Sunday, July 16, 2023

 

Covid-19 in France : ending in June 2023?




At the end of June 2023, Santé Publique France ceased publication of the last of the statistics tracking Covid-19 prevalence in France. According the the following announcement, after a two week adjustment period, new virological surveillance statistics will be published weekly at national, regional and departmental aggregation levels. They do not say what these new statistics will be.
The stated rationale is a "favorable epidemic context." 


The announcement [fr] from geodes.santepubliquefrance.fr


Dans un contexte d’épidémie favorable, à compter du 1er juillet 2023, le cadre juridique actuellement en cours prévoit l’arrêt du traitement des données personnelles issues de SI-DEP. Par conséquent, après une période transitoire d’ajustement de deux semaines, les nouveaux indicateurs de surveillance virologique seront publiés aux niveaux national, régional et départemental à une fréquence hebdomadaire.

Les consignes de saisie spécifiques dans SI-VIC seront levées à partir de cette date, les indicateurs hospitaliers ne seront plus disponibles.

Santé publique France maintient la surveillance de l’épidémie à travers son dispositif multi-sources. Les indicateurs relatifs à la surveillance génomique, aux recours aux associations SOS Médecins, aux urgences hospitalières et aux décès resteront disponibles.

The Last Measures

The last measures I'd been tracking were of lab test results (number of tests, positive tests, prevalence of positive tests per 100 000 people in the cohort) and weekly hospital admissions for covid-19, or for something else but positive for SARS-Cov-2, by region, per 100 000 residents.


Confirmed test results

The lab test results series were of diminished interest after the changes to limit access to tests beginning a couple of months ago because the statistics were no longer comparable to those of the three years acquired thus far (which already had some consistency issues linked to policy changes). A much smaller proportion of the population was now sampled, restricted to symptomatic cases (I believe), without contact tracing.  However, as testing continued in this new policy context, a history of several weeks of comparable data was constituted, enabling inference of recent dynamics.


The following graph, shown with a logarithmic scale, then again with an arithmetic scale (same data), presents the weekly prevalence rates (per 100 k in the cohort) since the trough following the Winter wave. Notable is that prevalence is higher the older the age cohort fairly uniformly through these several months. Rates are remarkably low in children and adolescents, not only due to much lower rates of testing, but also lower rates of positivity.





It might have been interesting to see how quickly, and how far, rates would drop for a few more weeks (until all groups had fewer than 10 per 100k, for instance), particularly since there has been little booster vaccination or massive infection for six months and we don't know how persistent immunity is now.


Hospital admission rates

The hospital admissions series did indeed indicate a favorable epidemic context on the whole. In particular, admissions levels in the three June weeks were under 1 per 100k for "with-or-for." This compares very favorably to the low of 2.4 achieved in February 2023 for two weeks, between the Winter wave and the Spring surge (much lower peak to the surge, at 3.8, compared to 14.8 in mid-December). 




Considering the net number of hospitalized patients with Covid-19, the number has declined as admissions have slowed, but is far from cleared, particularly in some regions. The last three weeks average about 14 (per 100k) nationwide, contrasted with 36.6 for the three peak weeks in December 2022. Back-of-the-envelope: there are about 650 x 100k population (65 million); 650 x 14 = 9100, which, while the lowest since late November 2021, is still well above the 4500 in August 2020 and 6400 in October 2021 (pre-Omicron times).






There is great variation among regions' hospitalization rates. Ile de France had over 34 per 100k during the last three weeks published, whereas Pays de la Loire had under 3 per 100k. This is also true of rates of new admissions, to a lesser degree; it's less over in some places than in others.




Note that while the aggregate level for France (broader orange line) is declining, it is doing so much more gradually than after previous peaks. Will it continue to decline?  How far?  What level of hospitalization will be the stable, "endemic" state. 

It is disappointing to me that we will not have this statistic either to track the eventual resolution of this epidemic situation.

Corroboration with Excess Mortality

Eurostat reports, among other things, monthly excess mortality by country for the EU since January 2020.  The data through May 2023 were published July 14, 2023 (not all countries' data available, notably Italy not yet available).  There is an evident rise in France in April and May, which seems to correspond to the "surge" in positive test results and hospital admissions noted above. If so, June should fall close to 0 (or below) given that the situation in June was better than the situation in February in the statistics discussed above. I have no idea for July since we have no visibility on current prevalence or hospital admission rates.  

