volumes and less aggressive fluid strategies in patients with respiratory failure.[19, 20]

The success of these large, multicenter research networks should become the paradigm for the study of common hospital problems, ranging from the conditions that result in admission on the medical service to the problems that have engendered surgical comanagement services. New data can be gleaned by studying the medical care system, by studying routinely gathered administrative and clinical data, or even better yet, by gathering prospective data on patients and their diseases. For hospitalized patients, a variety of unanswered questions remain regarding the epidemiology of common diseases, the value of diagnostic tests, the impact of various therapies and treatment protocols, and the incremental value of new technologies, ranging from self‐monitoring to handheld ultrasound. High‐quality research may address the genetic epidemiology of why one person is admitted with pneumococcal pneumonia, whereas family members seem perfectly healthy; which patients with a particular diagnosis might be managed for different lengths of time in different settings; what physical findings or diagnostic tests best stratify prognosis; what new technologies are truly worth their cost; and especially, what therapies really work.

I do not dispute that hospital medicine researchers should try to improve the current use of interventions that are deemed to be valuable right now. But if that is all the field does, it will be a huge disappointment. Hospitalists should not be relegated to being adherence police who spend their collective research energy finding ways to force themselves to follow recommendations based on data gathered by others.

Hospital medicine researchers are uniquely positioned to discover new information that will change what should be done and help create the quality metrics for the future. Unless both of these two goalsimproving the implementation of today's knowledge and generating new and better knowledgeare part of the research agenda, we run the risk that some of the best minds in internal medicine may, when all is said and done, have spent an inordinate number of IQ hours on what, a century from now, will be reminiscent of improving the quality of bloodletting.

The practice of bloodletting, performed using sticks, thorns, bones, or anything sharp, probably began in Egypt about 3,000 years ago.[1] The practice continued in Greece, where Hippocrates recommended bloodletting to balance the body's four humorsblood, phlegm, yellow bile, and black bileand continued during Roman times under the influence of Galen. In the United States, perhaps the most infamous use of bloodletting was when doctors reportedly bled as much as 5 U of blood from George Washington before he died from what was probably either acute epiglottitis or streptococcal pharyngitis.[2, 3] Although many infectious organisms, especially malaria parasites, require iron to proliferateand therefore may be less virulent in iron‐deficient people[4]acute near‐exsanguination undoubtedly did more harm than good in the elderly ex‐president.

But the practice of bloodletting continued and even flourished. In 1833 alone, France reportedly imported more than 40 million leeches to assist in bloodletting,[5] which oftentimes was thought to be sufficiently aggressive only when the patient actually fainted. Enthusiasm for bloodletting declined in the second half of the 19th century, influenced in part by a nonrandomized study that compared mortality rates among patients who were bled early in their illness with those who were bled later.[6] Nevertheless, Sir William Osler still recommended small amounts of bloodletting for pneumonia in his last edition of his famous textbook, The Principles and Practice of Medicine, published in 1920.[7] By 1927, however, the first edition of the Cecil's A Textbook of Medicine thankfully no longer recommended venesection except to treat conditions such as pulmonary edema.[8]

Why would I start this essay with a history of bloodletting? Surely, one might argue, nothing could be less relevant to a modern discussion of the quality of in‐hospital medical care. The substantial literature on quality improvement emphasizes the practical implementation of strategies to increase the appropriate adherence to processes that are known to improve outcomes. A number of common quality measures quickly come to mind: the use of aspirin, ‐blockers, angiotensin‐converting enzyme inhibitors, and statins in post‐myocardial infarction patients without contraindications,[9, 10] the rapid initiation of appropriate antibiotics to patients with community‐acquired pneumonia,[11] and early endoscopy for patients with acute upper gastrointestinal hemorrhage.[12] I could go on and on, listing in‐hospital interventions supported by class 1 evidence from more than one definitive randomized trial. In essentially all of these situations, the creation of quality metrics, often accompanied by measurement and feedback, have improved adherence and undoubtedly saved lives. But although adherence has improved, the explosion in evidence‐based medicine means that even the best hospitals may be in perpetual catch‐up mode as they try to ensure adherence with the next wave of improvement interventions.

