Cluster detection with random neighbourhood covering: Application to invasive Group A Streptococcal disease.

The rapid detection of outbreaks is a key step in the effective control and containment of infectious diseases. In particular, the identification of cases which might be epidemiologically linked is crucial in directing outbreak-containment efforts and shaping the intervention of public health authorities. Often this requires the detection of clusters of cases whose numbers exceed those expected by a background of sporadic cases. Quantifying exceedances rapidly is particularly challenging when only few cases are typically reported in a precise location and time. To address such important public health concerns, we present a general method which can detect spatio-temporal deviations from a Poisson point process and estimate the odds of an isolate being part of a cluster. This method can be applied to diseases where detailed geographical information is available. In addition, we propose an approach to explicitly take account of delays in microbial typing. As a case study, we considered invasive group A Streptococcus infection events as recorded and typed by Public Health England from 2015 to 2020.

Forgotten but not gone: Yersinia infections in England, 1975 to 2020.

BackgroundYersiniosis is one of the most common food-borne zoonoses in Europe, but there are large variations in the reported incidence between different countries.AimWe aimed to describe the trends and epidemiology of laboratory-confirmed Yersinia infections in England and estimate the average annual number of undiagnosed Yersinia enterocolitica cases, accounting for under-ascertainment.MethodsWe analysed national surveillance data on Yersinia cases reported by laboratories in England between 1975 and 2020 and enhanced surveillance questionnaires from patients diagnosed in a laboratory that has implemented routine Yersinia testing of diarrhoeic samples since 2016.ResultsThe highest incidence of Yersinia infections in England (1.4 cases per 100,000 population) was recorded in 1988 and 1989, with Y. enterocolitica being the predominant species. The reported incidence of Yersinia infections declined during the 1990s and remained low until 2016. Following introduction of commercial PCR at a single laboratory in the South East, the annual incidence increased markedly (13.6 cases per 100,000 population in the catchment area between 2017 and 2020). There were notable changes in age and seasonal distribution of cases over time. The majority of infections were not linked to foreign travel and one in five patients was admitted to hospital. We estimate that around 7,500 Y. enterocolitica infections may be undiagnosed in England annually.ConclusionsFindings suggest a considerable number of undiagnosed yersiniosis cases in England, with possibly important changes in the epidemiology. The apparently low incidence of yersiniosis in England is probably due to limited laboratory testing.

Contrasting factors associated with COVID-19-related ICU admission and death outcomes in hospitalised patients by means of Shapley values.

Identification of those at greatest risk of death due to the substantial threat of COVID-19 can benefit from novel approaches to epidemiology that leverage large datasets and complex machine-learning models, provide data-driven intelligence, and guide decisions such as intensive-care unit admission (ICUA). The objective of this study is two-fold, one substantive and one methodological: substantively to evaluate the association of demographic and health records with two related, yet different, outcomes of severe COVID-19 (viz., death and ICUA); methodologically to compare interpretations based on logistic regression and on gradient-boosted decision tree (GBDT) predictions interpreted by means of the Shapley impacts of covariates. Very different association of some factors, e.g., obesity and chronic respiratory diseases, with death and ICUA may guide review of practice. Shapley explanation of GBDTs identified varying effects of some factors among patients, thus emphasising the importance of individual patient assessment. The results of this study are also relevant for the evaluation of complex automated clinical decision systems, which should optimise prediction scores whilst remaining interpretable to clinicians and mitigating potential biases.

Pneumococcal Disease: A Systematic Review of Health Utilities, Resource Use, Costs, and Economic Evaluations of Interventions.

BACKGROUND: Pneumococcal diseases cause substantial mortality, morbidity, and economic burden. Evidence on data inputs for economic evaluations of interventions targeting pneumococcal disease is critical. OBJECTIVES: To summarize evidence on resource use, costs, health utilities, and cost-effectiveness for pneumococcal disease and associated interventions to inform future economic analyses. METHODS: We searched MEDLINE, Embase, Web of Science, CINAHL, PsycINFO, EconLit, and Cochrane databases for peer-reviewed studies in English on pneumococcal disease that reported health utilities using direct or indirect valuation methods, resource use, costs, or cost-effectiveness of intervention programs, and summarized the evidence descriptively. RESULTS: We included 383 studies: 9 reporting health utilities, 131 resource use, 160 economic costs of pneumococcal disease, 95 both resource use and costs, and 178 economic evaluations of pneumococcal intervention programs. Health state utility values ranged from 0 to 1 for both meningitis and otitis media and from 0.3 to 0.7 for both pneumonia and sepsis. Hospitalization was shortest for otitis media (range: 0.1-5 days) and longest for sepsis/septicemia (6-48). The main categories of costs reported were drugs, hospitalization, and household or employer costs. Resource use was reported in hospital length of stay and number of contacts with general practitioners. Costs and resource use significantly varied among population ages, disease conditions, and settings. Current vaccination programs for both adults and children, antibiotic use and outreach programs to promote vaccination, early disease detection, and educational programs are cost-effective in most countries. CONCLUSION: This study has generated a comprehensive repository of health economic evidence on pneumococcal disease that can be used to inform future economic evaluations of pneumococcal disease intervention programs.

