on 24-Jan-2020 (Fri)

The human host serves as a natural reservoir for M. tuberculosis. The ability of the organism to efficiently establish latent infection has enabled it to spread to nearly one-third of individuals worldwide

NATURAL HISTORY OF INFECTION — Inhalation of aerosol droplets containing M. tuberculosis with subsequent deposition in the lungs leads to one of four possible outcomes:

● Immediate clearance of the organism

● Primary disease: immediate onset of active disease

● Latent infection

● Reactivation disease: onset of active disease many years following a period of latent infection

Among individuals with latent infection and no underlying medical problems, reactivation disease occurs in approximately 5 to 10 percent of cases [2-4]. The risk of reactivation is markedly increased in patients with HIV and other medical conditions [5,6]. These outcomes are determined by the interplay of factors attributable to both the organism and the host

Among the approximately 5 to 10 percent of infected individuals who develop active disease, approximately half will do so within the first two to three years following infection.

The tubercle bacilli establish infection in the lungs after they are carried in droplets small enough to reach the alveolar space (5 to 10 microns). If the innate defense system of the host fails to eliminate the infection, the bacilli proliferate inside alveolar macrophages, which may migrate away from the lungs to enter other tissue

If the bacterial replication is not controlled, the tubercle enlarges and the bacilli enter local draining lymph nodes. This leads to lymphadenopathy, a characteristic manifestation of primary TB. The lesion (called Ghon focus) produced by the expansion of the tubercle into the lung parenchyma and lymph node enlargement or calcification together comprise the Ranke complex [12].

The bacilli continue to proliferate until an effective cell-mediated immune (CMI) response develops, usually 2 to 10 weeks following initial infection; this occurs in more than 90 percent of exposed infected individuals

In the lung, failure of the host to mount an effective CMI response and tissue repair leads to progressive destruction of lung. Tumor necrosis factor (TNF)-alpha, reactive oxygen and nitrogen intermediates, and the contents of cytotoxic cells (granzymes, perforin), which function to eliminate M. tuberculosis, may also contribute to collateral host cell damage and the development of caseating necrosis

Caseous necrosis is frequently associated with TB but can also be caused by other organisms, including syphilis, histoplasmosis, cryptococcosis, and coccidioidomycosis. (See related topics.)

Chronic disease is characterized by repeated episodes of healing by fibrotic changes around the lesions and tissue breakdown. Complete spontaneous eradication of the bacilli is rare

Reactivation disease — Reactivation TB results from proliferation of a previously latent bacteria seeded at the time of the primary infection. Among individuals with latent infection and no underlying medical problems, it has generally been estimated that the lifetime risk of reactivation disease after an index infection is 5 to 10 percent, with a 5 percent risk in the two to five years following infection and another 5 percent risk over the remaining lifetime [4]. A study from Australia of close contacts of TB cases reported a risk of 11 percent in the five years following infection; the risk rose to 14.5 percent after adjusting for death and loss to follow-up [14]

Immunosuppressive conditions associated with reactivation TB include:

● HIV infection and AIDS

● Chronic and end-stage kidney disease

● Diabetes mellitus

● Malignant lymphoma

● Corticosteroid use

● Inhibitors of TNF-alpha and its receptor

● Diminution in cell-mediated immunity associated with age

● Cigarette smoking [6,15,16]

The disease process in reactivation TB tends to be localized (in contrast with primary disease); in general, there is little regional lymph node involvement and less caseation. The lesion typically occurs at the lung apices, and disseminated disease is unusual unless the host is severely immunosuppressed.

Prior TB infection, contained as latent TB, confers some protection against subsequent TB disease [17]. One review evaluating 23 paired cohorts (total more than 19,000 individuals) noted that individuals with latent TB had 79 percent lower risk of progressive TB following reinfection compared with uninfected individuals [18].

The term "remote infection" refers to infection that progresses to active TB after ≥2 years [21]. Worldwide, particularly in high-burden countries, most clinical TB reflects recently transmitted infection (<2 years prior to development of active disease) [21]. Therefore, it has been proposed that treatment of latent TB infection be prioritized for those who are recently infected.

In low-burden countries, a higher proportion of TB cases occurs due to remote infection among foreign-born residents [22-24]. As an example, in the United States in 2016, a majority of foreign-born persons who developed TB had resided in the United States for more than five years before diagnosis [25].

M. tuberculosis belongs to the genus Mycobacterium, which includes more than 50 other species, often collectively referred to as nontuberculous mycobacteria. TB is defined as a disease caused by members of the M. tuberculosis complex, which include the tubercle bacillus (M. tuberculosis), M. bovis, M. africanum, M. microti, M. canetti, M. caprae, M. pinnipedii, and M. orygis [28,29].

Cell envelope — The cell envelope is a distinguishing feature of the organisms belonging to the genus Mycobacterium. The mycobacterial cell envelope is composed of a core of three macromolecules covalently linked to each other (peptidoglycan, arabinogalactan, and mycolic acids) and a lipopolysaccharide, lipoarabinomannan (LAM), which is thought to be anchored to the plasma membrane [30]. The outermost layer, the mycobacterial outer membrane (MOM), consists of a lipid bilayer structure [31].

Mycolic acid, a beta-hydroxy fatty acid, is the major constituent of the cell envelope, accounting for more than 50 percent by weight; this structure defines the genus. Glycolipids are attached to the outside of the envelope layer through a connection to the mycolic acid layer; proteins are also embedded in this cell wall complex. Glycolipid components are implicated in "cord formation," whereby TB bacilli clump together forming a serpiginous structure seen on microscopy [32].

Staining characteristics — The cell wall components give Mycobacterium its characteristic staining properties. The organism stains positive with Gram stain. The mycolic acid structure confers the ability to resist destaining by acid alcohol after being stained by certain aniline dyes, leading to the term acid-fast bacillus (AFB).

