Understanding clinical efficacy of drugs (1)

The clinical efficacy of any drug can be understood by visualizing the balance between its risk of producing limiting toxicity as compared to the chance of producing a beneficial clinical response. These two effects may or may not be mediated through the same receptor systems, and toxicity need not necessarily be due to a pharmacological overdose. Simplistically the clinically efficacy can be visualized by comparing two concentration-response curves (not necessarily parallel since they do not necessarily operate through the same receptor system), where one represents the beneficial response, and the other, the toxicity response.

The space between the 2 curves can actually plotted out to visually represent the clinical efficacy profile of the drug. Assuming a threshold of clinical efficacy exists, an empirical therapeutic 'window' may be identified, within which we can try and keep drug concentrations for optimal efficacy. I personally do not like the term window as it suggests the best approach might be to target the middle of that 'window'. I prefer 'therapeutic range' because the therapeutic strategy might actually be to exploit the range of concentrations so as to maximize the efficacy for the individual patient.

The shape of the clinical efficacy plot obviously will depend on the shape of the individual beneficial and toxicity curves. Likewise the therapeutic range will depend on the acceptable threshold of clinical efficacy for that specific drug.

Pharmacokinetic variability

Most doctors do not fully appreciate just how much variability there exists in the population with respect to how their patients respond to a standard dose of the drug they use. Even without considering the variability in terms of the pharmacodynamics of the response, there is already a lot of variability. Here is a simple simulation of what a 'typical' plasma concentration profile might look like after a 8 hourly dosing regimen.

Looks fairly 'normal' doesn't it? But here (below), I have simulated a set of random profiles of a theoretical drug from a fictitious set of 20 patients with a very reasonable range of pharmacokinetic properties (body weight, clearance, volumes of distribution, bioavailability). These are values that one can reasonably expect from any patient. Just look at the wide range of plasma concentration profiles that can be found in a set of 20 'patients'. Whether these are eventually clinically important or not will of course, depend on the specific nature and the toxicity profile of the drug in question, but recognizing that such variability exists, is the first step to being able to personalize therapy.

The issue of clarithromycin and increased cardiac deaths #5 - How can we deal with the variability?

Clinical efficacy is never only about therapeutic efficacy. It is always a balance between benefits and risks. Drug response variability shapes the clinical efficacy curve by altering the relative distance between benefits and risks along the dose or concentration range. Good therapeutics therefore is always about being able to manage the variability in drug response, so that the optimal dose or concentration for the patient can be selected that will maximize benefits and minimize risk.

If we just consider the anti-microbial effects.... we actually manage the variability rather badly. Theoretically, there is a therapeutic target that is based on maintaining drug concentrations at the target site (where the bugs are actually growing) above the minimum inhibitory concentrations (MIC). By drug concentrations we refer mainly to the trough concentrations. Therapeutically, the assumption is that if we dose according to published guidelines, we will achieve target concentrations at the site of action. In reality, we do not know that for certain. In fact, we have absolutely no idea whether we are dosing too much or too little, and basically act on faith that a certain dose (quantum and frequency) will allow trough concentrations at the site of action to exceed the MIC.

Therapeutic drug monitoring of clarithromycin had been proposed, but has never ever been taken up seriously for it to be included in routine patient management.

This is fine, if there were no serious toxicities, because you can administer more drug than really necessary just to make sure you have adequate concentrations at the target site. However, clarithromycin use does carry a serious risk of toxicity....namely sudden death. Therefore the correct dosing schedule is important to deliver only adequate levels of clarithromycin so that the troughs are over the MIC and the peaks do not approach the IC10 of IC20 of HERG channel blockade.

Currently we have no means to doing this.

On the flip side of the coin, there is the problem of cardiac toxicity. Although there is a considerable gap between routinely achieved levels of clarithromycin and the IC10 or IC20 of HERG channel blockade, obviously this gap can sometimes, though rarely, be crossed. The Danish study suggests this might be happening at least in 37 out of 1 million dosing regiments. How do we monitor and manage this? Routine ECG to look for QT prolongation would certainly be helpful. But this is almost never done during clarithromycin use.

Effectively therefore we have no means to manage variability where clarithromycin use is concerned. There is an over-dependence on published dosage guidelines, and faith in the adequacy and safety of these guidelines. Am I surprised by the association with cardiac deaths? Definitely not. Understanding the pharmacology of clarithromycin, this risk is predictable. The risk is not high. But one sudden death occurring in a relatively healthy individual, no matter how infrequent, is one death too many.

