High In-hospital Mortality After Percutaneous Endoscopic Gastrostomy: Results of a Nationwide Population-based Study

High In-hospital Mortality After Percutaneous Endoscopic Gastrostomy: Results of a Nationwide Population-based Study

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY 2013;11:1437–1444 High In-hospital Mortality After Percutaneous Endoscopic Gastrostomy: Results of a Nationw...

211KB Sizes 0 Downloads 10 Views

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY 2013;11:1437–1444

High In-hospital Mortality After Percutaneous Endoscopic Gastrostomy: Results of a Nationwide Population-based Study GAURAV ARORA,* DON ROCKEY,‡ and SAMIR GUPTA,§,储 *Division of Digestive and Liver Diseases, Department of Internal Medicine, University of Texas Southwestern Medical Center, Dallas, Texas; ‡Department of Medicine, Medical University of South Carolina, Charleston, South Carolina; §Department of Internal Medicine, Veterans Affairs San Diego Health Care System, San Diego, California; and 储University of California San Diego, La Jolla, California

BACKGROUND & AIMS:

It is important to carefully select patients who undergo endoscopic procedures, to optimize health care. Percutaneous endoscopic gastrostomy (PEG) is a frequently performed invasive endoscopic procedure that has been associated with high short-term mortality. We used a national database to determine the incidence of, and factors associated with, in-hospital mortality among patients undergoing PEG.

METHODS:

We conducted a nested, case-control, retrospective study using the US Nationwide Inpatient Sample (NIS) to analyze data from all hospitalizations in 2006 with an International Classification of Diseases, 9th revision, procedure code for PEG. Bivariate and multivariate logistic regression analyses were performed using demographic and clinical variables to identify predictors of in-hospital mortality after the procedure. A separate analysis using a propensity score matching technique was conducted to compare mortality with a control cohort. Results were validated in an independent analysis of 2007 NIS data.

RESULTS:

In-hospital mortality was 10.8% among 181,196 patients who underwent PEG in 2006 (95% confidence interval, 10.3%–11.3%). Odds of death increased with age (1%/y), congestive heart failure, renal failure, chronic pulmonary disease, coagulopathy, pulmonary circulation disorders, metastatic cancer, and liver disease. The indication for PEG was associated strongly with mortality. Women and patients with diabetes mellitus or paralysis had a lower risk of death. PEG was associated with slightly higher odds of in-hospital mortality compared with patients who did not undergo PEG. Qualitatively and quantitatively similar results were obtained when 2007 NIS data were analyzed.

CONCLUSIONS:

The mortality rate is almost 11% among hospital inpatients after PEG. We have identified factors that increase and decrease the risk of death after PEG; these factors could improve patient selection for those most likely to benefit from this procedure.

Keywords: National Priorities Partnership; Enteral Access for Feeding; Risk Factor; Complication.

See editorial on page 1451; see related article, Kurien M et al, on page 1445 in this issue of CGH.

T

he National Priorities Partnership, a multidisciplinary group focused on optimizing health care in the United States,1 has recognized a need to address the use of unnecessary endoscopy when potential for harm exceeds possible benefits. Percutaneous endoscopic gastrostomy (PEG) is a frequently performed invasive endoscopic procedure for providing enteral access for feeding. Prior studies have reported that short-term mortality after PEG may be as high as 25%.2– 4 This suggests that many patients receiving PEG are unlikely to survive long enough to realize a benefit from the procedure. However, these prior studies were not population-based, raising uncertainty regarding the potential magnitude of PEG overuse. Moreover, there are limited data on which patients are least likely to survive to benefit from PEG placement.

In fact, the American Society for Gastrointestinal Endoscopy has called for clarification of risk factors and for further study to elucidate predictors of mortality and morbidity that would enable patients and physicians to identify which patients are most likely to benefit from PEG.5 Because PEG is commonly used, and is invasive, expensive, and potentially associated with complications, research to inform practitioners about optimal PEG use is required. This is particularly important because less-invasive methods of enteric feeding are available and are effective. For example, a grace period of 30 to 60 Abbreviations used in this paper: CI, confidence interval; ICD-9 CM, International Classification of Diseases, 9th revision, clinical modification; NIS, Nationwide Inpatient Sample; PEG, percutaneous endoscopic gastrostomy; PSM, propensity score matching. © 2013 by the AGA Institute 1542-3565/$36.00 http://dx.doi.org/10.1016/j.cgh.2013.04.011

1438

ARORA ET AL

days feeding via a nasogastric tube before performing a PEG may achieve long-term enteral nutritional goals while maintaining a relatively low short-term mortality.6 – 8 There are also substantial cost implications, not only for the PEG procedure itself,9 but also those associated with complications as well as with actual feeding and care of the tubes. Finally, research in the area of PEG overuse may serve as a paradigm for optimizing appropriate use of all endoscopic procedures. To address the gap in the population-based study of outcomes after PEG and associated predictors of mortality, we aimed to determine the incidence of in-hospital mortality among patients undergoing PEG, and to examine factors associated with in-hospital, post-PEG mortality, using nationally representative data.

Methods Study Design, Setting, and Patient Selection We conducted a nested case-control study in a retrospective cohort of patients undergoing PEG (identified using the International Classification of Diseases, 9th revision, clinical modification [ICD-9 CM] procedure code 43.11), using a nationally representative data set of hospital admissions, the Nationwide Inpatient Sample (NIS). The NIS is an administrative claims database and is the largest all-payer inpatient care database in the United States.10 The NIS compiles data from more than 8 million inpatient stays from approximately 1000 hospitals (representing about 85% of all nonfederal hospitals). It is designed to approximate a 20% stratified probability sample of patients from all nonfederal acute-care hospitals in the United States. Validation and quality control of the NIS data are performed by an independent contractor. NIS data also have been validated extensively against the American Hospital Association Annual Survey and the National Hospital Discharge Survey and were found to be in good agreement.11 We used data from the 2006 NIS to derive our model of factors associated with post-PEG mortality, and the 2007 NIS data to validate those results. Cases were all adult patients (age, ⱖ18 y) who underwent a PEG and died during the same hospitalization. Controls were all adult patients who underwent a PEG and were alive at the time of discharge. Finally, to control for confounding by indication, we also performed a retrospective cohort study using propensity score matching (PSM) to derive a control cohort that had similar risk of in-hospital mortality as the PEG cohort, but did not actually undergo a PEG, and compared the in-hospital mortality between the 2 groups.

