Another summary of breaking science surrounding Ebola, with a strong skew towards the epidemiological. Apologies for the delayed release; hope it’s useful to you.
A. Modelling the Epidemic
A1. Data quality
Things appear to be getting increasingly murky. Case numbers shot up early in the week, when previously un-included events from earlier in the year were added to the totals in Liberia; and then dropped down at the end of the week as a number of suspected cases from Guinea were ruled out. Upshot: building models from the cumulative case curve, or even the day-to-day new case numbers is going to be increasingly difficult as these figures cease to even reflect truly new cases in each reporting period. And I will note that Hans Rosling is now in Monrovia working with the MoH, which may explain some of the changes in reporting.
A2. Epidemic trajectory
The reporting front. Following on from last week, there have been efforts to determine if the epidemic really is slowing in Liberia. While numbers of patients at hospitals, particularly around Monrovia, are definitely down, there has been suggestions that this may be due to healthcare avoidance by relatives who don’t want their relatives cremated. Numbers elsewhere do not appear to be falling. There is a nice overview of the past month’s epidemic (up to October 18th) in MMWR; it’s shocking to think how fast things have changed in just a few weeks.
The modelling front (theoretical aside).
As people have been worrying more about data quality, the use of models to predict case numbers is falling away. So I’m going to use this section to provide a brief primer on mass-action mathematical models for Ebola, which may be handy for the section below on treatment. A compartmental model is one in which each individual is in a single state at any given time, and can move between them according to set rules. These rules can be simple (every period, you move on with a 10% probability), or complicated (every period, you move on if you meet someone infectious and neither of you has been vaccinated, with a probability of 7% if you are under 5 years old, and 3% otherwise). As well as complex transition rules, you can also have simple or complex sets of compartments. A simple version is the linear one SEIR. Here everyone is susceptible (S), exposed but not symptomatic (E), Infectious (I) or Recovered (R). With Ebola, one can have multiple infectious states, e.g. when at home (I), in hospital (H) or at a funeral (F). This thought process led to Legrand et al‘s 2007 SEIHFR model, commonly used for modelling Ebola. A slight tweak on this in the past month was Camacho et al‘s model which keeps track of how individuals were infected. One can, however, get very complicated with these models, as seen in Pandey et al’s advanced-ninja model in Science this week. Such models track everything with great care and allow you to see exactly who is where when. Oh, and the mass-action part means we don’t restrict who can infect whom; we can change all that with agent-based models, which track every single person separately. They usually require serious computers.
My takeaway here though, is that is important not to be seduced by complexity: as my math modelling teachers always said, don’t trust the findings if the model is detailed, trust them if the level of complexity is sufficient to get at the research question, but no more so. And with that, back to the research.
The one paper looking at the “natural history” of the epidemic this week is that of Jeffrey Shaman and colleagues at Columbia. The authors use an SEIR model and incidence data up to 28 September to show similar estimates for R0 to other papers (1.89 in Guinea; 1.72 in Liberia; 1.45 in Sierra Leone). They also provide additional estimates of the incubation, infectious and symptoms-to-death time periods. The key message I take from the paper is that their forecasts of cases for the 6 weeks from the end of data collection have consistently undershot for Liberia, but not for Sierra Leone and Guinea. As the authors note, and in line with observations mentioned above, something different is going on in Liberia – either in terms of reporting or the epidemic trajectory. The big question is then, what? (Shaman paper; Shaman website for live updates)
A3. Epidemic parameters, various
Stephen Goldstein provides a great summary of what we know about infection risk in asymptomatic, infected individuals. Goldstein draws on research from the 1995 Kikwit outbreak showing no transmissions without physical contact during clinical illness (n=78) and from the Uganda 2000 outbreak showing barely detectable viral load in the first 2-3 days of symptoms, to suggest that risk of infection during incubation and early illness is probably magnitudes lower than later on.1
There were also a couple of important descriptive epidemiology papers this week.2 First, John Schieffelin and colleagues reviewed patient-level data from 106 Ebola-diagnosed patients Kenema hospital in Sierra Leone – an initial epicentre of the SL outbreak – from May and June 2014. As highlighted in the press, key predictors of mortality included older age (over 45 was highest risk, but dose-response so that even young adults were at increased risk to those <21 – this last, not statistically significant, but a 20% absolute survival difference) and viral load at presentation (again, dose-response). A prognostic score for fatality risk included these two factors, plus baseline temperature, diarrhoea, weakness, dizzyness, abdominal pain and three blood tests (blood urea nitrogen; creatinine; aspartate aminotransferase [AST]). (Scheiffelin paper)
Second, Angela Rasmussen and colleagues presented evidence suggesting that genetic factors may play a part in causing divergent outcomes in patients: specifically, haemorrhage and thus risk of death from Ebola occurs in mice with certain genetic variations. The authors suggest that finding similar variations in humans might be an important first step to reducing the mortality rate. (Rasmussen paper) I will admit that I am no expert in lab studies, and relied somewhat on articles such as this explain things to me. But I did connect this with something I saw previously: work from the 2000 Uganda outbreak showing a connection between HLA-B alleles and Ebola fatality.