Unfortunately, every now and then a lot of attention is paid to meeting a quality metric that turns out to be misguided. Perhaps the best recent in‐hospital example was the metric of prophylactic ‐blocker use before major noncardiac surgery. Although this recommendation initially appeared to be based on reasonable data,[13] the large Perioperative Ischemic Evaluation Study (POISE) trial showed that reductions in rates of myocardial infarction were more than offset by an increased risk of stroke and other complications; therefore, average‐risk patients actually did worse, not better, with the ‐blocker regimen used in the trial. Although some have questioned whether these results were a function of the precise ‐blocker regimen that was used, the results of POISE are actually remarkably consistent with prior data on the risk of myocardial infarction and stroke.[14, 15] What was really different was the relative importance of these and other end points in patients whose risk of cardiac death was lower than those of higher risk patients in prior studies. But more recently, an even more disturbing reality has emerged: a number of key reports on which the guidelines were based came from an investigator whose publications included data that could not be confirmed when his studies were reviewed by his home institution.[16] Regardless of the precise reasons, we no longer routinely recommend an intervention that at one time was a key quality indicator.

The ‐blocker fiasco brings me back to bloodletting. In the early 19th century, a hypothetical visionary physician interested in quality improvement would likely have looked for ways to improve the efficiency and reduce the cost of bloodletting. Perhaps the leeches could be bigger, hungrier, or applied in a more effective fashion? Or perhaps vacuum tubes would have been invented sooner?

Of course, I am overemphasizing to prove a point. I truly believe that more and more of what we recommend is based on solid evidence to document, at least for now, that we are doing the right thing. If we do it more often and in more people, net benefit will be realized.

What does all this mean for the future of hospital medicine and its emerging research endeavors? For me at least, the message is pretty clear. First, we must be careful not to over extrapolate from limited studies in high risk patients, or we will jump to more conclusions like we did with ‐blockers. Second, most advances in medicine require new and better data.

How can new clinical data be generated most quickly and efficiently? One‐off studies at individual institutions are logistically and financially challenging, whereas an enduring research infrastructure is a treasure that can study a series of questions as they arise. The Thrombolysis in Myocardial Infarction Study Group has published scores of papers looking at a series of interventions in patients with acute myocardial infarction and the acute coronary syndrome.[17, 18] The Acute Respiratory Distress Syndrome Network has demonstrated the value of lower tidal volumes and less aggressive fluid strategies in patients with respiratory failure.[19, 20]

The success of these large, multicenter research networks should become the paradigm for the study of common hospital problems, ranging from the conditions that result in admission on the medical service to the problems that have engendered surgical comanagement services. New data can be gleaned by studying the medical care system, by studying routinely gathered administrative and clinical data, or even better yet, by gathering prospective data on patients and their diseases. For hospitalized patients, a variety of unanswered questions remain regarding the epidemiology of common diseases, the value of diagnostic tests, the impact of various therapies and treatment protocols, and the incremental value of new technologies, ranging from self‐monitoring to handheld ultrasound. High‐quality research may address the genetic epidemiology of why one person is admitted with pneumococcal pneumonia, whereas family members seem perfectly healthy; which patients with a particular diagnosis might be managed for different lengths of time in different settings; what physical findings or diagnostic tests best stratify prognosis; what new technologies are truly worth their cost; and especially, what therapies really work.

I do not dispute that hospital medicine researchers should try to improve the current use of interventions that are deemed to be valuable right now. But if that is all the field does, it will be a huge disappointment. Hospitalists should not be relegated to being adherence police who spend their collective research energy finding ways to force themselves to follow recommendations based on data gathered by others.

Hospital medicine researchers are uniquely positioned to discover new information that will change what should be done and help create the quality metrics for the future. Unless both of these two goalsimproving the implementation of today's knowledge and generating new and better knowledgeare part of the research agenda, we run the risk that some of the best minds in internal medicine may, when all is said and done, have spent an inordinate number of IQ hours on what, a century from now, will be reminiscent of improving the quality of bloodletting.