Microscopy to detect AFB (using Ziehl-Neelsen or Kinyoun stain) is the most commonly used procedure to diagnose TB in the world, especially in developing countries with limited laboratory capacity. A specimen must contain at least 10⁴ colony forming units (CFU)/mL to yield a positive smear [33]. Microscopy of specimens stained with a fluorochrome dye (such as auramine O) provides an easier, more efficient, and approximately a 10-fold more sensitive alternative. However, microscopic detection of mycobacteria does not distinguish M. tuberculosis from nontuberculous mycobacteria.

Growth characteristics — A distinguishing feature of M. tuberculosis is its slow growth rate. In artificial media and animal tissues, its generation time is about 20 to 24 hours (as opposed to 20 minutes for organisms such as Escherichia coli).

Isolation in the laboratory — Artificial media used to cultivate M. tuberculosis include potato- and egg-based media, such as Middlebrook 7H10 or 7H11, or albumin in an agar base, such as the Löwenstein-Jensen medium [34]. A liquid medium, such as Middlebrook 7H9, is used for subcultures and for propagating the bacillus to extract DNA for molecular diagnostic and strain-typing procedures [35]. Three to four weeks are required to recover the organism, depending on the initial quantity of organisms in the specimen.

Identification of the organism — Once the organism is isolated, identification is based on morphologic and biochemical characteristics, although nucleic acid–based detection methods have obviated many of the conventional tests. M. tuberculosis is identified by its rough, nonpigmented, so-called "corded" colonies on albumin-based agars. It is typically positive in the niacin test, has a weak catalase activity, which is inactivated at 68ºC, and reduces nitrate [34]. (See "Diagnosis of pulmonary tuberculosis in adults".)

The only other major slow-growing Mycobacterium that is niacin test–positive is M. simiae. Although all members of the mycobacteria species produce niacin (usually undetectable by the niacin test), the differences in the activity of the enzymes involved in the salvage pathway of nicotinamide adenine dinucleotide (NAD) biosynthesis in M. tuberculosis determines niacin positivity in M. tuberculosis. Niacin accumulates in M. tuberculosis because nicotinamidase, which converts nicotinamide to niacin, is several-fold more active, and the enzyme that recycles niacin to produce NAD is less active than in members of most other mycobacterial species [43].

The niacin, nitrate reductase, and catalase tests are the three biochemical tests most frequently used to distinguish M. tuberculosis from other mycobacterial species [34]. Tests for pyrazinamidase production as well as susceptibility to thiophen-2-carboxylic acid hydrazide (TCH) will distinguish M. tuberculosis from M. bovis, another member of the M. tuberculosis complex. M. bovis does not express pyrazinamidase (or nicotinamidase) and is susceptible to less than 5 mcg/mL of TCH [34,44]. Clinical isolates of M. tuberculosis lacking pyrazinamidase activity have been described that contain nucleotide point mutations in the gene (pncA) that encodes pyrazinamidase; these isolates are resistant to pyrazinamide (PZA), one of the first-line drugs used to treat TB [45]. (See "Epidemiology and molecular mechanisms of drug-resistant tuberculosis".)

Many additional so-called virulence factors have been identified by molecular biology techniques. M. tuberculosis "virulence factors" are often defined as bacterial products whose disruption (based on comparison with wild-type M. tuberculosis) leads to diminished ability of the mutant to attach to or enter mammalian cells in vitro, decreased growth in vitro in artificial medium or inside mammalian cells, attenuation in an animal infection model, decreased ability to induce cytokines associated with disease in macrophages infected ex vivo, inability to evade host immune effector molecules, and inability to induce changes (eg, inhibition of metabolism) in cellular targets outside of itself [57,58].

The significance of M. tuberculosis entering epithelial cells in TB pathogenesis is not clear; most studies examining M. tuberculosis–mammalian cell interactions focus on professional phagocytic cells (macrophages and dendritic cells). It should be noted, however, that the initial site of lung infection by M. tuberculosis is the alveolar air space, which is composed of type I and type II pneumocytes; these are epithelial cells. Type I cells comprise about 96 percent of the alveolar surface area; type II cells cover about 4 percent of the surface area but comprise 60 percent of all the alveolar epithelial cells [69]. Thus, M. tuberculosis is most likely to encounter and enter pneumocytes before they are taken up by alveolar macrophages. Little is known about the in vivo interaction of M. tuberculosis with alveolar epithelial cells.

As noted below, lipids are a crucial carbon source of energy in vitro and in vivo and hence for M. tuberculosis' long-term survival inside a host. (See "Immunology of tuberculosis".)

Differences in structure, composition, and metabolism of the cell wall lipid molecules contribute to differences in clinical outcome in mammalian hosts [79]. In vivo, during chronic phase of infection, M. tuberculosis metabolizes lipids rather than carbohydrates [56,80]. This shift occurs via a bypass system (glyoxylate shunt) in the Krebs cycle when carbon substrates for glycolysis become limited. The enzymes involved are isocitrate lyases 1 and 2 (ICL1/2), which allow the utilization of fatty acids as a sole carbon source [81,82]. (See "Immunology of tuberculosis".)

Factors associated with intracellular and in vivo fate of M. tuberculosis — In mice, M. tuberculosis can be observed inside alveolar macrophages and dendritic cells in the lungs about 14 days after aerosol infection [99]. M. tuberculosis enters these cells after binding to a variety of receptors on these cells, including C-type lectin receptors (mannose receptor, DC-SIGN), scavenger receptors, and complement receptors [100]. It is believed that the engagement of certain receptors can determine the intracellular fate of M. tuberculosis.