Analytical Approach The primary outcome was in-hospital death. The analytical variables included demographic variables such as age, sex, and race (white vs nonwhite) as well as clinical variables that we hypothesized might be associated with post-PEG mortality, based on prior reported studies: congestive heart failure, chronic pulmonary disease, renal failure, coagulopathy, pulmonary circulation disorders, metastatic cancer, liver disease, diabetes mellitus, peripheral vascular disease, solid tumor without metastasis, leukemia/ lymphoma, paralysis, and AIDS. These comorbidities were defined using the HCUP comorbidity software (version 3.5; Rockville, MD)12; this program classifies various comorbidities using the Elixhauser et al13 classification based on ICD-9 codes (Supplementary Table 1). We also included the possible indication for PEG and

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

used the following hierarchy to assign one indication per PEG: head and neck cancer, stroke, head trauma, esophageal cancer, other neurologic conditions (Parkinson disease, myasthenia gravis, amyotrophic lateral sclerosis, multiple sclerosis, Lambert–Eaton syndrome, and unclassified oropharyngeal dysphagia), dementia, malnutrition, and unknown indication. These were defined using either the HCUP clinical classification software14 or individual ICD-9 codes (Supplementary Table 1). We also included the Charlson comorbidity score (Deyo adaptation) as an analytical variable.15 Other variables that were included as potential confounders were hospital location (urban vs rural), hospital bed size (large, medium, small), hospital teaching status (teaching vs nonteaching), primary expected payer (Medicare, Medicaid, private or other [self-pay, no-charge, other, or missing]), and median household income for patient’s zip code (quartiles). The 2006 NIS was used to estimate in-hospital mortality after PEG and elucidate factors associated with post-PEG mortality. All analyses were performed using SAS version 9.2 software (Cary, NC). All reported frequencies are weighted. We took into account the complex survey design of NIS data by incorporating sampling weights, cluster, and strata information in our analyses; survey means procedure was used for descriptive statistics and survey logistic procedure (for logistic regression) was used for bivariate and multivariate analyses. Statistical procedures recommended by the database source were used to derive correct estimates of the variance.16 Continuous variables were described using means (95% confidence interval [CI]) and categoric variables as percentages (95% CI). Only variables with a significant association (P ⬍ .05 in the bivariate analyses) with mortality were included in the multivariable model. The final multivariable model was constructed by excluding all variables with a P value of .05 or greater in the initial multivariable model. The Charlson score was not included in multivariable modeling given concern for multicollinearity and was modeled separately. We also assessed the immediate postprocedure mortality (within 1 or 2 d) because it may indicate mortality that is attributable directly to the procedure itself and/or sedation/ anesthesia used during the procedure. Finally, cognizant of the fact that patients who undergo PEG are not a homogenous group, we assessed mortality stratified by the indication as well. After completing all analyses using the 2006 NIS, we validated our estimate of in-hospital post-PEG mortality and our multivariable model using the 2007 NIS.

Propensity Score Matching We included indications for PEG (described earlier), significant comorbidities associated with in-hospital mortality after PEG (as derived from the multivariable model earlier), as well as other factors possibly associated with performance of PEG or in-hospital mortality (age, sex, length of hospital stay, admission source, and various hospital characteristics such as teaching status, location, and bed size) as variables in a nonparsimonious logistic regression model used for PSM, the outcome of which was receipt of PEG. Using the Greedy Algorithm17 for PSM, an equal number of unique controls (1:1) were derived. In-hospital mortality was compared between the 2 groups using conditional logistic regression. This study was ruled exempt from formal review by the University of Texas Southwestern Medical Center Institutional Review Board, given the use of deidentified, previously collected data.

November 2013

IN–HOSPITAL MORTALITY AFTER PEG

1439

Table 1. Baseline Patient Characteristics Variable

N (weighted)

Percentage (95% CI)

Total Died during hospitalization Sex Men Women Race White Nonwhite Missing Indication for PEG Stroke Other neurologic conditionsa Malnutrition Head and neck cancer Dementia Head trauma Esophageal cancer Unknown Comorbid conditions Chronic pulmonary disease Congestive heart failure Diabetes mellitus Renal failure Paralysis Coagulopathy Metastatic cancer Peripheral vascular disease Solid tumor without metastasis Pulmonary circulation disease Liver disease Lymphoma/leukemia AIDS Expected primary payer Medicare Private insurance Medicaid Other (includes self-pay, no-charge, other, and missing) Median household income for patient’s zip code, US $ 1–38,999 39,000–47,999 48,000–62,999 ⱖ63,000 Teaching hospital (vs nonteaching) Hospital bed size39 Small Medium Large Urban hospital (vs rural)

181,196 19,562

— 10.8 (10.3–11.3)

92,848 88,343

51.2 (50.4–52.1) 48.8 (47.9–49.6)

86,383 50,284 44,529

47.6 (44.0–51.3) 27.8 (24.7–30.8) 24.6 (20.2–29.0)

35,732 35,563 33,513 9309 8741 8266 2524 47,549

19.7 (19.0–20.5) 19.6 (18.6–20.6) 18.5 (17.4–19.5) 5.1 (4.5–5.8) 4.8 (4.4–5.2) 4.6 (3.8–5.3) 1.4 (1.2–1.5) 26.2 (25.1–27.4)

44,811 42,249 38,341 29,031 23,866 14,128 9245 8452 6556 4721 3201 1577 547

24.7 (23.8–25.6) 23.3 (22.3–24.3) 21.2 (20.3–22.0) 16.0 (15.3–16.7) 13.2 (12.6–13.7) 7.8 (7.3–8.3) 5.1 (4.7–5.5) 4.7 (4.4–5.0) 3.6 (3.4–3.9) 2.6 (2.4–2.8) 1.8 (1.6–1.9) 0.9 (0.8–1.0) 0.3 (0.2–0.4)