The impact of childhood pneumococcal vaccination on hospital admissions in England: a whole population observational study.

BACKGROUND: Pneumococcal infections are major causes of morbidity and mortality worldwide. We use routine hospital admissions data and time-series modelling analysis to estimate the impact of the seven and thirteen valent pneumococcal conjugate vaccines (PCV7 and PCV13) on hospital admissions due to pneumococcal disease in England. METHODS: Hospital admissions for pneumococcal meningitis, bacteraemia and pneumonia between January 1, 2003 and December 31, 2015 were identified from the national Hospital Episode Statistics database for all age groups in England. We model the impact of pneumococcal vaccination using interrupted time series analysis. Hospital admissions prior to vaccine introduction were extrapolated to predict the expected number of admissions in the absence of pneumococcal vaccines. Admissions avoided over time were estimated by comparing the fitted interrupted time series and the expected model for no vaccination in a Bayesian framework. RESULTS: Overall, there were 43,531 (95% credible interval (CrI): 36486-51,346) fewer hospital admissions due to bacteraemia, meningitis and pneumonia in England during the period from 2006 to 2015 than would have been expected if pneumococcal vaccines had not been implemented, with the majority of hospital admissions avoided due to pneumonia. Among young children reductions in meningitis were more common, while among adults reductions in pneumonia admissions were relatively more important, with no evidence for reduced bacteraemia and meningitis among older adults. We estimated that 981 (95% CrI: 391-2018), 749 (95% CrI: 295-1442) and 1464 (95% CrI: 793-2522) bacteraemia, meningitis and pneumonia related hospital admissions, respectively, were averted in children < 2 years of age. CONCLUSIONS: Substantial reductions in hospital admissions for bacteraemia, meningitis and pneumonia in England were estimated after the introduction of childhood vaccination, with indirect effects being responsible for most of the hospital admissions avoided.

A systematic review of source attribution of human campylobacteriosis using multilocus sequence typing.

BackgroundCampylobacter is a leading global cause of bacterial gastroenteritis, motivating research to identify sources of human infection. Population genetic studies have been increasingly applied to this end, mainly using multilocus sequence typing (MLST) data.ObjectivesThis review aimed to summarise approaches and findings of these studies and identify best practice lessons for this form of genomic epidemiology.MethodsWe systematically reviewed publications using MLST data to attribute human disease isolates to source. Publications were from January 2001, when this type of approach began. Searched databases included Scopus, Web of Science and PubMed. Information on samples and isolate datasets used, as well as MLST schemes and attribution algorithms employed, was obtained. Main findings were extracted, as well as any results' validation with subsequent correction for identified biases. Meta-analysis is not reported given high levels of heterogeneity.ResultsOf 2,109 studies retrieved worldwide, 25 were included, and poultry, specifically chickens, were identified as principal source of human infection. Ruminants (cattle or sheep) were consistently implicated in a substantial proportion of cases. Data sampling and analytical approaches varied, with five different attribution algorithms used. Validation such as self-attribution of isolates from known sources was reported in five publications. No publication reported adjustment for biases identified by validation.ConclusionsCommon gaps in validation and adjustment highlight opportunities to generate improved estimates in future genomic attribution studies. The consistency of chicken as the main source of human infection, across high income countries, and despite methodological variations, highlights the public health importance of this source.

Comparative efficacy of drugs for treating giardiasis: a systematic update of the literature and network meta-analysis of randomized clinical trials.

