Making sense of complaints – using AI to improve learning from patient complaints

Complaints are a “Big Problem”

Complaints to the NHS by patients or families are incredibly common, with the NHS receiving over 175,000 complaints about its services every year - this translates into over 3,300 per week. Our partner NHS trust (University Hospitals of Leicester NHS Trust) alone receives 4,500 complaints or concerns per year. Allied to this, the cost of litigation is steadily increasing, being over £2 billion to the wider NHS and £22 million to UHL alone, annually.

This high volume causes significant problems with the way in which complaints are managed and, most importantly, learnt from by hospitals and GP practices. Manually analysing and processing these complaints is incredibly labour-intensive and takes considerable time. As complaints can be very long and emotive, analysing them, responding to them and learning from them objectively, can be difficult.

As several high-profile reports to the British Government have found, this means:

• already distressed patients and families frequently suffer delays
• many complaint responses do not address all the key issues or questions that people have raised, causing further distress, cost to hospitals and staff time
• Perhaps most worryingly, hospitals cannot maximise the learning from past experiences.
• In turn, this means mistakes are repeated and patient care does not improve.

The solution – how can AI help?

We’re using a combination of automation and a branch of AI called Natural Language Processing. Although it’s still in development, it is based on previous work which we have successfully deployed for organisations such as the Care Quality Commission and Ministry of Defence.

Natural language processing (NLP) is a branch of artificial intelligence that allows computers to interpret human language by “reading” and then analysing large blocks of unstructured text. In brief, the system first pre-processes the data into a format in which it can be analysed, then classifies it and identifies the underlying issues in the complaint (“topic modelling”).

The aim is by using the AI system, we will address the issues discussed above, specifically by:-

1) Providing patients and relatives with:

a. a faster, more complete, and more accurate response to the issues raised in their complaints.

b. Reassurance that all the issues they raise are being identified and patterns are being viewed by senior leaders.

2) Helping hospitals

a. Accurately identify key problem areas in near-real time.

b. Improve the efficiency of complaint management resulting in:

i. improved experience for patients and relatives

ii. reduced burden on staff (identified by the Ombudsman as a key issue)

iii. reduced time clinical staff (e.g. consultants and ward sisters) spend dealing with complaints, thereby allowing them to focus on clinical care.

c. Focus quality improvement efforts where they are needed most, thereby improve patient outcomes.

d. Potentially reduce both the number of complaints and litigation costs, by improving learning from complaints and by improving patient satisfaction with the complaints process (as evidenced in the Ombudsman report).

Keynote

Key to note is that although it makes the human’s task much faster, easier and more objective, it does not remove the human oversight of what is obviously a sensitive area.

“AI in healthcare” usually conjures up images of robots on the wards or using computers to read CT scans. However, healthcare is so much more complex than this, and many of the ways in which AI will improve care and improve efficiency are in “back-room” tasks. Although less “glamourous”, they are no less important for both the patient and the hospital.

Author:

Dr. Tom Palser – Consultant Surgeon and clinical lead at Methods Analytics