The following figure was screen-grabbed from the Eurostat page EU excess mortality above the baseline in May 2023





Tags: :Covid-19 France 2023 epidemic monitoring

Thursday, April 06, 2023

 

French SARS-Cov-2 Hospital Stats at the end of March


There were some changes to the rules effective the beginning of February:
  • Self-isolation for 5 days (or more) when positive test results is no longer expected.
  • Contacts with positive cases are no longer expected to be tested after two days.  
From March, people under 65 and not subject to the risk factors (obesity, hypertension, diabetes, etc.) need a prescription from their doctor to get a free test.  And positive results do not automatically give them medical leave to stay home from school or work. Contact tracing has ceased, as far as I know.

In a nutshell, in my opinion, we're close to operating as we were at the start of the pandemic en 2020 when testing was limited by material capacity, asymptomatic and pre-symptomatic cases are treated as "not contagious," children were considered rarely troubled by the disease. 

So, how is it going?  The number of positive test results is lower, as is to be expected when access to tests is lower; but is gradually rising in most age cohorts.  The number of new hospitalizations is well below those this season in the past two years, but is not tending toward zero as one should expect of a seasonal, annual virus (like influenza) as it is now asserted to be by public health policy wonks.  How over is it?  That depends on one's age.

Positive Test Result Rates

The first chart is in a linear (normal) scale.  The high number during last year's waves boost the scale making it hard to see what's happening recently.  The second chart show the same data with a logarithmic scale on the y-axis. As expected, daily positives are down, due in part to reduced testing.  However, they have not approached zero.  Rates are substantially lower for young children and somewhat lower for teens than for older people. Apparent rate of infection increases with age.





Hospital Patients with Covid-19

In this chart, we see the number of Covid-19 patients in the hospital at the end of each week for the two years ending 31 March 2023. I chose these age groups  to distinguish people who have work and school obligations (0-69 years) and those who don't (70 and over). 

One sees that for the younger first group, after Omicron hit in late December 2021, "surge peaks" have been lower and lower (but baseline troughs have not) and the current situation is at about the same level as what they experienced in 2021 when masks were expected, many indoor venues were closed (restaurants, bars, discos, and more) and access to others was subject to limits and "pass sanitaire" (museums, e.g.). In the current context, without the mid-2021 safeguards, these young people appear to be at little more risk than they were then.



On the other hand, the elderly continue to occupy the hospitals, at levels double or more what they experienced during the mid-2021 restrictions. Roughly, the 25% of the population 70 years and older are 2/3 of the hospital Covid-19 patients since early 2022; that is a six times higher rate of serious illness in a group of people who shouldn't be exposed.

Looking at patient counts gives an idea of who is being treated, but the length of hospital stay may distort this picture: if older patients need three weeks of care and younger patients only three days, then equal rates of admissions will register seven times more older patients at any point in time.  The next charts show admissions  per million people in the bracket rather than occupants, over the past two years, for five twenty-year age brackets. The first is in a linear scale; because the high numbers in the oldest bracket dwarf the numbers in the younger brackets, the second shows the same data with a logarithmic scale. One sees that the rates for the 60-79 group are nearly ten times higher than for the under-60s.





Better news is that intensive care needs for Covid-19 patients are much lower than during many periods in the past. Deaths, too, are much lower.







Tags: : France, Covid-19, 2023

Thursday, March 02, 2023

 

French SARS-Cov-2 Hospital Stats for February 2023




In mid-February 2023, Sante Publique France announced a data quality problem in the hospital admissions statistics they receive.  They were not very specific in saying what the problem was, but their promise that "from February 8, 2023 on the admissions data are posted to the correct date" suggests the problem was with dates. Because I also track "cumulative number released to go home" (and "cumulative number of deaths") I think there was another, possibly related problem.


The Cumulative Releases to go Home

As these two partial screen captures show, there was a big decline in the number of patients who had gone home



Normally, a cumulative count of this sort is non-decreasing unless it is to remove prior entries which should not have been included. On February 3, 846 970 patients had gone home; on February 6, only 844 630 had: over 2 000 fewer!  By February 8, the count was back up to 848 428; the pace of daily increase has been more normal since.

