by Revere, cross-posted at Effect Measure

We continue our summary of the Institute of Medicine “Letter Report” on non-drug non-vaccine measures to slow or contain the spread of an influenza pandemic of a severity similar or worse than that of 1918 (see previous post on models here). The IOM report examined several analyses of historical data from 1918 to see if it was possible to obtain information on the effectiveness interventions on the pattern of outbreaks in various cities in the US. It is well known that both timing and severity varied a great deal in that pandemic. The goal was to see if differences in morbidity and mortality were related to specific actions taken in response.

The IOM/NAS panel heard from Dr. Howard Markel, Director for The Center for the History of Medicine at the University of Michigan. He and his colleagues have identified 16 non-pharmacetutical interventions (NPIs), including such things as closing schools, restricting public events and making influenza a reportable condition. The overall conclusion from studying 6 cities in 1918 is worth highlighting because it echoes what we have been saying here for two years (can you think of a better reason?):

From these case studies, Markel concluded that investment in public health infrastructure and the building of public trust by local health officials seemed to have facilitated the implementation of the interventions. (p. 15)

There was also this important qualification. The community had to be committed to carrying out the intervention for the entire length of the epidemic (none did) and that even then, from his data, there was no guarantee:

He also observed that “fatigue” was an important factor; in other words, communities which had to reinstitute interventions after having lifted them experienced pushback and noncompliance in the second phase of restrictions. Finally, he concluded that the community interventions may have lowered the peak death rate and that proactive and early implementation were associated with flatter epidemic curves, although there were examples of cities that implemented the strategies but still had severe epidemics.

Dr. Marc Lipsitch, a well respected flu modeler from the Harvard School of Public Health, presented his analysis of 17 cities to see if he and his colleagues could discern whether early intervention affected the pattern or outcome of the epidemic. Using newspaper and secondary sources, Lipsitch et al. studied 17 NPIs to see if timing of intervention made a difference. Early interventions were considered to be those started when as yet few people had died in the city. The panel noted that using mortality as a measure of effectiveness is complicated by the fact that case fatality exhibited almost a three fold difference in across different cities. The general conclusions were that early interventions were related to lower death rate peaks but less of a relation with total deaths. Early school closures were most related followed by cancellation of public events. No other interventions were associated with the pattern of the outbreak.

One feature of analyses, both historical and modeling, is that they often come up with different results, illustrated by unpublished data presented by another modeler, Dr. Neil Ferguson of Imperial College London. Again, the method was to take advantage of the differences in timing and nature of interventions in different cities with very different epidemic patterns. First, there was some concordance with Lipsitch’s results:

Lower peak mortality was correlated with “early” interventions–the same results that Lipsitch and colleagues (2006) found in their analysis. However, they also found that in terventions across the country were started within a few days of each other. The date when the epidemic reached the cities varied more. They also found that peak mortality was strongly correlated with the presence of two autumn peaks, but that total mortality was only weakly associated. These findings point to a major theoretical reason to explain why NPIs may have little impact on total mortality; that is, unless interventions are kept in place until there is no longer a threat of reintroduction, the interventions may delay when people get infected without having much impact on the total size of the epidemic (the total number of people infected). (p. 17, cite omitted)

Ferguson et al. also used historical data in a simple compartmental model to see if they could reproduce the epidemic curves (the development of cases over time), looking to see what intervention measures enabled them to do so best. It appears quite a few assumptions were needed regarding transmission and the effects of various interventions and these weren’t sufficient to get a good fit unless another variable to account for spontaneous changes of behavior (e.g., voluntarily restricting contact with others) was added. These behavior changes clearly do occur independently of mandated interventions and in Ferguson’s model their existence was required to get a decent model fit.

Ferguson et al. have interpreted their results to say that even simple models can reproduce the epidemic curves in cities with widely disparate experiences and that the best of these suggest a 40-45% transmission reduction. Early interventions that last the full duration of the epidemic are most effective but because they did not present data that separated the different interventions (e.g., school closures and mandatory masks), there is not as much information for policy makers as there might be.

Ferguson et al. presented a second set of analyses from French data from the 1980s which seemed to indicate that school closures made little difference, although this model is highly sensitive to assumptions about contact rates. Using the assumptions of the models, reductions in transmission from school closures were mainly in children, that is, the idea that infections in children drive the community rate didn’t seem to be true.

The panel tried to summarize the disparate findings from the modeling exercises (which we report here) and the historical analyses, and their rather circumspect conclusions in the face of the many analyses deserve quoting:

The models generally suggest that a combination of targeted antivirals and NPIs can de lay and flatten the epidemic peak, but the evidence is less convincing that they can reduce the overall size of the epidemic. Delay of the epidemic peak is critically important be cause it allows additional time for vaccine development and antiviral production. Lower ing the peak of the epidemic is crucial also because it can reduce the burden on healthcare infrastructure by avoiding an extremely large influx of patients. Another important find ing is that interventions will likely be most effective if they are initiated early in the epi demic and sustained until the threat of reintroduction of the virus has been eliminated. (p. 19)

The panel also notes that the historical data are limited because there are many differences between 1918 and 2006, among them much higher population densities today. But again, there overall conclusion is worth repeating (we can’t say it too many times):

[T]he committee believes that the finding from Markel and Wantz (2006) regarding the importance of a strong public health infrastructure in mitigating the epidemic likely remains true today. (p. 19)

Once again. The importance of a strong public health infrastructure.