Jose de Leon: Training psychiatrists to think like pharmacologists
27. Evidence-based versus personalized medicine


Jose de Leon’s response to Donald F. Klein’s response

            I would like to start by giving thanks to Dr. Klein for his kindness and by emphasizing my two major points of agreement with him: antipathy toward commercial funders of randomized clinical trials (RCTs) and the crucial need for study replication in psychopharmacology. Then I will address four of his points in order from my least to greatest disagreement: 1) his Utopian wish that NIH take over RCTs to approve drugs, 2) the use of longitudinal mixed-effect models rather than ANCOVA, 3) the relevance or lack thereof of outliers and 4) the lack of data on the relevance of pre-stratification using pharmacological mechanisms. Finally, I mention one underlying conceptual issue: 5) the differences between mathematical and mechanistic science in psychopharmacology. 


1) Utopian wish that NIH take over RCTs to approve drugs.  I would love for Dr. Klein’s Utopian wish to become reality, but I am pretty sure that pharmaceutical companies are not going to give up control of the RCTs conducted to gain drug approval. My version of a Utopian wish is much less revolutionary than Dr. Klein’s wish; I merely want pharmaceutical companies to give up control of the package insert (or prescribing information), particularly after the drug has become generic, so that interested scientists and clinicians, using a Wikipedia-type of system under FDA supervision, can update the information for each drug (de Leon, 2011).


2) The use of longitudinal mixed-effect models rather than ANCOVA.  This time I am not going to make complex statistical comments, but try to illustrate the difference with a practical example. ANCOVA allows one to establish whether a confounding variable has a significant effect or not, but it does not provide its effect size, which is what a clinician needs to know. I have never been fortunate enough to have millions for completing drug RCTs, but with the help of my statistician collaborator and friend, Dr. Diaz, we have used longitudinal mixed-effects models to explore the effect of drugs involved in drug-drug interactions (DDIs) on dosing using a few prospective studies and several naturalistic studies with data on therapeutic drug monitoring (TDM). An inhibitor that decreases the metabolism of drug A by 50% requires a correction factor of 0.5 (to compensate for the effect of the inhibitor, the dosage of drug A needs to be halved). An inducer that increases the metabolism of drug A by a factor of 2 requires a correction factor of 2.0 (to compensate for the effect of the inducer, the dosage of drug A needs to be doubled).

            Edoardo Spina, an Italian pharmacologist who has conducted six prospective clozapine DDI studies, was kind enough to provide his TDM database with 415 clozapine concentrations from 215 patients. Using a longitudinal mixed-effects model, we calculated that, after correcting for another drug, on average the correction factor for phenobarbital was 1.5, for paroxetine 0.77, for fluoxetine 0.70 and for fluvoxamine 0.28 (Diaz et al., 2008). Then Dr. Diaz decided, without asking my permission, to use the model to test for an interaction between valproate and smoking. It was a very humbling experience, given my pharmacokinetic knowledge of valproate. Perhaps talking about my ignorance of valproate pharmacokinetics is better since only recently have I begged to understand the complexity of valproate pharmacokinetics (Jackson et al., 2015). Anyway, Dr. Diaz’s interaction analysis led us to find that in smokers, valproate appears to be a mild clozapine inducer (1.3 correction factor) and in non-smokers it appears to be an inhibitor (0.86 correction factor). These findings seem to explain why in some prior published clozapine studies valproate behaved as an inhibitor and in others as an inducer (Diaz et al., 2008). When Dr. Spina provided us with a similar database on olanzapine, we could not see a similar valproate effect on olanzapine, but this database had no repeated olanzapine concentration data from the same individual, on and off valproate (Botts et al., 2008).  Repeated data from the same individual is required by the longitudinal mixed-effects model to distinguish between intra- and inter-individual variability. Then, using internal funding, Dr. Spina conducted a prospective study of 18 patients on olanzapine in which valproate was added. The longitudinal mixed-effects model allowed us to determine that, in effect, valproate is probably both an inducer and a competitive inhibitor of olanzapine (Spina et al., 2009). Currently, we believe that the inductive or inhibitor effects of valproate on olanzapine and clozapine are influenced by serum valproate concentrations, time and individual vulnerability (Spina et al., 2016).        


3) The relevance or lack thereof of outliers.  The relevance of outliers depends on their prevalence (de Leon, 2012). Let’s assume four prevalences: 50%, 10%, 1% and 0.1%. If the prevalence is 50%, (e.g., when sex has a major effect on a drug response), I propose that stratifying by sex during randomization is the best approach but, if it is not done, heterogeneity analyses after the completion of the RCT should have no problem with statistical power.  If outliers are 10%, heterogeneity analyses after the RCT are not likely to have statistical power, particularly in typical RCTs with a few hundred patients, and especially if different drug dosages are used. In this situation, the best approach would be to enrich the sample of outliers and stratify by the outlier status in the randomization. If 1% are outliers, it would be better to exclude them but, if they are not excluded, it is not likely that they would cause any relevant headache to the investigator after the completion of the RCT. If 0.1% are outliers, the headache is for the patients and their families, since nobody will study them and no recommendation for their treatment would be provided in the literature. Unfortunately, this is not a hypothetical situation. Most antidepressants are metabolized by CYP2D6 and/or CYP2C19. Approximately 0.1% of subjects in most races are probably missing both CYP2D6 and CYP2C19 (Johnson et al., 2006). Although I have no prospective data on these subjects, my lecture 6 on “Pharmacogenetic testing in psychiatry” recommends either bupropion, mirtazapine or reboxetine for these patients when they need antidepressants (vilazadone may be another option).


