Tag Archives: farming

Everybody knows about confounder but what is a collider?

Here is the best explanation of a collider written Julia Rohrer at www.the100.ci

Whenever X1 (conscientiousness) and X2 (intelligence) both cause Y (college attendance) in some manner, conditioning on Y will bias the relationship between X1 and X2 and potentially introduce a spurious association (or hide an existing link between X1 and X2, or exaggerate an existing link, or reverse the direction of the association…)

The cartoon makes it even clearer – confounder act on exposure and outcome, while collider condition on exposure and outcome.


The bullshit asymmetry principle

Really, did not know that, but agree immediately to the fact, that the amount of energy required to refute bullshit is an order of magnitude bigger than to produce it.

The kinds of people who bullshit are more likely to be the kinds of people who misrepresent evidence, avoid correcting their errors, and intimidate dissenters, so at some point the people who could shoot down the bullshit might decide it’s not worth the trouble: Why bother fight bullshit if the bullshitters are going to turn around and personally attack you? From this standpoint, once bullshit becomes “too big to fail,” it can stay around forever.

The bullshit asymmetry principle

Can criticial thinking been teached?

Yes, it can.

Already in 2017 there was a Lancet paper with the super-long title “Effects of the Informed Health Choices primary school intervention on the ability of children in Uganda to assess the reliability of claims about treatment effects: a cluster-randomised controlled trial”. The paper is extensively discussed at vox.com

Andy Oxman is obsessed with the study of bullshit health claims and how to prevent them from spreading.
For decades, he’s been trying to find ways to get adults to think critically about the latest diet fads, vaccine rumors, or “miracle cures.” But he realized these efforts are often in vain: Adults can be stubborn old dogs — resistant to learning new things and changing their minds.

So not only Germany but also Uganda has its own bullshit hypothesis.

Distance calculations are not plausible

Rompa et al. claim that

Asthma and atopy were inversely associated with presence of a farm within a radius of maximum 100m.

wich refers to their FIG 3

FIG 3 (click for zoom)
Original legend: Number of farms within a given radius was dichotomized for at least one farm in the radius versus no farms. Associations are calculated by logistic regression resulting in odds ratios (OR) with 95%- confidence intervals.

The authors probably want to say that a non farm child that lives within a smaller distance to a farm shows a stronger negative association. Unfortunately it is not clear from the methods how the categories have been exactly defined, including or excluding category borders? And why is the highest category of 1.000-10.000 excluded here? There is a negative association with asthma in ALL strata irrespective of distance – what is the reference? The increased atopy risk by a farm in a circle with 1.000m radius is never discussed. BTW I am also quite sure that this plot has been produced with some kind of drawing software and not with “R version 3.2.3” if you zoom into the picture.

For whatever reason the authors abandon the distance definition above in favor of some data-derived classification afterwards. Why?

[1] children living on a farm currently run by the family [2] children not living on a farm, but with regular contact to farms, meaning at least once a week for a period of 6 months minimum and [3] children without any contact to farms.

I have no idea what regular contact is. But lets have a look on FIG 2

FIG 2: also misaligned and implausible
Original legend: Distribution of environmental variables across exposure strata 1 = farm children, 2 = exposed non-farm children, 3 = non exposed non-farm children. Differences between the subgroups were statistically significant (p<0.01) except for ozone, altitude and distance to the next farm.

This is also not a facet_warp() as I would expect  from a R analysis but some manually cut & pasted figures where P<0.01 is contradicting the methods. FIG 2.9 basically says that the distance to the next farm is not different in groups of “farm exposure”.

The abstract

The environmental variables greenness, tree cover, soil sealing, altitude, air pollution differed not only between farm and non-farm children, but also between farm children with and without another farm nearby.

is therefore wrong (it maybe even trivial as farms of course have less soil sealing than villages). In any case, we are now trapped in a loop as according to Fig 1 the prevalence of asthma and atopy over the exposure strata should have been different.


Schutz vor Allergien auf dem Bauernhof?

