Pharmacology of non-steroidal anti-inflammatory agents

This chapter tries to address Section K4(i) of the 2023 CICM Primary Syllabus, where NSADIS are shoehorned into "describe the pharmacology of drugs used to treat pain", whereas in previous versions of the syllabus these drugs had their own entry. To be asked about them as a class is not something were have seen in CICM exam papers, but they certainly seem to trot out ibuprofen now and then, as for example here:

• Question 1 from the second paper of 2017 (ibuprofen vs. paracetamol)
• Question 3(p.2) from the second paper of 2007 (ibuprofen vs tramadol)

In summary:

• Absorption for all NSAIDs is excellent and they have good bioavailability
• Distribution is variable; ibuprofen has the smallest VOD (0.1L/kg)
• Highly protein bound: most NSAIDs are 90-99% protein bound
  • In overdose, saturation of all binding sites increases the free fraction of the drug
  • This means the total drug levels (eg. salycilate levels) are not meaningful
  • This also increases the free fraction available for harmful activity, as well as for clearance by urinary excretion and through dialysis
• Metabolism is mainly hepatic, except for parecoxib which is a pro-drug rapidly hydrolysed into its active form (valdecoxib). The vast majority of metabolites are inactive and renally excreted
• Half-life of older agents is short (10-2hrs) and of newer COX-2 selective agents is longer (8-12 hrs), but this has no relationship to the duration of their clinical activity
• Mechanism of action involves blocking the action of cyclooxygease (COX) enzyme isoforms:
  • COX-1 is constantly active and has a homeostatic role
  • COX-2 is induced by inflammatory cytokines
  • Analgesic activity is mediated by both enzymes:
    • inflammatory eicosanoids are responsible for increasing the sensitivity of nociceptors, and NSAIDs prevent this
  • COX-1 inhibitor and nonselective NSAID side effects:
    • GI ulceration (decreased gastric mucosal pH and mucus synthesis)
    • Acute kidney injury (microvascular renal dysfunction)
  • COX-2 inhibitor side effects:
    • Anti-inflammatory activity is mainly due to COX-2 inhibition
    • Prothrombotic side effects are due to COX-2 inhibition
    • CCF exacerbation and hypertension

Free articles which treat the reader's time with respect include Bacchi et al (2012)  and  Rao & Knaus (2008), in the sense that they are short and to the point. Especially useful is the COX enzyme section from Rao & Knaus. They fall short in the department of pharmacokinetics; those articles are for some reason all paywalled, and you'd have to pay Elsevier for Day (1988) or Springer-Verlag for Verbeeck et al (1983).

Classification of non-steroidal anti-inflammatory drugs

NSAIDs can be classified in two main ways, and probably the best is a classification system based on their COX inhibition activity, as this immediately tells you something about their performance and side effects. Thus:

• Strong nonselective NSAIDs:
  • Ibuprofen, diclofenac, aspirin, piroxicam, naproxen
• Weak nonselective NSAIDs:
  • Sodium salicylate, nabumetone
• Weakly COX-2 selective NSAIDs:
  • Celecoxib, meloxicam, nimesulide, etodolac
• Highly COX-2 selective NSAIDs
  • Rofecoxib

Or, you could classify them according to their structure, which has massive chemistry nerd energy but tells you absolutely nothing about the drug or its expected performance in the field:

• Salicylic acid derivates: acetylsalicylic acid (aspirin), sulfosalazine
• Para-aminophenol derivates: Paracetamol
• Indole and indene acetic acids: indomethacin, etodolac, sulindac
• Hetero-aryl acetic acids: diclofenac, ketorolac, tolmetin
• Aryl-propionic acids: ibuprofen, ketoprofen, flurbiprofen, naproxen,
• Anthranilic acids: mefenamic acid, meclofenanic acid
• Enolic acids: piroxicam, tenoxicam, meloxicam
• Alkanones: nabumetone
• Pyrazolidinediones: phynylbutazone, oxyphenylbutazone
• Diarylheterocyclic NSAIDs: celecoxib, rofecoxib, valdecoxib, lumiracoxib, parecoxib, eterocoxib.