130,419 25,531 17,115 8130

72.0 (70.5–73.5) 14.1 (13.1–15.1) 9.4 (8.5–10.4) 4.5 (3.8–5.1)

57,065 44,685 39,552 35,577 92,972

31.5 (29.0–34.0) 24.7 (22.9–26.4) 21.8 (20.3–23.3) 19.6 (17.2–22.0) 51.3 (48.4–54.3)

21,457 44,137 115,345 166,987

11.8 (10.2–13.5) 24.4 (22.1–26.6) 63.7 (61.0–66.3) 92.1 (91.0–93.3)

aIncludes

Parkinson disease, myasthenia gravis, amyotrophic lateral sclerosis, multiple sclerosis, Lambert–Eaton syndrome, and unclassified oropharyngeal dysphagia.

Dr Arora had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Results Patient Characteristics In-hospital mortality after PEG was 10.8% (95% CI, 10.3%–11.3%) among the 181,196 patients who underwent a PEG in the United States during 2006 (Table 1). The mean age

of all patients undergoing PEG was 71 years (95% CI, 70.7–71.9 y; range, 18 –116 y); 49% were women; and 47.6% were white (27.8% were nonwhite and for the remaining 24.6% race information was missing). As expected, the majority of patients were on Medicare (72%), and only a small minority were privately insured (14%); almost a third lived in a poor neighborhood. Procedural volume was divided equally between teaching and nonteaching hospitals; 92% of all PEGs were performed in urban hospitals. The comorbidity burden in the cohort was

1440

ARORA ET AL

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

metastatic cancer, and chronic pulmonary disease were associated with a 30% to 40% increased in-hospital mortality after PEG. Notably, paralysis was associated with a 50% decrease in the odds of dying in the hospital; female sex and diabetes mellitus also were associated with slightly decreased mortality. A number of comorbidities such as AIDS, lymphoma/leukemia, peripheral vascular disease, and solid tumor without metastasis were not associated with in-hospital mortality. Hospital-level variables and household income were not significant in the bivariate analysis and thus were not included in the multivariate analysis.

Multivariate Analyses

Figure 1. Post-PEG mortality stratified by indication (unadjusted). Indications for PEG were defined as per Supplementary Table 1; one indication was assigned per PEG using the hierarchy described in the Methods section. ca, cancer; HNC, head and neck cancer.

substantial as shown by a mean Charlson comorbidity index score of 2.4 (95% CI, 2.37–2.45; range, 0 –15); a quarter of all patients had congestive heart failure or chronic pulmonary disease. Malnutrition, stroke, or other neurologic conditions together accounted for the indication in 60% of the PEGs. Of the 19,562 patients who died, data on time from procedure to death was available for 79% (n ⫽ 15,475) and showed that the median number of days from PEG to death was 9 days (interquartile range, 14 d; range, 0 –259 d). Of those who died, 5% died within 1 day and 10% died within 2 days of the procedure. When stratified by the indication (Figure 1), the highest inhospital mortality was seen in the malnutrition group as well as the group in which the indication was unknown (⬃15%), and the lowest was in the head and neck cancer group (⬃3%).

Bivariate Analyses Increasing age (1% increase/y) and increasing Charlson score (7% increase per 1-unit increase) were associated with increased post-PEG mortality (Table 2). When compared with head and neck cancer, all other indications for PEG were associated with a 2- to 6-fold higher odds of death, malnutrition being the most striking. In addition, a 2-fold increased risk for post-PEG mortality was noted for patients with severe comorbid illnesses including congestive heart failure, renal failure, coagulopathy, and pulmonary circulation disease. Liver disease,

In the final multivariate model, after adjusting for all other variables in the model, the indication for PEG continued to be a highly significant predictor of in-hospital mortality, with almost similar risk associations as seen in the bivariate analysis. Metastatic cancer (90% increase), congestive heart failure (68% increase), renal failure (64% increase), pulmonary circulation disease (59% increase), liver disease (31% increase), and chronic pulmonary disease (10% increase) were statistically significantly associated with an increased risk of mortality (Table 2). Of note, age (1% increase/y) and coagulopathy (66% increase) also were associated with increased mortality. Conditions associated with decreased risk of in-hospital mortality after a PEG included paralysis (45% less risk), sex (women were 16% less likely than men to die), and diabetes mellitus (17% decreased risk). Medicaid patients had a 29% higher odds of death after adjusting for all other significant variables. The final multivariable model had a concordance index (c-index) (a measure of the area under the receiver operating curve) of 0.69. This model was superior to separate models that were derived using age alone (c-index, 0.54); Charlson score alone (c-index, 0.55); age and Charlson score together (c-index, 0.56); and age, Charlson score, and interaction of age and Charlson score (c-index, 0.56).

Validation With 2007 Nationwide Inpatient Sample As with the 2006 NIS, approximately 1 in 10 inpatients who underwent PEG did not survive to hospital discharge (9.5% among the 182,268 patients in the 2007 NIS dataset). Demographic and clinical characteristics, as well as predictors of post-PEG mortality, were also similar (Supplementary Tables 2 and 3).

Results From Propensity Score Matching Of the 181,196 patients who underwent a PEG, there were 150,140 who were matched individually to an equal number of unique controls from the NIS. Both cohorts were well matched on the variables used for PSM (Table 3). The inhospital mortality rate was slightly higher in those who underwent a PEG vs those who did not (10.5% vs 9.2%; odds ratio, 1.16; 95% CI, 1.10 –1.23). The corresponding number neededto-harm was 77. Results were qualitatively and quantitatively similar in the 2007 data set.

