on 01-Mar-2021 (Mon)

Although hypokalemia can be transiently induced by the entry of potassium into the cells, most cases result from unreplenished gastrointestinal or urinary losses due, for example, to vomiting, diarrhea, or diuretic therapy

Potassium is given cautiously in redistributive hypokalemia since the hypokalemia is transient and the administration of potassium can lead to rebound hyperkalemia when the underlying process is corrected and potassium moves back out of the cells.

It is also warranted for acute therapy in disorders such as hypokalemic or thyrotoxic periodic paralysis in which the hypokalemia is due to redistribution of potassium into the cells, often in association with significant symptoms

The severity of the manifestations of hypokalemia tends to be proportionate to the degree and duration of the reduction in serum potassium. Symptoms generally do not become manifest until the serum potassium is below 3.0 mEq/L, unless the serum potassium falls rapidly or the patient has a potentiating factor, such as a predisposition to arrhythmia due to the use of digitalis. Symptoms usually resolve with correction of the hypokalemia.

Muscle weakness usually does not occur at serum potassium concentrations above 2.5 mEq/L if the hypokalemia develops slowly [2]. However, significant muscle weakness can occur at serum potassium concentrations below 2.5 mEq/L or at higher values with hypokalemia of acute onset, as occurs in hypokalemic or thyrotoxic periodic paralysis. In addition, the pathophysiology of weakness in these disorders is more complex.

The pattern of weakness in hypokalemia is similar to that associated with hyperkalemia. Weakness usually begins in the lower extremities, progresses to the trunk and upper extremities, and can worsen to the point of paralysis.

In addition to causing muscle weakness, severe potassium depletion (serum potassium less than 2.5 mEq/L) can lead to muscle cramps, rhabdomyolysis, and myoglobinuria [4-7]. Potassium release from muscle cells during exercise normally mediates vasodilation and an appropriate increase in muscle blood flow [8]. Decreased potassium release due to profound hypokalemia can diminish blood flow to muscles during exertion, leading to ischemic rhabdomyolysis [8]. The clinical and pathologic abnormalities are reversible with potassium repletion [4]. A potential diagnostic problem is that the release of potassium from the cells with rhabdomyolysis can mask the severity of the underlying hypokalemia or even lead to normal or high values.

Other manifestations of muscle dysfunction due to hypokalemia include:

● Respiratory muscle weakness, which can be severe enough to result in respiratory failure and death.

● Involvement of gastrointestinal muscles, resulting in ileus and its associated symptoms of distension, anorexia, nausea, and vomiting.

A variety of arrhythmias may be seen in patients with hypokalemia. These include premature atrial complex (PAC; also referred to a premature atrial beat, premature supraventricular complex, or premature supraventricular beat) and premature ventricular beats, sinus bradycardia, paroxysmal atrial or junctional tachycardia, atrioventricular block, and ventricular tachycardia or fibrillation [2]. Hypokalemia produces characteristic changes on the ECG although they are not seen in all patients. There is depression of the ST segment, decrease in the amplitude of the T wave, and an increase in the amplitude of U waves which occur at the end of the T wave (waveform 1). U waves are often seen in the lateral precordial leads V4 to V6. Hypokalemia also prolongs the QT interval [11-13]

There is a large interpatient variability in the serum potassium concentration that is associated with progression of ECG changes or arrhythmias. In a carefully controlled trial of thiazide therapy (hydrochlorothiazide 50 mg/day), there was a two-fold increase in ventricular arrhythmias (as detected by Holter monitoring) in the small proportion of patients in whom the serum potassium concentration fell to or below 3.0 mEq/L [14]. In addition, the presence of concomitant factors, such as coronary ischemia, digitalis, increased beta-adrenergic activity, and magnesium depletion, can promote arrhythmias; the last two of these cofactors can also lower the serum potassium concentration

Epinephrine released during a stress response (as with coronary ischemia) drives potassium into the cells, possibly worsening preexisting hypokalemia. A similar effect can be seen with bronchodilator therapy with a beta-adrenergic agonist.

Hypokalemia may be associated with magnesium depletion (due, for example, to diuretics or diarrhea), both of which promote the development of arrhythmias. Hypokalemia and hypomagnesemia are associated with an increased risk of torsades de pointes, particularly in patients treated with drugs that prolong the QT interval or those with a genetic predisposition to the long QT syndrome. In addition to its direct proarrhythmic effect, hypomagnesemia can increase urinary potassium losses and lower the serum potassium concentration.