Can we do better? Yes.

The issue of clarithromycin and increased cardiac deaths #4 - Where are the potential sources of variability?

1. Bioavailability
Regardless of its touted lipophilicity, clarithromycin has a reported average bioavailability of only about 50%. Generally, as a guiding principle, the lower the bioavailability, the greater the potential for variability in systemic availability.

2. Uncertain target site concentrations
There are two associated problems here.

Firstly, clarithromycin has an elimination half life of about 3-5 hours at low doses and 5-7 at higher doses. At a 12 hourly dosing intervals, there will be significant fluctuations in the plasma concentration profile. Even if it is administered at 8 hourly intervals, and if half-life is assumed to be at the high end of the range, say 8 hours, there will be at least a 2 fold fluctuation between peaks and trough. While this may meet the needs of anti-bacterial efficacy (assuming we keep trough levels above MIC), the levels of the peaks may predispose to cardiac toxicity if it is able to inhibit HERG potassium channels. To some extent, we can mitigate the fluctuations by using extended release formulations, but this may be at the expense of even more variability in bioavailability.

[source:http://medlibrary.org/lib/images-rx/clarithromycin-24/img-0238.jpg]

Comparison between normal formulation and extended release formulations

Secondly, since we do not routinely measure either plasma or tissue concentrations, we have little idea if adequate concentrations are being achieved at the target site. Here, there is some more uncertainty. Tissue and cellular concentrations tend to be higher than plasma unbound concentrations, but concentrations in the extra-cellular fluid (where the bugs are) are variable and may be lower than unbound concentrations of clarithromycin. These are functions of variable protein binding and the variable net activities of specific influx and efflux membrane transporters.

Consequent upon the previous two points, the differential effects of clarithromycin on the bacteria and on HERG channels may be variable between individuals not only because they relate to different effect compartments but the latter may relate to heights of the peak while the former to trough concentrations being above the MIC.

[Source:http://www.fda.gov/ohrms/dockets/ac/03/briefing/Image66.gif]

Relationship between QT prolongation ad clarithromycin concentrations

Although the IC50 for clarithromycin on the HERG channel is about a 100 times higher than the MIC, arrhythmic risk is associated with lower extent of inhibition. Hence cardiac risk is seen at much lower IC10 or IC20 concentrations

Added to all these, is the uncertainty contributed by an active 14-OH metabolite of clarithromycin.

3. Inter-individual variability in pharmacokinetics
Clarithromycin is both a substrate and inhibitor of CYP3A4. This metabolic pathway is also responsible to generating the active 14-OH metabolite. Variable CYP3A4 activity therefore results in a variable mix of clarithromycin and its active 14-OH metabolite.

There is a very high extent of variabilty in CYP3A4 activity in any population studied. There are also significant differences in activity between men and women (women generally higher). While there are genetic polymorphisms associated with CYP3A4, no single genetic variant has been able to account for the variability within a population. On the other hand, CYP3A4 is also vulnerable to many food and drug interactions.

To make matters more complicated, clarithromycin inhibits its own metabolism by CYP3A4, and exhibits a non-linear pharmacokinetic profile.

4. Inter-individual variability in susceptibility to QT prolongation
The HERG potassium channel gene is genetically polymorphic and variants may predispose to variable susceptibility to QT prolongation. Added to this is the uncertainty about appropriate dosing regiments between different ethnic populations, who may have different body weights and distributional volumes, as well as different exposures to CYP3A4 food and drug interactions.

5. Variability in microbial susceptibility
Apart from differences in anti-microbial efficacy due to variability in drug permeation to target sites, bacteria do differ in how susceptible they are to concentrations of clarithromycin. While sensitive bacteria generally have MICs in easily achievable range, resistance genes have become more prevalent and differences in bacterial sensitivity has become more common.

6. Compliance issues
One must never forget the variability that may be caused by failure of the patient to medicate according to instructions, leading to highly irregular dosing intervals and therefore variable degree of fluctuations in circulating drug concentrations.


Taking all these uncertainties into consideration, the question is how to ensure the patient gets optimal dosing? Think about it.

[To be continued]

Optimization? What's that?