Background: Giardiasis is the commonest intestinal protozoal infection worldwide. The current first-choice therapy is metronidazole. Recently, other drugs with potentially higher efficacy or with fewer and milder side effects have increased in popularity, but evidence is limited by a scarcity of randomized controlled trials (RCTs) comparing the many treatment options available. Network meta-analysis (NMA) is a useful tool to compare multiple treatments when there is limited or no direct evidence available. Objectives: To compare the efficacy and side effects of all available drugs for the treatment of giardiasis. Methods: We selected all RCTs included in systematic reviews and expert reviews of all treatments for giardiasis published until 2014, extended the systematic literature search until 2016, and identified new studies by scanning reference lists for relevant studies. We then conducted an NMA of all available treatments for giardiasis by comparing parasitological cure (efficacy) and side effects. Results: We identified 60 RCTs from 58 reports (46 from published systematic reviews, 8 from reference lists and 4 from the updated systematic search). Data from 6714 patients, 18 treatments and 42 treatment comparisons were available. Tinidazole was associated with higher parasitological cure than metronidazole [relative risk (RR) 1.23, 95% CI 1.12-1.35] and albendazole (RR 1.35, 95% CI 1.21-1.50). Taking into consideration clinical efficacy, side effects and amount of the evidence, tinidazole was found to be the most effective drug. Conclusions: We provide additional evidence that single-dose tinidazole is the best available treatment for giardiasis in symptomatic and asymptomatic children and adults.

Core Genome Multilocus Sequence Typing Scheme for Stable, Comparative Analyses of Campylobacter jejuni and C. coli Human Disease Isolates.

Human campylobacteriosis, caused by Campylobacter jejuni and C. coli, remains a leading cause of bacterial gastroenteritis in many countries, but the epidemiology of campylobacteriosis outbreaks remains poorly defined, largely due to limitations in the resolution and comparability of isolate characterization methods. Whole-genome sequencing (WGS) data enable the improvement of sequence-based typing approaches, such as multilocus sequence typing (MLST), by substantially increasing the number of loci examined. A core genome MLST (cgMLST) scheme defines a comprehensive set of those loci present in most members of a bacterial group, balancing very high resolution with comparability across the diversity of the group. Here we propose a set of 1,343 loci as a human campylobacteriosis cgMLST scheme (v1.0), the allelic profiles of which can be assigned to core genome sequence types. The 1,343 loci chosen were a subset of the 1,643 loci identified in the reannotation of the genome sequence of C. jejuni isolate NCTC 11168, chosen as being present in >95% of draft genomes of 2,472 representative United Kingdom campylobacteriosis isolates, comprising 2,207 (89.3%) C. jejuni isolates and 265 (10.7%) C. coli isolates. Validation of the cgMLST scheme was undertaken with 1,478 further high-quality draft genomes, containing 150 or fewer contiguous sequences, from disease isolate collections: 99.5% of these isolates contained ≥95% of the 1,343 cgMLST loci. In addition to the rapid and effective high-resolution analysis of large numbers of diverse isolates, the cgMLST scheme enabled the efficient identification of very closely related isolates from a well-defined single-source campylobacteriosis outbreak.

Local genes for local bacteria: Evidence of allopatry in the genomes of transatlantic Campylobacter populations.

The genetic structure of bacterial populations can be related to geographical locations of isolation. In some species, there is a strong correlation between geographical distance and genetic distance, which can be caused by different evolutionary mechanisms. Patterns of ancient admixture in Helicobacter pylori can be reconstructed in concordance with past human migration, whereas in Mycobacterium tuberculosis it is the lack of recombination that causes allopatric clusters. In Campylobacter, analyses of genomic data and molecular typing have been successful in determining the reservoir host species, but not geographical origin. We investigated biogeographical variation in highly recombining genes to determine the extent of clustering between genomes from geographically distinct Campylobacter populations. Whole-genome sequences from 294 Campylobacter isolates from North America and the UK were analysed. Isolates from within the same country shared more recently recombined DNA than isolates from different countries. Using 15 UK/American closely matched pairs of isolates that shared ancestors, we identify regions that have frequently and recently recombined to test their correlation with geographical origin. The seven genes that demonstrated the greatest clustering by geography were used in an attribution model to infer geographical origin which was tested using a further 383 UK clinical isolates to detect signatures of recent foreign travel. Patient records indicated that in 46 cases, travel abroad had occurred <2 weeks prior to sampling, and genomic analysis identified that 34 (74%) of these isolates were of a non-UK origin. Identification of biogeographical markers in Campylobacter genomes will contribute to improved source attribution of clinical Campylobacter infection and inform intervention strategies to reduce campylobacteriosis.