One reason this statistic matters to me is to compute admissions by difference, rather than rely on declared admissions.  In principle, today's patients are yesterday's patients, less those released or deceased, plus those admitted today.  This is often the same number as the declared admissions for the day, but there are occasional differences.

Daily Declared Covid-19 Patient Admissions

There was also an anomalously large number (3,419) of declared admissions on February 8, 2023, coinciding with the large number of discharged. February 8, 2023, is the date after which, according to the warning message (at top of this post), data are posted to the correct date.




How This Impacts Tracking

First, it is important to realize that there are many fewer admissions (and discharges) on week-ends and holidays.  That is why it is common to use 7-day moving totals for tracking, to have a smoother trace. A big outlier (or error) thus messes up the trace for a week.  Two, one on Saturday offset by another the following Wednesday, create confusion for over ten days, until Thursday of the following week. That is what is reflected in orange trace in the graph below. The gap between the blue trace and the orange trace indicates growing number of patients (blue above orange) or reducing number of patients (orange above blue). The number of patients declined from about 25,000 in the last week of December to only 13,000 in the first week of February, and remains at that level.


Another observation is that the green trace, the number of positive test results scaled to match the level of hospital admissions (1 in 50) and did so until early December, is now far lower.  That suggests that far fewer cases not leading to hospitalisation are being detected. That is entirely likely, since first the labs doing the testing under-reported due to a commercial conflict with the government, then contact tracing and testing ended (1 February).

Geographic Diversity

The ups and downs discussed above were also apparent in the data at department level.  I track ten departments around the country, chosen for personal reasons I won't explain. For clarity, I've excluded the data from the perturbed days, whence the broad gap. Nowhere is there a new wave growing, but there is clearly on-going activity everywhere to varying degrees.





Tags: : France, Covid-19, 2023, Hospitals


Friday, January 20, 2023

 

SARS-COV-2 in French Hospitals, 2021-2022 continued



In my previous summary of quarterly statistics for 2021 and 2022 (yesterday) we were left with a question: why were discharges to go home so much higher in 2022-Q1 when patient-days were not so different from those in 2021-Q1?

To get some clues, I looked at the same statistics for each of the ten age brackets commonly reported.  However, I did not fetch datasets of thousands of numbers (730 days x 10 brackets per statistic), I used data I had already fetched and formatted with snapshots each Friday. The "discharged to return home" should be very close, varying from the all-patients only because my Fridays may not align perfectly with the quarter boundaries, and some (few) cases may lack age data so be missing from my figures.  To get a rough estimate of duration of  hospitalisations I need number of patient-days; that is where the lack of daily data may make a difference.  I summed the thirteen Friday levels of each quarter and multiplied by 7,  as if each level applied to every day of its week.  The results are quite close to the totals in the previous post, certainly good enough for the heavily-rounded duration estimates.

discharges by age


  group       2021-Q1      2022-Q1   '22 vs '21 
---------     --------    --------   ---------
 0-9 ans          568        9,323   +1541 %
   10-19          878        4,265   + 386 %
   20-29        2,805        9,478   + 238 %
   30-39        4,727       12,089   + 156 %
   40-49        6,959        8,490   +  22 %
   50-59       12,671       12,001   -   5 %
   60-69       17,398       17,734   +   2 %
   70-79       21,345       23,207   +   9 %
   80-89       24,166       28,175   +  17 %
90 et plus     11,262       14,620   +  30 % 

Observation: much of the increase year-on-year is indeed due to more younger patients.

Does stay duration vary by age?

  group       2022-Q1
--------     ---------
 0-9 ans       <4 days
   10-19       <7 days
   20-29        6 days
   30-39       <7 days
   40-49       11 days
   50-59       16 days
   60-69       20 days
   70-79       22 days
   80-89       22 days
90 et plus     22 days

Estimation made by dividing patient-days by number of patients discharged in the quarter.

This does change from quarter to quarter, but the progression from shortest for youngest to longest for oldest remains stable--except some slightly too-high figures for the 10-19 year old bracket during the last three quarters.  That teen group is the least-present, so perhaps the longer stays are because only exceptionally ill teens are involved.