Optimization refers to the situation where you can adjust the inputs into a system according to output functions, and eventually arrive at the best solution. In therapeutic terms, it is the process of calibrating the dosage of a drug according to the clinical response so that the best dosage for the patient can be arrived at depending on the therapeutic targets and the patient's individualized response.

This is really no different from the engineering concept of a control system.

When there is an input into a particular process, and no feedback is received about the outcome....this is called an 'open loop control system'. This is probably the least ideal of all control systems. It fundamentally assumes you know everything there is to know and that the decision taken about the initial input is already adequate. This happens in therapeutics when you have fixed dose regiments, and there is no feedback about the outcome. Consider the situation in most cancer chemotherapeutic regiments. Dosage regiments are pretty much determined at the outset, and the only feedback received is if the patient has obvious toxicity which requires cessation of therapy, i.e. switch off the system! This is essentially the pharmacogenomic approach towards 'personalized medicine'.

In other scenarios, there is possibility for the operator to make adjustments to the original input based on feedback received, but this takes place independent of the original model which determined the input. This is called an "open-loop feedback control system'. Most therapeutic scenarios are of this sort, where an initial decision about a starting dosage regiment is made based on starting knowledge and assumptions about the patient. Subsequently minor adjustments to the dosage regiment can be made by the physician depending on feedback received about the patient's clinical drug response. Where there is poor ability to receive feedback about clinical drug response, the system starts to flounder and approximates the simple open-loop control system.

The open-loop feedback control system can operate with varying degrees of sophistication. It can be empirical, where the response to feedback received is relatively intuitive and based on clinical judgement. Or it can be highly deterministic where the response is determined by precise mathematical (PK or PK-PD) models.

A more sophisticated model takes into consideration the uncertainty in the system. This is called a stochastic approach.

The ideal system is that of a closed loop system where the original models determining the original input is linked to, and continually modified by new feedback received. The following diagram represents a closed loop control system with warfarin as an example.

Modified from Applied pharmacokinetics & pharmacodynamics: principles of therapeutic drug monitoring. 1992 Michael E. Burton et al.

At top-right there is a population PK-PD model of warfarin. This represents what is known about how warfarin behaves in the population in which the patient exists. Together with the clinical model at top-left, of what the desired or target INR response is, a decision about the starting dosage regiment can be made.

Subsequently, feedback is regularly received about warfarin pharmacokinetics (bottom right) and INR response (bottom left). These feedback into the original models at top right and left, and continually adjust the model so that decisions are continually taken about how the dosage regiment can be adjusted.

This is optimization... and represents what personalized medicine ought to be.

Can it be done for all drugs? Yes, it can. But it requires that there are good population PK-PD models for the drug, and good biomarkers of response that can be used as feedback. It requires resources and effort.

Above all, it will require that physicians be prepared to put in the extra effort to optimize their therapy according the the patient's real requirements.

Predictive pharmacogenetics

Many people agonize about the predictivity of the science of pharmacogenetics. Somehow there is an expectation that pharmacogenetics should somehow lead to position where we can stop thinking.

Sounds a bit harsh but true.

Pharmacogenetics began as a science to understand the genetic basis for outlier behaviour. In much earlier experiences, outliers were characterized principally by phenotypic behaviour. As it is now, phenotypes tended to be classified in binary fashion - rapid/slow, fast/slow, extensive/poor. Such binary depictions of reality can only be predictive when the reaction or process in point is singular, critical or both. In some situations drug response can be described binarily as 'at risk for toxicity', or 'not at risk', e.g. G6PD deficiencies or HLA B*1502 for carbamazepine-SJS. For the most part however, pharmacogenetics data only helped explain genetic bases for limited aspects of drug response.

FDA classification: +, for information only; ++, recommended; +++, required

Gervasini et al, Eur J Clin Pharmacol (2010) 66:755–774

In highly controlled experiments, it can be easily shown that genetic variants can result in either loss or gain in function in specific processes related to drug response, e.g. drug metabolism and clearances. However since drug response/metabolism/

pharmacokinetics is seldom the result of a singular process, genotyping a genetic variant almost never provides a clear prediction of what the final drug response would be like.

