Evidence-Based Medicine Versus Personalized Medicine
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
GUEST EDITORIAL Evidence-Based Medicine Versus Personalized Medicine Are They Enemies? Jose de Leon, MD E vidence-based medicine (EBM) and personalized medicine (PM) are fashionable concepts fre- quently endorsed by medical and pharmacological journals. Everyone appears to agree that both should be implemented in the real world of clinical practice as soon as possible. The author believes that it will not be easy to implement both simultaneously because their approaches collide. Whereas the EBM approach focuses on using randomized clinical trials (RCTs) to establish the best treat- ment for the average patient and ignores the outliers, PM focuses on the outliers (Table 1). To appreciate this, one must reckon with the statistical concept of heterogeneity, which seems widely ignored. Feinstein1 (both a physician and a scientist with mathematical training) pointed out that statistical heterogeneity is largely overlooked by statistics textbooks. In the view of the author,2 it is not surprising that statistical heterogeneity is ignored. First, clinicians have no idea what sta- tistical heterogeneity means; they just want their statistician to analyze their data and produce the right answer. Second, statisticians receiving data from clinicians do not want to raise the possibility that the clinician’s data are heterogeneous because then it could not be analyzed with the usual statistical methods or at least could not be analyzed collectively. The author believes that, when dealing with drug response, EBM assumes that the ‘‘reality’’ of the drug response is statistically homogeneous; thus, an average response represents individuals quite well. Personalized medicine, on the other hand, assumes that reality is not homogeneous because the average response is not representative of a sample of individuals, and drug response is a statistically heterogeneous phe- nomenon (Table 1). How is this misunderstanding possible? Is EBM coming from Mars and PM from Venus? Yes. These 2 concepts have completely different origins; EBM was developed in the context of medical education, and PM in the context of clinical pharmacology. Moreover, the battle between EBM versus PM may just be another skirmish in the larger ongoing wars between those who think medicine is a science and those who think it is an art.3,4 The conflict can also be exemplified by other battles such as those between defenders of (1) empirical observation versus mechanistic dis- ease models,5 (2) probabilistic and empirical thinking versus deterministic and explanatory think- ing,6 (3) evaluation of interventions versus discovery and explanation,7 and (4) public health versus individual patient health.8,9 This article briefly describes (1) the history of EBM, (2) the recent realization that statistical heterogeneity may be an obstacle in EBM’s attempts to summarize medical practice, (3) the history of PM, (4) the author’s opinion that the level of heterogeneity or the lack thereof is that which dictates whether PM can easily coexist with EBM, (5) the history of attempts to explore drug response heterogeneity in psychiatry, and (6) proposed next steps for moving beyond this stalemate between EBM and PM. HISTORY OF EBM The developers of EBM, Sackett and coworkers,10 proposed that EBM had its origins in the first attempts to use statistics in clinical practice in 19th-century France. Evidence-based medicine has been compared in particular with the ‘‘medical observation’’ developed by Pierre Charles Alexandre Louis in Paris during the 1830s and 1840s, who crusaded against the use of bloodletting in pulmonary infections, using the observational method and numerical calculations.5,11 Gerber and coworkers12 argued that EBM is fundamentally different from the mere attempts to introduce statistics in medi- cine in 19th-century France, because its fundamental innovation is that EBM relies on and enhances From the University of Kentucky Mental Health Research Center at Eastern State Hospital, Lexington, KY; and Psychiatry and Neurosciences Research Group (CTS-549), Institute of Neurosciences, University of Granada, Granada, Spain. Received April 20, 2011; accepted after revision December 27, 2011. Reprints: Jose de Leon, MD, UK Mental Health Research Center at Eastern State Hospital, 627 W Fourth St, Lexington, KY 40508 (e-mail: jdeleon@uky.edu). Copyright * 2012 by Lippincott Williams & Wilkins ISSN: 0271-0749 DOI: 10.1097/JCP.0b013e3182491383 Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 www.psychopharmacology.com 153 Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Guest Editorial Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 TABLE 1. Statistical Assumptions on Drug Response Underlying EBM and PM EBM PM Statistical homogeneity Assumed to be present Assumed to be absent Statistical heterogeneity Ignored Assumed Mean Represents sample well Does not represent sample well Outliers Ignored Crucial Stratification before randomization Not needed Ideal for studying outliers Analyses for HTE Not needed Needed if stratification was not done HTE indicates heterogeneity in treatment effects. a more equal relationship between physicians rather than the a more experienced physician, ideally somebody who was an historical tradition of relying on experts. This appears to agree ‘‘expert.’’ The traditional approach in learning medicine was with one of Sackett’s13 lesser known articles, in which he that of rotating with a mentor who taught the student physician stressed that, when compared with the traditional method of the art of medicine; then the physician practiced by himself educating physicians, EBM ‘‘puts a much lower value on and acquired experience as he/she made his/her own mistakes authority.’’ (sometimes lethal to the patients) and, in the case of a few, Despite these disagreements about antecedents, everyone became a physician mentor. Thus, using the traditional way of agrees that EBM was mainly developed at McMaster University educating physicians, the older the physician, the wiser and in Canada in the 1980s and 1990s14Y16 by Guyatt and Sackett, more experienced he/she was supposed to be.25 The EBM ap- among others. Sackett contributed to EBM’s dissemination by proach has inverted this process because older physicians tend moving to Oxford University.14 Evidence-based medicine be- to be less experienced with the updates provided by the EBM came a mainstream concept in medicine after the publication of approach.25 2 articles, one in JAMA16 in 1992 and one in the British Medical Evidence-based medicine can be seen as the culmination Journal10 in 1996. According to Vandenbroucke,17 however, the of the introduction of the scientific method in medicine26 be- British Medical Journal, Annals of Internal Medicine, and cause, during the last 500 years, greater scientific knowledge JAMA have greatly championed EBM, whereas The Lancet and has progressively been introduced in the mentoring of physi- the New England Journal of Medicine have kept some distance. cians. In the 20th century, the development of the RCT approach In a recent review updating the progress of EBM, Guyatt and its progressive