This series is the second in a weekly round-up of stories I’ve seen relating to Ebola with a strong lean towards the scientific. This post is probably longer than last week’s, and certainly not as well organized. Maybe wider-ranging? Hopefully useful. Anyway, let’s dive in with a quick asterisk relating to:
- Data quality. Those building models are starting to seriously worry about the data going into them (cf GIGO). Up to mid-September, despite rapid growth in cases, contact tracing in Sierra Leone and Liberia has been pretty impressively strong: Caitlin Rivers at Virginia Polytechnic Institute and State University (i.e. Virginia Tech) showed that follow-up rates remain over 90% for known contacts. However, it seems that the two WHO data drops this week have a far higher ratio of new cases to fatalities than was expected, which has led model updates (see below). What is unclear for now is whether the change is due to the epidemic outstripping our ability to follow it, or things actually changing in terms of infections or deaths.
With the above proviso, here’s what I’ve seen in Ebola science this week:
- Epidemic size (update). Given the above concerns re varying data quality, estimates on epidemic size continue vary greatly. No-one I have seen has any certainty regarding what the overall population at risk is, and thus when the epidemic might “top-out” and stop growing exponentially. When contacts between individuals have community structure – i.e. villages, townships, any human setting really – we should expect to see growth spurts, leveling off and then renewed take-off (which would be one explanation for the recent jump in cases in Guinea). The implication is that even when growth curves appear to be slowing, we shouldn’t assume that exponential growth is really tapering off.
- Effective reproductive rate (update). The only scientific publication I saw this week on this topic was from Sherry Towers at Arizona State University, and colleagues, who have developed methods to deal with the irregular reporting of case/mortality data by WHO. Their approach provides ‘local’ estimates of exponential growth in case numbers over small time periods, and then converts them through a standard SEIR mass-action model into an estimate for Rt. Using data up to September 8th, they generate country Rt estimates of 1.6-2.3 for Guinea, 1.5-2.1 for Liberia and 1.2-1.3 for Sierra Leone depending on assumptions regarding incubation and infectious periods. They note that these numbers are in line with the previous work by Althaus and by Fisman et al. that I reported last week. (Towers paper)
- Case fatality rate (update). As mentioned above, the CFR based on current reporting data seems to be coming down from last week’s peak estimation of 80-85%, quite possibly to between 65 and 75% – based on work by David Fisman and Maia Majumder. But they both note that this may be an artifact of slower mortality reporting. I would add that the lockdown and house-to-house investigation in Sierra Leone is leading to the discovery of several unreported cases/deaths in Freetown, which is showing the massive gaps in data emerging. This is in line with the WHO’s estimate of massive under-reporting of cases and deaths.
- Intervention impact. This is the stuff that I’ve been waiting to see come into the open. If there’s one area in which mathematical models should have a massive comparative advantage in an outbreak, it’s predicting what impact different proposed interventions will have. Of course, in order to model impacts one needs a solid base model of what is happening now, and that model needs to be “as simple as possible, but no simpler”. Personally, I would love to be building geographically accurate models of contact networks in each country, but I would be pushing on a piece of string if I did so. So instead I’m watching others building faster and better. Anyway, digression aside, there’s a paper up on the arXiv as of Tuesday from a team at Virginia Tech led by Caitlin Rivers (see also above) and also including Eric Lofgren and Bryan Lewis, starting to dig into the question of ‘what might work’. In a model with three transmission hubs (homes, hospitals and funerals), they show that:
- improved contact tracing shifts infections from homes and funerals to hospitals as cases are brought into care;
- improved infection control within hospitals reduces healthcare infections; and
- a pharmaceutical intervention has a greater proportional impact on home and funeral infections, but overall is little more effective than contact tracing.
- Airborne transmission (update). Still lots of traffic on this. Still everyone is saying that the selective pressure for such a mutation is limited and the chances are very small. Some links to make that point. Just keep moving along people, nothing to see here.