Discussion By using a large, population-based sample of more than 180,000 inpatients who underwent PEG in the United States, we found that the in-hospital, post-PEG mortality rate was

November 2013

IN–HOSPITAL MORTALITY AFTER PEG

1441

Table 2. Results From the Bivariate and Multivariate Analysis of In-Hospital Mortality After PEG Variable

Unadjusted odds ratio (95% CI)

Adjusted odds ratio (95% CI)a

Age (per year increase) Women (vs men) White (vs nonwhite) Charlson score Indication for PEG Head and neck cancer Malnutrition Stroke Head trauma Esophageal cancer Dementia Other neurologic conditionsa Unknown Congestive heart failure Renal failure Coagulopathy Pulmonary circulation disease AIDS Liver disease Metastatic cancer Chronic pulmonary disease Lymphoma/leukemia Peripheral vascular disease Diabetes mellitus Solid tumor without metastasis Paralysis Expected primary payer Private insurance Medicare Medicaid Other (includes self-pay, no-charge, other, and missing) Median household income for patient’s zip code, US $ 1–38,999 39,000–47,999 48,000–62,999 ⱖ63,000 Teaching hospital (vs nonteaching) Hospital bed size Small Medium Large Urban hospital (vs rural)

1.01 (1.01–1.01) 0.90 (0.83–0.95) 0.94 (0.87–1.03) 1.07 (1.06–1.09)

1.01 (1.00–1.01) 0.84 (0.79–0.91) — —

1 [referent] 5.93 (4.43–7.93) 3.96 (2.94–5.34) 3.23 (2.32–4.49) 2.90 (1.83–4.58) 2.64 (1.89–3.68) 2.27 (1.65–3.12) 5.96 (4.42–8.04) 2.19 (2.04–2.36) 2.14 (1.97–2.32) 2.11 (1.90–2.33) 1.88 (1.63–2.18) 1.56 (0.94–2.60) 1.40 (1.09–1.78) 1.35 (1.18–1.54) 1.32 (1.22–1.43) 1.19 (0.86–1.64) 1.03 (0.87–1.23) 0.86 (0.78–0.93) 0.80 (0.64–1.00) 0.51 (0.45–0.58)

1 [referent] 5.25 (3.91–7.05) 4.68 (3.44–6.36) 4.15 (2.97–5.78) 2.45 (1.55–3.87) 2.60 (1.84–3.67) 2.29 (1.66–3.15) 5.23 (3.87–7.08) 1.68 (1.55–1.82) 1.64 (1.51–1.78) 1.66 (1.50–1.85) 1.59 (1.36–1.84) — 1.31 (1.02–1.67) 1.90 (1.65–2.19) 1.10 (1.01–1.20) — — 0.83 (0.76–0.90) — 0.55 (0.48–0.63)

1 [referent] 1.30 (1.17–1.45) 1.19 (1.03–1.37) 1.06 (0.88–1.28)

1 [referent] 1.01 (0.9–1.15) 1.29 (1.12–1.48) 1.17 (0.96–1.42)

1 [referent] 0.97 (0.87–1.07) 0.93 (0.84–1.03) 0.98 (0.86–1.11) 0.92 (0.82–1.02)

— — — — —

1 [referent] 0.86 (0.73–1.00) 0.83 (0.74–0.96) 0.95 (0.81–1.12)

— — — —

aAdjusting

for all other variables in the model.

high: 10.8%. Increasing age, congestive heart failure, renal failure, chronic pulmonary disease, coagulopathy, pulmonary circulation disorders, metastatic cancer, and liver disease were associated with increasing odds of post-PEG mortality. Patients with head and neck cancer as the indication for PEG had the lowest mortality and those with malnutrition as the indication had the highest. Further, only a small minority of those who died did so shortly after PEG, suggesting that the procedure itself is relatively safe. When compared with a control group matched on indications for PEG and important comorbidities, PEG was associated with worse in-hospital mortality (16% higher odds of in-hospital death); the absolute increase in mortality risk, however, was small. Our results imply poor selection of patients for PEG as shown by the following: overall in-hospital mortality of 10.8%; median time to death after PEG of 9 days; the fact that 10% of those who died, died within 2

days of the procedure; and higher mortality compared with controls (9.2%). Two previous US studies reporting short-term mortality after PEG were performed in specific populations: a study of 81,105 Medicare beneficiaries from 1991 reported a 15.3% inhospital and 23.9% 30-day mortality,2 and another study of 7369 US veterans found the in-hospital mortality to be 23.5%.4 We found lower in-hospital mortality than these studies; our results, however, are comparable with another study that evaluated 714 patients, of whom 11% died in-hospital after PEG, with 6% patients having died within 1 week of the procedure.18 These data, indicating a high in-hospital mortality after PEG, raise the possibility that patient selection is not ideal, or that this procedure is overused. Reasons for inappropriate patient selection are likely to be complex and include PEG placement for non– evidence-based

1442

ARORA ET AL

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

Table 3. Characteristics of the Case and Control Cohorts Used in the Propensity Score Analysis

In-hospital mortality Mean age, y Mean length of stay, d Women Head and neck cancer Malnutrition Stroke Head trauma Dementia Esophageal cancer Other neurologic conditionsa Congestive heart failure Diabetes mellitus Renal failure Coagulopathy Pulmonary circulation disease Metastatic cancer Liver disease Paralysis Expected primary payer Medicare Private Medicaid Other (includes self-pay, no-charge, other, and missing) Admission source Emergency room Another hospital Another facility including long-term care Routine Teaching hospital Hospital bed size Small Medium Large Urban location of hospital

PEG cohort, % (95% CI) (N ⫽ 150,140)

Control cohort, % (95% CI) (N ⫽ 150,140)

10.5 (10.0–11.1) 71.1 (70.5–71.7) 18.7 (18.2–19.3) 49.4 (48.5–50.3) 4.2 (3.8–4.7) 32.6 (31.0–34.1) 16.8 (16.0–17.5) 4.4 (3.6–5.2) 10.0 (9.2–10.7) 1.4 (1.3–1.6) 26.3 (25.1–27.5) 22.7 (21.8–23.7) 21.6 (20.6–22.5) 16.0 (15.3–16.7) 7.4 (6.9–7.8) 2.6 (2.4–2.8) 5.2 (4.8–5.6) 1.8 (1.7–2.0) 11.7 (11.2–12.2)

