Transcript model
THEORY AND PRACTICE OF INFECTIOUS DISEASE SURVEILLANCE Mark Woolhouse and many others Epidemiology Research Group Centre for Immunity, Infection & Evolution University of Edinburgh M.E.J. Woolhouse, University of Edinburgh, August 2013 TOWARDS ‘SMART’ SURVEILLANCE Using information on patterns of risk of infection to design more efficient (= less effort, lower cost) surveillance systems Topics • Targeted surveillance: FMD, HAIs • Noisy backgrounds: influenza • Unusual events: EIDs Theme • Model-based approaches to designing surveillance systems Pi 1 exp[ βij ] jI ( t ) M.E.J. Woolhouse, University of Edinburgh, August 2013 POST-EPIDEMIC SURVEILLANCE MODEL spatial microsimulation + non-detection risk 5x System diagnostic sensitivity with increasing number of sheep farms sampled showing risk-based sampling and random selection from surveillance zone M.E.J. Woolhouse, University of Edinburgh, August 2013 Handel et al. (2011) PLOS ONE NETWORK MODEL FOR HAI MODEL: stochastic, network SI Hospital ID (ranked) Patient movement network Time to infection (yrs) M.E.J. Woolhouse, University of Edinburgh, August 2013 Ciccolini et al. (submitted); van Bunnik et al. (in prep.) NETWORK MODEL FOR HAI Time to detection Hospitals affected RANDOM GREEDY 6x M.E.J. Woolhouse, University of Edinburgh, August 2013 8x Ciccolini et al. (submitted); van Bunnik et al. (in prep.) NETWORK MODEL FOR HAI Time to detection (days) STRAIN COMBINATION MODEL: MULTI-DRUG RESISTANCE SINGLE DOUBLE No. hospitals M.E.J. Woolhouse, University of Edinburgh, August 2013 van Bunnik et al. (in prep.) DETECTING HPAI IN POULTRY FLOCKS P (OUTBREAK) P (UNDETECTED) MODEL: Within-flock IBM + background mortality Probability of event P (+ SENTINELS) P (UNDETECTED) P (+ SENTINELS) Fraction birds protected M.E.J. Woolhouse, University of Edinburgh, August 2013 Savill et al. (2006) Nature DETECTING OUTBREAKS AGAINST A BACKGROUND PANDEMIC INFLUENZA IN SCOTLAND 2009 M.E.J. Woolhouse, University of Edinburgh, August 2013 OUTBREAK DETECTION DATA: Spatial WCR MODEL: Spatial IBM + Spatially explicit simulations: allocate cases to GPs by postcode given set probability of reporting Ferguson et al. (2006) Nature M.E.J. Woolhouse, University of Edinburgh, August 2013 Singh et al. (2010) BMC Publ Hlth OUTBREAK DETECTION Case Case reporting reproting rate rate Threshold method Cusum method specificity = 95% WCR THRESHOLD WCR method 0.5% Sen MDT 100 5 100 5 96 6 1.0% Sen MDT 100 4 100 5 97 5 5.0% Sen MDT 100 3 100 4 97 4 specificity = 99% CUSUM M.E.J. Woolhouse, University of Edinburgh, August 2013 0.5% Sen MDT 98 5 100 6 77 6 1.0% Sen MDT 100 5 100 5 92 6 5.0% Sen MDT 100 4 100 4 95 5 Singh et al. (2010) BMC Publ Hlth DETECTING PANDEMIC INFLUENZA 2009 12 wks What went wrong? 12 wks Asynchronous outbreaks Low R0 13 wks M.E.J. Woolhouse, University of Edinburgh, August 2013 Singh et al. (2010) BMC Publ Hlth SERO-SURVEILLANCE: EXPOSURE VS VACCINATION MODEL: age-time varying λ (MCMC fit) 1 in 3 people vaccinated already exposed M.E.J. Woolhouse, University of Edinburgh, August 2013 McLeish et al. (2011) PLOS ONE DETECTING PANDEMIC INFLUENZA • Better data – – – – More GPs (now 100s) More frequent reporting (daily) More reliable reporting Serosurveillance data • Better pandemic models • Cleverer algorithms M.E.J. Woolhouse, University of Edinburgh, August 2013 VIZIONS Wellcome Trust-Viet Nam Initiative on Zoonotic Infections AIMS: • Disease burden in a) hospital patients, b) high risk cohort • Outbreak detection algorithms • Identify drivers for disease emergence • Phylodynamics across species barriers • Bioinformatics methodologies M.E.J. Woolhouse, University of Edinburgh, August 2013 VIET NAM HOSPITAL DATA ~250,000 infectious disease admissions over 5 years S. N. pneumoniae meningitidis 6% 1% Others H.i.b 1% 6% UNKNOWN AETIOLOGIES JEV 23% 49% Co-infection TBM Dengue HSV virus 2% 2% 2% Dak Lak: dengue-like fevers Enteroviruses 2% 6% M.E.J. Woolhouse, University of Edinburgh, August 2013 OUTBREAK IDENTIFICATION ALGORITHMS M.E.J. Woolhouse, University of Edinburgh, August 2013 Bogich et al. (2011) Interface RISK (NOT DISEASE) MAPPING Institute of Medicine (2009) Chan et al. (2010) PNAS M.E.J. Woolhouse, University of Edinburgh, August 2013 CONCLUSIONS: BEING SMART • Risk is heterogeneous → targeting works • Smart surveillance is more efficient ‒ More efficient post-epidemic FMD surveillance √ ‒ Faster detection of HAIs √ ‒ Faster outbreak detection? ‒ Detection of novel infections/outbreaks? • Designing better surveillance systems is a challenging problem for modellers » More efficient surveillance and more effective interventions M.E.J. Woolhouse, University of Edinburgh, August 2013 Pi 1 exp[ βij ] jI ( t ) ACKNOWLEDGEMENTS Steve Baker (OUCRU), Paul Bessell, Marc Bonten (UTRECHT), Mark Bronsvoort, Bill Carman (NHSS), Margo Chase-Topping, Mariano Ciccolini, Peter Daszak (NEW YORK), T. Donker (GRONINGEN), Giles Edwards (SMRL), Jeremy Farrar (OUCRU), Neil Ferguson (IC), Eric Fèvre, Cheryl Gibbons, Ian Handel, Shona Kerr, Nigel McLeish, Jim McMenamin (HPS), Maia Rabaa, Chris Robertson (STRATHCLYDE), Nick Savill, Peter Simmonds, Brajendra Singh, Suzanne St Rose, Bram van Bunnik + Foresight and IOM/NAS committees, Generation Scotland FUNDING: Wellcome Trust, EC FP7, ICHAIR, SG, DEFRA, SFC, USAID, SIRN M.E.J. Woolhouse, University of Edinburgh, August 2013