- Convalescent serum. The idea of taking blood from recovered patients and giving it to current patients to provide antibodies to Ebola. While “serum” is used in layman conversation as something akin to ‘https://constrainedoptimization.wordpress.com/wp-admin/post-new.phpwonderdrug’ (cf flying a Brit to the US to provide blood), it has the specific scientific meaning of ‘blood plasma from which clotting factors have been removed’. Which hopefully makes it sound a lot less sci-fi. This would be a good thing since a comprehensive review of the science of serum (of which I will not pretend to parsed every word) by Heather Lander points out that evidence for its success is limited. Which reminds me of the hype about Zmapp a few weeks ago – recovery rate 5/7, with at least 5 individuals getting care in the US or Europe. It might work, but we don’t know yet. In any case, the WHO has OK’d convalescent serum as a first line approach in West Africa, citing the same Kikwit outbreak experience that Lander (and others) say points to serum being largely ineffective. Something to keep tabs on…
- Human resources for Ebola. I am no health systems researcher, but I really liked this pair of posts by Shane Granger on healthcare worker impacts of Ebola (discussion blog; data blog). Granger’s key message is the need not only for ‘loss of supply’ data, but also ‘unmet demand’ data; which seems essential to any effective response. But as a newbie to the field, I really liked his overview of categories of healthcare as they relate to this outbreak:
- Operationally critical job roles (OCJR). The linchpins who plan out everything else, with deep relevant knowledge. E.g.s senior medical staff, virologists, nurse practice managers.
- Critical job roles (CJR). Skilled individuals with expertise. E.g.s doctors, nurses, logistics specialists. These are the positions I see advertised through US/European health networks (and on Facebook) every day.
- Hard-to-fill (HtF). Not high-skill, but high-demand/low-supply positions. E.g. burial crews, ambulance crews, ward staff. Granger adds nuance by highlighting that this group may or may not overlap with Hard-to-replace (HtR): people in this category may die more often than the critical staff, leading to shortfalls, and also an ever-lower desire to take on such roles.
- Historical Ebola material. For those of you who would like some historical context, or just have a few more spare minutes, I would point you to some items that caught my eye this week.
- The Kikwit outbreak – see above – seems like a really important case to study, since it was an urban epidemic. I haven’t delved into the literature in depth, but there are papers out there considering lessons to be learned, e.g. Heymann et al. on international preparedness (Heymann was the WHO lead on SARS, now at LSHTM, Chatham House and PHE) and Hall et al. on medical preparedness. I would love to see something on lessons for community control…
- Dynamics of epidemics – within and between. Every mass-action model I’ve seen used in this outbreak seems to be built from the model in Legrand et al.’s 2007 paper. Built on the 1995 DRC and 2000 Uganda outbreaks, they find an R0 of 2.7 in both cases. Good background reading. For a different take, Thomas House has estimated the gap between outbreaks, and final case and fatality size, based on the past 24 outbreaks. It mainly seems to highlight the heterogeneity of past outbreaks, but their frequent occurrence. (House paper)
- What has Uganda done right? Tara Smith is always worth reading, but she provides a really nice overview of how a nation with endemic zoonotic breakouts (Uganda) has set up structures to manage and minimize harm from them. (Aetiology blog)
An aside on sourcing. With apologies, I haven’t cited every claim mentioned in this post, since many are based my reading of twitter posts by people who have been somewhere between highly trustworthy and heroic to date. Clearly all errors in this post are due to me, not them. A non-exhaustive list of people I place in this category, many of them also cited above:
- David Fisman for math modelling
- Stephen Goldstein for virology
- Ian Mackay for virology (see also Ian’s blog)
- Maia Majumder for math modelling (see also Maia’s blog)
- Caitlin Rivers for math modelling and data wrangling (see also Caitlin’s blog and Ebola github)
In journalism, I can also recommend Cédric Moro (often in French) and Umaru Fofana (Sierra Leone) for detailed up-to-the-minute reports on what is happening in the three countries mainly affected. I know there are many more out there, I just don’t have time to follow them all.
I’d be very interested in hearing about other people producing Ebola science – especially those tweeting results or discussion. As ever, I’m @harlingg