9.2 (8.8–9.6) 72.3 (71.8–72.8) 17.5 (16.9–18.1) 48.8 (48.2–49.5) 4.6 (4.3–4.9) 34.5 (32.8–36.2) 16.1 (15.5–16.7) 4.5 (4.0–4.9) 11.6 (11.0–12.2) 1.6 (1.4–1.7) 26.4 (25.4–27.3) 23.8 (23.0–24.6) 22.0 (21.2–22.8) 16.9 (16.2–17.6) 7.7 (7.2–8.2) 2.7 (2.4–2.9) 6.0 (5.6–6.3) 1.9 (1.8–2.1) 12.4 (11.8–13.0)

72.2 (70.7–73.7) 14.2 (13.1–15.3) 9.1 (8.1–10.0) 4.5 (3.9–5.2)

74.7 (73.5–76.0) 12.8 (12.1–13.5) 8.3 (7.4–9.2) 4.2 (3.5–4.8)

62.5 (60.0–65.0) 8.3 (7.0–9.6) 4.8 (3.7–5.9) 24.4 (22.6–26.2) 50.8 (47.7–53.8)

62.1 (60.0–64.1) 9.1 (7.9–10.3) 5.4 (4.5–6.3) 23.4 (21.8–25.0) 51.1 (48.5–53.7)

12.2 (10.5–13.9) 24.7 (22.4–27.0) 63.1 (60.4–65.8) 91.7 (90.5–92.9)

12.4 (11.0–13.7) 24.2 (22.1–26.2) 63.5 (61.1–65.8) 91.9 (91.0–92.8)

aIncludes

Parkinson disease, myasthenia gravis, amyotrophic lateral sclerosis, multiple sclerosis, Lambert–Eaton syndrome, and unclassified oropharyngeal dysphagia.

indications, inability to predict which patients are most likely to survive long enough to derive benefit from PEG, psychosocial factors, and perhaps even financial factors. Despite evidence showing that feeding tube insertion is not associated with prolonged survival, prevention of aspiration pneumonia, healing of decubitus ulcers, or improvement in quality of life,19 –21 more than one third of nursing home residents with advanced dementia have a feeding tube inserted.22 Patient knowledge about the risks and benefits of PEG appears to be limited.23 Family members often make the decision about PEG based on an emotional response to not let their loved one starve to death. However, family members of patients who died with a feeding tube in place were significantly less likely to report excellent end-of-life care than those without a feeding tube.24 Compounding the picture is the fact that physicians consistently overestimate the benefits of a feeding tube. Almost 40% of the family members in a study reported that their physician was strongly in favor of the feeding tube placement and 11% reported feeling pressured by the physician to insert the feeding tube.24 The personal, non– evidence-based views of many health

professionals regarding tube feeding for patients with dementia also may lead to PEG overuse.25 Financial incentives for PEG placement may exist. One study of nursing home residents with advanced cognitive impairment found that when these patients were admitted to acute care hospitals, increased feeding tube insertion was associated with for-profit ownership, larger hospital size, and greater intensive care unit use.26 In addition, the practice of nursing homes preferentially accepting patients with a PEG in place27 may put pressure on the hospitals to perform a PEG prematurely to expedite acceptance and transfer. Finally, the current fee-for-service system in the United States pays for the performance of a procedure but may not pay for the counseling required to avoid an unnecessary procedure and the associated cost. If PEG is overused, the consequences likely are substantial. First and foremost, PEG is an invasive procedure that carries with it the potential for complications. PEG overuse thus results in exposure of patients unlikely to benefit from the procedure to substantial harms. Moreover, if PEG is overused, there

November 2013

are significant financial concerns. The cost of a PEG has been estimated to be at least $2200.9 Based on our data, we conservatively project that more than 400 million dollars are spent annually on this procedure in the United States. Thus, even a low rate of PEG overuse (eg, 10%) has substantial cost implications. Prior research has identified some factors associated with post-PEG mortality. These include age, body mass index, hypoalbuminemia, cancer, Charlson comorbidity score, chronic obstructive pulmonary disease, diabetes, cirrhosis, sex, serum C-reactive protein level, renal failure, and cardiac disease.3,18,28 –35 However, these risk factors have been derived from variable samples and none have used a nationally representative sample. In contrast, by taking into account the complex survey design of a nationally representative database that includes outcomes from multiple types of hospitals throughout the United States, we have produced estimates and predictors of in-hospital mortality that likely are generalizable to the population of US patients who undergo a PEG in a hospital. The large sample size of the NIS database also enabled us to perform multivariate adjustment for important covariates that are known to impact in-hospital and postoperative mortality. Importantly, independent validation of results observed with 2006 NIS data by 2007 NIS analyses further strengthened our findings. We recognize several potential limitations of our study. First, the modest concordance index of 0.69 of the model in our study indicates that there still is an opportunity to improve prediction of in-hospital mortality after PEG. Second, laboratory values are not available in the NIS; this is especially of interest because albumin and C-reactive protein values may be important predictors of post-PEG mortality.32,36 Similarly, no information on functional status is available, and this is an important predictor of survival in older adults.37 The indication for PEG in our study is only an estimate and may not accurately reflect the actual reason for the procedure. The association of diabetes mellitus with lower odds of in-hospital mortality was an unexpected finding and likely is secondary to a bias in coding; it previously has been proposed that there is “a bias in coding in which the severity of overall patient illness inversely affects the likelihood that nonthreatening conditions are coded.”38 In other words, sicker patients are less likely to have all their diagnoses coded compared with those who are relatively healthy, leading to a systematic bias that may affect the validity of association of certain diagnostic groups with inhospital mortality. Paralysis associated with lower odds of inhospital mortality may be explained by the chronic, stable nature of the condition. Finally, the true measure of futility of a procedure may be better derived from 30-day mortality rather than in-hospital mortality; the former can be derived from study of patient-level data. In conclusion, using a large, population-based sample of inpatients undergoing PEG in the United States, we determined that approximately 1 in 10 inpatients undergoing PEG do not survive to hospital discharge. The high rate of post-PEG mortality suggests that some patients undergoing PEG may be exposed to the risks of the procedure without expectation of benefit. Future research should focus on improving the selection of patients for PEG for whom benefits are likely to outweigh risks. Overall, PEG-associated risks and costs require that

IN–HOSPITAL MORTALITY AFTER PEG

1443

researchers, policymakers, and physicians should focus on optimizing safe and appropriate PEG use.