Unfortunately the market place has misled many to believe that we can somehow construct a genetic testing panel that will predict with some degree of finality, what the patient's drug response and hence his drug dosage requirements will be. Hooray.... and we can therefore stop thinking! This line of thinking is clearly fallacious. The concept of 'personalized medicine' has been hijacked (biotech commercialism?) to refer to a one step genotyping approach to therapeutics when correctly it should refer to the ability to look at the patient in totality, i.e. the entire person - not just from the perspective of his constitutive make-up, but the totality of contributions of his altered physiology and environmental effects.

The more correct and rational approach is that of 'optimization', but the term sadly is far more mundane and less commercially sexy compared to 'personalized medicine'.

Warfarin - variability in response

Frequency distribution of warfarin daily dose requirement

Pharmacogenomics. 2009, 10 (12) :1955-1965

Warfarin presents a very good case study with respect to drug response variability, and the management of the uncertainty that surrounds the therapeutic use of warfarin.

Warfarin inhibits the reductase that recycles warfarin epoxide (Vit K epoxide reductase C1) so that it can be used again in the production of the Vit K dependent clotting factors. Conceptually very simple, but a number of things complicate this schematic. Firstly warfarin is optically active, and the two isomers, R and S warfarin, have different potencies and PK characteristics. The S warfarin has 5 times the potency of R warfarin and so often has been taken to represent the active ingredient of racemic warfarin. This is a convenient over-simplification, and it is by no means true that all warfarin activity is accounted for by only the S isomer. This is further

complicated by the fact that the isomers are metabolized preferentially by different CYP450 enzymes and have different elimination halflives.

S warfarin has quite a long halflife - an average of 40 hours. In some individuals it may be up to or longer than 60 hours. This means that after initiation of dosing, S warfarin doesn't achieve steady-state concentrations until about a week of dosing. Using a loading dose will get you closer to the steady-state concentrations, but will still need 5 halflives to settle into 'steady-state'. To make it worse, this does not even mean that warfarin's anticoagulant effects stabilize after one week. In fact the anticoagulant effects are not just dependent on warfarin kinetics but also on the kinetics of the clotting factors, which have their own halflives of elimination. This means that after warfarin steady-state is reached, some more time is required for the clotting factors, and consequently the fully anticoagulant effect, to settle into 'steady-state'. Simulations suggest that the whole process of anticoagulation may take up to about 2 weeks to reach steady-state.

Practically this means that the sooner you can settle into the correct maintenance dose, the sooner the patient will be at a stable level of anticoagulation. Every time you tweak the dose, it will require another 2 weeks to settle down. This makes dosage optimization particularly problematic.

The main sources of variability for warfarin response may be anticipated to relate to the following:

a] body weight

b] diet (Vit K supply, inhibitors.inducers of CYP enzymes),

c] smoking

d] genetics of CYP enzymes

-particularly CYP2C9 for S-warfarin, but cannot ignore other CYP enzymes involved with R warfarin.

e] genetics of CYP4F2 involved in breakdown of Vit K

f] genetics of Vit epoxide reductase complex 1 (VKORC1)

g] drug interaction with CYP enzymes

What saves the situation for warfarin is that it has an excellent direct measurement of drug response, - the INR (International Normalized Ratio) which directly measures the state of anticoagulation produced by warfarin. The INR allows a very convenient way to adjust warfarin dosages according to a 'target' level of response. This is called a target response strategy. For warfarin, drug level monitoring is of little use because of i) the delay in response because of the clotting factors halflives, ii) the presence of 2 active warfarin isomers, and iii) because there are different sensitivities to warfarin effects because of genetic variants affecting VKORC1.

Though the INR is a useful 'direct' measure of warfarin response, it is in reality only a 'surrogate' measure of the true warfarin efficacy, which is the eventual effect warfarin has in reducing morbidity and mortality associated with thromboembolism, strokes etc. These can only be assessed through monitoring therapeutic outcomes. However, these outcome measures, do not help us in the day to day optimization of the patient's warfarin dose.

Direct, surrogates or outcomes?

Students are quite often confused during discussions of these various types of efficacy measures. It is really not surprising, as many clinicians I discuss with also seem quite unclear about these concepts.

But these are to me quite important ideas; ideas which are quite often overlooked during drug development and the design of therapeutic regiments. And we do pay a price for neglecting them.

When we move from the bench to the bedside, we do lose the ability to assess drug response. Often we do not even begin to recognize just how much we have lost in our ability to do this. Yet it is vitally important for us to be able to do this because if we cannot, we will not be able to rationally manage our dosage regiments. This is one of the great difficulties in managing therapeutics.

