{"id":6188,"date":"2021-10-21T07:30:44","date_gmt":"2021-10-21T05:30:44","guid":{"rendered":"http:\/\/variances.eu\/?p=6188"},"modified":"2021-10-23T11:03:40","modified_gmt":"2021-10-23T09:03:40","slug":"covid-19%e2%80%8a-%e2%80%8aanatomy-of-a-methodology-crisis-why-surveys-of-prevalence-and-severity-are-still-needed","status":"publish","type":"post","link":"https:\/\/variances.eu\/?p=6188","title":{"rendered":"Covid-19\u200a\u2014\u200aanatomy of a methodology crisis. Why surveys of prevalence and severity are still needed"},"content":{"rendered":"
<\/h5>\n
Pour un suivi clair de la Covid-19\u200a\u2014\u200a pr\u00e9valence et s\u00e9v\u00e9rit\u00e9 \u00e0 mesurer par sondage<\/em><\/h5>\n

 <\/p>\n

L\u2019article qui suit est une version l\u00e9g\u00e8rement condens\u00e9e de celui publi\u00e9 <\/u><\/a><\/span>dans le journal en ligne towardsdatascience.com, <\/em>qui fait partie de medium.com[1] :<\/a><\/em><\/strong><\/p>\n

https:\/\/towardsdatascience.com\/covid-19-anatomy-of-a-methodology-crisis-it-all-starts-with-data-collection-e23fcbcba54a<\/span><\/span><\/a>[2]<\/a><\/p>\n

\n

Faute de mesure statistique robuste et continue des indicateurs de suivi \u00e9pid\u00e9miques, il est n\u00e9cessaire de faire des hypoth\u00e8ses sur les tendances de fond. L\u2019ambigu\u00eft\u00e9, ou l\u2019incertitude fondamentale sur les param\u00e8tres-cl\u00e9s des mod\u00e8les, conduit \u00e0 des d\u00e9bats parfois passionn\u00e9s, souvent malheureusement davantage centr\u00e9s sur les conclusions des mod\u00e8les et les politiques \u00e0 mener, que sur la m\u00e9thode de collecte et d\u2019analyse des donn\u00e9es.<\/p>\n

Aussi \u00e9trange que cela puisse para\u00eetre, pratiquement aucun pays \u2014 un contre-exemple notoire est le Royaume-Uni \u2014 n\u2019a r\u00e9alis\u00e9 de mesure al\u00e9atoire de la pr\u00e9valence pour la population dans son ensemble\u00a0; la s\u00e9v\u00e9rit\u00e9, c\u2019est-\u00e0-dire le risque d\u2019une infection grave en cas de contamination, l\u2019est encore moins.<\/p>\n

Pourtant, l\u2019ambigu\u00eft\u00e9, <\/em>c\u2019est-\u00e0-dire l\u2019incertitude sur les param\u00e8tres du mod\u00e8le, repr\u00e9sente un al\u00e9a suppl\u00e9mentaire vis-\u00e0-vis duquel personne n\u2019est neutre. L\u2019ambigu\u00eft\u00e9<\/em> conduit n\u00e9cessairement soit \u00e0 un pessimisme suppl\u00e9mentaire, soit \u00e0 un optimiste suppl\u00e9mentaire \u2014 par rapport \u00e0 la situation dans laquelle seul un risque,<\/em> c\u2019est-\u00e0-dire une distribution de probabilit\u00e9 connue, existe.<\/p>\n

Dans le cas de la COVID, l\u2019ambig\u00fcit\u00e9 peut \u00eatre r\u00e9solue de mani\u00e8re tr\u00e8s simple, par une mesure statistique robuste. Il s\u2019agit tout simplement de mener une enqu\u00eate al\u00e9atoire r\u00e9guli\u00e8re sur la pr\u00e9valence et la s\u00e9v\u00e9rit\u00e9 des infections. Le co\u00fbt d\u2019une telle enqu\u00eate serait inf\u00e9rieur \u00e0 un million d\u2019euros par an\u00a0!<\/p>\n