adoption by government drug agencies for and coworkers18 gave credit to Cochrane, Sackett, and Feinstein marketing drugs have made the knowledge gained from RCT for laying its foundations. They credited the Scottish epidemi- available and, more importantly, have led to the combination ologist Archie Cochrane for insisting that the clinical disci- of available RCT in the so-called meta-analyses. The realiza- plines should summarize evidence concerning their practices.19 tion that RCT and meta-analyses should be the cornerstone They acknowledged their colleague Sackett for developing for medical decisions and medical education has led to EBM. the teaching innovations. They also credited Feinstein for de- According to Feinstein,27 this new ‘‘faith’’ is expanded mainly fining the principles of quantitative clinical reasoning. It is in- by epidemiologists, whereas physicians usually defer to epi- teresting that Feinstein was recognized because he was critical demiologists or other experts, because they rarely master the of EBM until his death and helped to develop a patient-centered ‘‘secretive’’ art of summarizing average drug responses using approach.20 meta-analytic techniques. The believers in EBM rarely recog- According to Sackett and Rosenberg,21 the rapid growth nize a major practical problem: the RCTs that are used to test of RCTs, the slow pace of updating textbooks, the lack of time drug efficacy and gain Food and Drug Administration (FDA) physicians have for keeping up with journals, and the lack of approval for marketing usually deal with short-term drug re- efficacy of continual medical education in improving clinical sponse in otherwise healthy and uncomplicated patients who competence explain the need for the development of EBM. are also willing to enter RCT. Physicians, however, often deal According to Woolf,22 the main historical factor contributing with chronically ill patients who usually take multiple medica- to the introduction of EBM was the wild variation in medical tions and can be uncooperative in taking medications. Thus, practice for the same types of patients, including overuse (and physicians are frequently interested in drug response after sometimes underuse) of services leading to increased costs, a many months or years of treatment in all types of patients.28 major issue for Cochrane.20 Sackett and coworkers10 defined More recently, an attempt has been made to increase the repre- EBM as ‘‘the conscientious, explicit, and judicious use of cur- sentativeness of RCTs by conducting practical or pragmatic rent best evidence in making decisions about the care of in- trials29 that focus on effectiveness rather than efficacy, but there dividual patients. The practice of evidence-based medicine means are very few or no pragmatic clinical trials for most medi- integrating individual clinical expertise with the best available cal problems. The author works in an acute care hospital for external clinical evidence from systematic research.’’ Most would patients with severe mental illness; it has 2000 admissions per agree that the heart of EBM is the reliance on RCTs as the best year, and nearly all its patients would be excluded from prag- alternative for guiding medical knowledge.23 More recently, matic trials because of the risk of suicide attempts (half of the Reilly24 very wisely has acknowledged that EBM is 3 different patients have previously attempted suicide) and/or the inability/ things: (1) a scientific hypothesis, (2) an ever-evolving body of unwillingness to sign a complex consent form because of se- evidence, and (3) an idealized way of practicing medicine. vere cases of psychosis, mania, or depression.30 Thus, ‘‘current In the opinion of this writer, EBM is a definitively a de- EBM’’ does not include well-controlled studies for the types parture from the prior 2500 years in the sense that traditionally of very sick patients treated in his hospital. Knottnerus and the study of medicine was primarily based on mentorship with Dinant31 wisely stated that ‘‘medicine-based evidence is a 154 www.psychopharmacology.com * 2012 Lippincott Williams & Wilkins Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 Guest Editorial prerequisite for EBM’’ and that ‘‘future research methods must and not on scientific knowledge. The current use of the term find ways of accommodating clinical reality, not ignoring it.’’ personalized medicine can be explained by advances in clinical The sad truth is that the more difficult the patient, the less pharmacology.45Y48 Meyer45 has presented the most compre- evidence is available to treat him/her. It is very unlikely that hensive historical review of these pharmacological advances; RCTs will ever be conducted for the types of patients seen here, only a brief overview of how these advances have led to in the author’s hospital because these patients combine se- the development of PM is presented. At the beginning of the vere mental illnesses with extensive history of suicidality, high 20th century, around the time the word ‘‘genetic’’ was first levels of comorbid substance abuse, and high levels of medical used, Garrod49 hypothesized that there was a ‘‘chemical indi- comorbidity.32 viduality,’’ and ‘‘those idiosyncrasies with regard to drug’’ response may be explained by evolution. In the 1950s,45Y48 STATISTICAL HETEROGENEITY AND EBM pharmacologists observed that some severe adverse drug reac- tions (ADRs) occurred only in a small number of patients. These Although in 1997 Feinstein and Horwitz33 emphasized the ADRs included hemolytic anemia after the use of primaquine, problem of lack of homogeneity in RCTs, only recently have apnea with succinylcholine, and peripheral neuropathy with iso- statisticians34,35 started focusing on the issue that the average niazid. In 1959, Vogel50 coined the term pharmacogenetics. In patient may not be a good representative of all patients. There is the 1960s to 1970s45Y48 2 phenotypes, poor metabolizers and no general agreement on how to best deal with this complex extensive metabolizers, were described for several drugs includ- problem. The best and most sophisticated way is through strat- ing tricyclic antidepressants (TCAs), debrisoquine, sparteine, ification,34 separating patients according to a characteristic that and mephenytoin. In the 1980s to 1990s,45Y48 (1) the various has a crucial effect on drug response and then randomizing the genes associated with the cytochrome P450 (CYP) isoenzymes groups separately. The number of RCTs utilizing this coura- were discovered, (2) the allele variants associated with poor me- geous approach36 may increase in the future, but it is likely to tabolism were described, (3) the ultrarapid metabolizers were be a slow process if the RCTs are designed by drug companies; described in patients taking TCAs and were associated with a they may not be interested in taking risks because stratifica- gene duplication or multiplication, and (4) other pharmacoki- tion decreases statistical power and therefore may require larger netic genes began to be identified. recruitment samples. When heterogeneity is not considered a In the late 1990s, advances in genetics allowed parallel priori and resolved by a design using stratification, it