Hypokalemia reduces insulin secretion, which may play an important role in thiazide-associated diabetes. However, worsening glucose tolerance is much less common in the era of low-dose thiazide therapy (eg, 12.5 to 25 mg of hydrochlorothiazide)

The neuromuscular and cardiac symptoms induced by hypokalemia are related to alterations in the generation of the action potential [2]. The ease of generating an action potential (called membrane excitability) is related both to the magnitude of the resting membrane potential and to the activation state of membrane sodium channels. Opening the sodium channels leads to the passive diffusion of extracellular sodium into the cells, which is the primary step in this process.

According to the Nernst equation, the resting membrane potential is related to the ratio of the intracellular to the extracellular potassium concentration. In skeletal muscle, a reduction in the serum (extracellular) potassium concentration will increase this ratio and therefore hyperpolarize the cell membrane (that is, make the resting potential more electronegative); this impairs the ability of the muscle to depolarize and contract, leading to weakness. However, in some cardiac cells (such as Purkinje fibers in the conducting system), hypokalemia causes K2P1 channels, which are normally selective for potassium, to transport sodium into the cells, causing depolarization [16,17]. This leads to increased membrane excitability and arrhythmias.

Hypokalemia also delays ventricular repolarization by inhibiting the activity of potassium channels responsible for this component of the cardiac electrical cycle [12]. (See "Reentry and the development of cardiac arrhythmias".)

In addition, hypokalemia causes ventricular arrhythmias through downregulation of cardiac Na-K-ATPase activity [18,19]. This produces an increase in intracellular sodium, which impedes removal of intracellular calcium by the Na-Ca exchanger, leading to intracellular calcium overload. The ensuing activation of calmodulin kinase II activity reduces repolarization reserve by activating late sodium and calcium currents [20]. This, in turn, predisposes the heart to early afterdepolarization-associated arrhythmias, such as Torsades de pointes and polymorphic ventricular tachycardia [20]

The assessment of the hypokalemic patient begins with evaluation of muscle strength and obtaining an electrocardiogram to assess the cardiac consequences of the hypokalemia, with particular attention to the QT interval. At serum potassium concentrations below 2.5 mEq/L, severe muscle weakness and/or marked electrocardiographic changes may be present and require immediate treatment

Telemetry or continuous ECG monitoring is indicated for hypokalemic patients with a prolonged QT, other ECG changes associated with hypokalemia, and/or underlying cardiac issues that predispose to arrhythmia in the setting of hypokalemia (digoxin toxicity, myocardial infarction, underlying long QT syndrome, etc)

The urgency of therapy depends upon the severity of hypokalemia, associated and/or comorbid conditions, and the rate of decline in serum potassium concentration

The risk of arrhythmia from hypokalemia is highest in older patients, patients with organic heart disease, and patients on digoxin or antiarrhythmic drugs [23].

Potassium replacement is the mainstay of therapy in hypokalemia. Such therapy is clearly warranted in patients with hypokalemia due to renal or gastrointestinal losses. It should also be considered when hypokalemia is due to redistribution of potassium from the extracellular fluid into the cells (eg, hypokalemic periodic paralysis, insulin therapy) if serious complications such as paralysis, rhabdomyolysis, or arrhythmias are present or imminent.

Hypomagnesemia and redistributive hypokalemia — The underlying cause of the hypokalemia should be identified as quickly as possible, particularly the presence of hypomagnesemia or redistributive hypokalemia:

Patients with hypokalemia may also have hypomagnesemia due to concurrent loss with diarrhea or diuretic therapy or, in patients with hypomagnesemia as the primary abnormality, renal potassium wasting [24,25]. Such patients can be refractory to potassium replacement alone [26]. Thus, measurement of serum magnesium should be considered in patients with hypokalemia and, if present, hypomagnesemia should be treated.