Alert: subjective read ahead. My synthesis of the evidence I’ve read over the past few weeks, suggests to me that those infected with Ebola may fall into three categories: (a) sub-clinical, infected but never infectious; (b) rarely-fatal, infected but on an early trajectory to serious but manageable illness, even in settings with limited or no healthcare; and (c) often-fatal, infected and with a very high risk of death if not treated very early in high-quality healthcare settings. This doesn’t mean that genetics are destiny, but they may play a part in determining how serious illness is if someone is infected. I would also note that trajectories for children may be very different from adults, and would probably benefit from careful study as a separate population. </opinionation>
A quiet week here. But I will note that the National Geospatial-Intelligence Agency has unclassified its data for West Africa to provide a mapping platform for response efforts. Currently it looks like they have data for Guinea and Liberia, but not Sierra Leone. Maybe this will help someone?
And there is now a single source for data on all the historic Ebola outbreaks, often linking individual cases geospatially and temporally. Gathered by Simon Hay and his research group at Oxford. What a great resource.
C. Stopping the Epidemic
The above evidence on low blood-based viral load early in the disease process highlights the difficulty of effectively testing people at airports, or elsewhere. PCR (polymerase-chain reaction) tests, the typical gold standard for diagnosis, requires virus to be in the blood, but early on Ebola hangs out in the organs. Rapid blood tests are being developed, but since they are either PCR or antibody tests (the latter being even slower to emerge, since they require the body to respond to the virus), even fast tests aren’t likely to be effective in catching people early in their infections.
Two papers have come from Alison Galvani‘s team at Yale (in addition to the one from last week). In the first paper, Dan Yamin and colleagues used data from the Liberian Ministry of Health and Social Work, including contact tracing data on 246 cases from August 2014. They built an SEIR-type model with two infectious compartments (early and late) and allowed infectiousness (taken from 2000 outbreak data) and contact rates to vary daily for each individual. This was thus an individual-agent model. Two key findings arise from this paper. First, the authors highlight that survivors and non-survivors look very different in terms of how many secondary cases they generate (i.e. their own Re) as the viral load of non-survivors becomes much higher as their illness progresses. As a result, a strategy which suceded in isolating 75% of those patients who subsequently die within 4 days from symptom onset is likely to eliminate the disease. Alternatively, reducing all patients’ contact rate by 60% from the first day of symptoms would have the same impact. (Yamin paper)
The second paper, led by Abhishek Pandey and Katherine Atkins used a Legrand-style model but stratified each category by location (community, hospital, quarantine) and healthcare worker status (yes/no). They, in line with most other modelling work to date, found that only a combination of treatment, quarantine and safer funeral practices would be sufficient to curtail the epidemic.(Pandey paper)
Eric Lofgren and colleagues at Virginia Tech highlight that simply ramping up hospital provision is unlikely to do more than slow down the epidemic. Their paper considered differential effectiveness of specialized Ebola Treatment Centres (ETC), community-based centres and home-kit approaches, but even in optimistic scenarios other prevention interventions were needed to bring the epidemic under control. This finding is in line with previous papers showing that treatment alone is not going to stop the epidemic any time soon. (Lofgren paper on the arXiv)
Buttressing Lofgren and Pandey/Atkins’ findings, Anton Camacho and colleagues modelled case data from the original 1976 Ebola outbreak, using individual-level “line list” data. They found that transmission mostly fell away prior to hospital closure (the hospital was the focal point for infection in this outbreak). The authors point to behaviour change by residents as the most likely cause of epidemic die-out. (Camacho paper)
Which leaves us with an apparent divergence between the message that containment/treatment alone may or may not be sufficient to end the epidemic. One way of reading the Yamin paper is that changes in funeral practices and reduced home-based care are baked into the modelled intervention – since contact rates are being severely curtailed. I think my takeaways are: (i) there needs to be a significant reduction in contact rates to end the epidemic; (ii) this will require changes at every point in the Ebola illness process; and (iii) these changes aren’t impossible to envisage/carry out.