The practice of bloodletting, performed using sticks, thorns, bones, or anything sharp, probably began in Egypt about 3,000 years ago.[1] The practice continued in Greece, where Hippocrates recommended bloodletting to balance the body's four humorsblood, phlegm, yellow bile, and black bileand continued during Roman times under the influence of Galen. In the United States, perhaps the most infamous use of bloodletting was when doctors reportedly bled as much as 5 U of blood from George Washington before he died from what was probably either acute epiglottitis or streptococcal pharyngitis.[2, 3] Although many infectious organisms, especially malaria parasites, require iron to proliferateand therefore may be less virulent in iron‐deficient people[4]acute near‐exsanguination undoubtedly did more harm than good in the elderly ex‐president.

But the practice of bloodletting continued and even flourished. In 1833 alone, France reportedly imported more than 40 million leeches to assist in bloodletting,[5] which oftentimes was thought to be sufficiently aggressive only when the patient actually fainted. Enthusiasm for bloodletting declined in the second half of the 19th century, influenced in part by a nonrandomized study that compared mortality rates among patients who were bled early in their illness with those who were bled later.[6] Nevertheless, Sir William Osler still recommended small amounts of bloodletting for pneumonia in his last edition of his famous textbook, The Principles and Practice of Medicine, published in 1920.[7] By 1927, however, the first edition of the Cecil's A Textbook of Medicine thankfully no longer recommended venesection except to treat conditions such as pulmonary edema.[8]

Why would I start this essay with a history of bloodletting? Surely, one might argue, nothing could be less relevant to a modern discussion of the quality of in‐hospital medical care. The substantial literature on quality improvement emphasizes the practical implementation of strategies to increase the appropriate adherence to processes that are known to improve outcomes. A number of common quality measures quickly come to mind: the use of aspirin, ‐blockers, angiotensin‐converting enzyme inhibitors, and statins in post‐myocardial infarction patients without contraindications,[9, 10] the rapid initiation of appropriate antibiotics to patients with community‐acquired pneumonia,[11] and early endoscopy for patients with acute upper gastrointestinal hemorrhage.[12] I could go on and on, listing in‐hospital interventions supported by class 1 evidence from more than one definitive randomized trial. In essentially all of these situations, the creation of quality metrics, often accompanied by measurement and feedback, have improved adherence and undoubtedly saved lives. But although adherence has improved, the explosion in evidence‐based medicine means that even the best hospitals may be in perpetual catch‐up mode as they try to ensure adherence with the next wave of improvement interventions.

Unfortunately, every now and then a lot of attention is paid to meeting a quality metric that turns out to be misguided. Perhaps the best recent in‐hospital example was the metric of prophylactic ‐blocker use before major noncardiac surgery. Although this recommendation initially appeared to be based on reasonable data,[13] the large Perioperative Ischemic Evaluation Study (POISE) trial showed that reductions in rates of myocardial infarction were more than offset by an increased risk of stroke and other complications; therefore, average‐risk patients actually did worse, not better, with the ‐blocker regimen used in the trial. Although some have questioned whether these results were a function of the precise ‐blocker regimen that was used, the results of POISE are actually remarkably consistent with prior data on the risk of myocardial infarction and stroke.[14, 15] What was really different was the relative importance of these and other end points in patients whose risk of cardiac death was lower than those of higher risk patients in prior studies. But more recently, an even more disturbing reality has emerged: a number of key reports on which the guidelines were based came from an investigator whose publications included data that could not be confirmed when his studies were reviewed by his home institution.[16] Regardless of the precise reasons, we no longer routinely recommend an intervention that at one time was a key quality indicator.

The ‐blocker fiasco brings me back to bloodletting. In the early 19th century, a hypothetical visionary physician interested in quality improvement would likely have looked for ways to improve the efficiency and reduce the cost of bloodletting. Perhaps the leeches could be bigger, hungrier, or applied in a more effective fashion? Or perhaps vacuum tubes would have been invented sooner?

Of course, I am overemphasizing to prove a point. I truly believe that more and more of what we recommend is based on solid evidence to document, at least for now, that we are doing the right thing. If we do it more often and in more people, net benefit will be realized.

What does all this mean for the future of hospital medicine and its emerging research endeavors? For me at least, the message is pretty clear. First, we must be careful not to over extrapolate from limited studies in high risk patients, or we will jump to more conclusions like we did with ‐blockers. Second, most advances in medicine require new and better data.