Keine Frage, die Lebensbedingungen auf Bauernhöfen sind anders. Mehr Tiere, mehr Dreck, mehr frische Luft, vieles ist anders als in der Großstadt. Dass es hier auch weniger Allergien gibt, wird wohl an den Bedingungen liegen, die bereits an einer der ersten Studien vor 30 Jahren zu sehen waren.

Clin Exp Allergy. 1999 Jan;29(1):28-34. https://doi.org/10.1016/S0140-6736(01)06252-3 Hier nicht relevante Daten ausgegraut

Auch die Eltern hatten schon weniger Allergien. Der “protektive” Effekt kann also einfach dadurch erklärt werden, dass mit weniger Eltern als “Risikofaktor” auch weniger Kinder Allergien haben. Und warum die Eltern wohl weniger Allergien haben? Nun ja, mit Heuschnupfen wird man nicht gern im Heu arbeiten wollen. Auch das zeigen Studien ziemlich eindeutig.

Wenn man genau hinschaut, dann haben alle Bauernhofstudien immer wieder dieselbe Argumentationsstruktur: weil die Bedingung X dort so ist, dann kann die Folge Y auch auf die Bedingung X zurückgeführt werden. Allerdings machen immer mehr Beschreibungen von X die Story nicht glaubwürdiger.  Keine der jemals beschriebenen Bedingungen X, ist aus der Bauernhofsituation auf eine allgemeine Situation übertragbar gewesen, von einem einzigen verunglückten Versuch abgesehen.

Hier die Story mal erläutert an einem simulierten Datensatz – eine normal verteilte Allergiehäufigkeit und normal verteilte Endotoxinwerten. Zwischen beiden Variablen gibt es keine Korrelation.

Ausgangssituation: Jedes 10. Kind hat eine Allergie. Es gibt keine Korrelation zwischen Allergiehäufigkeit und Endotoxin.
Wir beginnen nun eine Studie im ländlichen Raum (dunkelgrün) und sind dabei vor allem an den Bauernhöfen interessiert (hellgrün), also nur der Region mit hoher Endotoxinbelastung.
Hier läuft aber schon seit längerem eine Wanderungsbewegung. Wer Heuschnupfen hat, wird nicht Heu machen können..
Somit fallen in dem oberen rechten Quadranten Allergiker weg und verteilen sich im dunkelgrünen oder grauen Bereich. Müssen nicht viele sein, 30% weniger reichen schon.
Wenn wir nun erneut eine Regressionsgleichung aufstellen, so gibt es eine negative Assoziation im ländlichen Bereich (kurze Linie) während in der Bevölkerung insgesamt die Verschiebung nicht besonders in das Gewicht fällt  (lange Linie).

Das ist nun genau das Ergebnis der Bauernhofstudien.


Natürlich kann ein hoher Endotoxin Spiegel auf den Bauernhöfen eine bestimmte Wirkung haben –  zumindest bei einigen Menschen und bei einigen Mäusen – aber mehr ist nach aktuellem Kenntnisstand sehr unwahrscheinlich.

Da die Lebensbedingungen auf dem Bauernhof angeblich protektiv sind, müsste es eigentlich Kinder geben, die eine Allergie haben müssten (zB mit doppelter Familienanamnese) aber nun keine Allergien bekommen hat. Solche Kinder gib es aber nicht, weil auch schon die Eltern keine Allergien hatten.

What is wrong with the 2011 NEJM paper?

N Engl J Med 2011;364:701-9 is another paper with 1000+ citations  that had a lasting impression on some but not all people.

First, I can’t remember of any study with such an enormous selection bias  where >94% of individuals have been lost.

Second, we should not forget that farm is not protective per se – farmers may just avoid a known allergy risk factor.  PARSIFAL participants in this study included Steiner schools — anthroposophic medicine mostly avoids vitamin D (ref). This is of course a major issue for any cross-sectional study that doesn’t take into account the temporality of events.

Third, in PARSIFAL dust from children’s mattresses were collected by vacuuming — it is not very likely that many helminthic eggs were transported  from stable to bedroom. In GABRIELA, only airborne dust samples  were collected which again may miss helminth eggs although being certainly present in stable dust.

Fourth, more  microbial exposure and more fungal taxa on farms are a trivial finding.