Supplementary Material Note: To access the supplementary material accompanying this article, visit the online version of Clinical Gastroenterology and Hepatology at www.cghjournal.org, and at http://dx.doi.org/10.1016/j.cgh.2013.04.011. References 1. National priorities partnership. Overuse. Available at: http:// www.nationalprioritiespartnership.org/uploadedFiles/NPP/Priorities/ 6.pdf. Accessed March 29, 2012. 2. Grant MD, Rudberg MA, Brody JA. Gastrostomy placement and mortality among hospitalized Medicare beneficiaries. JAMA 1998;279:1973–1976. 3. Johnston SD, Tham TC, Mason M. Death after PEG: results of the National Confidential Enquiry into Patient Outcome and Death. Gastrointest Endosc 2008;68:223–227. 4. Rabeneck L, Wray NP, Petersen NJ. Long-term outcomes of patients receiving percutaneous endoscopic gastrostomy tubes. J Gen Intern Med 1996;11:287–293. 5. ASGE Technology Committee, Kwon RS, Banerjee S, et al. Enteral nutrition access devices. Gastrointest Endosc 2010;72:236 – 248. 6. Cortez-Pinto H, Correia AP, Camilo ME, et al. Long-term management of percutaneous endoscopic gastrostomy by a nutritional support team. Clin Nutr 2002;21:27–31. 7. Lockett MA, Templeton ML, Byrne TK, et al. Percutaneous endoscopic gastrostomy complications in a tertiary-care center. Am Surg 2002;68:117–120. 8. Tham TC, Taitelbaum G, Carr-Locke DL. Percutaneous endoscopic gastrostomies: are they being done for the right reasons? QJM 1997;90:495– 496. 9. Callahan CM, Buchanan NN, Stump TE. Healthcare costs associated with percutaneous endoscopic gastrostomy among older adults in a defined community. J Am Geriatr Soc 2001;49:1525– 1529. 10. Overview of the Nationwide Inpatient Sample. Available at: http://www.hcup-us.ahrq.gov/nisoverview.jsp. Accessed April 4, 2012. 11. HCUP Methods Series. NIS comparison report. Available at: http://www.hcup-us.ahrq.gov/reports/methods/2008_01.pdf. Accessed April 4, 2012. 12. HCUP comorbidity software version 3.5. Available at: http:// www.hcup-us.ahrq.gov/toolssoftware/comorbidity/comoanaly2010. txt. Accessed April 4, 2012. 13. Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use with administrative data. Med Care 1998;36:8 –27. 14. HCUP clinical classification software. Available at: http:// www.hcup-us.ahrq.gov/toolssoftware/ccs/CCSUsersGuide.pdf. Accessed February 27, 2013. 15. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613– 619. 16. HCUP Methods Series. Calculating Nationwide Inpatient sample variances. Available at: http://www.hcup-us.ahrq.gov/reports/ methods/CalculatingNISVariances200106092005.pdf. Accessed April 4, 2012. 17. Parsons L. Performing a 1:N case-control match on propensity score. Available at: http://www2.sas.com/proceedings/sugi29/ 165-29.pdf. Accessed December 28, 2012. 18. Smith BM, Perring P, Engoren M, et al. Hospital and long-term outcome after percutaneous endoscopic gastrostomy. Surg Endosc 2008;22:74 – 80.

1444

ARORA ET AL

19. Finucane TE, Christmas C, Travis K. Tube feeding in patients with advanced dementia: a review of the evidence. JAMA 1999;282: 1365–1370. 20. Gillick MR. Rethinking the role of tube feeding in patients with advanced dementia. N Engl J Med 2000;342:206 –210. 21. Sampson EL, Candy B, Jones L. Enteral tube feeding for older people with advanced dementia. Cochrane Database Syst Rev 2009;2:CD007209. 22. Mitchell SL, Teno JM, Roy J, et al. Clinical and organizational factors associated with feeding tube use among nursing home residents with advanced cognitive impairment. JAMA 2003;290: 73– 80. 23. Berger JT, Hida S, Chen H, et al. Surrogate consent for percutaneous endoscopic gastrostomy. Arch Intern Med 2011;171: 178 –179. 24. Teno JM, Mitchell SL, Kuo SK, et al. Decision-making and outcomes of feeding tube insertion: a five-state study. J Am Geriatr Soc 2011;59:881– 886. 25. Hasan M, Meara RJ, Bhowmick BK, et al. Percutaneous endoscopic gastrostomy in geriatric patients: attitudes of health care professionals. Gerontology 1995;41:326 –331. 26. Teno JM, Mitchell SL, Gozalo PL, et al. Hospital characteristics associated with feeding tube placement in nursing home residents with advanced cognitive impairment. JAMA 2010;303: 544 –550. 27. Breier-Mackie SJ. PEGs and ethics revisited: a timely reflection in the wake of the Terri Schiavo case. Gastroenterol Nurs 2005; 28:292–297. 28. Lang A, Bardan E, Chowers Y, et al. Risk factors for mortality in patients undergoing percutaneous endoscopic gastrostomy. Endoscopy 2004;36:522–526. 29. Richter-Schrag HJ, Richter S, Ruthmann O, et al. Risk factors and complications following percutaneous endoscopic gastrostomy: a case series of 1041 patients. Can J Gastroenterol 2011;25: 201–206. 30. Zopf Y, Maiss J, Konturek P, et al. Predictive factors of mortality after PEG insertion: guidance for clinical practice. JPEN J Parenter Enteral Nutr 2011;35:50 –55. 31. Suzuki Y, Tamez S, Murakami A, et al. Survival of geriatric patients after percutaneous endoscopic gastrostomy in Japan. World J Gastroenterol 2010;16:5084 –5091. 32. Blomberg J, Lagergren P, Martin L, et al. Albumin and C-reactive protein levels predict short-term mortality after percutaneous endoscopic gastrostomy in a prospective cohort study. Gastrointest Endosc 2011;73:29 –36. 33. Tokunaga T, Kubo T, Ryan S, et al. Long-term outcome after placement of a percutaneous endoscopic gastrostomy tube. Geriatr Gerontol Int 2008;8:19 –23.