In a relative small subset of therapeutic situations we do have direct measurements of drug effect - such as in the management of hyper/hypotension or hyper/hypoglycaemia. The blood pressure and blood sugar responses serve us well. A similar possibility exists with the management of the INR using warfarin.

In many therapeutic situations, no clear biological marker of drug response exists; or the therapeutic aims is actually far more complex compared to the immediate aspects of drug response. In these situations the desired drug response may actually be an 'outcome' measure - such as the control of epilepsy or arrhythmia, or even the management of depression and psychoses. In such situations, a surrogate for drug action should be available that can allow real time management of drug dosages. In some situations, measuring drug concentrations, can provide you with a reasonable surrogate for managing dosages. This is referred to as having a 'target concentration strategy', or more popularly called therapeutic drug monitoring.

Admittedly, good surrogate measures are often not even available. This deficiency creates a therapeutic environment where 'therapeutists' are often limited to relatively fixed, or inflexible dosage regiments, and therefore cannot deal effectively with any patient variability in drug response. This could be as comforting as having your pilot fly blind.

The recent problems with rofecoxib (Vioxx) provides an interesting example. While initially developed for the management of inflammatory joint disease, the selective Cox2 inhibitor found a new use in the prevention of intestinal polyps. When used as an anti-inflammatory analgesic, rheumatologists could manage drug dosages through assessing pain relief, joint involvement etc. When it came to the prevention of intestinal polyps, the therapeutist essentially had to 'fly blind' using fixed dose regiments, since there was neither direct nor surrogate measures of drug action. Intestinal polyposis was at best an outcome measure that could only be assessed at the end of treatment periods. Were there patients who over over-dosed or under-dosed? Very likely. Could we have better managed the dosages, and consequently the risks of cardiovascular mortality? Quite likely; but we will never know now. Rofecoxib was eventually withdrawn from the market.

A pity, perhaps.

Measures of efficacy

In the previous posts we had considered the optimization of warfarin dosages. The genotyping offers an added dimension to the optimization process, but may not be really cost effective. The warfarin problem is probably not a good example for predictive genotyping. This is because it already has a very good biomarker of clinical response - the INR (International Normalized Ratio). Most other drugs do not have good measures of clinical response. The absence of such measures of efficacy makes the optimization much more challenging.

Some drugs like warfarin allow direct measurement of the therapeutic efficacy. So for warfarin, it is the INR. For antihypertensive drugs, it could be the blood pressure response. For an antiasthmatic bronchodilator, it could be the airway relaxation, FEV1 for example. For an antidiabetic drug, it could easily be the blood glucose response. There are many other examples.

But for many therapeutic areas, the clinical response is much less directly quantifiable. For example, in the use of an antiepileptic drug, without a good surrogate measure of response, one is never really sure whether appropriate amounts of drug have been used. Likewise, for an antibiotic or a chemotherapeutic agent, the inappropriateness of the dosing regiment may not be recognized until it is too late.

For such situations, the availability of a 'surrogate' measure of response may be critical in patient care. One such surrogate measure is the measurement of circulating drug concentrations. Crude though it may be, getting the patient into a 'therapeutic range' may provide some guidance as to whether or not appropriate dosages have been used. Such an approach is sometimes referred to as the 'target concentration strategy'. There are however limitations to this method and it is not always applicable.

For the target concentration strategy to work, there must be evidence that circulating concentrations bear a good relationship to therapeutic outcome.

Genes vs environment

Not an easy topic, but here is an interesting recent guest column by Sandra Aamodt and Sam Wang, in the New York Times discussing the complex interactions between genes and the environment. Although they discuss this from a largely neuropsychiatric perspective the lessons are widely applicable to therapeutics. We have far too many champions of the genetic approach who push ideas that genetic variability underlie everything that determines drug efficacy and toxicity. The underfunded environmental approaches go largely ignored because they use rather unexciting mundane technology, and produce results that tend not to generate patents.

IMO the gene only approach is clearly not valid. The challenge is how to tease out the various gene-environment interactions and to define them clearly so that they can eventually help us optimize our therapeutic regiments.

