home email us! sindicaci;ón

Jennifer Dixon on predicting risk of emergency admission

by Lloyd Davis

Jennifer is Director, Health Policy at the King’s Fund in the UK and is a board member of both the Audit Commission and the Healthcare Commission

Theme: risk segmentation and identifying populations at risk

We want to reduce preventable admissions, particularly among older people. There’s been a lot of guidance from NICE and elswhere and there have been funds for interventions. Meanwhile commissioners have been mandated to employ community matrons to carry out case management (though we’re not sure on what evidence this decision was made!)

So what we saw a real need for was a way to identify high risk, high cost patients and intervene before they got to a point of crisis.

We set up a project to find ways of predicting future high-risk patients. We did a literature review, confirming that the clinician’s hunch is probably not the most robust method of prediction, the best ways are statistical.

So we developed a first predictive model using inpatient data which we then developed using multiple sources.

Our model known as PARR uses HES and census data. We took two approaches: PARR1 focusing on emergency admissions for avoidable conditions (CHF, COPD, Diabetes, CAD, Sickle cell etc) that often lead to re-hospitalisation and then a second approach, PARR2 to broaden that to prevent emergency admissions.

We used statistical techniques to come up with a risk score from 0 to 100 (where 100 is certainty of admission). We took 5 years of data looking at year 4 we looked back at the previous 3 years and then correlated that data with actual admission in year 5.

No model works with 100% certainty but we were pleased to see that the proportion of false positives reduces at higher risk levels.

Not surprisingly we found that the high risk group contained a high preponderance of people over 75 and people from ethnic groups and a high proportion of people with chronic illness.

We’ve built an interactive tool now that you can use it in real-time so we’re suggesting to commissioners that they identify high-risk patients and interview them to find out why they have been admitted and then working with patients and their families to design a better intervention. We’re also encouraging PCTs to build evaluation methods into their adoption of the tool. The tool is freely available and downloadable from the internet. At least half of English PCTs are now using it

The final phase is to try to improve the model by adding in other data about individuals - eg A&E GP, Pharmacy and Outpatient activities

We looked at half a million patients from two PCTs, we split the sample to test, using logistic regression to model the predictive factors using 850 variables. With this one we just looked at the previous 2 years data.

This approach does identifiy new patients who had had no prior admission, so adding more data improves the prediction. It’s better but we have to consider whether it’s worth the cost of linking the data.

The highest risk patients have 20 times the average admissions and 24 times the average of emergency bed days so if you target your interventions lower down the pyramid you can make more of an impact than with the highest risk. We recommend that you don’t just focus on high risk patients.

The model also allows you to include your costs, so that you can predict savings and look for the most cost-effective risk management strategies.

PARR is now used across the country and the combined model is in a smaller number of PCTs because it’s so data intensive. The interventions are currently being evaluated. PARR is of growing usefulness for commissioners and regulators and we’re now looking at risk-adjusted person-based resource allocation which until now in the UK we’ve not been able to do.

Read more about this project on the King’s Fund website.

Here’s Jennifer talking to me between walking off stage and dashing to the airport, putting the project in a nutshell: http://www.worldhealthcareblog.org/lloydd/whcc-dixon.wmv

Watch Now:
.. ..
icon for podpress  Jennifer Dixon on the Kings Fund Risk Prediction Model [2:00m]: Play Now | Play in Popup | Download (116)

No comments yet »

Your comment

<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>