The risk of rebound hyperkalemia is particularly high in patients with hypokalemic or thyrotoxic periodic paralysis in whom rebound hyperkalemia has been described in 40 to 60 percent of treated attacks

When increased sympathetic tone is thought to play a major role, the administration of a nonspecific beta blocker, such as propranolol, should be considered. The greatest experience is with acute attacks of hypokalemic thyrotoxic periodic paralysis [31], although head injury and theophylline toxicity can also result in redistributive hypokalemia [32-35], presumably due to sympathetic activation and elevated epinephrine levels [36]. In such patients, high-dose oral propranolol or intravenous propranolol rapidly reverses the hypokalemia and paralysis seen in acute attacks, without rebound hyperkalemia

Estimation of the potassium deficit assumes that there is a normal distribution of potassium between the cells and the extracellular fluid. The most common settings in which this estimation does not apply is diabetic ketoacidosis or nonketotic hyperglycemia, and in redistributive causes of hypokalemia such as hypokalemic periodic paralysis.

Estimation of the potassium deficit and careful monitoring of the serum potassium helps to prevent hyperkalemia due to excessive supplementation. This is not an uncommon outcome in hospitalized patients since, in one report, one in six patients developed mild hyperkalemia following potassium administration for hypokalemia [38]. The risk of overcorrection is increased in patients with a reduced glomerular filtration rate.

In different studies, the serum potassium concentration fell by approximately 0.27 mEq/L for every 100 mEq reduction in total body potassium stores [3,37,39] and, in chronic hypokalemia, a potassium deficit of 200 to 400 mEq is required to lower the serum potassium concentration by 1 mEq/L [39]. However, these estimates are only an approximation of the amount of potassium replacement required to normalize the serum potassium concentration and careful monitoring is required

In diabetic ketoacidosis or a hyperosmolar hyperglycemic state (nonketotic hyperglycemia), hyperosmolality and insulin deficiency favor the movement of potassium out of cells. As a result, the serum potassium concentration at presentation may be normal or even elevated despite a marked potassium deficit due to urinary and, in some patients, gastrointestinal losses [40]. The initiation of insulin therapy and fluid replacement will lower the serum potassium toward the level appropriate for the potassium deficit. Potassium supplementation is usually begun once the serum potassium concentration is 4.5 mEq/L or lower.

Occasional patients with uncontrolled diabetes (eg, 6 percent in one study) have more marked potassium loss and are hypokalemic at presentation [41]. Such patients require aggressive potassium replacement (20 to 30 mEq/hour), which can be achieved by the addition of 40 to 60 mEq of potassium chloride to each liter of one-half isotonic saline. Since insulin will worsen the hypokalemia, insulin therapy should be delayed until the serum potassium is above 3.3 mEq/L to avoid possible complications of hypokalemia such as cardiac arrhythmias and respiratory muscle weakness.

Potassium bicarbonate or its precursors are preferred in patients with hypokalemia and metabolic acidosis (eg, renal tubular acidosis or diarrhea) [23,37]. Only potassium acetate is available for intravenous use.

Potassium phosphate should be considered only in the rarely seen patients with hypokalemia and hypophosphatemia, as might occur with proximal (type 2) renal tubular acidosis associated with Fanconi syndrome and phosphate wasting [37,42,43]

Potassium chloride is preferred in all other patients for two major reasons [1]:

• Patients with hypokalemia and metabolic alkalosis are often chloride depleted due, for example, to diuretic therapy or vomiting. In such patients, chloride depletion contributes to maintenance of the metabolic alkalosis by enhancing renal bicarbonate reabsorption and may contribute to potassium wasting as sodium is reabsorbed in exchange for secreted potassium rather than with chloride [3,44,45]. It has been estimated that administration of non-chloride-containing potassium salts in the presence of metabolic alkalosis results in the retention of only 40 percent as much potassium as the administration of potassium chloride [45]. (See "Pathogenesis of metabolic alkalosis", section on 'Chloride depletion' and "Treatment of metabolic alkalosis", section on 'Treatment'.)

• Potassium chloride raises the serum potassium concentration at a faster rate than potassium bicarbonate. Chloride is primarily an extracellular anion that does not enter cells to the same extent as bicarbonate, thereby promoting maintenance of the administered potassium in the extracellular fluid [46] In addition, potassium bicarbonate may partially offset the benefits of potassium administration by aggravating metabolic alkalosis, if present

Oral potassium chloride can be given in crystalline form (salt substitutes), as a liquid, or in a slow-release tablet or capsule. Salt substitutes contain 50 to 65 mEq per level teaspoon; they are safe, well tolerated, and much less expensive than the other preparations [47]. Liquid forms of potassium chloride are also inexpensive, but are often unpalatable. Nevertheless, they may be preferred in patients with an enteral feeding tube or who are unable to swallow tablets. Slow-release tablets are better tolerated, but they have been associated with gastrointestinal ulceration and bleeding, which have been ascribed to local accumulation of high concentrations of potassium [48]. The risk is relatively low, and even lower with microencapsulated preparations (eg, microK or Klor-Con) compared to wax matrix tablets [3]

