Paediatric tonsillectomy in England: A cohort study of clinical practice and outcomes using Hospital Episode Statistics data (2008-2019).

OBJECTIVES: To assess the safety of paediatric tonsillectomy procedures conducted in NHS hospitals in England between 2008 and 2019. DESIGN: Retrospective observational cohort study using Hospital Episode Statistics (HES) data. SETTING: Acute NHS trusts in England conducting paediatric tonsillectomy procedures. PARTICIPANTS: Children (≤16 years old) undergoing bilateral tonsillectomy. MAIN OUTCOME MEASURES: Number of tonsillectomies performed per year by procedural method. In-hospital complications including return to theatre for arrest of haemorrhage. Readmission within 28 days, including those for pain, haemorrhage and surgical arrest of haemorrhage. Long-term outcomes: all-cause mortality, revision tonsillectomy. RESULTS: A total of 318 453 paediatric tonsillectomies were performed from 2008 to 2019:278,772 dissection (87.5%) and 39 681 coblation (12.5%). The proportion of tonsillectomy performed using coblation increased from 7% in 2008/9 to 27% in 2018/9. Five patients died in hospital (including 4 due to respiratory complications). In-hospital complications occurred in 4202 children (1.3%), with the most frequent being haemorrhage. Within 28 days of tonsillectomy, 28 170 patients (8.8%) were readmitted and 7 deaths occurred. Readmission rates for haemorrhage and pain have increased since 2008. The proportion of children undergoing revision tonsillectomy procedures within 5 years following coblation tonsillectomy (1.4%) was approximately double that of dissection (0.6%). CONCLUSIONS: Clinical practice of paediatric tonsillectomy has changed in England over the past 11 years. The overall mortality rate associated with the procedure is 0.0037%. Differences in outcomes have been identified for different procedural methods. However, routine administrative data are limited in differentiating procedural detail (eg we are unable to differentiate intra or extra-capsular techniques from current clinical coding of tonsillectomy procedures). Therefore, prospective national data collection or more granular clinical coding is essential to capture relative outcomes of the different tonsillectomy methods and techniques being used in the NHS.

Comparison of identifiable and non-identifiable data linkage: health technology assessment of MitraClip using registry, administrative and mortality datasets.

OBJECTIVES: The UK MitraClip registry was commissioned by National Health Service (NHS) England to assess real-world outcomes from percutaneous mitral valve repair for mitral regurgitation using a new technology, MitraClip. This study aimed to determine longitudinal patient outcomes by linking to routine datasets: Hospital Episode Statistics (HES) Admitted Patient Care (APC) and Office of National Statistics. METHODS: Two methods of linkage were compared, using identifiable (NHS number, date of birth, postcode, gender) and non-identifiable data (hospital trust, age in years, admission, discharge and operation dates, operation and diagnosis codes). Outcome measures included: matching success, patient demographics, all-cause mortality and subsequent cardiac intervention. RESULTS: A total of 197 registry patients were eligible for matching with routine administrative data. Using identifiable linkage, a total of 187 patients (94.9%) were matched with the HES APC dataset. However, 21 matched individuals (11.2%) had inconsistencies across the datasets (eg, different gender) and were subsequently removed, leaving 166 (84.3%) for analysis. Using non-identifiable data linkage, a total of 170 patients (86.3%) were uniquely matched with the HES APC dataset.Baseline patient characteristics were not significantly different between the two methods of data linkage. The total number of deaths (all causes) identified from identifiable and non-identifiable linkage methods was 37 and 40, respectively, and the difference in subsequent cardiac interventions identified between the two methods was negligible. CONCLUSIONS: Patients from a bespoke clinical procedural registry were matched to routine administrative data using identifiable and non-identifiable methods with equivalent matching success rates, similar baseline characteristics and similar 2-year outcomes.

A universal SNP and small-indel variant caller using deep neural networks.

Despite rapid advances in sequencing technologies, accurately calling genetic variants present in an individual genome from billions of short, errorful sequence reads remains challenging. Here we show that a deep convolutional neural network can call genetic variation in aligned next-generation sequencing read data by learning statistical relationships between images of read pileups around putative variant and true genotype calls. The approach, called DeepVariant, outperforms existing state-of-the-art tools. The learned model generalizes across genome builds and mammalian species, allowing nonhuman sequencing projects to benefit from the wealth of human ground-truth data. We further show that DeepVariant can learn to call variants in a variety of sequencing technologies and experimental designs, including deep whole genomes from 10X Genomics and Ion Ampliseq exomes, highlighting the benefits of using more automated and generalizable techniques for variant calling.