Predicting Drug Interactions Outcomes—Do We Do Better Than Meteorologists?
"Prediction is very difficult. Especiallyif it's about the future." Thissounds like something Yogi Berramight have said, but it was actually theDanish physicist, Niels Bohr. Predictionis indeed difficult, especiallywhen one is trying to predict the clinicaloutcome of drug-drug interactionsin individual patients.
The deluge of information ondrug-drug interactions in the pastdecade—particularly in the area ofcytochrome P450 isozymes and morerecently ATP [adenosine triphosphate]-Binding Cassette (ABC) transporters—has demystified much of the seeminglyinconsistent behavior of interactingdrugs. We now can often predict whichdrugs are likely to interact with otherdrugs. Even in vitro studies of druginteractions are now providing usefuldata in assessing the interactive propertiesof drugs.
But these remarkable advances inour understanding of drug interactionmechanisms raise the question: Howuseful is this information for predictingthe clinical outcome in a patientwho begins taking a particular interactingdrug combination? The answer is,"Not very."This quandary—lack ofpredictability in clinical outcome—hasproven to be a particularly sticky pointin making clinical decisions aboutdrug interactions. We can, of course,often predict that the serum concentrationof a drug will be affected byanother drug. Clarithromycin, forexample, is very likely to increaseserum digoxin concentrations.1 Whatwe have trouble predicting is, forexample, of 10 people on digoxin whoare started on clarithromycin, which ofthem will develop clinical evidence ofdigoxin toxicity (and of those, whichwill be the most severe)?
Meteorologists have struggled withthis problem for years—as anyoneplanning a picnic is well aware—andthe similarities are striking. Althoughmeteorologists are often taken to taskfor their less-than-accurate forecasts,they actually do rather well in theirpredictions, their reputation forbungling being fueled by our selectiverecall. (We tend to remember their mistakesmuch more vividly than theircorrect forecasts, just as it appears tomost of us that we almost always selectthe wrong line at the bank.)
But what about drug interactions?Why is it so difficult to predict the clinicaloutcome when a patient takes aninteracting drug combination? It istrue that the clinical outcomes of a fewdrug interactions are relatively predictable.For example, if a patient stabilizedon carbamazepine begins takinga CYP3A4 inhibitor such aserythromycin, a high probability existsthat symptoms of carbamazepine toxicitywill appear. Similarly, meteorologistsare confident about some forecasts—on occasion they are willing tocall for a 90% chance of rain (notprone to taking unnecessary risks,however, they normally reserve "100%chance"for those times when it isalready raining).
But for most problems, either meteorologicalor pharmacological, the systemsare so complex—that is, subjectto so many variables—that truly reliablepredictions of outcome are notpossible. Is that likely to change soon?It does not appear likely. While meteorologistsare improving their predictionsby using better models and biggercomputers, they cannot accuratelymake specific predictions (eg, light rainwill begin in downtown Seattle at 9:27AM tomorrow, lasting until 11:52 AM).
Similarly, we are making progress inidentifying factors that affect the clinicaloutcome of drug interactions—pharmacogenetics, disease states, doseand duration of therapy, diet, dosingschedules, and the like. But accuratelypredicting the extent to which a particularpatient will have an adverseclinical outcome from a drug interactionusually takes far more than simplyknowing if the patient has identifiablerisk factors. Moreover, little doubtexists that some of what we currently"know"about drug interactions willeventually be proved false—and mostof the remainder will be found flawedby oversimplification. So we havemade a start, but only a rudimentaryone—the meteorological equivalent ofpredicting weather using only thebarometric pressure and humidity.
A Complex System
Weather has been used as an exampleof a chaotic system—completelydeterministic, but subject to an astronomicalnumber of variables. A subtlechange in one of these variables canlead to a chain reaction with majorchanges elsewhere in the world—theproverbial butterfly beating its wingsin New Delhi, India, causing a blizzardin Chicago, Illinois. Precise predictionof future weather in a particular placerequires absolute accuracy in the measurement of each of these variables—an obviously impossible task.
Predicting drug interaction outcomesis not so unruly, representinga "complex system"rather than achaotic one—the New Delhi butterflyis unlikely to cause a patient inChicago to bleed from a warfarindrug interaction. Nevertheless,while it is true that drug interactionoutcomes—like the weather—arecompletely deterministic, we knowonly a small fraction of the factorsaffecting the outcome in either case.Since we cannot begin to measureall of these variables—even if weknew what they were—we are stuckwith imprecise forecasts of bothweather and drug interaction outcomesfor the foreseeable future.
Given that predicting the clinicaloutcome of a drug interaction in aspecific patient is so often imprecise,what can we do until better modelsfor prediction are developed?
•Use the information we do have.Some risk factors for adversedrug interactions are known,and when they are, we shouldconsider them in making decisions—for example, the increasedrisk of hyperkalemia in apatient on an angiotensin-convertingenzyme inhibitor andpotassium-sparing diuretic whois also a diabetic with renalimpairment.2
•Do not make specific predictionswhen informing a prescriberabout a drug interactionin a patient. Even if the interactionis very likely to occur, it isusually best to point out thelarge variability in the clinicaloutcome of drug interactions.
•Do not make hasty decisionsbased on your own clinical experience.If you see a number ofpatients receiving a particularpair of interacting drugs withoutany adverse outcomes, rememberthe variability in outcome,and do not conclude—based on this informationalone—that the drug interactionis not clinically important.
Drs. Horn and Hansten are both professorsof pharmacy at the University of WashingtonSchool of Pharmacy. For an electronic versionof this article, including references ifany, visit www.hanstenandhorn.com.
For a list of references, send a stamped,self-addressed envelope to:References Department, Attn. A. Stahl,Pharmacy Times, 241 Forsgate Drive,Jamesburg, NJ 08831; or send an e-mailrequest to: email@example.com.