Increasing the intake of potassium-rich foods, such as oranges and bananas (table 1), is less effective, in part because dietary potassium is predominantly in the form of potassium phosphate or potassium citrate which, as mentioned earlier in this section, results in the retention of only 40 percent as much potassium as potassium chloride [45]. In addition, the potassium concentration is relatively low in fruit (eg, approximately 2.2 mEq/inch [0.9 mEq/cm] in bananas) [49]. As a result, it would take two to three bananas to provide 40 mEq

Stable patients with chronic diuretic therapy (at a fixed dose), primary aldosteronism (unless aldosterone secretion increases), or Gitelman or Bartter syndrome typically do not develop progressive hypokalemia because the increased urinary potassium losses are balanced by hypokalemia-induced potassium retention. The net effect is a new steady state in which potassium intake and output are in balance, with a lower-than-normal serum potassium concentration. A potassium-sparing diuretic is usually required in such patients who do not respond adequately to potassium supplementation.

There are two classes of potassium-sparing diuretics: blockers of the cortical collecting tubule sodium channels (amiloride and triamterene); and the aldosterone (mineralocorticoid receptor) antagonists (spironolactone and eplerenone). With respect to treating hypokalemia, these drugs are used in patients with renal potassium wasting in whom, as noted in the preceding section, potassium supplements may not be sufficiently effective

Amiloride is usually preferred to a mineralocorticoid receptor antagonist because it is better tolerated (see "Triamterene nephrotoxicity"). Primary aldosteronism is an important exception since spironolactone or eplerenone is preferred to block apparent adverse effects of excess aldosterone on the heart and vascular system in patients diagnosed with this disorder.

Among patients with heart failure, mineralocorticoid receptor antagonists should be given only if the serum creatinine is less than or equal to 2.5 mg/dL (221 micromol/L) in men and 2.0 mg/dL (177 micromol/L) in women and the serum potassium is less than 5.0 mEq/L.

Most hypokalemic patients have a serum potassium concentration of 3.0 to 3.4 mEq/L. This degree of potassium depletion usually produces no symptoms. Exceptions include patients with heart disease (particularly if they are taking digitalis or certain other antiarrhythmic drugs or are undergoing cardiac surgery [50,51]) and patients with cirrhosis, in whom hypokalemia can increase ammonia generation and promote the development of hepatic encephalopathy

Patients with gastrointestinal losses are treated with potassium chloride if they have metabolic alkalosis (as usually seen with vomiting) or a normal serum bicarbonate concentration, and with potassium bicarbonate (or potassium citrate or acetate) in the presence of metabolic acidosis (as seen with diarrhea or renal tubular acidosis). Treatment is usually started with 10 to 20 mEq of potassium given two to four times per day (20 to 80 mEq/day), depending upon the severity of the hypokalemia.

By contrast, potassium supplements at usual doses produce only modest elevations in serum potassium in patients with hypokalemia due to renal potassium wasting (eg, chronic diuretic therapy, primary aldosteronism). As soon as the serum potassium rises, there is less hypokalemia-induced potassium retention and most of the administered potassium is excreted in the urine. Thus, a potassium-sparing diuretic is likely to be more effective. Amiloride is usually preferred to a mineralocorticoid receptor antagonist because it is better tolerated. Primary aldosteronism is an important exception since spironolactone or eplerenone is preferred to block apparent adverse effects of excess aldosterone on the heart and vascular system

Potassium must be given more rapidly to patients with hypokalemia that is severe (serum potassium less than 2.5 to 3.0 mEq/L) or symptomatic (arrhythmias, marked muscle weakness, or rhabdomyolysis)

A potential diagnostic and therapeutic problem in patients with hypokalemia-induced rhabdomyolysis is that the release of potassium from the muscle cells can mask the severity of the underlying hypokalemia or even lead to normal or elevated values at presentation or after potassium supplementation. If the serum potassium is normal or elevated at baseline, it will not be possible to be certain that underlying hypokalemia was responsible for the rhabdomyolysis and initial potassium therapy in such patients is not warranted and may be dangerous

