Deprivation influences the emergency admission rate of ambulatory care sensitive conditions

Background

St James’ Hospital (SJH) serves as a secondary care centre for emergency admissions for its local Dublin catchment area of 270,000 adults. SJH’s catchment area is largely inner city with over 50% of patients considered to be deprived; hospitals serving deprived catchment areas face particular resource problems in meeting the needs of the population that they serve.^10,11 Emergency medical patients are admitted from the ED to an acute medical admission unit, the operation and outcome of which have been described elsewhere.^4,5

Data collection

For audit purposes we employed an anonymous patient database assembling core information about each clinical episode from elements contained on the patient administration system, the national hospital inpatient enquiry (HIPE) scheme, the patient electronic record, the emergency room and laboratory systems. HIPE is a national database of coded discharge summaries from acute public hospitals in Ireland.^12,13 Ireland used the International classification of diseases, 9th revision, clinical modification (ICD-9-CM) for both diagnosis and procedure coding from 1990 to 2005, and ICD-10-CM since then.

Data held on the database include the unique hospital number, admitting consultant, date of birth, gender, area of residence, principal and up to nine additional secondary diagnoses, principal and up to nine additional secondary procedures, and admission and discharge dates. Additional information cross-linked and automatically uploaded to the database includes physiological, haematological and biochemical parameters. For the present study data were related to all emergency general medical patients admitted to SJH from 2002 to 2013.

Approximately 11.5% of our patients stay >30 days with a median length of stay (LOS) of 55.0 days (inter-quartile range (IQR) 38.8–97.6) – many of these are awaiting nursing home beds or other social interventions rather than ongoing admissions for medical reasons. We have therefore chosen a truncated end-point (death or episode completed by the 30-day endpoint) for analysis, to avoid the additional confounding of non-medical reasons for continuing as an inpatient.

ACSCs

Discharge codes were interrogated to construct the total occurrence as well as the number of such ACS diagnoses present per patient. These codes were broadly by system: i) cardiovascular – angina, essential hypertension, congestive heart failure; ii) respiratory – chronic obstructive airways disease, asthma, pneumonia/influenza; iii) neurological – convulsions/epilepsy; iv) gastrointestinal – gastrointestinal bleeding/perforated ulcer, gastroenteritis/dehydration; v) diabetes; vi) renal – pyelonephritis; vii) infective – cellulitis, ENT infections, pelvic inflammatory diease; viii) haematological – anaemia; ix) nutritional deficiencies; x) gangrene; and xi) dental conditions. We categorised patients into five groups on this basis, zero meaning no ACSC present. As only 2.7% of patients had four or more, we grouped all patients with four or more ACSC present into the top category.

Deprivation indices

To classify deprivation, we used the public domain Irish National Deprivation Index for Health and Health Services Research, derived by the Small Area Health Research Unit at Trinity College Dublin.¹⁴ We have described the relevant methodology in more detail previously.¹¹ From the 2006 census returns, information by electoral division was available in respect of the total population within each unit area and the age dependency ratio (proportion of population <15 years or >64 years old). The incidence rate (rate/1,000 population) for emergency medical admissions for each electoral division was calculated by the average yearly admissions divided by the total population.

Statistical methods

Descriptive statistics were calculated for background demographic data, including mean/standard deviation (SD), median/ IQR or percentages. Comparisons between categorical variables and mortality were made using chi-square tests. We used a stepwise logistic regression model to predict 30-day in-hospital death, adjusting risk estimates for major predictor variables. We used margins to estimate and interpret adjusted predictions for sub-groups, while controlling for other variables, using computations of average marginal effects. In the multivariate model, we adjusted for previously determined outcome predictor variables of acute illness severity,^15,16 Charlson comorbidity index,¹⁷ sepsis status¹⁸ and chronic disabling disease score.¹⁹ For LOS data, we employed a zero-truncated Poisson regression model including predictive categorical variables in the model as a series of indicator variables. We used robust standard errors for the parameter estimates.²⁰ The Poisson regression coefficients are the log of the rate ratio, the rate at which events occur is called the incidence rate. Thus we can interpret the coefficients in terms of incidence rate ratios (IRRs).

Adjusted odds ratios (OR) and 95% confidence intervals (CI) or IRRs were calculated for those predictors that significantly entered the multivariate model. Statistical significance at p<0.05 was assumed throughout. Stata version 13.1 (Stata Corp, College Station, TX, USA) statistical software was used for analysis.