The summary of Dr. Ceire Costelloe's presentation from the Jun 11-12th 2019 event Data-driven systems medicine at Cardiff University Brain Research Imaging Centre.
Premium Call Girls In Jaipur {8445551418} ❤️VVIP SEEMA Call Girl in Jaipur Ra...
Dr. Ceire Costelloe (Imperial College London) - Data-driven systems medicine
1. Céire Costelloe BA(hons) MSc PhD CStat
Senior Lecturer, Director Global Digital Health unit
Data-driven systems medicine workshop, Cardiff 12th June
Using real world healthcare data to drive
precision medicine across the
UK healthcare economy
2.
3.
4. Background
• 25,000 patient die each year from infection caused by multi-drug resistant
bacteria in the EU
• Does antibiotic prescribing lead to increased prevalence of antibiotic resistant
infections?
• Primary care provides 80% of all antibiotic prescriptions
The bacterial challenge: time to react, Joint Technical Report from ECDC and EMA, Stockholm, September 2009
Centre for Disease Dynamics, Economics & Policy. State of the World’s Antibiotics, 2015. CDDEP: Washington, D.C
5. Association between primary care
antibiotic use and ABR in
E. coli isolates from adults
presenting to primary care with
suspected UTI
Costelloe et al., BMJ 2010;340:c2096
Community antibiotic use is associated with
antibiotic resistance in individual patients
6. Community antibiotic use is associated with antibiotic
resistant E.coli infections in children
Association between primary care antibiotic use and ABR in
E. coli isolates from children presenting to primary care with suspected UTI
Bryce A, Hay AD, Lane I,Thornton HV, Wootton M, Costelloe C. BMJ 2016
7. Costelloe et al. BMJ 2010;340:bmj.c2096
Antibiotic use is associated with antibiotic resistance in
healthy adults – evidence from trials
8. Data Linkage, Syndromic Surveillance and Modelling
Overall aims:
– To develop models of risk of infection focusing on urinary tract infections
(UTIs), pneumonia and bloodstream infection
– To investigate the epidemiology of AMR infections in relation to antibiotic
usage
➢ To utilise and link together existing local, regional and national
healthcare datasets to address research aims
10. Is it overuse of antibiotics that drives antibiotic resistance?
• Need to consider prescribing quality
• Need to consider the whole healthcare economy - interface between primary
and secondary care (community - hospital)
• Measuring levels of antibiotic prescribing for urinary tract infections
(UTIs) in primary care in England which does not adhere with national
prescribing guidelines - what impact does this have on blood steam
infection?
• Make use of healthcare data on a national scale
11. Antibiotic prescribing for UTI in primary care in England
which does not adhere with national prescribing guidelines
• UTIs are among the most common infections in primary care
• Resistance to trimethoprim (used to treat UTIs) is ~40% in England
• Up to 1/3 of antibiotic prescriptions in the EU do not follow guidelines
• Prescribing guidelines are one of the main components of Antimicrobial
Stewardship Programs
• Treatment failures for UTIs are the main recorded source of bloodstream
infections
1. Laupland, K.B., et al., Community-onset urinary tract infections: a population-based assessment. Infection, 2007. 35(3): p. 150-3.
2. Public Health England, English surveillance programme for antimicrobial utilisation and resistance (ESPAUR) 2010 to 2014 2015.
3. Davey, P., et al., Interventions to improve antibiotic prescribing practices for hospital inpatients. Cochrane Database Syst Rev, 2013. 4: p. Cd003543.
4. Ashiru-Oredope D, Sharland M, Charani E, McNulty C, Cooke J. Improving the quality of antibiotic prescribing in the NHS by developing a new Antimicrobial
Stewardship Programme: Start Smart—Then Focus. Journal of Antimicrobial Chemotherapy. 2012;67(suppl 1):i51-i63.
5. Wilson, J., et al., Trends among pathogens reported as causing bacteraemia in England, 2004–2008. Clinical Microbiology and Infection, 2011. 17(3): p. 451-458.