The cyclic di-nucleotides (CDNs) represent another group of PAMPs that serve as targets of the host cytoplasmic surveillance pathway [115,116]. Bacterial CDNs (cyclic di-AMP and cyclic-di-GMP) bind to stimulator of interferon genes (STING) and induce type I interferons (IFN-alpha/beta) which elicit autophagy, an innate immunity mechanism important for controlling intracellular infection [117,118]. M. tuberculosis produces cyclic-di-AMP from ATP or ADP [119]. M. tuberculosis that over-expresses cyclic di-AMP can activate the interferon regulatory factor pathway to generate increased amounts of IFN-beta, which leads to macrophage autophagy [120].

Once M. tuberculosis establishes an infection, an important pathogenic feature is the ability of the organism to establish latent infection, which can then give rise to reactivation disease. Since reactivation TB is the most common form of the disease, bacterial factors associated with latency and transition from latency to active disease are important virulence factors.

M. tuberculosis has a particular predilection for the lungs. In immunocompetent mice, virulent strains of M. tuberculosis grow progressively in the lungs but not in the spleen or liver [131]. Even in severe combined immunodeficient (SCID) mice, M. bovis BCG (a relatively avirulent [vaccine] strain) grows faster in the lungs than in other organs [132]. In the mouse model, fewer organisms are required to establish a lung lesion by the inhalation than by intravenous challenge [131]. None of the other pathogenic Mycobacterium species appears to share this tissue tropism. The factors associated with M. tuberculosis that facilitate this unique characteristic are unknown.

Metabolic factors associated with bacterial survival — An in vitro model has been developed that mimics the physiologic state of M. tuberculosis during latency in vivo [135]. When M. tuberculosis is grown under microaerophilic conditions, a state called nonreplicating, persistent (NRP1) is produced and glycine dehydrogenase activity is induced. In contrast, growth under anaerobic conditions produces a state called NRP2 in which glycine dehydrogenase activity decreases, but the organism still survives as long as the loss of oxygen occurred slowly and it passed through the NRP1 stage for a period of time. When oxygen is reintroduced to organisms grown anaerobically, the pathogen goes out of the NRP2 state. Such an in vitro system could potentially be used to examine differential gene expression and thus to identify bacterial factors specifically required under these growth conditions.

The above studies suggest that bacterial latency is associated with a hypoxic state in the host. Using a whole genome microarray, a large number of genes that are induced under defined hypoxic conditions were identified [142].

Differences in virulence of clinical isolates — Genotyping of M. tuberculosis isolates has demonstrated a number of clades that account for a large proportion of new TB cases in different geographic regions, suggesting that such strains may be more virulent than others. M. tuberculous lineages have been associated with clinical manifestations of disease. In one study, for example, East Asian lineage was less likely to be associated with extrapulmonary TB than Euro-American, Indo-Oceanic, or East-African Indian lineage [145]. Improved understanding of the role of MTB lineages may provide insights into pathogenicity, infectiousness, progression from infection to active disease, and, perhaps, response to treatment.

Another strain called PG004, responsible for a large cluster of TB cases in one northern California community, was not resistant to these effector molecules. Instead, in mice, this strain produced relatively mild lung disease, in which loosely organized granulomas apparently failed to limit the spread of infection and allowed the escape of M. tuberculosis into alveolar air spaces [63]. This study suggested that an M. tuberculosis strain that causes mild lung disease may allow individuals with subclinical disease to be under recognized in the community and therefore more time to spread infection in the population. Therefore, such strains would be overly represented in a community. This demonstrates that the predominance of an M. tuberculosis strain in a community is not necessarily a marker of enhanced pathogenicity (eg, transmissibility is not equivalent to virulence).

HOST FACTORS

Genetic susceptibility to infection — Genetic analysis of sibling pairs has been used to evaluate putative genetic markers for enhanced susceptibility to tuberculosis (TB) in populations in Africa [165]. Possible markers on chromosomes 15q and Xq were identified; the investigators speculate that finding a susceptibility gene on an X chromosome may partially explain the increased incidence of TB in males in some populations.

Acquired susceptibility to infection — Investigators examined the interferon (IFN)-gamma response pathway in three patients with severe, unexplained nontuberculous mycobacterial disease [169]. In all three patients, IFN-gamma was undetectable following stimulation of whole blood but was detectable when stimulated in the absence of the patients' own plasma. An autoantibody against IFN-gamma was isolated from the patients' plasma and was found to be capable of blocking the upregulation of tumor necrosis factor (TNF)-alpha production in response to endotoxin, in blocking induction of IFN-gamma–inducible genes, and in inhibiting upregulation of human leukocyte antigen (HLA) class II expression on peripheral blood mononuclear cells (PBMCs). These acquired defects in the IFN-gamma pathway may explain unusual susceptibilities to intracellular pathogens, including mycobacteria, in patients without underlying, genetically determined immunologic defects.

All of the host genes associated with TB susceptibility account for a tiny fraction of all TB cases identified in the world. The most important host factor that determines TB susceptibility to TB is HIV coinfection, followed by other immunosuppressive conditions, including cancer, diabetes, and immunosuppressive medications. Environmental factors, such as crowding, low socioeconomic status, poor access to healthcare, and family history, also contribute substantially to the incidence of TB worldwide, and these are important to understanding TB pathogenesis.