Integration of Genomic and Other Epidemiologic Data to Investigate and Control a Cross-Institutional Outbreak of Streptococcus pyogenes.

Single-strain outbreaks of Streptococcus pyogenes infections are common and often go undetected. In 2013, two clusters of invasive group A Streptococcus (iGAS) infection were identified in independent but closely located care homes in Oxfordshire, United Kingdom. Investigation included visits to each home, chart review, staff survey, microbiologic sampling, and genome sequencing. S. pyogenes emm type 1.0, the most common circulating type nationally, was identified from all cases yielding GAS isolates. A tailored whole-genome reference population comprising epidemiologically relevant contemporaneous isolates and published isolates was assembled. Data were analyzed independently using whole-genome multilocus sequencing and single-nucleotide polymorphism analyses. Six isolates from staff and residents of the homes formed a single cluster that was separated from the reference population by both analytical approaches. No further cases occurred after mass chemoprophylaxis and enhanced infection control. Our findings demonstrate the ability of 2 independent analytical approaches to enable robust conclusions from nonstandardized whole-genome analysis to support public health practice.

A quantitative review of healthcare professionals' questions to a local immunization advice service: 4299 enquiries from 3 years.

BACKGROUND: Immunization advice services can support health professionals by providing rapid access to accurate and reliable current information and advice. The Vaccine Advice for Clinicians Service (VACCSline) is a service for health professionals working within the Thames Valley Area of the UK. METHODS: We reviewed all 4299 enquiries received by VACCSline over 3 years. Queries were summarized by vaccine type and topic of enquiry. Associations with profession and workplace of the enquirer were tested using Fisher's exact tests. RESULTS: Incomplete immunization status and non-UK schedules were the most common topics of enquiry. Practice nurses were the main service users followed by doctors. Enquiries varied by professional role. Alterations to the immunization programme led to temporary changes to enquiry content and some more persistent adjustments in the balance of enquiries were identified, such as an increase in enquiries relating to vaccination in pregnancy. CONCLUSIONS: The content of enquiries to VACCSline is broad, confirming the need for immunizers to have a wide knowledge base and access to specialist advice to assist with complex scenarios. Systematic data capture provided intelligence to guide training and materials to support immunizers. A wider networked application of this approach could improve support for immunizers.

Partial Failure of Milk Pasteurization as a Risk for the Transmission of Campylobacter From Cattle to Humans.

BACKGROUND: Cattle are the second most common source of human campylobacteriosis. However, routes to account for this scale of transmission have not been identified. In contrast to chicken, red meat is not heavily contaminated at point of sale. Although effective pasteurization prevents milk-borne infection, apparently sporadic infections may include undetected outbreaks from raw or perhaps incompletely pasteurized milk. METHODS: A rise in Campylobacter gastroenteritis in an isolated population was investigated using whole-genome sequencing (WGS), an epidemiological study, and environmental investigations. RESULTS: A single strain was identified in 20 cases, clearly distinguishable from other local strains and a reference population by WGS. A case-case analysis showed association of infection with the outbreak strain and milk from a single dairy (odds ratio, 8; Fisher exact test P value = .023). Despite temperature records indicating effective pasteurization, mechanical faults likely to lead to incomplete pasteurization of part of the milk were identified by further testing and examination of internal components of dairy equipment. CONCLUSIONS: Here, milk distribution concentrated on a small area, including school-aged children with low background incidence of campylobacteriosis, facilitated outbreak identification. Low-level contamination of widely distributed milk would not produce as detectable an outbreak signal. Such hidden outbreaks may contribute to the substantial burden of apparently sporadic Campylobacter from cattle where transmission routes are not certain. The effective discrimination of outbreak isolates from a reference population using WGS shows that integrating these data and approaches into surveillance could support the detection as well as investigation of such outbreaks.

The long-term dynamics of Campylobacter colonizing a free-range broiler breeder flock: an observational study.

A free-range broiler breeder flock was studied in order to determine the natural patterns of Campylobacter colonization over a period of 63 weeks. Campylobacter sequence types (STs) were not mutually exclusive and on average colonized only 17.7% of the birds tested at any time. Campylobacter STs typically reached a peak in prevalence upon initial detection in the flock before tailing off, although the ST and antigenic flaA short variable region in combination were stable over a number of months. There was evidence that, with a couple of exceptions, the ecology of C. jejuni and C. coli differed, with the latter forming a more stable population. Despite being free range, no newly colonizing STs were detected over a 6-week period in autumn and a 10-week period in winter, towards the end of the study. There was limited evidence that those STs identified among broiler chicken flocks on the same farm site were likely to colonize the breeder flock earlier (R(2) 0.16, P 0.01). These results suggest that there is natural control of Campylobacter dynamics within a flock which could potentially be exploited in designing new intervention strategies, and that the two different species should perhaps be considered separately.