Even the minimally attentive reader will realise that the latter classification system integrates a lot more drugs than the former, for various reasons but mainly because many of the drugs do not fit neatly into the COX matrix, and some are grouped along with the nonsteroidal anti-inflammatories even though they themselves may have absolutely no intrinsic antiinflammatory activity (looking at you, paracetamol). In short, neither system is perfect. Fortunately, the ICU trainee is only likely to encounter a small selection of these substances clinically, and their properties are sufficiently similar that they can actually be discussed en masse as a class rather than individual drugs. That will be the objective for the rest of this chapter, picking a few of the usual suspects as illustrative examples. 

Pharmacokinetics of non-steroidal anti-inflammatory drugs

The ADME of NSAIDs is rather milquetoast in the sense that there is nothing there to really attract a person trying to create exam questions, if the task they put to themselves was "let's see if we can interrogate the candidate's understanding of basic pharmacokinetic principles". The drugs are well behaved and their pharmacokinetic profiles are free from any weird metabolic shenanigans. Observe:

	Route	Bio-avail.	pKa	VOD (L/kg)	Protein binding		Half-life (hrs)
Paracetamol	oral, IV, PR	80%	9.5	0.9L/kg	20%	Hepatic metabolism: major pathway is glucouronidation and sulfation; minor pathway involves the formation of a toxic metabolite (NAPQUI) which needs to be detoxified by conjugation with gluathione. All products of metabolism are renally excreted	2
Ibuprofen	Oral, PR, IV is available in some places	100%	4.9	0.1	99%	Almost 100% of the dose is metabolised in the liver: oxidation into water-soluble inactive metabolites. All products of metabolism are renally excreted	1.8
Diclofenac	Oral, PR	65%	4.2	1.4	99%	Hepatic metabolism, which converts 70% of the dose into water-soluble inactive metabolites, and 30% is excreted in the bile. Biliary excretion leads to enterohepatic recirculation.	1.2
Celecoxib	Oral only	80%	11.1	5-6	97%	Metabolised completely by the liver into hydoroxycelecoxib and carboxycelecoxib, neither of which has any COX activity. All products of metabolism are renally excreted	8-12
Parecoxib	Oral and IV	100%	6.7	0.8	98%	Undergoes rapid amine hydrolysis into valdecoxib, which is slowly metabolised in the liver into inactive metabolites.All products of valdecoxib metabolism are renally excreted	8-12

Administration is almost uniformly oral, in the sense that these drugs are marketed to users in the community rather than hospitals and intensive care units, but parenteral formulations (eg. parecoxib, paracetamol) are available. According to Irvine et al (2018), there does not appear to be any major challenges with creating parenteral options, other than extremely poor water solubility (and since when has that stopped anybody). The same properties that make it difficult to dissolve them in water make it easier to combine them with a waxy/oily carrier and offer them as a suppository.

Absorption is almost uniformly good. NSAIDs are weak acids that are lipid-soluble, which means that they will be non-ionised under most of the conditions available in the upper gastrointestinal tract, and should be completely absorbed. 

Bioavailability is largely excellent - hepatic metabolism of these drugs is relatively sluggish, and most of the dose will be able to make it into the bloodstream unmolested. 

Distribution differs across the class. Volumes of distribution are fairly diverse, ranging from 0.1L/kg for ibuprofen

to 5-6L/kg for celecoxib. This property seems to vary depending on the affinity of these NSAIDs for plasma proteins: ibuprofen commits rather firmly and completely to binding albumin, whereas others may promiscuously bond with other tissue proteins. All of them are highly protein-bound (well, the only black sheep of the class here is paracetamol, which is only 15-20% protein-bound, but it's not a "real" NSAID anyway, so that does not count). This has toxicological implications: there's only so much protein to go around, and a massive overdose will result in the saturation of binding sites. In such a scenario, the increase in the free fraction of the NSAID will make more molecules available to do harm. It will also expose more of these free molecules to the mechanisms of renal clearance, as well as to dialysis. Moreover, when one does a drug level (for example, in the case of salicylate, which has a widely available plasma assay), the total level ends up being rather meaningless, as it does not reveal the magnitude of the free fraction and will therefore be totally unrelated to the magnitude of clinical toxicity.