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

34. Abuksis G, Mor M, Segal N, et al. Percutaneous endoscopic gastrostomy: high mortality rates in hospitalized patients. Am J Gastroenterol 2000;95:128 –132. 35. Light VL, Slezak FA, Porter JA, et al. Predictive factors for early mortality after percutaneous endoscopic gastrostomy. Gastrointest Endosc 1995;42:330 –335. 36. Leeds J, McAlindon ME, Grant J, et al. Albumin level and patient age predict outcomes in patients referred for gastrostomy insertion: internal and external validation of a gastrostomy score and comparison with artificial neural networks. Gastrointest Endosc 2011;74:1033–1039 e1033. 37. Lee SJ, Lindquist K, Segal MR, et al. Development and validation of a prognostic index for 4-year mortality in older adults. JAMA 2006;295:801– 808. 38. van Walraven C, Austin PC, Jennings A, et al. A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care 2009;47:626 – 633. 39. Bed size categories, by region. Available at: http://www.hcup-us. ahrq.gov/db/nation/nis/NIS_Introduction_2006.jsp#table7. Accessed July 17, 2013.

Reprint requests Address requests for reprints to: Gaurav Arora, MD, MS, Division of Digestive and Liver Diseases, University of Texas Southwestern Medical Center, 5959 Harry Hines Boulevard, POB-1, Suite 520, Dallas, Texas 75390-8887. e-mail: [email protected]; fax: (214) 645-6294. Acknowledgments Presented in part at the oral plenary session of the American College of Gastroenterology annual scientific meeting, October 20, 2010, San Antonio, TX. Previously published in part as an abstract: Am J Gastroenterol 2010;105:S402. Conflicts of interest The authors disclose no conflicts. Funding Supported by the Dedman Family Scholarship in Clinical Care at the University of Texas Southwestern Medical Center (G.A.); and by a grant (KL2RR024983) titled, “North and Central Texas Clinical and Translational Science Initiative” from the National Center for Research Resources, a component of the National Institutes of Health, and the National Institutes of Health Roadmap for Medical Research. The contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Center for Research Resources or the National Institutes of Health.

1444.e1

ARORA ET AL

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

Supplementary Table 1. ICD-9 and V27 Diagnosis-related Group Codes Used to Define Comorbidities and Indications Variable

ICD-9 CM diagnosis codes

V27 DRGs

Congestive heart failure

398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 428.0–428.9

Pulmonary circulation disorders

415.11–415.19, 416.0–416.9, 417.9

Chronic pulmonary disease

490–492.8, 493.00–493.92, 494–494.1, 495.0– 505, 506.4 249.00–249.31, 250.00–250.33, 648.00–648.04, 249.40–249.91, 250.40–250.93, 775.1 403.01, 403.11, 403.91, 404.02, 404.03, 404.12, 404.13, 404.92, 404.93, 585.3, 585.4, 585.5, 585.6, 585.9, 586, V42.0, V45.1, V45.11, V45.12, V56.0–V56.32, V56.8 070.22, 070.23, 070.33, 070.44, 070.54, 456.0, 456.1, 456.20, 456.21, 571.0, 571.2, 571.3, 571.40–571.49, 571.5, 571.6, 571.8, 571.9, 572.3, 572.8, V42.7 042–044.9 200.00–202.38, 202.50–203.01, 203.02–203.82, 203.8–203.81, 238.6, 273.4 196.0–199.1, 789.51, 209.70, 209.71, 209.72, 209.73, 209.74, 209.75, 209.79, 789.51

Cardiac: 001–002, 215–238, 242–251, 253–254, 258–262, 265, 280–293, 296–298, 302–303, 306–313 Cardiac: 001–002, 215–238, 242–251, 253–254, 258–262, 265, 280–293, 296–298, 302–303, 306–313, or COPD asthma 190–192, 202–203 COPD asthma: 190–192, 202–203

Diabetes mellitus Renal failure

Liver disease

HIV and AIDS Lymphoma/leukemia Metastatic cancer

Solid tumor without metastasis

140.0–172.9, 174.0–175.9, 179–195.8, 209.00– 209.24, 209.25–209.3, 209.31, 209.32, 209.33, 209.34, 209.35, 209.36, 258.01–258.03

Coagulopathy

286.0–286.9, 287.1, 287.3–287.5, 289.84, 649.30–649.34 440–440.9, 441.00–441.9, 442.0–442.9, 443.1– 443.9, 444.21–444.22, 447.1, 449, 557.1, 557.9, V43.5 342.0–344.9, 438.20–438.53, 780.72

Peripheral vascular disease

Paralysis Malnutrition

Dementia Esophageal cancer Other neurologic conditions

Head and neck cancer Stroke Head trauma

Diabetes: 637–639 Kidney transplant, renal failure/dialysis: 652, 682–685

Liver: 420–425, 432–434, 441–446

HIV: 969–970, 974–977 Leukemia/lymphoma: 820–830, 834– 849 Cancer, lymphoma: 054, 055, 146– 148, 180–182, 374–376, 435–437, 542–544, 582–583, 597–599, 656– 658, 686–688, 715–716, 722–724, 736–741, 754–756, 826–830, 843– 849 Cancer, lymphoma: 054, 055, 146– 148, 180–182, 374–376, 435–437, 542–544, 582–583, 597–599, 656– 658, 686–688, 715–716, 722–724, 736–741, 754–756, 826–830, 843– 849 Coagulation disorders: 813 Peripheral vascular: 299–301