The inverse associations of the diversity scores with asthma were not confounded by status with respect to living on a farm because adjustment did not change the respective point estimates for asthma (Table 2), although the associations became nonsignificant.

Small sample size, borderline p-values even after a long fishing expedition?

What do these strange “probability” plots  really show – the probability of asthma or the probability to live on a farm?

N Engl J Med 2011;364:701-9 Figure 3 Does it refute any general effect of diversity?

The plots are misleading if adjustment for farm living does not change the parameter estimates for bacterial/fungal diversity.

Sixth – even many years later, the main findings of this study have not been independently replicated. There is not any single study that shows listeriosis (Listeria) or diphtheria (Corynebacterium)  to be protective.

Allergy protection on farms – why also studies in mice could have failed

There are  many immunological differences in humans and mice (follow my link) that are never discussed — not even in Science 2015;349/6252:1106.

Ignoring the long-standing paradox that endotoxin is also acting as a natural adjuvant to atopic inflammation, the credibility of the Science paper is further reduced.

IMHO it is also a  strange experimental condition to have all animals on a standard vitamin D diet – a known co-sensitizers – and looking then for A20 which is co-regulated by vitamin D ??

And here is the last argument – more than 90% of studies in mice fail to translate into humans.

Does a healthy worker effect explain the “allergy protection” at Bavarian farms?

Unfortunately most studies in the farming environment did not report the prevalence of parental history nor did they report the effect size of parental risk in the farming population. This is, however, a critical issue as the so called healthy worker effect (HEW) may be a rather trivial explanation of the results.

Specifically, it is a sampling bias: the kind of subjects that voluntarily enroll in a clinical trial and actually follow the experimental regimen are not representative of the general population. They can be expected, on average, to be healthier as they are concerned for their health [or as ill people already dropped out]

At least Braun-Fahrländer 1999 reported that allergic parents were seen much less at farms: allergic rhinitis 12.7% versus 29.4% (P=0.001). A history of allergy at farms is no more a risk factor as it is otherwise in the general population. How that?

Are there any studies in adults? I know of three studies (and one review Le Moual N 2008).

Leynaert 2001 showed only a slightly reduced prevalence of “allergy” (39.1% vs 41.5%, NS) while her table 4 is most interesting. The association started only after year 1960 which points towards misclassification as far as the analysis is not stratified by year of birth.

Remes 2002 showed a dose dependent effect decline between farming (36.2%) and controls (31.6%, P=0.075),

Perkin 2006 also found some significant lower prevalence in farmers 47.3% versus 57.7%, P<0.001.

A HWE is therefore likely.

There are six studies (Thelin 1994, Braback 2006, Chenard 2007, Thaon 2011, Elholm 2013 and Spierenburg 2015) that examined in detail a possible relationship of HWE, allergy and farming. Unfortunately the examination period in five of these studies is too short for any conclusion while Braback 2006 seems to be the only reliable study.

Source: Braback 2006

Also from this study, we can safely conclude, that there is a significant HWE.


Addendum 22 Nov 2019

It seems that I missed some papers on HWE and farming. For asthma it is important to discriminate atopic and non-atopic asthma.

Timm 2019: a hard to understand 3 generation study of unclear asthma  type. Point estimates of parental asthma on farm upbringing are not really a measure of HWE – shuffling exposure and outcome distorts temporality. Neither protection by farm nor HWE found but also selection bias (approx. 50% dropout) and non-differential bias for residency.

Eduard 2015: 1,964 farming students. Current asthma in farmers was 3.0% compared to 6.3% in farmers who had retired early. No HWE found at entry level but during farm work.

Farmers who had changed the type of production had an increased risk of current asthma (OR 9.8), compared with those who had not changed production, indicating a strong selection effect.

Vogelzang 1999: 400 pig farmers, X-sectional point estimates, not a  real HWE study, although HWE offered as explanation.

Health-based selection of nonasthmatics for pig farming, which tends to mask a work-related hazard for asthma, is offered as an explanation for these results.

Taken together – more data but not more knowledge.