is pos- genetic testing51 (or testing for multiple genes at the same time) sible to deal with it only a posteriori, using subgroup statistical through the use of the so-called DNA microarrays. Thus, the analyses to test for heterogeneity in treatment effects,35 which new term pharmacogenomics was coined for the study of all is a rather controversial subject among statisticians. genes that may influence drug response.52 In 1997, the devel- Regarding a posteriori classification, a statistician with opment of DNA microarray technology led Science53 to define long experience in working on industry RCTs, Senn,37Y41 has ‘‘personalized prescription’’ as ‘‘tailoring drugs to a patient’s provided a comprehensive critique on heterogeneity. He criti- genetic makeup’’ and to predict that personalized prescription cizes EBM’s defenders by reminding them that RCTs provide would ‘‘soon’’ reach clinical practice. The race for the human information only on averages and that the calculation of an av- genome and the political decision in 2000 by President Clinton erage number, such as the number needed to treat, assumes that that the race was over54 led to further hype in lay55 and scien- all patients benefit equally38,41; and he incorporates the think- tific journals56 about the promises of genetics to personalize ing from a PM expert in his solutions.42,43 Senn reminds37Y41 us medicine. The generalized use of personalized prescription that an RCT has 3 main sources of error variability besides the would begin in 2015, according to Time magazine,55 or 2020, differences between treatments (the average differences between according to JAMA.56 The first decade of the 21st century has treatments over all randomizations). They include (1) between- led to greater acknowledgement of difficulties and complexi- patient variability (the average differences between patients), (2) ties at 3 levels: (1) genetic science, (2) the use of genetic DNA patient-by-treatment interaction (the extent to which difference microarrays in clinical practice, and (3) the development of between treatment differs from one patient to another), and (3) other microarrays. The complexity of the increasing genetic within-patient error (the variability shown from treatment period knowledge led to the realization that the function of approxi- to treatment period when the same patient is given the same mately one third of human genes is unknown and that dupli- treatment).38 Senn38 says that parallel RCTs cannot distinguish cations and other variants, the so-called copy number variations between the 3 types of error variability and crossover RCTs and epigenetics, may influence drug response.57 The intro- (in which each patient receives only the treatment and control duction of the first DNA microarray approved by the FDA for conditions) cannot distinguish between patient-by-treatment pharmacogenetic testing in 20069,58 has led to very limited clini- interaction and within-patient error. He notes that differences in cal use for reasons that are complex.59 The further development drug response may include genetic variability but genetic vari- of DNA microarray technology has allowed the testing of mul- ability cannot exceed individual variability shown by patient-by- tiple other products besides DNA, including proteins, RNAs, treatment interaction in RCTs.38,41 In addition, the same data and lipids and to the development of new diagnostic branches cannot be used to test for differences between groups in RCTs such as ‘‘protenomics,’’ ‘‘transcriptomics,’’ and ‘‘metabonom- and then used to classify patients retrospectively.37,41 Senn ics.’’ All of these new techniques called ‘‘omics’’60 have brought et al40 also point out that the assumption that pharmacogenetics additional expectations about their use in PM and also the ra- is important in drug response is largely untested.42 pid realization of the complexities of the scientific and prac- tical issues involved in taking these microarrays to clinical HISTORY OF PM practice.61,62 Physicians have traditionally practiced PM in their at- In the time frame of 3 to 5 years, these technologies, tempts to decide the best treatment for each of their patients.44 microarrays and genotyping methods, continue to advance and However, they were not using the term ‘‘PM,’’ and they were make prior versions obsolete by becoming more powerful (pro- probably basing their traditional PM on their subjective pre- viding more and more dataVup to millions of pieces of data) ferences or their understanding of what was best for the patient, and progressively cheaper. To make these advances more suitable * 2012 Lippincott Williams & Wilkins www.psychopharmacology.com 155 Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Guest Editorial Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 for clinical practice, ‘‘better’’ and ‘‘cheaper’’ are important, but better representation of drug response than EBM, one must as- advances in making these tests easier to use are also crucial. In sume that a good number of subjects are outliers and should be that sense, the introduction of nanotechnology in genotyping excluded from the usual statistical analysis and that mean re- tests may be an important step.63 The FDA approved a warfarin sults do not represent them; therefore, outliers should be studied pharmacogenetic test using this technology in September 2007.64 separately (Table 1). This technology can provide genotyping results in a few hours, To illustrate, 3 outlier prevalences in the population are con- providing busy clinicians with rapid answers without the need sidered in this discussion as examples: less than 1% of the pop- of waiting 1 to 2 days for the usual genotyping test before starting ulation, 1% to 10%, and 50%. If the outliers are so rare as to a new medication. include less than 1% of the population, they are quite unlucky; In the view of the author,62 personalized prescription is it is very unlikely that well-designed studies using RCTs will the branch of PM focused on individualizing drug prescription. ever be conducted in them. This is the tragedy of EBM: the To predict drug response, one has to use genetic information more difficult the patient is, the more likely that he/she will be in the context of environmental and personal factors65; fur- ignored by the EBM approach.65 An example may clarify this: thermore, these other factors may be more important than ge- some very unlucky subjects do not have CYP2D6 and CYP2C19. netics in determining drug response with some drugs. Thus, They number approximately 1 per 1000, or fewer in each race.69 personalized prescription should be built around the idea that If you are one of these unusual subjects, it may be better for you pharmacology is a mechanistic science. Chemistry (and physics) not to need an antidepressant because almost all of them are is foundational in understanding pharmacological response, but mainly metabolized by 1 of these 2 enzymes (and frequently the pharmacologists who are familiar with pharmacological mech- other CYP is used as an auxiliary pathway).70 If you are a anisms at the cellular, tissue, and organic levels, rather than pharmacologist interested in antidepressants and are trying to chemists, are the ones who can predict drug response.66 Un- recruit 50 of these double poor metabolizers for a controlled fortunately, pharmacokinetic and pharmacodynamic mechanisms clinical trial, you have the Herculean task of needing to screen vary from drug to drug.62 In a view of personalized prescription 50,000 patients to identify 50 who meet the study criteria. based on pharmacological mechanisms, one can propose the On the other hand, when ‘‘unusual’’ subjects include 50% following62,67,68: (1) a drug’s pharmacokinetic and pharma- of the population, such as with gender, there may be ‘‘too many codynamic mechanisms are behind its efficacy and safety; (2) outliers.’’ Let’s imagine that one scientist believes males can genetic, environmental, and personal variables influence drug respond to a new drug but most females do not respond. He/she pharmacokinetic and pharmacodynamic mechanisms and through will have the difficult task of convincing a pharmaceutical com- them its efficacy and safety; (3) personalizing drug selection is pany to develop this drug for only half of the market, in this case, much more complex than personalizing drug dosing; (4) in the males, when the competing drugs are approved for both sexes. process of personalizing drug selection, eliminating a drug in the If the outliers (patients who show heterogeneity) account process is easier than choosing a drug; (5) personalizing dosing for 1% to 10% of the population, the possibility of studying them is easier when the drug follows linear kinetics and has a nar- is dependent on resources and is influenced by the number of row therapeutic window; and (6) personalizing dosing in wide- groups used in the randomization or for the tests of heteroge- therapeutic-window drugs may not make practical sense because neity in treatment effects, the percentage of outliers, and the use physicians may be very arbitrary in their selection of drug dosages. or nonuse of some kind of enrichment strategy for increasing the number of outliers randomized into the study. In summary, it is a question of balancing financial cost and the practicality STATISTICAL HETEROGENEITY OR LACK of RCTs. OF HOMOGENEITY AND PM The crucial role of the level of statistical heterogeneity in PM and its implications related to the statistical power of the THE HISTORY OF THE ATTEMPTS TO EXPLORE studies has not been addressed by the literature. The first ques- TREATMENT RESPONSE HETEROGENEITY tions regarding personalizing prescription for a specific drug re- IN PSYCHIATRY sponse are: Is it completely heterogeneous (each patient responds Psychiatry has a long history of trying to identify predictors differently)? Is it somewhat heterogeneous? Is it homogeneous of differential pharmacological response, but these attempts (most patients respond similarly to the drug)? Thus, knowing the began before the EBM approach was born, were not developed statistical distribution of drug response is absolutely crucial in in the context of RCTs, and were not tested in RCTs. These the design of PM studies. If drug response is absolutely het- attempts came from a different tradition, the mechanistic tra- erogeneous (it lacks homogeneity) and cannot be extrapolated dition. The developers of these models hypothesized that psy- from individual to individual, it is impossible to study or pre- chiatric drugs work by modifying specific pharmacological dict unless we focus on the prior response of the same drug in mechanisms in the brain. Two of the most important attempts are the same patient. As pharmacokinetic and pharmacodynamic briefly described: the existence of a serotonergic versus norad- mechanisms are fairly similar across all humans, probably be- renergic type of depression and the dexamethasone suppression cause of the effects of individual selection associated with evo- test (DST). lution, drugs with nonhomogeneous response do not appear to The monoamine hypothesis of depression was first for- exist. Thus, patients appear to be relatively homogeneous in mulated in the 1960s,71 and some evidence that different anti- drug response, with differing levels of heterogeneity. However, depressants may have differential effects on serotonin and heterogeneity varies from drug to drug because of the pecu- noradrenalin metabolites72 led to efforts to classify depressive liarities of variations in pharmacokinetic and pharmacodynamic patients according to pharmacological mechanisms. In that sense, mechanisms. If one wants to defend EBM as being a good rep- in 1975, Maas73 simplistically hypothesized that there are 2 resentation of drug response, one must assume that the mean groups of depressive patients: A (with disorders of norepineph- represents the population well and that the standard statistical rine systems) and B (with disorders of serotonin systems). This test can be properly used because the sample shows statistical approach was described early in the first attempts to develop homogeneity (Table 1). If one wants to defend PM as being a pharmacological guidelines in psychiatry,74 before the birth of 156 www.psychopharmacology.com * 2012 Lippincott Williams & Wilkins Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 Guest Editorial the EBM movement. These guidelines74 were developed by ex- companies have no interest in conducting studies once a drug is perts using a comprehensive approach including the mechanis- on the market and especially when it is off-patent. Companies tic and RCT approaches, trying to balance all biological and developing PM tests have very small research budgets when clinical data and therefore giving a lot of weight to data beyond compared with pharmaceutical companies.61 In theory, pharma- those obtained from RCTs. cogenetic studies may help to bring increased use of some of In psychiatry, the DST was the most important biomarker the classic antipsychotic or antidepressant drugs by reducing used as a potential index of heterogeneity of treatment response. ADRs83 or may even rescue drugs withdrawn from the market.84 The initial DST research studies in depression in the 1970s75,76 In practice, the lack of funding and the enormous cost of any led to enthusiasm with which mainstream psychiatry embraced type of study that may produce competition with approved drugs a biological test of the heterogeneity of pharmacological re- on the market and supported by the pharmaceutical company sponse. However, in 1987, a guideline published by the Amer- marketing are serious limitations. ican Psychiatric Association Task Force on Laboratory Tests77 indicated the lack of definitive data of the DST’s clinical use- Studying Variability Between Individuals fulness in selecting treatment. In 1988, Nierenberg and in Drug Response Feinstein78 used the history of the DST, a diagnostic test initially The literature describes 3 methods for exploring the dif- widely accepted and later rejected, as a cautionary example for ferences in drug response among individuals: (1) pharmacoge- avoiding similar problems in the future with diagnostic tests. netic twin studies, (2) repeated drug studies, and (3) replicate These 2 promising but, in the end, unsuccessful attempts crossover trials after RCTs. to use biological markers in