How can new clinical data be generated most quickly and efficiently? One‐off studies at individual institutions are logistically and financially challenging, whereas an enduring research infrastructure is a treasure that can study a series of questions as they arise. The Thrombolysis in Myocardial Infarction Study Group has published scores of papers looking at a series of interventions in patients with acute myocardial infarction and the acute coronary syndrome.[17, 18] The Acute Respiratory Distress Syndrome Network has demonstrated the value of lower tidal volumes and less aggressive fluid strategies in patients with respiratory failure.[19, 20]

The success of these large, multicenter research networks should become the paradigm for the study of common hospital problems, ranging from the conditions that result in admission on the medical service to the problems that have engendered surgical comanagement services. New data can be gleaned by studying the medical care system, by studying routinely gathered administrative and clinical data, or even better yet, by gathering prospective data on patients and their diseases. For hospitalized patients, a variety of unanswered questions remain regarding the epidemiology of common diseases, the value of diagnostic tests, the impact of various therapies and treatment protocols, and the incremental value of new technologies, ranging from self‐monitoring to handheld ultrasound. High‐quality research may address the genetic epidemiology of why one person is admitted with pneumococcal pneumonia, whereas family members seem perfectly healthy; which patients with a particular diagnosis might be managed for different lengths of time in different settings; what physical findings or diagnostic tests best stratify prognosis; what new technologies are truly worth their cost; and especially, what therapies really work.

I do not dispute that hospital medicine researchers should try to improve the current use of interventions that are deemed to be valuable right now. But if that is all the field does, it will be a huge disappointment. Hospitalists should not be relegated to being adherence police who spend their collective research energy finding ways to force themselves to follow recommendations based on data gathered by others.

Hospital medicine researchers are uniquely positioned to discover new information that will change what should be done and help create the quality metrics for the future. Unless both of these two goalsimproving the implementation of today's knowledge and generating new and better knowledgeare part of the research agenda, we run the risk that some of the best minds in internal medicine may, when all is said and done, have spent an inordinate number of IQ hours on what, a century from now, will be reminiscent of improving the quality of bloodletting.

The use of inferior vena cava (IVC) filters has increased substantially in recent years. These medical devices, which are used to prevent pulmonary embolism in patients considered to be at high risk of venous thromboembolism, were placed in 167,000 patients in 2007.¹ In 2012, it is estimated that 259,000 patients will undergo placement of an IVC filter, an increase of 55%.[1] Increasing use of IVC filters is attributable to the development of retrievable versions of the devices, which have expanded indications for use such as in bariatric surgery.

Unfortunately, the increase in the use of IVC filters has been accompanied by an increase in reports of adverse events in patients receiving them. The United States Food and Drug Administration (FDA) has received more than 900 adverse event reports involving IVC filters, prompting the agency to issue a warning about their use.[2] A prior study by our group demonstrated a lack of benefit of IVC filter insertion for the prevention of pulmonary embolism among bariatric surgery patients but lacked statistical power to prove harms associated with this practice.[3]

In the current study, we analyzed data from the prospective, statewide, clinical registry of the Michigan Bariatric Surgery Collaborative. Our study population now includes 35,477 bariatric surgery patients from 32 hospitals whose procedures were performed between 2006 and 2012. Since the publication of our prior study, the use of IVC filters in bariatric surgery has decreased significantly in Michigan. For this reason, our study population now includes many more high‐risk patients who did not undergo IVC filter placement, allowing us to match IVC filter patients to similarly high‐risk patients who did not receive IVC filters. We used these data to compare outcomes within 30 days of surgery, including rates of venous thromboembolism, overall serious complications, and death among patients who did and not receive IVC filters.

METHODS

RESULTS

Matching resulted in cohorts of IVC filter and control patients who were well balanced on all baseline characteristics (Table 1). In contrast, there were large and significant differences between IVC filter patients and unmatched control patients. For example, mean body mass index was 58 and 57 in the matched cohorts and 47 in the unmatched control patients. Prior history of venous thromboembolism was present in 39%, 39%, and 2% of the IVC filter, matched control, and unmatched control patients, respectively. With regard to procedure mix, unmatched control patients were less likely to have open gastric bypass and more likely to have adjustable gastric band procedures than IVC filter or matched control patients.