Addendum 21 Feb 2021

A new farming paper shows parental history basically same vein as Braun-Fahrländer 1999. If we could get also paternal history, this would explain the whole farming effect.




Claim to fame of the hygiene hypothesis

The recent encyclopedia article about the hygiene hypothesis seems to be well written. At least on the first instance … in reality it is more a novel than a scientific review.

For many years already, the hygiene hypothesis has been called an outdated concept; various times it was revised and transformed, and finally it gave birth to novel hypotheses.

In other words, the hypothesis has been rejected for being wrong . Even many revisions did not change that. There seem to be only one proven fact – the obsession of some authors with hygiene and nouvel Rousseauism.

Anyway, the hygiene hypothesis has promoted radical rethinking of infections, microbiota, and coevolution of mankind and microbes.

There is nothing radical in backward thinking. We still carry tons of microbes, freezer and antibiotics only did some qualitative but not so much quantitative changes,

With the advent of novel high-throughput sequencing technologies the human microbiome, which is sometimes called the ‘forgotten organ,’ has attracted much attention and is currently being implemented in a wider concept of self-foreign relationship, which may even include recognition of the nonmicrobial nonself as a vital stimulus to a well-developing immune system.

  1. So the interest is technology and not science driven.
  2. The microbiome is not an organ.
  3. The hype is already over.
  4. The Self is not defined by any bacterium.
  5. Most bacteria are excreted and not vital stimulus.

Given the many molecule classes regulating immune functions across individuals such as short RNAs, the hygiene hypothesis may eventually come back as a surprising explanation of the phenomena evoked by crowding, day care, sibship size, orofecally transmitted diseases, and respiratory infections.

Why that?
A comeback of the hygiene hypothesis by short RNA?
The listed phenomena are not intrinsically related, but are occuring only at the same time scale.

Even the old birth order effect might be rediscovered as epigenetic programming someday. Admittedly, these notions are entirely hypothetical, but without hypotheses, proven or not, science hardly advances.

So if David Strachan’s birth order effect would be really caused by  epigenetic programming – why would that be related to hygiene at all?

Science is is not so much about proven or unproven but about reasonable and non reasonable hypotheses.



ArXiv is operational since 1991, bioRxiv since 2013 and since 2019 there is now also medrxiv. More details  at https://www.bmj.com/content/365/bmj.l2301

The main arguments in favour of sharing work in its preliminary form are, firstly, that science works faster if work is made available sooner after it is completed and, secondly, that articles are improved by feedback from a wider group of readers, alongside formal peer review by a few experts. Simple estimates suggest that halving the delay to sharing a research result can double the speed at which research progresses. Ambitious research funders are now embracing preprints and other measures that aim to accelerate the pace of research.

Although there was a mixed reception in the beginning, see Science back in 2017

MedArXiv will have a hard time attracting preprints if mainstream medical journal editors decide they won’t publish final versions of the papers. Currently, The BMJ and The Lancet are among the few medical journals that have explicitly said that posting a preprint doesn’t preclude publication; Nature and Science, which both occasionally publish medical studies, have the same policy. But at the JAMA Network, which publishes a dozen journals, the issue is hotly debated.

@medRxiv opened on June 6. So far they have only 304 followers on Twitter (and no allergy paper in the archive).

As the current “Allergy” editor and the publisher (John Wiley and Sons A/S)  agreed to preprints last week, I have submited now my first preprint paper. Therefore, there are now 305 followers and 1 allergy paper :-)

The endotoxin phantom – discrepancies in the NEJM 2002 farm paper

N Engl J Med 2002; 347:869-877 made a strong point that the farm effect is mediated by endotoxin but could show only a 1,7 fold higher endotoxin exposure at farms.

N Engl J Med 2002; 347:869-877 table 1 shows environmental exposure between farm and non-farming households

LPS therefore will not explain the negative farm association as endotoxin effects are so much similar between farming and non-farming environment. BTW why is only the result of the total sample given here and not just the farm result?

N Engl J Med 2011; 364:701-709 table 2 endotoxin effect is not stronger at farms – more some unspecific immunosuppressive effect?