psychiatry are prime examples that The typical means of studying genetic versus environ- the history of addressing heterogeneity in treatment response mental influences in complex illness has been the use of twin in psychiatry has been rather unproductive. Thus, all recent studies to compare differences between dizygotic and monozy- American Psychiatric Association practice guidelines79 incor- gotic twins. In classic formulations, monozygotic twins are said porate EBM principles but also consider clinical consensus. to share 100% of their genome, whereas dizygotic twins are said These guidelines address treatment response heterogeneity, but to share on average only 50%; in both types of twins, the twin they correctly point out the limitations of the data. and his/her cotwin were believed to share similar levels of en- vironmental influences.85,86 Statistical models have been devel- oped allowing causal associations to be made and estimations NEXT STEPS FOR MOVING FORWARD FROM of heritability to be defined as the proportion of total variance THE CURRENT STALEMATE BETWEEN that is accounted for by genetic components.85 The assumption EBM AND PM of shared environment has always been dubious, but advances This editorial has presented the unsatisfactory status of med- in genetics are beginning to question the genetic assumptions ical science due to the stalemate between EBM and PM and of the twin models because monozygotic twins may not share presented a caricatured view of their traditions to keep the in- 100% of their genome, and they may have differences in copy terest of the reader. number variations or epigenetic changes.86 However, twin stud- To move forward, 3 issues appear important and are de- ies are considered to be a powerful method with high potential for scribed in the next 3 sections: (1) advances in scientific meth- studying some aspects of drug response, but there are prac- odology that would bring EBM and PM closer; (2) education of tical limitations because it is not ethical to administer drugs pro- the defenders of EBM and PM about the weaknesses of their spectively in long-term studies to healthy twins under controlled approaches and the virtues of the other side; and (3) the need for conditions. The twin model can be used for short studies with openness toward advances in science, in general, that may rescue little risk, particularly using single doses and focusing on phar- medicine from this stalemate. Many specific solutions are pre- macokinetic effects.87 As a matter of fact, recently, some twin sented in these 3 sections; unfortunately, many of these ideas pharmacological studies have focused on the pharmacokinetics are not likely to be implemented. Thus, the discussion of each of some medications or substances of abuse.85,88,89 Besides specific idea is concluded with a comment on its limitations. these controlled studies, ‘‘natural experiments’’ may occur when monozygotic twins take the same drug but show a discordant drug response (eg, ADRs). In summary, twin studies have some ADVANCES IN SCIENTIFIC METHODOLOGY THAT potential to explore the heritability of some aspects of drug re- WOULD BRING EBM AND PM CLOSER sponse and advance scientific knowledge in PM but are limited Additional development is needed in the following areas: by practical aspects.85 (1) scientific methodologies for PM, (2) knowledge of drug- Kalow and coworkers42,43 proposed investigating response response heterogeneity, (3) introduction of a PM approach to variation by repeating drug administration to the same individ- RCTs, and (4) use of a PM approach in the development of uals and comparing the variability of response within and be- guidelines for EBM. tween individuals (intraindividual vs interindividual variation). Interindividual variation in response to the same drug may have Development of Scientific Methodology for PM environmental and/or genetic reasons. Intraindividual drug var- Regarding the development of methodology for PM, it is iation cannot be explained by genetic variations; it can only be important to remember that RCTs are usually funded by in- explained by instability of measuring methods (test-retest reli- dustry, whereas most of the pharmacogenetic studies using ge- ability), chance, environment, or changes in genetic function in- netic testing to expand PM have been conducted by research duced by environment (epigenetics).43 For this type of study, it institutions.80 Two excellent reviews describe some possible types is important to use reliable measures, exclude chance by using of designs for pharmacogenetic studies including RCTs.81,82 As enough relatively large samples, and eliminate known environ- long as no funding sources are identified for these studies, this mental factors that influence drug response, including those that discussion of design types for researching methodological prog- may change drug response over time (such as drug tolerance, ress in PM is mostly a theoretical exercise. Grant agencies have which is probably mediated by epigenetic mechanisms).57 The traditionally left drug study funding to industry. Pharmaceutical comparison of intraindividual variation versus interindividual * 2012 Lippincott Williams & Wilkins www.psychopharmacology.com 157 Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Guest Editorial Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 variation provides a measure of heritability. More recently, double the risperidone dose compared with the other manic Kalow’s colleagues90 have (1) clarified the types of drugs that patients taking noninducers (lithium or valproate). can be studied using this method; (2) proposed that a compari- son of intersubject versus intrasubject variability in drug re- Personalizing EBM Guidelines sponse under minimal environmental exposure may provide an Guidelines for EBM can be more PM-friendly by (1) incor- upper-bound estimate of heritability of drug efficacy and safety; porating a more patient-centered approach, (2) incorporating a and (3) indicated that modest changes in population averages mechanistic pharmacological approach, and (3) rethinking the may underestimate the dramatic impact of a genetic variation hierarchy of grading recommendations, particularly for drug at the tails of a population. According to Dorne,91 the effects safety. of polymorphic enzyme outliers may contribute to huge differ- The editors of a family medicine journal101 have proposed ences in metabolism up to a factor of 52 times in CYP2D6 and that it is possible for guidelines to emphasize patient-centered of 26 times in CYP2C19. outcomes rather than focus on illness-centered outcomes. Clin- Senn,41 a statistician, has taken Kalow’s proposal42,43 one ical pharmacologists and pharmacists have incorporated mech- step further by proposing the use of what he calls replicate anistic knowledge in the development of pharmacogenetic crossover trials. They are a particular type of crossover design: guidelines102Y104 and EBM guidelines for diagnostic tests.105 The those in which the same treatment is given to the same subject approach is different than the approach of traditional EBM guide- more than once. A replicate crossover trial is the equivalent of lines because the emphasis