With regard to outcomes (Table 2, Figures 1 and 2), IVC filter patients had significantly higher rates of venous thromboembolism (1.9% vs 0.74%; OR, 2.7; 95% CI, 1.1‐6.3; P=0.027) and deep vein thrombosis (1.2% vs 0.37%, OR, 3.3; 95% CI, 1.1‐10.1; P=0.039) than matched control patients. Rates of pulmonary embolism were higher among IVC filter patients, but the difference was not statistically significant (0.84% vs 0.46%; OR, 2.0; 95% CI, 0.6‐6.5; P=0.232). Rates of pulmonary embolism were similar for patients with a low baseline risk of venous thromboembolism (0.27% vs 0.27%; OR, 1.0; 95% CI, 0.1‐7.7; P=0.965) but were higher for medium‐risk patients (2.1% vs 0.87%; OR, 2.5; 95% CI, 0.5‐12.7; P=0.288) and high‐risk patients (2.1% vs 0.97%; OR, 2.2; 95% CI, 0.2‐24.3; P=0.530).

Figure 1

Figure 2

Rates of other complications were also higher among IVC filter than matched control patients (Table 2 and Figure 3). There were significantly higher rates of complications (15.2% vs 11.6%; OR, 1.3; 95% CI, 1.0‐1.7; P=0.048), serious complications (5.8% vs 3.8%; OR, 1.6; 95% CI, 1.0‐2.4; P=0.031), and permanently disabling complications (1.2% vs 0.4%; OR, 4.3; 95% CI, 1.2‐15.6; P=0.028) among IVC filter patients. Rates of death (0.7% vs 0.1%; OR, 7.0; 95% CI, 0.9‐57.3; P=0.068) were also higher among IVC filter patients than matched control patients, but this difference was not statistically significant.

Figure 3

Among the 7 IVC filter patients who died, 4 had fatal pulmonary embolism, and 2 had IVC filter thrombosis/occlusion. Other IVC filter‐specific complications included IVC filter migration requiring heart valve replacement surgery in 1 patient, contrast‐induced nephropathy in 1 patient, IVC filter incision site infection in 1 patient, and technical difficulties removing a temporary IVC filter requiring that the device stay in place in 1 patient.

DISCUSSION

In this propensity matched, observational cohort study, we assessed the safety and effectiveness of prophylactic IVC filters among bariatric surgery patients. We found that patients with IVC filters had significantly worse outcomes than comparably high‐risk patients without IVC filters. Rates of venous thromboembolism were higher in the IVC filter patients, and a large proportion of the other complications among IVC filter patients were device related.

Our current study of IVC filters was prompted by an FDA advisory report regarding complications in patients receiving IVC filters.[2] The FDA's report was in turn prompted by a study indicating a high prevalence of strut fracture and embolization among 80 patients who received a certain type of retrievable IVC filter.[4] The FDA conveyed receiving 921 adverse events reports involving IVC filters between 2005 and 2009. Thirty‐six percent of these reports involved migration of the device, 16% were related to breakage and embolization of parts of the device, and 8% involved perforation of the IVC.

Research on the safety and efficacy of IVC filters in bariatric surgery patients has largely been limited to small, single‐center, case series or cohort studies.[5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] A systematic review of this literature concluded that the evidence was insufficient to recommend IVC filters for patients undergoing bariatric surgery.[16] A 2010 study by our group was the largest and only multicenter study of IVC filters in the bariatric surgery population. We found no benefit of IVC filters in a comparison of 542 gastric bypass patients with prophylactic IVC filters to 5,834 gastric bypass patients without prophylactic IVC filters.[3]

When interpreting the results of this study, a number of limitations should be considered. Our study was observational, so there is the potential for unmeasured confounding variables to have influenced our results. To minimize the risk of confounding, we used propensity scores to match IVC filter patients to comparably high‐risk control patients, resulting in study cohorts that were well balanced on all baseline variables. Although this method accounts for confounding on the variables for which there are data, there is still the possibility that an unknown confounder could affect our findings. For example, our clinical registry lacks data on hypercoagulable states, so it is possible that a higher proportion of IVC filter patients could have had this risk factor and therefore a higher baseline risk of venous thromboembolism. However, most patients with a hypercoagulable state would have had a prior history of venous thromboembolism, which is a variable included in our database that patients were matched on.