Tags: :

Thursday, January 19, 2023

 

SARS-COV-2 Hospitalisation Activity in France, 2021-2022



As 2022 came to a close, I believe it was not generally known how much Covid-19 was (and is) still occupying French hospitals after the "crisis" was declared over from 1 August 2022. Most people, it seems, have ceased trying to prevent contagion, and contagion happens, sometimes to vulnerable people who end up in the hospital (if they are lucky).

I have been tracking a few daily statistics on number of  Covid-19 hospital patients and producing a few plots so I was aware that the troughs between the peaks have been shallow --number of patients never gets very low -- since November 2021.  On the other hand, ICU occupancy is lower. We'll also look at deaths and a couple of back-of-the-envelope indicators.

First, some definitions:
  • hospital patient-days: for one day, the number of patients on that day; for longer periods, the sum of the patient counts.  Think of it as the number of nights a hotel would invoice at a per-bed rate.
  • ICU patient-days: analog for ICU patient counts.
  • Number discharged to go home: a cumulative count from a fixed start date; number within a period is the difference between the count on the day before the start and the last count.
  • Deaths: also a cumulative count from a fixed start date, same usage as discharges.
The data I used are from geodes.santepubliquefrance.fr.

Hospital Patient-days (quarterly)

                             Hospital 
                            Patient-days
2021-Q1       2,334,547
2021-Q2       1,968,283
2021-Q3         809,762
2021-Q4         895,495

2022-Q1       2,301,857      -1,40 %
2022-Q2       1,736,947     -11,75 %
2022-Q3       1,525,883     +88,44 %
2022-Q4       1,841,350    +105,62 %

Observation: there really is a lot more Covid in the hospitals in the second half of 2022.

ICU Patient-days (quarterly)


                ICU Patient-days
2021-Q1 311,976
2021-Q2 363,445
2021-Q3 145,860
2021-Q4 160,337
2022-Q1 261,587     -16,15 %
2022-Q2 112,307     -69,10 %
2022-Q3  89,380     -38,72 %
2022-Q4 101,018     -37,00 %

Observation: After the first Omicron surge in Q1, much less use of ICU, notable compared to the huge rise in the number of patients.

Deaths (in hospital)


               Hospital Deaths
2021-Q1          25,156
2021-Q2          14,214
2021-Q3           5,205
2021-Q4           6,739

2022-Q1          17,174      -31.7 %
2022-Q2           6,755      -52.5 %
2022-Q3           5,251      + 0.9 %
2022-Q4           6,459      - 4.2 %

Observation: much better outcomes than in 2021 in the first half of the year, and no worse in the second half despite much higher occupancy and less use of ICU.

Discharges to go Home

               Hospital Discharges  
2021-Q1         102,781
2021-Q2          84,751
2021-Q3          36,390  
2021-Q4          39,680 

2022-Q1         139,382      +35.6 %    
2022-Q2          79,362      - 7.3 %
2022-Q3          65,578      +80.2 %
2022-Q4          75,582      +90.5 %

Observation: The Q1 2022 performance needs some explaining. I suspect it is due to shorter stays, but I don't know why that would have been the case.  Milder Omicron? Younger patients? 

Hospital Stay Length

I know from some other tracking I do by age bracket weekly that younger patients "turn over" faster, and apparent stay duration appear to increase with age.  I don't think  I can readily check the age mix effect in 2022-Q1, however.

What I can propose is a rough calculation --rough because it supposes negligible changes to opening and closing counts--as follows: 
  • consider discharges + deaths as the number of patients whose occupancies are counted in patient-days
  • divide patient-days by this number to get an average stay duration
2021-Q1     18
2021-Q2     19
2021-Q3     19
2021-Q4     19

2022-Q1     14
2022-Q2     20
2022-Q3     21
2022-Q4     22

It does not answer the question of why, what happened, it just presents it differently.  I'll see what I can find and post it if I can.


Tags: :Covid-19, France, 2022, hospitals

Sunday, January 15, 2017

 

Phishing for Friends


This is about how impersonators may try to pass for some of the people you consider friends and acquaintances, and what I suggest one do about it, with little explanation of why they do so. Impersonation on Facebook does not imply "I was hacked" or that it is imperative to change password. It might just mean one or more of your friends is too friendly (and not wary enough) with strangers, that you are too free with information about who your friends are, or both.