4) The lack of data on the relevance of pre-stratification using pharmacological mechanisms.  I have spent many years studying risperidone pharmacokinetics but, although I may presume high proficiency in that area, I am afraid that I am completely incompetent in obtaining funding for RCTs. I have developed a prospective RCT design to study CYP2D6 poor metabolizers (PMs) and ultrarapid metabolizers (UMs), but I have not been able to convince anybody to fund the RCT. Since talk is “cheap,” I will recount a long history that demonstrates, on the other hand, that ignoring pre-stratification by pharmacological mechanisms probably cost the risperidone marketer millions of dollars by making one of their RCTs a partial failure (Yatham et al., 2003).

            According to the marketer’s story, risperidone is metabolized by CYP2D6 (Huang et al., 1993).  In 1996, I discovered that, in one patient from my unit, carbamazepine was decreasing his risperidone TDM by half, which suggested to me that carbamazepine was a risperidone inducer. It was a very unusual finding since CYP2D6 cannot be induced. Based on that case, I hypothesized that risperidone was metabolized by CYP3A4, too (de Leon and Bork, 1997).  DDIs from other patients appeared to suggest that, in effect, risperidone was metabolized by both CYP2D6 and CYP3A4 (Bork et al., 1999). When we presented the data at a meeting in 1997 (Bork et al., 1997), staff from the risperidone company appeared to indicate that they knew about these findings (de Leon, 2014). Then, in a better-controlled study, Spina et al. (2000) definitively established that carbamazepine is a risperidone inducer.  Finally, an in vitro study demonstrated that, as I had hypothesized, risperidone is metabolized by CYP3A4, too (Yasui-Furuk et al., 2001).  The risperidone marketer appeared to prefer ignoring the evidence that risperidone is also metabolized by CYP3A4 and that carbamazepine is a risperidone inducer.  When they conducted an RCT for risperidone as an adjunctive treatment for mania (Yatham et al., 2003), risperidone was effective in the lithium and valproate arms, but ignoring carbamazepine’s inductive effect was costly because risperidone was not more effective than placebo in the carbamazepine arm. By the way, after 20 years of reviewing the literature, including prospective DDI and TDM studies (Spina and de Leon, 2007; de Leon et al., 2012; Spina et al., 2016), we have pretty well established that giving carbamazepine is equivalent, on average, to reducing the risperidone dose by half, as the original case report suggested (de Leon and Bork, 1997). The risperidone marketer should have stratified the patients on carbamazepine to a pill with double the dose of risperidone. This design would have made the carbamazepine arm of the RCT better than placebo. 


5) The differences between mathematical and mechanistic science in psychopharmacology. 

I am afraid that Dr. Klein, as with most current psychiatric researchers, is too attached to the mathematical scientific approach in psychopharmacology and does not pay enough attention to the mechanistic scientific approach. I believe that science needs both the mathematical and mechanistic approaches (de Leon and De las Cuevas, in press).  This is not an original idea. To clarify my terminology, mechanisms are abstract concepts for organizing realities which cannot be demonstrated in experiments, but are needed for making hypotheses and interpreting experiments (de Leon and De las Cuevas, in press). Biology is a mechanistic science since it was founded by Aristotle (Leroi, 2014). Galileo gave a major push to mathematical science in the 17th century when he proposed that the laws of nature are mathematical.

            In 19th century France, Pierre Charles Alexandre Louis defended the mathematical approach in medicine. Benigno Risueño de Amador rejected the mathematical approach in medicine because for him medicine was an art. Claude Bernard defended an intermediate approach that mathematical approaches were important, particularly when mechanisms were not known. On the other hand, he thought that mathematical approaches are more limited in order to demonstrate mechanisms that are better established by experimental methods. My scientific approach follows Bernard’s ideas (de Leon and De las Cuevas, in press). These are three of Bernard’s sentences, translated by Morabia (2006), that exemplify his thinking: 1) “By destroying the biological characteristics of phenomena, the use of averages in physiology and medicine usually only gives apparent accuracy to the results”   2) “Statistics therefore apply only to cases in which the cause of the facts observed is still indeterminate” and 3) “But when determinism increases, statistics can no longer grasp and confine it within a limit of variations.”

            Later on, the combination of three types of mechanistic thinking (anatomoclinical, physiopathological and etiopathological thinking) led to the advance of medicine in the 20th century (Lain Entralgo, 1978). Developments in statistics in the 20th century allowed for the full implementation of mathematical science in medicine. In summary, in my view both mathematical and mechanistic approaches are needed in psychopharmacology.