The effects of changes in clinical care occurring during the time frame of this study should be considered in interpreting our findings. For example, bariatric surgery has been getting safer in general over time. Rates of death have fallen both in Michigan and in the rest of the country as bariatric surgeons have gained experience with this procedure. In Michigan during this time period, our group has developed and implemented a risk‐stratified, standardized approach to venous thromboembolism prophylaxis for patients undergoing bariatric surgery. For these reasons, we included the year of the procedure and the type of medical venous thromboembolism prophylaxis (unfractionated or low molecular weight heparin) used perioperatively as a matching variable in our analysis.

Another limitation that should be considered in interpreting our findings is statistical power. Although our study is the largest in this study population to date, many of the outcomes of interest are relatively rare. Considering the entire bariatric surgery population, rates of venous thromboembolism and death within 30 days are each less than 1%. Even in the high‐risk patients included in this analysis, there were a total of just 28 (1.3%) venous thromboembolism events and 8 (0.37) deaths. Nonetheless, our study did find significantly greater risks of multiple types of complications among patients receiving IVC filters.

Finally, our study captures events occurring within 30 days of bariatric surgery. Complications, including venous thromboembolism and other complications directly related to IVC filters, frequently occur after 30 days of bariatric surgery. Therefore, our study may be a conservative estimate of the risks associated with the use of IVC filters in bariatric surgery patients. Furthermore, certain brands of filters have been shown to be associated with higher risks of complications. Our study lacks data on the brand of IVC filter used and so cannot assess the extent to which this would affect our results.

CONCLUSIONS

In conclusion, our study indicates that IVC filters do not reduce the risk of pulmonary embolism in high‐risk bariatric surgery patients. They are also associated with other complications attributable to malfunctions of the device itself. We believe that the use of IVC filters among bariatric surgery patients should be discouraged.

Disclosure

This study was supported by a grant from the Agency for Healthcare Research and Quality (HS018050) and was presented at the Annual Meeting of the American Society for Metabolic and Bariatric Surgery (ASMBS), San Diego, California, June 20, 2012.

The use of inferior vena cava (IVC) filters has increased substantially in recent years. These medical devices, which are used to prevent pulmonary embolism in patients considered to be at high risk of venous thromboembolism, were placed in 167,000 patients in 2007.¹ In 2012, it is estimated that 259,000 patients will undergo placement of an IVC filter, an increase of 55%.[1] Increasing use of IVC filters is attributable to the development of retrievable versions of the devices, which have expanded indications for use such as in bariatric surgery.

Unfortunately, the increase in the use of IVC filters has been accompanied by an increase in reports of adverse events in patients receiving them. The United States Food and Drug Administration (FDA) has received more than 900 adverse event reports involving IVC filters, prompting the agency to issue a warning about their use.[2] A prior study by our group demonstrated a lack of benefit of IVC filter insertion for the prevention of pulmonary embolism among bariatric surgery patients but lacked statistical power to prove harms associated with this practice.[3]

In the current study, we analyzed data from the prospective, statewide, clinical registry of the Michigan Bariatric Surgery Collaborative. Our study population now includes 35,477 bariatric surgery patients from 32 hospitals whose procedures were performed between 2006 and 2012. Since the publication of our prior study, the use of IVC filters in bariatric surgery has decreased significantly in Michigan. For this reason, our study population now includes many more high‐risk patients who did not undergo IVC filter placement, allowing us to match IVC filter patients to similarly high‐risk patients who did not receive IVC filters. We used these data to compare outcomes within 30 days of surgery, including rates of venous thromboembolism, overall serious complications, and death among patients who did and not receive IVC filters.

METHODS

RESULTS

Matching resulted in cohorts of IVC filter and control patients who were well balanced on all baseline characteristics (Table 1). In contrast, there were large and significant differences between IVC filter patients and unmatched control patients. For example, mean body mass index was 58 and 57 in the matched cohorts and 47 in the unmatched control patients. Prior history of venous thromboembolism was present in 39%, 39%, and 2% of the IVC filter, matched control, and unmatched control patients, respectively. With regard to procedure mix, unmatched control patients were less likely to have open gastric bypass and more likely to have adjustable gastric band procedures than IVC filter or matched control patients.

