Multivariable logistic regression was used to model the effects of both race and ethnicity on the risk of death. The effects of race and ethnicity were adjusted for variations in comorbid disease and for conditions related to aspiration pneumonia that are present on admission, as well as for demographic and hospitalization characteristics. We analyzed differences among racial and ethnic groups in the frequency of categories of comorbid disease. Finally, we compared the magnitude of the effects of selected categories of comorbid disease on the risk of in-hospital death in each racial and ethnic group.
The study population included all patients discharged from California hospitals during the period from 1996 to 1999 with aspiration pneumonia (principal ICD-9-CM diagnosis code 507.0) listed as the principal cause of hospitalization. The following information was gathered from each patient record: unique patient identifier, age, race, sex, month of admission, year of admission, emergency admission, status at discharge, principal diagnosis, all secondary diagnoses, and a present-on-admission indicator for each reported secondary diagnosis. Kamagra Oral Jelly
Defining Comorbid Disease and Other Characteristics
A panel of five physicians reviewed the collection of secondary diagnoses reported as present on admission for patients with a hospitalization caused by aspiration pneumonia. Panelists used a software-support Delphi process to identify which ICD-9-CM codes could be used as indicators of comorbid disease and which ICD-9-CM codes could be used as indicators of conditions closely related to aspiration.
ICD-9-CM diagnoses reported as present on admission were organized into 260 clinically coherent categories using the structure of the Clinical Classification System (CCS) The CCS was developed by the federal Agency for Healthcare Research and Quality to create mutually exclusive disease categories from the ICD-9 taxonomy.
Panelists reviewed 260 categories of ICD-9-CM diagnoses grouped by the CCS. Within each group, panelists independently scored each diagnosis code on a nine-point scale to indicate whether they considered the diagnosis as likely to be closely related, equivocal, or unlikely to be closely related to the principal diagnosis of aspiration pneumonia. “Closely related” was conceptually defined as a condition likely to be a direct cause of aspiration pneumonia, directly caused by aspiration pneumonia, or related to aspiration pneumonia through a shared cause. Panelists could globally score all diagnoses within a group whenever appropriate.
Diagnosis codes for which the panel scores disagreed were reviewed and rescored by all panel members. We defined disagreement as any diagnosis code scored as both closely related and unlikely to be closely related by any two panel members,. Panelists independently rescored the diagnosis codes for which there was disagreement, using both their original score and the series of scores from all panel members as a guide. Panelists were blinded to the identity of all panelist scores except their own.
The results of the panelists’ review were used to divide all present at admission diagnoses into categories of comorbid disease and categories of conditions closely related to aspiration pneumonia, using the CCS structure to categorize individual ICD-9-CM codes. Diagnosis codes that the panelists agreed were closely related to aspiration pneumonia were grouped into categories using the CCS structure. All of the remaining diagnoses were grouped into categories of comorbid disease also using the CCS structure. A complete listing the specific ICD-9-CM diagnoses and categories defined by the panelists is available upon request from the authors.
Modeling Effect of Race and Ethnicity
Multivariable logistic regression was used to model the independent effects of both race and ethnicity on the risk of hospital death, while controlling for the confounding effects of differences in comorbid disease, conditions closely related to aspiration pneumonia, and other covariates that influence mortality risk. The regression model included adjustments for differences in age, sex, whether the hospitalization was an emergency admission, and whether the admission followed transfer from another acute care hospitalization. The effects of age were included in the model by categorizing age into decades to account for nonlinear effects. African Americans, Asians, and Native Americans were compared to white discharges for mortality outcomes. The effect of ethnicity was assessed by comparing Hispanics to non-Hispanics. The multivariable logistic regression model was developed using data from the 1996 through 1998 calendar years and validated by applying the fitted model equation to all discharges occurring in 1999, thus accounting for the potential of overfitting the model.
canada drugs pharmacy
The predictive accuracy of the multivariable logistic regression model was measured using both the С statistic and pseudo-R. The С statistic is equivalent to the area under the “receiver operating characteristic” curve for models with a dichotomous response variable, and it provides an estimate of the model’s ability to discriminate between observed instances of inpatient death and survival. A value of 0.5 indicates that the model provides no predictive discrimination, while a value of 1.0 indicates perfect discrimination by predicted risk of inpatient death. The pseudo-Rstatistic measures the amount of variability explained by the model. Pseudo-R is also the amount by which the average predicted probability of death for those discharges who died in the hospital exceeds the average predicted probability of death for those discharges who survived. Our multivariable logistic regression model demonstrates greater predictive accuracy than existing models designed to adjust for comorbid disease conditions.
Differences in lengths of stay by race or ethnicity have the potential to confound our analysis of mortality outcomes. To evaluate this possibility, we tested whether there were statistically significant differences in the mean length of stay by race or by ethnicity using single factor analysis of variance model.
Comorbid Disease Prevalence and Effects
Comorbid diseases are important predictors of in-hospital death, and controlling for these illnesses is essential for assessing the true effects of race and ethnicity. Differences in the prevalence and effects of comorbid disease by race and ethnicity were of particular interest. We hypothesized that the effects of race and ethnicity on mortality might be expressed in part by racial and ethnic differences in the effects of comorbid disease on mortality. To examine this possibility, we identified the 10 most commonly occurring categories of comorbid disease for which there were statistically significant differences in frequency by race and by ethnicity using the chi-square statistic. This test assesses only whether there is a statistically significant difference in the prevalence of the selected comorbid diseases by any race or ethnicity. buy levitra 20 mg
We measured the race-specific effects of each of the 10 selected categories of comorbid disease, respectively, by adding interaction terms to the original multivariable regression model for combinations of race and comorbid disease. We repeated this analysis for combinations of ethnicity and the 10 selected categories of comorbid disease. The statistical significance of the race- and ethnicity-specific effects was then calculated by a global test of the statistical significance of the interaction terms added to the model original model.
We supplemented the statistical tests of the interaction terms with an analysis of the specific effect of each of the 10 comorbid diseases in racial and ethnic subpopulations. Separate multivariable regression models were developed in subsets of the total population stratified by race and by ethnicity, respectively. In each subset, we compared the race-specific and ethnic-specific odds of in-hospital death for each category of comorbid disease adjusted for all of the covariates included in the original multivariable regression model. suhagra