L'infection tuberculeuse se définit par une multiplication bacillaire induisant une réponse immunitaire spécifique. La positivité d'un test immunitaire (intradermoréaction à la tuberculine [IDR] ou test in vitro mesurant la libération de l'interféron gamma) en est la marque

L'infection tuberculeuse latente est une infection au cours de laquelle la multiplication bacillaire est contrôlée efficacement par la réponse immunitaire spécifique : il n'y a ni signe radiologique ni signe clinique. La tuberculose-maladie est une infection au cours de laquelle la multiplication bacillaire se poursuit malgré la réponse immunitaire spécifique, aboutissant à des signes radiologiques, accompagnés ou non de signes cliniques. Le risque de progression immédiate de l'infection tuberculeuse vers la tuberculose-maladie est majoré chez les enfants âgés de moins de 5 ans (surtout ceux de moins de 2 ans), ainsi que chez les immunodéprimés.

Concernant l'enfant, il est estimé qu'environ 1 million d'enfants sont atteints de tuberculose chaque année dans le monde, occasionnant plus de 200 000 décès. Il est également estimé que, chaque année, 10 millions d'enfants deviennent orphelins à cause du décès d'un des deux parents par tuberculose

• En France, il y a environ 5 000 nouveaux cas de tuberculose-maladie chaque année.
  L'incidence en 2015 était de 7,1/100 000.
  Plus du tiers des nouveaux cas sont déclarés en Ile-de-France. Moins de 100 souches bacillaires identifiées chaque année sont MDR.
  
  
• Les cas pédiatriques (< 15 ans) représentent environ 5 % des cas déclarés (autour de 250 cas annuels), la moitié de ces cas étant observée chez l'enfant de moins de 5 ans.
• Chaque année, 2 à 3 cas de méningite tuberculeuse sont déclarés chez l'enfant de moins de 5 ans.
• Depuis l'arrêt de l'obligation vaccinale par le BCG au niveau national, on observe une légère augmentation des cas de tuberculose-maladie chez les enfants non vaccinés.
  Cette augmentation reste toutefois bien inférieure aux simulations qui avaient précédé les modifications du calendrier vaccinal

Enfin, certains enfants vont avoir un risque augmenté de progression immédiate vers la tuber- culose-maladie en cas d'infection :

• âge < 5 ans et surtout < 2 ans
• Immunodépression (quelque soit la cause)
• IRC avec hémodialyse

Ces facteurs sont cumulatifs.

Il est souvent proposé de considérer le risque d'infection comme significatif lorsque la durée de contact cumulée sur les 3 derniers mois est supérieure à 8 heures si le cas index est BAAR + et supérieure à 40 heures si le cas index est BAAR – et culture + .

Les premières approches concluantes ont été la découverte de gènes de susceptibilité aux infections graves, sur la voie IL-12/interféron γ. La vaccination par le BCG diminue le risque d'évolution vers la tuberculose-maladie et, sur- tout, le risque d'évolution vers une forme disséminée. Il n'est pas certain que le BCG prévienne l'infection elle-même (voir chapitre 43)

Étapes du dépistage (fig. 61.1) À la première évaluation, quel que soit l'âge de l'enfant :

Examen clinique complet

RxTho de face, facilement complétée par un profil si âge 5 ans (moins de 5 ans ?)

Chez l'enfant immunocompétent âgé de moins de 5 ans, une IDR à la tuberculine (Tubertest ® ) est également systématiquement réalisée lors de cette première évaluation.

Les règles d'interprétation de l'IDR dépendent de la vaccination par le BCG (tableau 61.1).

Les tests interféron gamma (IFN-γ) ne sont actuellement pas recommandés pour le dépistage des enfants d'âge < 5 ans (HAS, 2011).

En l'absence de critères d'infection, une nouvelle évaluation par IDR et radiographie de thorax sera réalisée 8 à 12 semaines après le dernier contact avec le cas index.
Durant cet intervalle, les enfants âgés de moins de 2 ans doivent bénéficier d'une prophylaxie, c'est-à-dire en pratique d'un traitement d'infection tuberculeuse latente

Chez l'enfant immunocompétent âgé de 5 ans ou plus, du fait du très faible risque de progression rapide vers la tuberculose-maladie, le test immunitaire n'est pas réalisé initialement si la radiographie de thorax est normale, mais seulement après un délai de 8 à 12 semaines après le dernier contact avec le cas index. Un test in vitro de libération d'IFN-γ (Quantiféron ® , SPOT-TB ® ) peut remplacer l'IDR. Tout test positif définit une infection tuberculeuse.

Chez l'enfant immunodéprimé, le risque de progression rapide vers la tuberculose-maladie est élevé. La stratégie de dépistage est donc superposable à celle réalisée chez l'enfant âgé de moins de 5 ans.

Les tests immunitaires ont souvent une sensibilité diminuée dans ces situations, mais doivent néanmoins être réalisés.

Une prophylaxie est indiquée pour tous ces enfants, entre les deux évaluations.

Toute radiographie anormale doit faire considérer le diagnostic de tuberculose-maladie, quel que soit le résultat des tests immunitaires.

Quel que soit l'âge de l'enfant, toute infection tuberculeuse latente (ITL) à bacille supposé sensible doit être traitée

Généralités Le diagnostic de tuberculose-maladie peut être difficile chez l'enfant, particulièrement chez le nourrisson. La tuberculose-maladie de l'enfant étant le plus souvent pauvre en bacilles, la preuve microbiologique fait souvent défaut, contrairement à la tuberculose de l'adulte. Le diagnostic est donc le plus souvent posé sur un faisceau d'arguments intégrant la notion de contage, la présence de signes cliniques, une réponse immunitaire spécifique positive (IDR ou test interféron gamma), et des anomalies radiologiques évocatrices

L'examen microscopique est positif chez moins de 20 % des enfants avec une tuberculose- maladie. La culture est positive dans moins de 50 % des cas. Malgré cette faible rentabilité, les recherches microbiologiques sont systématiques devant toute suspicion de tuberculose-maladie. Leur positivité imposera un dépistage autour de l'enfant

Imagerie thoracique

• La radiographie de thorax peut être d'emblée très évocatrice, en montrant :
  • Des adénopathies latérotrachéales droites
  • Médiastinales et hilaires, typiques de la tuberculose de l'enfant (voir fig. 60.5 au chapitre 60).
    • Ces adénopathies peuvent comprimer les voies aériennes adjacentes avec éventuellement atélectasie ou emphysème obstructif.
      
