Potentially avoidable admissions

There were a total of 15 million emergency admissions in the 3-year period 2008–2011, approximately 5 million in each year and increasing over time. Twenty-two percent (3 273 395/14 998 773) of these admissions were for conditions identified as rich in potentially avoidable admissions. Chest pain, abdominal pain, urinary tract infections, acute mental crisis and COPD accounted for over two-thirds of these potentially avoidable emergency admissions (table 1) and 29% occurred in people aged over 75 years old.

Variation in age sex adjusted admission rates for conditions rich in avoidable admissions by system

Two systems reported zero emergency admissions for 2010/11 because they had merged in 2010. We excluded them from the analysis, leaving 150 systems. The median age sex adjusted admission rate for conditions rich in avoidable admissions was 2258 (IQR 1808–2662), with a 3.4-fold variation between systems ranging from 1268 to 4359 and a 1.9-fold variation between the 10 and 90 percentiles. Geographical variation in the rate was apparent, with the highest rates clustering in the northwest of England and northeast London (figure 1). There was a strong correlation between the age sex adjusted admission rates for conditions rich in avoidable admissions and directly standardised emergency admission rates in 2009/10 (Pearson's R=0.88, p<0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Map of age sex adjusted admission rates for conditions rich in avoidable admissions for emergency and urgent care systems.

Factors affecting age sex adjusted admission rates for conditions rich in avoidable admissions

We identified routine data for some of the relevant factors (see online supplementary table S2).19–29 We could find no routinely available data on availability and quality of intermediate care services or integration between services for these systems. We tested each factor separately in a univariate linear regression (see online supplementary table S2). Then we undertook multiple linear regression to identify the combination of factors which predicted potentially avoidable admission rates (table 2). In the univariate linear regression, two factors explained a large amount of variation in the rate. Both were population factors related to deprivation: income deprivation and employment deprivation (see online supplementary table S2). In the multiple linear regression we undertook the block 1 analysis—which tested population, geographical and health factors—and identified two factors which predicted potentially avoidable admission rates: employment deprivation (population factor) and urban/rural status (geographical factor). The more deprived populations and more urban populations had higher rates. These two factors together predicted 75% of the variation, with employment deprivation alone predicting 72% (table 2). In the block 2 analysis—which tested factors modifiable by services—factors related to emergency departments, hospitals, emergency ambulance services and primary care explained further variation (R²=85%). Systems with higher rates of potentially avoidable admissions had higher attendance rates at emergency departments, higher rates of conversion of emergency department attendances to admissions, higher proportions of very short stay patients, higher rates of ambulance calls transported to hospital and better perceived access to GP (table 2). There was no evidence of multicollinearity in block 2: variance inflation factors varied between 1.1 and 1.4.

Some of the predictors in the multiple linear regression were correlated with others not included in the final regression. It is important to describe these correlations to facilitate interpretation of the final multiple regression. Employment deprivation was positively correlated with the prevalence of some of the 14 conditions (table 3). Urban/rural status was positively correlated with a number of factors, including the proportion of older people in a system and some primary care factors. Some of the factors in the multiple regression were correlated with each other, even though there was no multicollinearity: employment deprivation was negatively correlated with the proportion of ambulance callers transported to hospital; urban/rural status was positively correlated with perceptions of access to GP and the proportion of ambulance callers transported to hospital.

Summary of findings

Over a fifth of all emergency admissions in England between 2008 and 2011 were accounted for by 14 conditions that are likely to be rich in potentially avoidable admissions. There was considerable variation in age sex adjusted admission rates for conditions rich in avoidable admissions calculated for different emergency and urgent systems in England, with high rates clustering in the northwest and northeast London. A unique aspect of this study was consideration of the effect of the wider system of services offering emergency and urgent care on these emergency admission rates. Most of the variation in these potentially avoidable emergency admissions was explained by factors outside the control of emergency and urgent care system providers. Systems with higher rates of employment deprivation, and with urban geography, had higher rates of potentially avoidable emergency admissions. Employment deprivation alone explained 72% of the variation. This factor is likely to represent deprivation more generally because income deprivation had equally large explanatory power in the univariate regression. It is not clear from this study how deprivation affects admissions but it can be partly explained by higher levels of illness in deprived communities because deprivation was correlated with measures of disease prevalence.

Urban-based systems had higher rates of potentially avoidable emergency admissions than rural ones. However, the urban/rural factor was correlated with factors measuring access to GP and GP out of hours services, with rural areas having better perceived access to GPs. The urban/rural factor was also correlated with the proportion of ambulance calls not transported to hospital. This highlights the difficulty in interpretation of our findings (and those of other similar studies) because the causal pathways are unclear. Our hierarchical multiple regression assumes that services cannot modify their population or geography. Yet it may be that services in rural areas can tailor provision based on geography by keeping patients at home and in their community because of the distance to local hospitals, suggesting that services in urban areas could do the same. Or, it may be that services are able to operate in rural areas in ways that are simply not possible in urban areas.

There was evidence that variation in the operation and use of key emergency services—emergency departments, emergency ambulances, primary care and hospitals—accounted for a further amount of the remaining variation. The emergency department attendance rate, the conversion rate of emergency department attendances into hospital admissions, and the proportion of admissions with a very short length of stay were higher in some systems than others and were associated with higher rates of potentially avoidable admissions. The latter factor about short length of stay may be related to hospitals coding the same cases differently (as an admission or a discharge from emergency department) or managing the same cases differently because they are more or less able to discharge emergency department attendees back to the community. Some ambulance services kept a large proportion of callers away from hospitals, either by giving advice over the telephone or sending a response without then taking the person to hospital. There were no routinely available data on availability and quality of intermediate care services or integration between services for these systems, so we could not assess the effect of these potentially important factors.