Cela permettrait de rendre le processus de d\u00e9cision plus robuste\u00a0; cela permettrait aussi part d\u2019am\u00e9liorer la communication des politiques publiques. Et mettrait fin \u00e0 des oppositions st\u00e9riles qui, plus que des diff\u00e9rences d\u2019opinion, refl\u00e8tent parfois une attitude diff\u00e9rente face \u00e0 l\u2019ambig\u00fcit\u00e9.<\/p>\n

Abstract<\/strong>: Without reliable measures of prevalence (viral circulation) and severity (the probability that infections lead to severe forms such as hospitalisation), the medium-term impact of viral circulation cannot be assessed robustly. A random survey of prevalence and associated severity would greatly robustify current research efforts, facilitate decision-making, and foster transparency in public policy, for cost less than \u20ac1 million per year.<\/p>\n

 <\/p>\n

Introduction<\/strong><\/h5>\n
\"\"

Figure 1: Garbage in, garbage out. https:\/\/xkcd.com\/1838\/<\/p><\/div>\n

The naive view that sophisticated models can be applied regardless of the data quality explains why institutions with deep pockets neglect data collection.<\/p>\n

Yet the best efforts of modelers are hampered by data quality and relevance to answer a given question. Lack of reliable data leads to less robust model output, and in turn in practice to more debatable conclusions than desired, even possibly to distrust in public action.<\/p>\n

Deep pockets, at financial conglomerates, international institutions and governments, provide incentives to develop sophisticated models that potentially can use a variety of data sources, but robustness is not a characteristic of models on their own, but the ability to draw unambiguous conclusions to a question with the dataset and models at hands. Although parsimony is thought as a desirable model attribute to test simple models or hypotheses, a parsimonious model without robust data may not lead to robust predictions. Although data abundance and modern, complex models enable new insights or signals, sophistication has not prevented bankruptcies of financial conglomerates. In fact, unmanaged complexity is one of the causes of the 2008 financial crisis <\/strong>(see Greenspan, 2013<\/a><\/span><\/span> and the IMF\u2019s 2008 Global Financial Stability Report<\/a><\/span><\/span>).<\/p>\n

In the case of the COVID epidemic, the poor measurement of two fundamental variables, prevalence and the severity of infections, leads to non-robust predictions regarding the medium-term impact of viral circulation or any policy related to viral circulation. It thus appears necessary to robustify the current research effort with sound and reliable data collection. Our proposal involves a random survey of prevalence.<\/p>\n

Such a survey will permit tracking the viral transmission more precisely and for a cheaper cost than today. As a matter of fact, we reckon it can be implemented in each country or state with costs starting below one million euro per year. It will permit the monitoring of prevalence and severity in an unbiased way for the first time. The impact of viral circulation and governmental measures will be assessed in a transparent and non-debatable way. In turn, this will facilitate not only decision making but also communication and readability of public policies. This will foster transparency and confidence in vaccination campaigns and non-pharmaceutical interventions (such as quarantine).<\/p>\n

This approach can be replicated at the relevant administrative or geographical level (region, state or country).<\/p>\n

Note:<\/em> prevalence is the fraction of people currently infected and can be thought as the population-wide positivity rate; incidence refers to newly infected persons.<\/p>\n

I) A problem starts with a question and a model overview<\/h3>\n
I.1) Monitoring needs<\/h5>\n

The two essential components of the COVID epidemic can be summarised as the viral circulation (number of new infections) and its health impacts, both for individuals and for public health systems.<\/p>\n

A fraction of new infections will lead to hospitalisations. This fraction, one possible measure of the severity of infections, may fluctuate (potentially in either way, see below). It thus must<\/em><\/strong> be monitored.<\/p>\n

That the severity of infections varies in time is particularly true for the respiratory viruses, whose circulation can hardly be stopped.<\/p>\n

I.2) With insights from the theory<\/h5>\n

For the population, the average severity of viral infections evolves in time, potentially increasing or decreasing with viral mutations, and always decreasing with the rise in immunity that results from previous viral circulation. It cannot be assumed constant and thus must be monitored.<\/p>\n

Regarding immunity, it is essential to distinguish between immunity in the upper respiratory tract<\/em> (\u201cthe nose\u201d), and immunity of the lower respiratory tract<\/em> (lungs as well as other internal organs).<\/p>\n