  • Des opacités parenchymateuses alvéolaires sont également fréquentes.
  • Le classique complexe primaire, associant nodule parenchymateux d'un sommet et adénopathie satellite, est en fait rarement objectivé.
  • La présence de cavernes est rare chez l'enfant (sauf chez l'adolescent)

Le scanner thoracique avec injection de produit de contraste permet de préciser les aspects radiologiques. Les adénopathies tuberculeuses ont typiquement un centre hypodense (nécrose) et une périphérie discrètement rehaussée lors de l'injection de produit de contraste (aspect toutefois inconstant chez l'enfant)

Endoscopie bronchique

• Elle est nécessaire pour objectiver l'atteinte endobronchique, et est réalisée dès que le scanner suggère une compression des voies aériennes.
  
  
• Elle permet la mise en évidence de compressions extrinsèques par les adénopathies, les réactions inflammatoires de la paroi bronchique sous forme de granulome, et d'éventuelle fistulisation ganglionnaire avec irruption de caséum.
  
  
• Les prélèvements endobronchiques, ou le lavage bronchoalvéolaire, n'ont pas de meilleur rendement microbiologique chez l'enfant que trois tubages gastriques consécutifs. L'endoscopie bronchique n'est donc pas réalisée à but uniquement microbiologique

IGRAs are in vitro blood tests of cell-mediated immune response; they measure T cell release of interferon-gamma following stimulation by antigens unique to Mycobacterium tuberculosis and a few other mycobacteria [1].

IGRAs are surrogate markers of M. tuberculosis infection and indicate a cellular immune response to M. tuberculosis protein antigens secreted by M. tuberculosis, but not by most non-tuberculous mycobacteria (NTM) or by Mycobacterium bovis Bacille Calmette-Guérin (BCG). In most individuals infected with M. tuberculosis, mononuclear cells in the blood release interferon-gamma when stimulated with antigens derived from M. tuberculosis

To perform an IGRA test, a blood sample is incubated with antigens and controls [2]. These tests require cells to remain viable and functional from the time of blood draw to completion of the incubation period with antigens.

IGRAs cannot distinguish between latent infection and active TB disease and should not be used for diagnosis of active TB, which is a microbiologic diagnosis. A positive IGRA result may not necessarily indicate active TB, and a negative IGRA result may not rule out active TB [3,4]

IGRAs appear to be unaffected by most infections with environmental NTM, which can cause false-positive TSTs [6]. In addition, limited data suggest that IGRA results are not affected by intravesical BCG for bladder cancer [7]

estimate the basic reproductive number of the infection (𝑅𝑅 0 ) to be significantly greater than one. We estimate it to be between 3.6 and 4.0, indicating that 72-75% of transmissions must be prevented by control measures for infections to stop increasing.

only 5.1% (95%CI, 4.8–5.5) of infections in Wuhan are identified

we expect further outbreaks to occur in other Chinese cities, and that infections will continue to be exported to international destinations at an increasing rate. In 14 days’ time (4 February 2020), our model predicts the number of infected people in Wuhan to be greater than 250 thousand (prediction interval, 164,602 to 351,396). We predict the cities with the largest outbreaks elsewhere in China to be Shanghai, Beijing, Guangzhou, Chongqing and Chengdu. We also predict that by 4 Feb 2020, the countries or special administrative regions at greatest risk of importing infections through air travel are Thailand, Japan, Taiwan, Hong Kong, and South Korea.

travel restrictions from and to Wuhan city are unlikely to be effective in halting transmission across China; with a 99% effective reduction in travel, the size of the epidemic outside of Wuhan may only be reduced by 24.9% on 4 February

On 31 December 2019, Chinese authorities alerted the World Health Organization (WHO) to an outbreak of pneumonia of unknown aetiology in Wuhan City, Hubei Province, China. A novel strain of coronavirus (2019-nCoV) was subsequently isolated from a patient on 7 January 2020 (Tan et al. 2020). Most cases from the initial cluster had epidemiological links with a live animal market (Huanan South China Seafood Market), suggesting a possible zoonotic origin (World Health Organization 2020a). However, the definitive source of the virus is unknown. Infections in family clusters as well as in healthcare workers confirm the occurrence of human-to-human transmission, though the extent of this mode of transmission is unclear.

Cases with travel history to Wuhan have also been reported in other Chinese provinces, including large cities such as Beijing, Shanghai and Shenzhen, as well as other countries, including Thailand (n=4), Japan (n=1), South Korea (n=1), Taiwan (n=1), Hong Kong (n=2), Macau (n=2) and the United States (n=1).

Presenting symptoms of [2019-nCoV] cases reported include fever, cough, and shortness of breath (World Health Organization 2020c).