12. Study aim
• To describe the characteristics of a cohort of adult patients receiving antibiotic
treatment for UTIs in primary care as well as reporting the frequencies of:
– antibiotic type,
– antibiotic dose and
– antibiotic duration of UTI treatment
• To investigate the effect of non-adherent prescribing in primary care for UTI
patients on bloodstream infection hospital admissions.
13. Patient records from 10% of English
GP practices
Records for every admission to an
English NHS hospital
Data from all death certificates registered
in England and Wales
Neighbourhood deprivation
data for all of England
19. Uncomplicated & recurrent UTIs
Antibiotic Dosage Frequency Percent
Trimethoprim 200mg 292,432 56.9%
Nitrofurantoin 50mg 75,907 14.8%
Cefalexin 500mg 22,405 4.4%
Cefalexin 250mg 19,448 3.8%
Nitrofurantoin 100mg 14,266 2.8%
Amoxicillin 250mg 14,124 2.8%
Co-amoxiclav 250mg/125mg 13,131 2.6%
Amoxicillin 500mg 10,712 2.1%
Co-amoxiclav 500mg/125mg 8,251 1.6%
Ciprofloxacin 500mg 6,874 1.3%
Ciprofloxacin 250mg 6,664 1.3%
Study period: 2008 to 2017
Total patients: 376,314
Total UTIs: 721,763
Antibiotic/dose frequency of UTI prescriptions (2008-2017)Regional distribution of UTI patients in cohort (2008-2014)
12969
40738
46146
9031
54462
58885
49621
41402
47808
10342
East Midlands London North West South East
Coast
West Midlands
Total recurrent UTIs: 207,473
(28.8%)
Mean age: 59.92 (SD 19.6)
Female: 83.7% Male: 16.3%
20. Antibiotic Dosage Percent
Trimethoprim 200mg 41.6%
Nitrofurantoin 50mg 20.7%
Cefalexin 500mg 6.3%
Cefalexin 250mg 5.6%
Amoxicillin 250mg 4.2%
Co-amoxiclav 250mg/125mg 3.5%
Amoxicillin 500mg 2.9%
Co-amoxiclav 500mg/125mg 1.9%
Nitrofurantoin 100mg 1.9%
Antibiotic
duration Frequency
3 days 1.10%
5 days 47.20%
7 days 48.60%
Numeric daily dose = 2 in 98% of UTIs
✗
✗
✓ ✓
✓
Frequencies of antibiotics prescribed in
female non-adherent prescriptions
Antibiotic Dosage Percent
Trimethoprim 200mg 30.0%
Nitrofurantoin 50mg 14.1%
Cefalexin 500mg 7.7%
Ciprofloxacin 500mg 5.9%
Co-amoxiclav 250mg/125mg 4.9%
Cefalexin 250mg 4.7%
Amoxicillin 500mg 4.1%
Nitrofurantoin 100mg 4.0%
Ciprofloxacin 250mg 3.9%
Frequencies of antibiotics prescribed in
male non-adherent prescriptions
✓ ✓ Numeric daily dose = 2 in 98% of UTIs
✓
Antibiotic
duration Frequency
3 days 35.2%
5 days 58.9%
7 days 0.64%
✗
✗
Uncomplicated UTIs
21. Measuring patient outcomes following a non-adherent
UTI prescription
Non-adherent
prescribing for UTIs
<-> BSI risk
Levels of non-adherent
prescribing for UTIs in
1º care
Descriptive analysis
Linkage with HES/ONS
data to measure risk of
BSI admission in 60
days
Logistic regression and
survival analysis
22. What can we do – Informing interventions
• Local level initiatives: clinical decision support tools
• National initiatives: Reduce inappropriate prescribing
• Evaluating local and national initiatives
23. Imperial College healthcare Trust
• Five hospitals across North West London
• Treating ~2 million patients per year
• Together with Imperial College London forms Academic
Health Science Centre
• One of 11 NIHR funded Biomedical Research Centres
across UK
24.
25. Sepsis CQUIN introduced in 2015/2016 to improve
sepsis screening
Assess the performance of the digital sepsis alert
Evaluate the effect of sepsis alert on patient outcomes
– mortality, length of stay, appropriate antibiotic
therapy
Precision Medicine and the Impact of digital alerting for
sepsis
26. Evaluate a sepsis alerting intervention
§ Challenge - no gold standard for diagnosing sepsis
§ approximately 20% of patients who have an alert have a blood
culture.