Metabolism is largely hepatic for all of these agents, with a few notable features. Most NSAIDs undergo extensive hepatic metabolism and are completely biotransformed into water-soluble metabolites. The unusual are:

• Diclofenac: of which only about 70% is metabolised into really excreted metabolites, and about 30%  is cleared via biliary excretion, which results in enterohepatic recirculation: gut bacteria break down the excreted glucuronides, liberating the original drug molecule for reabsorption.
• Parecoxib: which is itself an inactive pro-drug, hydrolised rapidly into valdecoxib, a highly active COX_2 inhibitor with poor water solubility. Over the first 20 or so minutes, most of the dose of parecoxib is converted in this way. Valdecoxib is then metabolised in the liver like a normal NSAID, with really excreted inactive metabolites.

Paracetamol also deserves a special mention here, as its metabolism is mainly by glucuronidation and sulfation (~70%), with a minor pathway that becomes important in overdose when the other enzymes are saturated. This side-pathway via the CYPE21 system produces NAPQI, a toxic byproduct that needs to be conjugated with glutathione in order to make it into something benign.  

Clearance of inactive metabolites is boring and renal. Only with massive overdose do we ever get to consider things like urinary ion trapping, where the weakly acidic NSAID may be purposefully made more water-soluble by alkalinising the urine (thus preventing it from returning into the circulation). Normally,  none of this plays any role. The enterohepatic recirculation of diclofenac may also make it susceptible to multi-dose charcoal, as it decreases the amount of drug available for reabsorption.

Half-life is polarised between older and newer agents. Old drugs like ibuprofen diclofenac and paracetamol have very short half lives (with a much longer duration of effect due to their COX-inhibiting effects). Newer agents like celecoxib and valdecoxib have longer elimination half-lives, allowing for once-daily dosing in the case of the latter. In the event that anybody anywhere requires a longer list of half-lives with more detail, this one from Day et al (1983) could be of use, even though it omits the 'coxibs (as they weren't invented yet). Fortunately, there's really no reason to learn any of this, as the plasma half-life is completely unrelated to the drug's half-life in the compartment of interest (say, the synovial fluid). Moreover, it has no relationship to the duration of their binding to their molecular target, or to the decreased levels of inflammatory mediators which result from their primary mechanism of action. Which brings us to:

Mechanism of action of non-steroidal anti-inflammatory drugs

Inhibition of cyclooxygenase enzymes is the defining pharmacokinetic characteristic for these drugs, and their selectivity for its various isoforms can be used to classify them.  Fitzpatrick et al (2004) and Hawkey (2001) are a couple of  good overviews of what these enzymes do, in case anybody is interested in more detail than what follows:

