The researchers in the Informatics team based at Glasgow, Dundee and Strathclyde Universities have constructed linked data sets using routine healthcare data in Scotland to derive risk prediction models for the main health care associated infections - S.aureus bacteraemia, E.colibacteraemia and Clostridium difficile. For all three organisms there are now prototype risk factor analyses and risk prediction models. The implementation of these models in clinical practice is developing, and studies have taken place with practitioners on the most appropriate ways to incorporate the predictions into clinical practice. The team has also investigated delivery mechanisms for the risk prediction models through mobile phone applications, internet web pages and embedding within GP software. Currently the Team is engaged in seeking additional funding to pursue these mechanisms.

Risk factors for acquiring Clostridium difficile infection (CDI), S. aureus bacteraemias (SAB) and E. coli bacteraemias in hospitals using routine data

  • Using routine linked national datasets from 9,798 cases and 29,394 concurrent hospital controls from 2009 to 2015 to investigate risk factors for S. aureus bacteraemias. The principal risk factors identified for an increased risk were increasing deprivation of the patient, increased co-morbidity, care home resident, hospitalisations (including emergency, high dependency unit and intensive care admissions) in the past year, reflecting patients who are frail. An increasing volume of antibiotics prescribed in the previous 90 days was protective, more so for antibiotics deemed not to be protective for S. aureus – penicillin and amoxicillin – than for other antibiotics which were deemed to be protective.
  • A similar study was carried out for E. coli bacteraemias which identified 24,073 hospitalised cases and 72,219 controls, over the same period. These identified similar risk factors for an increased risk of E. coli bacteraemia as for S. aureus bacteraemia with the addition of an increased risk associated with the admissions to hospital due to diseases of the genitourinary system and digestive system and infections with decreased risks for other admission reasons. Furthermore, previous procedures and investigations were associated with an increased risk. Prescriptions of antibiotics deemed protective against E. coli: co-amoxiclav, trimethoprim, nitrofurantoin and any quinolone, in the 90 days before admission had a 2-fold increase in risk of E. coli bacteraemia.
  • An updated analysis of hospital acquired E. coli bacteraemias between 2011 and 2017 was carried out by an ISD analysts under supervision to develop a risk prediction model and also to investigate risk factors for resistant and non-resistant infections. There were 638 resistant cases and 5292 non-resistant cases and 35,233 controls. For any hospital acquired E. coli bacteraemia, the parsimonious risk factor model developed using LASSO had similar types of risk factors as above. The sensitivity and specificity of the risk model were 72% and 75% respectively. A prior E. coli bacteraemia was associated with the risk of a resistant E. coli bacteraemia but this variable was not retained in the risk prediction models for a resistant E. coli bacteraemia. The prediction models were similar for resistant and non-resistant infections.
  • The Clostridium difficile study used data from 2010 until 2013 and was based upon 930 unique cases of Healthcare Associated-CDI, with onset in hospital and no hospital discharge in the 12 weeks prior to index admission, linked with 1810 matched controls. The principal findings were the continued impact of antimicrobial prescribing in the community in the 6 months prior to admission to hospital, particularly if the CDI onset was within one week of admission to hospital, where a prescription of any of clindamycin, cephalosporins, fluoroquinolones and co-amoxiclav was associated with a 2.5 fold increase in risk.
  • There are common aspects to identifying the patients at most risk of these infections. The risks increase with the general frailty of the patient and the strength of the effects of comorbidity, number of previous hospitalisations in the previous year admission for a care home, transfer from another hospital are of a similar magnitude over the three infection types. As these conditions largely describe the population, which are admitted to hospital most frequently, there is limited scope for developing a risk prediction model which will be effective at identifying on admission the patients at greatest risk of specific infections.
  • The risk assessment work has developed a spatial aspect through the PhD project of Florence Tydeman. Recently she has completed (i) an analysis of GP prescribing of antibiotics in Scotland for 2016-17 data to identify high prescribing practices and some of the factors associated with the high levels of prescribing. A second study will be completed within a few months. This is a study of the association between GP prescribing of antibiotics and Clostridium difficile infection in Wales. Both of these studies reveal 4-fold variation in prescription rates for antibiotics across GP practices and that, in Wales, there is an association with CDI. Spatial aspects of the risk of CDI in Scotland will be investigated over the next year.
  • As a result of the no cost extension to March 2021 some of the risk prediction analyses for community acquisition have been postponed until next year to permit the researchers to work on the ECONI study.

