Integrated disease surveillance and care delivery: stories from polio (and weather) Mike Famulare
This talk -A story from the history of quantitative weather forecasting -A short tour of the polio surveillance system -Interaction of surveillance and analysis to understand -polio vulnerability -polio incidence -vaccination campaign planning Key point -Good modeling – timely, interpretable, generalizable, predictive – requires understanding the data generating process
How do we get here? my computer
First numerical weather “forecast” “Perhaps some day … it will be possible to advance the computations faster than the weather advances and at a cost less than the saving to mankind due to the information gained. But that is a dream.” In between shifts driving an ambulance, Lewis Fry Richardson numerically integrates physical equations to compute the change in air pressure over 6 hours on May 20, 1910 near Munich -He predicted a rise in pressure equivalent to going from the eye of a hurricane to a sunny day, when there was actually no change.
1946: weather balloon network Integrated Global Radiosonde Network
1950: first sensible numerical weather “forecast” -ENIAC 24-hr forecast for evolution of deep low pressure over central USA -24 hours of in computer lab to calculate 24 hours of pressure field evolution -Typical forecast was outperformed by assuming today will be like yesterday, but it’s competitive January 30, 1949Observed January 31, 1949 Computed January 31, 1949 January 30, 1949
How did the polio program develop? 1930s-1960s -Basic research into polio epidemiology and control Pan American Health Organization (PAHO) decides to eliminate polio from the Americas. -Develop reporting system -Epidemiologists investigate every paralytic case -Lab network established -Standardized data with centralized reporting -National immunization days introduced Transition focus from polio to all AFP -Virological isolation becomes diagnostic gold standard -“Negative notification” introduced Complete expansion to Asia, Africa, and Middle East Eradication is driven by the info in negative samples Birmingham et al JID: 1997;175(Suppll):S “Spearheading partners” -WHO -US CDC -Rotary International -UNICEF -National health systems -BMGF Who does the work? -> 10 million vaccinators -> 1 million HCW recognize and report AFP -35,000 surveillance officers -6,000 lab, technical consultant, admin staff Budget breakdown -50% for buying and delivering vaccines -40% supports permanent staff, technical assistants, infrastructure -10% surveillance activities
Key features of the polio surveillance system Acute flaccid paralysis (AFP) surveillance. Geographically comprehensive. Peak polio AFP incidence is roughly 30 per 100,000 children under age 15 years. -1 case for every few hundred to few thousand infections AFP surveillance adequacy standard today is 1 per 100,000 children under 15 years old, almost all non-polio. Focus on AFP (and not just polio) provides representative samples of polio vulnerability and maintains surveillance capacity.
Detecting outbreaks -Somalia, Kenya, and Ethiopia experienced a wild polio outbreak in Back then, we captured the data as delivered to polio partners in “real time” First case reported on May 10 with paralysis onset on April 13. Was there a surveillance failure? Probably not. Phylogenetic tree, CDC, as of May 28, 2013 Distributed as ed pdf Vaccination response ramped up by May 30 ( communication: GPEI/PolioThisWeek) May 26 -Supplemental immunization activities (SIAs) began in Kenya and Somalia -First response planned in Ethiopia in district with large ethnic Somali population Data in WHO HQ as of May 28, 2013 Delivered to partners as ed pdf Confirmed cases begin to accumulate
How many people have been infected? June 28, 2013 How big will the outbreak be? July 15, 2013 Last case Aug 11, 2014 Forecasting outbreaks as fast as the data allow Population susceptible June 28, ,000 children (40%) with zero doses of polio vaccine Vaccine coverage -SIR model: first SIA immunized 50% of targeted susceptible population -Monitoring: 65-85% coverage
Infected regions are spatially complex but social integrated. Why did the prediction work as well as it did? -The outbreak stayed within a single ethnic population (that spanned country borders). -We understand the sampling frame well enough. -The breaking exponential gives information about the size of the participating population. Cases in inaccessible regions Mar – Aug 2013
Using every sample: understanding vulnerability Polio vaccination history is collected from every AFP case -Circa : India begins to use AFP vaccination data to augment direct measures of vaccination coverage -By 2014, AFP data is the “silver standard” for measuring effectiveness of vaccination efforts Nigeria -Ability of vaccination campaigns to deliver doses improved over time BMC Med :60 Upfill-Brown, Voorman, Chabot-Couture, Shuaib, Lyons
Interpreting zeros: has Nigeria eliminated polio transmission? March 2015: no polio cases of any type in Nigeria for 4 months But, cases are rare relative to infections, so how long should we wait before we believe polio is gone? 2014
Surveillance and dynamics near elimination permit a simple model (1)The observable cohort (susceptible to paralysis) is essential for sustained transmission. (2)Paralysis is a rare symptom of polio infection and we expect to observe most cases. (3)A linear model is appropriate because susceptible depletion can be ignored.
Surveillance and dynamics near elimination permit a simple model (4)A spatial model is not required. -Elimination is a country-scale question. -The surveillance system can find polio anywhere. -Dynamics near elimination are simple. (5)The model near elimination can be parameterized by data far from elimination. -Transmission intensity is stationary in the places that make cases WHO AFRO - Nigeria PLoS ONE 10(8): e arXiv:
Has Nigeria eliminated wild poliovirus? May 16, 2015 Initialize the model based on recent cases Consider a few scenarios for surveillance and immunity Type 1 very likely eliminated from Nigeria (and all of Africa) Type 3 very likely eradicated globally Persistent vaccine-derived type 2 (cVDPV2) -March 2015 forecast predicted certain persistence into mid-year -new case at 54 th percentile of the prediction interval -Update: -“green” forecast was best forecast -elimination by April 2016 is promising Case R eff = 0.6 R eff = 0.85 R eff = 1 50% surveillance March 31, April 18, 2016 PLoS ONE 10(8): e Famulare. arXiv:
Forecasting and immunity maps inform operations In April 2015, 7 vaccination campaigns are planned from July 2015 to March tOPV (containing types 1,2,3) and 3 bOPV (containing types 1,3) How should the plan account for shifting risks for type 1 vs type 2 transmission? BMC Med :60 Upfill-Brown, Voorman, Chabot-Couture, Shuaib, Lyons 2014 estimated immunity snapshot Type 1 Type 2 Projected immunityModel Output All tOPV schedule: -Loss in type 1 immunity small compared to gain in type 2 Policy impact Implemented schedule -7 tOPV BMC Med :92 Upfill-Brown, Lyons, Pate, Shuaib, Hu, Eckhoff, Chabot-Couture
Closed loop integration of surveillance, analysis, intervention This analysis thread spans 4 years of engagement with Nigeria’s polio program. -Vulnerability mapping -Dynamical incidence modeling -Intervention planning These efforts are a small part of the larger effort to implement surveillance, use data effectively, and deliver vaccines. -More to see in the Vaccine-Preventable Diseases breakout session. The loop works as well as it does because we understand (decently well) -the sampling frame -the disease dynamics -predictive correlations of vulnerability data with polio incidence -how to turn data and analysis into effective interventions
WHO-UNICEF joint reporting process Polio is not an ideal model for surveillance and intervention Decision-making structure is neither hierarchical nor cellular. Institutional ability to use information is highly variable. Data sharing is difficult, even among established partners. There are few mechanisms to make data useful at the point of collection. Privacy is mainly protected by the difficulty of sharing data. Huge resources for one disease when many have related epidemiology.