Personalized medicine vs Genome-based medicine

I have a very good series of meeting as part of the Health Sciences Authority, Singapore (HSA) team with the Japanese Pharmaceutical and Medical Devices Agency (PMDA), Ministry of Health, Labour and Welfare (MHLW), and the National Institute of Health Sciences (NIHS), as well as the RIKEN Center for Genomics Medicine (CGM).

I learnt a lot just talking the the various agencies. But the highlight for the meeting for me was catching that fleeting comment by Prof Yusuke Nakamura, Director of Riken CGM, that there was a difference between "personalized medicine" and genome-based medicine", and that RIKEN had more of a focus on "genome-based medicine". I thought that was incredibly insightful. So many of the luminaries in PGx toss around the term "personalized medicine" almost as a justification for spending their multimillion $$ budgets but totally missing the point that what they are after really isn't personalizing therapeutics.

In our analogy of the F1 race car driver, it's really no different from developing better and more precise fuel injection systems, or tires that are better suited for the road surface. Technology is great. But after all that we musn't forget that one still needs to navigate and drive fast to get to the finish line, ... with the best time. That is personalized medicine.

See recent editorial: Personalized medicine: are we there yet?

The Great Durian Poll outcome

Many thanks to all who participated. We managed a half decent 66 responses... :).

I must say I was somewhat surprised by the results. As a non-durian lover I was expecting to see a much clearer separation of lovers and haters, with perhaps a more distinct bimodality in the distribution, somewhat like the taster/non-taster distribution. Instead we had a kind of log-normal distribution, like the CYP3A4/5 one.

There are a couple of possible reasons for this. One is that there may be a sampling bias, i.e. non-durian lovers aren't that motivated to participate. Secondly, the category axis is an ordinal scale, so even though I tried to space out the responses as 'equally' as I can imagine them to be, there is no certainty that the categories have equal intervals. It could well be that there is a larger separation at the "So-so only....no big deal" category.

In any case, it was an interesting exercise. As has been shown many times before, the lack of a clear 'bi- or poly-modal' distribution does not necessarily exclude any genetic bases for the interindividual differences.

Regardless of the genetic or lack of genetic basis for the interindividual difference, (and assuming the sample represents all of Singapore) there is an interesting lesson for us here...

Firstly, because of the preponderance of durian lovers in the sample, one can say Singaporeans generally love durians...passionately...though they can mostly live without it. Secondly and perhaps more importantly, is to recognize that despite such an overwhelming support for the spikey fruit, there are regulations that protect the interests of the people represented by right tail of the distribution. You don't allow the fruit on airplanes, in cars, shopping centres and restaurents. It is such a common sense thing to do, so we kinda take it for granted. It is actually a very common phenomenon. In a classroom, for example, the (good) teacher's attention is often focused on the poor students, or the bright spark...and not on the majority of the students who (on average) do not have any problems.

We have the same situation in dealing with therapeutic problems. Most dosage regimens are designed for the average patient (central tendency, remember?), and we know (or should know) that the patients who develop problems are those at the tails of the distribution...either inadequate response, or too much response/toxicity. Our mental focus should really be on helping the patients in the tails of distribution achieve an appropriate therapeutic response. Yet physicians often forget this and assume that the recommended (average) dose will meet the needs of all the patients they treat.

The challenge for us is in helping physicians identify which of the patients reside in the tails. This is where pharmacogenetics come in.

There is a nice review, "Pharmacogenetics - Tailoring Treatment for the Outliers" in the New England Journal of Medicine by Woodcock and Lesko that deals with this specific issue. It also reminds us of what Sir William Osler had shared over a hundred years ago: "If it were not for the great variability among individuals, medicine might as well be a science and not an art." Paradoxically, medicine is now at a stage of development where dealing with this variability has become much more of a science.

Rat poison and the F1 driver - the warfarin story

The anticoagulant warfarin actually started life as rat poison. Chemically, it is derived from a natural plant product, coumarin. The way it acts is by inhibiting the enzyme Vitamin K epoxide reductase, and in so doing reduce formation of various Vit K dependent clotting factors.

So what's the deal about F1 drivers?

Well... like F1 drivers, the physician using warfarin needs to keep his eye on the road. Too little warfarin, and there is inadequate therapeutic anticoagulation; too much warfarin and the patient may suffer a catastrophic bleed. Fortunately, he has a way to do this. The Prothrombin Time and other derived measures such as the International Normalized Ratio (INR) provide a heads up to the physician about how much anticoagulation has been provided for the patient. By keeping his eye on the INR, the physician can adjust the dose of warfarin to provide just the right range on anticoagulation the patient needs. This is important, because the warfarin requirements for every patient differ, and the warfarin dose needs to be 'individualized'.