Particular caution must be exercised when repleting potassium in patients with a concurrent disorder that, when treated, will tend to drive potassium into the cells and worsen the hypokalemia. The two main examples are insulin therapy in diabetic ketoacidosis or nonketotic hyperglycemia, and bicarbonate therapy in metabolic acidosis with a normal anion gap

Potassium repletion is most easily accomplished orally but can be given intravenously. The serum potassium concentration can transiently rise by as much as 1 to 1.5 mEq/L after an oral dose of 40 to 60 mEq, and by as much as 2.5 to 3.5 mEq/L after 135 to 160 mEq [52,53]. The serum potassium concentration will then fall back toward baseline over a few hours, as most of the exogenous potassium is taken up by the cells [54]. A patient with a serum potassium concentration of 2.0 mEq/L, for example, may have a 400 to 800 mEq potassium deficit [39]. In patients with severe hypokalemia, potassium chloride can be given orally in doses of 40 mEq, three to four times per day or, particularly in patients also treated with intravenous potassium, 20 mEq every two to three hours.

Careful monitoring is essential in patients treated with potassium. We suggest that the serum potassium should initially be measured every two to four hours to ascertain the response to therapy. If tolerated, this regimen should be continued until the serum potassium concentration is persistently above 3.0 to 3.5 mEq/L and symptoms or signs attributable to hypokalemia have resolved. Thereafter, the dose and frequency of administration can be reduced to that used in mild to moderate hypokalemia since aggressive repletion is no longer required and gastric irritation can be avoided.

A saline rather than a dextrose solution should be used for initial therapy since the administration of dextrose stimulates the release of insulin which drives extracellular potassium into the cells. This can lead to a transient 0.2 to 1.4 mEq/L reduction in the serum potassium concentration, particularly if the solution contains only 20 mEq/L of potassium [3,55]. The transient reduction in serum potassium can induce arrhythmias in susceptible patients, such as those taking digitalis [55].

Furthermore, treatment with insulin and intravenous fluids will exacerbate the hypokalemia and minimize the efficacy of potassium repletion. Thus, insulin therapy should be delayed until the serum potassium is above 3.3 mEq/L to avoid possible complications of hypokalemia such as cardiac arrhythmias, cardiac arrest, and respiratory muscle weakness.

Although isotonic saline is often the initial replacement fluid used in treating diabetic ketoacidosis or nonketotic hyperglycemia, the addition of potassium will make this a hypertonic fluid (since potassium is as osmotically active as sodium), thereby delaying reversal of the hyperosmolality. Thus, 40 to 60 mEq of potassium per liter in one-half isotonic saline is preferred.

In contrast to patients with marked potassium depletion, patients with hypokalemia due to potassium redistribution (eg, hypokalemic periodic paralysis) have no potassium deficit and even low rates of potassium administration can result in hyperkalemia once the redistributed potassium returns to the extracellular fluid. In a report of patients with hypokalemic thyrotoxic periodic paralysis (baseline serum potassium concentration 2.0 mEq/L), administration of potassium at a rate of 10 mEq/hour (80 mEq/L) resulted in hyperkalemia (>5.5 mEq/L) in 40 percent of patients, one-half of whom had ECG changes [30]. Appropriate therapy in these patients is discussed separately

Pain and phlebitis can occur during parenteral infusion of potassium into a peripheral vein. This primarily occurs at rates above 10 mEq/hour, but can be seen at lower rates. If pain occurs, either the infusion rate or, preferably, the potassium concentration should be reduced.

Rates above 20 mEq/hour are highly irritating to peripheral veins. Such high rates should be infused into a large central vein or into multiple peripheral veins.

For patients with severe hypokalemia due to gastrointestinal or renal losses, the recommended maximum rate of potassium administration is 10 to 20 mEq/hour in most patients.

● In any 1000 mL-sized container of appropriate non-dextrose fluid, we suggest a maximum of 60 mEq of potassium.

● In a small-volume mini-bag of 100 to 200 mL of water that is to be infused into a peripheral vein, we suggest 10 mEq of potassium.

● In a small-volume mini-bag of 100 mL of water that is to be infused into a large central vein, we suggest a maximum of 40 mEq of potassium.