With regard to outcomes (Table 2, Figures 1 and 2), IVC filter patients had significantly higher rates of venous thromboembolism (1.9% vs 0.74%; OR, 2.7; 95% CI, 1.1‐6.3; P=0.027) and deep vein thrombosis (1.2% vs 0.37%, OR, 3.3; 95% CI, 1.1‐10.1; P=0.039) than matched control patients. Rates of pulmonary embolism were higher among IVC filter patients, but the difference was not statistically significant (0.84% vs 0.46%; OR, 2.0; 95% CI, 0.6‐6.5; P=0.232). Rates of pulmonary embolism were similar for patients with a low baseline risk of venous thromboembolism (0.27% vs 0.27%; OR, 1.0; 95% CI, 0.1‐7.7; P=0.965) but were higher for medium‐risk patients (2.1% vs 0.87%; OR, 2.5; 95% CI, 0.5‐12.7; P=0.288) and high‐risk patients (2.1% vs 0.97%; OR, 2.2; 95% CI, 0.2‐24.3; P=0.530).

Figure 1

Figure 2

Rates of other complications were also higher among IVC filter than matched control patients (Table 2 and Figure 3). There were significantly higher rates of complications (15.2% vs 11.6%; OR, 1.3; 95% CI, 1.0‐1.7; P=0.048), serious complications (5.8% vs 3.8%; OR, 1.6; 95% CI, 1.0‐2.4; P=0.031), and permanently disabling complications (1.2% vs 0.4%; OR, 4.3; 95% CI, 1.2‐15.6; P=0.028) among IVC filter patients. Rates of death (0.7% vs 0.1%; OR, 7.0; 95% CI, 0.9‐57.3; P=0.068) were also higher among IVC filter patients than matched control patients, but this difference was not statistically significant.

Figure 3

Among the 7 IVC filter patients who died, 4 had fatal pulmonary embolism, and 2 had IVC filter thrombosis/occlusion. Other IVC filter‐specific complications included IVC filter migration requiring heart valve replacement surgery in 1 patient, contrast‐induced nephropathy in 1 patient, IVC filter incision site infection in 1 patient, and technical difficulties removing a temporary IVC filter requiring that the device stay in place in 1 patient.

DISCUSSION

In this propensity matched, observational cohort study, we assessed the safety and effectiveness of prophylactic IVC filters among bariatric surgery patients. We found that patients with IVC filters had significantly worse outcomes than comparably high‐risk patients without IVC filters. Rates of venous thromboembolism were higher in the IVC filter patients, and a large proportion of the other complications among IVC filter patients were device related.

Our current study of IVC filters was prompted by an FDA advisory report regarding complications in patients receiving IVC filters.[2] The FDA's report was in turn prompted by a study indicating a high prevalence of strut fracture and embolization among 80 patients who received a certain type of retrievable IVC filter.[4] The FDA conveyed receiving 921 adverse events reports involving IVC filters between 2005 and 2009. Thirty‐six percent of these reports involved migration of the device, 16% were related to breakage and embolization of parts of the device, and 8% involved perforation of the IVC.

Research on the safety and efficacy of IVC filters in bariatric surgery patients has largely been limited to small, single‐center, case series or cohort studies.[5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15] A systematic review of this literature concluded that the evidence was insufficient to recommend IVC filters for patients undergoing bariatric surgery.[16] A 2010 study by our group was the largest and only multicenter study of IVC filters in the bariatric surgery population. We found no benefit of IVC filters in a comparison of 542 gastric bypass patients with prophylactic IVC filters to 5,834 gastric bypass patients without prophylactic IVC filters.[3]

When interpreting the results of this study, a number of limitations should be considered. Our study was observational, so there is the potential for unmeasured confounding variables to have influenced our results. To minimize the risk of confounding, we used propensity scores to match IVC filter patients to comparably high‐risk control patients, resulting in study cohorts that were well balanced on all baseline variables. Although this method accounts for confounding on the variables for which there are data, there is still the possibility that an unknown confounder could affect our findings. For example, our clinical registry lacks data on hypercoagulable states, so it is possible that a higher proportion of IVC filter patients could have had this risk factor and therefore a higher baseline risk of venous thromboembolism. However, most patients with a hypercoagulable state would have had a prior history of venous thromboembolism, which is a variable included in our database that patients were matched on.