Immunisation errors reported to a vaccine advice service: intelligence to improve practice.

BACKGROUND: The success of immunisation programmes depends on the quality with which they are administered. The Vaccine Advice for CliniCians Service (VACCSline) is an advice service to support immunisers and promote excellence in immunisation practice, through specialist guidance and local education, covering a catchment population of two million people. All enquiries are recorded onto a database and categorised. Vaccine error is selected when a vaccine has not been prepared or administered according to national recommendations or relevant expert guidance. METHOD: All enquiries from 2009 to 2011, categorised on the VACCSline database as 'vaccine error' were analysed and subjected to a detailed free-text review. RESULTS: Of 4301 enquiries, 158 (3.7%) concerned vaccine errors. The greatest frequency of errors, 145 (92.9%) concerned immunisations delivered in primary care services; 92% of all errors occurred during either vaccine selection and preparation or history checking and scheduling. Administration of the wrong vaccine was the most frequent error recorded in 33.3% of reports. A shared first letter of the vaccine name was noted to occur in 13 error reports in which the incorrect vaccine was inadvertently administered. Consultations involving pairs of siblings were associated with various errors in seven enquiries. Failure to revaccinate after spillage (seven reports) showed a widespread knowledge gap in this area. CONCLUSION: Advice line enquiries provide intelligence to alert immunisers to the errors that are commonly reported and may serve to highlight processes that predispose to errors, thus informing immuniser training and updating.

Duck liver-associated outbreak of Campylobacteriosis among humans, United Kingdom, 2011.

Campylobacter- spp.-related gastroenteritis in diners at a catering college restaurant was associated with consumption of duck liver pâté. Population genetic analysis indicated that isolates from duck samples were typical of isolates from farmed poultry. Campylobacter spp. contamination of duck liver may present a hazard similar to the increasingly recognized contamination of chicken liver.

Real-time genomic epidemiological evaluation of human Campylobacter isolates by use of whole-genome multilocus sequence typing.

Sequence-based typing is essential for understanding the epidemiology of Campylobacter infections, a major worldwide cause of bacterial gastroenteritis. We demonstrate the practical and rapid exploitation of whole-genome sequencing to provide routine definitive characterization of Campylobacter jejuni and Campylobacter coli for clinical and public health purposes. Short-read data from 384 Campylobacter clinical isolates collected over 4 months in Oxford, United Kingdom, were assembled de novo. Contigs were deposited at the pubMLST.org/campylobacter website and automatically annotated for 1,667 loci. Typing and phylogenetic information was extracted and comparative analyses were performed for various subsets of loci, up to the level of the whole genome, using the Genome Comparator and Neighbor-net algorithms. The assembled sequences (for 379 isolates) were diverse and resembled collections from previous studies of human campylobacteriosis. Small subsets of very closely related isolates originated mainly from repeated sampling from the same patients and, in one case, likely laboratory contamination. Much of the within-patient variation occurred in phase-variable genes. Clinically and epidemiologically informative data can be extracted from whole-genome sequence data in real time with straightforward, publicly available tools. These analyses are highly scalable, are transparent, do not require closely related genome reference sequences, and provide improved resolution (i) among Campylobacter clonal complexes and (ii) between very closely related isolates. Additionally, these analyses rapidly differentiated unrelated isolates, allowing the detection of single-strain clusters. The approach is widely applicable to analyses of human bacterial pathogens in real time in clinical laboratories, with little specialist training required.

Widespread acquisition of antimicrobial resistance among Campylobacter isolates from UK retail poultry and evidence for clonal expansion of resistant lineages.