I received an invitation to become "Friends" on a popular social network today, once again. I recognized the name, it was that of someone who might conceivably want to become more intimate, since we had friends and experiences in common, but why now? And I have seen enough impersonators of other friends by now to be wary: I always check out the available information on the source of the invitation before replying, now, but have not always done so. It was clearly a new account, with only two friends as yet.

It wasn't long ago that I received a similar invitation, but from someone with whom I was already connected (if one believes the name given). It was easier to check that said person was already in my contacts list, refuse the invitation, report the invitation to "the powers that be," and post a message on the original's stream to warn any other friends; the warning may not be very helpful, who knows who will see it in time (before accepting the impersonator' invitation) but one hopes. And I hope he was not actually the original trying to replace his account (he hasn't responded one way or the other, so I have a slight lingering doubt).

How does this happen? Common reaction is "I've been hacked! Must change password!!" Let's think about it. The new profile, that of the imitator, typically shows cover page and profile portrait pictures, and may not show much else, as if it is the account of someone starting over because they got locked out somehow. The imitator also has a list of people to invite to connect. The pictures are easy, on a network like Facebook they are necessarily public--anybody can see and copy them, knowing whose they are. I presume that the imitator/impersonator gets access to the list of friends of the impersonated, then sends invitations. Many of us leave our friends lists visible to at least some others, for various reasons. The friends list may not be public, that is one of the privacy settings one may tighten; one may even make it completely only-me private. Important (as stated in the sequence of texts on the page cited above)
If you’re friends with someone, you’ll be on their friend list and it’s up to that friend who sees their list.

The impersonation does not require "being hacked" or having an impersonator acquire access to one's account. It only requires an impersonator having access to a list of people to "invite" to connect.
The impersonator might, if the impersonated person keeps the list my-eyes-only or friends-only, use the friends lists of one of the friends of the impersonated (or more friends' lists, particularly if they overlap). So one is a candidate for impersonation if any or several of one's friends share their lists of friends, even if one does not, oneself. And the impersonator is probably, depending on sharing policies, a friend or a friend of a friend, and I am tempted to add "new" to friend unless there is some reason to imagine that they have been dormant and suddenly decided to start impersonating.

Why do they bother? That I find harder to say. What is clear is that with their impersonation account, they will have a view of all information their new "friends" share only with friends, be that photos, timeline posts, contact information, friends list. They will also enter the category of "friends of friends" of all those new friends, and gain access to everything other friends of those new friends share with that category. There is thus a wide range of information they might be seeking, and they might just be trawling to see what they can happen upon, and where they can extend their imposture-based web.

What to do about it as individual users is not a simple question to answer. I suggest three axes: what one shares even only with friends, one's privacy settings for information imposters might exploit, and one's vigilance when one receives invitations.

First, one should recognize the risk of making a mistake and connecting with an imposter.  What they see you cannot make them un-see.  Same as for friends with whom you later have a quarrel and end your connection.  As my grandmother used to say, twirl your tongue in your mouth some number of times before speaking (to give yourself time to recognize something better not said); same goes for sharing, especially since everything you share in digital media is recorded and impossible to deny.

Second, privacy settings for friends list, especially. One can make one's friends list private, and check that all one's friends do, too, and estrange any friends that don't: that will make it much harder for the potential imposter to know who to invite. It may be dissuasive. But one might lose a lot of friends that way, and one might--as one's friends might--have good reason to display one's friends list: to help disambiguate for people who are looking for you (my room-mate from 1975-6 I still haven't found, and his name is much more common than mine so I'm counting on him to find me) and want to be sure they have the right one. One of the things that soured the invitation today was the lack of friends; if the phisher is lucky, that may change, and snowball as people recognize the others who have already "arrived."But had the friends not been visible to me at all, I would have considered that suspicious and would have been at least as wary of the invitation.

What one can and should do is not click on "accept invitation" until one has vetted the source of the invitation. In some cases it is easy because one has communication through other channels (real life, telephone, etc.) or the invitation is apparently redundant (connection already existed). In others it may be harder, and I have failed in at least one instance when I recognized a dozen friends in the new account's list and assumed it was a replacement account (someone was hacked or locked out and began anew) rather than an imposter; I realized too late, and someone may have learned whatever I share with friends and nobody else. When in doubt, hesitate, wonder what you'll lose by waiting, think about who to ask, and then do ask.

