Personalised medicine in the UK

Smart laser Sintering: Deep Learning-Powered powder bed fusion 3D printing in precision medicine

Medicines remain ineffective for over 50% of patients due to conventional mass production methods with fixed drug dosages. Three-dimensional (3D) printing, specifically selective laser sintering (SLS), offers a potential solution to this challenge, allowing the manufacturing of small, personalized batches of medication. Despite its simplicity and suitability for upscaling to large-scale production, SLS was not designed for pharmaceutical manufacturing and necessitates a time-consuming, trial-and-error adaptation process. In response, this study introduces a deep learning model trained on a variety of features to identify the best feature set to represent drugs and polymeric materials for the prediction of the printability of drug-loaded formulations using SLS. The proposed model demonstrates success by achieving 90% accuracy in predicting printability. Furthermore, explainability analysis unveils materials that facilitate SLS printability, offering invaluable insights for scientists to optimize SLS formulations, which can be expanded to other disciplines. This represents the first study in the field to develop an interpretable, uncertainty-optimized deep learning model for predicting the printability of drug-loaded formulations. This paves the way for accelerating formulation development, propelling us into a future of personalized medicine with unprecedented manufacturing precision.

Extended optical treatment versus early patching with an intensive patching regimen in children with amblyopia in Europe (EuPatch): a multicentre, randomised controlled trial

BACKGROUND

Amblyopia, the most common visual impairment of childhood, is a public health concern. An extended period of optical treatment before patching is recommended by the clinical guidelines of several countries. The aim of this study was to compare an intensive patching regimen, with and without extended optical treatment (EOT), in a randomised controlled trial.

METHODS

EuPatch was a randomised controlled trial conducted in 30 hospitals in the UK, Greece, Austria, Germany, and Switzerland. Children aged 3-8 years with newly detected, untreated amblyopia (defined as an interocular difference ≥0·30 logarithm of the minimum angle of resolution [logMAR] best corrected visual acuity [BCVA]) due to anisometropia, strabismus, or both were eligible. Participants were randomly assigned (1:1) via a computer-generated sequence to either the EOT group (18 weeks of glasses use before patching) or to the early patching group (3 weeks of glasses use before patching), stratified for type and severity of amblyopia. All participants were initially prescribed an intensive patching regimen (10 h/day, 6 days per week), supplemented with motivational materials. The patching period was up to 24 weeks. Participants, parents or guardians, assessors, and the trial statistician were not masked to treatment allocation. The primary outcome was successful treatment (ie, ≤0·20 logMAR interocular difference in BCVA) after 12 weeks of patching. Two primary analyses were conducted: the main analysis included all participants, including those who dropped out, but excluded those who did not provide outcome data at week 12 and remained on the study; the other analysis imputed this missing data. All eligible and randomly assigned participants were assessed for adverse events. This study is registered with the International Standard Randomised Controlled Trial Number registry (ISRCTN51712593) and is no longer recruiting.

FINDINGS

Between June 20, 2013, and March 12, 2020, after exclusion of eight participants found ineligible after detailed screening, we randomly assigned 334 participants (170 to the EOT group and 164 to the early patching group), including 188 (56%) boys, 146 (44%) girls, and two (1%) participants whose sex was not recorded. 317 participants (158 in the EOT group and 159 in the early patching group) were analysed for the primary outcome without imputation of missing data (median follow-up time 42 weeks [IQR 42] in the EOT group vs 27 weeks [27] in the early patching group). 24 (14%) of 170 participants in the EOT group and ten (6%) of 164 in the early patching group were excluded or dropped out of the study, mostly due to loss to follow-up and withdrawal of consent; ten (6%) in the EOT group and three (2%) in the early patching group missed the 12 week visit but remained on the study. A higher proportion of participants in the early patching group had successful treatment (107 [67%] of 159) than those in the EOT group (86 [54%] of 158; 13% difference; p=0·019) after 12 weeks of patching. No serious adverse events related to the interventions occurred.

INTERPRETATION

The results from this trial indicate that early patching is more effective than EOT for the treatment of most children with amblyopia. Our findings also provide data for the personalisation of amblyopia treatments.

FUNDING

Action Medical Research, NIHR Clinical Research Network, and Ulverscroft Foundation.

Trust and Trade-Offs in Sharing Data for Precision Medicine: A National Survey of Singapore

BACKGROUND

Precision medicine (PM) programs typically use broad consent. This approach requires maintenance of the social license and public trust. The ultimate success of PM programs will thus likely be contingent upon understanding public expectations about data sharing and establishing appropriate governance structures. There is a lack of data on public attitudes towards PM in Asia.

METHODS

The aim of the research was to measure the priorities and preferences of Singaporeans for sharing health-related data for PM. We used adaptive choice-based conjoint analysis (ACBC) with four attributes: uses, users, data sensitivity and consent. We recruited a representative sample of n = 1000 respondents for an in-person household survey.

RESULTS

Of the 1000 respondents, 52% were female and majority were in the age range of 40-59 years (40%), followed by 21-39 years (33%) and 60 years and above (27%). A total of 64% were generally willing to share de-identified health data for IRB-approved research without re-consent for each study. Government agencies and public institutions were the most trusted users of data. The importance of the four attributes on respondents' willingness to share data were: users (39.5%), uses (28.5%), data sensitivity (19.5%), consent (12.6%). Most respondents found it acceptable for government agencies and hospitals to use de-identified data for health research with broad consent. Our sample was consistent with official government data on the target population with 52% being female and majority in the age range of 40-59 years (40%), followed by 21-39 years (33%) and 60 years and above (27%).