Current clinical and epidemiological data are insufficient to understand the full extent of the transmission potential of the epidemic.

the majority of air travel departing Wuhan is domestic (87.2% of bookings, Jan 2017)

deterministic SEIR metapopulation transmission model

We modelled the period from 1 January 2020 when local authorities closed the wet market implicated as the zoonotic source of human infection (World Health Organization 2020a).

only considered human-to- human transmission in our model, and made the assumption that following the closure of the market, no further zoonotic infection contributed to the epidemic dynamics

assuming that travellers are drawn randomly from the origin population

assumed that the incubation period was 4 days, based on an estimate for SARS, a related coronavirus (Lessler et al. 2009)

Fitting was achieved by predicting the number of infections in all Chinese cities and other countries having onset between 1 and 21 January 2020, and maximizing the likelihood using the optim function in the R statistical language (R Core Team 2019)

projections make strong assumptions: that no control interventions are instigated; that the key epidemiological variables driving epidemic dynamics remain constant; that travel behaviour within China and to the rest of the world continues as per our mobility estimates; finally, we only consider travel by air, and do not include land transportation, particularly via trains

We [Read et al. 2020] estimate the number of people infected in Wuhan from 1 January to 18 January was 4,764 (prediction interval, 3,969–5,817); this is similar to other published estimates (Imai et al. 2020) and highlights the estimated rapid growth of the epidemic

By 4 February, there will be an elevated risk of importations into other countries, most notably to Thailand (average of 15.0 imports d -1 ), Japan (7.8), Taiwan (6.3), Hong Kong (5.8); South Korea (5.5), USA (4.5), Malaysia (4.1), Singapore (3.2), Australia (2.9) and Vietnam (2.7)

We explored the potential impact of a reduction in travel from and to Wuhan, by reducing the appropriate airline traffic by 50%, 80%, 90%, 95% and 99%. Our model predicts that these reductions in travel results in 12.6%, 20.1%, 22.6%, 23.9% and 24.9% reduction in infections, respectively, elsewhere in China by 4 February.

In line with other modelling studies of travel restrictions: reducing travel only serves to delay the epidemic reaching other locations, rather than suppressing the spread entirely

Our estimates of the basic reproductive number for this novel coronavirus are higher than most estimates reported for SARS and MERS-CoV, but similar to some estimates from subsets of data in the early period of SARS. For the SARS coronavirus, estimates ranged from 1.1 to 4.2 with most estimates between 2 and 3 (Bauch et al. 2005).

spatial component of our model is dependent upon only airline travel; the model does not include rail and road transportation, so we may underestimate local connectivity

do not attempt to account for any dynamic changes in control or other factors than may influence transmission, nor changes in surveillance and reporting effort

modelling of Wuhan travel restrictions does not account for stochastic effects, and so may underestimate the potential effect of travel restrictions as a consequence

Earlier novel coronavirus (SARS and MERS-CoV) outbreaks found evidence for substantial heterogeneity in reproductive numbers between individuals (Cauchemez et al. 2016; Bauch et al. 2005; Chowell et al. 2004). In our analysis, we assume that there is little heterogeneity

A threat to the accuracy of these projections is if a substantial proportion of infection has been due to multiple exposures to animals that has been curtailed in some way.

In software development life cycle (SDLC), artifact usually refers to "things" that are produced by people involved in the process.

Examples of artifacts would be design documents, data models, workflow diagrams, test matrices and plans, setup scripts.

In software, like an archaeological site, any thing that is created could be an artifact.

In systems that are used to take actions in the real world, it can be useful to set and enforce action limits as a sanity check. Action limits for systems should be broad enough not to trigger spuriously. If the system hits a limit for a given action, automated alerts should fire and trigger manual intervention or investigation.

This is not to say that machine learning is bad, or even that technical debt is something to be avoided at all costs. It may be reasonable to take on moderate technical debt for the benefit of moving quickly in the short term, but this must be recognized and accounted for lest it quickly grow unmanageable.

Perhaps the most important insight to be gained is that technical debt is an issue that both engineers and researchers need to be aware of.

Research solutions that provide a tiny accuracy benefit at the cost of massive increases in system complexity are rarely wise practice. Even the addition of one or two seemingly innocuous data dependencies can slow further progress.

CS is an in- creasingly broad and diverse field. It combines aspects of mathematical reasoning, engineering methodology, and the empirical approaches of the scientific method. The empirical com- ponents are clearly on the upswing, in part because the computer systems we construct have become so large that analytic techniques cannot properly describe their properties, because the systems now dynamically adjust to the difficult-to-predict needs of a diverse user community, and because the sys- tems can learn from vast datasets and large numbers of interactive sessions that provide continuous feedback

we note that in the terminology of Pasteur’s Quadrant, 11 we do “use-inspired basic” and “pure ap- plied” (CS) research.

Recent articles, such as those by Leifer et al. 8 and Enkel et al., 6 illustrate related issues on how firms do research and catalyze innovation

The goal of research at Google is to bring significant, practical benefits to our users, and to do so rapidly, within a few years at most.

Research happens throughout Google, exploring techni- cal innovations whose implementation is risky, and may well fail.

Sometimes,

research at Google operates in entire-

ly new spaces, but most frequently,

the goals are major advances in areas

where the bar is already high, but there is still potential for new methods

In particular, we often observed the incorrect application of over-sampling techniques to circumvent data imbalance issues [11],

we try to divide the referred works according to different potential issues, to the best of our ability (see Section 2). After a short discussion on the use of over-sampling techniques (Section 3)

(Section 4) is devoted to reporting our own results on the TPE- HGDB dataset for the various models, engineered features, and over-sampling techniques available in literature. Moreover, we investigate the actual impact of applying over-sampling, when performed correctly, and the added value of tun- ing the over-sampling algorithm and its corresponding hyper-parameters.

The state-of-the-art methods for rela- tion classification are primarily based on Convolutional or Recurrent Neural Net- works.... In this paper, we pro- pose a model that both leverages the pre- trained BERT language model and incor- porates information from the target entities to tackle the relation classification task. ... We achieve significant improvement over the state-of-the-art method on the SemEval- 2010 task 8 relational dataset.