• Cyclooxygenases are a family of enzymes responsible for the formation of prostanoids out of arachidonic acid.
  • Arachidonic acid is a 20-carbon fatty acid that is embedded in cell membranes
  • Its metabolites (eicosanoids, after "eicoso", i.e. twenty) are numerous, and can originate from numerous non-COX-related pathways (eg. when arachidonic acid is metabolised by lipoxygenase).
  • Most of its metabolites have inflammatory or immunoregulatory roles
  • Pharmacologically relevant isoforms of COX include COX-1 and COX-2
  • COX enzymes have two functional sites, one of which is the cyclooxygenase site that converts arachidonic acid to prostaglandin G2, the other being the peroxidase site that reduces PGG2 to prostaglandin H2. This other site is occasionally confusingly referred to as POX, which makes you think that it is a whole separate enzyme.
• COX-1: the "constitutive" COX enzyme:
  • This is a ubiquitous enzyme which is present in basically all tissues, and its role is usually described as "housekeeping", i.e. it maintains various homeostatic baselines.
  • Pharmacologically relevant roles include:
    • The synthesis of PGE2 and PGI2, which promote increased bicarbonate and mucus secretion in the gastric mucosa
    • The synthesis of thromboxane A2 in platelets, which contributes to platelet aggregation
    • The synthesis of PGI2 in vascular endothelium, which inhibits platelet aggregation and which has a vasodilator effect
  • The active molecular binding site is small, which means only smaller molecules (eg. salicylates, ibuprofen, diclofenac) can bind there, whereas the larger COX-2 selective inhibitors cannot (Hawkey, 2001).
• COX-2: the induceable COX enzyme:
  • Of the two main isoforms, this is the one most involved in immune function and inflammation, as it is induced by the action of cytokines (IL-1, TNFα)
  • Pharmacologically relevant roles include:
    • The synthesis of PGE2 and PGI2 in inflamed or infected tissue, where they increase vascular permeability, produce oedema and sensitive peripheral nociceptors by decreasing their threshold
    • The synthesis of PGE2 in the hypothalamus, which sets the hypothalamic thermoregulatory set point to a higher temperature, producing fever
• Prostaglandin synthesis, in gross oversimplification, is:
  • Phospholipase A2 is the enzyme responsible for catabolism of membrane phospholipids to form, among other things, arachidonic acid.
  • Arachidonic acid gains two molecules of O₂ to form PGG2 under the influence of the cyclooxygenase site of a COX enzyne
  • PGG2 2 is reduced to PGH2 by two electrons by the actions of the peroxidase, sometimes abbreviated as POX.
  • PGH2 can then make thromboxane A2, prostaglandins (including PGE2) and prostacyclin, each pathway requiring its own specific enzyme.
• The interaction of NSAIDs with the COX enzyme system is isoform-dependent:
  • COX-1 inhibitors tend to bind to COX-2 in an irreversible fashion, which explains their prolonged duration of antiplatelet effect
  • COX-2 inhibitors tend to bind to COX-2 in a reversible fashion
• NSAIDs have some variability in the mechanisms of action:
  • Some NSAIDs, such as aspirin, irreversibly acetylate the enzyme and disable the cyclooxygenase (but not the peroxidase) function.
  • Other NSAIDs (most, including ibuprofen and indomethacin) reversibly compete with arachidonic acid for the COX active site (Vane & Botting, 1996) 

In summary, the specific analgesia which results from the use of NSAIDs is a mainly peripheral action (that's one of the reasons one can achieve a similar amount of relief by taking a diclofenac tablet and by smearing oneself with a diclofenac ointment). To paraphrase Cashman (1996), 

• Prostanoids bind to nociceptor nerve endings (usually via G-protein coupled receptors)
• This does not directly result in depolarisation of the membrane
• Instead, they indirectly increase the sensitivity of the receptor to other inflammatory stimuli, eg. bradykinin and histamine
• NSAIDs decrease the synthesis of these mediators and therefore decrease the sensitivity of the nociceptors.

To be different, and also to highlight how it really does not fit into this chapter, paracetamol does not really affect peripheral cyclooxygenase function in any major way "Unclear and multiple" or "complex and yet to be determined" is how professional articles explain its mechanism of action. The uncertainty is sufficiently blank that CICM examiners commented positively on it in their response to Question 1 from the second paper of 2017, suggesting that it would be important for trainees to admit we have no idea how paracetamol works. Sharma & Mehta (2014) offered a series of best guess theories, including "effects on prostaglandin production, and on serotonergic, opioid, nitric oxide (NO), and cannabinoid pathways". 

Spectrum of COX selectivity among NSAIDs

No piece of writing about the NSAIDs would be complete without a bar graph of COX selectivity, ranking agents from most selective to least selective. Here's a good one from Warner & Mitchell (2004), rotated slightly for convenience. 