BACKGROUND: Antimicrobial resistance is increasing among clinical Campylobacter cases and is common among isolates from other sources, specifically retail poultry - a major source of human infection. In this study the antimicrobial susceptibility of isolates from a UK-wide survey of Campylobacter in retail poultry in 2001 and 2004-5 was investigated. The occurrence of phenotypes resistant to tetracycline, quinolones (ciprofloxacin and naladixic acid), erythromycin, chloramphenicol and aminoglycosides was quantified. This was compared with a phylogeny for these isolates based upon Multi Locus Sequence Typing (MLST) to investigate the pattern of antimicrobial resistance acquisition. RESULTS: Antimicrobial resistance was present in all lineage clusters, but statistical testing showed a non-random distribution. Erythromycin resistance was associated with Campylobacter coli. For all antimicrobials tested, resistant isolates were distributed among relatively distant lineages indicative of widespread acquisition. There was also evidence of clustering of resistance phenotypes within lineages; indicative of local expansion of resistant strains. CONCLUSIONS: These results are consistent with the widespread acquisition of antimicrobial resistance among chicken associated Campylobacter isolates, either through mutation or horizontal gene transfer, and the expansion of these lineages as a proportion of the population. As Campylobacter are not known to multiply outside of the host and long-term carriage in humans is extremely infrequent in industrialized countries, the most likely location for the proliferation of resistant lineages is in farmed chickens.

Marked host specificity and lack of phylogeographic population structure of Campylobacter jejuni in wild birds.

Zoonotic pathogens often infect several animal species, and gene flow among populations infecting different host species may affect the biological traits of the pathogen including host specificity, transmissibility and virulence. The bacterium Campylobacter jejuni is a widespread zoonotic multihost pathogen, which frequently causes gastroenteritis in humans. Poultry products are important transmission vehicles to humans, but the bacterium is common in other domestic and wild animals, particularly birds, which are a potential infection source. Population genetic studies of C. jejuni have mainly investigated isolates from humans and domestic animals, so to assess C. jejuni population structure more broadly and investigate host adaptation, 928 wild bird isolates from Europe and Australia were genotyped by multilocus sequencing and compared to the genotypes recovered from 1366 domestic animal and human isolates. Campylobacter jejuni populations from different wild bird species were distinct from each other and from those from domestic animals and humans, and the host species of wild bird was the major determinant of C. jejuni genotype, while geographic origin was of little importance. By comparison, C. jejuni differentiation was restricted between more phylogenetically diverse farm animals, indicating that domesticated animals may represent a novel niche for C. jejuni and thereby driving the evolution of those bacteria as they exploit this niche. Human disease is dominated by isolates from this novel domesticated animal niche.

Progressive genome-wide introgression in agricultural Campylobacter coli.

Hybridization between distantly related organisms can facilitate rapid adaptation to novel environments, but is potentially constrained by epistatic fitness interactions among cell components. The zoonotic pathogens Campylobacter coli and C. jejuni differ from each other by around 15% at the nucleotide level, corresponding to an average of nearly 40 amino acids per protein-coding gene. Using whole genome sequencing, we show that a single C. coli lineage, which has successfully colonized an agricultural niche, has been progressively accumulating C. jejuni DNA. Members of this lineage belong to two groups, the ST-828 and ST-1150 clonal complexes. The ST-1150 complex is less frequently isolated and has undergone a substantially greater amount of introgression leading to replacement of up to 23% of the C. coli core genome as well as import of novel DNA. By contrast, the more commonly isolated ST-828 complex bacteria have 10-11% introgressed DNA, and C. jejuni and nonagricultural C. coli lineages each have <2%. Thus, the C. coli that colonize agriculture, and consequently cause most human disease, have hybrid origin, but this cross-species exchange has so far not had a substantial impact on the gene pools of either C. jejuni or nonagricultural C. coli. These findings also indicate remarkable interchangeability of basic cellular machinery after a prolonged period of independent evolution.

Informing antimicrobial stewardship with explainable AI.

The accuracy and flexibility of artificial intelligence (AI) systems often comes at the cost of a decreased ability to offer an intuitive explanation of their predictions. This hinders trust and discourage adoption of AI in healthcare, exacerbated by concerns over liabilities and risks to patients' health in case of misdiagnosis. Providing an explanation for a model's prediction is possible due to recent advances in the field of interpretable machine learning. We considered a data set of hospital admissions linked to records of antibiotic prescriptions and susceptibilities of bacterial isolates. An appropriately trained gradient boosted decision tree algorithm, supplemented by a Shapley explanation model, predicts the likely antimicrobial drug resistance, with the odds of resistance informed by characteristics of the patient, admission data, and historical drug treatments and culture test results. Applying this AI-based system, we found that it substantially reduces the risk of mismatched treatment compared with the observed prescriptions. The Shapley values provide an intuitive association between observations/data and outcomes; the associations identified are broadly consistent with expectations based on prior knowledge from health specialists. The results, and the ability to attribute confidence and explanations, support the wider adoption of AI in healthcare.