Intravenous potassium is most often infused in a peripheral vein at concentrations of 20 to 60 mEq/L in a non-dextrose-containing saline solution. Use of an infusion pump is preferred to prevent overly rapid potassium administration in any intravenous container with more than 40 mEq of potassium or if the desired rate of potassium administration is more than 10 mEq/hour. For patients with severe hypokalemia, administration in a large central vein is preferred if this access is available.

L'hypokaliémie est définie comme une concentration plasmatique inférieure à 3,5 mmol/l. Il s'agit du trouble électrolytique le plus fréquent en pratique clinique. Il est présent chez 20% des patients hospitalisés et chez 10-40% des patients sous traitement de diurétiques thiazidiques.

En général, une hypokaliémie discrète à modérée (3,0-3,5 mmol/l) est bien tolérée chez les personnes en bonne santé, toutefois elle peut être vitale surtout chez les patients avec une pathologie cardiaque sous-jacente

Lorsque l'hypokaliémie est modérée à sévère (3 à 2,5 mmol/l), elle peut s'accompagner de symptômes non spécifiques (faiblesse musculaire, myalgies, sensation de fatigue). Quand la kaliémie est inférieure à 2,0 mmol/l, on peut observer une atteinte musculaire avec rhabdomyolyse, tétraparésie, atteinte des fibres musculaires lisses avec iléus paralytique, rétention urinaire et finalement arrêt respiratoire

En cas de pathologie cardiaque sous-jacente, une hypokaliémie discrète à modérée (3,0-3,5 mmol/l) peut quand même induire des arythmies cardiaques en relation avec la rapidité de son installation.

L'augmentation du gradient transcellulaire de potassium entraîne une augmentation du potentiel de repos membranaire avec comme conséquences une augmentation de la durée du potentiel d'action et de l'excitabilité myocardique (apparition d'une onde U). Les anomalies à l'ECG peuvent aller de l'onde U à des arythmies plus importantes (fibrillation auriculaire, extrasystoles, torsades de pointes et fibrillation ventriculaire)

Certains auteurs préconisent de maintenir une kaliémie supérieure à 4,5 mmol/l chez les patients présentant une insuffisance cardiaque ou une cardiopathie ischémique.³

Les b2 mimétiques, la théophylline et la caféine entraînent une hypokaliémie par stimulation de la pompe Na⁺,K⁺,-ATPase. L'insuline augmente l'activité de la Na⁺,K⁺,-ATPase, provoquant une hypokaliémie transitoire lors de son administration chez les patients diabétiques, néanmoins cette hypokaliémie n'a pas de conséquences cliniques, sauf en cas d'intoxication à l'insuline et lors du traitement de l'acidocétose diabétique. Le vérapamil est le seul des anticalciques qui peut parfois être responsable d'une hypokaliémie modérée

Lors de l'hypokaliémie dans le contexte de l'alcalose métabolique et respiratoire, le shift cellulaire permet de garder l'électroneutralité intracellulaire

La paralysie périodique hypokaliémique familiale est une affection génétique rare qui modifie la sensibilité au canal calcique, et est responsable d'hypokaliémies à la fin d'un effort intense ou d'un repas riche en glucides. Le mécanisme n'est pas clairement établi.

Lors de phénomènes d'anabolisme cellulaire comme lors de la correction d'une anémie mégaloblastique, on peut observer une hypokaliémie liée à l'utilisation du potassium dans la synthèse cellulaire

Une diminution des apports potassiques à moins de 1g par jour, soit 25 mmol par jour, peut induire une hypokaliémie transitoire car l'adaptation rénale n'est pas immédiate.

L'anorexie ou une alimentation riche en hydrates de carbone associée à une consommation chronique d'alcool sont les causes principales d'hypokaliémie d'origine alimentaire.

La concentration de potassium dans les selles normales est environ 80 à 100 mmol/l et la perte quotidienne normale est de 10 mmol/l.

Les vomissements peuvent également être responsables d'une hypokaliémie mais dont le mécanisme est d'origine rénale par alcalose métabolique de contraction et stimulation de la sécrétion d'aldostérone

Les diurétiques de l'anse agissent au niveau de la branche ascendante de Henle par inhibition du cotransporteur Na⁺-K⁺-2Cl^- (NKCC2) et les diurétiques thiazidiques au niveau du tubule contourné distal par inhibition du cotransporteur Na⁺-Cl^- (NCC1). La sécrétion de potassium dépend de la charge sodée au niveau distal, de l'aldostérone et du flux dans la lumière tubulaire. La kaliurie dans cette situation est donc élevée

Les diurétiques thiazidiques sont plus souvent associés à une hypokaliémie plus sévère que les diurétiques de l'anse. ⁷

La pénicilline et ses dérivés, lorsqu'ils sont administrés par voie intraveineuse à hautes doses, peuvent entraîner une perte rénale de potassium par augmentation du flux tubulaire distal.