§ Sepsis coding in administrative data is affected by clinician
awareness and national policy
27. ICD-10 coding for sepsis pre/post digital sepsis alert
Dec 16 June 17 Dec 17 June 18
28. Evaluate a sepsis alerting intervention (PreMiSS)
• A natural experiment using causal inference methods
• The Cerner module ‘alerts’ medical staff if patients’ symptoms could be sepsis
• Key patient outcomes
– mortality (30 days and 7 day)
– Post infection length of stay
– Process measures – microbiology test within 24 hours (+/-) of alert
– Antibiotics within one hour of alert
29. Evidence of organ dysfunction
SPB <90 mmHg (30 hours)
Lactate >2.0 mmol/L (12 hours)
Bilirubin: 2.0 mg/dL and<10.0 mg/dL (30 hours)
Creatinine: Increase of ≥0.5 mg/dL from base-line
(72 hours)
Suspicion of
sepsis
Suspicion of
severe sepsis
Evidence of infection response
Temperature >38.3o
C or<36o
C
Heart rate >95 beats/min
Respiratory rate ≥ 22 breaths/min
WBC >12,000 or <4000 cells/mm3
Glucose <141 and >200 mg/dL
≥3 criteria
infection response
≥2 criteria
infection response
&
≥1 criteria
organ dysfunction
≥2 criteria
30. • visible to clinicans - ‘intervention’ group
• running silently and not visible - control group natural experiment
31. Methods
Key outcomes – informed by national targets
q In-hospital mortality within 30-days
q Prolonged hospital stay ≥ 7 days
q IV antibiotics within one hour of alert
Methodology
Inverse probability of treatment weighted multivariable logistic regression was used to
adjust for confounders.
32. Patient population
Length of stay
Patients who alert in the ED
9988
Antibiotic use
Patients who need IV
antibiotics
6563
Mortality
All patients who alerted
21,183
33. Results
Length of stay is a binary outcome based on the NHS definition of a stranded patients.
Timely antibiotics are defined as within one hour of the alert – sample is patients who received
antibiotics within 24 hours
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
34. Results
Length of stay is a binary outcome based on the NHS definition of a stranded patients.
Timely antibiotics are defined as within one hour of the alert – sample is patients who received
antibiotics within 24 hours
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
35. Results
Length of stay is a binary outcome based on the NHS definition of a stranded patients.
Timely antibiotics are defined as within one hour of the alert – sample is patients who received
antibiotics within 24 hours
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
36. Results
Length of stay is a binary outcome based on the NHS definition of a stranded patients.
Timely antibiotics are defined as within one hour of the alert – sample is patients who received
antibiotics within 24 hours
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
37. Results
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
OR (95% CI) OR (95% CI) OR (95% CI)
Unadjusted 0.67 (0.67, 0.90) 0.97 (0.89, 1.05) 1.38 (1.22, 1.55)
Adjusted (reg) 0.79 (0.67, 0.93) 0.97 (0.87, 1.05) 1.70 (1.43, 1.95)
Adjusted (IPTW) 0.76 (0.70, 0.84) 0.93 (0.88, 0.99) 1.71 (1.57, 1.87)
RR (95% CI) RR (95% CI) RR (95% CI)
Adjusted (IPTW) 0.76 (0.70, 0.84) 0.96 (0.93, 0.99) 1.35 (1.28 to 1.41)
38. Results
Death Extended LOS Timely ABX
Control Live Control Live Control Live
Total encounters 15061 6671 4494 5494 1927 2695
Number events 959 339 1846 2209 712 1204
% events 6.4 5.1 41.1 40.2 36.9 44.7
OR (95% CI) OR (95% CI) OR (95% CI)
Unadjusted 0.67 (0.67, 0.90) 0.97 (0.89, 1.05) 1.38 (1.22, 1.55)
Adjusted (reg) 0.79 (0.67, 0.93) 0.97 (0.87, 1.05) 1.70 (1.43, 1.95)
Adjusted (IPTW) 0.76 (0.70, 0.84) 0.93 (0.88, 0.99) 1.71 (1.57, 1.87)
RR (95% CI) RR (95% CI) RR (95% CI)
Adjusted (IPTW) 0.76 (0.70, 0.84) 0.96 (0.93, 0.99) 1.35 (1.28 to 1.41)
39. Results
The introduction of the sepsis alert into wards across ICHT resulted in a 24%
reduction in risk of death, 29% increase in receiving timely antibiotics in
the ED and a 7% reduction in risk of a long length of stay for patients
admitted through the ED.