More recently, various other biomarkers enable the physician to make educated guesses about the dosage requirement for the patient. These are genetic markers related to the rate of metabolic degradation of warfarin through cytochrome P450 2C9 (not many such problems in our Chinese population) and the genetically reduced sensitivity of the Vitamin K epoxide reductase C1 subunit (VKORC1). However, using these genetic biomarkers only provide an improved starting dose. Once the race car engine starts, the F1 driver will still have to manage the therapeutic process through keeping a close eye on the INR.

There have been many discussions about the genotyping of patients prior to dosing with warfarin. I have no doubt to its usefulness in helping us to understand the patient a lot better. But becasue there is already a good efficacy marker (the INR) for us to titrate the patients dosing against, the improved starting point may only be of theoretical benefit. I think most of the benefit will come in situations when you need to deliver very fast anticoagulation. Where time is not on an essence, genotyping would likely not be a cost effective option.

Therapeutics and the F1 race.....

There are quite a few variables operating in the context of an F1 race. Where the driver is concerned there are questions about his mental alertness, situational awareness, speed of reflexes, knowledge, experience and even his risk for appetite. Clearly there are also variables related to machine performance and conditions affecting road and track. All this variability create exciting conditions and an unpredictable outcome. Yet, no matter who wins eventually, most car-machine partnerships perform outstanding well.

The operation of any high performance machine system requires complex and sophisticated servo-systems operating at multiple levels. At its most simplistic level, the race car driver need to sense the speed of the car and adjust the speed through an interplay of acceleration and deceleration.

Precision therapeutics is really not any different. The physician must be able to recognize the effect of his procedures (drug regiment) and modulate these through the adjustment of his procedures (drugs dosages, for example). What is surprising is how many physicians do not recognize that they need to do this. They function like race car drivers who don't know their car, have no speedometer and are blinded. What's worse....don't even know they have brake and accelerator pedals.

For them, the patient is represented by a virtual description of a population average. Their expectation of a therapeutic effect is a relatively crude measurement of eventual outcome - cured, didn't work...died(often they don't even recognize they are driving blind). And they don't seem to realize, they can actually adjust drug doses in scientifically rational ways.

We can actually do a whole lot better than that. And many physicians have.

Highlighted report: Validation of VKORC1 and CYP2C9 genotypes on interindividual warfarin maintenance dose

Validation of VKORC1 and CYP2C9 genotypes on interindividual warfarin maintenance dose: a prospective study in Chinese patients.

Huang, Sheng-Wen, Chen, Hai-Sheng, Wang, Xian-Qun, Huang, Ling, Xu, Ding-Li, Hu, Xiao-Jia, Huang, Zhi-Hui, He, Yong, Chen, Kai-Ming, Xiang, Dao-Kang, Zou, Xiao-Ming, Li, Qiang, Ma, Li-Qin, Wang, Hao-Fei, Chen, Bao-Lin, Li, Liang, Jia, Yan-Kai, Xu, Xiang-Min

Objectives: To develop a warfarin-dosing algorithm that could be combined with pharmacogenomic and demographic factors, and to evaluate its effectiveness in a randomized prospective controlled clinical trial.

Methods: A pharmacogenetics-based dosing model was derived using retrospective data from 266 Chinese patients and multiple linear regression analysis. To prospectively validate this model, 156 patients with an operation of heart valve replacement were enrolled and randomly assigned to the group of pharmacogenetics-guided or traditional dosing for warfarin therapy. All patients were followed up for 50 days after initiation of warfarin therapy. The log-rank test was compared with the time-to-event (Kaplan-Meier) curves. Cox proportional hazards-regression model was used to assess the hazard ratio of the time to reach stable dose.

Results: The linear regression model derived from the pharmacogenomic model correlated with 54.1% of warfarin dosing variance. The final multiple linear regression model included age, body surface area, VKORC1, and CYP2C9 genotype. The study showed that the hazard ratio for the time to reach stable dose was 1.932 for the traditional dosing group versus the model-based group and a close and highly significant relationship was observed to exist between the predicted and the actual warfarin dose (R2=0.454).