L'amphotéricine B augmente la perméabilité de la membrane apicale au potassium entraînant une hypokaliémie et une hypomagnésémie.

En cas d'alcalose métabolique, on distingue celle à volume extracellulaire diminué (diurétiques, vomissements importants, sonde naso-gastrique, dérivations digestives hautes). La perte de chlore et la déplétion volémique sont à l'origine de l'hypokaliémie par stimulation du système rénine-angiotensine-aldostérone (hyperaldostéronisme secondaire). L'administration de chlore suffit en général à corriger la kaliémie : alcalose métabolique répondant au chlore.⁸

En cas de volume extracellulaire normal ou augmenté, on doit suspecter un hyperaldostéronisme primaire (syndrome de Conn) ou une sténose des artères rénales. Cette hypokaliémie ne répond pas au chlore mais au traitement de la pathologie de base.^9,10

Certaines anomalies génétiques rares touchant le rein comme le syndrome de Liddle s'accompagnant d'une hypervolémie avec hypertension (mutation avec gain de fonction canal ENaC), ou le syndrome de Bartter (mutation du cotransporteur Na⁺-K⁺-2Cl^- apical, canal K⁺ apical et canal Cl^- basolatéral au niveau de l'anse de Henle) et le syndrome de Gittelman (mutation du cotransporteur Na⁺-Cl⁺ du contourné distal) s'accompagnant d'une contraction volémique sont également des causes d'hypokaliémie.

L'acidose tubulaire distale (type I), proximale (type II) ou mixte (type III) figurent parmi les rares causes d'hypokaliémie associée à une acidose métabolique.

La concentration de potassium intracellulaire est maintenue par la pompe Na⁺,K⁺,-ATPase. En cas d'hypomagnésémie, l'activité de la pompe est altérée et la concentration extracellulaire de potassium augmente avec comme conséquence une adaptation rénale sous la forme d'une kaliurèse accrue et une hypokaliémie secondaire. De plus, l'hypomagnésémie stimule le système rénine-angiotensine-aldostérone. Les aminoglycosides, le foscarnet et le cisplatine peuvent provoquer une déplétion en magnésium, et entraîner une hypokaliémie.

L'anamnèse, la prise de médicaments et l'évaluation du volume extracellulaire par un examen clinique minutieux sont essentielles dans la démarche diagnostique d'une hypokaliémie.

La mesure du potassium urinaire sur un spot est alors utile. Lorsqu'il est inférieur à 20 mmol/l, ceci est compatible avec des pertes extrarénales de potassium, un shift cellulaire ou une diminution des apports alimentaires

La première est représentée par le cas d'un hyperaldostéronisme secondaire sur hypovolémie (diarrhées profuses). L'hypokaliémie est d'origine digestive mais la kaliurèse est haute en raison de la stimulation du système rénine-angiotensine-aldostérone. Dans la seconde, la kaliurèse haute signe l'origine rénale de l'hypokaliémie (syndrome de Conn, tubulopathies). La mesure du chlore urinaire permet de différencier ces deux possibilités. En cas d'hypovolémie, le chlore urinaire est bas.

La décision de corriger rapidement une hypokaliémie dépend du status clinique, des comorbidités et de la sévérité de l'hypokaliémie.