One sees this thing a lot, which might give one the impression that it is somehow meaningful, but in fact it is only a fairly vague guide to the expected analgesic properties of these drugs, as many other factors need to be taken into consideration (eg. their penetration to the site of anti-inflammatory activity). Fortunately, we do have some empirical methods of doing this sort of comparison, borrowing the methodology from the world of opioids (where everything is compared to oral morphine). In the same fashion everything can be compared to the raw dose required to produce the same effect as 100mg of diclofenac:

The first column of this table from Moore (2020) demonstrates the DDD (defined daily dose) required to achieve the same analgesia as 100mg of diclofenac, which is a high bar (because that is a GORD-inducingly generous dose which one should only take if one is suffering from the serious musculoskeletal consequences of relocating adult tree ferns in a Corolla). From this table one develops the impression that one would require to take 200mg of celecoxib or 1200mg of ibuprofen to achieve the same analgesic effect, but in fact these data mostly come from Riendeau et al (2001) and Cryer & Feldman (1998), which were in vitro enzyme studies, and therefore functionally no different to the pointless bar graph in the preceding text.  Moreover even a short period of contemplation will lead the reader to conclude that it would be rather difficult to compare these agents in a way that covers all the diverse circumstances of their use. However this does bring up the topic of side effects, and these are probably more relevant to COX selectivity, and so:

Side-effects of non-steroidal anti-inflammatory drugs

A great resource for this specific topic is Wongrakpanich et al (2018), which focuses mainly on the NSAID side effect profile among the elderly, but to be fair a) the side effects in the young are not much different, and b) most of our patients are elderly.

Myocardial infarction and stroke: by intefering with the balance of prothrombotic and antiplatelet prostanoids (i.e. decreasing the production of PGI2, a vasodilator with antiplatelet properties), COX-2 selective NSAIDs can increase the risk of cardiovascular adverse events. The overall strength of COX-2 inhibition seems to be the most important determinant of cardiovascular risk (Hinz et al, 2007). They may also increase blood pressure, accelerate atherosclerosis, and decrease the cardioprotective effect of statins.  Rofecoxib and valdecoxib were so bad at this that they have been withdrawn from the market, and parecoxib exists only by virtue of stern warnings to give it as a single-shot perioperative dose.

The joys of extensive postmarkting data have allowed a glimpse at the magnitude of this risk, and to define it somewhat better. Bally et al (2017), for example, looked back on 446 763 individual experiences, of which around 60,000 ended up with myocardial infarction. The infarction was taken as the main outcome event, and the NSAID use pattern leading up to it was used to quantify the dose of exposure. The authors expressed their analysis in the form of intimidatingly massive tables, which would hurt the eyes of the reader, and which are therefore left behind this link. A better way to represent these data would probably be something like this weird plot, where the risk is represented by the fatness of the line:

where:

• Rofecoxib (low dose ≤25 mg, high dose >25 mg)
• Naproxen (low dose ≤750 mg, high dose >750 mg)
• Ibuprofen (low dose ≤1200 mg, high dose >1200 mg)
• Diclofenac (low dose ≤100 mg, high dose >100 mg)
• Celecoxib (low dose ≤200 mg, high dose >200 mg)

The magnitude of the relative risk increase is not displayed in this graph (because that kind of visual clutter is what professional peer-reviwed publications are for) but it does remain consistent across the bars, and for the purpose of context, in the case of high dose rofecoxib, it was 2.65. 

Nephrotoxicity: COX-1 controls the expression of vasodilator molecules in the microcirculation of the kidney, and excessive vasoconstriction here could definitely produce a decrease in the glomerular filtration rate. Weirdly, the extent to which this happens does not seem to be particularly closely related to the COX-selectivity of the drug. When Ungprasert et al (2015) performed a meta-analysis of these substances' influence on the risk of developing AKI, they found that pooled risk ratios were basically the same for both selective and unselective inhibitors (and modest, only about 1.5-2.0)

Exacerbation of CCF: Following from the above, agents which decrease the glomerular filtration rate will also exacerbate any existing propensity towards fluid retention. It also does not help that the reduced delivery of salt to the macula densa will result in RAAS activation, the retention of sodium and water, vasopressin release and hypertension by action of angiotensin. Clearly, all of this will produce a situation where patients with CCF will suffer from worsening CCF symptoms, and borderline people previously functioning well in the community will decompensate, present to hospital, and be diagnosed with CCF for the first time. From Page & Henry (2000), it would appear that being on regular NSAIDs doubles your risk of being hospitalised for CCF.

