Mortality and Time to Endoscopy in Upper GI Hemorrhage
This cross-sectional retrospective study used hospital discharge data from the Nationwide Inpatient Sample (NIS) from 2002 to 2007. The NIS is a database of hospital in-patient stays and is a part of the Healthcare Cost and Utilization Project (HCUP), which is sponsored by the Agency for Healthcare Research and Quality (AHRQ). The NIS is the largest all-payer in-patient care database in the US. The most recent database contains data from approximately eight million hospital stays, from 1044 hospitals in 40 different states. It is the only national hospital database containing charge information on all patients, regardless of payer, including persons covered by Medicare, Medicaid, private insurance and the uninsured.
We analysed all adult patients (≥18 years old) who had a primary or secondary diagnosis of AVH or NVUGIH, as identified by International Classification of Diseases, 9th Revision (ICD-9) code, between 2002 and 2007. The NIS database stores up to 15 diagnosis codes and 15 procedure codes for each visit. It is important to note that in administrative claims data, the primary code may not be the 'chief complaint' and often represents the medical problem which can be billed at the highest level.
Patients were excluded if they lived in one of the 10 states that do not participate in the NIS. Records that did not contain details on in-patient death, race, and day of procedure were removed from the analysis, which represented 169 002 patients (27.9%) with NVUGIH and 11 154 (23.8%) patients with AVH.
The ICD-9 codes used for diagnoses of AVH and NVUGIH are listed in Supporting Information Tables S1 and S2. The broad selection of ICD-9 codes used for the analysis was intentionally chosen to retrospectively analyse the patient population as they present to the emergency department, when the aetiology of their UGIH has a broad differential diagnosis prior to endoscopy. Haematemesis and melaena were not included unless one of the other diagnoses was an ICD-9 code listed in supporting information Tables S1 or S2. UGIH that resulted from medication use (NSAIDs) were not sub-categorised in the analysis because this data is not included in the NIS database. Finally, UGIH that occurred during or as a result of medical care received while in the hospital could not be distinguished as the NIS datasets are completed at the time of discharge and not based upon admission diagnoses. The ICD-9 procedure codes for OGD are listed in supporting information Table S3.
The multivariate regression analysis initially included gender, race category, age category, insurance payer, hospital details (region, size, location, ownership), comorbidities, income by zip code, OGD within 1 day of admission, surgical intervention and OGD during hospitalisation. Several variables were not found to be statistically significant or had a significantly low Wald score, and were not included in the final regression model which is described in the Results section.
Certain variables were modified for the analysis. Specifically, race was categorised into White, African American, Hispanic and a new category – Other – that encompassed all other races. Age and number of comorbidities were converted from scale into categorical variables. Age categories (<40, 40–60, 60–80, >80) were stratified around the median ages of the populations for NVUGIH (66.3) and AVH (55.3). Finally, the comorbidity index was calculated using the Deyo-Charlson Modification ICD-9 codes, a well-documented comorbidity index for administrative data.
The study subjects were described on various descriptive measures by their means ± standard deviations, simple proportions and percentages. Basic demographic variables selected for the analysis included age, gender, race, hospital details, comorbidity and primary insurance type. Population frequencies and basic management outcomes were described. Continuous variables, group means and standard deviations were calculated accordingly. Chi squared tests and the Mantel Haenzel Chi squared tests were used to see if there were significant differences between subjects with AVH and NVUGIH for mortality and for receipt of OGD within 1 day of admission. When appropriate, analysis of variance methods (anova) was used at the 5% level of significance to compare values of interest. Next, we compared risk factors for mortality and risk factors for receipt of OGD within 1 day of admission for both AVH and NVUGIH, using logistic regression. Predictive models for mortality or receipt of OGD within 1 day of admission were developed using multiple logistic regression analysis using backward regression procedure. All statistical analysis was run first on the entire sample and then separately for men and then women. All statistical analyses were performed using spss v16.0 software for Windows (SPSS Inc., Chicago, IL, USA).