CONCLUSIONS

While a significant body of prior research focuses on preferences for consent, our conjoint analysis found consent was the least important attribute for sharing data. Our findings suggest the social license for PM data sharing in Singapore currently supports linking health and genomic data, sharing with public institutions for health research and quality improvement; but does not support sharing with private health insurers or for private commercial use.

Generating high-fidelity synthetic patient data for assessing machine learning healthcare software

There is a growing demand for the uptake of modern artificial intelligence technologies within healthcare systems. Many of these technologies exploit historical patient health data to build powerful predictive models that can be used to improve diagnosis and understanding of disease. However, there are many issues concerning patient privacy that need to be accounted for in order to enable this data to be better harnessed by all sectors. One approach that could offer a method of circumventing privacy issues is the creation of realistic synthetic data sets that capture as many of the complexities of the original data set (distributions, non-linear relationships, and noise) but that does not actually include any real patient data. While previous research has explored models for generating synthetic data sets, here we explore the integration of resampling, probabilistic graphical modelling, latent variable identification, and outlier analysis for producing realistic synthetic data based on UK primary care patient data. In particular, we focus on handling missingness, complex interactions between variables, and the resulting sensitivity analysis statistics from machine learning classifiers, while quantifying the risks of patient re-identification from synthetic datapoints. We show that, through our approach of integrating outlier analysis with graphical modelling and resampling, we can achieve synthetic data sets that are not significantly different from original ground truth data in terms of feature distributions, feature dependencies, and sensitivity analysis statistics when inferring machine learning classifiers. What is more, the risk of generating synthetic data that is identical or very similar to real patients is shown to be low.

Has the introduction of direct oral anticoagulants (DOACs) in England increased emergency admissions for bleeding conditions? A longitudinal ecological study

OBJECTIVE

There is concern about long-term safety of direct oral coagulants (DOACs) in clinical practice. Our aim was to investigate whether the introduction of DOACs compared with vitamin-K antagonists in England was associated with a change in admissions for bleeding or thromboembolic complications.

SETTING

5508 General practitioner (GP) practices in England between 2011 and 2016.

PARTICIPANTS

All GP practices in England with a registered population size of greater than 1000 that had data for all 6 years.

MAIN OUTCOME MEASURE

The rate of emergency admissions to hospital for bleeding or thromboembolism, per 100 000 population for each GP practice in England.

MAIN EXPOSURE MEASURE

The annual number of DOAC items prescribed for each GP practice population as a proportion of all anticoagulant items prescribed.

DESIGN

This longitudinal ecological study used panel regression models to investigate the association between trends in DOAC prescribing within GP practice populations and trends in emergency admission rates for bleeding and thromboembolic conditions, while controlling for confounders.

RESULTS

For each additional 10% of DOACs prescribed as a proportion of all anticoagulants, there was a 0.9% increase in bleeding complications (rate ratio 1.008 95% CI 1.003 to 1.013). The introduction of DOACs between 2011 and 2016 was associated with additional 4929 (95% CI 2489 to 7370) emergency admissions for bleeding complications. Increased DOAC prescribing was associated with a slight decline in admission for thromboembolic conditions.

CONCLUSION

Our data show that the rapid increase in prescribing of DOACs after changes in National Institute for Health and Care Excellence guidelines in 2014 may have been associated with a higher rate of emergency admissions for bleeding conditions. These consequences need to be considered in assessing the benefits and costs of the widespread use of DOACs.

Personalising medicine in inflammatory bowel disease-current and future perspectives

Up to 25% of inflammatory bowel disease (IBD) presents during childhood, often with severe and extensive disease, leading to significant morbidity including delayed growth and nutritional impairment. The classical approach to management has centred on differentiation into Crohn's disease (CD) or ulcerative colitis (UC), with subsequent treatment based on symptoms, results and complications. However, IBD is a heterogeneous condition with substantial variation in phenotype, disease course and outcome, so whilst effective treatment exists one size does not fit all. The ability to predict disease course at diagnosis, alongside tailoring medications based on response gives the potential for a more 'personalised approach'. The move to a pre-emptive strategy to prevent IBD-related complications, whilst simultaneously minimising side effects and long-term toxicity from therapy, particularly in those with relatively indolent disease, has the potential to revolutionise care. In very early-onset IBD, personalised approaches to diagnosis and management have become the standard of treatment enabling clinicians to significantly alter the outcomes of the few children with monogenic disease. However, the promise of discoveries in genomics, microbiome and transcriptomics in paediatric IBD has not yet translated to clinical application for the vast majority of patients. Despite this, the opportunity presents itself to apply data gathered at diagnosis and follow-up to predict which patients are likely to progress to complicated disease, which will respond well and which will require additional therapy. Using complex mathematics and innovative, cutting-edge machine learning (ML) techniques gives the potential to use this data to develop personalised clinical care algorithms to treat patients more effectively, reduce toxicity and improve outcome. In this review, we will consider current management of paediatric IBD, discuss how precision medicine is making inroads into clinical practice already, examine the contemporary studies applying data to stratify patients and explore how future management may be revolutionised by personalisation with clinical, genomic and other multi-omic data.