is not so much on the availability doing ‘‘n-of-1 trials’’ in all the patients included in the RCTs. He of RCTs to support these recommendations but also incorpo- also indicated that the data need to be analyzed using mixed- rates the available mechanistic information in the guidelines. effects models. This study design is possible only in drugs used Obviously, if one thinks that RCTs are the only way to establish for chronic illnesses, with reversible effects and no need for long reliable knowledge in medicine, these types of guidelines in- treatments.38 More importantly, it will make RCTs much more corporating mechanistic knowledge are worthless. expensive. Mant92 opines that it is naive to think that replicate Vandenbroucke,106Y109 a physician with epidemiological crossover trials are a feasible solution and that we should resign training, has emphasized that the hierarchy of innovation in med- ourselves to using observational studies to answer questions icine is exactly the opposite of the hierarchy used by EBM in about applicability to individual patients. guidelines. Randomized clinical trials are the least innovative, In summary, pharmacogenetic twin studies, repeated drug and anecdotal findings are the most innovative. More importantly, administration studies, and adding n-of-1 trials after RCTs can he emphasized that EBM forgets that RCTs were developed to be used to establish an upper-bound estimate for individual test drug efficacy, not drug safety. As a matter of fact, Ioannidis110 variations (and possibly for heritability) of different aspects for proposes that ADRs are ‘‘neglected, restricted, distorted, and some drug responses, but they have rarely been used and have silenced’’ in RCTs. According to Vandenbroucke,106Y109 RCTs, some practical limitations. Thus, the current status of RCTs is because of their nature, are limited in detecting ADRs because that they can tell us which treatments are effective but not nec- they include (1) a small number of patients (a few hundred), (2) a essarily which patient should receive them.93,94 limited population (patients with comorbidities and poor health are excluded), and (3) a relatively short duration (usually a few Personalizing RCTs months). Randomization is not helpful or important for unex- Predictions based on mechanistic knowledge of genetic pected and unpredictable ADRs. As a matter of fact, observational or environmental influences on drug responses can be incor- studies (case reports, case series, and pharmacoepidemiological porated in the design of RCTs. An example would be to stratify studies) have been crucial in establishing new ADR knowl- patients by genetic markers according to our knowledge of phar- edge.109 Thus, strictly from the best scientific methodological macogenetics. The CYP2D6 gene variations are the best un- point of view, EBM’s hierarchy of putting RCTs at the top for derstood; one extreme is the poor metabolizer who does not drug efficacy may be correct, but for drug safety, RCTs may not have any CYP2D6 because of lack of functional genes, and the be particularly helpful. This is demonstrated by the withdrawals other extreme is the ultrarapid metabolizer who has too much from the market of many drugs that were proven to be safe in CYP2D6 enzyme because of 3 or more active genes. Ideally, RCTs and were not safe when used in the general population.111 RCTs stratified by CYP2D6 genotype should be carried out It is also demonstrated by an aldosterone RCT that suggested for classic antidepressant and antipsychotic drugs that are off- that this drug was associated with decreased mortality in ad- patent, but these RCTs will not be supported by the industry. vanced congestive heart failure, leading to an EBM guideline Furthermore, the industry has eliminated CYP2D6 drugs from and a subsequent increase in hospitalizations with no reductions their pipeline9; no future CYP2D6 drugs will be studied using in mortality when aldosterone was used in the general population CYP2D6 knowledge to stratify RCTs. of this type of patient.112 These examples of major problems in Our knowledge of the influence of inhibitors and inducers extrapolating from the patients studied in RCTs to the general can also be incorporated in RCTs. The industry, however, usu- population reflect that RCTs may have major problems with ally prefers to exclude patients taking inducers or inhibitors external validity.113 from their RCTs and ignore the issue completely. A good ex- ample of ‘‘ignorance’’ in psychiatry is the failure of risperidone as an adjunct treatment in mania compared with placebo when EDUCATING EBM AND PM DEFENDERS risperidone was added to carbamazepine in an RCT.95 A case REGARDING THEIR WEAKNESSES AND report,96 case series,97,98 and in vitro studies99,100 had demon- THE VIRTUES OF THE OTHER SIDE strated that carbamazepine is an inducer of risperidone me- It is fitting to be reminded of Reilly’s24 description that tabolism, but the pharmaceutical company’s ignorance of the EBM is an idealized way of practicing medicine. Four physi- literature led to a partial failure (in the carbamazepine arm) in cians with epidemiological expertise are included in this edi- this RCT.95 To obtain a positive outcome in the carbamazepine torial and the accompanying references: Feinstein,1,20,27,33,78 arm, the company should have paid attention to this prior mech- Ioannidis,60,110,114Y116 Sackett,10,13,16,21 and Vandenbroucke.5Y7, 17,106Y109,117 anistic literature and randomized the carbamazepine patients to If all physicians were like them, it would be very 158 www.psychopharmacology.com * 2012 Lippincott Williams & Wilkins Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 Guest Editorial easy to implement EBM. Unfortunately, we (physicians) do not Any pharmacologist knows that selective serotonin reuptake in- have the superb mathematical minds of these 4 experts. Evidence- hibitors (SSRIs) frequently exhibit no linear relationship be- based medicine assumes that physicians have the time and knowl- tween dosage and plasma concentration and wide ranges between edge to perform complex computer searches and complete therapeutic and toxic doses, and some are powerful self-inhibitors meta-analyses to resolve their clinical questions. The reality is of their metabolism.65 This means that SSRIs are not good that completing meta-analyses is not ‘‘easy’’ science. A statisti- candidates for CYP genotyping.65 The TCAs are much better cian has recently complained that several of the published meta- candidates for CYP genotyping. However, when the epidemi- analyses in top medical journals have significant flaws.118 ologists from the Agency for Healthcare Research and Qual- The writer spends a substantial part of his time helping ity120,121 decided to examine the clinical utility of CYP practicing psychiatrists and psychiatry residents with their prac- genotyping in psychiatry, they selected SSRIs instead of TCAs. tical pharmacological problems. This includes considerable ef- Not surprisingly, the outcome was not good for PM defend- fort in teaching them to (1) use PubMed to search for reviews ers.120,121 Pharmacokinetic knowledge in the CYP area would and meta-analyses and (2) use pharmacokinetic and