Methods
This cross-sectional retrospective study used hospital discharge data from the Nationwide Inpatient Sample (NIS) from 2002 to 2007. The NIS is a database of hospital in-patient stays and is a part of the Healthcare Cost and Utilization Project (HCUP), which is sponsored by the Agency for Healthcare Research and Quality (AHRQ). The NIS is the largest all-payer in-patient care database in the US. The most recent database contains data from approximately eight million hospital stays, from 1044 hospitals in 40 different states. It is the only national hospital database containing charge information on all patients, regardless of payer, including persons covered by Medicare, Medicaid, private insurance and the uninsured.
We analysed all adult patients (≥18 years old) who had a primary or secondary diagnosis of AVH or NVUGIH, as identified by International Classification of Diseases, 9th Revision (ICD-9) code, between 2002 and 2007. The NIS database stores up to 15 diagnosis codes and 15 procedure codes for each visit. It is important to note that in administrative claims data, the primary code may not be the 'chief complaint' and often represents the medical problem which can be billed at the highest level.
Patients were excluded if they lived in one of the 10 states that do not participate in the NIS. Records that did not contain details on in-patient death, race, and day of procedure were removed from the analysis, which represented 169 002 patients (27.9%) with NVUGIH and 11 154 (23.8%) patients with AVH.
The ICD-9 codes used for diagnoses of AVH and NVUGIH are listed in Supporting Information Tables S1 and S2. The broad selection of ICD-9 codes used for the analysis was intentionally chosen to retrospectively analyse the patient population as they present to the emergency department, when the aetiology of their UGIH has a broad differential diagnosis prior to endoscopy. Haematemesis and melaena were not included unless one of the other diagnoses was an ICD-9 code listed in supporting information Tables S1 or S2. UGIH that resulted from medication use (NSAIDs) were not sub-categorised in the analysis because this data is not included in the NIS database. Finally, UGIH that occurred during or as a result of medical care received while in the hospital could not be distinguished as the NIS datasets are completed at the time of discharge and not based upon admission diagnoses. The ICD-9 procedure codes for OGD are listed in supporting information Table S3.
Statistical Analysis
The multivariate regression analysis initially included gender, race category, age category, insurance payer, hospital details (region, size, location, ownership), comorbidities, income by zip code, OGD within 1 day of admission, surgical intervention and OGD during hospitalisation. Several variables were not found to be statistically significant or had a significantly low Wald score, and were not included in the final regression model which is described in the Results section.
Certain variables were modified for the analysis. Specifically, race was categorised into White, African American, Hispanic and a new category – Other – that encompassed all other races. Age and number of comorbidities were converted from scale into categorical variables. Age categories (<40, 40–60, 60–80, >80) were stratified around the median ages of the populations for NVUGIH (66.3) and AVH (55.3). Finally, the comorbidity index was calculated using the Deyo-Charlson Modification ICD-9 codes, a well-documented comorbidity index for administrative data.
The study subjects were described on various descriptive measures by their means ± standard deviations, simple proportions and percentages. Basic demographic variables selected for the analysis included age, gender, race, hospital details, comorbidity and primary insurance type. Population frequencies and basic management outcomes were described. Continuous variables, group means and standard deviations were calculated accordingly. Chi squared tests and the Mantel Haenzel Chi squared tests were used to see if there were significant differences between subjects with AVH and NVUGIH for mortality and for receipt of OGD within 1 day of admission. When appropriate, analysis of variance methods (anova) was used at the 5% level of significance to compare values of interest. Next, we compared risk factors for mortality and risk factors for receipt of OGD within 1 day of admission for both AVH and NVUGIH, using logistic regression. Predictive models for mortality or receipt of OGD within 1 day of admission were developed using multiple logistic regression analysis using backward regression procedure. All statistical analysis was run first on the entire sample and then separately for men and then women. All statistical analyses were performed using spss v16.0 software for Windows (SPSS Inc., Chicago, IL, USA).
SHARE
