The National Health Service (NHS) uses the Friends and Family Test (FFT) to gather patient feedback aimed at improving service quality and satisfaction. The current workflow involves FFT feedback processed manually by the Patient Experience team to identify sentiment and theme, which is a time-consuming and unsustainable approach given the large volume of data. To address this, we proposed and implemented a two-phase solution that is adaptable with limited computational resources that has no access to GPUs: (1) training and evaluation of lightweight, fine-tuned transformer-based models using routinely collected, expert-reviewed feedback data, and (2) deployment of these models within an automated pipeline for sentiment prediction and theme classification, involving a human-in-the-loop verification process. Our experiments and results validated on expert-annotated data collected over a period of 3.5 years achieve the best performance with an F1-score of 88.14% in sentiment classification and 71.47% in theme classification. This paper details the implementation, testing, and monitoring of our proposed solution in a paediatric healthcare setting, highlighting the ability to generate near-real-time results. Our deployed solution has substantially decreased the time needed to manually process approximately 2000 feedback pieces each month by delivering automated outcomes daily, a task that previously required several days for completion. In addition, we discuss the broader implications of our findings and potential enhancements for scalable deployment across NHS trusts. Source code will be open-sourced and published here: https://github.com/gosh-dre/FFT-NLP-Pipeline.

UNLABELLED: The high cost of genotyping single nucleotide polymorphisms (SNPs) generally prohibits the systematic mapping of entire genetic linkage regions in order to find the polymorphisms associated with increased risk of disease. In practice, SNPs are selected at approximately equal spacing across the linkage region to try to locate a SNP lying in the haplotype block of the disease SNP. The size of the haplotype block may not be known, however, and SNPs taken from public domain sources may not in fact be polymorphic. Our program will choose a subset of the SNPs in a linkage region so as to maximize the expected proportion of the sequence that lies within a given distance of a real SNP. AVAILABILITY: The software is available, free of charge, for academic use on request from the authors. SUPPLEMENTARY INFORMATION: www.oxagen.co.uk