Bork JA, Rogers T, Wedlund P, Chou W, de Leon J. Risperidone metabolism and drug interaction in treatment refractory patients. New Research Program and Abstracts (NR 117). American Psychiatric Association 150th Annual Meeting. San Diego, CA. May 30-June 4, 1997.


Bork JA, Rogers T, Wedlund PJ, de Leon J. A pilot study on risperidone metabolism: the role of cytochromes P450 2D6 and 3A. Journal of Clinical Psychiatry 1999; 60: 469-476.


Botts S, Diaz FJ, Santoro V, Spina E, Muscatello MR, Cogollo M, Castro FE, de  Leon J. Estimating the effects of co-medications on plasma olanzapine concentrations by using a mixed model. Progress in Neuropsychopharmacology & Biological Psychiatry 2008; 32: 1453-1458.


de Leon J. Highlights of drug package inserts and the website DailyMed: the need for further improvement in package inserts to help busy prescribers. Journal of Clinical Psychopharmacology 2011; 31: 263-265.


de Leon J. Evidence-based medicine versus personalized medicine: are they enemies? Journal of

Clinical Psychopharmacology 2012, 32: 153-164. Pre-published free version:


de Leon J. False-negative studies may systematically contaminate the literature on the effects of inducers in neuropsychopharmacology: part II: focus  on bipolar disorder. Journal of Clinical Psychopharmacology 2014; 34: 291-296.


de Leon J, Bork J. Risperidone and cytochrome P450 3A. Journal of Clinical Psychiatry 1997; 58: 450.


de Leon J, De las Cuevas C. The art of pharmacotherapy: reflections on pharmacophobia. Journal of Clinical Psychopharmacology  (in press).


de Leon J, Santoro V, D'Arrigo C, Spina E. Interactions between antiepileptics and second-generation antipsychotics. Expert Opinion Drug Metabolism and Toxicology 2012; 8: 311-334.


Diaz FJ, Santoro V, Spina E, Cogollo M, Rivera TE, Botts S, de Leon J. Estimating the size of the effects of co-medications on plasma clozapine concentrations using a model that controls for clozapine doses and confounding variables. Pharmacopsychiatry 2008; 41: 81-91.


Huang ML, Van Peer A, Woestenborghs R, De Coster R, Heykants J, Jansen AA, Zylicz Z, Visscher HW, Jonkman JH. Pharmacokinetics of the novel antipsychotic agent risperidone and the prolactin response in healthy subjects. Clinical Pharmacology & Therapeutics 1993; 54: 257-268.


Jackson J, McCollum B, Ognibene J, Diaz FJ, de Leon J. Three patients needing high doses of

valproic acid to get therapeutic concentrations. Case Report in Psychiatry 2015; 2015: 542862.


Johnson M, Markham-Abedi C, Susce MT, Murray-Carmichael E, McCollum S, de Leon J. A poor metabolizer for cytochromes P450 2D6 and 2C19: a case report on antidepressant treatment. CNS Spectrums 2006; 11: 757-760.


Lain Entralgo P. Historia de la Medicina. Barcelona, Spain: Salvat; 1978.


Leroi A. The Lagoon: How Aristotle Invented Science. New York, NY: Viking; 2014.


Morabia A. Claude Bernard was a 19th century proponent of medicine based on evidence. Journal of Clinical Epidemiology 2006; 59: 1150-1154.


Spina E, Avenoso A, Facciolà G, Salemi M, Scordo MG, Giacobello T, Madia AG, Perucca E. Plasma concentrations of risperidone and 9-hydroxyrisperidone: effect of comedication with carbamazepine or valproate. Therapeutic Drug Monitoring 2000; 22: 481-485.


Spina E, D'Arrigo C, Santoro V, Muscatello MR, Pandolfo G, Zoccali R, Diaz FJ, de Leon J. Effect of valproate on olanzapine plasma concentrations in patients with bipolar or schizoaffective disorder. Theraputic Drug Monitoring 2009; 31: 758-763.


Spina E, de Leon J. Metabolic drug interactions with newer antipsychotics: a comparative review. Basic Clinical Pharmacology and Toxicology 2007;100: 4-22.


Spina E, Hiemke C, de Leon J. Assessing drug-drug interactions through therapeutic drug monitoring when administering oral second-generation antipsychotics. Expert Opinion Drug Metabolism and Toxicology 2016; 12: 407-422.


Yasui-Furukori N, Hidestrand M, Spina E, Facciolá G, Scordo MG, Tybring G. Different enantioselective 9-hydroxylation of risperidone by the two human CYP2D6 and CYP3A4 enzymes. Drug Metabolism and Disposition 2001; 29: 1263-1268.


Yatham LN, Grossman F, Augustyns I, Vieta E, Ravindran A. Mood stabilisers plus risperidone or placebo in the treatment of acute mania. International, double-blind, randomised controlled trial. British Journal of Psychiatry 2003; 182: 141-147.

Jose de Leon

March 2, 2017