Cerebrovascular: 020–022, 034–38, 064–072

260, 261, 262, 263.0, 263.1, 263.2, 263.8, 263.9, 269.8, 269.9, 799.4, V121, 783.21, 783.22, 783.3, 783.41, 783.7, 783.9, V85.0 290.0, 290.10, 290.40, 294.10, 294.11, 331.0, 331.11, 331.19, 331.2, 331.82 150.0, 150.1, 150.2, 150.3, 150.4, 150.5, 150.8, 150.9, 151.0 787.2, 787.20, 787.21, 787.22, 787.23, 787.24, 787.29, 538, 558.1, V41.6, 340, 335.20, 332.0, 332.1, 341.8, 341.9, 358.00, 358.01, 358.1 CCS 1114 CCS 10914 CCS 23314

CCS, Clinical Classification Software; COPD, chronic obstructive pulmonary disease; ICD-9, CM, International Classification of Diseases, 9th revision, Clinical Modification.

November 2013

IN–HOSPITAL MORTALITY AFTER PEG

1444.e2

Supplementary Table 2. Descriptive Results From the 2007 NIS Analysis Variable

N (weighted)

Percentage (95% CI)

Total Died during hospitalization Sex Men Women Race White Nonwhite Missing Indication for PEG Stroke Other neurologic conditionsa Malnutrition Head and neck cancer Dementia Head trauma Esophageal cancer Unknown Comorbid conditions Chronic pulmonary disease Congestive heart failure Diabetes mellitus Renal failure Paralysis Coagulopathy Metastatic cancer Peripheral vascular disease Solid tumor without metastasis Pulmonary circulation disease Liver disease Lymphoma/leukemia AIDS Expected primary payer Medicare Private insurance Medicaid Other (includes self-pay, no-charge, other, and missing) Median household income for patient’s zip code, US $ 1–38,999 39,000–47,999 48,000–62,999 ⱖ63,000 Teaching hospital (vs nonteaching) Hospital bed size Small Medium Large Urban hospital (vs rural)

182,268 17,370

— 9.5 (9.1–10.0)

94,502 87,640

51.8 (51.0–52.7) 48.1 (47.2–48.9)

87,424 50,176 44,668

48.0 (44.7–51.3) 27.5 (24.7–30.4) 24.5 (20.4–28.6)

36,554 38,113 34,410 10,031 8401 7891 2794 44,073

20.0 (19.1–21.0) 20.9 (19.9–21.9) 18.9 (17.8–19.9) 5.5 (4.8–6.2) 4.6 (4.2–5.0) 4.3 (3.6–5.0) 1.5 (1.3–1.7) 24.2 (23.2–25.1)

43,904 39,199 39,052 29,201 23,541 14,484 10,823 9455 7166 5416 3204 1865 436

24.1 (23.1–25.0) 21.5 (20.5–22.5) 21.4 (20.6–22.2) 16.0 (15.3–16.7) 12.9 (12.3–13.5) 7.9 (7.5–8.4) 5.9 (5.2–6.7) 5.2 (4.8–5.5) 3.9 (3.6–4.2) 3.0 (3.0–3.2) 1.8 (1.6–1.9) 1.0 (0.9–1.1) 0.2 (0.2–0.3)

129,079 28,075 17,073 8040

70.8 (69.2–72.4) 15.4 (14.1–16.6) 9.4 (8.4–10.3) 4.4 (3.7–5.1)

58,154 44,207 38,610 36,256 93,030

31.9 (29.3–34.5) 24.2 (22.3–26.2) 21.2 (19.6–22.8) 19.9 (17.1–22.6) 51.0 (48.0–54.1)

18,091 44,655 119,333 167,468

9.9 (8.5–11.4) 24.5 (22.3–26.7) 65.5 (62.9–68.0) 91.9 (90.7–93.0)

aIncludes

Parkinson disease, myasthenia gravis, amyotrophic lateral sclerosis, multiple sclerosis, Lambert-Eaton syndrome, and unclassified oropharyngeal dysphagia.

1444.e3

ARORA ET AL

CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 11, No. 11

Supplementary Table 3. Results From the Validation of Multivariate Analysis From the 2006 NIS in the 2007 NIS Database Variable

Adjusted OR (95% CI)a

Age (per year increase) Women (vs men) White (vs nonwhite) Charlson score Indication for PEG Head and neck cancer Malnutrition Stroke Head trauma Esophageal cancer Dementia Other neurologic conditionsa Unknown Congestive heart failure Renal failure Coagulopathy Pulmonary circulation disease AIDS Liver disease Metastatic cancer Chronic pulmonary disease Lymphoma/leukemia Peripheral vascular disease Diabetes mellitus Solid tumor without metastasis Paralysis Expected primary payer Private insurance Medicare Medicaid Other (includes self-pay, no-charge, other, and missing) Median household income for patient’s zip code, US $ 1–38,999 39,000–47,999 48,000–62,999 ⱖ63,000 Teaching hospital (vs nonteaching) Hospital bed size Small Medium Large Urban hospital (vs rural)

1.01 (1.01–1.02) 0.85 (0.79–0.91) — —

aAdjusting

for all other variables in the model.

1 [referent] 5.38 (4.02–7.20) 4.09 (3.05–5.49) 3.56 (2.49–5.08) 1.45 (0.90–2.36) 2.92 (2.08–4.10) 2.48 (1.83–3.37) 5.68 (4.26–7.57) 1.54 (1.42–1.68) 1.51 (1.37–1.65) 1.68 (1.49–1.89) 1.51 (1.24–1.84) — 1.51 (1.18–1.93) 1.96 (1.68–2.28) 1.19 (1.09–1.30) — — 0.75 (0.68–0.83) — 0.55 (0.47–0.64) 1 [referent] 1.11 (0.97–1.28) 1.43 (1.20–1.71) 1.51 (1.19–1.93)

— — — — — — — — —