Conclusion: A pharmacogenetics-based dosing algorithm has been developed for improvement in the time to reach the stable dosing of warfarin. This model may be useful in helping the clinicians to prescribe warfarin with greater safety and efficiency.

See also:

Estimation of the Warfarin Dose with Clinical and Pharmacogenetic Data.
The International Warfarin Pharmacogenetics Consortium.

BACKGROUND: Genetic variability among patients plays an important role in determining the dose of warfarin that should be used when oral anticoagulation is initiated, but practical methods of using genetic information have not been evaluated in a diverse and large population. We developed and used an algorithm for estimating the appropriate warfarin dose that is based on both clinical and genetic data from a broad population base. METHODS: Clinical and genetic data from 4043 patients were used to create a dose algorithm that was based on clinical variables only and an algorithm in which genetic information was added to the clinical variables. In a validation cohort of 1009 subjects, we evaluated the potential clinical value of each algorithm by calculating the percentage of patients whose predicted dose of warfarin was within 20% of the actual stable therapeutic dose; we also evaluated other clinically relevant indicators. RESULTS: In the validation cohort, the pharmacogenetic algorithm accurately identified larger proportions of patients who required 21 mg of warfarin or less per week and of those who required 49 mg or more per week to achieve the target international normalized ratio than did the clinical algorithm (49.4% vs. 33.3%, P<0.001,>/=49 mg per week). CONCLUSIONS: The use of a pharmacogenetic algorithm for estimating the appropriate initial dose of warfarin produces recommendations that are significantly closer to the required stable therapeutic dose than those derived from a clinical algorithm or a fixed-dose approach. The greatest benefits were observed in the 46.2% of the population that required 21 mg or less of warfarin per week or 49 mg or more per week for therapeutic anticoagulation.

Cost-effectiveness of using pharmacogenetic information in warfarin dosing for patients with nonvalvular atrial fibrillation.
Eckman MH, Rosand J, Greenberg SM, Gage BF.
University of Cincinnati Medical Center, Cincinnati, OH 45267-0535, USA. mark.eckman@uc.edu

BACKGROUND: Variants in genes involved in warfarin metabolism and sensitivity affect individual warfarin requirements and the risk for bleeding. Testing for these variant alleles might allow more personalized dosing of warfarin during the induction phase. In 2007, the U.S. Food and Drug Administration changed the labeling for warfarin (Coumadin, Bristol-Myers Squibb, Princeton, New Jersey), suggesting that clinicians consider genetic testing before initiating therapy. OBJECTIVE: To examine the cost-effectiveness of genotype-guided dosing versus standard induction of warfarin therapy for patients with nonvalvular atrial fibrillation. DESIGN: Markov state transition decision model. DATA SOURCES: MEDLINE searches and bibliographies from relevant articles of literature published in English. TARGET POPULATION: Outpatients or inpatients requiring initiation of warfarin therapy. The base case was a man age 69 years with newly diagnosed nonvalvular atrial fibrillation and no contraindications to warfarin therapy. TIME HORIZON: Lifetime. PERSPECTIVE: Societal. INTERVENTION: Genotype-guided dosing consisting of genotyping for CYP2C9*2, CYP2C9*3, and/or VKORC1 versus standard warfarin induction. OUTCOME MEASURES: Effectiveness was measured in quality-adjusted life-years (QALYs), and costs were in 2007 U.S. dollars. RESULTS: In the base case, genotype-guided dosing resulted in better outcomes, but at a relatively high cost. Overall, the marginal cost-effectiveness of testing exceeded $170 000 per QALY. On the basis of current data and cost of testing (about $400), there is only a 10% chance that genotype-guided dosing is likely to be cost-effective (that is, <$50 000 per QALY). Sensitivity analyses revealed that for genetic testing to cost less than $50 000 per QALY, it would have to be restricted to patients at high risk for hemorrhage or meet the following optimistic criteria: prevent greater than 32% of major bleeding events, be available within 24 hours, and cost less than $200. LIMITATION: Few published studies describe the effect of genotype-guided dosing on major bleeding events, and although these studies show a trend toward decreased bleeding, the results are not statistically significant. CONCLUSION: Warfarin-related genotyping is unlikely to be cost-effective for typical patients with nonvalvular atrial fibrillation, but may be cost-effective in patients at high risk for hemorrhage who are starting warfarin therapy.