The task of relation classification is to predict se- mantic relations between pairs of nominals. Given a sequence of text (usually a sentence) s and a pair of nominals e 1 and e 2 , the objective is to identify the relation between e 1 and e 2

Recently, deep neural networks have applied to relation classification (Socher et al., 2012; Zeng et al., 2014; Yu et al., 2014; dos Santos et al., 2015; Shen and Huang, 2016; Lee et al., 2019). These methods usually use some features derived from lexical resources such as Word-Net or NLP tools such as dependency parsers and named entity rec- ognizers (NER).

Language model pre-training has been shown to be effective for improving many natural language processing tasks (Dai and Le, 2015; Peters et al., 2017; Radford et al., 2018; Ruder and Howard, 2018; Devlin et al., 2018). The pretrained model BERT proposed by (Devlin et al., 2018) has es- pecially significant impact.

he tasks that BERT has been applied to are typically modeled as clas- sification problems and sequence labeling prob- lems. It has also been applied to the SQuAD ques- tion answering (Rajpurkar et al., 2016) problem, in which the objective is to find the starting point and ending point of an answer span

We insert special tokens before and after the target entities before feeding the text to BERT for fine-tuning, in order to identify the locations of the two target entities and transfer the information into the BERT model.

We then locate the positions of the two target entities in the out- put embedding from BERT model. We use their embeddings as well as the sentence encoding (em- bedding of the special first token in the setting of BERT) as the input to a multi-layer neural network for classification. By this way, it captures both the semantics of the sentence and the two target enti- ties to better fit the relation classification task.

MVRNN model (Socher et al., 2012) applies a recursive neural network (RNN) to relation clas- sification. They assign a matrix-vector represen- tation to every node in a parse tree and com- pute the representation for the complete sentence from bottom up according to the syntactic struc- ture of the parse tree. (Zeng et al., 2014) pro- pose a CNN model by incorporating both word embeddings and position features as input. Then they concatenate lexical features and the output from CNN into a single vector and feed them into a softmax layer for prediction. (Yu et al., 2014) propose a Factor-based Compositional Em- bedding Model (FCM) by constructing sentence- level and substructure embeddings from word em- beddings, through dependency trees and named entities. (Santos et al., 2015) tackle the relation classification task by ranking with a convolutional neural network named CR-CNN. Their loss func- tion is based on pairwise ranking. In our work, we take advantage of a pre-trained language model for the relation classification task, without relying on CNN or RNN architecutures. (Shen and Huang, 2016) utilize a CNN encoder in conjunction with a sentence representation that weights the words by attention between the target entities and the words in the sentence to perform relation classification. (Wang et al., 2016) propose a convolutional neu- ral network architecture with two levels of atten- tion in order to catch the patterns in heterogeneous contexts to classify relations. (Lee et al., 2019) de- velop an end-to-end recurrent neural model which incorporates an entity-aware attention mechanism with a latent entity typing for relation classifica- tion. There are some related work on the relation ex- traction based on distant supervision, for example, (Mintz et al., 2009; Hoffmann et al., 2011; Lin et al., 2016; Ji et al., 2017; Wu et al., 2019). The difference between relation classification on reg- ular data and on distantly supervised data is that the latter may contain a large number of noisy la- bels. In this paper, we focus on the regular relation classification problem, without noisy labels.

The pre-trained BERT model (Devlin et al., 2018) is a multi-layer bidirectional Transformer encoder (Vaswani et al., 2017).

The Transformer – a model that uses attention to boost the speed with which these models can be trained.

The Transformers outperforms the Google Neural Machine Translation model in specific tasks. The biggest benefit, however, comes from how The Transformer lends itself to parallelization.

The Transformer was proposed in the paper Attention is All You Need. A TensorFlow implementation of it is available as a part of the Tensor2Tensor package. Harvard’s NLP group created a guide annotating the paper with PyTorch implementation.

Popping open that Optimus Prime goodness, we see an encoding component, a decoding component, and connections between them.

The encoding component is a stack of encoders (the paper stacks six of them on top of each other – there’s nothing magical about the number six, one can definitely experiment with other arrangements). The decoding component is a stack of decoders of the same number.

The encoders are all identical in structure (yet they do not share weights). Each one is broken down into two sub-layers:

The encoder’s inputs first flow through a self-attention layer – a layer that helps the encoder look at other words in the input sentence as it encodes a specific word.

The outputs of the self-attention layer are fed to a feed-forward neural network. The exact same feed-forward network is independently applied to each position.

he decoder has both those layers, but between them is an attention layer that helps the decoder focus on relevant parts of the input sentence (similar what attention does in seq2seq models).

we begin by turning each input word into a vector using an embedding algorithm.

Each word is embedded into a vector of size 512.

The embedding only happens in the bottom-most encoder. The abstraction that is common to all the encoders is that they receive a list of vectors each of the size 512 – In the bottom encoder that would be the word embeddings, but in other encoders, it would be the output of the encoder that’s directly below. The size of this list is hyperparameter we can set – basically it would be the length of the longest sentence in our training dataset.

the word in each position flows through its own path in the encoder. There are dependencies between these paths in the self-attention layer. The feed-forward layer does not have those dependencies, however, and thus the various paths can be executed in parallel while flowing through the feed-forward layer.

Say the following sentence is an input sentence we want to translate:

”The animal didn't cross the street because it was too tired”

What does “it” in this sentence refer to? Is it referring to the street or to the animal? It’s a simple question to a human, but not as simple to an algorithm.

When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”.

Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.

Be sure to check out the Tensor2Tensor notebook where you can load a Transformer model, and examine it using this interactive visualization.

The first step in calculating self-attention is to create three vectors from each of the encoder’s input vectors (in this case, the embedding of each word). So for each word, we create a Query vector, a Key vector, and a Value vector. These vectors are created by multiplying the embedding by three matrices that we trained during the training process.