Implementation of Risk Prediction Models in Clinical Practice

  • Linkages for hospital acquired E. coli bacteraemia and community acquired Clostridium difficile have progressed to the construction of preliminary risk models and to an investigation into the implementation of these models in clinical practice. Work on the Clostridium difficile risk tool is most advanced and the use of such a model in clinical practice has been investigated
  • A prototype tool for GPs to use to assess risk of patients getting Clostridium difficile as a possible consequence of antibiotic prescribing was developed in collaboration with GPs and other health professionals. Over a period of 12 months we engaged in a longitudinal co-design journey with clinical prescribing champions. Using interviews, observation, and co-design workshops we captured user requirements and mapped facilitators and barriers to the implementation of the tool in practice.
  • The results confirmed that a more detailed and nuanced understanding of health care practices are an essential step in the process of the user centred design and development of digital tools to aid decision making in primary care.
  • Methods are required that capture (i) personal motivations of the workforce that intend to use the tool, (ii) context of use, (iii) interactions with existing systems in real time, (iv) readiness to use tools in existing workflows.
  • The extended and focused co-design approach highlighted the differing requirements of potential users – GP’s, Pharmacists and Nurse Prescribers - as well as the gaps in what the users would like to have relative to the information that the statistical models can provide.
  • A manuscript describing this implementation project is in draft form, as is a second manuscript describing the components of the statistical prediction model.
  • Work on implementation of the risk models is proceeding through the PhD thesis of Ansu Joseph who has started the second round of interviews to get feedback on the prototype and understand the perception of C. difficile from a wider cohort, including primary and secondary care clinicians.
  • The team has also been involved with detailed discussions with a company which writes software for GP systems to fund a follow on study to update the model using data available in GP Systems.

Collaborative Working within SHAIPI and NHS Scotland

  • One of the main challenges of SHAPI has been the development of strong links between the researchers in the three areas – the genomics team, the informatics team, and the infection prevention and control teams. A second challenge has been the links with NHS Scotland and the development of increased research capacity within NHS Scotland through the translation research activities of SHAIPI.
  • There are now very strong links with the informatics teams and the infection prevention and control teams through joint studies – ECONI and SOAP – and also with the informatics team and the genomics researchers through the studies involving S. aureus bacteraemia, E. colibacteraemia and Clostridium difficile infection which are linked to epidemiological data. Both of these links also involve NHS analysts and epidemiologists from Health Protection Scotland and Information Services Division, where the analysts and epidemiologists are involved in meetings, but crucially are also carrying out some of the research.
  • The ECONI study is a major example of these linked research studies which also has a strong translational element. The research team is led by researchers at GCU with support from statisticians at Strathclyde. This is a very complex study involving raw data collection in hospitals and from patients once they have left hospital, linkage to national data sets and, as an addition, the collection of material on the infections for genomic analysis. Researchers from all 3 SHAIPI work streams are now contributing to this study.
  • The ECONI study is led by GCU and funded by HPS. This epidemiology study is divided into four parallel phases. Phase one is an incidence cohort study of laboratory confirmed HAI in two Scottish hospitals (one large teaching and one large general). Data collection began in April 2018 and will complete in June 2019. Interim analysis of the Phase one incidence data is underway. Linkage to routine data sets is being undertaken. The Phase 2 is a nested case control study and involves recruitment of patients and will began in April 2018. Recruitment to this phase of the study has been limited due to the underlying health of the patients in hospital. However sufficient data has been collected to address the study research questions. Phase 3 will investigate the long term consequences of HAI on the quality of life and economic outcomes for the patients, data collection will complete in September 2018 for this phase. This includes a qualitative interview with patients who developed HAI. The final phase which will develop a Markov microsimulation model to support decision making for infection prevention and control. Screening for carbapenemase producing Enterobacteriaceae (CPE) is in draft. Ben Parcel who is part of the SHAIPI collaboration has been collaborating in costing screening strategies for CPE. ECONI will report in January 2020. Preliminary results of this study are available to the research team and full results will be available early in 2020. Final data is not expected to be available until late summer of 2019 and it is largely due to imbedding the analytical work within NHS Scotland and through researchers working side by side with NHS staff over the course of the project that we anticipate completing the study on time.
  • A protocol has been developed and is awaiting approval for an add on study (which is not part of the ECONI deliverables for HPS) which will allow for the collection of isolates which cause E. coli or S. aureus HAI, where available, to be saved, sent and sequenced as part of the above study to allow for the micro-epidemiology within a single hospital to be investigated. This work will involve Martin McHugh, Stephen Fox, and Cosmika Goswami who are all part of the SHAIPI collaboration.
  • The strong links between GCU and HPS have contributed to the development of the study and the strong links, through SHAIPI, between Strathclyde and HPS have enabled the analysis team to expand to involve researchers and analysts within NHS Scotland. There are weekly meetings of SHAIPI researchers at Strathclyde and GCU with analysts at ISD. The NHS analyst is carrying out the data linkage, contributing to the development of the analysis plan, and is responsible for the major analysis of the risk of acquiring a HAI in hospital. SHAIPI researchers also have honorary contracts within HPS and go there on a weekly basis to access the initial data and to work closely with the ISD analysts.
  • The informatics team has continued to engage fully with NHS Scotland through joint working on a day to day basis and training courses. SHAIPI researchers and ISD analysts work closely on the ancillary ECONI study and the preliminary analysis of the first 6 months data is complete.
  • Over the past year SHAIPI researchers in the informatics team have delivered 3 training courses in statistics to HPS epidemiologists and ISD analysts.
  • Mentoring of NHS Scotland staff is one of the key paths for translational research activities and this has led to the initiation of a new SHAIPI project on the identification of risk factors associated with the recurrence of S. aureus bacteraemia. This is to be carried out by Elaine Glass at HPS. A similar mentoring model was used to assist Donald Bunyan at HPS in his study of mortality following an E. colibacteraemia infection.
  • The translational element of SHAIPI research is also being delivered by an industrial placement of a PhD Student, Florence Tydeman, at Guy’s and St Thomas’ hospital in London to work on metabolomics analyses.