On estime que pour chaque augmentation de 0,3 mmol/l de la kaliémie, il est nécessaire d'administrer 100 mmol de potassium sous forme de chlorure de potassium.^11,12

En cas d'hypokaliémie sévère, ou symptomatique, l'administration de potassium par voie intraveineuse est nécessaire. Elle ne doit pas dépasser les 20 mmol/h et sous surveillance rythmique.¹² En cas de perfusion par voie périphérique, l'administration maximale est de 40 mmol par litre de solution

Une substitution en magnésium est souhaitée en cas d'hypokaliémie sévère

Since its introduction halfway the 20 th century, the Finite Element Method (FEM) has gained an immense amount of popularity as an engineering analysis tool. As of today, engineering fields such as aerospace and automotive rely heavily on FEM to compute the mechanical behaviour of structures under various types of loading. Despite its significant computational expense, engineers have gone as far as to model entire aircraft and other complex structures with this method. How- ever, it is often undesired to perform a FEM analysis of a full-scale model each time when a small component is changed. Instead of performing analyses on a fully-meshed model, a complex structure may be split into multiple sections based on a method called substructuring. Traditionally, this technique combines coarse global models with high-fidelity local substructure models. These substructures are subject to appropriate boundary conditions, obtained from analyses on the global scale. Essentially, there are three advantages to substructuring: an easier division of labour, a reduction in model preparation time and a reduction in computational time.

An emerging field that may assist in improving the efficiency and accuracy of how homogenised substructure representations are set up, is machine learning.

The current master’s thesis aims to develop a framework that integrates ANN and FEM tech- niques, as to provide a novel approach to homogenised substructure modelling. More specifically, this work strives to establish a procedure which allows to construct a homogenised representation of any substructure regardless of the structural complexity (e.g. inclusions or cut-outs) and ma- terial complexity (e.g. damage progression and failure). To achieve this, traditional techniques to obtain the homogenised behaviour of structural sections are replaced by a constitutive model that is captured in an ANN. With this in mind, the main research objective for this thesis is identified as follows: The objective within the time-span of this thesis is to improve the accuracy and/or computational efficiency of substructuring in FEM by implementing a constitutive model based on an artificial neural network, to be trained on a detailed and repre- sentative data set. To achieve this goal, a set of four research questions are formulated to guide and structure the re- search. The first one addresses background knowledge on ANNs, whereas the remaining questions focus on the actual research topic. I What artificial neural network set-ups, architectures and training procedures are suitable to model constitutive behaviour? Furthermore, how can a sufficiently representative training data set be constructed from a minimum amount of simulations? II To what extent can an artificial neural network be used to capture complex homogenised behaviour of a substructure? Furthermore, what are the conditions such that a substructure is fit for usage with the proposed method? III Is it feasible to integrate an artificial neural network into a finite element analysis procedure as a constitutive model? IV If the previous question is answered positively, what are the gains in computational time and/or accuracy from using the artificial neural network-based set-up?

Machine learning is a broad term that includes all algorithms which are able to extract patterns from a data set.

This subsection aims to provide an intuitive definition of a machine learning problem, which is the fundamental concept behind self-learning computational units such as ANNs. Each problem consists of three essential components which are: the task, the performance measure and the experience, as explained by Goodfellow et al. [3]. Mitchell [4] motivates the presence of a natural relation between these entities: "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P , if its performance at tasks in T , as measured by P , improves with experience E."

The type of performance measurement is to be chosen by the user. It is a critical part of the machine learning problem design which defines how certain types and magnitudes of errors translate into a learning step. For problems with continuous outputs, it is common to use a cost function such as the ones described in Subsection 2.2.3. The choice should always be made with respect to a specific problem, as it is often desired to treat each output error differently. For example, it may be so that an error value above a certain threshold for output a is considered unacceptable and heavily penalised (the learning process then puts ’extra effort’ in reducing this error). In the same system, output b may be less critical and feature a simple linear or logarithmic penalty [3]. Another method of measuring performance is to quantify the total accuracy of a trained com- putational unit on a data set. As explained in Subsection 2.2.1, labeled data is always divided in at least a training and a testing data set in the context of machine learning and neural networks. Measuring the accuracy of the trained computational unit on these sets serves as an indicator of how well the labels are predicted. Based on these values, one can make adjustments to the hyper- parameters (settings of the computational unit and training process) if necessary. For example, a low training and testing accuracy can be an indicator that the training time is too short or a lack of complexity in the model. A low testing accuracy in combination with a high training accuracy may be due to over-fitting of noisy training data (covered in Subsection 2.2.5).

As the name suggests, the fundamentals of artificial neurons (also called ’nodes’) and ANNs lie in theories based on how neurons in the human brain operate and cooperate. Whilst the processes inside the brain are still not fully understood, scientists all over the world have been intrigued by its capability to understand and predict even the most non-linear relations in data [7].

train machine learning systems

The choices made by the engineer concerning the data are one of the most critical aspects of building a well-functioning ANN [14].