pharma- have saved the effort and expense of completing a comprehen- codynamic knowledge to resolve their questions. He has had sive review and a meta-analysis, but researchers with an EBM some success in these tasks, but he has had no success in teach- background120,121 did not see the need to review the literature ing them basic statistical principles such that they could in- on the pharmacokinetic science of SSRIs. telligently assess the typical article published in a psychiatric Medical science includes epidemiological principles and journal. The McMaster University students and physicians may basic mechanistic science; medical practice is the difficult art be particularly sophisticated in statistics and epidemiology, but of taking this knowledge to clinical practice with individual the average US practicing physician and resident may not be patients. Because of the complexity of the task, physicians need as capable. This is why Feinstein27 insisted that the science of to rely on and learn from epidemiologists and basic scientists, EBM was not developed by physician scientists but rather by including pharmacologists. Thus, it is obvious that medicine scientists who were not physicians. A deceptively easy solution needs both EBM and PM, and collaboration between defenders would be to increase the mathematical training of prospective of EBM and PM is needed to advance medicine. Defenders physicians. In the United States, premedicine curricula and med- of each view must acknowledge the limitations of their ap- ical school entrance examinations emphasize biology, chemistry, proach, their lack of expertise in the other approach, and that and physics, with little emphasis on statistics or mathematical collaboration is needed. Teagarden122 used a clinician’s way of thinking. One can propose substituting the emphasis on physics thinking by describing a risk-benefit analysis to justify the use with an emphasis on statistics. Is this likely to happen? No. Is of warfarin pharmacogenetics despite the current lack of avail- this wise? That is uncertain. The psychology of science119 is in able RCTs. its initial stages, but there may be a dichotomy of personality traits among different individuals according to their career se- lections. Clinicians tend to be people-oriented, and scientists tend ADVANCES IN SCIENCE IN GENERAL THAT MAY to be oriented to things.119 Mathematicians, engineers, and phys- RESCUE MEDICINE FROM THIS STALEMATE icists may have lesser people skills than social scientists. It would Three major new advances in science may be relevant for not be surprising, then, if increasing the mathematical require- those who think that ‘‘outside’’ help could resolve this stalemate ments for medical students led to different populations of pro- in medicine: (1) the progressive understanding of how science spective students, with less interest in people and fewer social advances, along with its limitations; (2) a new way of scientific skills. thinking called ‘‘complexity’’; and (3) the initial steps in using The reality is that the application of pharmacological sci- the scientific approach for defining an expert. entific advances is a complex process. In the United States, An interesting advance that is finally being accepted by where there are few physicians with formal clinical pharmaco- the experts is that scientific thinking (1) has been very successful logical training (versus European countries that have clinical but has its weaknesses, (2) has no completely clear demarcations pharmacology residencies), the mechanistic science of phar- from ‘‘pseudoscientific’’ thinking, and (3) is not so different macology is mainly developed by pharmacologist scientists and from the thinking used in other sophisticated human activities some pharmacists with basic skills; physician scientists make (such as law or the arts).117,123Y125 In summary, science is a com- minimal contributions. In the area of RCTs, some US physician plex trial-and-error historical process led by experts, the scien- scientists may be the first authors of articles, but most of the tists.117,123Y125 Thus, the influence of subjective and personal complex statistical designs and analyses are done by biostat- issues cannot be avoided, to the point that major advances in isticians or epidemiologists working for the industry. The science can be explained by a complex mix of (1) ‘‘charge’’ (dis- meta-analyses of pharmacological studies are usually authored coverers solve problems that are quite obvious but in which the by biostatisticians or epidemiologists. On the other hand, in the way to solve the problem is not so clear), (2) ‘‘challenge’’ (dis- United States, most of the prescribing is done by practicing coveries are explained by a new acknowledgment of the lim- physicians (or physician extenders) who have very limited phar- itations of scientific thinking, and (3) ‘‘chance’’ (serendipitous macological and statistical training. If these practicing clinicians findings made by ‘‘prepared minds’’).126 Ioannidis,114Y116 a phy- have access to a pharmacist and are wise, they consult the phar- sician with mathematical training, has conceptualized for med- macist, who could provide a more comprehensive and mecha- ical science the limitations of scientific thinking and the nistic pharmacological view in treating problematic patients. In difficulties in advancing it. In a series of seminal articles, he has summary, the author believes that using EBM and conducting reminded us that the majority of published medical findings are meta-analyses for making decisions about drug prescription are false, including many of those obtained through RCTs.114,115 an ‘‘ideal’’ way of practicing medicine but are beyond the abil- He also stated that ‘‘important’’ articles and ‘‘important’’ jour- ities of current US physicians (and physician extenders). nals are not free from the problem of spreading false findings; Many of the scientists practicing EBM and writing meta- moreover, they are even more prone to it because of the com- analyses appear not to understand that theories and mechanis- petition to publish ‘‘significant’’ findings.115 The ‘‘boring job’’ tic interpretations are a fundamental aspect of medical practice. of replication,127 the very cornerstone of establishing a scientific * 2012 Lippincott Williams & Wilkins www.psychopharmacology.com 159 Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Guest Editorial Journal of Clinical Psychopharmacology & Volume 32, Number 2, April 2012 finding as truth, is discouraged. Finally, Ioannidis pointed out single patient by using a special type of randomized design that significant biases, both personal and financial, contribute (called sequential multiple assignment randomized trial). An- to the dissemination of false findings.116 In this context, it seems other regimen, a regression-based analysis called Q-learning, important to remind the defenders of the PM and EBM ap- may allow exploring similar data in nonrandomized naturalistic proaches that their fields may not be so different in that both are long-term treatment designs.136 full of false findings; both are biased, but their varying traditions An even more ‘‘futuristic’’ scientific development may