Notice that these new vectors are smaller in dimension than the embedding vector. Their dimensionality is 64, while the embedding and encoder input/output vectors have dimensionality of 512. They don’t HAVE to be smaller, this is an architecture choice to make the computation of multiheaded attention (mostly) constant.

What are the “query”, “key”, and “value” vectors?

They’re abstractions that are useful for calculating and thinking about attention.

Définition d'inflammation granulomateuse

Il ne faut pas confondre un « granulome inflammatoire » et une « inflammation granulomateuse ».

• Le « granulome inflammatoire » est un ensemble d'éléments cellulaires inflammatoires, associant des leucocytes et des macrophages, visible sur un prélèvement tissulaire, sans architecture particulière.
  
  
• L'« inflammation granulomateuse » est une dénomination plus restrictive d'une lésion limitée, d'aspect nodulaire. Elle est majoritairement constituée de cellules mononucléées histiocytaires (macrophages, cellules épithélioïdes et/ou cellules géantes multinucléées) et de lymphocytes, avec participation de fibroblastes. Cette lésion correspond au granulome épithélioïde et gigantocellulaire. C'est la traduction morphologique visible d'une réaction immunitaire à médiation cellulaire de type Th1

The second step in calculating self-attention is to calculate a score. Say we’re calculating the self-attention for the first word in this example, “Thinking”. We need to score each word of the input sentence against this word.

The score determines how much focus to place on other parts of the input sentence as we encode a word at a certain position.

Ces macrophages infectés sécrètent des médiateurs solubles permettant le recrutement d'autres populations cellulaires, telles que les neutrophiles, les cellules dendritiques (DC). Les DC sont des cellules présentatrices de l'antigène capables de migrer vers les organes lymphoïdes secondaires afin de recruter les lymphocytes T. De retour sur le site d'infection, les différent es populations cellulaires vont s'organiser afin de former un granulome nécessaire au confinement de l'infection

Les différentes étapes de cette réaction (fig. 20.1) :

• La cellule présentatrice d'antigène sécrète de l'IL-12 induisant une différenciation de type Th1 au niveau des lymphocytes T, pour ensuite leur présenter l'antigène
• Les lymphocytes Th1, en reconnaissant l'antigène, sécrètent de l'IL-2 et de l'interféron γ (IFN-γ)
• L'IFN-γ active les macrophages provoquant une augmentation de leur bactéricidie et une sécrétion de TNF-α
• Le TNF-α recrute de nombreux autres macrophages qui vont changer de morphologie (transformation en cellules épithélioïdes et formation de cellules géantes multinucléées) et former ainsi le granulome épithélioïde et gigantocellulaire

• L'IFN-γ et le TNF-α sont les médiateurs essentiels de cette réaction immunitaire.
  
  
• Toute réaction immunitaire Th1, quelle qu'en soit la cause, se traduira par la présence de granulomes épithélioïdes.
  
  
• Ce mécanisme explique le principe des tests de type QuantiFERON ® pour le diagnostic de tuberculose, qui repose sur la mise en évidence d'une sécrétion d'IFN-γ par les lymphocytes du patient en présence d'antigènes spécifiques de M. tuberculosis, ainsi que le risque de développement d'une tuberculose maladie chez des patients avec infection latente traités par anti-TNF-α.

The third and forth steps are to divide the scores by 8 (the square root of the dimension of the key vectors used in the paper – 64. This leads to having more stable gradients. There could be other possible values here, but this is the default), then pass the result through a softmax operation.

Softmax normalizes the scores so they’re all positive and add up to 1.

La pathogénicité des mycobactéries n'est pas liée à la sécrétion de toxines ou d'enzymes. Elle dépend de leur capacité à résister au pouvoir bactéricide des macrophages et donc de la rapidité de la mise en place d'une réponse immunitaire cellulaire T-dépendante.

This softmax score determines how much how much each word will be expressed at this position. Clearly the word at this position will have the highest softmax score, but sometimes it’s useful to attend to another word that is relevant to the current word.

Une bactérie Ziehl positive est un BAAR, et cette propriété est commune à toutes les mycobactéries

The fifth step is to multiply each value vector by the softmax score (in preparation to sum them up). The intuition here is to keep intact the values of the word(s) we want to focus on, and drown-out irrelevant words (by multiplying them by tiny numbers like 0.001, for example).

The sixth step is to sum up the weighted value vectors. This produces the output of the self-attention layer at this position (for the first word).

The paper further refined the self-attention layer by adding a mechanism called “multi-headed” attention.

L'immunité à médiation cellulaire (type Th1) se développe en deux à trois semaines et induit la formation de granulomes épithélioïdes ± gigantocellulaires avec possibilité de nécrose caséeuse centrale au niveau du foyer pulmonaire primaire et du ganglion, permettant le plus souvent de limiter la multiplication du BK. Les éléments du « complexe primaire », foyer granulomateux pulmonaire initial associé à une adénopathie satellite de drainage, ne sont presque jamais biopsiés.

En cas de biopsie, il faut envoyer un fragment en anatomie pathologique pour recherche, sur les prélèvements fixés en formol et inclus en paraffine, de granulomes épithélioïdes et gigantocellulaires avec nécrose caséeuse centrale et coloration de Ziehl pour mise en évidence des bacilles. Cette coloration doit être faite dès lors que le diagnostic de tuberculose est suspecté (fig. 20.4), mais c'est une coloration peu performante et sa négativité n'élimine pas le diagnostic +++ (les bacilles sont très peu nombreux dans les lésions granulomateuses et dans la nécrose caséeuse)

N.B. : les ganglions doivent faire l'objet d'une biopsie-exérèse (risque de fistulisation après ponction). Les prélèvements tissulaires sont toujours à partager entre la bactériologie et l'anatomie pathologique