Strengths and weaknesses

The strengths of the study were that it focused on admissions for conditions rich in avoidable admissions and moved beyond previous research which focused on primary care factors explaining emergency admissions. That is, it took a system-wide focus, recognising that a range of services can contribute to avoiding admission. There were six limitations. First, factors could only be tested if the data were available routinely by PCT and there will be other factors affecting admission that we were unable to include. Importantly some factors will be related to the population, which healthcare managers are unable to change, such as severity of illness. A key missing variable was distance to hospital because this has been shown to explain variation in emergency admission rates between GPs, with higher emergency admissions for practices close to a hospital.7 We could not generate a distance variable because PCT populations often do not use a single hospital but a number of hospitals within or outside the PCT boundaries. However, the geographical variable in our analysis may have addressed some of this issue in that distance to hospital is likely to be longer in rural than urban areas. Other important missing variables were related to social services, nursing homes, intermediate care, and system-level variables such as integration and coordination between services. Second, a large number of predictors were tested (n=30) for the number of subjects (n=150) and this can result in some factors being significant by chance and an over-optimistic regression with regards to goodness of fit. Third, the final regression was dependent on the variables that were tested for inclusion. For example, at an earlier stage of the study we included length of stay <2 days rather than <1 day because we were concerned that coding differences might account for variation. Later, we decided that length of stay less than 2 days might reflect discharge policy rather than admission policy so changed it to less than 1 day. By making this decision, the rate of attendance at emergency departments became a statistically significant addition to the regression. Fourth, it was not clear exactly how the factors affected admission rates. For example, high emergency department attendance rates might indicate a lack of availability or accessibility of alternative sources of urgent care in systems for example, minor injury units or walk-in centres, or that an emergency department was located in a densely populated area with short travel distance to the emergency department. Higher conversion rates from emergency department attendance to admissions might be caused by risk adverse decision making in the emergency department or lack of access to alternative services to admission at the emergency department. The short length of stay might indicate coding differences between hospitals for the same types of cases, or different management practices related to the emergency department, or simply that avoidable admissions tend to stay in hospital for less than a day. The next stage of our wider study is case studies of six systems with high and low admission rates for conditions rich in avoidable admissions to study systems in depth and understand how emergency and urgent care services are managed. Fifth, data errors in routine data may account for some of the variation. We substituted the median for very high values for two factors. The findings were the same when we included these values. Finally, regressions can overestimate correlation if the dependent and independent variables share the same denominator. Many of our variables were based on the PCT population. Taking the logarithm of the independent variables removes this problem. This made little difference to the findings. For example, employment deprivation explained 70% rather than 72% of variation.

Comparison with other studies

Deprivation has been identified as an important factor within multiple regressions, explaining variation in emergency admissions.7 ,9 ,30–34 It has also been identified as explaining very large amounts of the variation in emergency admissions between GPs,30–33 although this is not always the case.35 Disability living allowance explained 62% of variation in standardised emergency admission rates between primary care groups,30 socio-demographic patient characteristics accounted for 45% of variation in emergency admissions between GPs,31 age and deprivation explained 64% of variation between practices,32 and age, sex, ethnicity and deprivation explained 64% of preventable hospital admissions.33 The reasons proposed for deprivation explaining so much variation were that people in deprived areas may be more ill, may present later or may not have social support at home.31 We have shown that deprivation and prevalence of disease are correlated and indeed recent evidence shows a strong association between deprivation and multimorbidity.36 There is likely to be more to deprivation than high levels of morbidity because in our regression deprivation explained much more variation than disease prevalence. Researchers suggest that people from deprived communities may have difficulty negotiating access to services when they are ill.37 In the context of our study, this may result in under-use of existing community-based services which could help people to avoid a hospital admission.

Primary care factors have been tested in previous studies and found to explain small amounts of variation in emergency admissions7 ,9 ,31 ,35 or none at all.32 ,33 Aspects of primary care included patients reporting being able to see a particular GP,7 having access to a GP within 2 days,9 or being able to book an appointment with a preferred doctor.36 These aspects of perceived access to in-hours GP are relevant to avoidable emergency admissions because people with acute exacerbations may not be able to access urgent appointments at their GP or may perceive this to be a barrier. Indeed in our univariate regression, perceptions of good access to a GP within 48 h were related to lower admission rates. This factor was included in the final multiple regression but the relationship was in the opposite direction. This might be explained by the inclusion of emergency department attendance rates which in other studies were predicted by satisfaction with telephone access to GP38 and amount of support from a primary care team.39 That is, perceived poor access with GP can increase emergency department attendances which in turn result in emergency admissions, and perceived good access to urgent GP may also increase emergency admissions. One study found that the hospital which patients were admitted to explained 10% of variation,31 and we found that the hospital may have a significant effect in terms of admitting people who stay for less than a day.

Implications for policymakers and service providers

We found that deprivation accounts for a large amount of variation in admissions of selected health conditions where admission could be avoided by emergency and urgent care systems. Deprivation is strongly associated with multimorbidity and we have not been able to disentangle complex causal pathways using our study design. It is important to focus on what healthcare providers can do to reduce these avoidable admissions. We suggest investigating whether short-stay emergency admissions are related to coding differences or hospital management of emergency department attendances, why some emergency departments convert more attendances to admissions than others, why some populations think they have better access to urgent GP appointments than others, and whether ambulance services in urban areas are able to match those in rural areas by transporting lower proportions of callers to hospital. It is important not to treat population deprivation as a nuisance within this type of analysis. A key finding here is that the greatest potential for health services to reduce avoidable emergency admissions lies with focusing interventions to reduce admissions on deprived communities.