be result in different biases. Working together and using the strengths science advancement to the point that one can study complex of the other side to balance their weaknesses appear wise be- human concepts such as ‘‘what is an expert.’’137 The scientific cause, according to Naylor,128 ‘‘clinical medicine seems to con- study of how individuals become experts should be important sist of a few things we know, a few things we think we know (but for both: (1) the defenders of EBM, who believe that training probably don’t), and lots of things we don’t know at all.’’ with an ‘‘expert’’ was a key tradition in the ‘‘art’’ of medicine and A new current in scientific thinking is called ‘‘complex- that EBM is ‘‘revolutionary’’ in its attempt to make all physi- ity,’’129,130 reflecting the complexity of scientific exploration cians more expert and more scientific; and (2) the enemies of and the need for complex computer models. This theory and its EBM, who believe that EBM is a part of social process in which accompanying computer models are being introduced in medi- the ‘‘old’’ authoritarian structures are hidden behind statistical cine, the so-called ‘‘network’’ medicine.131 Network analysis may techniques that still need to be interpreted by ‘‘experts.’’93,138 have potential to explore the myriad of genes that might con- Physicians do not talk about who is an expert, but they tribute to PM. Two applications are briefly discussed here: (1) demonstrate it with their behavior. In a simplistic way, in clini- gene-gene interactions and (2) ethnicity. When one is sure that cal medicine, ‘‘experts’’ are probably those physicians who are a single gene has important and definitive effects on response consulted on difficult cases by other physicians from the same to a specific drug, the task of probing it is ‘‘relatively’’ simple; specialty. From the point of view of medical trainees, ‘‘experts’’ one only needs to establish how all personal, environmental, are those who are more frequently chosen as ‘‘mentors’’ by the and epigenetic changes modify the activity of the gene and de- young physicians who want to learn medicine. In the scientific termine the clinical relevance of these changes. Therefore, one area, because of large amounts of data available from publica- assumes that this gene has ‘‘large effect sizes’’ on the drug’s tions, a more ‘‘statistical’’ and ‘‘numerical’’ approach toward de- safety or efficacy so that other genes can be ignored. However, termining who is an expert is possible by testing how influential the ‘‘elephant in the living room’’ of PM is how to deal with the the researcher’s articles are. An example of these attempts may possibility that myriads of gene-gene interactions are likely to be the index ‘‘h’’ developed by Hirsch139 to compare, using be important. The problem of dealing with gene-gene interac- one number, the most frequently quoted authors in a scientific tions may be mind-boggling if one moves from 2 to 3 genes to area. As English is definitively established as the lingua franca 20 or 50 genes interacting. Network analyses,132 by developing for disseminating research, and journals from countries other gene pathways, may be a reasonable way of dealing with un- than majority-English countries are increasingly well-read by known complex gene-gene interactions. researchers, this type of index may help to establish the identi- From the genetic point of view, the concepts of race or ties of the ‘‘current’’ world experts in a specific area of publi- ethnicity do not make sense, but there are obvious genetic dif- cation. Unfortunately, there is no guarantee that the current ferences related to ancestry, because of the evolution of human medical experts would be the ones considered ‘‘experts’’ when populations.133 Unfortunately, what we know about the func- the researchers of the next century are looking back. Moreover, tionality of genes used in PM (eg, CYP2D6 and CYP2C19) history tells us that great discoverers are usually ignored by their cannot be predicted easily by ancestry markers. Network anal- contemporaries. The scientific approach has not been particu- ysis134 has been used recently to control for ancestry, but it is larly successful in studying and explaining some of the more not known whether this will be a successful method to deal with complex concepts of human life, such as expertise and creative this complex issue. Badcott135 has argued that putting too much spirit, but researchers from different areas including educational faith in the idea of multiple genes facilitating PM may be mis- sciences are trying to define what an ‘‘expert’’ is.137 Even with leading. The aleatory process of evolution ensures that, with the these advances in understanding what an expert is and how to possible exception of monozygotic twins, every human being is train them, one must hold out the possibility that inherent lim- genetically unique and that predictive power may be enhanced itations exist in what science can offer for better training experts by knowledge and technology, but is ultimately constrained by among physicians. According to Polanyi,125 a thinker who be- insuperable probabilistic considerations. It is illusory to relate gan as a physician, then became an internationally known chem- population-level probabilistic causation with individual-level ist and finally a philosopher of science, the process of learning causation. medicine with a mentor requires some implicit learning, which Only time will tell if these ‘‘complexity’’ and ‘‘network’’ is by nature difficult to translate into words and is learned only models are a passing fad or a contribution to individual patient by example.140 care through the incorporation of complex information from multiple sources, the final goal of PM. Certainly, it seems hard to convince current practicing physicians that in the future spe- CONCLUSIONS cific treatment decisions will be made by ignoring their ‘‘sub- Prior articles have described the tension between EBM and jective’’ experience and introducing hundreds or thousands of PM regarding economic issues141 and practical implementation pieces of data from genetic or other biomarkers into a computer in the clinical environment142Y146; one has even proposed that model that will ‘‘magically’’ provide the answer. The word ‘‘mag- EBM has ‘‘been effectively sidelined and marginalized’’ by ically’’ is used here in the sense that physicians will have little PM.147 clue as to how the software works or was developed. Another This editorial focuses on historical and statistical issues that new statistical methodology that may be relevant but is even cannot be ignored in understanding this tension. The more we harder to discuss than network analysis, because of its novelty, believe that PM is needed for one drug, the more this means that is the so-called dynamic treatment regimens.136 These regimens drug response is not homogeneous and that an unmodified EBM allow individualization of treatment (type, dosage, timing) of a approach will not be helpful. If we decide that EBM is the way 160 www.psychopharmacology.com * 2012 Lippincott Williams & Wilkins Copyright © 2012 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
You can also read