So, what is the reason for the lower IL10 capacity in Figure 2D after LPS exposure – exhaustion, adaptation? And even more important: What is the reason for the lower allergy rate at farms?

If we now go back to table 1 there is a 5,0 fold Der p1 excess in farm – much higher than the 1,7 fold increase of endotoxin. As nearly all allergens have helminth homologues –  the question is what did the authors really measure? Da Costa Santiago 2015 has a nice table that could give an answer.

J Immunol. 2015 Jan 1;194(1):93-100 shows Der p1 homologs

Der p1 is a cysteine protease of 25kDa that has homologues for example in Loa loa. Unfortuneately Loa loa is not an explanation at Bavarian farms – as it is more common in tropical Africa. Cysteine proteases are nevertheless a big tool for helminths –  in Schistosoma japonicum a cathepsin B2 cysteine protease is considered the main penetration tool.

Usually cysteine proteases are not allergenic, but the excessive (and rather isolated) rise combined with a reduced Der  p1 sensitization in the children,  is definitely an unusual finding. The Dermatophagoides pteronyssinus habitat is cosmopolitan, house dust,  also influenced by altitude, but there is no known reasons for this excess in farms. Is the Der p1 value just a false positive and has it helminthic and not mite origin?

I am therefore blasting Der p1  sequence (P08176) against all known cow endoparasite genomes. Fasciola has a strong hit with CL6, a papain family cysteine protease with e+48.

Wormbase blastp results – e+ values are somewhat lower than my local blast result at the same locus

With identity values between 40% and 70% there could well be a confusion of mite Der p1 and  helminth (Fasciola?) CL6, which should be tested for cross-reactivity.

The story even gets more interesting when reading more about parasites

Helminth parasites have complicated life cycles … at the same time as skewing the immune system toward a Th2-driven response, they have a general suppressive effect on the host immune system that prevents their elimination and reduces immune-mediated tissue damage. It has been suggested that cytokines of the anti-inflammatory network, particularly IL-10 and transforming growth factor- (TGF-), that are produced in response to continual stimulation of the immune system by parasite antigens, are pivotal to regulating the damage they cause and that, coincidentally, these have a bystander protective affect against allergic reactions.

This is exactly what N Engl J Med 2002; 347:869-877 showed: increasingly exhausted IL10 capacity. May the x-axis label should be CL6 instead of LPS?

N Engl J Med 2002; 347:869-877 Figure 2. Smoothed Plots of the Log-Transformed Capacity of Peripheral-Blood Leukocytes (PBL) to Produce Interleukin-10 (Panel D) after Stimulation with Lipopolysaccharide (LPS) or Staphylococcal Enterotoxin B (SEB) in Relation to the Log-Transformed Endotoxin-Load Values.



The Amish paradox in NEJM 2016 explained

A recent study in the NEJM found remarkable differences in the asthma prevalence between Amish and Hutterite populations. The lifestyle of both communities is similar but their farming practice is distinct as the Amish follow a more traditional style of outdoor grazing whereas the Hutterities use industrialized farming practices. Gene expression data in the Amish children have been interpreted as „intense exposure to microbes“ because protection of experimental asthma by Amish derived house dust was nearly abrogated in mice deficient for MyD88.

Any helminth exposure has been excluded due to low IgE and eosinophil counts in the children while I still think that this could be an explanation in particular as the attempt to show an effect of bacterial exposure was unsuccessful since the discovery of the farming effect.

One difference between conventional stable (Hutterites) and outdoor grazing (Amish) is the higher helminthic infection rate on pasture, mainly with Fasciola, Ostertagia, Eimeria, Cooperia, Dictyocaulus and Trichostrongylos species.

Infected cattle rarely demonstrate clinical disease, while it is known that Fasciola (as for example Schistosome) has numerous immunosuppressive functions in the host. IgE is not always raised as Fasciola can degrade human immunoglobulin or even induce eosinophil apoptosis.

Re-analysis of Gene Expression Network using string-db.org (String Consortium 2019). The gene expression network in Amish children {Stein et al., 2016, #73074} in the upper area has similarities with the network observed in sheep after Fasciola infection {Fu et al., 2017, #6751} module #1 and #3, in the lower plot.