Platelet inhibition often discourages the use of NSAIDs in various postoperative populations, who might benefit most from some non-nauseating analgesic options. Schafer (1995) and Tsoupras et al (2024) describe this thromboxane-dependent platelet aggregation effect in some detail. Each study done on this subject seems to be using a different methodology to assess each agent, and usually assesses no more than one or two agents at a time, making their results more difficult to pool. The best that could be done is an unsatisfying table like this:

NSAID	AA-induced Aggregation	ADP-induced Aggregation	Collagen-induced Aggregation
Ibuprofen	Strong inhibition	Moderate inhibition	Moderate inhibition
Diclofenac	Strong inhibition	Minimal/No inhibition	Minimal inhibition
Ketoprofen	Moderate inhibition	Strong inhibition	Strong inhibition (animal studies)
Naproxen	Strong inhibition	Moderate inhibition	Moderate inhibition
Paracetamol	Weak inhibition	Minimal inhibition	Minimal inhibition

Cronberg et al (1984), an ancient paper that remains relevant for as long as the drugs remain on the market, reported a more objective method of assessment, where the platelet aggregation was measured by an adrenaline assay in a primitive aggegometer. Considering that truly preposterous doses were used (1g aspirin, 100mg indomethacin, 200mg diclofenac, 800mg ibuprofen), the results are unimpressive:

From this, we can see that after 1 day, aspirin reduced the aggregation to less than a third of its original value, whereas for the rest, the platelet activity only dropped by about half and then returned to normal later that same day. That is fair, one may say, and suggests that we should not be using these agents close to the date of your spinal operation. But what of the selective inhibitors? Graff et al (2007) looked at this in non-selected patients, and Scott et al (2014) specifically through the lens of neurosurgery, observed the effects on platelet aggregation for rofecoxib, celecoxib, and meloxicam. Minimal effects were observed, such that even the neurosurgical author was forced to admit that "these findings do suggest that patients on COX-2 inhibitors can likely be safely taken for surgery immediately". Celecoxib specifically had been given in vast doses (1200mg/day) to healthy volunteers with basically no effect on platelet function.

Gastric ulceration: NSAIDs have multiple effects which result in an increased risk of GI bleeding, which are expertly summarised by Musumba et al (2009). The most commonly quoted exam answer would probably be their COX-1 inhibitory effect (leading to decreased acid production and bicarbonate secretion) but in reality there are numerous mechanisms all of which lead to the degradation of the protective barrier.  The risk of bleeding from an upper GI source is something like four to five times higher in a population of chronic NSAID consumers, according to Wongrakpanich et al. That risk is often simplified as 1% over 6 months. Rodríguez & Tolosa (2007) went further by ranking these agents along a scale of relative risk increase, which resulted in the following table:

But what if you co-treat them with a PPI, one might ask? Surely nobody would leave their mucosa defenceless, knowing that NSAIDs are about to be blasting at it. Ray et al (2007)  looked at this question and established that the concomitant use of a PPI reduces the risk by half. Interestingly, and not surprisingly, selective inhibitors both caused fewer GI side effects and benefited least from the PPI.

A commonly encountered question is, does the oral formulation somehow increase the risk of gastric ulceration, just because it passes through the stomach? On one hand, one might expect that any systemic absorption of the drug would lead to the same level of GI side-effects, as the same circulatory system delivers the drug everywhere. On the other hand, one cannot deny that NSAIDs are well absorbed by the stomach and duodenum, being highly lipid-soluble, which means the most vulnerable site of potential ulceration will be maximally saturated with their evil molecules. Thus, according to Musumba et al (2009), there is both a systemic and a topical effect. After entering the gastric mucosal cells, NSAID molecules can be trapped there by ion trapping, resulting in a concentrated and sustained local effect. In other words, yes the enteric route matters.