The effects of changes in clinical care occurring during the time frame of this study should be considered in interpreting our findings. For example, bariatric surgery has been getting safer in general over time. Rates of death have fallen both in Michigan and in the rest of the country as bariatric surgeons have gained experience with this procedure. In Michigan during this time period, our group has developed and implemented a risk‐stratified, standardized approach to venous thromboembolism prophylaxis for patients undergoing bariatric surgery. For these reasons, we included the year of the procedure and the type of medical venous thromboembolism prophylaxis (unfractionated or low molecular weight heparin) used perioperatively as a matching variable in our analysis.

Another limitation that should be considered in interpreting our findings is statistical power. Although our study is the largest in this study population to date, many of the outcomes of interest are relatively rare. Considering the entire bariatric surgery population, rates of venous thromboembolism and death within 30 days are each less than 1%. Even in the high‐risk patients included in this analysis, there were a total of just 28 (1.3%) venous thromboembolism events and 8 (0.37) deaths. Nonetheless, our study did find significantly greater risks of multiple types of complications among patients receiving IVC filters.

Finally, our study captures events occurring within 30 days of bariatric surgery. Complications, including venous thromboembolism and other complications directly related to IVC filters, frequently occur after 30 days of bariatric surgery. Therefore, our study may be a conservative estimate of the risks associated with the use of IVC filters in bariatric surgery patients. Furthermore, certain brands of filters have been shown to be associated with higher risks of complications. Our study lacks data on the brand of IVC filter used and so cannot assess the extent to which this would affect our results.

CONCLUSIONS

In conclusion, our study indicates that IVC filters do not reduce the risk of pulmonary embolism in high‐risk bariatric surgery patients. They are also associated with other complications attributable to malfunctions of the device itself. We believe that the use of IVC filters among bariatric surgery patients should be discouraged.

Disclosure

This study was supported by a grant from the Agency for Healthcare Research and Quality (HS018050) and was presented at the Annual Meeting of the American Society for Metabolic and Bariatric Surgery (ASMBS), San Diego, California, June 20, 2012.

The use of inferior vena cava (IVC) filters has increased substantially in recent years. These medical devices, which are used to prevent pulmonary embolism in patients considered to be at high risk of venous thromboembolism, were placed in 167,000 patients in 2007.¹ In 2012, it is estimated that 259,000 patients will undergo placement of an IVC filter, an increase of 55%.[1] Increasing use of IVC filters is attributable to the development of retrievable versions of the devices, which have expanded indications for use such as in bariatric surgery.

Unfortunately, the increase in the use of IVC filters has been accompanied by an increase in reports of adverse events in patients receiving them. The United States Food and Drug Administration (FDA) has received more than 900 adverse event reports involving IVC filters, prompting the agency to issue a warning about their use.[2] A prior study by our group demonstrated a lack of benefit of IVC filter insertion for the prevention of pulmonary embolism among bariatric surgery patients but lacked statistical power to prove harms associated with this practice.[3]

In the current study, we analyzed data from the prospective, statewide, clinical registry of the Michigan Bariatric Surgery Collaborative. Our study population now includes 35,477 bariatric surgery patients from 32 hospitals whose procedures were performed between 2006 and 2012. Since the publication of our prior study, the use of IVC filters in bariatric surgery has decreased significantly in Michigan. For this reason, our study population now includes many more high‐risk patients who did not undergo IVC filter placement, allowing us to match IVC filter patients to similarly high‐risk patients who did not receive IVC filters. We used these data to compare outcomes within 30 days of surgery, including rates of venous thromboembolism, overall serious complications, and death among patients who did and not receive IVC filters.

METHODS

RESULTS

Matching resulted in cohorts of IVC filter and control patients who were well balanced on all baseline characteristics (Table 1). In contrast, there were large and significant differences between IVC filter patients and unmatched control patients. For example, mean body mass index was 58 and 57 in the matched cohorts and 47 in the unmatched control patients. Prior history of venous thromboembolism was present in 39%, 39%, and 2% of the IVC filter, matched control, and unmatched control patients, respectively. With regard to procedure mix, unmatched control patients were less likely to have open gastric bypass and more likely to have adjustable gastric band procedures than IVC filter or matched control patients.