Addendum

On Facebook, one can use the "Report" procedure, initiating it on the suspect profile page. It is accessed by clicking on the little "..." button next to"Following", "Message" in the lower right corner of the cover photo area.  For yesterday's invitation, I did so, choosing "fake" as my objection.  Facebook replied just a few hours later,
Thanks for your report - you did the right thing by letting us know about this. After reviewing the profile you reported, we've decided to follow up with the account owner directly. When we think a profile may be fake or pretending to be someone else, we ask them to confirm their identity. If we see something on a profile that goes against the Facebook Community Standards, we remove it.

Tags: :

Sunday, January 08, 2017

 

Waze and Means


I have used the Waze application (for mobile phones with the Android OS) for a few road trips.  I like its "updatedness"--timeliness, its provision of information on incidents and solicitation of revisions from later drivers. I dislike the inefficiency caused by poor mobile phone coverage, and suspect that Waze avoids routing through such areas even if they might provide advantageous itineraries.

I have been using ViaMichelin to prepare driving plans for many years; sometimes I would print out the whole long itinerary, sometimes I would prepare (and print or hand copy) my own distilled version of the key turns and changes of direction. That mostly worked well. However, there were a couple of times their route instructions referred to signs which did not match those we saw (I had trouble navigating one junction with a competent navigator reading the instructions to me), and they tend to use street names for which one has difficulty finding signs, if there are any. Their instructions through Besançon pretty much always leave me lost in the middle of town wondering which way to head out. Once I pulled in to a bakery's parking lot and asked the first person who came by for directions; her first response was "show me your itinerary," presuming I had one (ViaMichelin or Mappy or something else) because, well, one should have a computer-issued itinerary, at that time in the progress of technology. Then one of the last times I tried to get through Besançon with a pre-calculated itinerary, I hailed a couple of young men I supposed competent to answer my question about where next to turn while we were waiting at a traffic light; they suggested we pull over to discuss it, and I accepted. They did not rob me, at knife point or otherwise, they indicated a right way to go, and suggested I buy I GPS navigation aide.

I don't consider a GPS navigation aide worthwhile for the little I travel. But when I was buying a next car, I did not reject one just because it had been equiped with a Garmin GPS navigation aide; nor did I commit to buying the Garmin map updates. For the few automobile trips I make to places I haven't been I can buy paper maps from IGN and others.

I used the Garmin GPS navigator on a trip a few months ago. I was annoyed by its insistance on using theoretically faster roads even if that meant a detour and a toll to pay. I later learned how to set it to avoid toll roads, but not how to accept them for a worthwhile time savings. It did not have real-time traffic information. It was no better than ViaMichelin for traversing a town center like that of Carpentras, which I think took me three loops (twenty minutes or more) to succeed. And then it took me up a mountain to a closed road, then around and down through a tourist-crowded village. I'm inclined to use it for details when close, not for choice of longer distance roads and routes.


The application Waze for "smartphones" equipped with GPS offers an alternative to devices like the Garmin navigator/navigon. It has the advantage of enabling users to annotate current conditions, providing reports of vehicles on the shoulder, dead animals on the road, mobile radar monitors, congestion, and so on. Or, for those who pass later, indicating whether or not the condition is still the case. And not asking for money.
Waze benefits from the interaction with its users, whether that be to monitor their progress and deduce driving conditions, or to provide a set of notifications of distractions and dangers.
Waze has a problem with areas having poor or no mobile phone coverage. It cannot provide information about current conditions if it cannot receive bulletins, which is understandable, but worse, it seemingly cannot track one's progress with GPS-only data, it needs to check back via a phone/data link to get server-side comments and recommendations. When coverage comes and goes, it may not "know" whether one is on the right road, may beep frequently while recommending to get on some road or other--which may be the one one is currently on. {comment from 2015)
To avoid this inconvenience, Waze may well--I would--avoid recommending routes through areas with poor mobile phone coverage, so as to prevent disappointment and frustration of users who expect constant tracking of their progress and next instructions. But then how does one navigate from Pirmasens to Niederbronn or Bitche?
Probably, Waze will transfer more data and software to the phone to navigate seat-of-the-pants and log and take notes and feed back recommendations, decisions, and outcomes later. But that is just a guess. Tags: :

This page is powered by Blogger. Isn't yours?