Measuring patient complexity in hospital admissions Multiple analytical approaches with one consistent message

In our recent briefing Emergency hospital admissions in England: which may be avoidable and how? we looked at the growing demand for emergency care in NHS hospitals. We found that emergency admissions had increased by half and that pressure on hospitals was compounded by the increasing age of patients being admitted.  We also wanted to tell the story of the increasing complexity of patients’ health needs that might arise from an ageing population. So, one of the issues we grappled with in our analysis was how best to measure the mixture of acute and chronic conditions and other characteristics that make a patient more complex to care for. Surprisingly there isn’t a universally agreed way to define the complexity of patients’ health needs from the routine hospital data we had access to.

Defining complexity

One possible approach was to measure the number of selected long-term conditions a patient has – conditions such as cancer, diabetes, asthma, or heart disease. This method was used previously in Health Foundation analysis. It is a good indicator of the level of health need across the population and will certainly influence health care spending, but it doesn’t tell the whole story of a patient’s complexity. We needed a method that took in the range of other acute events like hip fracture, or pneumonia, as well long-term conditions and other risk factors.

We decided to use the administrative codes for diseases, health conditions and events during a hospital stay. These are known as the international classification of disease codes (ICD-10), and there are 22 chapters grouping ICD-10 codes into distinct disease and health areas. In hospital episode statistics data, patients have many codes logged in their record, including reason for admission and other codes recording underlying conditions, complications, treatments and procedures. We counted the number of distinct chapters the codes fell into. Using this method, we defined a patient with three health conditions as having codes falling into three distinct chapters; whereas a patient with three codes falling into one chapter had one health condition.

The assumption underlying this approach was that patients with health conditions and events in multiple disease areas were more complex than a patient who had a condition recorded in only one disease area. It also allowed us to record acute events and conditions. Complex patients may have to see multiple disease specialists during their stay, and it may be more difficult to manage their condition successfully in the community.

What are the most common conditions?

One of our really striking findings was the big increase in the number of patients with five or more health conditions. Examining these patients more closely, we found that the most common conditions (in 14% of patients) were diseases of the circulatory system, including heart disease. The next most common conditions (12% of patients) were diseases of the endocrine system, such as diabetes, followed by diseases of the respiratory system (9% of patients), including conditions such as asthma and emphysema.

What are the limitations of our methodology?

There are, however, limitations with using the hospital episode statistics dataset to calculate patient complexity. It is primarily used to calculate the cost of care for commissioners and isn’t designed as a research dataset. As a result, it doesn’t record everything we’d like it to. There’s no record of the severity of illness for example, nor any patient assessment of his or her health and wellbeing. There is also a lot of variability in how consistently different diseases are coded over time and in the level of detail recorded on each admission. Finally, analytical coders record patient conditions in the dataset by scouring through patient notes – if the doctor or nurse fails to record a condition it won’t be recorded on the system.

The striking growth in emergency admissions

We wanted to check whether the method we used changed the picture of complexity we saw. Four of the health categories (chapters 19-22) we included in our analysis covered external causes of injury, morbidity and other more general factors that could influence contact with the health service. We included only chapters 1-18 and reran our analysis to check whether it made a difference to our findings. We found that fewer patients were recorded with five or more health conditions. However, we still saw the same sharp increase in admissions for people with multiple conditions. In fact, one in four patients admitted to hospital as an emergency in 2015/16 had five or more health conditions according to this definition, compared with just one in twenty in 2006/07. Looked at in this way, there has been a growth in admissions for these patients of 362% over ten years – more striking even than the analysis we included in the briefing.

However we calculated it, our findings told the same story – that there’s been a massive growth in emergency admissions from patients with multiple health conditions. This has big implications for the number of beds we have in hospitals to provide emergency care, and what support we can provide in the community. The reality is that we may need extra resources in our hospitals, and in our communities to care for our ageing population. Improving the way we measure complexity in the hospital datasets, with more information on the severity of illness, for example, would allow us to better use these resources to design solutions to improve patient care.

Sarah Deeny (@SarahDeeny) is Assistant Director of Data Analytics at the Health Foundation