These results suggest an important clinical benefit of the alert - in one month, a
potential increase of 14 (95%CI:8 to 20) patients a month receiving timely
antibiotics in the ED, and a reduction in long hospital stays for 9 (95%CI:2 to
15 days) patients admitted through the ED.
bioRxiv 637967; doi: https://doi.org/10.1101/637967
40. Conclusion
Introduction of the sepsis alerting system appears to have changed
behaviour resulting in improved outcomes for patients
What is the mechanism of improvement?
• Highlights severely ill (potentially sepsis) patients
• Improves/changes dialogue between nurses and doctors
• Raises the question ‘is it sepsis?’
• Associated treatment plan
• Training on using the alert raised the sepsis agenda
41. Conclusions
Clinically as well as statistically meaningful improvements in patient
outcomes
Introduction of the sepsis alerting system appears to have changed
behaviour resulting in improved outcomes for patients
• Highlights severely ill (potentially sepsis) patients
• Improves/changes dialogue between nurses and doctors
• Raised the question ‘is it sepsis’
• Coupled with a treatment plan
• Training on using the alert raised the sepsis agenda
42. What next?
• Impact of introduction of sepsis alerting in acute trusts using different
algorithms (NIHR Health informatics collaborative)
– Impact on patient outcomes
• Mortality
• Length of stay
• Readmissions
• Longer term impacts
– Impact on process measures
• Timely (& appropriate) antibiotics
• Blood cultures
44. Sepsis alert fires
Alert fires
• severe sepsis or
• sepsis
Notification to nurse
IF logged on to a computer
Alert window
visible to Drs
If in that patient’s record
Dr closes alert
Treatment plans/ confirmation
buttons/ ‘not sepsis’ buttons
In ED spreadsheet
with colours
45. Methodological Approach
• Initial analysis of the two cohorts suggested we were not comparing
similar groups and we risked confounding.
• PS methods rely on a model of the treatment given confounders
• In order to preserve a complete sample, we used propensity score
weighting (inverse probability of treatment weighting - IPTW).
• Each patient encounter was reweighted based on the probability of
alerting in the silent and live phase.
• Multivariable logistic models were used to determine the propensity
scores.
• Adopted a doubly robust method – confounders also included in
logistic models of outcome
46. COMORBIDITIES
Based on any relevant ICD-
10 code appearing in the
discharge diagnosis codes
MI; CHF; PVD; Stroke;
Dementia; Pulmonary;
Rheumatic; PUD; Liver
(mild); Liver (severe);
Diabetes; Diabetes
(complex); Paralysis; Renal;
Metastatic cancer; HIV.
ETHNICITY
Based on the following
groupings:
White; Asian; Black; other;
and not known.
SEX
AGE
Grouped into 18-44; 45-64; 65-69; 70-74;
75-79; 80-84; and 85+
DEPRIVATION
Measured as the deprivation score of
the patient’s GP practice, obtained by
matching patients to their registered
GP. If patients did not have a
registered GP or the practice was not
included in the PHE practice profiles a
‘missing’ categorisation was allocated.
There are therefore six deprivation
categories, with Quintile 1 being the
least deprived.
47. SEASON OF ADMISSION PATIENT SEVERITY
NEWS SCORE
NEWS = 0 : Zero
1≤ NEWS<5 :Low
5≤ NEWS<7 : Medium
7≥ NEWS : High
A NEWS score is available for
19599 of the patients (90% of the
patients).
HOSPITAL OF ADMISSION
PATIENT SEVERITY
ALERT STATUS
Severe sepsis or sepsis