Listen, Learn, Improve: Using advanced data analytics to embed patient experience feedback as a

near real-time quality improvement initiative

Clinical Effectiveness Patient Safety

Patient Experience

Quality Trifecta

• Fixated on scores• Focus on response rates• Free text-comments not

analysed

Staff appreciate feedback in the patients’ own words as this makes the comments seem more ‘real’

Frontline staff do not have access to this data, only used by at divisional/directorate level (given little priority)

High volume of free-text comments – lack of time

Need innovative method to analyse free-text data

• Period: January – July 2017

• Inpatient, Outpatient, A&E, Maternity

– “What did we do well? & “What could we do better?”

• Total – 65,973 (x2)

Hopper et al, 895

Greaves et al, 6,412

Maramba et al, 3,426, avg. 43.98 words

Doing-Harris et al, 51,234

5%

Manual Classification

(PExF Domain and Sentiment) MK

Manual Classification

SHW

Manual Classification

DM

Total comments January -July (IP, OP, A&E, Mat)

>60,000 10%

Train different learners

SVM (Support Vector Machine)

Blind classification

(stratified sample 500)

Inpatient

Outpatient

Prediction (PExF Domain

and Sentiment)

A&E

Maternity

• Encouraging early dialogue in Transition and Continuity in Inpatients

• Facilitating better Access to Care in Accident and Emergency

• Improving Physical Comfort in Outpatients

Early Outcomes

Staff Perspectives

• Technical efficiencies through reduction in labour costs and relocating staff time supporting QI projects

• Time spent on manual processing 6,000 responses was four days compared to 15 minutes using the NLP/ML algorithm, p<0.001

• Weak correlation 0.45 with average sentiment and FFT score

Applying analytics and QI in parallel