PLoS has taken a second unusual step in responding to Ebola. In addition to their rapid-turnaround PLoS Currents they have now decided to blog-publish pre-peer-review papers they think urgent and likely to be accepted, via their Speaking of Medicine blog. The first paper up is on a topic of great interest to me: the social setting of rural Sierra Leone and how it feeds the epidemic. Paul Richards and colleagues draw on a mass of data collected over the past four years to show how trust (or lack thereof in national institutions), marital patterns, funeral practices and inter-village dependencies (for marriage, school, work, markets) help to propagate the epidemic. The authors suggest the need to convey accurate information, make hospital attendance truly attractive and to leverage family/village structures to protect people from infection by involving them in decision-making for care practices. (Richards paper pre-acceptance).
Reading this article, especially the migration parts, reminded me of the Economist article last week highlighting the potential of cellphone data to help in understanding epidemic spread, and how it efforts are being stymied by a lack of experience in managing data sharing. A sad read, and a journalistic follow-up to this earlier commentary on the subject.
A second, briefer, social determinants publication this week was this letter by Robert Synder and colleagues, highlighting the role of urban slums in propagating the epidemic, and how their residents are often highly mobile but not via formal channels. Thus, the poorest are least able to avoid infection, and also likely to then pass their infection on. A reminder that social policies to alleviate urban poverty can be great health policies too.
D1. International travel restrictions/Quarantine
Short and sweet here: a long but thoroughly argued piece by Judy Stone caught my eye this week. She highlights, among other things, the inconsistency of quarantining returning HCW with no specific risk but not similar HCW who treated patients in the United States. I imagine the argument would be that conditions are better here than there; but at some level we probably are going to have to trust HCW to know when they are/are not at risk of developing infection. And given the benefits of early treatment, there’s a fair amount of self interest in turning yourself in if your temperature spikes. And as others have noted, quarantines in West Africa don’t appear to have worked very well. While in a perfect world, quarantines will act as fire-breaks, as ever we have to consider their effectiveness (on-the-ground) not their efficacy (in-the-laboratory-in-our-heads), and balance this against the economic and human rights costs.
D2. Risk communication
I don’t know quite what to make of this one, but I want to put it out there. Jody Lanard and Peter Sandman have a new blogpost positing that messaging around Ebola needs to be far more focused on the worst-case scenario. As with much of this epidemic, there is a balance between overreacting and underreacting; since we don’t know what is going to happen, we have to make educated guesses at the probability of various scenarios. Lanard and Sandman argue that the downside of overreacting is less than the downside of underreacting. I think they underestimate the impact of scaring the public; something about the intelligence of a mob? There’s a debate to be had here, but I think that at present the product of the probability of an epidemic in the United States (which it appears we don’t have the data to estimate accurately, but it’s pretty darn low) and the potential damage caused by highlighting the impact of such an epidemic doesn’t make a convincing case for going negative on this one.
And to offset this perspective, I offer you a letter using the 2003 SARS outbreak to highlight that data presentation can be crucial for how information is understood. The authors note that daily figures are easier to comprehend than cumulative numbers, conveying what is happening now, rather than everything that has happened to date – and thus allowing readers to update their understanding of the margin, often more important than the mean/total. A quick point that made me think about how I display data…
- I felt I needed to link to this New Yorker piece by Richard Preston, he of The Hot Zone. I should note that many people find his tone unreasonably sensationalizing – especially since he purports to be providing non-fiction. But he is part of the Ebola conversation, so a link is provided, along with one to one of his strongest critics.
- And a shorter article by Helen Epstein, mostly on falling hospital numbers in Liberia. While I don’t always agree with Epstein’s messages, her health journalism is amongst the very best out there.
- And the Lancet held an Ebola twitter chat and then Storified the resulting discussion. Very high-calibre respondents make this a great resource for getting up to speed on some of the key debates.
If you have seen something I’ve missed (or links are broken), you can reach me @harlingg. And as ever, thank you to all those tweeting Ebola science out there, from whom I glean so much.
1 I will add a footnote here (for my own reference really) about the work of Daniel Bausch and colleagues, also in the 2000 outbreak, showing the ability to extract viral markers from many clinical specimens, but very few hospital surfaces.
2 I use the term “descriptive” as a differentiator from modelling papers, rather than to denigrate them in contrast to, say, “causal”.